linux_dsm_epyc7002/drivers/mmc/host/mmci.c

1973 lines
48 KiB
C
Raw Normal View History

/*
* linux/drivers/mmc/host/mmci.c - ARM PrimeCell MMCI PL180/1 driver
*
* Copyright (C) 2003 Deep Blue Solutions, Ltd, All Rights Reserved.
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
* Copyright (C) 2010 ST-Ericsson SA
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/init.h>
#include <linux/ioport.h>
#include <linux/device.h>
#include <linux/io.h>
#include <linux/interrupt.h>
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/delay.h>
#include <linux/err.h>
#include <linux/highmem.h>
#include <linux/log2.h>
#include <linux/mmc/pm.h>
#include <linux/mmc/host.h>
#include <linux/mmc/card.h>
#include <linux/mmc/slot-gpio.h>
#include <linux/amba/bus.h>
#include <linux/clk.h>
#include <linux/scatterlist.h>
#include <linux/of.h>
#include <linux/regulator/consumer.h>
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
#include <linux/dmaengine.h>
#include <linux/dma-mapping.h>
#include <linux/amba/mmci.h>
#include <linux/pm_runtime.h>
#include <linux/types.h>
#include <linux/pinctrl/consumer.h>
#include <asm/div64.h>
#include <asm/io.h>
#include "mmci.h"
#include "mmci_qcom_dml.h"
#define DRIVER_NAME "mmci-pl18x"
static unsigned int fmax = 515633;
static struct variant_data variant_arm = {
.fifosize = 16 * 4,
.fifohalfsize = 8 * 4,
.datalength_bits = 16,
.pwrreg_powerup = MCI_PWR_UP,
.f_max = 100000000,
.reversed_irq_handling = true,
.mmcimask1 = true,
.start_err = MCI_STARTBITERR,
.opendrain = MCI_ROD,
};
static struct variant_data variant_arm_extended_fifo = {
.fifosize = 128 * 4,
.fifohalfsize = 64 * 4,
.datalength_bits = 16,
.pwrreg_powerup = MCI_PWR_UP,
.f_max = 100000000,
.mmcimask1 = true,
.start_err = MCI_STARTBITERR,
.opendrain = MCI_ROD,
};
static struct variant_data variant_arm_extended_fifo_hwfc = {
.fifosize = 128 * 4,
.fifohalfsize = 64 * 4,
.clkreg_enable = MCI_ARM_HWFCEN,
.datalength_bits = 16,
.pwrreg_powerup = MCI_PWR_UP,
.f_max = 100000000,
.mmcimask1 = true,
.start_err = MCI_STARTBITERR,
.opendrain = MCI_ROD,
};
static struct variant_data variant_u300 = {
.fifosize = 16 * 4,
.fifohalfsize = 8 * 4,
.clkreg_enable = MCI_ST_U300_HWFCEN,
.clkreg_8bit_bus_enable = MCI_ST_8BIT_BUS,
.datalength_bits = 16,
.datactrl_mask_sdio = MCI_DPSM_ST_SDIOEN,
.st_sdio = true,
.pwrreg_powerup = MCI_PWR_ON,
.f_max = 100000000,
.signal_direction = true,
.pwrreg_clkgate = true,
.pwrreg_nopower = true,
.mmcimask1 = true,
.start_err = MCI_STARTBITERR,
.opendrain = MCI_OD,
};
static struct variant_data variant_nomadik = {
.fifosize = 16 * 4,
.fifohalfsize = 8 * 4,
.clkreg = MCI_CLK_ENABLE,
.clkreg_8bit_bus_enable = MCI_ST_8BIT_BUS,
.datalength_bits = 24,
.datactrl_mask_sdio = MCI_DPSM_ST_SDIOEN,
.st_sdio = true,
.st_clkdiv = true,
.pwrreg_powerup = MCI_PWR_ON,
.f_max = 100000000,
.signal_direction = true,
.pwrreg_clkgate = true,
.pwrreg_nopower = true,
.mmcimask1 = true,
.start_err = MCI_STARTBITERR,
.opendrain = MCI_OD,
};
static struct variant_data variant_ux500 = {
.fifosize = 30 * 4,
.fifohalfsize = 8 * 4,
.clkreg = MCI_CLK_ENABLE,
.clkreg_enable = MCI_ST_UX500_HWFCEN,
.clkreg_8bit_bus_enable = MCI_ST_8BIT_BUS,
.clkreg_neg_edge_enable = MCI_ST_UX500_NEG_EDGE,
.datalength_bits = 24,
.datactrl_mask_sdio = MCI_DPSM_ST_SDIOEN,
.st_sdio = true,
.st_clkdiv = true,
.pwrreg_powerup = MCI_PWR_ON,
.f_max = 100000000,
.signal_direction = true,
.pwrreg_clkgate = true,
.busy_detect = true,
.busy_dpsm_flag = MCI_DPSM_ST_BUSYMODE,
.busy_detect_flag = MCI_ST_CARDBUSY,
.busy_detect_mask = MCI_ST_BUSYENDMASK,
.pwrreg_nopower = true,
.mmcimask1 = true,
.start_err = MCI_STARTBITERR,
.opendrain = MCI_OD,
};
static struct variant_data variant_ux500v2 = {
.fifosize = 30 * 4,
.fifohalfsize = 8 * 4,
.clkreg = MCI_CLK_ENABLE,
.clkreg_enable = MCI_ST_UX500_HWFCEN,
.clkreg_8bit_bus_enable = MCI_ST_8BIT_BUS,
.clkreg_neg_edge_enable = MCI_ST_UX500_NEG_EDGE,
.datactrl_mask_ddrmode = MCI_DPSM_ST_DDRMODE,
.datalength_bits = 24,
.datactrl_mask_sdio = MCI_DPSM_ST_SDIOEN,
.st_sdio = true,
.st_clkdiv = true,
.blksz_datactrl16 = true,
.pwrreg_powerup = MCI_PWR_ON,
.f_max = 100000000,
.signal_direction = true,
.pwrreg_clkgate = true,
.busy_detect = true,
.busy_dpsm_flag = MCI_DPSM_ST_BUSYMODE,
.busy_detect_flag = MCI_ST_CARDBUSY,
.busy_detect_mask = MCI_ST_BUSYENDMASK,
.pwrreg_nopower = true,
.mmcimask1 = true,
.start_err = MCI_STARTBITERR,
.opendrain = MCI_OD,
};
static struct variant_data variant_stm32 = {
.fifosize = 32 * 4,
.fifohalfsize = 8 * 4,
.clkreg = MCI_CLK_ENABLE,
.clkreg_enable = MCI_ST_UX500_HWFCEN,
.clkreg_8bit_bus_enable = MCI_ST_8BIT_BUS,
.clkreg_neg_edge_enable = MCI_ST_UX500_NEG_EDGE,
.datalength_bits = 24,
.datactrl_mask_sdio = MCI_DPSM_ST_SDIOEN,
.st_sdio = true,
.st_clkdiv = true,
.pwrreg_powerup = MCI_PWR_ON,
.f_max = 48000000,
.pwrreg_clkgate = true,
.pwrreg_nopower = true,
};
static struct variant_data variant_qcom = {
.fifosize = 16 * 4,
.fifohalfsize = 8 * 4,
.clkreg = MCI_CLK_ENABLE,
.clkreg_enable = MCI_QCOM_CLK_FLOWENA |
MCI_QCOM_CLK_SELECT_IN_FBCLK,
.clkreg_8bit_bus_enable = MCI_QCOM_CLK_WIDEBUS_8,
.datactrl_mask_ddrmode = MCI_QCOM_CLK_SELECT_IN_DDR_MODE,
.data_cmd_enable = MCI_CPSM_QCOM_DATCMD,
.blksz_datactrl4 = true,
.datalength_bits = 24,
.pwrreg_powerup = MCI_PWR_UP,
.f_max = 208000000,
.explicit_mclk_control = true,
.qcom_fifo = true,
.qcom_dml = true,
.mmcimask1 = true,
.start_err = MCI_STARTBITERR,
.opendrain = MCI_ROD,
.init = qcom_variant_init,
};
/* Busy detection for the ST Micro variant */
static int mmci_card_busy(struct mmc_host *mmc)
{
struct mmci_host *host = mmc_priv(mmc);
unsigned long flags;
int busy = 0;
spin_lock_irqsave(&host->lock, flags);
if (readl(host->base + MMCISTATUS) & host->variant->busy_detect_flag)
busy = 1;
spin_unlock_irqrestore(&host->lock, flags);
return busy;
}
/*
* Validate mmc prerequisites
*/
static int mmci_validate_data(struct mmci_host *host,
struct mmc_data *data)
{
if (!data)
return 0;
if (!is_power_of_2(data->blksz)) {
dev_err(mmc_dev(host->mmc),
"unsupported block size (%d bytes)\n", data->blksz);
return -EINVAL;
}
return 0;
}
static void mmci_reg_delay(struct mmci_host *host)
{
/*
* According to the spec, at least three feedback clock cycles
* of max 52 MHz must pass between two writes to the MMCICLOCK reg.
* Three MCLK clock cycles must pass between two MMCIPOWER reg writes.
* Worst delay time during card init is at 100 kHz => 30 us.
* Worst delay time when up and running is at 25 MHz => 120 ns.
*/
if (host->cclk < 25000000)
udelay(30);
else
ndelay(120);
}
/*
* This must be called with host->lock held
*/
static void mmci_write_clkreg(struct mmci_host *host, u32 clk)
{
if (host->clk_reg != clk) {
host->clk_reg = clk;
writel(clk, host->base + MMCICLOCK);
}
}
/*
* This must be called with host->lock held
*/
static void mmci_write_pwrreg(struct mmci_host *host, u32 pwr)
{
if (host->pwr_reg != pwr) {
host->pwr_reg = pwr;
writel(pwr, host->base + MMCIPOWER);
}
}
/*
* This must be called with host->lock held
*/
static void mmci_write_datactrlreg(struct mmci_host *host, u32 datactrl)
{
/* Keep busy mode in DPSM if enabled */
datactrl |= host->datactrl_reg & host->variant->busy_dpsm_flag;
if (host->datactrl_reg != datactrl) {
host->datactrl_reg = datactrl;
writel(datactrl, host->base + MMCIDATACTRL);
}
}
/*
* This must be called with host->lock held
*/
static void mmci_set_clkreg(struct mmci_host *host, unsigned int desired)
{
struct variant_data *variant = host->variant;
u32 clk = variant->clkreg;
/* Make sure cclk reflects the current calculated clock */
host->cclk = 0;
if (desired) {
if (variant->explicit_mclk_control) {
host->cclk = host->mclk;
} else if (desired >= host->mclk) {
clk = MCI_CLK_BYPASS;
if (variant->st_clkdiv)
clk |= MCI_ST_UX500_NEG_EDGE;
host->cclk = host->mclk;
} else if (variant->st_clkdiv) {
/*
* DB8500 TRM says f = mclk / (clkdiv + 2)
* => clkdiv = (mclk / f) - 2
* Round the divider up so we don't exceed the max
* frequency
*/
clk = DIV_ROUND_UP(host->mclk, desired) - 2;
if (clk >= 256)
clk = 255;
host->cclk = host->mclk / (clk + 2);
} else {
/*
* PL180 TRM says f = mclk / (2 * (clkdiv + 1))
* => clkdiv = mclk / (2 * f) - 1
*/
clk = host->mclk / (2 * desired) - 1;
if (clk >= 256)
clk = 255;
host->cclk = host->mclk / (2 * (clk + 1));
}
clk |= variant->clkreg_enable;
clk |= MCI_CLK_ENABLE;
/* This hasn't proven to be worthwhile */
/* clk |= MCI_CLK_PWRSAVE; */
}
/* Set actual clock for debug */
host->mmc->actual_clock = host->cclk;
if (host->mmc->ios.bus_width == MMC_BUS_WIDTH_4)
clk |= MCI_4BIT_BUS;
if (host->mmc->ios.bus_width == MMC_BUS_WIDTH_8)
clk |= variant->clkreg_8bit_bus_enable;
if (host->mmc->ios.timing == MMC_TIMING_UHS_DDR50 ||
host->mmc->ios.timing == MMC_TIMING_MMC_DDR52)
clk |= variant->clkreg_neg_edge_enable;
mmci_write_clkreg(host, clk);
}
static void
mmci_request_end(struct mmci_host *host, struct mmc_request *mrq)
{
writel(0, host->base + MMCICOMMAND);
BUG_ON(host->data);
host->mrq = NULL;
host->cmd = NULL;
mmc_request_done(host->mmc, mrq);
}
static void mmci_set_mask1(struct mmci_host *host, unsigned int mask)
{
void __iomem *base = host->base;
struct variant_data *variant = host->variant;
if (host->singleirq) {
unsigned int mask0 = readl(base + MMCIMASK0);
mask0 &= ~MCI_IRQ1MASK;
mask0 |= mask;
writel(mask0, base + MMCIMASK0);
}
if (variant->mmcimask1)
writel(mask, base + MMCIMASK1);
host->mask1_reg = mask;
}
static void mmci_stop_data(struct mmci_host *host)
{
mmci_write_datactrlreg(host, 0);
mmci_set_mask1(host, 0);
host->data = NULL;
}
static void mmci_init_sg(struct mmci_host *host, struct mmc_data *data)
{
unsigned int flags = SG_MITER_ATOMIC;
if (data->flags & MMC_DATA_READ)
flags |= SG_MITER_TO_SG;
else
flags |= SG_MITER_FROM_SG;
sg_miter_start(&host->sg_miter, data->sg, data->sg_len, flags);
}
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
/*
* All the DMA operation mode stuff goes inside this ifdef.
* This assumes that you have a generic DMA device interface,
* no custom DMA interfaces are supported.
*/
#ifdef CONFIG_DMA_ENGINE
static void mmci_dma_setup(struct mmci_host *host)
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
{
const char *rxname, *txname;
host->dma_rx_channel = dma_request_slave_channel(mmc_dev(host->mmc), "rx");
host->dma_tx_channel = dma_request_slave_channel(mmc_dev(host->mmc), "tx");
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
/* initialize pre request cookie */
host->next_data.cookie = 1;
/*
* If only an RX channel is specified, the driver will
* attempt to use it bidirectionally, however if it is
* is specified but cannot be located, DMA will be disabled.
*/
if (host->dma_rx_channel && !host->dma_tx_channel)
host->dma_tx_channel = host->dma_rx_channel;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (host->dma_rx_channel)
rxname = dma_chan_name(host->dma_rx_channel);
else
rxname = "none";
if (host->dma_tx_channel)
txname = dma_chan_name(host->dma_tx_channel);
else
txname = "none";
dev_info(mmc_dev(host->mmc), "DMA channels RX %s, TX %s\n",
rxname, txname);
/*
* Limit the maximum segment size in any SG entry according to
* the parameters of the DMA engine device.
*/
if (host->dma_tx_channel) {
struct device *dev = host->dma_tx_channel->device->dev;
unsigned int max_seg_size = dma_get_max_seg_size(dev);
if (max_seg_size < host->mmc->max_seg_size)
host->mmc->max_seg_size = max_seg_size;
}
if (host->dma_rx_channel) {
struct device *dev = host->dma_rx_channel->device->dev;
unsigned int max_seg_size = dma_get_max_seg_size(dev);
if (max_seg_size < host->mmc->max_seg_size)
host->mmc->max_seg_size = max_seg_size;
}
if (host->ops && host->ops->dma_setup)
host->ops->dma_setup(host);
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
}
/*
* This is used in or so inline it
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
* so it can be discarded.
*/
static inline void mmci_dma_release(struct mmci_host *host)
{
if (host->dma_rx_channel)
dma_release_channel(host->dma_rx_channel);
if (host->dma_tx_channel)
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
dma_release_channel(host->dma_tx_channel);
host->dma_rx_channel = host->dma_tx_channel = NULL;
}
static void mmci_dma_unmap(struct mmci_host *host, struct mmc_data *data)
{
struct dma_chan *chan;
if (data->flags & MMC_DATA_READ)
chan = host->dma_rx_channel;
else
chan = host->dma_tx_channel;
dma_unmap_sg(chan->device->dev, data->sg, data->sg_len,
mmc_get_dma_dir(data));
}
static void mmci_dma_data_error(struct mmci_host *host)
{
dev_err(mmc_dev(host->mmc), "error during DMA transfer!\n");
dmaengine_terminate_all(host->dma_current);
host->dma_in_progress = false;
host->dma_current = NULL;
host->dma_desc_current = NULL;
host->data->host_cookie = 0;
mmci_dma_unmap(host, host->data);
}
static void mmci_dma_finalize(struct mmci_host *host, struct mmc_data *data)
{
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
u32 status;
int i;
/* Wait up to 1ms for the DMA to complete */
for (i = 0; ; i++) {
status = readl(host->base + MMCISTATUS);
if (!(status & MCI_RXDATAAVLBLMASK) || i >= 100)
break;
udelay(10);
}
/*
* Check to see whether we still have some data left in the FIFO -
* this catches DMA controllers which are unable to monitor the
* DMALBREQ and DMALSREQ signals while allowing us to DMA to non-
* contiguous buffers. On TX, we'll get a FIFO underrun error.
*/
if (status & MCI_RXDATAAVLBLMASK) {
mmci_dma_data_error(host);
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (!data->error)
data->error = -EIO;
} else if (!data->host_cookie) {
mmci_dma_unmap(host, data);
}
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
/*
* Use of DMA with scatter-gather is impossible.
* Give up with DMA and switch back to PIO mode.
*/
if (status & MCI_RXDATAAVLBLMASK) {
dev_err(mmc_dev(host->mmc), "buggy DMA detected. Taking evasive action.\n");
mmci_dma_release(host);
}
mmc: core/mmci: restore pre/post_req behaviour commit 64b12a68a9f74bb32d8efd7af1ad8a2ba02fc884 "mmc: core: fix prepared requests while doing bkops" is fixing a bug in the wrong way. A bug in the MMCI device driver is fixed by amending the MMC core. Thinking about it: what the pre- and post-callbacks are doing is to essentially map and unmap SG lists for DMA transfers. Why would we not be able to do that just because a BKOPS command is sent inbetween? Having to unprepare/prepare the next asynchronous request for DMA seems wrong. Looking the backtrace in that commit we can see what the real problem actually is: mmci_data_irq() is calling mmci_dma_unmap() twice which is goung to call arm_dma_unmap_sg() twice and v7_dma_inv_range() twice for the same sglist and that will crash. This happens because a request is prepared, then a BKOPS is sent. The IRQ completing the BKOPS command goes through mmci_data_irq() and thinks that a DMA operation has just been completed because dma_inprogress() reports true. It then proceeds to unmap the sglist. But that was wrong! dma_inprogress() should NOT be true because no DMA was actually in progress! We had just prepared the sglist, and the DMA channel dma_current has been configured, but NOT started! Because of this, the sglist is already unmapped when we get our actual data completion IRQ, and we are unmapping the sglist once more, and we get this crash. Therefore, we need to revert this solution pushing the problem to the core and causing problems, and instead augment the implementation such that dma_inprogress() only reports true if some DMA has actually been started. After this we can keep the request prepared during the BKOPS and we need not unprepare/reprepare it. Fixes: 64b12a68a9f7 ("mmc: core: fix prepared requests while doing bkops") Cc: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-01-27 21:04:54 +07:00
host->dma_in_progress = false;
host->dma_current = NULL;
host->dma_desc_current = NULL;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
}
/* prepares DMA channel and DMA descriptor, returns non-zero on failure */
static int __mmci_dma_prep_data(struct mmci_host *host, struct mmc_data *data,
struct dma_chan **dma_chan,
struct dma_async_tx_descriptor **dma_desc)
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
{
struct variant_data *variant = host->variant;
struct dma_slave_config conf = {
.src_addr = host->phybase + MMCIFIFO,
.dst_addr = host->phybase + MMCIFIFO,
.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES,
.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES,
.src_maxburst = variant->fifohalfsize >> 2, /* # of words */
.dst_maxburst = variant->fifohalfsize >> 2, /* # of words */
.device_fc = false,
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
};
struct dma_chan *chan;
struct dma_device *device;
struct dma_async_tx_descriptor *desc;
int nr_sg;
unsigned long flags = DMA_CTRL_ACK;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (data->flags & MMC_DATA_READ) {
conf.direction = DMA_DEV_TO_MEM;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
chan = host->dma_rx_channel;
} else {
conf.direction = DMA_MEM_TO_DEV;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
chan = host->dma_tx_channel;
}
/* If there's no DMA channel, fall back to PIO */
if (!chan)
return -EINVAL;
/* If less than or equal to the fifo size, don't bother with DMA */
if (data->blksz * data->blocks <= variant->fifosize)
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
return -EINVAL;
device = chan->device;
nr_sg = dma_map_sg(device->dev, data->sg, data->sg_len,
mmc_get_dma_dir(data));
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (nr_sg == 0)
return -EINVAL;
if (host->variant->qcom_dml)
flags |= DMA_PREP_INTERRUPT;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
dmaengine_slave_config(chan, &conf);
desc = dmaengine_prep_slave_sg(chan, data->sg, nr_sg,
conf.direction, flags);
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (!desc)
goto unmap_exit;
*dma_chan = chan;
*dma_desc = desc;
return 0;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
unmap_exit:
dma_unmap_sg(device->dev, data->sg, data->sg_len,
mmc_get_dma_dir(data));
return -ENOMEM;
}
static inline int mmci_dma_prep_data(struct mmci_host *host,
struct mmc_data *data)
{
/* Check if next job is already prepared. */
if (host->dma_current && host->dma_desc_current)
return 0;
/* No job were prepared thus do it now. */
return __mmci_dma_prep_data(host, data, &host->dma_current,
&host->dma_desc_current);
}
static inline int mmci_dma_prep_next(struct mmci_host *host,
struct mmc_data *data)
{
struct mmci_host_next *nd = &host->next_data;
return __mmci_dma_prep_data(host, data, &nd->dma_chan, &nd->dma_desc);
}
static int mmci_dma_start_data(struct mmci_host *host, unsigned int datactrl)
{
int ret;
struct mmc_data *data = host->data;
ret = mmci_dma_prep_data(host, host->data);
if (ret)
return ret;
/* Okay, go for it. */
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
dev_vdbg(mmc_dev(host->mmc),
"Submit MMCI DMA job, sglen %d blksz %04x blks %04x flags %08x\n",
data->sg_len, data->blksz, data->blocks, data->flags);
mmc: core/mmci: restore pre/post_req behaviour commit 64b12a68a9f74bb32d8efd7af1ad8a2ba02fc884 "mmc: core: fix prepared requests while doing bkops" is fixing a bug in the wrong way. A bug in the MMCI device driver is fixed by amending the MMC core. Thinking about it: what the pre- and post-callbacks are doing is to essentially map and unmap SG lists for DMA transfers. Why would we not be able to do that just because a BKOPS command is sent inbetween? Having to unprepare/prepare the next asynchronous request for DMA seems wrong. Looking the backtrace in that commit we can see what the real problem actually is: mmci_data_irq() is calling mmci_dma_unmap() twice which is goung to call arm_dma_unmap_sg() twice and v7_dma_inv_range() twice for the same sglist and that will crash. This happens because a request is prepared, then a BKOPS is sent. The IRQ completing the BKOPS command goes through mmci_data_irq() and thinks that a DMA operation has just been completed because dma_inprogress() reports true. It then proceeds to unmap the sglist. But that was wrong! dma_inprogress() should NOT be true because no DMA was actually in progress! We had just prepared the sglist, and the DMA channel dma_current has been configured, but NOT started! Because of this, the sglist is already unmapped when we get our actual data completion IRQ, and we are unmapping the sglist once more, and we get this crash. Therefore, we need to revert this solution pushing the problem to the core and causing problems, and instead augment the implementation such that dma_inprogress() only reports true if some DMA has actually been started. After this we can keep the request prepared during the BKOPS and we need not unprepare/reprepare it. Fixes: 64b12a68a9f7 ("mmc: core: fix prepared requests while doing bkops") Cc: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-01-27 21:04:54 +07:00
host->dma_in_progress = true;
dmaengine_submit(host->dma_desc_current);
dma_async_issue_pending(host->dma_current);
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (host->variant->qcom_dml)
dml_start_xfer(host, data);
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
datactrl |= MCI_DPSM_DMAENABLE;
/* Trigger the DMA transfer */
mmci_write_datactrlreg(host, datactrl);
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
/*
* Let the MMCI say when the data is ended and it's time
* to fire next DMA request. When that happens, MMCI will
* call mmci_data_end()
*/
writel(readl(host->base + MMCIMASK0) | MCI_DATAENDMASK,
host->base + MMCIMASK0);
return 0;
}
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
static void mmci_get_next_data(struct mmci_host *host, struct mmc_data *data)
{
struct mmci_host_next *next = &host->next_data;
WARN_ON(data->host_cookie && data->host_cookie != next->cookie);
WARN_ON(!data->host_cookie && (next->dma_desc || next->dma_chan));
host->dma_desc_current = next->dma_desc;
host->dma_current = next->dma_chan;
next->dma_desc = NULL;
next->dma_chan = NULL;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
}
static void mmci_pre_request(struct mmc_host *mmc, struct mmc_request *mrq)
{
struct mmci_host *host = mmc_priv(mmc);
struct mmc_data *data = mrq->data;
struct mmci_host_next *nd = &host->next_data;
if (!data)
return;
BUG_ON(data->host_cookie);
if (mmci_validate_data(host, data))
return;
if (!mmci_dma_prep_next(host, data))
data->host_cookie = ++nd->cookie < 0 ? 1 : nd->cookie;
}
static void mmci_post_request(struct mmc_host *mmc, struct mmc_request *mrq,
int err)
{
struct mmci_host *host = mmc_priv(mmc);
struct mmc_data *data = mrq->data;
if (!data || !data->host_cookie)
return;
mmci_dma_unmap(host, data);
if (err) {
struct mmci_host_next *next = &host->next_data;
struct dma_chan *chan;
if (data->flags & MMC_DATA_READ)
chan = host->dma_rx_channel;
else
chan = host->dma_tx_channel;
dmaengine_terminate_all(chan);
mmc: mmci: fix mmci_post_request If the post request is cancelling the channel and descriptor and which are equal to host->dma_current and host->dma_desc_current respectively, then it makes sense to reset these pointers to NULL, so that the driver does not reference it. Also the host_cookie can be reset to 0 in cases of error, so that the core could reissue the same mmc_request. This patch was tested with 'mmc: core: fix prepared requests while doing bkops' to fix the below issue. mmci-pl18x 12400000.sdcc: error during DMA transfer! Unable to handle kernel paging request at virtual address 40000000 pgd = c0204000 [40000000] *pgd=00000000 Internal error: Oops: 805 [#1] SMP ARM Modules linked in: ipv6 ath6kl_sdio ath6kl_core CPU: 0 PID: 0 Comm: swapper/0 Tainted: G W 3.17.0-rc7-linaro-multi-v7 #1 task: c0c9d7e0 ti: c0c92000 task.ti: c0c92000 PC is at v7_dma_inv_range+0x34/0x4c LR is at __dma_page_dev_to_cpu+0x80/0x100 pc : [<c021efc0>] lr : [<c021af18>] psr: 400f0193 sp : c0c93e20 ip : c0c9a478 fp : c08ea538 r10: c0c9f548 r9 : 00000002 r8 : e97d9000 r7 : 00000200 r6 : c0c9d504 r5 : c0db0880 r4 : 00000000 r3 : 0000003f r2 : 00000040 r1 : 40000200 r0 : 40000000 Flags: nZcv IRQs off FIQs on Mode SVC_32 ISA ARM Segment kernel Control: 10c5787d Table: a9ef406a DAC: 00000015 Process swapper/0 (pid: 0, stack limit = 0xc0c92250) Stack: (0xc0c93e20 to 0xc0c94000) 3e20: c021f058 e9a17178 e9a171bc e99dfd6c 00000001 00000001 e995de10 00000002 3e40: 00000000 c021b574 00000000 c04bc4a4 00000000 e9b49ac0 c0ce6e6c e99dfda4 3e60: 00000088 e9810780 c0d8291c c072ea58 00000000 c072d3fc 00000000 c072f534 3e80: 00000000 e9b49ac0 00000100 c0c9a444 00000088 c072f6b4 c072f5d4 e9d40080 3ea0: e98107dc 00000000 00000000 c0280a60 00000000 7d55bf61 e9810780 e98107dc 3ec0: 00000000 f0002000 c0d460e8 c0d460e8 c0c92000 c0280b60 e9810780 c0ce7190 3ee0: 00000000 c028369c c02835f4 00000088 00000088 c0280278 c0c8ec70 c020f080 3f00: f000200c c0c9a958 c0c93f28 c02088e4 c04bd630 c04bd5bc 200f0013 ffffffff 3f20: c0c93f5c c0212800 00000001 a987c000 c0c93f3c c04bd574 00000000 0000015b 3f40: ea7a0e40 00000000 c0d460e8 c0d460e8 c0c92000 c08ea538 29b12000 c0c93f70 3f60: c04bd630 c04bd5bc 200f0013 ffffffff c04bd574 c071bd24 7d50c9b4 c0719a44 3f80: 7d50c9b4 0000015b c0c9a498 c0c92028 c0c9a498 c0c9a4fc ea7a0e40 c0c8ee38 3fa0: c0d460e8 c0276198 00000000 c0d8291a 00000000 c0c9a400 00000000 c0be0bc4 3fc0: ffffffff ffffffff c0be05f8 00000000 00000000 c0c533d8 c0d82ed4 c0c9a47c 3fe0: c0c533d4 c0c9e870 8020406a 511f06f0 00000000 80208074 00000000 00000000 [<c021efc0>] (v7_dma_inv_range) from [<c021af18>] (__dma_page_dev_to_cpu+0x80/0x100) [<c021af18>] (__dma_page_dev_to_cpu) from [<c021b574>] (arm_dma_unmap_sg+0x5c/0x84) [<c021b574>] (arm_dma_unmap_sg) from [<c072ea58>] (mmci_dma_unmap.isra.16+0x60/0x74) [<c072ea58>] (mmci_dma_unmap.isra.16) from [<c072f534>] (mmci_data_irq+0x1fc/0x29c) [<c072f534>] (mmci_data_irq) from [<c072f6b4>] (mmci_irq+0xe0/0x114) [<c072f6b4>] (mmci_irq) from [<c0280a60>] (handle_irq_event_percpu+0x78/0x134) [<c0280a60>] (handle_irq_event_percpu) from [<c0280b60>] (handle_irq_event+0x44/0x64) [<c0280b60>] (handle_irq_event) from [<c028369c>] (handle_fasteoi_irq+0xa8/0x1a8) [<c028369c>] (handle_fasteoi_irq) from [<c0280278>] (generic_handle_irq+0x2c/0x3c) [<c0280278>] (generic_handle_irq) from [<c020f080>] (handle_IRQ+0x40/0x90) [<c020f080>] (handle_IRQ) from [<c02088e4>] (gic_handle_irq+0x38/0x68) [<c02088e4>] (gic_handle_irq) from [<c0212800>] (__irq_svc+0x40/0x54) Exception stack(0xc0c93f28 to 0xc0c93f70) 3f20: 00000001 a987c000 c0c93f3c c04bd574 00000000 0000015b 3f40: ea7a0e40 00000000 c0d460e8 c0d460e8 c0c92000 c08ea538 29b12000 c0c93f70 3f60: c04bd630 c04bd5bc 200f0013 ffffffff [<c0212800>] (__irq_svc) from [<c04bd5bc>] (msm_cpu_pm_enter_sleep+0x48/0x4c) [<c04bd5bc>] (msm_cpu_pm_enter_sleep) from [<c071bd24>] (qcom_lpm_enter_spc+0x20/0x2c) [<c071bd24>] (qcom_lpm_enter_spc) from [<c0719a44>] (cpuidle_enter_state+0x44/0xf0) [<c0719a44>] (cpuidle_enter_state) from [<c0276198>] (cpu_startup_entry+0x1f4/0x238) [<c0276198>] (cpu_startup_entry) from [<c0be0bc4>] (start_kernel+0x384/0x390) Code: 1e070f3e e1110003 e1c11003 1e071f3e (ee070f36) ---[ end trace cf6cb3f6432c9834 ]--- Kernel panic - not syncing: Fatal exception in interrupt Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2014-10-08 18:25:17 +07:00
if (host->dma_desc_current == next->dma_desc)
host->dma_desc_current = NULL;
mmc: core/mmci: restore pre/post_req behaviour commit 64b12a68a9f74bb32d8efd7af1ad8a2ba02fc884 "mmc: core: fix prepared requests while doing bkops" is fixing a bug in the wrong way. A bug in the MMCI device driver is fixed by amending the MMC core. Thinking about it: what the pre- and post-callbacks are doing is to essentially map and unmap SG lists for DMA transfers. Why would we not be able to do that just because a BKOPS command is sent inbetween? Having to unprepare/prepare the next asynchronous request for DMA seems wrong. Looking the backtrace in that commit we can see what the real problem actually is: mmci_data_irq() is calling mmci_dma_unmap() twice which is goung to call arm_dma_unmap_sg() twice and v7_dma_inv_range() twice for the same sglist and that will crash. This happens because a request is prepared, then a BKOPS is sent. The IRQ completing the BKOPS command goes through mmci_data_irq() and thinks that a DMA operation has just been completed because dma_inprogress() reports true. It then proceeds to unmap the sglist. But that was wrong! dma_inprogress() should NOT be true because no DMA was actually in progress! We had just prepared the sglist, and the DMA channel dma_current has been configured, but NOT started! Because of this, the sglist is already unmapped when we get our actual data completion IRQ, and we are unmapping the sglist once more, and we get this crash. Therefore, we need to revert this solution pushing the problem to the core and causing problems, and instead augment the implementation such that dma_inprogress() only reports true if some DMA has actually been started. After this we can keep the request prepared during the BKOPS and we need not unprepare/reprepare it. Fixes: 64b12a68a9f7 ("mmc: core: fix prepared requests while doing bkops") Cc: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-01-27 21:04:54 +07:00
if (host->dma_current == next->dma_chan) {
host->dma_in_progress = false;
mmc: mmci: fix mmci_post_request If the post request is cancelling the channel and descriptor and which are equal to host->dma_current and host->dma_desc_current respectively, then it makes sense to reset these pointers to NULL, so that the driver does not reference it. Also the host_cookie can be reset to 0 in cases of error, so that the core could reissue the same mmc_request. This patch was tested with 'mmc: core: fix prepared requests while doing bkops' to fix the below issue. mmci-pl18x 12400000.sdcc: error during DMA transfer! Unable to handle kernel paging request at virtual address 40000000 pgd = c0204000 [40000000] *pgd=00000000 Internal error: Oops: 805 [#1] SMP ARM Modules linked in: ipv6 ath6kl_sdio ath6kl_core CPU: 0 PID: 0 Comm: swapper/0 Tainted: G W 3.17.0-rc7-linaro-multi-v7 #1 task: c0c9d7e0 ti: c0c92000 task.ti: c0c92000 PC is at v7_dma_inv_range+0x34/0x4c LR is at __dma_page_dev_to_cpu+0x80/0x100 pc : [<c021efc0>] lr : [<c021af18>] psr: 400f0193 sp : c0c93e20 ip : c0c9a478 fp : c08ea538 r10: c0c9f548 r9 : 00000002 r8 : e97d9000 r7 : 00000200 r6 : c0c9d504 r5 : c0db0880 r4 : 00000000 r3 : 0000003f r2 : 00000040 r1 : 40000200 r0 : 40000000 Flags: nZcv IRQs off FIQs on Mode SVC_32 ISA ARM Segment kernel Control: 10c5787d Table: a9ef406a DAC: 00000015 Process swapper/0 (pid: 0, stack limit = 0xc0c92250) Stack: (0xc0c93e20 to 0xc0c94000) 3e20: c021f058 e9a17178 e9a171bc e99dfd6c 00000001 00000001 e995de10 00000002 3e40: 00000000 c021b574 00000000 c04bc4a4 00000000 e9b49ac0 c0ce6e6c e99dfda4 3e60: 00000088 e9810780 c0d8291c c072ea58 00000000 c072d3fc 00000000 c072f534 3e80: 00000000 e9b49ac0 00000100 c0c9a444 00000088 c072f6b4 c072f5d4 e9d40080 3ea0: e98107dc 00000000 00000000 c0280a60 00000000 7d55bf61 e9810780 e98107dc 3ec0: 00000000 f0002000 c0d460e8 c0d460e8 c0c92000 c0280b60 e9810780 c0ce7190 3ee0: 00000000 c028369c c02835f4 00000088 00000088 c0280278 c0c8ec70 c020f080 3f00: f000200c c0c9a958 c0c93f28 c02088e4 c04bd630 c04bd5bc 200f0013 ffffffff 3f20: c0c93f5c c0212800 00000001 a987c000 c0c93f3c c04bd574 00000000 0000015b 3f40: ea7a0e40 00000000 c0d460e8 c0d460e8 c0c92000 c08ea538 29b12000 c0c93f70 3f60: c04bd630 c04bd5bc 200f0013 ffffffff c04bd574 c071bd24 7d50c9b4 c0719a44 3f80: 7d50c9b4 0000015b c0c9a498 c0c92028 c0c9a498 c0c9a4fc ea7a0e40 c0c8ee38 3fa0: c0d460e8 c0276198 00000000 c0d8291a 00000000 c0c9a400 00000000 c0be0bc4 3fc0: ffffffff ffffffff c0be05f8 00000000 00000000 c0c533d8 c0d82ed4 c0c9a47c 3fe0: c0c533d4 c0c9e870 8020406a 511f06f0 00000000 80208074 00000000 00000000 [<c021efc0>] (v7_dma_inv_range) from [<c021af18>] (__dma_page_dev_to_cpu+0x80/0x100) [<c021af18>] (__dma_page_dev_to_cpu) from [<c021b574>] (arm_dma_unmap_sg+0x5c/0x84) [<c021b574>] (arm_dma_unmap_sg) from [<c072ea58>] (mmci_dma_unmap.isra.16+0x60/0x74) [<c072ea58>] (mmci_dma_unmap.isra.16) from [<c072f534>] (mmci_data_irq+0x1fc/0x29c) [<c072f534>] (mmci_data_irq) from [<c072f6b4>] (mmci_irq+0xe0/0x114) [<c072f6b4>] (mmci_irq) from [<c0280a60>] (handle_irq_event_percpu+0x78/0x134) [<c0280a60>] (handle_irq_event_percpu) from [<c0280b60>] (handle_irq_event+0x44/0x64) [<c0280b60>] (handle_irq_event) from [<c028369c>] (handle_fasteoi_irq+0xa8/0x1a8) [<c028369c>] (handle_fasteoi_irq) from [<c0280278>] (generic_handle_irq+0x2c/0x3c) [<c0280278>] (generic_handle_irq) from [<c020f080>] (handle_IRQ+0x40/0x90) [<c020f080>] (handle_IRQ) from [<c02088e4>] (gic_handle_irq+0x38/0x68) [<c02088e4>] (gic_handle_irq) from [<c0212800>] (__irq_svc+0x40/0x54) Exception stack(0xc0c93f28 to 0xc0c93f70) 3f20: 00000001 a987c000 c0c93f3c c04bd574 00000000 0000015b 3f40: ea7a0e40 00000000 c0d460e8 c0d460e8 c0c92000 c08ea538 29b12000 c0c93f70 3f60: c04bd630 c04bd5bc 200f0013 ffffffff [<c0212800>] (__irq_svc) from [<c04bd5bc>] (msm_cpu_pm_enter_sleep+0x48/0x4c) [<c04bd5bc>] (msm_cpu_pm_enter_sleep) from [<c071bd24>] (qcom_lpm_enter_spc+0x20/0x2c) [<c071bd24>] (qcom_lpm_enter_spc) from [<c0719a44>] (cpuidle_enter_state+0x44/0xf0) [<c0719a44>] (cpuidle_enter_state) from [<c0276198>] (cpu_startup_entry+0x1f4/0x238) [<c0276198>] (cpu_startup_entry) from [<c0be0bc4>] (start_kernel+0x384/0x390) Code: 1e070f3e e1110003 e1c11003 1e071f3e (ee070f36) ---[ end trace cf6cb3f6432c9834 ]--- Kernel panic - not syncing: Fatal exception in interrupt Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2014-10-08 18:25:17 +07:00
host->dma_current = NULL;
mmc: core/mmci: restore pre/post_req behaviour commit 64b12a68a9f74bb32d8efd7af1ad8a2ba02fc884 "mmc: core: fix prepared requests while doing bkops" is fixing a bug in the wrong way. A bug in the MMCI device driver is fixed by amending the MMC core. Thinking about it: what the pre- and post-callbacks are doing is to essentially map and unmap SG lists for DMA transfers. Why would we not be able to do that just because a BKOPS command is sent inbetween? Having to unprepare/prepare the next asynchronous request for DMA seems wrong. Looking the backtrace in that commit we can see what the real problem actually is: mmci_data_irq() is calling mmci_dma_unmap() twice which is goung to call arm_dma_unmap_sg() twice and v7_dma_inv_range() twice for the same sglist and that will crash. This happens because a request is prepared, then a BKOPS is sent. The IRQ completing the BKOPS command goes through mmci_data_irq() and thinks that a DMA operation has just been completed because dma_inprogress() reports true. It then proceeds to unmap the sglist. But that was wrong! dma_inprogress() should NOT be true because no DMA was actually in progress! We had just prepared the sglist, and the DMA channel dma_current has been configured, but NOT started! Because of this, the sglist is already unmapped when we get our actual data completion IRQ, and we are unmapping the sglist once more, and we get this crash. Therefore, we need to revert this solution pushing the problem to the core and causing problems, and instead augment the implementation such that dma_inprogress() only reports true if some DMA has actually been started. After this we can keep the request prepared during the BKOPS and we need not unprepare/reprepare it. Fixes: 64b12a68a9f7 ("mmc: core: fix prepared requests while doing bkops") Cc: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-01-27 21:04:54 +07:00
}
mmc: mmci: fix mmci_post_request If the post request is cancelling the channel and descriptor and which are equal to host->dma_current and host->dma_desc_current respectively, then it makes sense to reset these pointers to NULL, so that the driver does not reference it. Also the host_cookie can be reset to 0 in cases of error, so that the core could reissue the same mmc_request. This patch was tested with 'mmc: core: fix prepared requests while doing bkops' to fix the below issue. mmci-pl18x 12400000.sdcc: error during DMA transfer! Unable to handle kernel paging request at virtual address 40000000 pgd = c0204000 [40000000] *pgd=00000000 Internal error: Oops: 805 [#1] SMP ARM Modules linked in: ipv6 ath6kl_sdio ath6kl_core CPU: 0 PID: 0 Comm: swapper/0 Tainted: G W 3.17.0-rc7-linaro-multi-v7 #1 task: c0c9d7e0 ti: c0c92000 task.ti: c0c92000 PC is at v7_dma_inv_range+0x34/0x4c LR is at __dma_page_dev_to_cpu+0x80/0x100 pc : [<c021efc0>] lr : [<c021af18>] psr: 400f0193 sp : c0c93e20 ip : c0c9a478 fp : c08ea538 r10: c0c9f548 r9 : 00000002 r8 : e97d9000 r7 : 00000200 r6 : c0c9d504 r5 : c0db0880 r4 : 00000000 r3 : 0000003f r2 : 00000040 r1 : 40000200 r0 : 40000000 Flags: nZcv IRQs off FIQs on Mode SVC_32 ISA ARM Segment kernel Control: 10c5787d Table: a9ef406a DAC: 00000015 Process swapper/0 (pid: 0, stack limit = 0xc0c92250) Stack: (0xc0c93e20 to 0xc0c94000) 3e20: c021f058 e9a17178 e9a171bc e99dfd6c 00000001 00000001 e995de10 00000002 3e40: 00000000 c021b574 00000000 c04bc4a4 00000000 e9b49ac0 c0ce6e6c e99dfda4 3e60: 00000088 e9810780 c0d8291c c072ea58 00000000 c072d3fc 00000000 c072f534 3e80: 00000000 e9b49ac0 00000100 c0c9a444 00000088 c072f6b4 c072f5d4 e9d40080 3ea0: e98107dc 00000000 00000000 c0280a60 00000000 7d55bf61 e9810780 e98107dc 3ec0: 00000000 f0002000 c0d460e8 c0d460e8 c0c92000 c0280b60 e9810780 c0ce7190 3ee0: 00000000 c028369c c02835f4 00000088 00000088 c0280278 c0c8ec70 c020f080 3f00: f000200c c0c9a958 c0c93f28 c02088e4 c04bd630 c04bd5bc 200f0013 ffffffff 3f20: c0c93f5c c0212800 00000001 a987c000 c0c93f3c c04bd574 00000000 0000015b 3f40: ea7a0e40 00000000 c0d460e8 c0d460e8 c0c92000 c08ea538 29b12000 c0c93f70 3f60: c04bd630 c04bd5bc 200f0013 ffffffff c04bd574 c071bd24 7d50c9b4 c0719a44 3f80: 7d50c9b4 0000015b c0c9a498 c0c92028 c0c9a498 c0c9a4fc ea7a0e40 c0c8ee38 3fa0: c0d460e8 c0276198 00000000 c0d8291a 00000000 c0c9a400 00000000 c0be0bc4 3fc0: ffffffff ffffffff c0be05f8 00000000 00000000 c0c533d8 c0d82ed4 c0c9a47c 3fe0: c0c533d4 c0c9e870 8020406a 511f06f0 00000000 80208074 00000000 00000000 [<c021efc0>] (v7_dma_inv_range) from [<c021af18>] (__dma_page_dev_to_cpu+0x80/0x100) [<c021af18>] (__dma_page_dev_to_cpu) from [<c021b574>] (arm_dma_unmap_sg+0x5c/0x84) [<c021b574>] (arm_dma_unmap_sg) from [<c072ea58>] (mmci_dma_unmap.isra.16+0x60/0x74) [<c072ea58>] (mmci_dma_unmap.isra.16) from [<c072f534>] (mmci_data_irq+0x1fc/0x29c) [<c072f534>] (mmci_data_irq) from [<c072f6b4>] (mmci_irq+0xe0/0x114) [<c072f6b4>] (mmci_irq) from [<c0280a60>] (handle_irq_event_percpu+0x78/0x134) [<c0280a60>] (handle_irq_event_percpu) from [<c0280b60>] (handle_irq_event+0x44/0x64) [<c0280b60>] (handle_irq_event) from [<c028369c>] (handle_fasteoi_irq+0xa8/0x1a8) [<c028369c>] (handle_fasteoi_irq) from [<c0280278>] (generic_handle_irq+0x2c/0x3c) [<c0280278>] (generic_handle_irq) from [<c020f080>] (handle_IRQ+0x40/0x90) [<c020f080>] (handle_IRQ) from [<c02088e4>] (gic_handle_irq+0x38/0x68) [<c02088e4>] (gic_handle_irq) from [<c0212800>] (__irq_svc+0x40/0x54) Exception stack(0xc0c93f28 to 0xc0c93f70) 3f20: 00000001 a987c000 c0c93f3c c04bd574 00000000 0000015b 3f40: ea7a0e40 00000000 c0d460e8 c0d460e8 c0c92000 c08ea538 29b12000 c0c93f70 3f60: c04bd630 c04bd5bc 200f0013 ffffffff [<c0212800>] (__irq_svc) from [<c04bd5bc>] (msm_cpu_pm_enter_sleep+0x48/0x4c) [<c04bd5bc>] (msm_cpu_pm_enter_sleep) from [<c071bd24>] (qcom_lpm_enter_spc+0x20/0x2c) [<c071bd24>] (qcom_lpm_enter_spc) from [<c0719a44>] (cpuidle_enter_state+0x44/0xf0) [<c0719a44>] (cpuidle_enter_state) from [<c0276198>] (cpu_startup_entry+0x1f4/0x238) [<c0276198>] (cpu_startup_entry) from [<c0be0bc4>] (start_kernel+0x384/0x390) Code: 1e070f3e e1110003 e1c11003 1e071f3e (ee070f36) ---[ end trace cf6cb3f6432c9834 ]--- Kernel panic - not syncing: Fatal exception in interrupt Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2014-10-08 18:25:17 +07:00
next->dma_desc = NULL;
next->dma_chan = NULL;
mmc: mmci: fix mmci_post_request If the post request is cancelling the channel and descriptor and which are equal to host->dma_current and host->dma_desc_current respectively, then it makes sense to reset these pointers to NULL, so that the driver does not reference it. Also the host_cookie can be reset to 0 in cases of error, so that the core could reissue the same mmc_request. This patch was tested with 'mmc: core: fix prepared requests while doing bkops' to fix the below issue. mmci-pl18x 12400000.sdcc: error during DMA transfer! Unable to handle kernel paging request at virtual address 40000000 pgd = c0204000 [40000000] *pgd=00000000 Internal error: Oops: 805 [#1] SMP ARM Modules linked in: ipv6 ath6kl_sdio ath6kl_core CPU: 0 PID: 0 Comm: swapper/0 Tainted: G W 3.17.0-rc7-linaro-multi-v7 #1 task: c0c9d7e0 ti: c0c92000 task.ti: c0c92000 PC is at v7_dma_inv_range+0x34/0x4c LR is at __dma_page_dev_to_cpu+0x80/0x100 pc : [<c021efc0>] lr : [<c021af18>] psr: 400f0193 sp : c0c93e20 ip : c0c9a478 fp : c08ea538 r10: c0c9f548 r9 : 00000002 r8 : e97d9000 r7 : 00000200 r6 : c0c9d504 r5 : c0db0880 r4 : 00000000 r3 : 0000003f r2 : 00000040 r1 : 40000200 r0 : 40000000 Flags: nZcv IRQs off FIQs on Mode SVC_32 ISA ARM Segment kernel Control: 10c5787d Table: a9ef406a DAC: 00000015 Process swapper/0 (pid: 0, stack limit = 0xc0c92250) Stack: (0xc0c93e20 to 0xc0c94000) 3e20: c021f058 e9a17178 e9a171bc e99dfd6c 00000001 00000001 e995de10 00000002 3e40: 00000000 c021b574 00000000 c04bc4a4 00000000 e9b49ac0 c0ce6e6c e99dfda4 3e60: 00000088 e9810780 c0d8291c c072ea58 00000000 c072d3fc 00000000 c072f534 3e80: 00000000 e9b49ac0 00000100 c0c9a444 00000088 c072f6b4 c072f5d4 e9d40080 3ea0: e98107dc 00000000 00000000 c0280a60 00000000 7d55bf61 e9810780 e98107dc 3ec0: 00000000 f0002000 c0d460e8 c0d460e8 c0c92000 c0280b60 e9810780 c0ce7190 3ee0: 00000000 c028369c c02835f4 00000088 00000088 c0280278 c0c8ec70 c020f080 3f00: f000200c c0c9a958 c0c93f28 c02088e4 c04bd630 c04bd5bc 200f0013 ffffffff 3f20: c0c93f5c c0212800 00000001 a987c000 c0c93f3c c04bd574 00000000 0000015b 3f40: ea7a0e40 00000000 c0d460e8 c0d460e8 c0c92000 c08ea538 29b12000 c0c93f70 3f60: c04bd630 c04bd5bc 200f0013 ffffffff c04bd574 c071bd24 7d50c9b4 c0719a44 3f80: 7d50c9b4 0000015b c0c9a498 c0c92028 c0c9a498 c0c9a4fc ea7a0e40 c0c8ee38 3fa0: c0d460e8 c0276198 00000000 c0d8291a 00000000 c0c9a400 00000000 c0be0bc4 3fc0: ffffffff ffffffff c0be05f8 00000000 00000000 c0c533d8 c0d82ed4 c0c9a47c 3fe0: c0c533d4 c0c9e870 8020406a 511f06f0 00000000 80208074 00000000 00000000 [<c021efc0>] (v7_dma_inv_range) from [<c021af18>] (__dma_page_dev_to_cpu+0x80/0x100) [<c021af18>] (__dma_page_dev_to_cpu) from [<c021b574>] (arm_dma_unmap_sg+0x5c/0x84) [<c021b574>] (arm_dma_unmap_sg) from [<c072ea58>] (mmci_dma_unmap.isra.16+0x60/0x74) [<c072ea58>] (mmci_dma_unmap.isra.16) from [<c072f534>] (mmci_data_irq+0x1fc/0x29c) [<c072f534>] (mmci_data_irq) from [<c072f6b4>] (mmci_irq+0xe0/0x114) [<c072f6b4>] (mmci_irq) from [<c0280a60>] (handle_irq_event_percpu+0x78/0x134) [<c0280a60>] (handle_irq_event_percpu) from [<c0280b60>] (handle_irq_event+0x44/0x64) [<c0280b60>] (handle_irq_event) from [<c028369c>] (handle_fasteoi_irq+0xa8/0x1a8) [<c028369c>] (handle_fasteoi_irq) from [<c0280278>] (generic_handle_irq+0x2c/0x3c) [<c0280278>] (generic_handle_irq) from [<c020f080>] (handle_IRQ+0x40/0x90) [<c020f080>] (handle_IRQ) from [<c02088e4>] (gic_handle_irq+0x38/0x68) [<c02088e4>] (gic_handle_irq) from [<c0212800>] (__irq_svc+0x40/0x54) Exception stack(0xc0c93f28 to 0xc0c93f70) 3f20: 00000001 a987c000 c0c93f3c c04bd574 00000000 0000015b 3f40: ea7a0e40 00000000 c0d460e8 c0d460e8 c0c92000 c08ea538 29b12000 c0c93f70 3f60: c04bd630 c04bd5bc 200f0013 ffffffff [<c0212800>] (__irq_svc) from [<c04bd5bc>] (msm_cpu_pm_enter_sleep+0x48/0x4c) [<c04bd5bc>] (msm_cpu_pm_enter_sleep) from [<c071bd24>] (qcom_lpm_enter_spc+0x20/0x2c) [<c071bd24>] (qcom_lpm_enter_spc) from [<c0719a44>] (cpuidle_enter_state+0x44/0xf0) [<c0719a44>] (cpuidle_enter_state) from [<c0276198>] (cpu_startup_entry+0x1f4/0x238) [<c0276198>] (cpu_startup_entry) from [<c0be0bc4>] (start_kernel+0x384/0x390) Code: 1e070f3e e1110003 e1c11003 1e071f3e (ee070f36) ---[ end trace cf6cb3f6432c9834 ]--- Kernel panic - not syncing: Fatal exception in interrupt Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2014-10-08 18:25:17 +07:00
data->host_cookie = 0;
}
}
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
#else
/* Blank functions if the DMA engine is not available */
static void mmci_get_next_data(struct mmci_host *host, struct mmc_data *data)
{
}
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
static inline void mmci_dma_setup(struct mmci_host *host)
{
}
static inline void mmci_dma_release(struct mmci_host *host)
{
}
static inline void mmci_dma_finalize(struct mmci_host *host,
struct mmc_data *data)
{
}
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
static inline void mmci_dma_data_error(struct mmci_host *host)
{
}
static inline int mmci_dma_start_data(struct mmci_host *host, unsigned int datactrl)
{
return -ENOSYS;
}
#define mmci_pre_request NULL
#define mmci_post_request NULL
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
#endif
static void mmci_start_data(struct mmci_host *host, struct mmc_data *data)
{
struct variant_data *variant = host->variant;
unsigned int datactrl, timeout, irqmask;
unsigned long long clks;
void __iomem *base;
int blksz_bits;
dev_dbg(mmc_dev(host->mmc), "blksz %04x blks %04x flags %08x\n",
data->blksz, data->blocks, data->flags);
host->data = data;
host->size = data->blksz * data->blocks;
data->bytes_xfered = 0;
clks = (unsigned long long)data->timeout_ns * host->cclk;
do_div(clks, NSEC_PER_SEC);
timeout = data->timeout_clks + (unsigned int)clks;
base = host->base;
writel(timeout, base + MMCIDATATIMER);
writel(host->size, base + MMCIDATALENGTH);
blksz_bits = ffs(data->blksz) - 1;
BUG_ON(1 << blksz_bits != data->blksz);
if (variant->blksz_datactrl16)
datactrl = MCI_DPSM_ENABLE | (data->blksz << 16);
else if (variant->blksz_datactrl4)
datactrl = MCI_DPSM_ENABLE | (data->blksz << 4);
else
datactrl = MCI_DPSM_ENABLE | blksz_bits << 4;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (data->flags & MMC_DATA_READ)
datactrl |= MCI_DPSM_DIRECTION;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (host->mmc->card && mmc_card_sdio(host->mmc->card)) {
u32 clk;
datactrl |= variant->datactrl_mask_sdio;
/*
* The ST Micro variant for SDIO small write transfers
* needs to have clock H/W flow control disabled,
* otherwise the transfer will not start. The threshold
* depends on the rate of MCLK.
*/
if (variant->st_sdio && data->flags & MMC_DATA_WRITE &&
(host->size < 8 ||
(host->size <= 8 && host->mclk > 50000000)))
clk = host->clk_reg & ~variant->clkreg_enable;
else
clk = host->clk_reg | variant->clkreg_enable;
mmci_write_clkreg(host, clk);
}
if (host->mmc->ios.timing == MMC_TIMING_UHS_DDR50 ||
host->mmc->ios.timing == MMC_TIMING_MMC_DDR52)
datactrl |= variant->datactrl_mask_ddrmode;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
/*
* Attempt to use DMA operation mode, if this
* should fail, fall back to PIO mode
*/
if (!mmci_dma_start_data(host, datactrl))
return;
/* IRQ mode, map the SG list for CPU reading/writing */
mmci_init_sg(host, data);
if (data->flags & MMC_DATA_READ) {
irqmask = MCI_RXFIFOHALFFULLMASK;
/*
* If we have less than the fifo 'half-full' threshold to
* transfer, trigger a PIO interrupt as soon as any data
* is available.
*/
if (host->size < variant->fifohalfsize)
irqmask |= MCI_RXDATAAVLBLMASK;
} else {
/*
* We don't actually need to include "FIFO empty" here
* since its implicit in "FIFO half empty".
*/
irqmask = MCI_TXFIFOHALFEMPTYMASK;
}
mmci_write_datactrlreg(host, datactrl);
writel(readl(base + MMCIMASK0) & ~MCI_DATAENDMASK, base + MMCIMASK0);
mmci_set_mask1(host, irqmask);
}
static void
mmci_start_command(struct mmci_host *host, struct mmc_command *cmd, u32 c)
{
void __iomem *base = host->base;
dev_dbg(mmc_dev(host->mmc), "op %02x arg %08x flags %08x\n",
cmd->opcode, cmd->arg, cmd->flags);
if (readl(base + MMCICOMMAND) & MCI_CPSM_ENABLE) {
writel(0, base + MMCICOMMAND);
mmci_reg_delay(host);
}
c |= cmd->opcode | MCI_CPSM_ENABLE;
if (cmd->flags & MMC_RSP_PRESENT) {
if (cmd->flags & MMC_RSP_136)
c |= MCI_CPSM_LONGRSP;
c |= MCI_CPSM_RESPONSE;
}
if (/*interrupt*/0)
c |= MCI_CPSM_INTERRUPT;
if (mmc_cmd_type(cmd) == MMC_CMD_ADTC)
c |= host->variant->data_cmd_enable;
host->cmd = cmd;
writel(cmd->arg, base + MMCIARGUMENT);
writel(c, base + MMCICOMMAND);
}
static void
mmci_data_irq(struct mmci_host *host, struct mmc_data *data,
unsigned int status)
{
/* Make sure we have data to handle */
if (!data)
return;
/* First check for errors */
if (status & (MCI_DATACRCFAIL | MCI_DATATIMEOUT |
host->variant->start_err |
MCI_TXUNDERRUN | MCI_RXOVERRUN)) {
u32 remain, success;
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
/* Terminate the DMA transfer */
if (dma_inprogress(host))
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
mmci_dma_data_error(host);
/*
* Calculate how far we are into the transfer. Note that
* the data counter gives the number of bytes transferred
* on the MMC bus, not on the host side. On reads, this
* can be as much as a FIFO-worth of data ahead. This
* matters for FIFO overruns only.
*/
remain = readl(host->base + MMCIDATACNT);
success = data->blksz * data->blocks - remain;
dev_dbg(mmc_dev(host->mmc), "MCI ERROR IRQ, status 0x%08x at 0x%08x\n",
status, success);
if (status & MCI_DATACRCFAIL) {
/* Last block was not successful */
success -= 1;
data->error = -EILSEQ;
} else if (status & MCI_DATATIMEOUT) {
data->error = -ETIMEDOUT;
} else if (status & MCI_STARTBITERR) {
data->error = -ECOMM;
} else if (status & MCI_TXUNDERRUN) {
data->error = -EIO;
} else if (status & MCI_RXOVERRUN) {
if (success > host->variant->fifosize)
success -= host->variant->fifosize;
else
success = 0;
data->error = -EIO;
}
data->bytes_xfered = round_down(success, data->blksz);
}
if (status & MCI_DATABLOCKEND)
dev_err(mmc_dev(host->mmc), "stray MCI_DATABLOCKEND interrupt\n");
if (status & MCI_DATAEND || data->error) {
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
if (dma_inprogress(host))
mmci_dma_finalize(host, data);
mmci_stop_data(host);
if (!data->error)
/* The error clause is handled above, success! */
data->bytes_xfered = data->blksz * data->blocks;
if (!data->stop || host->mrq->sbc) {
mmci_request_end(host, data->mrq);
} else {
mmci_start_command(host, data->stop, 0);
}
}
}
static void
mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd,
unsigned int status)
{
void __iomem *base = host->base;
bool sbc;
if (!cmd)
return;
sbc = (cmd == host->mrq->sbc);
/*
* We need to be one of these interrupts to be considered worth
* handling. Note that we tag on any latent IRQs postponed
* due to waiting for busy status.
*/
if (!((status|host->busy_status) &
(MCI_CMDCRCFAIL|MCI_CMDTIMEOUT|MCI_CMDSENT|MCI_CMDRESPEND)))
return;
/*
* ST Micro variant: handle busy detection.
*/
if (host->variant->busy_detect) {
bool busy_resp = !!(cmd->flags & MMC_RSP_BUSY);
/* We are busy with a command, return */
if (host->busy_status &&
(status & host->variant->busy_detect_flag))
return;
/*
* We were not busy, but we now got a busy response on
* something that was not an error, and we double-check
* that the special busy status bit is still set before
* proceeding.
*/
if (!host->busy_status && busy_resp &&
!(status & (MCI_CMDCRCFAIL|MCI_CMDTIMEOUT)) &&
(readl(base + MMCISTATUS) & host->variant->busy_detect_flag)) {
mmc: mmci: avoid clearing ST Micro busy end interrupt mistakenly This fixes a race condition that may occur whenever ST micro busy end interrupt is raised just after being unmasked but before leaving mmci interrupt context. A dead-lock has been found if connecting mmci ST Micro variant whose amba id is 0x10480180 to some new eMMC that supports internal caches. Whenever mmci driver enables cache control by programming eMMC's EXT_CSD register, block driver may request to flush the eMMC internal caches causing mmci driver to send a MMC_SWITCH command to the card with FLUSH_CACHE operation. And because busy end interrupt may be mistakenly cleared while not yet processed, this mmc request may never complete. As a result, mmcqd task may be stuck forever. Here is an instance caught by lockup detector which shows that mmcqd task was hung while waiting for mmc_flush_cache command to complete: .. [ 240.251595] INFO: task mmcqd/1:52 blocked for more than 120 seconds. [ 240.257973] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.263109] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.270955] mmcqd/1 D c047504c 0 52 2 0x00000000 [ 240.277359] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.284418] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.292191] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.300491] [<c0476040>] (wait_for_common) from [<c02efc1c>] (mmc_wait_for_req_done+0x4c/0x13c) [ 240.309224] [<c02efc1c>] (mmc_wait_for_req_done) from [<c02efd90>] (mmc_wait_for_cmd+0x64/0x84) [ 240.317953] [<c02efd90>] (mmc_wait_for_cmd) from [<c02f5b14>] (__mmc_switch+0xa4/0x2a8) [ 240.325964] [<c02f5b14>] (__mmc_switch) from [<c02f5d40>] (mmc_switch+0x28/0x30) [ 240.333389] [<c02f5d40>] (mmc_switch) from [<c02f0984>] (mmc_flush_cache+0x54/0x80) [ 240.341073] [<c02f0984>] (mmc_flush_cache) from [<c02ff0c4>] (mmc_blk_issue_rq+0x114/0x4e8) [ 240.349459] [<c02ff0c4>] (mmc_blk_issue_rq) from [<c03008d4>] (mmc_queue_thread+0xc0/0x180) [ 240.357844] [<c03008d4>] (mmc_queue_thread) from [<c003cf90>] (kthread+0xdc/0xf4) [ 240.365339] [<c003cf90>] (kthread) from [<c0010068>] (ret_from_fork+0x14/0x2c) .. .. [ 240.664311] INFO: task partprobe:564 blocked for more than 120 seconds. [ 240.670943] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.676078] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.683922] partprobe D c047504c 0 564 486 0x00000000 [ 240.690318] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.697396] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.705149] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.713446] [<c0476040>] (wait_for_common) from [<c01f3300>] (submit_bio_wait+0x58/0x64) [ 240.721571] [<c01f3300>] (submit_bio_wait) from [<c01fbbd8>] (blkdev_issue_flush+0x60/0x88) [ 240.729957] [<c01fbbd8>] (blkdev_issue_flush) from [<c010ff84>] (blkdev_fsync+0x34/0x44) [ 240.738083] [<c010ff84>] (blkdev_fsync) from [<c0109594>] (do_fsync+0x3c/0x64) [ 240.745319] [<c0109594>] (do_fsync) from [<c000ffc0>] (ret_fast_syscall+0x0/0x3c) .. Here is the detailed sequence showing when this issue may happen: 1) At probe time, mmci device is initialized and card busy detection based on DAT[0] monitoring is enabled. 2) Later during run time, since card reported to support internal caches, a MMCI_SWITCH command is sent to eMMC device with FLUSH_CACHE operation. On receiving this command, eMMC may enter busy state (for a relatively short time in the case of the dead-lock). 3) Then mmci interrupt is raised and mmci_irq() is called: MMCISTATUS register is read and is equal to 0x01000440. So the following status bits are set: - MCI_CMDRESPEND (= 6) - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is 0x3FF, status variable is set to 0x00000040 and BIT MCI_CMDRESPEND is cleared by writing MMCICLEAR register. Then mmci_cmd_irq() is called. Considering the following conditions: - host->busy_status is 0, - this is a "busy response", - reading again MMCISTATUS register gives 0x1000400, MMCIMASK0 is updated to unmask MCI_ST_BUSYEND bit. Thus, MMCIMASK0 is set to 0x010003FF and host->busy_status is set to wait for busy end completion. Back again in status loop of mmci_irq(), we quickly go through mmci_data_irq() as there are no data in that case. And we finally go through following test at the end of while(status) loop: /* * Don't poll for busy completion in irq context. */ if (host->variant->busy_detect && host->busy_status) status &= ~host->variant->busy_detect_flag; Because status variable is not yet null (is equal to 0x40), we do not leave interrupt context yet but we loop again into while(status) loop. So we run across following steps: a) MMCISTATUS register is read again and this time is equal to 0x01000400. So that following bits are set: - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is equal to 0x010003FF: b) status variable is set to 0x01000000. c) MCI_ST_CARDBUSY bit is cleared by writing MMCICLEAR register. Then, mmci_cmd_irq() is called one more time. Since host->busy_status is set and that MCI_ST_CARDBUSY is set in status variable, we just return from this function. Back again in mmci_irq(), status variable is set to 0 and we finally leave the while(status) loop. As a result we leave interrupt context, waiting for busy end interrupt event. Now, consider that busy end completion is raised IN BETWEEN steps 3.a) and 3.c). In such a case, we may mistakenly clear busy end interrupt at step 3.c) while it has not yet been processed. This will result in mmc command to wait forever for a busy end completion that will never happen. To fix the problem, this patch implements the following changes: Considering that the mmci seems to be triggering the IRQ on both edges while monitoring DAT0 for busy completion and that same status bit is used to monitor start and end of busy detection, special care must be taken to make sure that both start and end interrupts are always cleared one after the other. 1) Clearing of card busy bit is moved in mmc_cmd_irq() function where unmasking of busy end bit is effectively handled. 2) Just before unmasking busy end event, busy start event is cleared by writing card busy bit in MMCICLEAR register. 3) Finally, once we are no more busy with a command, busy end event is cleared writing again card busy bit in MMCICLEAR register. This patch has been tested with the ST Accordo5 machine, not yet supported upstream but relies on the mmci driver. Signed-off-by: Sarang Mairal <sarang.mairal@garmin.com> Signed-off-by: Jean-Nicolas Graux <jean-nicolas.graux@st.com> Reviewed-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Ulf Hansson <ulf.hansson@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-02-07 18:12:41 +07:00
/* Clear the busy start IRQ */
writel(host->variant->busy_detect_mask,
host->base + MMCICLEAR);
/* Unmask the busy end IRQ */
writel(readl(base + MMCIMASK0) |
host->variant->busy_detect_mask,
base + MMCIMASK0);
/*
* Now cache the last response status code (until
* the busy bit goes low), and return.
*/
host->busy_status =
status & (MCI_CMDSENT|MCI_CMDRESPEND);
return;
}
/*
* At this point we are not busy with a command, we have
mmc: mmci: avoid clearing ST Micro busy end interrupt mistakenly This fixes a race condition that may occur whenever ST micro busy end interrupt is raised just after being unmasked but before leaving mmci interrupt context. A dead-lock has been found if connecting mmci ST Micro variant whose amba id is 0x10480180 to some new eMMC that supports internal caches. Whenever mmci driver enables cache control by programming eMMC's EXT_CSD register, block driver may request to flush the eMMC internal caches causing mmci driver to send a MMC_SWITCH command to the card with FLUSH_CACHE operation. And because busy end interrupt may be mistakenly cleared while not yet processed, this mmc request may never complete. As a result, mmcqd task may be stuck forever. Here is an instance caught by lockup detector which shows that mmcqd task was hung while waiting for mmc_flush_cache command to complete: .. [ 240.251595] INFO: task mmcqd/1:52 blocked for more than 120 seconds. [ 240.257973] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.263109] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.270955] mmcqd/1 D c047504c 0 52 2 0x00000000 [ 240.277359] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.284418] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.292191] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.300491] [<c0476040>] (wait_for_common) from [<c02efc1c>] (mmc_wait_for_req_done+0x4c/0x13c) [ 240.309224] [<c02efc1c>] (mmc_wait_for_req_done) from [<c02efd90>] (mmc_wait_for_cmd+0x64/0x84) [ 240.317953] [<c02efd90>] (mmc_wait_for_cmd) from [<c02f5b14>] (__mmc_switch+0xa4/0x2a8) [ 240.325964] [<c02f5b14>] (__mmc_switch) from [<c02f5d40>] (mmc_switch+0x28/0x30) [ 240.333389] [<c02f5d40>] (mmc_switch) from [<c02f0984>] (mmc_flush_cache+0x54/0x80) [ 240.341073] [<c02f0984>] (mmc_flush_cache) from [<c02ff0c4>] (mmc_blk_issue_rq+0x114/0x4e8) [ 240.349459] [<c02ff0c4>] (mmc_blk_issue_rq) from [<c03008d4>] (mmc_queue_thread+0xc0/0x180) [ 240.357844] [<c03008d4>] (mmc_queue_thread) from [<c003cf90>] (kthread+0xdc/0xf4) [ 240.365339] [<c003cf90>] (kthread) from [<c0010068>] (ret_from_fork+0x14/0x2c) .. .. [ 240.664311] INFO: task partprobe:564 blocked for more than 120 seconds. [ 240.670943] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.676078] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.683922] partprobe D c047504c 0 564 486 0x00000000 [ 240.690318] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.697396] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.705149] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.713446] [<c0476040>] (wait_for_common) from [<c01f3300>] (submit_bio_wait+0x58/0x64) [ 240.721571] [<c01f3300>] (submit_bio_wait) from [<c01fbbd8>] (blkdev_issue_flush+0x60/0x88) [ 240.729957] [<c01fbbd8>] (blkdev_issue_flush) from [<c010ff84>] (blkdev_fsync+0x34/0x44) [ 240.738083] [<c010ff84>] (blkdev_fsync) from [<c0109594>] (do_fsync+0x3c/0x64) [ 240.745319] [<c0109594>] (do_fsync) from [<c000ffc0>] (ret_fast_syscall+0x0/0x3c) .. Here is the detailed sequence showing when this issue may happen: 1) At probe time, mmci device is initialized and card busy detection based on DAT[0] monitoring is enabled. 2) Later during run time, since card reported to support internal caches, a MMCI_SWITCH command is sent to eMMC device with FLUSH_CACHE operation. On receiving this command, eMMC may enter busy state (for a relatively short time in the case of the dead-lock). 3) Then mmci interrupt is raised and mmci_irq() is called: MMCISTATUS register is read and is equal to 0x01000440. So the following status bits are set: - MCI_CMDRESPEND (= 6) - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is 0x3FF, status variable is set to 0x00000040 and BIT MCI_CMDRESPEND is cleared by writing MMCICLEAR register. Then mmci_cmd_irq() is called. Considering the following conditions: - host->busy_status is 0, - this is a "busy response", - reading again MMCISTATUS register gives 0x1000400, MMCIMASK0 is updated to unmask MCI_ST_BUSYEND bit. Thus, MMCIMASK0 is set to 0x010003FF and host->busy_status is set to wait for busy end completion. Back again in status loop of mmci_irq(), we quickly go through mmci_data_irq() as there are no data in that case. And we finally go through following test at the end of while(status) loop: /* * Don't poll for busy completion in irq context. */ if (host->variant->busy_detect && host->busy_status) status &= ~host->variant->busy_detect_flag; Because status variable is not yet null (is equal to 0x40), we do not leave interrupt context yet but we loop again into while(status) loop. So we run across following steps: a) MMCISTATUS register is read again and this time is equal to 0x01000400. So that following bits are set: - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is equal to 0x010003FF: b) status variable is set to 0x01000000. c) MCI_ST_CARDBUSY bit is cleared by writing MMCICLEAR register. Then, mmci_cmd_irq() is called one more time. Since host->busy_status is set and that MCI_ST_CARDBUSY is set in status variable, we just return from this function. Back again in mmci_irq(), status variable is set to 0 and we finally leave the while(status) loop. As a result we leave interrupt context, waiting for busy end interrupt event. Now, consider that busy end completion is raised IN BETWEEN steps 3.a) and 3.c). In such a case, we may mistakenly clear busy end interrupt at step 3.c) while it has not yet been processed. This will result in mmc command to wait forever for a busy end completion that will never happen. To fix the problem, this patch implements the following changes: Considering that the mmci seems to be triggering the IRQ on both edges while monitoring DAT0 for busy completion and that same status bit is used to monitor start and end of busy detection, special care must be taken to make sure that both start and end interrupts are always cleared one after the other. 1) Clearing of card busy bit is moved in mmc_cmd_irq() function where unmasking of busy end bit is effectively handled. 2) Just before unmasking busy end event, busy start event is cleared by writing card busy bit in MMCICLEAR register. 3) Finally, once we are no more busy with a command, busy end event is cleared writing again card busy bit in MMCICLEAR register. This patch has been tested with the ST Accordo5 machine, not yet supported upstream but relies on the mmci driver. Signed-off-by: Sarang Mairal <sarang.mairal@garmin.com> Signed-off-by: Jean-Nicolas Graux <jean-nicolas.graux@st.com> Reviewed-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Ulf Hansson <ulf.hansson@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-02-07 18:12:41 +07:00
* not received a new busy request, clear and mask the busy
* end IRQ and fall through to process the IRQ.
*/
if (host->busy_status) {
mmc: mmci: avoid clearing ST Micro busy end interrupt mistakenly This fixes a race condition that may occur whenever ST micro busy end interrupt is raised just after being unmasked but before leaving mmci interrupt context. A dead-lock has been found if connecting mmci ST Micro variant whose amba id is 0x10480180 to some new eMMC that supports internal caches. Whenever mmci driver enables cache control by programming eMMC's EXT_CSD register, block driver may request to flush the eMMC internal caches causing mmci driver to send a MMC_SWITCH command to the card with FLUSH_CACHE operation. And because busy end interrupt may be mistakenly cleared while not yet processed, this mmc request may never complete. As a result, mmcqd task may be stuck forever. Here is an instance caught by lockup detector which shows that mmcqd task was hung while waiting for mmc_flush_cache command to complete: .. [ 240.251595] INFO: task mmcqd/1:52 blocked for more than 120 seconds. [ 240.257973] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.263109] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.270955] mmcqd/1 D c047504c 0 52 2 0x00000000 [ 240.277359] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.284418] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.292191] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.300491] [<c0476040>] (wait_for_common) from [<c02efc1c>] (mmc_wait_for_req_done+0x4c/0x13c) [ 240.309224] [<c02efc1c>] (mmc_wait_for_req_done) from [<c02efd90>] (mmc_wait_for_cmd+0x64/0x84) [ 240.317953] [<c02efd90>] (mmc_wait_for_cmd) from [<c02f5b14>] (__mmc_switch+0xa4/0x2a8) [ 240.325964] [<c02f5b14>] (__mmc_switch) from [<c02f5d40>] (mmc_switch+0x28/0x30) [ 240.333389] [<c02f5d40>] (mmc_switch) from [<c02f0984>] (mmc_flush_cache+0x54/0x80) [ 240.341073] [<c02f0984>] (mmc_flush_cache) from [<c02ff0c4>] (mmc_blk_issue_rq+0x114/0x4e8) [ 240.349459] [<c02ff0c4>] (mmc_blk_issue_rq) from [<c03008d4>] (mmc_queue_thread+0xc0/0x180) [ 240.357844] [<c03008d4>] (mmc_queue_thread) from [<c003cf90>] (kthread+0xdc/0xf4) [ 240.365339] [<c003cf90>] (kthread) from [<c0010068>] (ret_from_fork+0x14/0x2c) .. .. [ 240.664311] INFO: task partprobe:564 blocked for more than 120 seconds. [ 240.670943] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.676078] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.683922] partprobe D c047504c 0 564 486 0x00000000 [ 240.690318] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.697396] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.705149] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.713446] [<c0476040>] (wait_for_common) from [<c01f3300>] (submit_bio_wait+0x58/0x64) [ 240.721571] [<c01f3300>] (submit_bio_wait) from [<c01fbbd8>] (blkdev_issue_flush+0x60/0x88) [ 240.729957] [<c01fbbd8>] (blkdev_issue_flush) from [<c010ff84>] (blkdev_fsync+0x34/0x44) [ 240.738083] [<c010ff84>] (blkdev_fsync) from [<c0109594>] (do_fsync+0x3c/0x64) [ 240.745319] [<c0109594>] (do_fsync) from [<c000ffc0>] (ret_fast_syscall+0x0/0x3c) .. Here is the detailed sequence showing when this issue may happen: 1) At probe time, mmci device is initialized and card busy detection based on DAT[0] monitoring is enabled. 2) Later during run time, since card reported to support internal caches, a MMCI_SWITCH command is sent to eMMC device with FLUSH_CACHE operation. On receiving this command, eMMC may enter busy state (for a relatively short time in the case of the dead-lock). 3) Then mmci interrupt is raised and mmci_irq() is called: MMCISTATUS register is read and is equal to 0x01000440. So the following status bits are set: - MCI_CMDRESPEND (= 6) - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is 0x3FF, status variable is set to 0x00000040 and BIT MCI_CMDRESPEND is cleared by writing MMCICLEAR register. Then mmci_cmd_irq() is called. Considering the following conditions: - host->busy_status is 0, - this is a "busy response", - reading again MMCISTATUS register gives 0x1000400, MMCIMASK0 is updated to unmask MCI_ST_BUSYEND bit. Thus, MMCIMASK0 is set to 0x010003FF and host->busy_status is set to wait for busy end completion. Back again in status loop of mmci_irq(), we quickly go through mmci_data_irq() as there are no data in that case. And we finally go through following test at the end of while(status) loop: /* * Don't poll for busy completion in irq context. */ if (host->variant->busy_detect && host->busy_status) status &= ~host->variant->busy_detect_flag; Because status variable is not yet null (is equal to 0x40), we do not leave interrupt context yet but we loop again into while(status) loop. So we run across following steps: a) MMCISTATUS register is read again and this time is equal to 0x01000400. So that following bits are set: - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is equal to 0x010003FF: b) status variable is set to 0x01000000. c) MCI_ST_CARDBUSY bit is cleared by writing MMCICLEAR register. Then, mmci_cmd_irq() is called one more time. Since host->busy_status is set and that MCI_ST_CARDBUSY is set in status variable, we just return from this function. Back again in mmci_irq(), status variable is set to 0 and we finally leave the while(status) loop. As a result we leave interrupt context, waiting for busy end interrupt event. Now, consider that busy end completion is raised IN BETWEEN steps 3.a) and 3.c). In such a case, we may mistakenly clear busy end interrupt at step 3.c) while it has not yet been processed. This will result in mmc command to wait forever for a busy end completion that will never happen. To fix the problem, this patch implements the following changes: Considering that the mmci seems to be triggering the IRQ on both edges while monitoring DAT0 for busy completion and that same status bit is used to monitor start and end of busy detection, special care must be taken to make sure that both start and end interrupts are always cleared one after the other. 1) Clearing of card busy bit is moved in mmc_cmd_irq() function where unmasking of busy end bit is effectively handled. 2) Just before unmasking busy end event, busy start event is cleared by writing card busy bit in MMCICLEAR register. 3) Finally, once we are no more busy with a command, busy end event is cleared writing again card busy bit in MMCICLEAR register. This patch has been tested with the ST Accordo5 machine, not yet supported upstream but relies on the mmci driver. Signed-off-by: Sarang Mairal <sarang.mairal@garmin.com> Signed-off-by: Jean-Nicolas Graux <jean-nicolas.graux@st.com> Reviewed-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Ulf Hansson <ulf.hansson@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-02-07 18:12:41 +07:00
writel(host->variant->busy_detect_mask,
host->base + MMCICLEAR);
writel(readl(base + MMCIMASK0) &
~host->variant->busy_detect_mask,
base + MMCIMASK0);
host->busy_status = 0;
}
}
host->cmd = NULL;
if (status & MCI_CMDTIMEOUT) {
cmd->error = -ETIMEDOUT;
} else if (status & MCI_CMDCRCFAIL && cmd->flags & MMC_RSP_CRC) {
cmd->error = -EILSEQ;
} else {
cmd->resp[0] = readl(base + MMCIRESPONSE0);
cmd->resp[1] = readl(base + MMCIRESPONSE1);
cmd->resp[2] = readl(base + MMCIRESPONSE2);
cmd->resp[3] = readl(base + MMCIRESPONSE3);
}
if ((!sbc && !cmd->data) || cmd->error) {
if (host->data) {
/* Terminate the DMA transfer */
if (dma_inprogress(host))
mmci_dma_data_error(host);
mmci_stop_data(host);
}
mmci_request_end(host, host->mrq);
} else if (sbc) {
mmci_start_command(host, host->mrq->cmd, 0);
} else if (!(cmd->data->flags & MMC_DATA_READ)) {
mmci_start_data(host, cmd->data);
}
}
static int mmci_get_rx_fifocnt(struct mmci_host *host, u32 status, int remain)
{
return remain - (readl(host->base + MMCIFIFOCNT) << 2);
}
static int mmci_qcom_get_rx_fifocnt(struct mmci_host *host, u32 status, int r)
{
/*
* on qcom SDCC4 only 8 words are used in each burst so only 8 addresses
* from the fifo range should be used
*/
if (status & MCI_RXFIFOHALFFULL)
return host->variant->fifohalfsize;
else if (status & MCI_RXDATAAVLBL)
return 4;
return 0;
}
static int mmci_pio_read(struct mmci_host *host, char *buffer, unsigned int remain)
{
void __iomem *base = host->base;
char *ptr = buffer;
u32 status = readl(host->base + MMCISTATUS);
int host_remain = host->size;
do {
int count = host->get_rx_fifocnt(host, status, host_remain);
if (count > remain)
count = remain;
if (count <= 0)
break;
/*
* SDIO especially may want to send something that is
* not divisible by 4 (as opposed to card sectors
* etc). Therefore make sure to always read the last bytes
* while only doing full 32-bit reads towards the FIFO.
*/
if (unlikely(count & 0x3)) {
if (count < 4) {
unsigned char buf[4];
ioread32_rep(base + MMCIFIFO, buf, 1);
memcpy(ptr, buf, count);
} else {
ioread32_rep(base + MMCIFIFO, ptr, count >> 2);
count &= ~0x3;
}
} else {
ioread32_rep(base + MMCIFIFO, ptr, count >> 2);
}
ptr += count;
remain -= count;
host_remain -= count;
if (remain == 0)
break;
status = readl(base + MMCISTATUS);
} while (status & MCI_RXDATAAVLBL);
return ptr - buffer;
}
static int mmci_pio_write(struct mmci_host *host, char *buffer, unsigned int remain, u32 status)
{
struct variant_data *variant = host->variant;
void __iomem *base = host->base;
char *ptr = buffer;
do {
unsigned int count, maxcnt;
maxcnt = status & MCI_TXFIFOEMPTY ?
variant->fifosize : variant->fifohalfsize;
count = min(remain, maxcnt);
/*
* SDIO especially may want to send something that is
* not divisible by 4 (as opposed to card sectors
* etc), and the FIFO only accept full 32-bit writes.
* So compensate by adding +3 on the count, a single
* byte become a 32bit write, 7 bytes will be two
* 32bit writes etc.
*/
iowrite32_rep(base + MMCIFIFO, ptr, (count + 3) >> 2);
ptr += count;
remain -= count;
if (remain == 0)
break;
status = readl(base + MMCISTATUS);
} while (status & MCI_TXFIFOHALFEMPTY);
return ptr - buffer;
}
/*
* PIO data transfer IRQ handler.
*/
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 20:55:46 +07:00
static irqreturn_t mmci_pio_irq(int irq, void *dev_id)
{
struct mmci_host *host = dev_id;
struct sg_mapping_iter *sg_miter = &host->sg_miter;
struct variant_data *variant = host->variant;
void __iomem *base = host->base;
u32 status;
status = readl(base + MMCISTATUS);
dev_dbg(mmc_dev(host->mmc), "irq1 (pio) %08x\n", status);
do {
unsigned int remain, len;
char *buffer;
/*
* For write, we only need to test the half-empty flag
* here - if the FIFO is completely empty, then by
* definition it is more than half empty.
*
* For read, check for data available.
*/
if (!(status & (MCI_TXFIFOHALFEMPTY|MCI_RXDATAAVLBL)))
break;
if (!sg_miter_next(sg_miter))
break;
buffer = sg_miter->addr;
remain = sg_miter->length;
len = 0;
if (status & MCI_RXACTIVE)
len = mmci_pio_read(host, buffer, remain);
if (status & MCI_TXACTIVE)
len = mmci_pio_write(host, buffer, remain, status);
sg_miter->consumed = len;
host->size -= len;
remain -= len;
if (remain)
break;
status = readl(base + MMCISTATUS);
} while (1);
sg_miter_stop(sg_miter);
/*
* If we have less than the fifo 'half-full' threshold to transfer,
* trigger a PIO interrupt as soon as any data is available.
*/
if (status & MCI_RXACTIVE && host->size < variant->fifohalfsize)
mmci_set_mask1(host, MCI_RXDATAAVLBLMASK);
/*
* If we run out of data, disable the data IRQs; this
* prevents a race where the FIFO becomes empty before
* the chip itself has disabled the data path, and
* stops us racing with our data end IRQ.
*/
if (host->size == 0) {
mmci_set_mask1(host, 0);
writel(readl(base + MMCIMASK0) | MCI_DATAENDMASK, base + MMCIMASK0);
}
return IRQ_HANDLED;
}
/*
* Handle completion of command and data transfers.
*/
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 20:55:46 +07:00
static irqreturn_t mmci_irq(int irq, void *dev_id)
{
struct mmci_host *host = dev_id;
u32 status;
int ret = 0;
spin_lock(&host->lock);
do {
status = readl(host->base + MMCISTATUS);
if (host->singleirq) {
if (status & host->mask1_reg)
mmci_pio_irq(irq, dev_id);
status &= ~MCI_IRQ1MASK;
}
/*
mmc: mmci: avoid clearing ST Micro busy end interrupt mistakenly This fixes a race condition that may occur whenever ST micro busy end interrupt is raised just after being unmasked but before leaving mmci interrupt context. A dead-lock has been found if connecting mmci ST Micro variant whose amba id is 0x10480180 to some new eMMC that supports internal caches. Whenever mmci driver enables cache control by programming eMMC's EXT_CSD register, block driver may request to flush the eMMC internal caches causing mmci driver to send a MMC_SWITCH command to the card with FLUSH_CACHE operation. And because busy end interrupt may be mistakenly cleared while not yet processed, this mmc request may never complete. As a result, mmcqd task may be stuck forever. Here is an instance caught by lockup detector which shows that mmcqd task was hung while waiting for mmc_flush_cache command to complete: .. [ 240.251595] INFO: task mmcqd/1:52 blocked for more than 120 seconds. [ 240.257973] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.263109] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.270955] mmcqd/1 D c047504c 0 52 2 0x00000000 [ 240.277359] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.284418] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.292191] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.300491] [<c0476040>] (wait_for_common) from [<c02efc1c>] (mmc_wait_for_req_done+0x4c/0x13c) [ 240.309224] [<c02efc1c>] (mmc_wait_for_req_done) from [<c02efd90>] (mmc_wait_for_cmd+0x64/0x84) [ 240.317953] [<c02efd90>] (mmc_wait_for_cmd) from [<c02f5b14>] (__mmc_switch+0xa4/0x2a8) [ 240.325964] [<c02f5b14>] (__mmc_switch) from [<c02f5d40>] (mmc_switch+0x28/0x30) [ 240.333389] [<c02f5d40>] (mmc_switch) from [<c02f0984>] (mmc_flush_cache+0x54/0x80) [ 240.341073] [<c02f0984>] (mmc_flush_cache) from [<c02ff0c4>] (mmc_blk_issue_rq+0x114/0x4e8) [ 240.349459] [<c02ff0c4>] (mmc_blk_issue_rq) from [<c03008d4>] (mmc_queue_thread+0xc0/0x180) [ 240.357844] [<c03008d4>] (mmc_queue_thread) from [<c003cf90>] (kthread+0xdc/0xf4) [ 240.365339] [<c003cf90>] (kthread) from [<c0010068>] (ret_from_fork+0x14/0x2c) .. .. [ 240.664311] INFO: task partprobe:564 blocked for more than 120 seconds. [ 240.670943] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.676078] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.683922] partprobe D c047504c 0 564 486 0x00000000 [ 240.690318] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.697396] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.705149] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.713446] [<c0476040>] (wait_for_common) from [<c01f3300>] (submit_bio_wait+0x58/0x64) [ 240.721571] [<c01f3300>] (submit_bio_wait) from [<c01fbbd8>] (blkdev_issue_flush+0x60/0x88) [ 240.729957] [<c01fbbd8>] (blkdev_issue_flush) from [<c010ff84>] (blkdev_fsync+0x34/0x44) [ 240.738083] [<c010ff84>] (blkdev_fsync) from [<c0109594>] (do_fsync+0x3c/0x64) [ 240.745319] [<c0109594>] (do_fsync) from [<c000ffc0>] (ret_fast_syscall+0x0/0x3c) .. Here is the detailed sequence showing when this issue may happen: 1) At probe time, mmci device is initialized and card busy detection based on DAT[0] monitoring is enabled. 2) Later during run time, since card reported to support internal caches, a MMCI_SWITCH command is sent to eMMC device with FLUSH_CACHE operation. On receiving this command, eMMC may enter busy state (for a relatively short time in the case of the dead-lock). 3) Then mmci interrupt is raised and mmci_irq() is called: MMCISTATUS register is read and is equal to 0x01000440. So the following status bits are set: - MCI_CMDRESPEND (= 6) - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is 0x3FF, status variable is set to 0x00000040 and BIT MCI_CMDRESPEND is cleared by writing MMCICLEAR register. Then mmci_cmd_irq() is called. Considering the following conditions: - host->busy_status is 0, - this is a "busy response", - reading again MMCISTATUS register gives 0x1000400, MMCIMASK0 is updated to unmask MCI_ST_BUSYEND bit. Thus, MMCIMASK0 is set to 0x010003FF and host->busy_status is set to wait for busy end completion. Back again in status loop of mmci_irq(), we quickly go through mmci_data_irq() as there are no data in that case. And we finally go through following test at the end of while(status) loop: /* * Don't poll for busy completion in irq context. */ if (host->variant->busy_detect && host->busy_status) status &= ~host->variant->busy_detect_flag; Because status variable is not yet null (is equal to 0x40), we do not leave interrupt context yet but we loop again into while(status) loop. So we run across following steps: a) MMCISTATUS register is read again and this time is equal to 0x01000400. So that following bits are set: - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is equal to 0x010003FF: b) status variable is set to 0x01000000. c) MCI_ST_CARDBUSY bit is cleared by writing MMCICLEAR register. Then, mmci_cmd_irq() is called one more time. Since host->busy_status is set and that MCI_ST_CARDBUSY is set in status variable, we just return from this function. Back again in mmci_irq(), status variable is set to 0 and we finally leave the while(status) loop. As a result we leave interrupt context, waiting for busy end interrupt event. Now, consider that busy end completion is raised IN BETWEEN steps 3.a) and 3.c). In such a case, we may mistakenly clear busy end interrupt at step 3.c) while it has not yet been processed. This will result in mmc command to wait forever for a busy end completion that will never happen. To fix the problem, this patch implements the following changes: Considering that the mmci seems to be triggering the IRQ on both edges while monitoring DAT0 for busy completion and that same status bit is used to monitor start and end of busy detection, special care must be taken to make sure that both start and end interrupts are always cleared one after the other. 1) Clearing of card busy bit is moved in mmc_cmd_irq() function where unmasking of busy end bit is effectively handled. 2) Just before unmasking busy end event, busy start event is cleared by writing card busy bit in MMCICLEAR register. 3) Finally, once we are no more busy with a command, busy end event is cleared writing again card busy bit in MMCICLEAR register. This patch has been tested with the ST Accordo5 machine, not yet supported upstream but relies on the mmci driver. Signed-off-by: Sarang Mairal <sarang.mairal@garmin.com> Signed-off-by: Jean-Nicolas Graux <jean-nicolas.graux@st.com> Reviewed-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Ulf Hansson <ulf.hansson@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-02-07 18:12:41 +07:00
* We intentionally clear the MCI_ST_CARDBUSY IRQ (if it's
* enabled) in mmci_cmd_irq() function where ST Micro busy
* detection variant is handled. Considering the HW seems to be
* triggering the IRQ on both edges while monitoring DAT0 for
* busy completion and that same status bit is used to monitor
* start and end of busy detection, special care must be taken
* to make sure that both start and end interrupts are always
* cleared one after the other.
*/
status &= readl(host->base + MMCIMASK0);
mmc: mmci: avoid clearing ST Micro busy end interrupt mistakenly This fixes a race condition that may occur whenever ST micro busy end interrupt is raised just after being unmasked but before leaving mmci interrupt context. A dead-lock has been found if connecting mmci ST Micro variant whose amba id is 0x10480180 to some new eMMC that supports internal caches. Whenever mmci driver enables cache control by programming eMMC's EXT_CSD register, block driver may request to flush the eMMC internal caches causing mmci driver to send a MMC_SWITCH command to the card with FLUSH_CACHE operation. And because busy end interrupt may be mistakenly cleared while not yet processed, this mmc request may never complete. As a result, mmcqd task may be stuck forever. Here is an instance caught by lockup detector which shows that mmcqd task was hung while waiting for mmc_flush_cache command to complete: .. [ 240.251595] INFO: task mmcqd/1:52 blocked for more than 120 seconds. [ 240.257973] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.263109] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.270955] mmcqd/1 D c047504c 0 52 2 0x00000000 [ 240.277359] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.284418] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.292191] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.300491] [<c0476040>] (wait_for_common) from [<c02efc1c>] (mmc_wait_for_req_done+0x4c/0x13c) [ 240.309224] [<c02efc1c>] (mmc_wait_for_req_done) from [<c02efd90>] (mmc_wait_for_cmd+0x64/0x84) [ 240.317953] [<c02efd90>] (mmc_wait_for_cmd) from [<c02f5b14>] (__mmc_switch+0xa4/0x2a8) [ 240.325964] [<c02f5b14>] (__mmc_switch) from [<c02f5d40>] (mmc_switch+0x28/0x30) [ 240.333389] [<c02f5d40>] (mmc_switch) from [<c02f0984>] (mmc_flush_cache+0x54/0x80) [ 240.341073] [<c02f0984>] (mmc_flush_cache) from [<c02ff0c4>] (mmc_blk_issue_rq+0x114/0x4e8) [ 240.349459] [<c02ff0c4>] (mmc_blk_issue_rq) from [<c03008d4>] (mmc_queue_thread+0xc0/0x180) [ 240.357844] [<c03008d4>] (mmc_queue_thread) from [<c003cf90>] (kthread+0xdc/0xf4) [ 240.365339] [<c003cf90>] (kthread) from [<c0010068>] (ret_from_fork+0x14/0x2c) .. .. [ 240.664311] INFO: task partprobe:564 blocked for more than 120 seconds. [ 240.670943] Not tainted 4.1.13-00510-g9d91424 #2 [ 240.676078] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 240.683922] partprobe D c047504c 0 564 486 0x00000000 [ 240.690318] [<c047504c>] (__schedule) from [<c04754a0>] (schedule+0x40/0x98) [ 240.697396] [<c04754a0>] (schedule) from [<c0477d40>] (schedule_timeout+0x148/0x188) [ 240.705149] [<c0477d40>] (schedule_timeout) from [<c0476040>] (wait_for_common+0xa4/0x170) [ 240.713446] [<c0476040>] (wait_for_common) from [<c01f3300>] (submit_bio_wait+0x58/0x64) [ 240.721571] [<c01f3300>] (submit_bio_wait) from [<c01fbbd8>] (blkdev_issue_flush+0x60/0x88) [ 240.729957] [<c01fbbd8>] (blkdev_issue_flush) from [<c010ff84>] (blkdev_fsync+0x34/0x44) [ 240.738083] [<c010ff84>] (blkdev_fsync) from [<c0109594>] (do_fsync+0x3c/0x64) [ 240.745319] [<c0109594>] (do_fsync) from [<c000ffc0>] (ret_fast_syscall+0x0/0x3c) .. Here is the detailed sequence showing when this issue may happen: 1) At probe time, mmci device is initialized and card busy detection based on DAT[0] monitoring is enabled. 2) Later during run time, since card reported to support internal caches, a MMCI_SWITCH command is sent to eMMC device with FLUSH_CACHE operation. On receiving this command, eMMC may enter busy state (for a relatively short time in the case of the dead-lock). 3) Then mmci interrupt is raised and mmci_irq() is called: MMCISTATUS register is read and is equal to 0x01000440. So the following status bits are set: - MCI_CMDRESPEND (= 6) - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is 0x3FF, status variable is set to 0x00000040 and BIT MCI_CMDRESPEND is cleared by writing MMCICLEAR register. Then mmci_cmd_irq() is called. Considering the following conditions: - host->busy_status is 0, - this is a "busy response", - reading again MMCISTATUS register gives 0x1000400, MMCIMASK0 is updated to unmask MCI_ST_BUSYEND bit. Thus, MMCIMASK0 is set to 0x010003FF and host->busy_status is set to wait for busy end completion. Back again in status loop of mmci_irq(), we quickly go through mmci_data_irq() as there are no data in that case. And we finally go through following test at the end of while(status) loop: /* * Don't poll for busy completion in irq context. */ if (host->variant->busy_detect && host->busy_status) status &= ~host->variant->busy_detect_flag; Because status variable is not yet null (is equal to 0x40), we do not leave interrupt context yet but we loop again into while(status) loop. So we run across following steps: a) MMCISTATUS register is read again and this time is equal to 0x01000400. So that following bits are set: - MCI_DATABLOCKEND (= 10) - MCI_ST_CARDBUSY (= 24) Since MMCIMASK0 register is equal to 0x010003FF: b) status variable is set to 0x01000000. c) MCI_ST_CARDBUSY bit is cleared by writing MMCICLEAR register. Then, mmci_cmd_irq() is called one more time. Since host->busy_status is set and that MCI_ST_CARDBUSY is set in status variable, we just return from this function. Back again in mmci_irq(), status variable is set to 0 and we finally leave the while(status) loop. As a result we leave interrupt context, waiting for busy end interrupt event. Now, consider that busy end completion is raised IN BETWEEN steps 3.a) and 3.c). In such a case, we may mistakenly clear busy end interrupt at step 3.c) while it has not yet been processed. This will result in mmc command to wait forever for a busy end completion that will never happen. To fix the problem, this patch implements the following changes: Considering that the mmci seems to be triggering the IRQ on both edges while monitoring DAT0 for busy completion and that same status bit is used to monitor start and end of busy detection, special care must be taken to make sure that both start and end interrupts are always cleared one after the other. 1) Clearing of card busy bit is moved in mmc_cmd_irq() function where unmasking of busy end bit is effectively handled. 2) Just before unmasking busy end event, busy start event is cleared by writing card busy bit in MMCICLEAR register. 3) Finally, once we are no more busy with a command, busy end event is cleared writing again card busy bit in MMCICLEAR register. This patch has been tested with the ST Accordo5 machine, not yet supported upstream but relies on the mmci driver. Signed-off-by: Sarang Mairal <sarang.mairal@garmin.com> Signed-off-by: Jean-Nicolas Graux <jean-nicolas.graux@st.com> Reviewed-by: Linus Walleij <linus.walleij@linaro.org> Tested-by: Ulf Hansson <ulf.hansson@linaro.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2017-02-07 18:12:41 +07:00
if (host->variant->busy_detect)
writel(status & ~host->variant->busy_detect_mask,
host->base + MMCICLEAR);
else
writel(status, host->base + MMCICLEAR);
dev_dbg(mmc_dev(host->mmc), "irq0 (data+cmd) %08x\n", status);
if (host->variant->reversed_irq_handling) {
mmci_data_irq(host, host->data, status);
mmci_cmd_irq(host, host->cmd, status);
} else {
mmci_cmd_irq(host, host->cmd, status);
mmci_data_irq(host, host->data, status);
}
/*
* Don't poll for busy completion in irq context.
*/
if (host->variant->busy_detect && host->busy_status)
status &= ~host->variant->busy_detect_flag;
ret = 1;
} while (status);
spin_unlock(&host->lock);
return IRQ_RETVAL(ret);
}
static void mmci_request(struct mmc_host *mmc, struct mmc_request *mrq)
{
struct mmci_host *host = mmc_priv(mmc);
unsigned long flags;
WARN_ON(host->mrq != NULL);
mrq->cmd->error = mmci_validate_data(host, mrq->data);
if (mrq->cmd->error) {
mmc_request_done(mmc, mrq);
return;
}
spin_lock_irqsave(&host->lock, flags);
host->mrq = mrq;
if (mrq->data)
mmci_get_next_data(host, mrq->data);
if (mrq->data && mrq->data->flags & MMC_DATA_READ)
mmci_start_data(host, mrq->data);
if (mrq->sbc)
mmci_start_command(host, mrq->sbc, 0);
else
mmci_start_command(host, mrq->cmd, 0);
spin_unlock_irqrestore(&host->lock, flags);
}
static void mmci_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
{
struct mmci_host *host = mmc_priv(mmc);
struct variant_data *variant = host->variant;
u32 pwr = 0;
unsigned long flags;
int ret;
if (host->plat->ios_handler &&
host->plat->ios_handler(mmc_dev(mmc), ios))
dev_err(mmc_dev(mmc), "platform ios_handler failed\n");
switch (ios->power_mode) {
case MMC_POWER_OFF:
if (!IS_ERR(mmc->supply.vmmc))
mmc_regulator_set_ocr(mmc, mmc->supply.vmmc, 0);
if (!IS_ERR(mmc->supply.vqmmc) && host->vqmmc_enabled) {
regulator_disable(mmc->supply.vqmmc);
host->vqmmc_enabled = false;
}
break;
case MMC_POWER_UP:
if (!IS_ERR(mmc->supply.vmmc))
mmc_regulator_set_ocr(mmc, mmc->supply.vmmc, ios->vdd);
/*
* The ST Micro variant doesn't have the PL180s MCI_PWR_UP
* and instead uses MCI_PWR_ON so apply whatever value is
* configured in the variant data.
*/
pwr |= variant->pwrreg_powerup;
break;
case MMC_POWER_ON:
if (!IS_ERR(mmc->supply.vqmmc) && !host->vqmmc_enabled) {
ret = regulator_enable(mmc->supply.vqmmc);
if (ret < 0)
dev_err(mmc_dev(mmc),
"failed to enable vqmmc regulator\n");
else
host->vqmmc_enabled = true;
}
pwr |= MCI_PWR_ON;
break;
}
if (variant->signal_direction && ios->power_mode != MMC_POWER_OFF) {
/*
* The ST Micro variant has some additional bits
* indicating signal direction for the signals in
* the SD/MMC bus and feedback-clock usage.
*/
pwr |= host->pwr_reg_add;
if (ios->bus_width == MMC_BUS_WIDTH_4)
pwr &= ~MCI_ST_DATA74DIREN;
else if (ios->bus_width == MMC_BUS_WIDTH_1)
pwr &= (~MCI_ST_DATA74DIREN &
~MCI_ST_DATA31DIREN &
~MCI_ST_DATA2DIREN);
}
if (variant->opendrain) {
if (ios->bus_mode == MMC_BUSMODE_OPENDRAIN)
pwr |= variant->opendrain;
} else {
/*
* If the variant cannot configure the pads by its own, then we
* expect the pinctrl to be able to do that for us
*/
if (ios->bus_mode == MMC_BUSMODE_OPENDRAIN)
pinctrl_select_state(host->pinctrl, host->pins_opendrain);
else
pinctrl_select_state(host->pinctrl, host->pins_default);
}
/*
* If clock = 0 and the variant requires the MMCIPOWER to be used for
* gating the clock, the MCI_PWR_ON bit is cleared.
*/
if (!ios->clock && variant->pwrreg_clkgate)
pwr &= ~MCI_PWR_ON;
if (host->variant->explicit_mclk_control &&
ios->clock != host->clock_cache) {
ret = clk_set_rate(host->clk, ios->clock);
if (ret < 0)
dev_err(mmc_dev(host->mmc),
"Error setting clock rate (%d)\n", ret);
else
host->mclk = clk_get_rate(host->clk);
}
host->clock_cache = ios->clock;
spin_lock_irqsave(&host->lock, flags);
mmci_set_clkreg(host, ios->clock);
mmci_write_pwrreg(host, pwr);
mmci_reg_delay(host);
spin_unlock_irqrestore(&host->lock, flags);
}
static int mmci_get_cd(struct mmc_host *mmc)
{
struct mmci_host *host = mmc_priv(mmc);
struct mmci_platform_data *plat = host->plat;
unsigned int status = mmc_gpio_get_cd(mmc);
if (status == -ENOSYS) {
if (!plat->status)
return 1; /* Assume always present */
status = plat->status(mmc_dev(host->mmc));
}
return status;
}
static int mmci_sig_volt_switch(struct mmc_host *mmc, struct mmc_ios *ios)
{
int ret = 0;
if (!IS_ERR(mmc->supply.vqmmc)) {
switch (ios->signal_voltage) {
case MMC_SIGNAL_VOLTAGE_330:
ret = regulator_set_voltage(mmc->supply.vqmmc,
2700000, 3600000);
break;
case MMC_SIGNAL_VOLTAGE_180:
ret = regulator_set_voltage(mmc->supply.vqmmc,
1700000, 1950000);
break;
case MMC_SIGNAL_VOLTAGE_120:
ret = regulator_set_voltage(mmc->supply.vqmmc,
1100000, 1300000);
break;
}
if (ret)
dev_warn(mmc_dev(mmc), "Voltage switch failed\n");
}
return ret;
}
static struct mmc_host_ops mmci_ops = {
.request = mmci_request,
.pre_req = mmci_pre_request,
.post_req = mmci_post_request,
.set_ios = mmci_set_ios,
.get_ro = mmc_gpio_get_ro,
.get_cd = mmci_get_cd,
.start_signal_voltage_switch = mmci_sig_volt_switch,
};
static int mmci_of_parse(struct device_node *np, struct mmc_host *mmc)
{
struct mmci_host *host = mmc_priv(mmc);
int ret = mmc_of_parse(mmc);
if (ret)
return ret;
if (of_get_property(np, "st,sig-dir-dat0", NULL))
host->pwr_reg_add |= MCI_ST_DATA0DIREN;
if (of_get_property(np, "st,sig-dir-dat2", NULL))
host->pwr_reg_add |= MCI_ST_DATA2DIREN;
if (of_get_property(np, "st,sig-dir-dat31", NULL))
host->pwr_reg_add |= MCI_ST_DATA31DIREN;
if (of_get_property(np, "st,sig-dir-dat74", NULL))
host->pwr_reg_add |= MCI_ST_DATA74DIREN;
if (of_get_property(np, "st,sig-dir-cmd", NULL))
host->pwr_reg_add |= MCI_ST_CMDDIREN;
if (of_get_property(np, "st,sig-pin-fbclk", NULL))
host->pwr_reg_add |= MCI_ST_FBCLKEN;
if (of_get_property(np, "mmc-cap-mmc-highspeed", NULL))
mmc->caps |= MMC_CAP_MMC_HIGHSPEED;
if (of_get_property(np, "mmc-cap-sd-highspeed", NULL))
mmc->caps |= MMC_CAP_SD_HIGHSPEED;
return 0;
}
static int mmci_probe(struct amba_device *dev,
const struct amba_id *id)
{
struct mmci_platform_data *plat = dev->dev.platform_data;
struct device_node *np = dev->dev.of_node;
struct variant_data *variant = id->data;
struct mmci_host *host;
struct mmc_host *mmc;
int ret;
/* Must have platform data or Device Tree. */
if (!plat && !np) {
dev_err(&dev->dev, "No plat data or DT found\n");
return -EINVAL;
}
if (!plat) {
plat = devm_kzalloc(&dev->dev, sizeof(*plat), GFP_KERNEL);
if (!plat)
return -ENOMEM;
}
mmc = mmc_alloc_host(sizeof(struct mmci_host), &dev->dev);
if (!mmc)
return -ENOMEM;
ret = mmci_of_parse(np, mmc);
if (ret)
goto host_free;
host = mmc_priv(mmc);
host->mmc = mmc;
/*
* Some variant (STM32) doesn't have opendrain bit, nevertheless
* pins can be set accordingly using pinctrl
*/
if (!variant->opendrain) {
host->pinctrl = devm_pinctrl_get(&dev->dev);
if (IS_ERR(host->pinctrl)) {
dev_err(&dev->dev, "failed to get pinctrl");
ret = PTR_ERR(host->pinctrl);
goto host_free;
}
host->pins_default = pinctrl_lookup_state(host->pinctrl,
PINCTRL_STATE_DEFAULT);
if (IS_ERR(host->pins_default)) {
dev_err(mmc_dev(mmc), "Can't select default pins\n");
ret = PTR_ERR(host->pins_default);
goto host_free;
}
host->pins_opendrain = pinctrl_lookup_state(host->pinctrl,
MMCI_PINCTRL_STATE_OPENDRAIN);
if (IS_ERR(host->pins_opendrain)) {
dev_err(mmc_dev(mmc), "Can't select opendrain pins\n");
ret = PTR_ERR(host->pins_opendrain);
goto host_free;
}
}
host->hw_designer = amba_manf(dev);
host->hw_revision = amba_rev(dev);
dev_dbg(mmc_dev(mmc), "designer ID = 0x%02x\n", host->hw_designer);
dev_dbg(mmc_dev(mmc), "revision = 0x%01x\n", host->hw_revision);
host->clk = devm_clk_get(&dev->dev, NULL);
if (IS_ERR(host->clk)) {
ret = PTR_ERR(host->clk);
goto host_free;
}
ret = clk_prepare_enable(host->clk);
if (ret)
goto host_free;
if (variant->qcom_fifo)
host->get_rx_fifocnt = mmci_qcom_get_rx_fifocnt;
else
host->get_rx_fifocnt = mmci_get_rx_fifocnt;
host->plat = plat;
host->variant = variant;
host->mclk = clk_get_rate(host->clk);
/*
* According to the spec, mclk is max 100 MHz,
* so we try to adjust the clock down to this,
* (if possible).
*/
if (host->mclk > variant->f_max) {
ret = clk_set_rate(host->clk, variant->f_max);
if (ret < 0)
goto clk_disable;
host->mclk = clk_get_rate(host->clk);
dev_dbg(mmc_dev(mmc), "eventual mclk rate: %u Hz\n",
host->mclk);
}
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
host->phybase = dev->res.start;
host->base = devm_ioremap_resource(&dev->dev, &dev->res);
if (IS_ERR(host->base)) {
ret = PTR_ERR(host->base);
goto clk_disable;
}
if (variant->init)
variant->init(host);
/*
* The ARM and ST versions of the block have slightly different
* clock divider equations which means that the minimum divider
* differs too.
* on Qualcomm like controllers get the nearest minimum clock to 100Khz
*/
if (variant->st_clkdiv)
mmc->f_min = DIV_ROUND_UP(host->mclk, 257);
else if (variant->explicit_mclk_control)
mmc->f_min = clk_round_rate(host->clk, 100000);
else
mmc->f_min = DIV_ROUND_UP(host->mclk, 512);
/*
* If no maximum operating frequency is supplied, fall back to use
* the module parameter, which has a (low) default value in case it
* is not specified. Either value must not exceed the clock rate into
* the block, of course.
*/
if (mmc->f_max)
mmc->f_max = variant->explicit_mclk_control ?
min(variant->f_max, mmc->f_max) :
min(host->mclk, mmc->f_max);
else
mmc->f_max = variant->explicit_mclk_control ?
fmax : min(host->mclk, fmax);
dev_dbg(mmc_dev(mmc), "clocking block at %u Hz\n", mmc->f_max);
/* Get regulators and the supported OCR mask */
ret = mmc_regulator_get_supply(mmc);
if (ret)
goto clk_disable;
if (!mmc->ocr_avail)
mmc->ocr_avail = plat->ocr_mask;
else if (plat->ocr_mask)
dev_warn(mmc_dev(mmc), "Platform OCR mask is ignored\n");
/* We support these capabilities. */
mmc->caps |= MMC_CAP_CMD23;
/*
* Enable busy detection.
*/
if (variant->busy_detect) {
mmci_ops.card_busy = mmci_card_busy;
/*
* Not all variants have a flag to enable busy detection
* in the DPSM, but if they do, set it here.
*/
if (variant->busy_dpsm_flag)
mmci_write_datactrlreg(host,
host->variant->busy_dpsm_flag);
mmc->caps |= MMC_CAP_WAIT_WHILE_BUSY;
mmc->max_busy_timeout = 0;
}
mmc->ops = &mmci_ops;
/* We support these PM capabilities. */
mmc->pm_caps |= MMC_PM_KEEP_POWER;
/*
* We can do SGIO
*/
mmc->max_segs = NR_SG;
/*
* Since only a certain number of bits are valid in the data length
* register, we must ensure that we don't exceed 2^num-1 bytes in a
* single request.
*/
mmc->max_req_size = (1 << variant->datalength_bits) - 1;
/*
* Set the maximum segment size. Since we aren't doing DMA
* (yet) we are only limited by the data length register.
*/
mmc->max_seg_size = mmc->max_req_size;
/*
* Block size can be up to 2048 bytes, but must be a power of two.
*/
mmc->max_blk_size = 1 << 11;
/*
* Limit the number of blocks transferred so that we don't overflow
* the maximum request size.
*/
mmc->max_blk_count = mmc->max_req_size >> 11;
spin_lock_init(&host->lock);
writel(0, host->base + MMCIMASK0);
if (variant->mmcimask1)
writel(0, host->base + MMCIMASK1);
writel(0xfff, host->base + MMCICLEAR);
/*
* If:
* - not using DT but using a descriptor table, or
* - using a table of descriptors ALONGSIDE DT, or
* look up these descriptors named "cd" and "wp" right here, fail
* silently of these do not exist
*/
if (!np) {
ret = mmc_gpiod_request_cd(mmc, "cd", 0, false, 0, NULL);
if (ret == -EPROBE_DEFER)
goto clk_disable;
ret = mmc_gpiod_request_ro(mmc, "wp", 0, false, 0, NULL);
if (ret == -EPROBE_DEFER)
goto clk_disable;
}
ret = devm_request_irq(&dev->dev, dev->irq[0], mmci_irq, IRQF_SHARED,
DRIVER_NAME " (cmd)", host);
if (ret)
goto clk_disable;
if (!dev->irq[1])
host->singleirq = true;
else {
ret = devm_request_irq(&dev->dev, dev->irq[1], mmci_pio_irq,
IRQF_SHARED, DRIVER_NAME " (pio)", host);
if (ret)
goto clk_disable;
}
writel(MCI_IRQENABLE, host->base + MMCIMASK0);
amba_set_drvdata(dev, mmc);
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
dev_info(&dev->dev, "%s: PL%03x manf %x rev%u at 0x%08llx irq %d,%d (pio)\n",
mmc_hostname(mmc), amba_part(dev), amba_manf(dev),
amba_rev(dev), (unsigned long long)dev->res.start,
dev->irq[0], dev->irq[1]);
mmci_dma_setup(host);
pm_runtime_set_autosuspend_delay(&dev->dev, 50);
pm_runtime_use_autosuspend(&dev->dev);
mmc_add_host(mmc);
pm_runtime_put(&dev->dev);
return 0;
clk_disable:
clk_disable_unprepare(host->clk);
host_free:
mmc_free_host(mmc);
return ret;
}
static int mmci_remove(struct amba_device *dev)
{
struct mmc_host *mmc = amba_get_drvdata(dev);
if (mmc) {
struct mmci_host *host = mmc_priv(mmc);
struct variant_data *variant = host->variant;
/*
* Undo pm_runtime_put() in probe. We use the _sync
* version here so that we can access the primecell.
*/
pm_runtime_get_sync(&dev->dev);
mmc_remove_host(mmc);
writel(0, host->base + MMCIMASK0);
if (variant->mmcimask1)
writel(0, host->base + MMCIMASK1);
writel(0, host->base + MMCICOMMAND);
writel(0, host->base + MMCIDATACTRL);
ARM: mmci: add dmaengine-based DMA support Based on a patch from Linus Walleij. Add dmaengine based support for DMA to the MMCI driver, using the Primecell DMA engine interface. The changes over Linus' driver are: - rename txsize_threshold to dmasize_threshold, as this reflects the purpose more. - use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'. - clean up requesting of dma channels. - don't release a single channel twice when it's shared between tx and rx. - get rid of 'dma_enable' bool - instead check whether the channel is NULL. - detect incomplete DMA at the end of a transfer. Some DMA controllers (eg, PL08x) are unable to be configured for scatter DMA and also listen to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI. They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the final burst/words, PL08x does not transfer the last few words. - map and unmap DMA buffers using the DMA engine struct device, not the MMCI struct device - the DMA engine is doing the DMA transfer, not us. - avoid double-unmapping of the DMA buffers on MMCI data errors. - don't check for negative values from the dmaengine tx submission function - Dan says this must never fail. - use new dmaengine helper functions rather than using the ugly function pointers directly. - allow DMA code to be fully optimized away using dma_inprogress() which is defined to constant 0 if DMA engine support is disabled. - request maximum segment size from the DMA engine struct device and set this appropriately. - removed checking of buffer alignment - the DMA engine should deal with its own restrictions on buffer alignment, not the individual DMA engine users. - removed setting DMAREQCTL - this confuses some DMA controllers as it causes LBREQ to be asserted for the last seven transfers, rather than six SREQ and one LSREQ. - removed burst setting - the DMA controller should not burst past the transfer size required to complete the DMA operation. Tested-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2011-01-12 02:35:53 +07:00
mmci_dma_release(host);
clk_disable_unprepare(host->clk);
mmc_free_host(mmc);
}
return 0;
}
#ifdef CONFIG_PM
static void mmci_save(struct mmci_host *host)
{
unsigned long flags;
spin_lock_irqsave(&host->lock, flags);
writel(0, host->base + MMCIMASK0);
if (host->variant->pwrreg_nopower) {
writel(0, host->base + MMCIDATACTRL);
writel(0, host->base + MMCIPOWER);
writel(0, host->base + MMCICLOCK);
}
mmci_reg_delay(host);
spin_unlock_irqrestore(&host->lock, flags);
}
static void mmci_restore(struct mmci_host *host)
{
unsigned long flags;
spin_lock_irqsave(&host->lock, flags);
if (host->variant->pwrreg_nopower) {
writel(host->clk_reg, host->base + MMCICLOCK);
writel(host->datactrl_reg, host->base + MMCIDATACTRL);
writel(host->pwr_reg, host->base + MMCIPOWER);
}
writel(MCI_IRQENABLE, host->base + MMCIMASK0);
mmci_reg_delay(host);
spin_unlock_irqrestore(&host->lock, flags);
}
static int mmci_runtime_suspend(struct device *dev)
{
struct amba_device *adev = to_amba_device(dev);
struct mmc_host *mmc = amba_get_drvdata(adev);
if (mmc) {
struct mmci_host *host = mmc_priv(mmc);
pinctrl_pm_select_sleep_state(dev);
mmci_save(host);
clk_disable_unprepare(host->clk);
}
return 0;
}
static int mmci_runtime_resume(struct device *dev)
{
struct amba_device *adev = to_amba_device(dev);
struct mmc_host *mmc = amba_get_drvdata(adev);
if (mmc) {
struct mmci_host *host = mmc_priv(mmc);
clk_prepare_enable(host->clk);
mmci_restore(host);
pinctrl_pm_select_default_state(dev);
}
return 0;
}
#endif
static const struct dev_pm_ops mmci_dev_pm_ops = {
SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
pm_runtime_force_resume)
SET_RUNTIME_PM_OPS(mmci_runtime_suspend, mmci_runtime_resume, NULL)
};
static const struct amba_id mmci_ids[] = {
{
.id = 0x00041180,
.mask = 0xff0fffff,
.data = &variant_arm,
},
{
.id = 0x01041180,
.mask = 0xff0fffff,
.data = &variant_arm_extended_fifo,
},
{
.id = 0x02041180,
.mask = 0xff0fffff,
.data = &variant_arm_extended_fifo_hwfc,
},
{
.id = 0x00041181,
.mask = 0x000fffff,
.data = &variant_arm,
},
/* ST Micro variants */
{
.id = 0x00180180,
.mask = 0x00ffffff,
.data = &variant_u300,
},
{
.id = 0x10180180,
.mask = 0xf0ffffff,
.data = &variant_nomadik,
},
{
.id = 0x00280180,
.mask = 0x00ffffff,
.data = &variant_nomadik,
},
{
.id = 0x00480180,
.mask = 0xf0ffffff,
.data = &variant_ux500,
},
{
.id = 0x10480180,
.mask = 0xf0ffffff,
.data = &variant_ux500v2,
},
{
.id = 0x00880180,
.mask = 0x00ffffff,
.data = &variant_stm32,
},
/* Qualcomm variants */
{
.id = 0x00051180,
.mask = 0x000fffff,
.data = &variant_qcom,
},
{ 0, 0 },
};
MODULE_DEVICE_TABLE(amba, mmci_ids);
static struct amba_driver mmci_driver = {
.drv = {
.name = DRIVER_NAME,
.pm = &mmci_dev_pm_ops,
},
.probe = mmci_probe,
.remove = mmci_remove,
.id_table = mmci_ids,
};
module_amba_driver(mmci_driver);
module_param(fmax, uint, 0444);
MODULE_DESCRIPTION("ARM PrimeCell PL180/181 Multimedia Card Interface driver");
MODULE_LICENSE("GPL");