drm/amdgpu/sriov: For sriov runtime, use kiq to do invalidate tlb
Subsystems (2)
PATCH 1/2 drm/amdgpu/sriov: For sriov runtime, use kiq to do invalidate tlb
To avoid the tlb flush not interrupted by world switch, use kiq and one command to do tlb invalidate.
v2:
Add firmware version checking.
v3:
Refine the code, and move the firmware checking into gfx_v9_0_ring_emit_reg_write_reg_wait.
v4:
Change the name amdgpu_kiq_invalidate_tlb to amdgpu_kiq_reg_write_reg_wait. Remove the in_interrupt.
Refine code.
SWDEV-161497
--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 3 -- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 15 +++++++- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 60 ++++++++++++++++++++++++++++++++ 4 files changed, 78 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +4 -0
@@ -212,6 +212,10 @@ enum amdgpu_kiq_irq { AMDGPU_CP_KIQ_IRQ_LAST }; +#define MAX_KIQ_REG_WAIT 5000/* in usecs, 5ms */ +#define MAX_KIQ_REG_BAILOUT_INTERVAL 5/* in msecs, 5ms */ +#define MAX_KIQ_REG_TRY 20 +int amdgpu_device_ip_set_clockgating_state (void *dev,enum amd_ip_block_type block_type,enum amd_clockgating_state state);
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +0 -3
@@ -22,9 +22,6 @@*/ #include "amdgpu.h" -#define MAX_KIQ_REG_WAIT 5000/* in usecs, 5ms */ -#define MAX_KIQ_REG_BAILOUT_INTERVAL 5/* in msecs, 5ms */ -#define MAX_KIQ_REG_TRY 20uint64_t amdgpu_csa_vaddr (struct amdgpu_device *adev) {
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +14 -1
@@ -4348,8 +4348,21 @@ static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, uint32_t ref, uint32_t mask) { int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); +struct amdgpu_device *adev = ring->adev; +bool fw_version_ok = false; -if (amdgpu_sriov_vf (ring->adev)) +if (ring->funcs->type ==AMDGPU_RING_TYPE_GFX ) { +if ((adev->gfx.me_fw_version >=0x0000009c ) && (adev->gfx.me_feature_version >=42 ) + && (adev->gfx.pfp_fw_version >=0x000000b1 ) && (adev->gfx.pfp_feature_version >=42 )) + fw_version_ok = true; + }else { +if ((adev->gfx.mec_fw_version >=0x00000193 ) && (adev->gfx.mec_feature_version >=42 )) + fw_version_ok = true; + } + + fw_version_ok = (adev->asic_type ==CHIP_VEGA10 ) ? fw_version_ok : false; + +if (amdgpu_sriov_vf (adev) && fw_version_ok)gfx_v9_0_wait_reg_mem (ring, usepfp,0 ,1 , reg0, reg1, ref, mask,0x20 );else
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +60 -0
78 lines not on this page · raw message
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): 1534337675-25039-1-git-send-email-Emily.Deng-5C7GfCeVMHo@public.gmane.org
PATCH 2/2 drm/amdgpu: For bare metal, use kiq to do invalidate tlb
Remove the amdgpu_sriov_vf checking.
--- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 8 +++----- 2 files changed, 4 insertions(+), 6 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +1 -1
@@ -4362,7 +4362,7 @@ static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, fw_version_ok = (adev->asic_type ==CHIP_VEGA10 ) ? fw_version_ok : false; -if (amdgpu_sriov_vf (adev) && fw_version_ok) +if (fw_version_ok)gfx_v9_0_wait_reg_mem (ring, usepfp,0 ,1 , reg0, reg1, ref, mask,0x20 );else
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +3 -5
@@ -393,11 +393,9 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,struct amdgpu_vmhub *hub = &adev->vmhub[i];u32 tmp =gmc_v9_0_get_invalidate_req (vmid); -if (amdgpu_sriov_vf (adev)) { - r =amdgpu_kiq_reg_write_reg_wait (adev, hub, eng, tmp, vmid); -if (!r) -continue ; - } + r =amdgpu_kiq_reg_write_reg_wait (adev, hub, eng, tmp, vmid); +if (!r) +continue ;WREG32_NO_KIQ (hub->vm_inv_eng0_req + eng, tmp);
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Am 15.08.2018 um 14:54 schrieb Emily Deng:
Quoted 60 lines
To avoid the tlb flush not interrupted by world switch, use kiq and one command to do tlb invalidate.
v2:
Add firmware version checking.v3:
Refine the code, and move the firmware checking intogfx_v9_0_ring_emit_reg_write_reg_wait.v4:
Change the nameamdgpu_kiq_invalidate_tlbtoamdgpu_kiq_reg_write_reg_wait. Remove thein_interrupt.
Refine code.SWDEV-161497
Signed-off-by Emily Deng <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 3 -- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 15 +++++++- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 60 ++++++++++++++++++++++++++++++++ 4 files changed, 78 insertions(+), 4 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +4
-0@@ -212,6 +212,10 @@ enum amdgpu_kiq_irq { AMDGPU_CP_KIQ_IRQ_LAST }; +#define MAX_KIQ_REG_WAIT 5000/* in usecs, 5ms */ +#define MAX_KIQ_REG_BAILOUT_INTERVAL 5/* in msecs, 5ms */ +#define MAX_KIQ_REG_TRY 20 +int amdgpu_device_ip_set_clockgating_state (void *dev,enum amd_ip_block_type block_type,enum amd_clockgating_state state);drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +0
-3@@ -22,9 +22,6 @@*/ #include "amdgpu.h" -#define MAX_KIQ_REG_WAIT 5000/* in usecs, 5ms */ -#define MAX_KIQ_REG_BAILOUT_INTERVAL 5/* in msecs, 5ms */ -#define MAX_KIQ_REG_TRY 20uint64_t amdgpu_csa_vaddr (struct amdgpu_device *adev) {drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +0
-0@@ -4348,8 +4348,21 @@ static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, uint32_t ref, uint32_t mask)
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, · line 4357
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +6
-1{int usepfp = (ring->funcs->type ==AMDGPU_RING_TYPE_GFX ); +struct amdgpu_device *adev = ring->adev; +bool fw_version_ok = false; -if (amdgpu_sriov_vf (ring->adev)) +if (ring->funcs->type ==AMDGPU_RING_TYPE_GFX ) { +if ((adev->gfx.me_fw_version >=0x0000009c ) && (adev->gfx.me_feature_version >=42 ) + && (adev->gfx.pfp_fw_version >=0x000000b1 ) && (adev->gfx.pfp_feature_version >=42 )) + fw_version_ok = true;
The indentation here is wrong and the && belongs at the end of the line.
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, · line 4365
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +8
-0+ }else { +if ((adev->gfx.mec_fw_version >=0x00000193 ) && (adev->gfx.mec_feature_version >=42 )) + fw_version_ok = true; + } + + fw_version_ok = (adev->asic_type ==CHIP_VEGA10 ) ? fw_version_ok : false; + +if (amdgpu_sriov_vf (adev) && fw_version_ok)
Please squash the second patch into this one. The goal is to use the same code path on both SRIOV and bare metal.
I would rather separate our this change into a patch and then make the TLB invalidation the second patch.
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid) · line 315
Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +0
-0gfx_v9_0_wait_reg_mem (ring, usepfp,0 ,1 , reg0, reg1, ref, mask,0x20 ); elsedrivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +2
-0@@ -311,6 +311,59 @@ static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid)return req; } +signed long amdgpu_kiq_reg_write_reg_wait (struct amdgpu_device *adev,struct amdgpu_vmhub *hub, +unsigned eng,u32 req,uint32_t vmid)
That is still specialized, e.g. you assume that you work with a VMHUB here instead of specifying the registers, req and value separately.
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid) · line 325
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +10
-0+{ +signed long r, cnt =0 ; +unsigned long flags; +uint32_t seq; +struct amdgpu_kiq *kiq = &adev->gfx.kiq; +struct amdgpu_ring *ring = &kiq->ring; + +if (!ring->ready) { +return -EINVAL ; + }
Please drop the {} here.
Christian.
Quoted 55 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +42
-0+ +spin_lock_irqsave (&kiq->ring_lock, flags); + +amdgpu_ring_alloc (ring,32 ); +amdgpu_ring_emit_reg_write_reg_wait (ring, hub->vm_inv_eng0_req + eng, + hub->vm_inv_eng0_ack + eng, + req,1 << vmid); +amdgpu_fence_emit_polling (ring, &seq); +amdgpu_ring_commit (ring); +spin_unlock_irqrestore (&kiq->ring_lock, flags); + + r =amdgpu_fence_wait_polling (ring, seq,MAX_KIQ_REG_WAIT ); + +/* don't wait anymore for gpu reset case because this way may +* block gpu_recover() routine forever, e.g. this virt_kiq_rreg +* is triggered in TTM and ttm_bo_lock_delayed_workqueue() will +* never return if we keep waiting in virt_kiq_rreg, which cause +* gpu_recover() hang there. +* +* also don't wait anymore for IRQ context +* */ +if (r <1 && (adev->in_gpu_reset ||in_interrupt ())) + goto failed_kiq; + +might_sleep (); + +while (r <1 && cnt++ <MAX_KIQ_REG_TRY ) { +msleep (MAX_KIQ_REG_BAILOUT_INTERVAL ); + r =amdgpu_fence_wait_polling (ring, seq,MAX_KIQ_REG_WAIT ); + } + +if (cnt >MAX_KIQ_REG_TRY ) + goto failed_kiq; + +return 0 ; + +failed_kiq: +pr_err ("failed to invalidate tlb with kiq\n" ); +return r; +} +/* * GART * VMID 0 is the physical GPU addresses as used by the kernel. @@ -332,6 +385,7 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,/* Use register 17 for GART */ const unsigned eng =17 ;unsigned i, j; +int r;spin_lock (&adev->gmc.invalidate_lock);@@ -339,6 +393,12 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,struct amdgpu_vmhub *hub = &adev->vmhub[i];u32 tmp =gmc_v9_0_get_invalidate_req (vmid);
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, · line 404
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +6
-0+if (amdgpu_sriov_vf (adev)) { + r =amdgpu_kiq_reg_write_reg_wait (adev, hub, eng, tmp, vmid); +if (!r) +continue ; + } +WREG32_NO_KIQ (hub->vm_inv_eng0_req + eng, tmp);/* Busy wait for ACK.*/
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx