linux_dsm_epyc7002/drivers/usb/musb/musb_cppi41.c

715 lines
19 KiB
C
Raw Normal View History

#include <linux/device.h>
#include <linux/dma-mapping.h>
#include <linux/dmaengine.h>
#include <linux/sizes.h>
#include <linux/platform_device.h>
#include <linux/of.h>
#include "musb_core.h"
#define RNDIS_REG(x) (0x80 + ((x - 1) * 4))
#define EP_MODE_AUTOREG_NONE 0
#define EP_MODE_AUTOREG_ALL_NEOP 1
#define EP_MODE_AUTOREG_ALWAYS 3
#define EP_MODE_DMA_TRANSPARENT 0
#define EP_MODE_DMA_RNDIS 1
#define EP_MODE_DMA_GEN_RNDIS 3
#define USB_CTRL_TX_MODE 0x70
#define USB_CTRL_RX_MODE 0x74
#define USB_CTRL_AUTOREQ 0xd0
#define USB_TDOWN 0xd8
struct cppi41_dma_channel {
struct dma_channel channel;
struct cppi41_dma_controller *controller;
struct musb_hw_ep *hw_ep;
struct dma_chan *dc;
dma_cookie_t cookie;
u8 port_num;
u8 is_tx;
u8 is_allocated;
u8 usb_toggle;
dma_addr_t buf_addr;
u32 total_len;
u32 prog_len;
u32 transferred;
u32 packet_sz;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
struct list_head tx_check;
int tx_zlp;
};
#define MUSB_DMA_NUM_CHANNELS 15
struct cppi41_dma_controller {
struct dma_controller controller;
struct cppi41_dma_channel rx_channel[MUSB_DMA_NUM_CHANNELS];
struct cppi41_dma_channel tx_channel[MUSB_DMA_NUM_CHANNELS];
struct musb *musb;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
struct hrtimer early_tx;
struct list_head early_tx_list;
u32 rx_mode;
u32 tx_mode;
u32 auto_req;
};
static void save_rx_toggle(struct cppi41_dma_channel *cppi41_channel)
{
u16 csr;
u8 toggle;
if (cppi41_channel->is_tx)
return;
if (!is_host_active(cppi41_channel->controller->musb))
return;
csr = musb_readw(cppi41_channel->hw_ep->regs, MUSB_RXCSR);
toggle = csr & MUSB_RXCSR_H_DATATOGGLE ? 1 : 0;
cppi41_channel->usb_toggle = toggle;
}
static void update_rx_toggle(struct cppi41_dma_channel *cppi41_channel)
{
struct musb_hw_ep *hw_ep = cppi41_channel->hw_ep;
struct musb *musb = hw_ep->musb;
u16 csr;
u8 toggle;
if (cppi41_channel->is_tx)
return;
if (!is_host_active(musb))
return;
musb_ep_select(musb->mregs, hw_ep->epnum);
csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
toggle = csr & MUSB_RXCSR_H_DATATOGGLE ? 1 : 0;
/*
* AM335x Advisory 1.0.13: Due to internal synchronisation error the
* data toggle may reset from DATA1 to DATA0 during receiving data from
* more than one endpoint.
*/
if (!toggle && toggle == cppi41_channel->usb_toggle) {
csr |= MUSB_RXCSR_H_DATATOGGLE | MUSB_RXCSR_H_WR_DATATOGGLE;
musb_writew(cppi41_channel->hw_ep->regs, MUSB_RXCSR, csr);
dev_dbg(cppi41_channel->controller->musb->controller,
"Restoring DATA1 toggle.\n");
}
cppi41_channel->usb_toggle = toggle;
}
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
static bool musb_is_tx_fifo_empty(struct musb_hw_ep *hw_ep)
{
u8 epnum = hw_ep->epnum;
struct musb *musb = hw_ep->musb;
void __iomem *epio = musb->endpoints[epnum].regs;
u16 csr;
musb_ep_select(musb->mregs, hw_ep->epnum);
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
csr = musb_readw(epio, MUSB_TXCSR);
if (csr & MUSB_TXCSR_TXPKTRDY)
return false;
return true;
}
static void cppi41_dma_callback(void *private_data);
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
static void cppi41_trans_done(struct cppi41_dma_channel *cppi41_channel)
{
struct musb_hw_ep *hw_ep = cppi41_channel->hw_ep;
struct musb *musb = hw_ep->musb;
void __iomem *epio = hw_ep->regs;
u16 csr;
if (!cppi41_channel->prog_len ||
(cppi41_channel->channel.status == MUSB_DMA_STATUS_FREE)) {
/* done, complete */
cppi41_channel->channel.actual_len =
cppi41_channel->transferred;
cppi41_channel->channel.status = MUSB_DMA_STATUS_FREE;
cppi41_channel->channel.rx_packet_done = true;
/*
* transmit ZLP using PIO mode for transfers which size is
* multiple of EP packet size.
*/
if (cppi41_channel->tx_zlp && (cppi41_channel->transferred %
cppi41_channel->packet_sz) == 0) {
musb_ep_select(musb->mregs, hw_ep->epnum);
csr = MUSB_TXCSR_MODE | MUSB_TXCSR_TXPKTRDY;
musb_writew(epio, MUSB_TXCSR, csr);
}
musb_dma_completion(musb, hw_ep->epnum, cppi41_channel->is_tx);
} else {
/* next iteration, reload */
struct dma_chan *dc = cppi41_channel->dc;
struct dma_async_tx_descriptor *dma_desc;
enum dma_transfer_direction direction;
u32 remain_bytes;
cppi41_channel->buf_addr += cppi41_channel->packet_sz;
remain_bytes = cppi41_channel->total_len;
remain_bytes -= cppi41_channel->transferred;
remain_bytes = min(remain_bytes, cppi41_channel->packet_sz);
cppi41_channel->prog_len = remain_bytes;
direction = cppi41_channel->is_tx ? DMA_MEM_TO_DEV
: DMA_DEV_TO_MEM;
dma_desc = dmaengine_prep_slave_single(dc,
cppi41_channel->buf_addr,
remain_bytes,
direction,
DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
if (WARN_ON(!dma_desc))
return;
dma_desc->callback = cppi41_dma_callback;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
dma_desc->callback_param = &cppi41_channel->channel;
cppi41_channel->cookie = dma_desc->tx_submit(dma_desc);
dma_async_issue_pending(dc);
if (!cppi41_channel->is_tx) {
musb_ep_select(musb->mregs, hw_ep->epnum);
csr = musb_readw(epio, MUSB_RXCSR);
csr |= MUSB_RXCSR_H_REQPKT;
musb_writew(epio, MUSB_RXCSR, csr);
}
}
}
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
static enum hrtimer_restart cppi41_recheck_tx_req(struct hrtimer *timer)
{
struct cppi41_dma_controller *controller;
struct cppi41_dma_channel *cppi41_channel, *n;
struct musb *musb;
unsigned long flags;
enum hrtimer_restart ret = HRTIMER_NORESTART;
controller = container_of(timer, struct cppi41_dma_controller,
early_tx);
musb = controller->musb;
spin_lock_irqsave(&musb->lock, flags);
list_for_each_entry_safe(cppi41_channel, n, &controller->early_tx_list,
tx_check) {
bool empty;
struct musb_hw_ep *hw_ep = cppi41_channel->hw_ep;
empty = musb_is_tx_fifo_empty(hw_ep);
if (empty) {
list_del_init(&cppi41_channel->tx_check);
cppi41_trans_done(cppi41_channel);
}
}
if (!list_empty(&controller->early_tx_list) &&
!hrtimer_is_queued(&controller->early_tx)) {
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
ret = HRTIMER_RESTART;
hrtimer_forward_now(&controller->early_tx,
ktime_set(0, 20 * NSEC_PER_USEC));
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
}
spin_unlock_irqrestore(&musb->lock, flags);
return ret;
}
static void cppi41_dma_callback(void *private_data)
{
struct dma_channel *channel = private_data;
struct cppi41_dma_channel *cppi41_channel = channel->private_data;
struct musb_hw_ep *hw_ep = cppi41_channel->hw_ep;
struct musb *musb = hw_ep->musb;
unsigned long flags;
struct dma_tx_state txstate;
u32 transferred;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
bool empty;
spin_lock_irqsave(&musb->lock, flags);
dmaengine_tx_status(cppi41_channel->dc, cppi41_channel->cookie,
&txstate);
transferred = cppi41_channel->prog_len - txstate.residue;
cppi41_channel->transferred += transferred;
dev_dbg(musb->controller, "DMA transfer done on hw_ep=%d bytes=%d/%d\n",
hw_ep->epnum, cppi41_channel->transferred,
cppi41_channel->total_len);
update_rx_toggle(cppi41_channel);
if (cppi41_channel->transferred == cppi41_channel->total_len ||
transferred < cppi41_channel->packet_sz)
cppi41_channel->prog_len = 0;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
empty = musb_is_tx_fifo_empty(hw_ep);
if (empty) {
cppi41_trans_done(cppi41_channel);
} else {
struct cppi41_dma_controller *controller;
/*
* On AM335x it has been observed that the TX interrupt fires
* too early that means the TXFIFO is not yet empty but the DMA
* engine says that it is done with the transfer. We don't
* receive a FIFO empty interrupt so the only thing we can do is
* to poll for the bit. On HS it usually takes 2us, on FS around
* 110us - 150us depending on the transfer size.
* We spin on HS (no longer than than 25us and setup a timer on
* FS to check for the bit and complete the transfer.
*/
controller = cppi41_channel->controller;
if (musb->g.speed == USB_SPEED_HIGH) {
unsigned wait = 25;
do {
empty = musb_is_tx_fifo_empty(hw_ep);
if (empty)
break;
wait--;
if (!wait)
break;
udelay(1);
} while (1);
empty = musb_is_tx_fifo_empty(hw_ep);
if (empty) {
cppi41_trans_done(cppi41_channel);
goto out;
}
}
list_add_tail(&cppi41_channel->tx_check,
&controller->early_tx_list);
usb: musb: Ensure that cppi41 timer gets armed on premature DMA TX irq Some TI chips raise the DMA complete interrupt before the actual transfer has been completed. The code tries to busy wait for a few microseconds and if that fails it arms an hrtimer to recheck. So far so good, but that has the following issue: CPU 0 CPU1 start_next_transfer(RQ1); DMA interrupt if (premature_irq(RQ1)) if (!hrtimer_active(timer)) hrtimer_start(timer); hrtimer expires timer->state = CALLBACK_RUNNING; timer->fn() cppi41_recheck_tx_req() complete_request(RQ1); if (requests_pending()) start_next_transfer(RQ2); DMA interrupt if (premature_irq(RQ2)) if (!hrtimer_active(timer)) hrtimer_start(timer); timer->state = INACTIVE; The premature interrupt of request2 on CPU1 does not arm the timer and therefor the request completion never happens because it checks for !hrtimer_active(). hrtimer_active() evaluates: timer->state != HRTIMER_STATE_INACTIVE which of course evaluates to true in the above case as timer->state is CALLBACK_RUNNING. That's clearly documented: * A timer is active, when it is enqueued into the rbtree or the * callback function is running or it's in the state of being migrated * to another cpu. But that's not what the code wants to check. The code wants to check whether the timer is queued, i.e. whether its armed and waiting for expiry. We have a helper function for this: hrtimer_is_queued(). This evaluates: timer->state & HRTIMER_STATE_QUEUED So in the above case this evaluates to false and therefor forces the DMA interrupt on CPU1 to call hrtimer_start(). Use hrtimer_is_queued() instead of hrtimer_active() and evrything is good. Reported-by: Torben Hohn <torbenh@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org Signed-off-by: Felipe Balbi <balbi@ti.com>
2014-06-21 04:41:24 +07:00
if (!hrtimer_is_queued(&controller->early_tx)) {
2014-06-20 05:20:44 +07:00
unsigned long usecs = cppi41_channel->total_len / 10;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
hrtimer_start_range_ns(&controller->early_tx,
2014-06-20 05:20:44 +07:00
ktime_set(0, usecs * NSEC_PER_USEC),
20 * NSEC_PER_USEC,
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
HRTIMER_MODE_REL);
}
}
out:
spin_unlock_irqrestore(&musb->lock, flags);
}
static u32 update_ep_mode(unsigned ep, unsigned mode, u32 old)
{
unsigned shift;
shift = (ep - 1) * 2;
old &= ~(3 << shift);
old |= mode << shift;
return old;
}
static void cppi41_set_dma_mode(struct cppi41_dma_channel *cppi41_channel,
unsigned mode)
{
struct cppi41_dma_controller *controller = cppi41_channel->controller;
u32 port;
u32 new_mode;
u32 old_mode;
if (cppi41_channel->is_tx)
old_mode = controller->tx_mode;
else
old_mode = controller->rx_mode;
port = cppi41_channel->port_num;
new_mode = update_ep_mode(port, mode, old_mode);
if (new_mode == old_mode)
return;
if (cppi41_channel->is_tx) {
controller->tx_mode = new_mode;
musb_writel(controller->musb->ctrl_base, USB_CTRL_TX_MODE,
new_mode);
} else {
controller->rx_mode = new_mode;
musb_writel(controller->musb->ctrl_base, USB_CTRL_RX_MODE,
new_mode);
}
}
static void cppi41_set_autoreq_mode(struct cppi41_dma_channel *cppi41_channel,
unsigned mode)
{
struct cppi41_dma_controller *controller = cppi41_channel->controller;
u32 port;
u32 new_mode;
u32 old_mode;
old_mode = controller->auto_req;
port = cppi41_channel->port_num;
new_mode = update_ep_mode(port, mode, old_mode);
if (new_mode == old_mode)
return;
controller->auto_req = new_mode;
musb_writel(controller->musb->ctrl_base, USB_CTRL_AUTOREQ, new_mode);
}
static bool cppi41_configure_channel(struct dma_channel *channel,
u16 packet_sz, u8 mode,
dma_addr_t dma_addr, u32 len)
{
struct cppi41_dma_channel *cppi41_channel = channel->private_data;
struct dma_chan *dc = cppi41_channel->dc;
struct dma_async_tx_descriptor *dma_desc;
enum dma_transfer_direction direction;
struct musb *musb = cppi41_channel->controller->musb;
unsigned use_gen_rndis = 0;
dev_dbg(musb->controller,
"configure ep%d/%x packet_sz=%d, mode=%d, dma_addr=0x%llx, len=%d is_tx=%d\n",
cppi41_channel->port_num, RNDIS_REG(cppi41_channel->port_num),
packet_sz, mode, (unsigned long long) dma_addr,
len, cppi41_channel->is_tx);
cppi41_channel->buf_addr = dma_addr;
cppi41_channel->total_len = len;
cppi41_channel->transferred = 0;
cppi41_channel->packet_sz = packet_sz;
cppi41_channel->tx_zlp = (cppi41_channel->is_tx && mode) ? 1 : 0;
/*
* Due to AM335x' Advisory 1.0.13 we are not allowed to transfer more
* than max packet size at a time.
*/
if (cppi41_channel->is_tx)
use_gen_rndis = 1;
if (use_gen_rndis) {
/* RNDIS mode */
if (len > packet_sz) {
musb_writel(musb->ctrl_base,
RNDIS_REG(cppi41_channel->port_num), len);
/* gen rndis */
cppi41_set_dma_mode(cppi41_channel,
EP_MODE_DMA_GEN_RNDIS);
/* auto req */
cppi41_set_autoreq_mode(cppi41_channel,
EP_MODE_AUTOREG_ALL_NEOP);
} else {
musb_writel(musb->ctrl_base,
RNDIS_REG(cppi41_channel->port_num), 0);
cppi41_set_dma_mode(cppi41_channel,
EP_MODE_DMA_TRANSPARENT);
cppi41_set_autoreq_mode(cppi41_channel,
EP_MODE_AUTOREG_NONE);
}
} else {
/* fallback mode */
cppi41_set_dma_mode(cppi41_channel, EP_MODE_DMA_TRANSPARENT);
cppi41_set_autoreq_mode(cppi41_channel, EP_MODE_AUTOREG_NONE);
len = min_t(u32, packet_sz, len);
}
cppi41_channel->prog_len = len;
direction = cppi41_channel->is_tx ? DMA_MEM_TO_DEV : DMA_DEV_TO_MEM;
dma_desc = dmaengine_prep_slave_single(dc, dma_addr, len, direction,
DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
if (!dma_desc)
return false;
dma_desc->callback = cppi41_dma_callback;
dma_desc->callback_param = channel;
cppi41_channel->cookie = dma_desc->tx_submit(dma_desc);
cppi41_channel->channel.rx_packet_done = false;
save_rx_toggle(cppi41_channel);
dma_async_issue_pending(dc);
return true;
}
static struct dma_channel *cppi41_dma_channel_allocate(struct dma_controller *c,
struct musb_hw_ep *hw_ep, u8 is_tx)
{
struct cppi41_dma_controller *controller = container_of(c,
struct cppi41_dma_controller, controller);
struct cppi41_dma_channel *cppi41_channel = NULL;
u8 ch_num = hw_ep->epnum - 1;
if (ch_num >= MUSB_DMA_NUM_CHANNELS)
return NULL;
if (is_tx)
cppi41_channel = &controller->tx_channel[ch_num];
else
cppi41_channel = &controller->rx_channel[ch_num];
if (!cppi41_channel->dc)
return NULL;
if (cppi41_channel->is_allocated)
return NULL;
cppi41_channel->hw_ep = hw_ep;
cppi41_channel->is_allocated = 1;
return &cppi41_channel->channel;
}
static void cppi41_dma_channel_release(struct dma_channel *channel)
{
struct cppi41_dma_channel *cppi41_channel = channel->private_data;
if (cppi41_channel->is_allocated) {
cppi41_channel->is_allocated = 0;
channel->status = MUSB_DMA_STATUS_FREE;
channel->actual_len = 0;
}
}
static int cppi41_dma_channel_program(struct dma_channel *channel,
u16 packet_sz, u8 mode,
dma_addr_t dma_addr, u32 len)
{
int ret;
struct cppi41_dma_channel *cppi41_channel = channel->private_data;
int hb_mult = 0;
BUG_ON(channel->status == MUSB_DMA_STATUS_UNKNOWN ||
channel->status == MUSB_DMA_STATUS_BUSY);
if (is_host_active(cppi41_channel->controller->musb)) {
if (cppi41_channel->is_tx)
hb_mult = cppi41_channel->hw_ep->out_qh->hb_mult;
else
hb_mult = cppi41_channel->hw_ep->in_qh->hb_mult;
}
channel->status = MUSB_DMA_STATUS_BUSY;
channel->actual_len = 0;
if (hb_mult)
packet_sz = hb_mult * (packet_sz & 0x7FF);
ret = cppi41_configure_channel(channel, packet_sz, mode, dma_addr, len);
if (!ret)
channel->status = MUSB_DMA_STATUS_FREE;
return ret;
}
static int cppi41_is_compatible(struct dma_channel *channel, u16 maxpacket,
void *buf, u32 length)
{
struct cppi41_dma_channel *cppi41_channel = channel->private_data;
struct cppi41_dma_controller *controller = cppi41_channel->controller;
struct musb *musb = controller->musb;
if (is_host_active(musb)) {
WARN_ON(1);
return 1;
}
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
if (cppi41_channel->hw_ep->ep_in.type != USB_ENDPOINT_XFER_BULK)
return 0;
if (cppi41_channel->is_tx)
return 1;
/* AM335x Advisory 1.0.13. No workaround for device RX mode */
return 0;
}
static int cppi41_dma_channel_abort(struct dma_channel *channel)
{
struct cppi41_dma_channel *cppi41_channel = channel->private_data;
struct cppi41_dma_controller *controller = cppi41_channel->controller;
struct musb *musb = controller->musb;
void __iomem *epio = cppi41_channel->hw_ep->regs;
int tdbit;
int ret;
unsigned is_tx;
u16 csr;
is_tx = cppi41_channel->is_tx;
dev_dbg(musb->controller, "abort channel=%d, is_tx=%d\n",
cppi41_channel->port_num, is_tx);
if (cppi41_channel->channel.status == MUSB_DMA_STATUS_FREE)
return 0;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
list_del_init(&cppi41_channel->tx_check);
if (is_tx) {
csr = musb_readw(epio, MUSB_TXCSR);
csr &= ~MUSB_TXCSR_DMAENAB;
musb_writew(epio, MUSB_TXCSR, csr);
} else {
csr = musb_readw(epio, MUSB_RXCSR);
csr &= ~(MUSB_RXCSR_H_REQPKT | MUSB_RXCSR_DMAENAB);
musb_writew(epio, MUSB_RXCSR, csr);
csr = musb_readw(epio, MUSB_RXCSR);
if (csr & MUSB_RXCSR_RXPKTRDY) {
csr |= MUSB_RXCSR_FLUSHFIFO;
musb_writew(epio, MUSB_RXCSR, csr);
musb_writew(epio, MUSB_RXCSR, csr);
}
}
tdbit = 1 << cppi41_channel->port_num;
if (is_tx)
tdbit <<= 16;
do {
musb_writel(musb->ctrl_base, USB_TDOWN, tdbit);
ret = dmaengine_terminate_all(cppi41_channel->dc);
} while (ret == -EAGAIN);
musb_writel(musb->ctrl_base, USB_TDOWN, tdbit);
if (is_tx) {
csr = musb_readw(epio, MUSB_TXCSR);
if (csr & MUSB_TXCSR_TXPKTRDY) {
csr |= MUSB_TXCSR_FLUSHFIFO;
musb_writew(epio, MUSB_TXCSR, csr);
}
}
cppi41_channel->channel.status = MUSB_DMA_STATUS_FREE;
return 0;
}
static void cppi41_release_all_dma_chans(struct cppi41_dma_controller *ctrl)
{
struct dma_chan *dc;
int i;
for (i = 0; i < MUSB_DMA_NUM_CHANNELS; i++) {
dc = ctrl->tx_channel[i].dc;
if (dc)
dma_release_channel(dc);
dc = ctrl->rx_channel[i].dc;
if (dc)
dma_release_channel(dc);
}
}
static void cppi41_dma_controller_stop(struct cppi41_dma_controller *controller)
{
cppi41_release_all_dma_chans(controller);
}
static int cppi41_dma_controller_start(struct cppi41_dma_controller *controller)
{
struct musb *musb = controller->musb;
struct device *dev = musb->controller;
struct device_node *np = dev->of_node;
struct cppi41_dma_channel *cppi41_channel;
int count;
int i;
int ret;
count = of_property_count_strings(np, "dma-names");
if (count < 0)
return count;
for (i = 0; i < count; i++) {
struct dma_chan *dc;
struct dma_channel *musb_dma;
const char *str;
unsigned is_tx;
unsigned int port;
ret = of_property_read_string_index(np, "dma-names", i, &str);
if (ret)
goto err;
if (!strncmp(str, "tx", 2))
is_tx = 1;
else if (!strncmp(str, "rx", 2))
is_tx = 0;
else {
dev_err(dev, "Wrong dmatype %s\n", str);
goto err;
}
ret = kstrtouint(str + 2, 0, &port);
if (ret)
goto err;
ret = -EINVAL;
if (port > MUSB_DMA_NUM_CHANNELS || !port)
goto err;
if (is_tx)
cppi41_channel = &controller->tx_channel[port - 1];
else
cppi41_channel = &controller->rx_channel[port - 1];
cppi41_channel->controller = controller;
cppi41_channel->port_num = port;
cppi41_channel->is_tx = is_tx;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
INIT_LIST_HEAD(&cppi41_channel->tx_check);
musb_dma = &cppi41_channel->channel;
musb_dma->private_data = cppi41_channel;
musb_dma->status = MUSB_DMA_STATUS_FREE;
musb_dma->max_len = SZ_4M;
dc = dma_request_slave_channel(dev, str);
if (!dc) {
dev_err(dev, "Failed to request %s.\n", str);
ret = -EPROBE_DEFER;
goto err;
}
cppi41_channel->dc = dc;
}
return 0;
err:
cppi41_release_all_dma_chans(controller);
return ret;
}
void dma_controller_destroy(struct dma_controller *c)
{
struct cppi41_dma_controller *controller = container_of(c,
struct cppi41_dma_controller, controller);
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
hrtimer_cancel(&controller->early_tx);
cppi41_dma_controller_stop(controller);
kfree(controller);
}
struct dma_controller *dma_controller_create(struct musb *musb,
void __iomem *base)
{
struct cppi41_dma_controller *controller;
int ret = 0;
if (!musb->controller->of_node) {
dev_err(musb->controller, "Need DT for the DMA engine.\n");
return NULL;
}
controller = kzalloc(sizeof(*controller), GFP_KERNEL);
if (!controller)
goto kzalloc_fail;
usb: musb: musb_cppi41: handle pre-mature TX complete interrupt The TX-complete interrupt of the CPPI41 on AM335x fires too early. Adding a loop and counting how long it takes until the MUSB_TXCSR_TXPKTRDY bit is cleared I see FS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadc54002, len=1514 is_tx=1 |cppi41_dma_callback() 74 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8802, len=1514 is_tx=1 |cppi41_dma_callback() 66 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadcd8002, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=64, mode=0, dma_addr=0xadf55802, len=1514 is_tx=1 |cppi41_dma_callback() 136 loops avg: 110 - 150us HS: |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xaca6f002, len=1514 is_tx=1 |cppi41_dma_callback() 0 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f802, len=1514 is_tx=1 |cppi41_dma_callback() 2 loops |musb-hdrc musb-hdrc.0.auto: configure ep1/80 packet_sz=512, mode=0, dma_addr=0xadd6f002, len=1514 is_tx=1 |cppi41_dma_callback() 13 loops avg: 2us for the same test case. One loop means a udelay(1). The delay seems to depend on the packet size. On HS the bit is always cleared for small packet sizes while on FS it is never the case, it mostly around 110us. This testing has been performed with g_ether (musb as device) and using BULK transfers. INTR transfers are way more fun: during init the gadget sends a INT packet to the host and cppi41 says "transfer done" shortly after. The MUSB_TXCSR_TXPKTRDY bit is set even seconds later. The reason is that the host did not try to receive it, it does so after the interface (on host side) has been configured. Until this happens, that packet remains in musb's FIFO. To fix this, two things are done: - No DMA transfers for INT based endpoints. These transfer are usually very small and rare so it is likely better to skip the DMA engine and stuff the four bytes directly into the FIFO - on HS we poll up to 25us and hope that bit goes away. If not we setup a hrtimer to poll for it. The 140us delay is a rule of thumb. In FS the command | ping 10.10.10.10 -c1 -s65130 creates about 44 1514bytes transfers. About 19 of them need a second timer to complete. Reported-by: Bin Liu <b-liu@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Felipe Balbi <balbi@ti.com>
2013-11-12 22:37:47 +07:00
hrtimer_init(&controller->early_tx, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
controller->early_tx.function = cppi41_recheck_tx_req;
INIT_LIST_HEAD(&controller->early_tx_list);
controller->musb = musb;
controller->controller.channel_alloc = cppi41_dma_channel_allocate;
controller->controller.channel_release = cppi41_dma_channel_release;
controller->controller.channel_program = cppi41_dma_channel_program;
controller->controller.channel_abort = cppi41_dma_channel_abort;
controller->controller.is_compatible = cppi41_is_compatible;
ret = cppi41_dma_controller_start(controller);
if (ret)
goto plat_get_fail;
return &controller->controller;
plat_get_fail:
kfree(controller);
kzalloc_fail:
if (ret == -EPROBE_DEFER)
return ERR_PTR(ret);
return NULL;
}