From 22c01cc48301f6974868bd4a7b03e29883da1103 Mon Sep 17 00:00:00 2001
From: Anatoli Antonovitch <anatoli.antonovitch@amd.com>
Date: Thu, 3 Sep 2015 11:13:31 -0400
Subject: drm/amdgpu: execution barrier after fence v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Insert wait for reg mem after EOP to fix potential issue with vm context switch
v2: move wait to vm_flush() use equal instead of greater than.
Signed-off-by: Anatoli Antonovitch <anatoli.antonovitch@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
(limited to 'drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 53f07439a512..0f979abca235 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -3965,6 +3965,7 @@ static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
amdgpu_ring_write(ring, lower_32_bits(seq));
amdgpu_ring_write(ring, upper_32_bits(seq));
+
}
/**
@@ -4044,6 +4045,17 @@ static void gfx_v8_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[ring->idx];
+ 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 */
+ 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) |
--
cgit v1.2.3
From 20a85ff846ffed84fba8637abbb6b1c96436c0ac Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Sat, 5 Sep 2015 11:59:50 +0200
Subject: drm/amdgpu: use write confirm for vm_flush()
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Make sure the CP waits for the write to be confirmed before
invalidating.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 0f979abca235..d6d330bc6daa 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -4059,7 +4059,8 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
- WRITE_DATA_DST_SEL(0)));
+ WRITE_DATA_DST_SEL(0)) |
+ WR_CONFIRM);
if (vm_id < 8) {
amdgpu_ring_write(ring,
(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vm_id));
--
cgit v1.2.3
From 72d7668b5ba5180b651e8a07dd6ed62e4e26f207 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Thu, 3 Sep 2015 17:34:59 +0200
Subject: drm/amdgpu: export reservation_object from dmabuf to ttm (v2)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Adds an extra argument to amdgpu_bo_create, which is only used in amdgpu_prime.c.
Port of radeon commit 831b6966a60fe72d85ae3576056b4e4e0775b112.
v2: fix up kfd.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index d6d330bc6daa..78e5900d71cd 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -868,7 +868,7 @@ static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
r = amdgpu_bo_create(adev,
adev->gfx.mec.num_mec *adev->gfx.mec.num_pipe * MEC_HPD_SIZE * 2,
PAGE_SIZE, true,
- AMDGPU_GEM_DOMAIN_GTT, 0, NULL,
+ AMDGPU_GEM_DOMAIN_GTT, 0, NULL, NULL,
&adev->gfx.mec.hpd_eop_obj);
if (r) {
dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
@@ -995,21 +995,21 @@ static int gfx_v8_0_sw_init(void *handle)
/* reserve GDS, GWS and OA resource for gfx */
r = amdgpu_bo_create(adev, adev->gds.mem.gfx_partition_size,
PAGE_SIZE, true,
- AMDGPU_GEM_DOMAIN_GDS, 0,
+ AMDGPU_GEM_DOMAIN_GDS, 0, NULL,
NULL, &adev->gds.gds_gfx_bo);
if (r)
return r;
r = amdgpu_bo_create(adev, adev->gds.gws.gfx_partition_size,
PAGE_SIZE, true,
- AMDGPU_GEM_DOMAIN_GWS, 0,
+ AMDGPU_GEM_DOMAIN_GWS, 0, NULL,
NULL, &adev->gds.gws_gfx_bo);
if (r)
return r;
r = amdgpu_bo_create(adev, adev->gds.oa.gfx_partition_size,
PAGE_SIZE, true,
- AMDGPU_GEM_DOMAIN_OA, 0,
+ AMDGPU_GEM_DOMAIN_OA, 0, NULL,
NULL, &adev->gds.oa_gfx_bo);
if (r)
return r;
@@ -3106,7 +3106,7 @@ static int gfx_v8_0_cp_compute_resume(struct amdgpu_device *adev)
sizeof(struct vi_mqd),
PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_GTT, 0, NULL,
- &ring->mqd_obj);
+ NULL, &ring->mqd_obj);
if (r) {
dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
return r;
--
cgit v1.2.3
From 5c3422b0b135b46c8dca9c1d909c1ae84f3561bd Mon Sep 17 00:00:00 2001
From: "monk.liu" <monk.liu@amd.com>
Date: Wed, 23 Sep 2015 13:49:58 +0800
Subject: drm/amdgpu: sync ce and me with SWITCH_BUFFER(2)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
we used to adopt wait_reg_mem to let CE wait before DE finish page
updating, but from Tonga+, CE doesn't support wait_reg_mem package so
this logic no longer works.
so here is another approach to do same thing:
Insert two of SWITCH_BUFFER at both front and end of vm_flush can
guarantee that CE not go further to process IB_const before vm_flush
done.
Insert two of SWITCH_BUFFER also works on CI, so remove legency method
to sync CE and ME
v2:
Insert double SWITCH_BUFFER at front of vm flush as well.
Signed-off-by: monk.liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 58 ++++++++---------------------------
1 file changed, 12 insertions(+), 46 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 78e5900d71cd..cb4f68f53f24 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -940,12 +940,6 @@ static int gfx_v8_0_sw_init(void *handle)
return r;
}
- r = amdgpu_wb_get(adev, &adev->gfx.ce_sync_offs);
- if (r) {
- DRM_ERROR("(%d) gfx.ce_sync_offs wb alloc failed\n", r);
- return r;
- }
-
/* set up the gfx ring */
for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
ring = &adev->gfx.gfx_ring[i];
@@ -1033,8 +1027,6 @@ static int gfx_v8_0_sw_fini(void *handle)
for (i = 0; i < adev->gfx.num_compute_rings; i++)
amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
- amdgpu_wb_free(adev, adev->gfx.ce_sync_offs);
-
gfx_v8_0_mec_fini(adev);
return 0;
@@ -4006,41 +3998,6 @@ static bool gfx_v8_0_ring_emit_semaphore(struct amdgpu_ring *ring,
return true;
}
-static void gfx_v8_0_ce_sync_me(struct amdgpu_ring *ring)
-{
- struct amdgpu_device *adev = ring->adev;
- u64 gpu_addr = adev->wb.gpu_addr + adev->gfx.ce_sync_offs * 4;
-
- /* instruct DE to set a magic number */
- amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
- amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
- WRITE_DATA_DST_SEL(5)));
- amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
- amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
- amdgpu_ring_write(ring, 1);
-
- /* let CE wait till condition satisfied */
- amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
- amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
- WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
- WAIT_REG_MEM_FUNCTION(3) | /* == */
- WAIT_REG_MEM_ENGINE(2))); /* ce */
- amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
- amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
- amdgpu_ring_write(ring, 1);
- amdgpu_ring_write(ring, 0xffffffff);
- amdgpu_ring_write(ring, 4); /* poll interval */
-
- /* instruct CE to reset wb of ce_sync to zero */
- amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
- amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
- WRITE_DATA_DST_SEL(5) |
- WR_CONFIRM));
- amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
- amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
- amdgpu_ring_write(ring, 0);
-}
-
static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
unsigned vm_id, uint64_t pd_addr)
{
@@ -4057,6 +4014,14 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
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));
+ amdgpu_ring_write(ring, 0);
+ amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+ amdgpu_ring_write(ring, 0);
+ }
+
amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
WRITE_DATA_DST_SEL(0)) |
@@ -4096,9 +4061,10 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
/* sync PFP to ME, otherwise we might get invalid PFP reads */
amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
amdgpu_ring_write(ring, 0x0);
-
- /* synce CE with ME to prevent CE fetch CEIB before context switch done */
- gfx_v8_0_ce_sync_me(ring);
+ amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+ amdgpu_ring_write(ring, 0);
+ amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+ amdgpu_ring_write(ring, 0);
}
}
--
cgit v1.2.3