[PATCH] drm/amdgpu/gfx7: fix pipeline sync

List overview All Threads
Download

newer

older

re: drm/ttm: Refuse to fault...

[PATCH 1/3] drm: Drop crtc...

Alex Deucher

9 May 2016 9 May '16

7:35 p.m.

From: Chunming Zhou David1.Zhou@amd.com

Need to wait on the fence as well.

Signed-off-by: Chunming Zhou David1.Zhou@amd.com Reviewed-by: Monk Liu monk.liu@amd.com Signed-off-by: Alex Deucher alexander.deucher@amd.com --- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 25 +++++++++++++------------ 1 file changed, 13 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index 92d8061..f4068f7 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -3054,6 +3054,19 @@ static int gfx_v7_0_cp_resume(struct amdgpu_device *adev) static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) { int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX); + uint32_t seq = ring->fence_drv.sync_seq; + uint64_t addr = ring->fence_drv.gpu_addr; + + amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); + amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */ + WAIT_REG_MEM_FUNCTION(3) | /* equal */ + WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ + amdgpu_ring_write(ring, addr & 0xfffffffc); + amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); + amdgpu_ring_write(ring, seq); + amdgpu_ring_write(ring, 0xffffffff); + amdgpu_ring_write(ring, 4); /* poll interval */ + if (usepfp) { /* synce CE with ME to prevent CE fetch CEIB before context switch done */ amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); @@ -3081,18 +3094,6 @@ static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring, unsigned vm_id, uint64_t pd_addr) { int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX); - uint32_t seq = ring->fence_drv.sync_seq; - uint64_t addr = ring->fence_drv.gpu_addr; - - amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); - amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */ - WAIT_REG_MEM_FUNCTION(3) | /* equal */ - WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ - amdgpu_ring_write(ring, addr & 0xfffffffc); - amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); - amdgpu_ring_write(ring, seq); - amdgpu_ring_write(ring, 0xffffffff); - amdgpu_ring_write(ring, 4); /* poll interval */

amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |

-- 2.5.5

Show replies by date

Alex Deucher

9 May 9 May

7:35 p.m.

New subject: [PATCH] drm/amdgpu: move preamble IB handling into common code

From: Christian König christian.koenig@amd.com

This fixes the handling which was completely broken when you ad more than one preamble IB.

Signed-off-by: Christian König christian.koenig@amd.com Reviewed-by: Alex Deucher alexander.deucher@amd.com Signed-off-by: Alex Deucher alexander.deucher@amd.com --- drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 8 +++++++- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 4 ---- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 4 ---- 3 files changed, 7 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index f670519..d6f8592 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -125,6 +125,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, struct fence *hwf; struct amdgpu_vm *vm = NULL; unsigned i, patch_offset = ~0; + bool skip_preamble;

int r = 0;

@@ -172,9 +173,14 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, /* always set cond_exec_polling to CONTINUE */ *ring->cond_exe_cpu_addr = 1;

+ skip_preamble = ring->current_ctx == ctx; old_ctx = ring->current_ctx; for (i = 0; i < num_ibs; ++i) { - ib = &ibs[i]; + + /* drop preamble IBs if we don't have a context switch */ + if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && skip_preamble) + continue; + amdgpu_ring_emit_ib(ring, ib); ring->current_ctx = ctx; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index f4068f7..d82fa96 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -2036,10 +2036,6 @@ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, u32 header, control = 0; u32 next_rptr = ring->wptr + 5;

- /* drop the CE preamble IB for the same context */ - if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && !need_ctx_switch) - return; - if (need_ctx_switch) next_rptr += 2;

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 7886228..9a0b6df 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -5652,10 +5652,6 @@ static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, u32 header, control = 0; u32 next_rptr = ring->wptr + 5;

- /* drop the CE preamble IB for the same context */ - if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && !need_ctx_switch) - return; - if (need_ctx_switch) next_rptr += 2;

-- 2.5.5

Alex Deucher

7:35 p.m.

New subject: [PATCH] drm/amdgpu: move context switch handling into common code v2

From: Christian König christian.koenig@amd.com

It was a source of bugs to repeat that in each IP version.

v2: rename parameter

Signed-off-by: Christian König christian.koenig@amd.com Reviewed-by: Alex Deucher alexander.deucher@amd.com Signed-off-by: Alex Deucher alexander.deucher@amd.com --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 16 +++++++--------- drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h | 2 +- drivers/gpu/drm/amd/amdgpu/cik_sdma.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 9 ++++----- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 9 ++++----- drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c | 2 +- drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c | 2 +- drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c | 2 +- 12 files changed, 25 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 4c103e9..11d7d4b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -283,7 +283,7 @@ struct amdgpu_ring_funcs { int (*parse_cs)(struct amdgpu_cs_parser *p, uint32_t ib_idx); /* command emit functions */ void (*emit_ib)(struct amdgpu_ring *ring, - struct amdgpu_ib *ib); + struct amdgpu_ib *ib, bool ctx_switch); void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr, uint64_t seq, unsigned flags); void (*emit_pipeline_sync)(struct amdgpu_ring *ring); @@ -2220,7 +2220,7 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring) #define amdgpu_ring_get_rptr(r) (r)->funcs->get_rptr((r)) #define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r)) #define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r)) -#define amdgpu_ring_emit_ib(r, ib) (r)->funcs->emit_ib((r), (ib)) +#define amdgpu_ring_emit_ib(r, ib, c) (r)->funcs->emit_ib((r), (ib), (c)) #define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r)) #define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr)) #define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags)) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index d6f8592..88b8fda 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -121,18 +121,16 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, { struct amdgpu_device *adev = ring->adev; struct amdgpu_ib *ib = &ibs[0]; - uint64_t ctx, old_ctx; struct fence *hwf; struct amdgpu_vm *vm = NULL; unsigned i, patch_offset = ~0; - bool skip_preamble; + bool skip_preamble, need_ctx_switch;

int r = 0;

if (num_ibs == 0) return -EINVAL;

- ctx = ibs->ctx; if (job) /* for domain0 job like ring test, ibs->job is not assigned */ vm = job->vm;

@@ -156,7 +154,6 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, patch_offset = amdgpu_ring_init_cond_exec(ring);

if (vm) { - /* do context switch */ r = amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr, ib->gds_base, ib->gds_size, ib->gws_base, ib->gws_size, @@ -173,16 +170,17 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, /* always set cond_exec_polling to CONTINUE */ *ring->cond_exe_cpu_addr = 1;

- skip_preamble = ring->current_ctx == ctx; - old_ctx = ring->current_ctx; + skip_preamble = ring->current_ctx == ib->ctx; + need_ctx_switch = ring->current_ctx != ib->ctx; for (i = 0; i < num_ibs; ++i) { + ib = &ibs[i];

/* drop preamble IBs if we don't have a context switch */ if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && skip_preamble) continue;

- amdgpu_ring_emit_ib(ring, ib); - ring->current_ctx = ctx; + amdgpu_ring_emit_ib(ring, ib, need_ctx_switch); + need_ctx_switch = false; }

if (ring->funcs->emit_hdp_invalidate) @@ -191,7 +189,6 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, r = amdgpu_fence_emit(ring, &hwf); if (r) { dev_err(adev->dev, "failed to emit fence (%d)\n", r); - ring->current_ctx = old_ctx; if (ib->vm_id) amdgpu_vm_reset_id(adev, ib->vm_id); amdgpu_ring_undo(ring); @@ -212,6 +209,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, if (patch_offset != ~0 && ring->funcs->patch_cond_exec) amdgpu_ring_patch_cond_exec(ring, patch_offset);

+ ring->current_ctx = ibs->ctx; amdgpu_ring_commit(ring); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c index a28d7e9..62cf732 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c @@ -761,7 +761,7 @@ out: * @ib: the IB to execute * */ -void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) +void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib, bool ctx_switch) { amdgpu_ring_write(ring, VCE_CMD_IB); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h index ef99d23..40d0650 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h @@ -34,7 +34,7 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, bool direct, struct fence **fence); void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp); int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx); -void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib); +void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib, bool ctx_switch); void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, unsigned flags); int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring); diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c index df824f8..6c2aa2b 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c @@ -210,7 +210,7 @@ static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (CIK). */ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { u32 extra_bits = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index d82fa96..189ef2b 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -2030,13 +2030,12 @@ static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring, * on the gfx ring for execution by the GPU. */ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { - bool need_ctx_switch = ring->current_ctx != ib->ctx; u32 header, control = 0; u32 next_rptr = ring->wptr + 5;

- if (need_ctx_switch) + if (ctx_switch) next_rptr += 2;

next_rptr += 4; @@ -2047,7 +2046,7 @@ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, amdgpu_ring_write(ring, next_rptr);

/* insert SWITCH_BUFFER packet before first IB in the ring frame */ - if (need_ctx_switch) { + if (ctx_switch) { amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); amdgpu_ring_write(ring, 0); } @@ -2070,7 +2069,7 @@ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, }

static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { u32 header, control = 0; u32 next_rptr = ring->wptr + 5; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 9a0b6df..0d556c9 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -5646,13 +5646,12 @@ static void gfx_v8_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) }

static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { - bool need_ctx_switch = ring->current_ctx != ib->ctx; u32 header, control = 0; u32 next_rptr = ring->wptr + 5;

- if (need_ctx_switch) + if (ctx_switch) next_rptr += 2;

next_rptr += 4; @@ -5663,7 +5662,7 @@ static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, amdgpu_ring_write(ring, next_rptr);

/* insert SWITCH_BUFFER packet before first IB in the ring frame */ - if (need_ctx_switch) { + if (ctx_switch) { amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); amdgpu_ring_write(ring, 0); } @@ -5686,7 +5685,7 @@ static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, }

static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { u32 header, control = 0; u32 next_rptr = ring->wptr + 5; diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index 6be2c0f..de94adb 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c @@ -242,7 +242,7 @@ static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (VI). */ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { u32 vmid = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c index b3dab09..ca2aee3 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c @@ -400,7 +400,7 @@ static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (VI). */ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { u32 vmid = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c index 46a3976..a75ffb5b 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c @@ -489,7 +489,7 @@ static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0)); amdgpu_ring_write(ring, ib->gpu_addr); diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c index b96486c..ecb8101 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c @@ -539,7 +539,7 @@ static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0)); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c index 892bdac..a43f1a7 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c @@ -631,7 +631,7 @@ static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib) + struct amdgpu_ib *ib, bool ctx_switch) { amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0)); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));

-- 2.5.5

Alex Deucher

7:35 p.m.

New subject: [PATCH] drm/amdgpu: move the context from the IBs into the job

From: Christian König christian.koenig@amd.com

We only have one context for all IBs.

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 11d7d4b..fcc1a9d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -743,7 +743,6 @@ struct amdgpu_ib { struct amdgpu_user_fence *user; unsigned vm_id; uint64_t vm_pd_addr; - uint64_t ctx; uint32_t gds_base, gds_size; uint32_t gws_base, gws_size; uint32_t oa_base, oa_size; @@ -1262,6 +1261,7 @@ struct amdgpu_job { struct fence *fence; /* the hw fence */ uint32_t num_ibs; void *owner; + uint64_t ctx; struct amdgpu_user_fence uf; }; #define to_amdgpu_job(sched_job) \ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 87ec113..2895d63 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -741,7 +741,6 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,

ib->length_dw = chunk_ib->ib_bytes / 4; ib->flags = chunk_ib->flags; - ib->ctx = parser->ctx->rings[ring->idx].entity.fence_context; j++; }

@@ -840,6 +839,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, union drm_amdgpu_cs *cs) { struct amdgpu_ring *ring = p->job->ring; + struct amd_sched_entity *entity = &p->ctx->rings[ring->idx].entity; struct fence *fence; struct amdgpu_job *job; int r; @@ -848,16 +848,16 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, p->job = NULL;

r = amd_sched_job_init(&job->base, &ring->sched, - &p->ctx->rings[ring->idx].entity, - amdgpu_job_timeout_func, - amdgpu_job_free_func, - p->filp, &fence); + entity, amdgpu_job_timeout_func, + amdgpu_job_free_func, + p->filp, &fence); if (r) { amdgpu_job_free(job); return r; }

job->owner = p->filp; + job->ctx = entity->fence_context; p->fence = fence_get(fence); cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, fence); job->ibs[job->num_ibs - 1].sequence = cs->out.handle; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index 88b8fda..dacbd2e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -121,18 +121,26 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, { struct amdgpu_device *adev = ring->adev; struct amdgpu_ib *ib = &ibs[0]; - struct fence *hwf; - struct amdgpu_vm *vm = NULL; - unsigned i, patch_offset = ~0; bool skip_preamble, need_ctx_switch; + unsigned patch_offset = ~0; + struct amdgpu_vm *vm; + struct fence *hwf; + uint64_t ctx;

+ unsigned i; int r = 0;

if (num_ibs == 0) return -EINVAL;

- if (job) /* for domain0 job like ring test, ibs->job is not assigned */ + /* ring tests don't use a job */ + if (job) { vm = job->vm; + ctx = job->ctx; + } else { + vm = NULL; + ctx = 0; + }

if (!ring->ready) { dev_err(adev->dev, "couldn't schedule ib\n"); @@ -170,8 +178,8 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, /* always set cond_exec_polling to CONTINUE */ *ring->cond_exe_cpu_addr = 1;

- skip_preamble = ring->current_ctx == ib->ctx; - need_ctx_switch = ring->current_ctx != ib->ctx; + skip_preamble = ring->current_ctx == ctx; + need_ctx_switch = ring->current_ctx != ctx; for (i = 0; i < num_ibs; ++i) { ib = &ibs[i];

@@ -209,7 +217,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, if (patch_offset != ~0 && ring->funcs->patch_cond_exec) amdgpu_ring_patch_cond_exec(ring, patch_offset);

- ring->current_ctx = ibs->ctx; + ring->current_ctx = ctx; amdgpu_ring_commit(ring); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 917c6f3..a0961f2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -122,14 +122,13 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, return -EINVAL;

r = amd_sched_job_init(&job->base, &ring->sched, - entity, - amdgpu_job_timeout_func, - amdgpu_job_free_func, - owner, &fence); + entity, amdgpu_job_timeout_func, + amdgpu_job_free_func, owner, &fence); if (r) return r;

job->owner = owner; + job->ctx = entity->fence_context; *f = fence_get(fence); amd_sched_entity_push_job(&job->base);

-- 2.5.5

Alex Deucher

7:35 p.m.

New subject: [PATCH] drm/amdgpu: move VM fields into job

From: Christian König christian.koenig@amd.com

They are the same for all IBs.

Signed-off-by: Christian König christian.koenig@amd.com Reviewed-by: Alex Deucher alexander.deucher@amd.com Signed-off-by: Alex Deucher alexander.deucher@amd.com --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 15 +++++++------- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 36 +++++++++++++++------------------ drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 19 +++++++++-------- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 12 ++--------- drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 3 ++- drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h | 3 ++- drivers/gpu/drm/amd/amdgpu/cik_sdma.c | 5 +++-- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 10 +++++---- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 10 +++++---- drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c | 5 +++-- drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 5 +++-- drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c | 3 ++- drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c | 3 ++- drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c | 3 ++- 14 files changed, 66 insertions(+), 66 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index fcc1a9d..7cff077 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -283,7 +283,8 @@ struct amdgpu_ring_funcs { int (*parse_cs)(struct amdgpu_cs_parser *p, uint32_t ib_idx); /* command emit functions */ void (*emit_ib)(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch); + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch); void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr, uint64_t seq, unsigned flags); void (*emit_pipeline_sync)(struct amdgpu_ring *ring); @@ -741,11 +742,6 @@ struct amdgpu_ib { uint64_t gpu_addr; uint32_t *ptr; struct amdgpu_user_fence *user; - unsigned vm_id; - uint64_t vm_pd_addr; - uint32_t gds_base, gds_size; - uint32_t gws_base, gws_size; - uint32_t oa_base, oa_size; uint32_t flags; /* resulting sequence number */ uint64_t sequence; @@ -1262,6 +1258,11 @@ struct amdgpu_job { uint32_t num_ibs; void *owner; uint64_t ctx; + unsigned vm_id; + uint64_t vm_pd_addr; + uint32_t gds_base, gds_size; + uint32_t gws_base, gws_size; + uint32_t oa_base, oa_size; struct amdgpu_user_fence uf; }; #define to_amdgpu_job(sched_job) \ @@ -2220,7 +2221,7 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring) #define amdgpu_ring_get_rptr(r) (r)->funcs->get_rptr((r)) #define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r)) #define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r)) -#define amdgpu_ring_emit_ib(r, ib, c) (r)->funcs->emit_ib((r), (ib), (c)) +#define amdgpu_ring_emit_ib(r, ib, vm_id, c) (r)->funcs->emit_ib((r), (ib), (vm_id), (c)) #define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r)) #define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr)) #define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags)) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 2895d63..9ab2f08 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -473,6 +473,9 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, goto error_validate;

if (p->bo_list) { + struct amdgpu_bo *gds = p->bo_list->gds_obj; + struct amdgpu_bo *gws = p->bo_list->gws_obj; + struct amdgpu_bo *oa = p->bo_list->oa_obj; struct amdgpu_vm *vm = &fpriv->vm; unsigned i;

@@ -481,6 +484,19 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,

p->bo_list->array[i].bo_va = amdgpu_vm_bo_find(vm, bo); } + + if (gds) { + p->job->gds_base = amdgpu_bo_gpu_offset(gds); + p->job->gds_size = amdgpu_bo_size(gds); + } + if (gws) { + p->job->gws_base = amdgpu_bo_gpu_offset(gws); + p->job->gws_size = amdgpu_bo_size(gws); + } + if (oa) { + p->job->oa_base = amdgpu_bo_gpu_offset(oa); + p->job->oa_size = amdgpu_bo_size(oa); + } }

error_validate: @@ -744,26 +760,6 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, j++; }

- /* add GDS resources to first IB */ - if (parser->bo_list) { - struct amdgpu_bo *gds = parser->bo_list->gds_obj; - struct amdgpu_bo *gws = parser->bo_list->gws_obj; - struct amdgpu_bo *oa = parser->bo_list->oa_obj; - struct amdgpu_ib *ib = &parser->job->ibs[0]; - - if (gds) { - ib->gds_base = amdgpu_bo_gpu_offset(gds); - ib->gds_size = amdgpu_bo_size(gds); - } - if (gws) { - ib->gws_base = amdgpu_bo_gpu_offset(gws); - ib->gws_size = amdgpu_bo_size(gws); - } - if (oa) { - ib->oa_base = amdgpu_bo_gpu_offset(oa); - ib->oa_size = amdgpu_bo_size(oa); - } - } /* wrap the last IB with user fence */ if (parser->job->uf.bo) { struct amdgpu_ib *ib = &parser->job->ibs[parser->job->num_ibs - 1]; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index dacbd2e..201aceb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -74,8 +74,6 @@ int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm, ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo); }

- ib->vm_id = 0; - return 0; }

@@ -147,7 +145,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, return -EINVAL; }

- if (vm && !ibs->vm_id) { + if (vm && !job->vm_id) { dev_err(adev->dev, "VM IB without ID\n"); return -EINVAL; } @@ -162,10 +160,10 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, patch_offset = amdgpu_ring_init_cond_exec(ring);

if (vm) { - r = amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr, - ib->gds_base, ib->gds_size, - ib->gws_base, ib->gws_size, - ib->oa_base, ib->oa_size); + r = amdgpu_vm_flush(ring, job->vm_id, job->vm_pd_addr, + job->gds_base, job->gds_size, + job->gws_base, job->gws_size, + job->oa_base, job->oa_size); if (r) { amdgpu_ring_undo(ring); return r; @@ -187,7 +185,8 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && skip_preamble) continue;

- amdgpu_ring_emit_ib(ring, ib, need_ctx_switch); + amdgpu_ring_emit_ib(ring, ib, job ? job->vm_id : 0, + need_ctx_switch); need_ctx_switch = false; }

@@ -197,8 +196,8 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, r = amdgpu_fence_emit(ring, &hwf); if (r) { dev_err(adev->dev, "failed to emit fence (%d)\n", r); - if (ib->vm_id) - amdgpu_vm_reset_id(adev, ib->vm_id); + if (job && job->vm_id) + amdgpu_vm_reset_id(adev, job->vm_id); amdgpu_ring_undo(ring); return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index a0961f2..8ea68d0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -142,23 +142,15 @@ static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)

struct fence *fence = amdgpu_sync_get_fence(&job->sync);

- if (fence == NULL && vm && !job->ibs->vm_id) { + if (fence == NULL && vm && !job->vm_id) { struct amdgpu_ring *ring = job->ring; - unsigned i, vm_id; - uint64_t vm_pd_addr; int r;

r = amdgpu_vm_grab_id(vm, ring, &job->sync, &job->base.s_fence->base, - &vm_id, &vm_pd_addr); + &job->vm_id, &job->vm_pd_addr); if (r) DRM_ERROR("Error getting VM ID (%d)\n", r); - else { - for (i = 0; i < job->num_ibs; ++i) { - job->ibs[i].vm_id = vm_id; - job->ibs[i].vm_pd_addr = vm_pd_addr; - } - }

fence = amdgpu_sync_get_fence(&job->sync); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c index 62cf732..5f83d76 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c @@ -761,7 +761,8 @@ out: * @ib: the IB to execute * */ -void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib, bool ctx_switch) +void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { amdgpu_ring_write(ring, VCE_CMD_IB); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h index 40d0650..f40cf76 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h @@ -34,7 +34,8 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, bool direct, struct fence **fence); void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp); int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx); -void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib, bool ctx_switch); +void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch); void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, unsigned flags); int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring); diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c index 6c2aa2b..518dca4 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c @@ -210,9 +210,10 @@ static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (CIK). */ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { - u32 extra_bits = ib->vm_id & 0xf; + u32 extra_bits = vm_id & 0xf; u32 next_rptr = ring->wptr + 5;

while ((next_rptr & 7) != 4) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index 189ef2b..7f18a53 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -2030,7 +2030,8 @@ static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring, * on the gfx ring for execution by the GPU. */ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { u32 header, control = 0; u32 next_rptr = ring->wptr + 5; @@ -2056,7 +2057,7 @@ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, else header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);

- control |= ib->length_dw | (ib->vm_id << 24); + control |= ib->length_dw | (vm_id << 24);

amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, @@ -2069,7 +2070,8 @@ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, }

static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { u32 header, control = 0; u32 next_rptr = ring->wptr + 5; @@ -2084,7 +2086,7 @@ static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,

header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);

- control |= ib->length_dw | (ib->vm_id << 24); + control |= ib->length_dw | (vm_id << 24);

amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 0d556c9..92647fb 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -5646,7 +5646,8 @@ static void gfx_v8_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) }

static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { u32 header, control = 0; u32 next_rptr = ring->wptr + 5; @@ -5672,7 +5673,7 @@ static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, else header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);

- control |= ib->length_dw | (ib->vm_id << 24); + control |= ib->length_dw | (vm_id << 24);

amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, @@ -5685,7 +5686,8 @@ static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, }

static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { u32 header, control = 0; u32 next_rptr = ring->wptr + 5; @@ -5701,7 +5703,7 @@ static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,

header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);

- control |= ib->length_dw | (ib->vm_id << 24); + control |= ib->length_dw | (vm_id << 24);

amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index de94adb..f4c3130 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c @@ -242,9 +242,10 @@ static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (VI). */ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { - u32 vmid = ib->vm_id & 0xf; + u32 vmid = vm_id & 0xf; u32 next_rptr = ring->wptr + 5;

while ((next_rptr & 7) != 2) diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c index ca2aee3..063f08a 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c @@ -400,9 +400,10 @@ static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) * Schedule an IB in the DMA ring (VI). */ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { - u32 vmid = ib->vm_id & 0xf; + u32 vmid = vm_id & 0xf; u32 next_rptr = ring->wptr + 5;

while ((next_rptr & 7) != 2) diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c index a75ffb5b..f075514 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c @@ -489,7 +489,8 @@ static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0)); amdgpu_ring_write(ring, ib->gpu_addr); diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c index ecb8101..e0a76a8 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c @@ -539,7 +539,8 @@ static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0)); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c index a43f1a7..c9929d6 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c @@ -631,7 +631,8 @@ static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring) * Write ring commands to execute the indirect buffer */ static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, bool ctx_switch) + struct amdgpu_ib *ib, + unsigned vm_id, bool ctx_switch) { amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0)); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));

-- 2.5.5

Alex Deucher

7:35 p.m.

New subject: [PATCH] drm/amdgpu: fix and cleanup user fence handling v2

From: Christian König christian.koenig@amd.com

We leaked the BO in the error pass, additional to that we only have one user fence for all IBs in a job.

v2: remove white space changes

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 7cff077..a9dfb8e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -368,13 +368,6 @@ struct amdgpu_fence_driver { #define AMDGPU_FENCE_FLAG_64BIT (1 << 0) #define AMDGPU_FENCE_FLAG_INT (1 << 1)

-struct amdgpu_user_fence { - /* write-back bo */ - struct amdgpu_bo *bo; - /* write-back address offset to bo start */ - uint32_t offset; -}; - int amdgpu_fence_driver_init(struct amdgpu_device *adev); void amdgpu_fence_driver_fini(struct amdgpu_device *adev); void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev); @@ -741,10 +734,7 @@ struct amdgpu_ib { uint32_t length_dw; uint64_t gpu_addr; uint32_t *ptr; - struct amdgpu_user_fence *user; uint32_t flags; - /* resulting sequence number */ - uint64_t sequence; };

enum amdgpu_ring_type { @@ -1219,7 +1209,7 @@ void amdgpu_ring_fini(struct amdgpu_ring *ring); struct amdgpu_cs_chunk { uint32_t chunk_id; uint32_t length_dw; - uint32_t *kdata; + void *kdata; };

struct amdgpu_cs_parser { @@ -1263,7 +1253,12 @@ struct amdgpu_job { uint32_t gds_base, gds_size; uint32_t gws_base, gws_size; uint32_t oa_base, oa_size; - struct amdgpu_user_fence uf; + + /* user fence handling */ + struct amdgpu_bo *uf_bo; + uint32_t uf_offset; + uint64_t uf_sequence; + }; #define to_amdgpu_job(sched_job) \ container_of((sched_job), struct amdgpu_job, base) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 9ab2f08..2bbeeb0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -87,33 +87,30 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, }

static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p, - struct amdgpu_user_fence *uf, - struct drm_amdgpu_cs_chunk_fence *fence_data) + struct drm_amdgpu_cs_chunk_fence *data, + uint32_t *offset) { struct drm_gem_object *gobj; - uint32_t handle;

- handle = fence_data->handle; gobj = drm_gem_object_lookup(p->adev->ddev, p->filp, - fence_data->handle); + data->handle); if (gobj == NULL) return -EINVAL;

- uf->bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj)); - uf->offset = fence_data->offset; - - if (amdgpu_ttm_tt_get_usermm(uf->bo->tbo.ttm)) { - drm_gem_object_unreference_unlocked(gobj); - return -EINVAL; - } - - p->uf_entry.robj = amdgpu_bo_ref(uf->bo); + p->uf_entry.robj = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj)); p->uf_entry.priority = 0; p->uf_entry.tv.bo = &p->uf_entry.robj->tbo; p->uf_entry.tv.shared = true; p->uf_entry.user_pages = NULL; + *offset = data->offset;

drm_gem_object_unreference_unlocked(gobj); + + if (amdgpu_ttm_tt_get_usermm(p->uf_entry.robj->tbo.ttm)) { + amdgpu_bo_unref(&p->uf_entry.robj); + return -EINVAL; + } + return 0; }

@@ -124,8 +121,8 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) union drm_amdgpu_cs *cs = data; uint64_t *chunk_array_user; uint64_t *chunk_array; - struct amdgpu_user_fence uf = {}; unsigned size, num_ibs = 0; + uint32_t uf_offset = 0; int i; int ret;

@@ -200,7 +197,8 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) goto free_partial_kdata; }

- ret = amdgpu_cs_user_fence_chunk(p, &uf, (void *)p->chunks[i].kdata); + ret = amdgpu_cs_user_fence_chunk(p, p->chunks[i].kdata, + &uf_offset); if (ret) goto free_partial_kdata;

@@ -219,7 +217,10 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) if (ret) goto free_all_kdata;

- p->job->uf = uf; + if (p->uf_entry.robj) { + p->job->uf_bo = amdgpu_bo_ref(p->uf_entry.robj); + p->job->uf_offset = uf_offset; + }

kfree(chunk_array); return 0; @@ -377,7 +378,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, INIT_LIST_HEAD(&duplicates); amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);

- if (p->job->uf.bo) + if (p->uf_entry.robj) list_add(&p->uf_entry.tv.head, &p->validated);

if (need_mmap_lock) @@ -760,17 +761,11 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, j++; }

- /* wrap the last IB with user fence */ - if (parser->job->uf.bo) { - struct amdgpu_ib *ib = &parser->job->ibs[parser->job->num_ibs - 1]; - - /* UVD & VCE fw doesn't support user fences */ - if (parser->job->ring->type == AMDGPU_RING_TYPE_UVD || - parser->job->ring->type == AMDGPU_RING_TYPE_VCE) - return -EINVAL; - - ib->user = &parser->job->uf; - } + /* UVD & VCE fw doesn't support user fences */ + if (parser->job->uf_bo && ( + parser->job->ring->type == AMDGPU_RING_TYPE_UVD || + parser->job->ring->type == AMDGPU_RING_TYPE_VCE)) + return -EINVAL;

return 0; } @@ -856,7 +851,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, job->ctx = entity->fence_context; p->fence = fence_get(fence); cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, fence); - job->ibs[job->num_ibs - 1].sequence = cs->out.handle; + job->uf_sequence = cs->out.handle;

trace_amdgpu_cs_ioctl(job); amd_sched_entity_push_job(&job->base); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index 201aceb..34e3542 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -203,10 +203,11 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, }

/* wrap the last IB with fence */ - if (ib->user) { - uint64_t addr = amdgpu_bo_gpu_offset(ib->user->bo); - addr += ib->user->offset; - amdgpu_ring_emit_fence(ring, addr, ib->sequence, + if (job && job->uf_bo) { + uint64_t addr = amdgpu_bo_gpu_offset(job->uf_bo); + + addr += job->uf_offset; + amdgpu_ring_emit_fence(ring, addr, job->uf_sequence, AMDGPU_FENCE_FLAG_64BIT); }

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 8ea68d0..f0dafa5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -97,7 +97,7 @@ void amdgpu_job_free(struct amdgpu_job *job) amdgpu_sa_bo_free(job->adev, &job->ibs[i].sa_bo, f); fence_put(job->fence);

- amdgpu_bo_unref(&job->uf.bo); + amdgpu_bo_unref(&job->uf_bo); amdgpu_sync_free(&job->sync);

if (!job->base.use_sched)

-- 2.5.5

3293

Age (days ago)

3293

Last active (days ago)

dri-devel@lists.freedesktop.org

5 comments

1 participants

tags (0)

participants (1)

Alex Deucher