linux_dsm_epyc7002/drivers/gpu/drm/radeon/radeon_kms.c

930 lines
28 KiB
C
Raw Normal View History

drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
/*
* Copyright 2008 Advanced Micro Devices, Inc.
* Copyright 2008 Red Hat Inc.
* Copyright 2009 Jerome Glisse.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
* Authors: Dave Airlie
* Alex Deucher
* Jerome Glisse
*/
#include <linux/pm_runtime.h>
#include <linux/slab.h>
#include <linux/uaccess.h>
#include <linux/vga_switcheroo.h>
#include <drm/drm_fb_helper.h>
#include <drm/drm_file.h>
#include <drm/drm_ioctl.h>
#include <drm/drm_pci.h>
#include <drm/radeon_drm.h>
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
#include "radeon.h"
#include "radeon_asic.h"
#if defined(CONFIG_VGA_SWITCHEROO)
bool radeon_has_atpx(void);
#else
static inline bool radeon_has_atpx(void) { return false; }
#endif
/**
* radeon_driver_unload_kms - Main unload function for KMS.
*
* @dev: drm dev pointer
*
* This is the main unload function for KMS (all asics).
* It calls radeon_modeset_fini() to tear down the
* displays, and radeon_device_fini() to tear down
* the rest of the device (CP, writeback, etc.).
* Returns 0 on success.
*/
void radeon_driver_unload_kms(struct drm_device *dev)
{
struct radeon_device *rdev = dev->dev_private;
if (rdev == NULL)
return;
if (rdev->rmmio == NULL)
goto done_free;
if (radeon_is_px(dev)) {
pm_runtime_get_sync(dev->dev);
pm_runtime_forbid(dev->dev);
}
radeon_acpi_fini(rdev);
radeon_modeset_fini(rdev);
radeon_device_fini(rdev);
done_free:
kfree(rdev);
dev->dev_private = NULL;
}
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
/**
* radeon_driver_load_kms - Main load function for KMS.
*
* @dev: drm dev pointer
* @flags: device flags
*
* This is the main load function for KMS (all asics).
* It calls radeon_device_init() to set up the non-display
* parts of the chip (asic init, CP, writeback, etc.), and
* radeon_modeset_init() to set up the display parts
* (crtcs, encoders, hotplug detect, etc.).
* Returns 0 on success, error on failure.
*/
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags)
{
struct radeon_device *rdev;
int r, acpi_status;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
if (!radeon_si_support) {
switch (flags & RADEON_FAMILY_MASK) {
case CHIP_TAHITI:
case CHIP_PITCAIRN:
case CHIP_VERDE:
case CHIP_OLAND:
case CHIP_HAINAN:
dev_info(dev->dev,
"SI support disabled by module param\n");
return -ENODEV;
}
}
if (!radeon_cik_support) {
switch (flags & RADEON_FAMILY_MASK) {
case CHIP_KAVERI:
case CHIP_BONAIRE:
case CHIP_HAWAII:
case CHIP_KABINI:
case CHIP_MULLINS:
dev_info(dev->dev,
"CIK support disabled by module param\n");
return -ENODEV;
}
}
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
rdev = kzalloc(sizeof(struct radeon_device), GFP_KERNEL);
if (rdev == NULL) {
return -ENOMEM;
}
dev->dev_private = (void *)rdev;
/* update BUS flag */
if (pci_find_capability(dev->pdev, PCI_CAP_ID_AGP)) {
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
flags |= RADEON_IS_AGP;
} else if (pci_is_pcie(dev->pdev)) {
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
flags |= RADEON_IS_PCIE;
} else {
flags |= RADEON_IS_PCI;
}
if ((radeon_runtime_pm != 0) &&
radeon_has_atpx() &&
((flags & RADEON_IS_IGP) == 0) &&
drm/radeon: Fix oops upon driver load on PowerXpress laptops Nicolai Stange reports the following oops which is caused by dereferencing rdev->pdev before it's subsequently set by radeon_device_init(). Fix it. BUG: unable to handle kernel NULL pointer dereference at 00000000000007cb IP: radeon_driver_load_kms+0xeb/0x230 [radeon] ... Call Trace: drm_dev_register+0x146/0x1d0 [drm] drm_get_pci_dev+0x9a/0x180 [drm] radeon_pci_probe+0xb8/0xe0 [radeon] local_pci_probe+0x45/0xa0 pci_device_probe+0x14f/0x1a0 driver_probe_device+0x29c/0x450 __driver_attach+0xdf/0xf0 ? driver_probe_device+0x450/0x450 bus_for_each_dev+0x6c/0xc0 driver_attach+0x1e/0x20 bus_add_driver+0x170/0x270 driver_register+0x60/0xe0 ? 0xffffffffc0508000 __pci_register_driver+0x4c/0x50 drm_pci_init+0xeb/0x100 [drm] ? vga_switcheroo_register_handler+0x6a/0x90 ? 0xffffffffc0508000 radeon_init+0x98/0xb6 [radeon] do_one_initcall+0x52/0x1a0 ? __vunmap+0x81/0xb0 ? kmem_cache_alloc_trace+0x159/0x1b0 ? do_init_module+0x27/0x1f8 do_init_module+0x5f/0x1f8 load_module+0x27ce/0x2be0 SYSC_finit_module+0xdf/0x110 ? SYSC_finit_module+0xdf/0x110 SyS_finit_module+0xe/0x10 do_syscall_64+0x67/0x150 entry_SYSCALL64_slow_path+0x25/0x25 Fixes: 7ffb0ce31cf9 ("drm/radeon: Don't register Thunderbolt eGPU with vga_switcheroo") Reported-and-tested-by: Nicolai Stange <nicstange@gmail.com> Signed-off-by: Lukas Wunner <lukas@wunner.de> Link: http://patchwork.freedesktop.org/patch/msgid/cfb91ba052af06117137eec0637543a2626a7979.1495135190.git.lukas@wunner.de
2017-05-19 02:33:44 +07:00
!pci_is_thunderbolt_attached(dev->pdev))
flags |= RADEON_IS_PX;
/* radeon_device_init should report only fatal error
* like memory allocation failure or iomapping failure,
* or memory manager initialization failure, it must
* properly initialize the GPU MC controller and permit
* VRAM allocation
*/
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
r = radeon_device_init(rdev, dev, dev->pdev, flags);
if (r) {
dev_err(&dev->pdev->dev, "Fatal error during GPU init\n");
goto out;
}
/* Again modeset_init should fail only on fatal error
* otherwise it should provide enough functionalities
* for shadowfb to run
*/
r = radeon_modeset_init(rdev);
if (r)
dev_err(&dev->pdev->dev, "Fatal error during modeset init\n");
/* Call ACPI methods: require modeset init
* but failure is not fatal
*/
if (!r) {
acpi_status = radeon_acpi_init(rdev);
if (acpi_status)
dev_dbg(&dev->pdev->dev,
"Error during ACPI methods call\n");
}
if (radeon_is_px(dev)) {
dev_pm_set_driver_flags(dev->dev, DPM_FLAG_NEVER_SKIP);
pm_runtime_use_autosuspend(dev->dev);
pm_runtime_set_autosuspend_delay(dev->dev, 5000);
pm_runtime_set_active(dev->dev);
pm_runtime_allow(dev->dev);
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
}
out:
if (r)
radeon_driver_unload_kms(dev);
return r;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
}
/**
* radeon_set_filp_rights - Set filp right.
*
* @dev: drm dev pointer
* @owner: drm file
* @applier: drm file
* @value: value
*
* Sets the filp rights for the device (all asics).
*/
static void radeon_set_filp_rights(struct drm_device *dev,
struct drm_file **owner,
struct drm_file *applier,
uint32_t *value)
{
struct radeon_device *rdev = dev->dev_private;
mutex_lock(&rdev->gem.mutex);
if (*value == 1) {
/* wants rights */
if (!*owner)
*owner = applier;
} else if (*value == 0) {
/* revokes rights */
if (*owner == applier)
*owner = NULL;
}
*value = *owner == applier ? 1 : 0;
mutex_unlock(&rdev->gem.mutex);
}
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
/*
* Userspace get information ioctl
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
*/
/**
* radeon_info_ioctl - answer a device specific request.
*
* @rdev: radeon device pointer
* @data: request object
* @filp: drm filp
*
* This function is used to pass device specific parameters to the userspace
* drivers. Examples include: pci device id, pipeline parms, tiling params,
* etc. (all asics).
* Returns 0 on success, -EINVAL on failure.
*/
static int radeon_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
{
struct radeon_device *rdev = dev->dev_private;
struct drm_radeon_info *info = data;
struct radeon_mode_info *minfo = &rdev->mode_info;
uint32_t *value, value_tmp, *value_ptr, value_size;
uint64_t value64;
struct drm_crtc *crtc;
int i, found;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
value_ptr = (uint32_t *)((unsigned long)info->value);
value = &value_tmp;
value_size = sizeof(uint32_t);
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
switch (info->request) {
case RADEON_INFO_DEVICE_ID:
*value = dev->pdev->device;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
break;
case RADEON_INFO_NUM_GB_PIPES:
*value = rdev->num_gb_pipes;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
break;
case RADEON_INFO_NUM_Z_PIPES:
*value = rdev->num_z_pipes;
break;
case RADEON_INFO_ACCEL_WORKING:
/* xf86-video-ati 6.13.0 relies on this being false for evergreen */
if ((rdev->family >= CHIP_CEDAR) && (rdev->family <= CHIP_HEMLOCK))
*value = false;
else
*value = rdev->accel_working;
break;
case RADEON_INFO_CRTC_FROM_ID:
if (copy_from_user(value, value_ptr, sizeof(uint32_t))) {
DRM_ERROR("copy_from_user %s:%u\n", __func__, __LINE__);
return -EFAULT;
}
for (i = 0, found = 0; i < rdev->num_crtc; i++) {
crtc = (struct drm_crtc *)minfo->crtcs[i];
if (crtc && crtc->base.id == *value) {
struct radeon_crtc *radeon_crtc = to_radeon_crtc(crtc);
*value = radeon_crtc->crtc_id;
found = 1;
break;
}
}
if (!found) {
DRM_DEBUG_KMS("unknown crtc id %d\n", *value);
return -EINVAL;
}
break;
case RADEON_INFO_ACCEL_WORKING2:
if (rdev->family == CHIP_HAWAII) {
if (rdev->accel_working) {
if (rdev->new_fw)
*value = 3;
else
*value = 2;
} else {
*value = 0;
}
} else {
*value = rdev->accel_working;
}
break;
case RADEON_INFO_TILING_CONFIG:
if (rdev->family >= CHIP_BONAIRE)
*value = rdev->config.cik.tile_config;
else if (rdev->family >= CHIP_TAHITI)
*value = rdev->config.si.tile_config;
else if (rdev->family >= CHIP_CAYMAN)
*value = rdev->config.cayman.tile_config;
else if (rdev->family >= CHIP_CEDAR)
*value = rdev->config.evergreen.tile_config;
else if (rdev->family >= CHIP_RV770)
*value = rdev->config.rv770.tile_config;
else if (rdev->family >= CHIP_R600)
*value = rdev->config.r600.tile_config;
else {
DRM_DEBUG_KMS("tiling config is r6xx+ only!\n");
return -EINVAL;
}
break;
case RADEON_INFO_WANT_HYPERZ:
/* The "value" here is both an input and output parameter.
* If the input value is 1, filp requests hyper-z access.
* If the input value is 0, filp revokes its hyper-z access.
*
* When returning, the value is 1 if filp owns hyper-z access,
* 0 otherwise. */
if (copy_from_user(value, value_ptr, sizeof(uint32_t))) {
DRM_ERROR("copy_from_user %s:%u\n", __func__, __LINE__);
return -EFAULT;
}
if (*value >= 2) {
DRM_DEBUG_KMS("WANT_HYPERZ: invalid value %d\n", *value);
return -EINVAL;
}
radeon_set_filp_rights(dev, &rdev->hyperz_filp, filp, value);
break;
case RADEON_INFO_WANT_CMASK:
/* The same logic as Hyper-Z. */
if (copy_from_user(value, value_ptr, sizeof(uint32_t))) {
DRM_ERROR("copy_from_user %s:%u\n", __func__, __LINE__);
return -EFAULT;
}
if (*value >= 2) {
DRM_DEBUG_KMS("WANT_CMASK: invalid value %d\n", *value);
return -EINVAL;
}
radeon_set_filp_rights(dev, &rdev->cmask_filp, filp, value);
break;
case RADEON_INFO_CLOCK_CRYSTAL_FREQ:
/* return clock value in KHz */
if (rdev->asic->get_xclk)
*value = radeon_get_xclk(rdev) * 10;
else
*value = rdev->clock.spll.reference_freq * 10;
break;
case RADEON_INFO_NUM_BACKENDS:
if (rdev->family >= CHIP_BONAIRE)
*value = rdev->config.cik.max_backends_per_se *
rdev->config.cik.max_shader_engines;
else if (rdev->family >= CHIP_TAHITI)
*value = rdev->config.si.max_backends_per_se *
rdev->config.si.max_shader_engines;
else if (rdev->family >= CHIP_CAYMAN)
*value = rdev->config.cayman.max_backends_per_se *
rdev->config.cayman.max_shader_engines;
else if (rdev->family >= CHIP_CEDAR)
*value = rdev->config.evergreen.max_backends;
else if (rdev->family >= CHIP_RV770)
*value = rdev->config.rv770.max_backends;
else if (rdev->family >= CHIP_R600)
*value = rdev->config.r600.max_backends;
else {
return -EINVAL;
}
break;
case RADEON_INFO_NUM_TILE_PIPES:
if (rdev->family >= CHIP_BONAIRE)
*value = rdev->config.cik.max_tile_pipes;
else if (rdev->family >= CHIP_TAHITI)
*value = rdev->config.si.max_tile_pipes;
else if (rdev->family >= CHIP_CAYMAN)
*value = rdev->config.cayman.max_tile_pipes;
else if (rdev->family >= CHIP_CEDAR)
*value = rdev->config.evergreen.max_tile_pipes;
else if (rdev->family >= CHIP_RV770)
*value = rdev->config.rv770.max_tile_pipes;
else if (rdev->family >= CHIP_R600)
*value = rdev->config.r600.max_tile_pipes;
else {
return -EINVAL;
}
break;
case RADEON_INFO_FUSION_GART_WORKING:
*value = 1;
break;
case RADEON_INFO_BACKEND_MAP:
if (rdev->family >= CHIP_BONAIRE)
*value = rdev->config.cik.backend_map;
else if (rdev->family >= CHIP_TAHITI)
*value = rdev->config.si.backend_map;
else if (rdev->family >= CHIP_CAYMAN)
*value = rdev->config.cayman.backend_map;
else if (rdev->family >= CHIP_CEDAR)
*value = rdev->config.evergreen.backend_map;
else if (rdev->family >= CHIP_RV770)
*value = rdev->config.rv770.backend_map;
else if (rdev->family >= CHIP_R600)
*value = rdev->config.r600.backend_map;
else {
return -EINVAL;
}
break;
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
case RADEON_INFO_VA_START:
/* this is where we report if vm is supported or not */
if (rdev->family < CHIP_CAYMAN)
return -EINVAL;
*value = RADEON_VA_RESERVED_SIZE;
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
break;
case RADEON_INFO_IB_VM_MAX_SIZE:
/* this is where we report if vm is supported or not */
if (rdev->family < CHIP_CAYMAN)
return -EINVAL;
*value = RADEON_IB_VM_MAX_SIZE;
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
break;
case RADEON_INFO_MAX_PIPES:
if (rdev->family >= CHIP_BONAIRE)
*value = rdev->config.cik.max_cu_per_sh;
else if (rdev->family >= CHIP_TAHITI)
*value = rdev->config.si.max_cu_per_sh;
else if (rdev->family >= CHIP_CAYMAN)
*value = rdev->config.cayman.max_pipes_per_simd;
else if (rdev->family >= CHIP_CEDAR)
*value = rdev->config.evergreen.max_pipes;
else if (rdev->family >= CHIP_RV770)
*value = rdev->config.rv770.max_pipes;
else if (rdev->family >= CHIP_R600)
*value = rdev->config.r600.max_pipes;
else {
return -EINVAL;
}
break;
case RADEON_INFO_TIMESTAMP:
if (rdev->family < CHIP_R600) {
DRM_DEBUG_KMS("timestamp is r6xx+ only!\n");
return -EINVAL;
}
value = (uint32_t*)&value64;
value_size = sizeof(uint64_t);
value64 = radeon_get_gpu_clock_counter(rdev);
break;
case RADEON_INFO_MAX_SE:
if (rdev->family >= CHIP_BONAIRE)
*value = rdev->config.cik.max_shader_engines;
else if (rdev->family >= CHIP_TAHITI)
*value = rdev->config.si.max_shader_engines;
else if (rdev->family >= CHIP_CAYMAN)
*value = rdev->config.cayman.max_shader_engines;
else if (rdev->family >= CHIP_CEDAR)
*value = rdev->config.evergreen.num_ses;
else
*value = 1;
break;
case RADEON_INFO_MAX_SH_PER_SE:
if (rdev->family >= CHIP_BONAIRE)
*value = rdev->config.cik.max_sh_per_se;
else if (rdev->family >= CHIP_TAHITI)
*value = rdev->config.si.max_sh_per_se;
else
return -EINVAL;
break;
case RADEON_INFO_FASTFB_WORKING:
*value = rdev->fastfb_working;
break;
case RADEON_INFO_RING_WORKING:
if (copy_from_user(value, value_ptr, sizeof(uint32_t))) {
DRM_ERROR("copy_from_user %s:%u\n", __func__, __LINE__);
return -EFAULT;
}
switch (*value) {
case RADEON_CS_RING_GFX:
case RADEON_CS_RING_COMPUTE:
*value = rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready;
break;
case RADEON_CS_RING_DMA:
*value = rdev->ring[R600_RING_TYPE_DMA_INDEX].ready;
*value |= rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX].ready;
break;
case RADEON_CS_RING_UVD:
*value = rdev->ring[R600_RING_TYPE_UVD_INDEX].ready;
break;
case RADEON_CS_RING_VCE:
*value = rdev->ring[TN_RING_TYPE_VCE1_INDEX].ready;
break;
default:
return -EINVAL;
}
break;
case RADEON_INFO_SI_TILE_MODE_ARRAY:
if (rdev->family >= CHIP_BONAIRE) {
value = rdev->config.cik.tile_mode_array;
value_size = sizeof(uint32_t)*32;
} else if (rdev->family >= CHIP_TAHITI) {
value = rdev->config.si.tile_mode_array;
value_size = sizeof(uint32_t)*32;
} else {
DRM_DEBUG_KMS("tile mode array is si+ only!\n");
return -EINVAL;
}
break;
case RADEON_INFO_CIK_MACROTILE_MODE_ARRAY:
if (rdev->family >= CHIP_BONAIRE) {
value = rdev->config.cik.macrotile_mode_array;
value_size = sizeof(uint32_t)*16;
} else {
DRM_DEBUG_KMS("macrotile mode array is cik+ only!\n");
return -EINVAL;
}
break;
case RADEON_INFO_SI_CP_DMA_COMPUTE:
*value = 1;
break;
case RADEON_INFO_SI_BACKEND_ENABLED_MASK:
if (rdev->family >= CHIP_BONAIRE) {
*value = rdev->config.cik.backend_enable_mask;
} else if (rdev->family >= CHIP_TAHITI) {
*value = rdev->config.si.backend_enable_mask;
} else {
DRM_DEBUG_KMS("BACKEND_ENABLED_MASK is si+ only!\n");
}
break;
case RADEON_INFO_MAX_SCLK:
if ((rdev->pm.pm_method == PM_METHOD_DPM) &&
rdev->pm.dpm_enabled)
*value = rdev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk * 10;
else
*value = rdev->pm.default_sclk * 10;
break;
case RADEON_INFO_VCE_FW_VERSION:
*value = rdev->vce.fw_version;
break;
case RADEON_INFO_VCE_FB_VERSION:
*value = rdev->vce.fb_version;
break;
case RADEON_INFO_NUM_BYTES_MOVED:
value = (uint32_t*)&value64;
value_size = sizeof(uint64_t);
value64 = atomic64_read(&rdev->num_bytes_moved);
break;
case RADEON_INFO_VRAM_USAGE:
value = (uint32_t*)&value64;
value_size = sizeof(uint64_t);
value64 = atomic64_read(&rdev->vram_usage);
break;
case RADEON_INFO_GTT_USAGE:
value = (uint32_t*)&value64;
value_size = sizeof(uint64_t);
value64 = atomic64_read(&rdev->gtt_usage);
break;
case RADEON_INFO_ACTIVE_CU_COUNT:
if (rdev->family >= CHIP_BONAIRE)
*value = rdev->config.cik.active_cus;
else if (rdev->family >= CHIP_TAHITI)
*value = rdev->config.si.active_cus;
else if (rdev->family >= CHIP_CAYMAN)
*value = rdev->config.cayman.active_simds;
else if (rdev->family >= CHIP_CEDAR)
*value = rdev->config.evergreen.active_simds;
else if (rdev->family >= CHIP_RV770)
*value = rdev->config.rv770.active_simds;
else if (rdev->family >= CHIP_R600)
*value = rdev->config.r600.active_simds;
else
*value = 1;
break;
case RADEON_INFO_CURRENT_GPU_TEMP:
/* get temperature in millidegrees C */
if (rdev->asic->pm.get_temperature)
*value = radeon_get_temperature(rdev);
else
*value = 0;
break;
case RADEON_INFO_CURRENT_GPU_SCLK:
/* get sclk in Mhz */
if (rdev->pm.dpm_enabled)
*value = radeon_dpm_get_current_sclk(rdev) / 100;
else
*value = rdev->pm.current_sclk / 100;
break;
case RADEON_INFO_CURRENT_GPU_MCLK:
/* get mclk in Mhz */
if (rdev->pm.dpm_enabled)
*value = radeon_dpm_get_current_mclk(rdev) / 100;
else
*value = rdev->pm.current_mclk / 100;
break;
case RADEON_INFO_READ_REG:
if (copy_from_user(value, value_ptr, sizeof(uint32_t))) {
DRM_ERROR("copy_from_user %s:%u\n", __func__, __LINE__);
return -EFAULT;
}
if (radeon_get_allowed_info_register(rdev, *value, value))
return -EINVAL;
break;
case RADEON_INFO_VA_UNMAP_WORKING:
*value = true;
break;
case RADEON_INFO_GPU_RESET_COUNTER:
*value = atomic_read(&rdev->gpu_reset_counter);
break;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
default:
DRM_DEBUG_KMS("Invalid request %d\n", info->request);
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
return -EINVAL;
}
if (copy_to_user(value_ptr, (char*)value, value_size)) {
DRM_ERROR("copy_to_user %s:%u\n", __func__, __LINE__);
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
return -EFAULT;
}
return 0;
}
/*
* Outdated mess for old drm with Xorg being in charge (void function now).
*/
/**
* radeon_driver_lastclose_kms - drm callback for last close
*
* @dev: drm dev pointer
*
* Switch vga_switcheroo state after last close (all asics).
*/
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
void radeon_driver_lastclose_kms(struct drm_device *dev)
{
drm_fb_helper_lastclose(dev);
vga_switcheroo: initial implementation (v15) Many new laptops now come with 2 gpus, one to be used for low power modes and one for gaming/on-ac applications. These GPUs are typically wired to the laptop panel and VGA ports via a multiplexer unit which is controlled via ACPI methods. 4 combinations of systems typically exist - with 2 ACPI methods. Intel/ATI - Lenovo W500/T500 - use ATPX ACPI method ATI/ATI - some ASUS - use ATPX ACPI Method Intel/Nvidia - - use _DSM ACPI method Nvidia/Nvidia - - use _DSM ACPI method. TODO: This patch adds support for the ATPX method and initial bits for the _DSM methods that need to written by someone with access to the hardware. Add a proper non-debugfs interface - need to get some proper testing first. v2: add power up/down support for both devices on W500 puts i915/radeon into D3 and cuts power to radeon. v3: redo probing methods, no DMI list, drm devices call to register with switcheroo, it tries to find an ATPX method on any device and once there is two devices + ATPX it inits the switcher. v4: ATPX msg handling using buffers - should work on more machines v5: rearchitect after more mjg59 discussion - move ATPX handling to radeon driver. v6: add file headers + initial nouveau bits (to be filled out). v7: merge delayed switcher code. v8: avoid suspend/resume of gpu that is off v9: rearchitect - mjg59 is always right. - move all ATPX code to radeon, should allow simpler DSM also proper ATRM handling v10: add ATRM support for radeon BIOS, add mutex to lock vgasr_priv v11: fix bug in resuming Intel for 2nd time. v12: start fixing up nvidia code blindly. v13: blindly guess at finishing nvidia code v14: remove radeon audio hacks - fix up intel resume more like upstream v15: clean up printks + remove unnecessary igd/dis pointers mount debugfs /sys/kernel/debug/vgaswitcheroo/switch - should exist if ATPX detected + 2 cards. DIS - immediate change to discrete IGD - immediate change to IGD DDIS - delayed change to discrete DIGD - delayed change to IGD ON - turn on not in use OFF - turn off not in use Tested on W500 (Intel/ATI) and T500 (Intel/ATI) Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-02-01 12:38:10 +07:00
vga_switcheroo_process_delayed_switch();
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
}
/**
* radeon_driver_open_kms - drm callback for open
*
* @dev: drm dev pointer
* @file_priv: drm file
*
* On device open, init vm on cayman+ (all asics).
* Returns 0 on success, error on failure.
*/
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
int radeon_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
{
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
struct radeon_device *rdev = dev->dev_private;
int r;
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
file_priv->driver_priv = NULL;
r = pm_runtime_get_sync(dev->dev);
if (r < 0)
return r;
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
/* new gpu have virtual address space support */
if (rdev->family >= CHIP_CAYMAN) {
struct radeon_fpriv *fpriv;
struct radeon_vm *vm;
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
fpriv = kzalloc(sizeof(*fpriv), GFP_KERNEL);
if (unlikely(!fpriv)) {
r = -ENOMEM;
goto out_suspend;
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
}
if (rdev->accel_working) {
vm = &fpriv->vm;
r = radeon_vm_init(rdev, vm);
if (r) {
kfree(fpriv);
goto out_suspend;
}
r = radeon_bo_reserve(rdev->ring_tmp_bo.bo, false);
if (r) {
radeon_vm_fini(rdev, vm);
kfree(fpriv);
goto out_suspend;
}
/* map the ib pool buffer read only into
* virtual address space */
vm->ib_bo_va = radeon_vm_bo_add(rdev, vm,
rdev->ring_tmp_bo.bo);
r = radeon_vm_bo_set_addr(rdev, vm->ib_bo_va,
RADEON_VA_IB_OFFSET,
RADEON_VM_PAGE_READABLE |
RADEON_VM_PAGE_SNOOPED);
if (r) {
radeon_vm_fini(rdev, vm);
kfree(fpriv);
goto out_suspend;
}
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
}
file_priv->driver_priv = fpriv;
}
out_suspend:
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
return r;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
}
/**
* radeon_driver_postclose_kms - drm callback for post close
*
* @dev: drm dev pointer
* @file_priv: drm file
*
* On device close, tear down hyperz and cmask filps on r1xx-r5xx
* (all asics). And tear down vm on cayman+ (all asics).
*/
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
void radeon_driver_postclose_kms(struct drm_device *dev,
struct drm_file *file_priv)
{
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
struct radeon_device *rdev = dev->dev_private;
pm_runtime_get_sync(dev->dev);
mutex_lock(&rdev->gem.mutex);
if (rdev->hyperz_filp == file_priv)
rdev->hyperz_filp = NULL;
if (rdev->cmask_filp == file_priv)
rdev->cmask_filp = NULL;
mutex_unlock(&rdev->gem.mutex);
radeon_uvd_free_handles(rdev, file_priv);
radeon_vce_free_handles(rdev, file_priv);
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
/* new gpu have virtual address space support */
if (rdev->family >= CHIP_CAYMAN && file_priv->driver_priv) {
struct radeon_fpriv *fpriv = file_priv->driver_priv;
struct radeon_vm *vm = &fpriv->vm;
int r;
if (rdev->accel_working) {
r = radeon_bo_reserve(rdev->ring_tmp_bo.bo, false);
if (!r) {
if (vm->ib_bo_va)
radeon_vm_bo_rmv(rdev, vm->ib_bo_va);
radeon_bo_unreserve(rdev->ring_tmp_bo.bo);
}
radeon_vm_fini(rdev, vm);
}
drm/radeon: GPU virtual memory support v22 Virtual address space are per drm client (opener of /dev/drm). Client are in charge of virtual address space, they need to map bo into it by calling DRM_RADEON_GEM_VA ioctl. First 16M of virtual address space is reserved by the kernel. Once using 2 level page table we should be able to have a small vram memory footprint for each pt (there would be one pt for all gart, one for all vram and then one first level for each virtual address space). Plan include using the sub allocator for a common vm page table area and using memcpy to copy vm page table in & out. Or use a gart object and copy things in & out using dma. v2: agd5f fixes: - Add vram base offset for vram pages. The GPU physical address of a vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete cards and the physical bus address of the stolen memory on integrated chips. - VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1 v3: agd5f: - integrate with the semaphore/multi-ring stuff v4: - rebase on top ttm dma & multi-ring stuff - userspace is now in charge of the address space - no more specific cs vm ioctl, instead cs ioctl has a new chunk v5: - properly handle mem == NULL case from move_notify callback - fix the vm cleanup path v6: - fix update of page table to only happen on valid mem placement v7: - add tlb flush for each vm context - add flags to define mapping property (readable, writeable, snooped) - make ring id implicit from ib->fence->ring, up to each asic callback to then do ring specific scheduling if vm ib scheduling function v8: - add query for ib limit and kernel reserved virtual space - rename vm->size to max_pfn (maximum number of page) - update gem_va ioctl to also allow unmap operation - bump kernel version to allow userspace to query for vm support v9: - rebuild page table only when bind and incrementaly depending on bo referenced by cs and that have been moved - allow virtual address space to grow - use sa allocator for vram page table - return invalid when querying vm limit on non cayman GPU - dump vm fault register on lockup v10: agd5f: - Move the vm schedule_ib callback to a standalone function, remove the callback and use the existing ib_execute callback for VM IBs. v11: - rebase on top of lastest Linus v12: agd5f: - remove spurious backslash - set IB vm_id to 0 in radeon_ib_get() v13: agd5f: - fix handling of RADEON_CHUNK_ID_FLAGS v14: - fix va destruction - fix suspend resume - forbid bo to have several different va in same vm v15: - rebase v16: - cleanup left over of vm init/fini v17: agd5f: - cs checker v18: agd5f: - reworks the CS ioctl to better support multiple rings and VM. Rather than adding a new chunk id for VM, just re-use the IB chunk id and add a new flags for VM mode. Also define additional dwords for the flags chunk id to define the what ring we want to use (gfx, compute, uvd, etc.) and the priority. v19: - fix cs fini in weird case of no ib - semi working flush fix for ni - rebase on top of sa allocator changes v20: agd5f: - further CS ioctl cleanups from Christian's comments v21: agd5f: - integrate CS checker improvements v22: agd5f: - final cleanups for release, only allow VM CS on cayman Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2012-01-06 10:11:05 +07:00
kfree(fpriv);
file_priv->driver_priv = NULL;
}
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
}
/*
* VBlank related functions.
*/
/**
* radeon_get_vblank_counter_kms - get frame count
*
* @dev: drm dev pointer
* @pipe: crtc to get the frame count from
*
* Gets the frame count on the requested crtc (all asics).
* Returns frame count on success, -EINVAL on failure.
*/
u32 radeon_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe)
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
{
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2) commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core more fragile to drivers which don't update hw vblank counters and vblank timestamps in sync with firing of the vblank irq and essentially at leading edge of vblank. This exposed a problem with radeon-kms/amdgpu-kms which do not satisfy above requirements: The vblank irq fires a few scanlines before start of vblank, but programmed pageflips complete at start of vblank and vblank timestamps update at start of vblank, whereas the hw vblank counter increments only later, at start of vsync. This leads to problems like off by one errors for vblank counter updates, vblank counters apparently going backwards or vblank timestamps apparently having time going backwards. The net result is stuttering of graphics in games, or little hangs, as well as total failure of timing sensitive applications. See bug #93147 for an example of the regression on Linux 4.4-rc: https://bugs.freedesktop.org/show_bug.cgi?id=93147 This patch tries to align all above events better from the viewpoint of the drm core / of external callers to fix the problem: 1. The apparent start of vblank is shifted a few scanlines earlier, so the vblank irq now always happens after start of this extended vblank interval and thereby drm_update_vblank_count() always samples the updated vblank count and timestamp of the new vblank interval. To achieve this, the reporting of scanout positions by radeon_get_crtc_scanoutpos() now operates as if the vblank starts radeon_crtc->lb_vblank_lead_lines before the real start of the hw vblank interval. This means that the vblank timestamps which are based on these scanout positions will now update at this earlier start of vblank. 2. The driver->get_vblank_counter() function will bump the returned vblank count as read from the hw by +1 if the query happens after the shifted earlier start of the vblank, but before the real hw increment at start of vsync, so the counter appears to increment at start of vblank in sync with the timestamp update. 3. Calls from vblank irq-context and regular non-irq calls are now treated identical, always simulating the shifted vblank start, to avoid inconsistent results for queries happening from vblank irq vs. happening from drm_vblank_enable() or vblank_disable_fn(). 4. The radeon_flip_work_func will delay mmio programming a pageflip until the start of the real vblank iff it happens to execute inside the shifted earlier start of the vblank, so pageflips now also appear to execute at start of the shifted vblank, in sync with vblank counter and timestamp updates. This to avoid some races between updates of vblank count and timestamps that are used for swap scheduling and pageflip execution which could cause pageflips to execute before the scheduled target vblank. The lb_vblank_lead_lines "fudge" value is calculated as the size of the display controllers line buffer in scanlines for the given video mode: Vblank irq's are triggered by the line buffer logic when the line buffer refill for a video frame ends, ie. when the line buffer source read position enters the hw vblank. This means that a vblank irq could fire at most as many scanlines before the current reported scanout position of the crtc timing generator as the number of scanlines the line buffer can maximally hold for a given video mode. This patch has been successfully tested on a RV730 card with DCE-3 display engine and on a evergreen card with DCE-4 display engine, in single-display and dual-display configuration, with different video modes. A similar patch is needed for amdgpu-kms to fix the same problem. Limitations: - Line buffer sizes in pixels are hard-coded on < DCE-4 to a value i just guessed to be high enough to work ok, lacking info on the true sizes atm. Fixes: fdo#93147 Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Michel Dänzer <michel.daenzer@amd.com> Cc: Harry Wentland <Harry.Wentland@amd.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net> (v2) Refine radeon_flip_work_func() for better efficiency: In radeon_flip_work_func, replace the busy waiting udelay(5) with event lock held by a more performance and energy efficient usleep_range() until at least predicted true start of hw vblank, with some slack for scheduler happiness. Release the event lock during waits to not delay other outputs in doing their stuff, as the waiting can last up to 200 usecs in some cases. Retested on DCE-3 and DCE-4 to verify it still works nicely. (v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-26 02:14:31 +07:00
int vpos, hpos, stat;
u32 count;
struct radeon_device *rdev = dev->dev_private;
if (pipe >= rdev->num_crtc) {
DRM_ERROR("Invalid crtc %u\n", pipe);
return -EINVAL;
}
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2) commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core more fragile to drivers which don't update hw vblank counters and vblank timestamps in sync with firing of the vblank irq and essentially at leading edge of vblank. This exposed a problem with radeon-kms/amdgpu-kms which do not satisfy above requirements: The vblank irq fires a few scanlines before start of vblank, but programmed pageflips complete at start of vblank and vblank timestamps update at start of vblank, whereas the hw vblank counter increments only later, at start of vsync. This leads to problems like off by one errors for vblank counter updates, vblank counters apparently going backwards or vblank timestamps apparently having time going backwards. The net result is stuttering of graphics in games, or little hangs, as well as total failure of timing sensitive applications. See bug #93147 for an example of the regression on Linux 4.4-rc: https://bugs.freedesktop.org/show_bug.cgi?id=93147 This patch tries to align all above events better from the viewpoint of the drm core / of external callers to fix the problem: 1. The apparent start of vblank is shifted a few scanlines earlier, so the vblank irq now always happens after start of this extended vblank interval and thereby drm_update_vblank_count() always samples the updated vblank count and timestamp of the new vblank interval. To achieve this, the reporting of scanout positions by radeon_get_crtc_scanoutpos() now operates as if the vblank starts radeon_crtc->lb_vblank_lead_lines before the real start of the hw vblank interval. This means that the vblank timestamps which are based on these scanout positions will now update at this earlier start of vblank. 2. The driver->get_vblank_counter() function will bump the returned vblank count as read from the hw by +1 if the query happens after the shifted earlier start of the vblank, but before the real hw increment at start of vsync, so the counter appears to increment at start of vblank in sync with the timestamp update. 3. Calls from vblank irq-context and regular non-irq calls are now treated identical, always simulating the shifted vblank start, to avoid inconsistent results for queries happening from vblank irq vs. happening from drm_vblank_enable() or vblank_disable_fn(). 4. The radeon_flip_work_func will delay mmio programming a pageflip until the start of the real vblank iff it happens to execute inside the shifted earlier start of the vblank, so pageflips now also appear to execute at start of the shifted vblank, in sync with vblank counter and timestamp updates. This to avoid some races between updates of vblank count and timestamps that are used for swap scheduling and pageflip execution which could cause pageflips to execute before the scheduled target vblank. The lb_vblank_lead_lines "fudge" value is calculated as the size of the display controllers line buffer in scanlines for the given video mode: Vblank irq's are triggered by the line buffer logic when the line buffer refill for a video frame ends, ie. when the line buffer source read position enters the hw vblank. This means that a vblank irq could fire at most as many scanlines before the current reported scanout position of the crtc timing generator as the number of scanlines the line buffer can maximally hold for a given video mode. This patch has been successfully tested on a RV730 card with DCE-3 display engine and on a evergreen card with DCE-4 display engine, in single-display and dual-display configuration, with different video modes. A similar patch is needed for amdgpu-kms to fix the same problem. Limitations: - Line buffer sizes in pixels are hard-coded on < DCE-4 to a value i just guessed to be high enough to work ok, lacking info on the true sizes atm. Fixes: fdo#93147 Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Michel Dänzer <michel.daenzer@amd.com> Cc: Harry Wentland <Harry.Wentland@amd.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net> (v2) Refine radeon_flip_work_func() for better efficiency: In radeon_flip_work_func, replace the busy waiting udelay(5) with event lock held by a more performance and energy efficient usleep_range() until at least predicted true start of hw vblank, with some slack for scheduler happiness. Release the event lock during waits to not delay other outputs in doing their stuff, as the waiting can last up to 200 usecs in some cases. Retested on DCE-3 and DCE-4 to verify it still works nicely. (v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-26 02:14:31 +07:00
/* The hw increments its frame counter at start of vsync, not at start
* of vblank, as is required by DRM core vblank counter handling.
* Cook the hw count here to make it appear to the caller as if it
* incremented at start of vblank. We measure distance to start of
* vblank in vpos. vpos therefore will be >= 0 between start of vblank
* and start of vsync, so vpos >= 0 means to bump the hw frame counter
* result by 1 to give the proper appearance to caller.
*/
if (rdev->mode_info.crtcs[pipe]) {
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2) commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core more fragile to drivers which don't update hw vblank counters and vblank timestamps in sync with firing of the vblank irq and essentially at leading edge of vblank. This exposed a problem with radeon-kms/amdgpu-kms which do not satisfy above requirements: The vblank irq fires a few scanlines before start of vblank, but programmed pageflips complete at start of vblank and vblank timestamps update at start of vblank, whereas the hw vblank counter increments only later, at start of vsync. This leads to problems like off by one errors for vblank counter updates, vblank counters apparently going backwards or vblank timestamps apparently having time going backwards. The net result is stuttering of graphics in games, or little hangs, as well as total failure of timing sensitive applications. See bug #93147 for an example of the regression on Linux 4.4-rc: https://bugs.freedesktop.org/show_bug.cgi?id=93147 This patch tries to align all above events better from the viewpoint of the drm core / of external callers to fix the problem: 1. The apparent start of vblank is shifted a few scanlines earlier, so the vblank irq now always happens after start of this extended vblank interval and thereby drm_update_vblank_count() always samples the updated vblank count and timestamp of the new vblank interval. To achieve this, the reporting of scanout positions by radeon_get_crtc_scanoutpos() now operates as if the vblank starts radeon_crtc->lb_vblank_lead_lines before the real start of the hw vblank interval. This means that the vblank timestamps which are based on these scanout positions will now update at this earlier start of vblank. 2. The driver->get_vblank_counter() function will bump the returned vblank count as read from the hw by +1 if the query happens after the shifted earlier start of the vblank, but before the real hw increment at start of vsync, so the counter appears to increment at start of vblank in sync with the timestamp update. 3. Calls from vblank irq-context and regular non-irq calls are now treated identical, always simulating the shifted vblank start, to avoid inconsistent results for queries happening from vblank irq vs. happening from drm_vblank_enable() or vblank_disable_fn(). 4. The radeon_flip_work_func will delay mmio programming a pageflip until the start of the real vblank iff it happens to execute inside the shifted earlier start of the vblank, so pageflips now also appear to execute at start of the shifted vblank, in sync with vblank counter and timestamp updates. This to avoid some races between updates of vblank count and timestamps that are used for swap scheduling and pageflip execution which could cause pageflips to execute before the scheduled target vblank. The lb_vblank_lead_lines "fudge" value is calculated as the size of the display controllers line buffer in scanlines for the given video mode: Vblank irq's are triggered by the line buffer logic when the line buffer refill for a video frame ends, ie. when the line buffer source read position enters the hw vblank. This means that a vblank irq could fire at most as many scanlines before the current reported scanout position of the crtc timing generator as the number of scanlines the line buffer can maximally hold for a given video mode. This patch has been successfully tested on a RV730 card with DCE-3 display engine and on a evergreen card with DCE-4 display engine, in single-display and dual-display configuration, with different video modes. A similar patch is needed for amdgpu-kms to fix the same problem. Limitations: - Line buffer sizes in pixels are hard-coded on < DCE-4 to a value i just guessed to be high enough to work ok, lacking info on the true sizes atm. Fixes: fdo#93147 Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Michel Dänzer <michel.daenzer@amd.com> Cc: Harry Wentland <Harry.Wentland@amd.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net> (v2) Refine radeon_flip_work_func() for better efficiency: In radeon_flip_work_func, replace the busy waiting udelay(5) with event lock held by a more performance and energy efficient usleep_range() until at least predicted true start of hw vblank, with some slack for scheduler happiness. Release the event lock during waits to not delay other outputs in doing their stuff, as the waiting can last up to 200 usecs in some cases. Retested on DCE-3 and DCE-4 to verify it still works nicely. (v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-26 02:14:31 +07:00
/* Repeat readout if needed to provide stable result if
* we cross start of vsync during the queries.
*/
do {
count = radeon_get_vblank_counter(rdev, pipe);
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2) commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core more fragile to drivers which don't update hw vblank counters and vblank timestamps in sync with firing of the vblank irq and essentially at leading edge of vblank. This exposed a problem with radeon-kms/amdgpu-kms which do not satisfy above requirements: The vblank irq fires a few scanlines before start of vblank, but programmed pageflips complete at start of vblank and vblank timestamps update at start of vblank, whereas the hw vblank counter increments only later, at start of vsync. This leads to problems like off by one errors for vblank counter updates, vblank counters apparently going backwards or vblank timestamps apparently having time going backwards. The net result is stuttering of graphics in games, or little hangs, as well as total failure of timing sensitive applications. See bug #93147 for an example of the regression on Linux 4.4-rc: https://bugs.freedesktop.org/show_bug.cgi?id=93147 This patch tries to align all above events better from the viewpoint of the drm core / of external callers to fix the problem: 1. The apparent start of vblank is shifted a few scanlines earlier, so the vblank irq now always happens after start of this extended vblank interval and thereby drm_update_vblank_count() always samples the updated vblank count and timestamp of the new vblank interval. To achieve this, the reporting of scanout positions by radeon_get_crtc_scanoutpos() now operates as if the vblank starts radeon_crtc->lb_vblank_lead_lines before the real start of the hw vblank interval. This means that the vblank timestamps which are based on these scanout positions will now update at this earlier start of vblank. 2. The driver->get_vblank_counter() function will bump the returned vblank count as read from the hw by +1 if the query happens after the shifted earlier start of the vblank, but before the real hw increment at start of vsync, so the counter appears to increment at start of vblank in sync with the timestamp update. 3. Calls from vblank irq-context and regular non-irq calls are now treated identical, always simulating the shifted vblank start, to avoid inconsistent results for queries happening from vblank irq vs. happening from drm_vblank_enable() or vblank_disable_fn(). 4. The radeon_flip_work_func will delay mmio programming a pageflip until the start of the real vblank iff it happens to execute inside the shifted earlier start of the vblank, so pageflips now also appear to execute at start of the shifted vblank, in sync with vblank counter and timestamp updates. This to avoid some races between updates of vblank count and timestamps that are used for swap scheduling and pageflip execution which could cause pageflips to execute before the scheduled target vblank. The lb_vblank_lead_lines "fudge" value is calculated as the size of the display controllers line buffer in scanlines for the given video mode: Vblank irq's are triggered by the line buffer logic when the line buffer refill for a video frame ends, ie. when the line buffer source read position enters the hw vblank. This means that a vblank irq could fire at most as many scanlines before the current reported scanout position of the crtc timing generator as the number of scanlines the line buffer can maximally hold for a given video mode. This patch has been successfully tested on a RV730 card with DCE-3 display engine and on a evergreen card with DCE-4 display engine, in single-display and dual-display configuration, with different video modes. A similar patch is needed for amdgpu-kms to fix the same problem. Limitations: - Line buffer sizes in pixels are hard-coded on < DCE-4 to a value i just guessed to be high enough to work ok, lacking info on the true sizes atm. Fixes: fdo#93147 Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Michel Dänzer <michel.daenzer@amd.com> Cc: Harry Wentland <Harry.Wentland@amd.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net> (v2) Refine radeon_flip_work_func() for better efficiency: In radeon_flip_work_func, replace the busy waiting udelay(5) with event lock held by a more performance and energy efficient usleep_range() until at least predicted true start of hw vblank, with some slack for scheduler happiness. Release the event lock during waits to not delay other outputs in doing their stuff, as the waiting can last up to 200 usecs in some cases. Retested on DCE-3 and DCE-4 to verify it still works nicely. (v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-26 02:14:31 +07:00
/* Ask radeon_get_crtc_scanoutpos to return vpos as
* distance to start of vblank, instead of regular
* vertical scanout pos.
*/
stat = radeon_get_crtc_scanoutpos(
dev, pipe, GET_DISTANCE_TO_VBLANKSTART,
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2) commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core more fragile to drivers which don't update hw vblank counters and vblank timestamps in sync with firing of the vblank irq and essentially at leading edge of vblank. This exposed a problem with radeon-kms/amdgpu-kms which do not satisfy above requirements: The vblank irq fires a few scanlines before start of vblank, but programmed pageflips complete at start of vblank and vblank timestamps update at start of vblank, whereas the hw vblank counter increments only later, at start of vsync. This leads to problems like off by one errors for vblank counter updates, vblank counters apparently going backwards or vblank timestamps apparently having time going backwards. The net result is stuttering of graphics in games, or little hangs, as well as total failure of timing sensitive applications. See bug #93147 for an example of the regression on Linux 4.4-rc: https://bugs.freedesktop.org/show_bug.cgi?id=93147 This patch tries to align all above events better from the viewpoint of the drm core / of external callers to fix the problem: 1. The apparent start of vblank is shifted a few scanlines earlier, so the vblank irq now always happens after start of this extended vblank interval and thereby drm_update_vblank_count() always samples the updated vblank count and timestamp of the new vblank interval. To achieve this, the reporting of scanout positions by radeon_get_crtc_scanoutpos() now operates as if the vblank starts radeon_crtc->lb_vblank_lead_lines before the real start of the hw vblank interval. This means that the vblank timestamps which are based on these scanout positions will now update at this earlier start of vblank. 2. The driver->get_vblank_counter() function will bump the returned vblank count as read from the hw by +1 if the query happens after the shifted earlier start of the vblank, but before the real hw increment at start of vsync, so the counter appears to increment at start of vblank in sync with the timestamp update. 3. Calls from vblank irq-context and regular non-irq calls are now treated identical, always simulating the shifted vblank start, to avoid inconsistent results for queries happening from vblank irq vs. happening from drm_vblank_enable() or vblank_disable_fn(). 4. The radeon_flip_work_func will delay mmio programming a pageflip until the start of the real vblank iff it happens to execute inside the shifted earlier start of the vblank, so pageflips now also appear to execute at start of the shifted vblank, in sync with vblank counter and timestamp updates. This to avoid some races between updates of vblank count and timestamps that are used for swap scheduling and pageflip execution which could cause pageflips to execute before the scheduled target vblank. The lb_vblank_lead_lines "fudge" value is calculated as the size of the display controllers line buffer in scanlines for the given video mode: Vblank irq's are triggered by the line buffer logic when the line buffer refill for a video frame ends, ie. when the line buffer source read position enters the hw vblank. This means that a vblank irq could fire at most as many scanlines before the current reported scanout position of the crtc timing generator as the number of scanlines the line buffer can maximally hold for a given video mode. This patch has been successfully tested on a RV730 card with DCE-3 display engine and on a evergreen card with DCE-4 display engine, in single-display and dual-display configuration, with different video modes. A similar patch is needed for amdgpu-kms to fix the same problem. Limitations: - Line buffer sizes in pixels are hard-coded on < DCE-4 to a value i just guessed to be high enough to work ok, lacking info on the true sizes atm. Fixes: fdo#93147 Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Michel Dänzer <michel.daenzer@amd.com> Cc: Harry Wentland <Harry.Wentland@amd.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net> (v2) Refine radeon_flip_work_func() for better efficiency: In radeon_flip_work_func, replace the busy waiting udelay(5) with event lock held by a more performance and energy efficient usleep_range() until at least predicted true start of hw vblank, with some slack for scheduler happiness. Release the event lock during waits to not delay other outputs in doing their stuff, as the waiting can last up to 200 usecs in some cases. Retested on DCE-3 and DCE-4 to verify it still works nicely. (v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-26 02:14:31 +07:00
&vpos, &hpos, NULL, NULL,
&rdev->mode_info.crtcs[pipe]->base.hwmode);
} while (count != radeon_get_vblank_counter(rdev, pipe));
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2) commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core more fragile to drivers which don't update hw vblank counters and vblank timestamps in sync with firing of the vblank irq and essentially at leading edge of vblank. This exposed a problem with radeon-kms/amdgpu-kms which do not satisfy above requirements: The vblank irq fires a few scanlines before start of vblank, but programmed pageflips complete at start of vblank and vblank timestamps update at start of vblank, whereas the hw vblank counter increments only later, at start of vsync. This leads to problems like off by one errors for vblank counter updates, vblank counters apparently going backwards or vblank timestamps apparently having time going backwards. The net result is stuttering of graphics in games, or little hangs, as well as total failure of timing sensitive applications. See bug #93147 for an example of the regression on Linux 4.4-rc: https://bugs.freedesktop.org/show_bug.cgi?id=93147 This patch tries to align all above events better from the viewpoint of the drm core / of external callers to fix the problem: 1. The apparent start of vblank is shifted a few scanlines earlier, so the vblank irq now always happens after start of this extended vblank interval and thereby drm_update_vblank_count() always samples the updated vblank count and timestamp of the new vblank interval. To achieve this, the reporting of scanout positions by radeon_get_crtc_scanoutpos() now operates as if the vblank starts radeon_crtc->lb_vblank_lead_lines before the real start of the hw vblank interval. This means that the vblank timestamps which are based on these scanout positions will now update at this earlier start of vblank. 2. The driver->get_vblank_counter() function will bump the returned vblank count as read from the hw by +1 if the query happens after the shifted earlier start of the vblank, but before the real hw increment at start of vsync, so the counter appears to increment at start of vblank in sync with the timestamp update. 3. Calls from vblank irq-context and regular non-irq calls are now treated identical, always simulating the shifted vblank start, to avoid inconsistent results for queries happening from vblank irq vs. happening from drm_vblank_enable() or vblank_disable_fn(). 4. The radeon_flip_work_func will delay mmio programming a pageflip until the start of the real vblank iff it happens to execute inside the shifted earlier start of the vblank, so pageflips now also appear to execute at start of the shifted vblank, in sync with vblank counter and timestamp updates. This to avoid some races between updates of vblank count and timestamps that are used for swap scheduling and pageflip execution which could cause pageflips to execute before the scheduled target vblank. The lb_vblank_lead_lines "fudge" value is calculated as the size of the display controllers line buffer in scanlines for the given video mode: Vblank irq's are triggered by the line buffer logic when the line buffer refill for a video frame ends, ie. when the line buffer source read position enters the hw vblank. This means that a vblank irq could fire at most as many scanlines before the current reported scanout position of the crtc timing generator as the number of scanlines the line buffer can maximally hold for a given video mode. This patch has been successfully tested on a RV730 card with DCE-3 display engine and on a evergreen card with DCE-4 display engine, in single-display and dual-display configuration, with different video modes. A similar patch is needed for amdgpu-kms to fix the same problem. Limitations: - Line buffer sizes in pixels are hard-coded on < DCE-4 to a value i just guessed to be high enough to work ok, lacking info on the true sizes atm. Fixes: fdo#93147 Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Michel Dänzer <michel.daenzer@amd.com> Cc: Harry Wentland <Harry.Wentland@amd.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net> (v2) Refine radeon_flip_work_func() for better efficiency: In radeon_flip_work_func, replace the busy waiting udelay(5) with event lock held by a more performance and energy efficient usleep_range() until at least predicted true start of hw vblank, with some slack for scheduler happiness. Release the event lock during waits to not delay other outputs in doing their stuff, as the waiting can last up to 200 usecs in some cases. Retested on DCE-3 and DCE-4 to verify it still works nicely. (v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-26 02:14:31 +07:00
if (((stat & (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE)) !=
(DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE))) {
DRM_DEBUG_VBL("Query failed! stat %d\n", stat);
}
else {
DRM_DEBUG_VBL("crtc %u: dist from vblank start %d\n",
pipe, vpos);
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2) commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core more fragile to drivers which don't update hw vblank counters and vblank timestamps in sync with firing of the vblank irq and essentially at leading edge of vblank. This exposed a problem with radeon-kms/amdgpu-kms which do not satisfy above requirements: The vblank irq fires a few scanlines before start of vblank, but programmed pageflips complete at start of vblank and vblank timestamps update at start of vblank, whereas the hw vblank counter increments only later, at start of vsync. This leads to problems like off by one errors for vblank counter updates, vblank counters apparently going backwards or vblank timestamps apparently having time going backwards. The net result is stuttering of graphics in games, or little hangs, as well as total failure of timing sensitive applications. See bug #93147 for an example of the regression on Linux 4.4-rc: https://bugs.freedesktop.org/show_bug.cgi?id=93147 This patch tries to align all above events better from the viewpoint of the drm core / of external callers to fix the problem: 1. The apparent start of vblank is shifted a few scanlines earlier, so the vblank irq now always happens after start of this extended vblank interval and thereby drm_update_vblank_count() always samples the updated vblank count and timestamp of the new vblank interval. To achieve this, the reporting of scanout positions by radeon_get_crtc_scanoutpos() now operates as if the vblank starts radeon_crtc->lb_vblank_lead_lines before the real start of the hw vblank interval. This means that the vblank timestamps which are based on these scanout positions will now update at this earlier start of vblank. 2. The driver->get_vblank_counter() function will bump the returned vblank count as read from the hw by +1 if the query happens after the shifted earlier start of the vblank, but before the real hw increment at start of vsync, so the counter appears to increment at start of vblank in sync with the timestamp update. 3. Calls from vblank irq-context and regular non-irq calls are now treated identical, always simulating the shifted vblank start, to avoid inconsistent results for queries happening from vblank irq vs. happening from drm_vblank_enable() or vblank_disable_fn(). 4. The radeon_flip_work_func will delay mmio programming a pageflip until the start of the real vblank iff it happens to execute inside the shifted earlier start of the vblank, so pageflips now also appear to execute at start of the shifted vblank, in sync with vblank counter and timestamp updates. This to avoid some races between updates of vblank count and timestamps that are used for swap scheduling and pageflip execution which could cause pageflips to execute before the scheduled target vblank. The lb_vblank_lead_lines "fudge" value is calculated as the size of the display controllers line buffer in scanlines for the given video mode: Vblank irq's are triggered by the line buffer logic when the line buffer refill for a video frame ends, ie. when the line buffer source read position enters the hw vblank. This means that a vblank irq could fire at most as many scanlines before the current reported scanout position of the crtc timing generator as the number of scanlines the line buffer can maximally hold for a given video mode. This patch has been successfully tested on a RV730 card with DCE-3 display engine and on a evergreen card with DCE-4 display engine, in single-display and dual-display configuration, with different video modes. A similar patch is needed for amdgpu-kms to fix the same problem. Limitations: - Line buffer sizes in pixels are hard-coded on < DCE-4 to a value i just guessed to be high enough to work ok, lacking info on the true sizes atm. Fixes: fdo#93147 Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Michel Dänzer <michel.daenzer@amd.com> Cc: Harry Wentland <Harry.Wentland@amd.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net> (v2) Refine radeon_flip_work_func() for better efficiency: In radeon_flip_work_func, replace the busy waiting udelay(5) with event lock held by a more performance and energy efficient usleep_range() until at least predicted true start of hw vblank, with some slack for scheduler happiness. Release the event lock during waits to not delay other outputs in doing their stuff, as the waiting can last up to 200 usecs in some cases. Retested on DCE-3 and DCE-4 to verify it still works nicely. (v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-26 02:14:31 +07:00
/* Bump counter if we are at >= leading edge of vblank,
* but before vsync where vpos would turn negative and
* the hw counter really increments.
*/
if (vpos >= 0)
count++;
}
}
else {
/* Fallback to use value as is. */
count = radeon_get_vblank_counter(rdev, pipe);
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2) commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core more fragile to drivers which don't update hw vblank counters and vblank timestamps in sync with firing of the vblank irq and essentially at leading edge of vblank. This exposed a problem with radeon-kms/amdgpu-kms which do not satisfy above requirements: The vblank irq fires a few scanlines before start of vblank, but programmed pageflips complete at start of vblank and vblank timestamps update at start of vblank, whereas the hw vblank counter increments only later, at start of vsync. This leads to problems like off by one errors for vblank counter updates, vblank counters apparently going backwards or vblank timestamps apparently having time going backwards. The net result is stuttering of graphics in games, or little hangs, as well as total failure of timing sensitive applications. See bug #93147 for an example of the regression on Linux 4.4-rc: https://bugs.freedesktop.org/show_bug.cgi?id=93147 This patch tries to align all above events better from the viewpoint of the drm core / of external callers to fix the problem: 1. The apparent start of vblank is shifted a few scanlines earlier, so the vblank irq now always happens after start of this extended vblank interval and thereby drm_update_vblank_count() always samples the updated vblank count and timestamp of the new vblank interval. To achieve this, the reporting of scanout positions by radeon_get_crtc_scanoutpos() now operates as if the vblank starts radeon_crtc->lb_vblank_lead_lines before the real start of the hw vblank interval. This means that the vblank timestamps which are based on these scanout positions will now update at this earlier start of vblank. 2. The driver->get_vblank_counter() function will bump the returned vblank count as read from the hw by +1 if the query happens after the shifted earlier start of the vblank, but before the real hw increment at start of vsync, so the counter appears to increment at start of vblank in sync with the timestamp update. 3. Calls from vblank irq-context and regular non-irq calls are now treated identical, always simulating the shifted vblank start, to avoid inconsistent results for queries happening from vblank irq vs. happening from drm_vblank_enable() or vblank_disable_fn(). 4. The radeon_flip_work_func will delay mmio programming a pageflip until the start of the real vblank iff it happens to execute inside the shifted earlier start of the vblank, so pageflips now also appear to execute at start of the shifted vblank, in sync with vblank counter and timestamp updates. This to avoid some races between updates of vblank count and timestamps that are used for swap scheduling and pageflip execution which could cause pageflips to execute before the scheduled target vblank. The lb_vblank_lead_lines "fudge" value is calculated as the size of the display controllers line buffer in scanlines for the given video mode: Vblank irq's are triggered by the line buffer logic when the line buffer refill for a video frame ends, ie. when the line buffer source read position enters the hw vblank. This means that a vblank irq could fire at most as many scanlines before the current reported scanout position of the crtc timing generator as the number of scanlines the line buffer can maximally hold for a given video mode. This patch has been successfully tested on a RV730 card with DCE-3 display engine and on a evergreen card with DCE-4 display engine, in single-display and dual-display configuration, with different video modes. A similar patch is needed for amdgpu-kms to fix the same problem. Limitations: - Line buffer sizes in pixels are hard-coded on < DCE-4 to a value i just guessed to be high enough to work ok, lacking info on the true sizes atm. Fixes: fdo#93147 Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Michel Dänzer <michel.daenzer@amd.com> Cc: Harry Wentland <Harry.Wentland@amd.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net> (v2) Refine radeon_flip_work_func() for better efficiency: In radeon_flip_work_func, replace the busy waiting udelay(5) with event lock held by a more performance and energy efficient usleep_range() until at least predicted true start of hw vblank, with some slack for scheduler happiness. Release the event lock during waits to not delay other outputs in doing their stuff, as the waiting can last up to 200 usecs in some cases. Retested on DCE-3 and DCE-4 to verify it still works nicely. (v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-26 02:14:31 +07:00
DRM_DEBUG_VBL("NULL mode info! Returned count may be wrong.\n");
}
return count;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
}
/**
* radeon_enable_vblank_kms - enable vblank interrupt
*
* @dev: drm dev pointer
* @crtc: crtc to enable vblank interrupt for
*
* Enable the interrupt on the requested crtc (all asics).
* Returns 0 on success, -EINVAL on failure.
*/
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
int radeon_enable_vblank_kms(struct drm_device *dev, int crtc)
{
struct radeon_device *rdev = dev->dev_private;
unsigned long irqflags;
int r;
if (crtc < 0 || crtc >= rdev->num_crtc) {
DRM_ERROR("Invalid crtc %d\n", crtc);
return -EINVAL;
}
spin_lock_irqsave(&rdev->irq.lock, irqflags);
rdev->irq.crtc_vblank_int[crtc] = true;
r = radeon_irq_set(rdev);
spin_unlock_irqrestore(&rdev->irq.lock, irqflags);
return r;
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
}
/**
* radeon_disable_vblank_kms - disable vblank interrupt
*
* @dev: drm dev pointer
* @crtc: crtc to disable vblank interrupt for
*
* Disable the interrupt on the requested crtc (all asics).
*/
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
void radeon_disable_vblank_kms(struct drm_device *dev, int crtc)
{
struct radeon_device *rdev = dev->dev_private;
unsigned long irqflags;
if (crtc < 0 || crtc >= rdev->num_crtc) {
DRM_ERROR("Invalid crtc %d\n", crtc);
return;
}
spin_lock_irqsave(&rdev->irq.lock, irqflags);
rdev->irq.crtc_vblank_int[crtc] = false;
radeon_irq_set(rdev);
spin_unlock_irqrestore(&rdev->irq.lock, irqflags);
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
}
const struct drm_ioctl_desc radeon_ioctls_kms[] = {
DRM_IOCTL_DEF_DRV(RADEON_CP_INIT, drm_invalid_op, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
DRM_IOCTL_DEF_DRV(RADEON_CP_START, drm_invalid_op, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
DRM_IOCTL_DEF_DRV(RADEON_CP_STOP, drm_invalid_op, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
DRM_IOCTL_DEF_DRV(RADEON_CP_RESET, drm_invalid_op, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
DRM_IOCTL_DEF_DRV(RADEON_CP_IDLE, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_CP_RESUME, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_RESET, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_FULLSCREEN, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_SWAP, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_CLEAR, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_VERTEX, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_INDICES, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_TEXTURE, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_STIPPLE, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_INDIRECT, drm_invalid_op, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
DRM_IOCTL_DEF_DRV(RADEON_VERTEX2, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_CMDBUF, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_GETPARAM, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_FLIP, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_ALLOC, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_FREE, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_INIT_HEAP, drm_invalid_op, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
DRM_IOCTL_DEF_DRV(RADEON_IRQ_EMIT, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_IRQ_WAIT, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_SETPARAM, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_SURF_ALLOC, drm_invalid_op, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_SURF_FREE, drm_invalid_op, DRM_AUTH),
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
/* KMS */
DRM_IOCTL_DEF_DRV(RADEON_GEM_INFO, radeon_gem_info_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_CREATE, radeon_gem_create_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_MMAP, radeon_gem_mmap_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_SET_DOMAIN, radeon_gem_set_domain_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_PREAD, radeon_gem_pread_ioctl, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_GEM_PWRITE, radeon_gem_pwrite_ioctl, DRM_AUTH),
DRM_IOCTL_DEF_DRV(RADEON_GEM_WAIT_IDLE, radeon_gem_wait_idle_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_CS, radeon_cs_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_INFO, radeon_info_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_SET_TILING, radeon_gem_set_tiling_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_GET_TILING, radeon_gem_get_tiling_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_BUSY, radeon_gem_busy_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_VA, radeon_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_OP, radeon_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_USERPTR, radeon_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
drm/radeon: introduce kernel modesetting for radeon hardware Add kernel modesetting support to radeon driver, use the ttm memory manager to manage memory and DRM/GEM to provide userspace API. In order to avoid backward compatibility issue and to allow clean design and code the radeon kernel modesetting use different code path than old radeon/drm driver. When kernel modesetting is enabled the IOCTL of radeon/drm driver are considered as invalid and an error message is printed in the log and they return failure. KMS enabled userspace will use new API to talk with the radeon/drm driver. The new API provide functions to create/destroy/share/mmap buffer object which are then managed by the kernel memory manager (here TTM). In order to submit command to the GPU the userspace provide a buffer holding the command stream, along this buffer userspace have to provide a list of buffer object used by the command stream. The kernel radeon driver will then place buffer in GPU accessible memory and will update command stream to reflect the position of the different buffers. The kernel will also perform security check on command stream provided by the user, we want to catch and forbid any illegal use of the GPU such as DMA into random system memory or into memory not owned by the process supplying the command stream. This part of the code is still incomplete and this why we propose that patch as a staging driver addition, future security might forbid current experimental userspace to run. This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX (radeon up to X1950). Works is underway to provide support for R6XX, R7XX and newer hardware (radeon from HD2XXX to HD4XXX). Authors: Jerome Glisse <jglisse@redhat.com> Dave Airlie <airlied@redhat.com> Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Jerome Glisse <jglisse@redhat.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Alex Deucher <alexdeucher@gmail.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-06-05 19:42:42 +07:00
};
int radeon_max_kms_ioctl = ARRAY_SIZE(radeon_ioctls_kms);