linux_dsm_epyc7002/drivers/gpu/drm/radeon/r600_dma.c
Matthew Dawson 04db4caf5c drm/radeon: Avoid double gpu reset by adding a timeout on IB ring tests.
When the radeon driver resets a gpu, it attempts to test whether all the
rings can successfully handle an IB.  If these rings fail to respond, the
process will wait forever.  Another gpu reset can't happen at this point,
as the current reset holds a lock required to do so.  Instead, make all
the IB tests run with a timeout, so the system can attempt to recover
in this case.

While this doesn't fix the underlying issue with card resets failing, it
gives the system a higher chance of recovering.  These timeouts have been
confirmed to help both a Tathi and Hawaii card recover after a gpu reset.

This also adds a new function, radeon_fence_wait_timeout, that behaves like
fence_wait_timeout.  It is used instead of fence_wait_timeout as it continues
to work during a reset.  radeon_fence_wait is changed to be implemented
using this function.

V2:
 - Changed the timeout to 1s, as the default 10s from radeon_wait_timeout was
too long.  A timeout of 100ms was tested and found to be too short.
 - Changed radeon_fence_wait_timeout to behave more like fence_wait_timeout.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Matthew Dawson <matthew@mjdsystems.ca>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2016-02-10 14:17:15 -05:00

497 lines
14 KiB
C

/*
* Copyright 2013 Advanced Micro Devices, Inc.
*
* 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: Alex Deucher
*/
#include <drm/drmP.h>
#include "radeon.h"
#include "radeon_asic.h"
#include "r600d.h"
u32 r600_gpu_check_soft_reset(struct radeon_device *rdev);
/*
* DMA
* Starting with R600, the GPU has an asynchronous
* DMA engine. The programming model is very similar
* to the 3D engine (ring buffer, IBs, etc.), but the
* DMA controller has it's own packet format that is
* different form the PM4 format used by the 3D engine.
* It supports copying data, writing embedded data,
* solid fills, and a number of other things. It also
* has support for tiling/detiling of buffers.
*/
/**
* r600_dma_get_rptr - get the current read pointer
*
* @rdev: radeon_device pointer
* @ring: radeon ring pointer
*
* Get the current rptr from the hardware (r6xx+).
*/
uint32_t r600_dma_get_rptr(struct radeon_device *rdev,
struct radeon_ring *ring)
{
u32 rptr;
if (rdev->wb.enabled)
rptr = rdev->wb.wb[ring->rptr_offs/4];
else
rptr = RREG32(DMA_RB_RPTR);
return (rptr & 0x3fffc) >> 2;
}
/**
* r600_dma_get_wptr - get the current write pointer
*
* @rdev: radeon_device pointer
* @ring: radeon ring pointer
*
* Get the current wptr from the hardware (r6xx+).
*/
uint32_t r600_dma_get_wptr(struct radeon_device *rdev,
struct radeon_ring *ring)
{
return (RREG32(DMA_RB_WPTR) & 0x3fffc) >> 2;
}
/**
* r600_dma_set_wptr - commit the write pointer
*
* @rdev: radeon_device pointer
* @ring: radeon ring pointer
*
* Write the wptr back to the hardware (r6xx+).
*/
void r600_dma_set_wptr(struct radeon_device *rdev,
struct radeon_ring *ring)
{
WREG32(DMA_RB_WPTR, (ring->wptr << 2) & 0x3fffc);
}
/**
* r600_dma_stop - stop the async dma engine
*
* @rdev: radeon_device pointer
*
* Stop the async dma engine (r6xx-evergreen).
*/
void r600_dma_stop(struct radeon_device *rdev)
{
u32 rb_cntl = RREG32(DMA_RB_CNTL);
if (rdev->asic->copy.copy_ring_index == R600_RING_TYPE_DMA_INDEX)
radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
rb_cntl &= ~DMA_RB_ENABLE;
WREG32(DMA_RB_CNTL, rb_cntl);
rdev->ring[R600_RING_TYPE_DMA_INDEX].ready = false;
}
/**
* r600_dma_resume - setup and start the async dma engine
*
* @rdev: radeon_device pointer
*
* Set up the DMA ring buffer and enable it. (r6xx-evergreen).
* Returns 0 for success, error for failure.
*/
int r600_dma_resume(struct radeon_device *rdev)
{
struct radeon_ring *ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
u32 rb_cntl, dma_cntl, ib_cntl;
u32 rb_bufsz;
int r;
WREG32(DMA_SEM_INCOMPLETE_TIMER_CNTL, 0);
WREG32(DMA_SEM_WAIT_FAIL_TIMER_CNTL, 0);
/* Set ring buffer size in dwords */
rb_bufsz = order_base_2(ring->ring_size / 4);
rb_cntl = rb_bufsz << 1;
#ifdef __BIG_ENDIAN
rb_cntl |= DMA_RB_SWAP_ENABLE | DMA_RPTR_WRITEBACK_SWAP_ENABLE;
#endif
WREG32(DMA_RB_CNTL, rb_cntl);
/* Initialize the ring buffer's read and write pointers */
WREG32(DMA_RB_RPTR, 0);
WREG32(DMA_RB_WPTR, 0);
/* set the wb address whether it's enabled or not */
WREG32(DMA_RB_RPTR_ADDR_HI,
upper_32_bits(rdev->wb.gpu_addr + R600_WB_DMA_RPTR_OFFSET) & 0xFF);
WREG32(DMA_RB_RPTR_ADDR_LO,
((rdev->wb.gpu_addr + R600_WB_DMA_RPTR_OFFSET) & 0xFFFFFFFC));
if (rdev->wb.enabled)
rb_cntl |= DMA_RPTR_WRITEBACK_ENABLE;
WREG32(DMA_RB_BASE, ring->gpu_addr >> 8);
/* enable DMA IBs */
ib_cntl = DMA_IB_ENABLE;
#ifdef __BIG_ENDIAN
ib_cntl |= DMA_IB_SWAP_ENABLE;
#endif
WREG32(DMA_IB_CNTL, ib_cntl);
dma_cntl = RREG32(DMA_CNTL);
dma_cntl &= ~CTXEMPTY_INT_ENABLE;
WREG32(DMA_CNTL, dma_cntl);
if (rdev->family >= CHIP_RV770)
WREG32(DMA_MODE, 1);
ring->wptr = 0;
WREG32(DMA_RB_WPTR, ring->wptr << 2);
WREG32(DMA_RB_CNTL, rb_cntl | DMA_RB_ENABLE);
ring->ready = true;
r = radeon_ring_test(rdev, R600_RING_TYPE_DMA_INDEX, ring);
if (r) {
ring->ready = false;
return r;
}
if (rdev->asic->copy.copy_ring_index == R600_RING_TYPE_DMA_INDEX)
radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
return 0;
}
/**
* r600_dma_fini - tear down the async dma engine
*
* @rdev: radeon_device pointer
*
* Stop the async dma engine and free the ring (r6xx-evergreen).
*/
void r600_dma_fini(struct radeon_device *rdev)
{
r600_dma_stop(rdev);
radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
}
/**
* r600_dma_is_lockup - Check if the DMA engine is locked up
*
* @rdev: radeon_device pointer
* @ring: radeon_ring structure holding ring information
*
* Check if the async DMA engine is locked up.
* Returns true if the engine appears to be locked up, false if not.
*/
bool r600_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
{
u32 reset_mask = r600_gpu_check_soft_reset(rdev);
if (!(reset_mask & RADEON_RESET_DMA)) {
radeon_ring_lockup_update(rdev, ring);
return false;
}
return radeon_ring_test_lockup(rdev, ring);
}
/**
* r600_dma_ring_test - simple async dma engine test
*
* @rdev: radeon_device pointer
* @ring: radeon_ring structure holding ring information
*
* Test the DMA engine by writing using it to write an
* value to memory. (r6xx-SI).
* Returns 0 for success, error for failure.
*/
int r600_dma_ring_test(struct radeon_device *rdev,
struct radeon_ring *ring)
{
unsigned i;
int r;
unsigned index;
u32 tmp;
u64 gpu_addr;
if (ring->idx == R600_RING_TYPE_DMA_INDEX)
index = R600_WB_DMA_RING_TEST_OFFSET;
else
index = CAYMAN_WB_DMA1_RING_TEST_OFFSET;
gpu_addr = rdev->wb.gpu_addr + index;
tmp = 0xCAFEDEAD;
rdev->wb.wb[index/4] = cpu_to_le32(tmp);
r = radeon_ring_lock(rdev, ring, 4);
if (r) {
DRM_ERROR("radeon: dma failed to lock ring %d (%d).\n", ring->idx, r);
return r;
}
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
radeon_ring_write(ring, lower_32_bits(gpu_addr));
radeon_ring_write(ring, upper_32_bits(gpu_addr) & 0xff);
radeon_ring_write(ring, 0xDEADBEEF);
radeon_ring_unlock_commit(rdev, ring, false);
for (i = 0; i < rdev->usec_timeout; i++) {
tmp = le32_to_cpu(rdev->wb.wb[index/4]);
if (tmp == 0xDEADBEEF)
break;
DRM_UDELAY(1);
}
if (i < rdev->usec_timeout) {
DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
} else {
DRM_ERROR("radeon: ring %d test failed (0x%08X)\n",
ring->idx, tmp);
r = -EINVAL;
}
return r;
}
/**
* r600_dma_fence_ring_emit - emit a fence on the DMA ring
*
* @rdev: radeon_device pointer
* @fence: radeon fence object
*
* Add a DMA fence packet to the ring to write
* the fence seq number and DMA trap packet to generate
* an interrupt if needed (r6xx-r7xx).
*/
void r600_dma_fence_ring_emit(struct radeon_device *rdev,
struct radeon_fence *fence)
{
struct radeon_ring *ring = &rdev->ring[fence->ring];
u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
/* write the fence */
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0, 0));
radeon_ring_write(ring, addr & 0xfffffffc);
radeon_ring_write(ring, (upper_32_bits(addr) & 0xff));
radeon_ring_write(ring, lower_32_bits(fence->seq));
/* generate an interrupt */
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0, 0));
}
/**
* r600_dma_semaphore_ring_emit - emit a semaphore on the dma ring
*
* @rdev: radeon_device pointer
* @ring: radeon_ring structure holding ring information
* @semaphore: radeon semaphore object
* @emit_wait: wait or signal semaphore
*
* Add a DMA semaphore packet to the ring wait on or signal
* other rings (r6xx-SI).
*/
bool r600_dma_semaphore_ring_emit(struct radeon_device *rdev,
struct radeon_ring *ring,
struct radeon_semaphore *semaphore,
bool emit_wait)
{
u64 addr = semaphore->gpu_addr;
u32 s = emit_wait ? 0 : 1;
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SEMAPHORE, 0, s, 0));
radeon_ring_write(ring, addr & 0xfffffffc);
radeon_ring_write(ring, upper_32_bits(addr) & 0xff);
return true;
}
/**
* r600_dma_ib_test - test an IB on the DMA engine
*
* @rdev: radeon_device pointer
* @ring: radeon_ring structure holding ring information
*
* Test a simple IB in the DMA ring (r6xx-SI).
* Returns 0 on success, error on failure.
*/
int r600_dma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
{
struct radeon_ib ib;
unsigned i;
unsigned index;
int r;
u32 tmp = 0;
u64 gpu_addr;
if (ring->idx == R600_RING_TYPE_DMA_INDEX)
index = R600_WB_DMA_RING_TEST_OFFSET;
else
index = CAYMAN_WB_DMA1_RING_TEST_OFFSET;
gpu_addr = rdev->wb.gpu_addr + index;
r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
if (r) {
DRM_ERROR("radeon: failed to get ib (%d).\n", r);
return r;
}
ib.ptr[0] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1);
ib.ptr[1] = lower_32_bits(gpu_addr);
ib.ptr[2] = upper_32_bits(gpu_addr) & 0xff;
ib.ptr[3] = 0xDEADBEEF;
ib.length_dw = 4;
r = radeon_ib_schedule(rdev, &ib, NULL, false);
if (r) {
radeon_ib_free(rdev, &ib);
DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
return r;
}
r = radeon_fence_wait_timeout(ib.fence, false, usecs_to_jiffies(
RADEON_USEC_IB_TEST_TIMEOUT));
if (r < 0) {
DRM_ERROR("radeon: fence wait failed (%d).\n", r);
return r;
} else if (r == 0) {
DRM_ERROR("radeon: fence wait timed out.\n");
return -ETIMEDOUT;
}
r = 0;
for (i = 0; i < rdev->usec_timeout; i++) {
tmp = le32_to_cpu(rdev->wb.wb[index/4]);
if (tmp == 0xDEADBEEF)
break;
DRM_UDELAY(1);
}
if (i < rdev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
} else {
DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
r = -EINVAL;
}
radeon_ib_free(rdev, &ib);
return r;
}
/**
* r600_dma_ring_ib_execute - Schedule an IB on the DMA engine
*
* @rdev: radeon_device pointer
* @ib: IB object to schedule
*
* Schedule an IB in the DMA ring (r6xx-r7xx).
*/
void r600_dma_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
{
struct radeon_ring *ring = &rdev->ring[ib->ring];
if (rdev->wb.enabled) {
u32 next_rptr = ring->wptr + 4;
while ((next_rptr & 7) != 5)
next_rptr++;
next_rptr += 3;
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
radeon_ring_write(ring, next_rptr);
}
/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
* Pad as necessary with NOPs.
*/
while ((ring->wptr & 7) != 5)
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_INDIRECT_BUFFER, 0, 0, 0));
radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
radeon_ring_write(ring, (ib->length_dw << 16) | (upper_32_bits(ib->gpu_addr) & 0xFF));
}
/**
* r600_copy_dma - copy pages using the DMA engine
*
* @rdev: radeon_device pointer
* @src_offset: src GPU address
* @dst_offset: dst GPU address
* @num_gpu_pages: number of GPU pages to xfer
* @resv: reservation object to sync to
*
* Copy GPU paging using the DMA engine (r6xx).
* Used by the radeon ttm implementation to move pages if
* registered as the asic copy callback.
*/
struct radeon_fence *r600_copy_dma(struct radeon_device *rdev,
uint64_t src_offset, uint64_t dst_offset,
unsigned num_gpu_pages,
struct reservation_object *resv)
{
struct radeon_fence *fence;
struct radeon_sync sync;
int ring_index = rdev->asic->copy.dma_ring_index;
struct radeon_ring *ring = &rdev->ring[ring_index];
u32 size_in_dw, cur_size_in_dw;
int i, num_loops;
int r = 0;
radeon_sync_create(&sync);
size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
num_loops = DIV_ROUND_UP(size_in_dw, 0xFFFE);
r = radeon_ring_lock(rdev, ring, num_loops * 4 + 8);
if (r) {
DRM_ERROR("radeon: moving bo (%d).\n", r);
radeon_sync_free(rdev, &sync, NULL);
return ERR_PTR(r);
}
radeon_sync_resv(rdev, &sync, resv, false);
radeon_sync_rings(rdev, &sync, ring->idx);
for (i = 0; i < num_loops; i++) {
cur_size_in_dw = size_in_dw;
if (cur_size_in_dw > 0xFFFE)
cur_size_in_dw = 0xFFFE;
size_in_dw -= cur_size_in_dw;
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, 0, cur_size_in_dw));
radeon_ring_write(ring, dst_offset & 0xfffffffc);
radeon_ring_write(ring, src_offset & 0xfffffffc);
radeon_ring_write(ring, (((upper_32_bits(dst_offset) & 0xff) << 16) |
(upper_32_bits(src_offset) & 0xff)));
src_offset += cur_size_in_dw * 4;
dst_offset += cur_size_in_dw * 4;
}
r = radeon_fence_emit(rdev, &fence, ring->idx);
if (r) {
radeon_ring_unlock_undo(rdev, ring);
radeon_sync_free(rdev, &sync, NULL);
return ERR_PTR(r);
}
radeon_ring_unlock_commit(rdev, ring, false);
radeon_sync_free(rdev, &sync, fence);
return fence;
}