static CSA for SRIOV serials
Subsystems (2)
cover static CSA for SRIOV serials
per device CSA is needed by world switch preemption for SRIOV virtualization we create CSA per device, and map it on each VM in vm_init, and update CSA BO bo_va in each submission.
*** BLURB HERE ***
Monk Liu (7):
drm/amdgpu:add new file for SRIOV
drm/amdgpu:new field members for SRIOV
drm/amdgpu:impl static CSA alloc/dealloc
drm/amdgpu:impl static CSA map/unmap routines
drm/amdgpu:set static csa function
drm/amdgpu:alloc/dealloc csa accordingly
drm/amdgpu:map/unmap static csa accordingly
drivers/gpu/drm/amd/amdgpu/Makefile | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 14 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 142 +++++++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 21 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 11 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 + drivers/gpu/drm/amd/amdgpu/vi.c | 3 + 8 files changed, 201 insertions(+), 1 deletion(-) create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
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): 1483948982-27056-1-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org
PATCH 1/7 drm/amdgpu:add new file for SRIOV
--- drivers/gpu/drm/amd/amdgpu/Makefile | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+), 1 deletion(-) create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
drivers/gpu/drm/amd/amdgpu/Makefile +1 -1
@@ -30,7 +30,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \ atombios_encoders.o amdgpu_sa.o atombios_i2c.o \ amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \ amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \ - amdgpu_gtt_mgr.o amdgpu_vram_mgr.o + amdgpu_gtt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o # add asic specific block amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c new file +23 -0
@@ -0,0 +1,23 @@ +/* +* Copyright 2016 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. +* +*/ +#include "amdgpu.h"
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): 1483948982-27056-2-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 1/7] drm/amdgpu:add new file for SRIOV
Am 09.01.2017 um 09:02 schrieb Monk Liu:
Quoted 2 lines
Change-Id I7e21c52e5b0c7ef39a706a3191293f3d7b5a7e08Signed-off-by Monk Liu <[email protected]>
Just adding the file without any functionality doesn't make much sense, please squash into the other commits.
Christian.
Quoted 39 lines
--- drivers/gpu/drm/amd/amdgpu/Makefile | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+), 1 deletion(-)create mode 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.cdrivers/gpu/drm/amd/amdgpu/Makefile +1
-1@@ -30,7 +30,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \ atombios_encoders.o amdgpu_sa.o atombios_i2c.o \ amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \ amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \ - amdgpu_gtt_mgr.o amdgpu_vram_mgr.o + amdgpu_gtt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o # add asic specific block amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c new file +14
-0@@ -0,0 +1,23 @@ +/* +* Copyright 2016 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
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c line 23
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c new file +9
-0+* 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. +* +*/ +#include "amdgpu.h"
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 2/7 drm/amdgpu:new field members for SRIOV
In reply to a message not in the archive--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 19 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 ++ 2 files changed, 21 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +19 -0
@@ -28,9 +28,28 @@#define AMDGPU_SRIOV_CAPS_ENABLE_IOV (1 << 1)/* sr-iov is enabled on this GPU */ #define AMDGPU_SRIOV_CAPS_IS_VF (1 << 2)/* this GPU is a virtual function */ #define AMDGPU_PASSTHROUGH_MODE (1 << 3)/* thw whole GPU is pass through for VM */ + +struct amdgpu_vm ;/* GPU virtualization */ struct amdgpu_virt {uint32_t caps; +uint32_t csa_size; +struct amdgpu_bo *csa_obj; +uint64_t csa_vmid0_addr; +uint64_t gds_vmid0_addr; +int (*allocate_csa)(struct amdgpu_device *adev); +void (*deallocate_csa)(struct amdgpu_device *adev); +int (*map_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); +void (*unmap_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); +}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va; +/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr; };#define amdgpu_sriov_enabled (adev) \
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +2 -0
@@ -29,6 +29,7 @@#include "gpu_scheduler.h" #include "amdgpu_sync.h" #include "amdgpu_ring.h" +#include "amdgpu_virt.h" struct amdgpu_bo_va ;struct amdgpu_job ;
@@ -111,6 +112,7 @@ struct amdgpu_vm {/* client id */ u64 client_id; +struct amdgpu_vm_virt vm_virt; };struct amdgpu_vm_id {
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): 1483948982-27056-3-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 09.01.2017 um 09:02 schrieb Monk Liu:
Quoted 18 lines
Change-Id Ife0eff7b13b8b5946f005a39f6ecb8db1cb72c38Signed-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 19 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 ++ 2 files changed, 21 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +2
-0@@ -28,9 +28,28 @@#define AMDGPU_SRIOV_CAPS_ENABLE_IOV (1 << 1)/* sr-iov is enabled on this GPU */ #define AMDGPU_SRIOV_CAPS_IS_VF (1 << 2)/* this GPU is a virtual function */ #define AMDGPU_PASSTHROUGH_MODE (1 << 3)/* thw whole GPU is pass through for VM */ + +struct amdgpu_vm ;/* GPU virtualization */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 43
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +8
-0struct amdgpu_virt {uint32_t caps; +uint32_t csa_size; +struct amdgpu_bo *csa_obj; +uint64_t csa_vmid0_addr; +uint64_t gds_vmid0_addr; +int (*allocate_csa)(struct amdgpu_device *adev); +void (*deallocate_csa)(struct amdgpu_device *adev); +int (*map_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); +void (*unmap_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm);
Why callbacks for this?
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;
Please put that directly into the amdgpu_vm structure.
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;
That should be constant, shouldn't it?
Regards,
Christian.
Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0};#define amdgpu_sriov_enabled (adev) \drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -29,6 +29,7 @@#include "gpu_scheduler.h" #include "amdgpu_sync.h" #include "amdgpu_ring.h" +#include "amdgpu_virt.h" struct amdgpu_bo_va ;struct amdgpu_job ;
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h struct amdgpu_vm { · line 118
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -111,6 +112,7 @@ struct amdgpu_vm {/* client id */ u64 client_id; +struct amdgpu_vm_virt vm_virt; };struct amdgpu_vm_id {
_______________________________________________
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): 9f6b9ab8-3bb1-7a57-75d9-70d23d5dd74b-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
答复: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;
That should be constant, shouldn't it?
[ML] do you mean "const uint64_t vm_csa_addr" ? of cause not if that's your point ... compiling error on that
and strictly speaking that's not constant, the address is calculated ...
________________________________
发件人: Christian König <[email protected]> 发送时间: 2017年1月9日 18:57:06
收件人: Liu, Monk; [email protected]
主题: Re: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 09.01.2017 um 09:02 schrieb Monk Liu:
Quoted 18 lines
Change-Id Ife0eff7b13b8b5946f005a39f6ecb8db1cb72c38Signed-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 19 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 ++ 2 files changed, 21 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +2
-0@@ -28,9 +28,28 @@#define AMDGPU_SRIOV_CAPS_ENABLE_IOV (1 << 1)/* sr-iov is enabled on this GPU */ #define AMDGPU_SRIOV_CAPS_IS_VF (1 << 2)/* this GPU is a virtual function */ #define AMDGPU_PASSTHROUGH_MODE (1 << 3)/* thw whole GPU is pass through for VM */ + +struct amdgpu_vm ;/* GPU virtualization */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 43
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +8
-0struct amdgpu_virt {uint32_t caps; +uint32_t csa_size; +struct amdgpu_bo *csa_obj; +uint64_t csa_vmid0_addr; +uint64_t gds_vmid0_addr; +int (*allocate_csa)(struct amdgpu_device *adev); +void (*deallocate_csa)(struct amdgpu_device *adev); +int (*map_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); +void (*unmap_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm);
Why callbacks for this?
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;
Please put that directly into the amdgpu_vm structure.
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;
That should be constant, shouldn't it?
Regards,
Christian.
Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0};#define amdgpu_sriov_enabled (adev) \drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -29,6 +29,7 @@#include "gpu_scheduler.h" #include "amdgpu_sync.h" #include "amdgpu_ring.h" +#include "amdgpu_virt.h" struct amdgpu_bo_va ;struct amdgpu_job ;
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h struct amdgpu_vm { · line 118
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -111,6 +112,7 @@ struct amdgpu_vm {/* client id */ u64 client_id; +struct amdgpu_vm_virt vm_virt; };struct amdgpu_vm_id {
_______________________________________________
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): BY2PR1201MB11106E7ABC5FF87AF1F6F67A84670-O28G1zQ8oGliQkyLPkmea2rFom/aUZj6nBOFsp37pqbUKgpGm//BTAC/[email protected]
Re: 答复: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 10.01.2017 um 04:03 schrieb Liu, Monk:
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;
Quoted 8 lines
That should be constant, shouldn't it?
[ML] do you mean "const
uint64_tvm_csa_addr" ? of cause not if that's your point ... compiling error on thatand strictly speaking that's not constant, the address is calculated ...
No that wasn't what I meant. The address is constant for all VM's in the system, isn't it?
Just add a define similar to AMDGPU_VA_RESERVED_SIZE and stop calculating it all together (BTW: AMDGPU_VA_RESERVED_SIZE should probably be moved to amdgpu_vm.h).
Regards,
Christian.
Quoted 83 lines
------------------------------------------------------------------------
*发件人:* Christian König <[email protected]> *发送时间:* 2017年1月9日 18:57:06
*收件人:* Liu, Monk; [email protected] *主题:* Re: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 09.01.2017 um 09:02 schrieb Monk Liu:Review
Quoted 8 lines
Change-Id Ife0eff7b13b8b5946f005a39f6ecb8db1cb72c38Signed-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 19 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 ++ 2 files changed, 21 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.hdrivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 28
Quoted 5 lines
index 0d821d9..5aa7f0c 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0@@ -28,9 +28,28 @@#define AMDGPU_SRIOV_CAPS_ENABLE_IOV (1 << 1)/* sr-iov is enabled on this GPU */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 29
Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0#define AMDGPU_SRIOV_CAPS_IS_VF (1 << 2)/* this GPU is a virtual function */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 30
Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0#define AMDGPU_PASSTHROUGH_MODE (1 << 3)/* thw whole GPU is pass through for VM */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h
Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +10
-0+ +struct amdgpu_vm ;/* GPU virtualization */ struct amdgpu_virt {uint32_t caps; +uint32_t csa_size; +struct amdgpu_bo *csa_obj; +uint64_t csa_vmid0_addr; +uint64_t gds_vmid0_addr; +int (*allocate_csa)(struct amdgpu_device *adev); +void (*deallocate_csa)(struct amdgpu_device *adev); +int (*map_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); + void (*unmap_csa)(struct amdgpu_device *adev, struct amdgpu_vm*vm);
Why callbacks for this?
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;Please put that directly into the
amdgpu_vmstructure.drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;That should be constant, shouldn't it?
Regards,
Christian.Review
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0};#define amdgpu_sriov_enabled (adev) \(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hQuoted 18 lines
index 42a629b..27cbcbc 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +2
-0@@ -29,6 +29,7 @@#include "gpu_scheduler.h" #include "amdgpu_sync.h" #include "amdgpu_ring.h" +#include "amdgpu_virt.h" struct amdgpu_bo_va ;struct amdgpu_job ;@@ -111,6 +112,7 @@ struct amdgpu_vm {/* client id */ u64 client_id; +struct amdgpu_vm_virt vm_virt; };
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h struct amdgpu_vm { · line 118
Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0struct amdgpu_vm_id {_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
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): aec66a0f-0fd7-1e02-f160-999fc76de4a6-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
答复: 答复: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
NAH, for AI/VI the gds_addr is the bottom half of CSA area (8kb), but we don't know for later chips ...
I prefer keep a member field in the struct for flexibility ...
regards with put those member into amdgpu_vm structure, if you really prefer that way I'm fine by moving them, not too important anyway.
BR Monk
________________________________
发件人: amd-gfx <[email protected]> 代表 Christian König <[email protected]> 发送时间: 2017年1月10日 17:51:46
收件人: Liu, Monk; [email protected]
主题: Re: 答复: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 10.01.2017 um 04:03 schrieb Liu, Monk:
Review
Quoted 3 lines
(quoted diff) +3
-0+ /* virtual MC address of CSA & GDS for each VM */ + uint64_t vm_csa_addr; + uint64_t vm_gds_addr;
That should be constant, shouldn't it?
[ML] do you mean "const uint64_t vm_csa_addr" ? of cause not if that's your point ... compiling error on that
and strictly speaking that's not constant, the address is calculated ...
No that wasn't what I meant. The address is constant for all VM's in the system, isn't it?
Just add a define similar to AMDGPU_VA_RESERVED_SIZE and stop calculating it all together (BTW: AMDGPU_VA_RESERVED_SIZE should probably be moved to amdgpu_vm.h).
Regards,
Christian.
________________________________
发件人: Christian König <[email protected]><mailto:[email protected]> 发送时间: 2017年1月9日 18:57:06
收件人: Liu, Monk; [email protected]<mailto:[email protected]> 主题: Re: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 09.01.2017 um 09:02 schrieb Monk Liu:
Quoted 18 lines
Change-Id Ife0eff7b13b8b5946f005a39f6ecb8db1cb72c38Signed-off-by Monk Liu <[email protected]><mailto:[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 19 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 ++ 2 files changed, 21 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +2
-0@@ -28,9 +28,28 @@#define AMDGPU_SRIOV_CAPS_ENABLE_IOV (1 << 1)/* sr-iov is enabled on this GPU */ #define AMDGPU_SRIOV_CAPS_IS_VF (1 << 2)/* this GPU is a virtual function */ #define AMDGPU_PASSTHROUGH_MODE (1 << 3)/* thw whole GPU is pass through for VM */ + +struct amdgpu_vm ;/* GPU virtualization */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 43
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +8
-0struct amdgpu_virt {uint32_t caps; +uint32_t csa_size; +struct amdgpu_bo *csa_obj; +uint64_t csa_vmid0_addr; +uint64_t gds_vmid0_addr; +int (*allocate_csa)(struct amdgpu_device *adev); +void (*deallocate_csa)(struct amdgpu_device *adev); +int (*map_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); +void (*unmap_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm);
Why callbacks for this?
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;
Please put that directly into the amdgpu_vm structure.
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;
That should be constant, shouldn't it?
Regards,
Christian.
Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0};#define amdgpu_sriov_enabled (adev) \drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -29,6 +29,7 @@#include "gpu_scheduler.h" #include "amdgpu_sync.h" #include "amdgpu_ring.h" +#include "amdgpu_virt.h" struct amdgpu_bo_va ;struct amdgpu_job ;
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h struct amdgpu_vm { · line 118
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -111,6 +112,7 @@ struct amdgpu_vm {/* client id */ u64 client_id; +struct amdgpu_vm_virt vm_virt; };struct amdgpu_vm_id {
_______________________________________________
amd-gfx mailing list [email protected]<mailto:[email protected]>
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
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): BY2PR1201MB1110F2B6E7FB093F369ACA2884670-O28G1zQ8oGliQkyLPkmea2rFom/aUZj6nBOFsp37pqbUKgpGm//BTAC/[email protected]
Re: 答复: 答复: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Quoted 2 lines
NAH, for AI/VI the
gds_addris the bottom half of CSA area (8kb), but we don't know for later chips ...
In this case we shouldn't store the offset anywhere, but rather just calculate it in the chip specific code from the CSA address when it is needed.
Quoted 2 lines
I prefer keep a member field in the struct for flexibility ...
We can still change it if we find that we need more flexibility here. Usually we try to avoid flexibility for the cost of complexity if we don't have a good reason for it.
It just complicates the code and over the long term creates maintenance problems because you end of with a lot of only halve way implemented features.
Regards,
Christian.
Am 10.01.2017 um 10:57 schrieb Liu, Monk:
Quoted 32 lines
NAH, for AI/VI the
gds_addris the bottom half of CSA area (8kb), but we don't know for later chips ...I prefer keep a member field in the struct for flexibility ...
regards with put those member into
amdgpu_vmstructure, if you really prefer that way I'm fine by moving them, not too important anyway.BR Monk
------------------------------------------------------------------------
*发件人:* amd-gfx <amd-gfx-bounces-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org> 代表 Christian König <[email protected]> *发送时间:* 2017年1月10日 17:51:46
*收件人:* Liu, Monk; [email protected]
*主题:* Re: 答复: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 10.01.2017 um 04:03 schrieb Liu, Monk:Quoted 5 lines
Review
Quoted 4 lines
(quoted diff) +3
-0+/* virtual MC address of CSA & GDS for each VM */ + uint64_t vm_csa_addr; + uint64_t vm_gds_addr;Quoted 8 lines
That should be constant, shouldn't it?
[ML] do you mean "const
uint64_tvm_csa_addr" ? of cause not if that's your point ... compiling error on thatand strictly speaking that's not constant, the address is calculated ...
Quoted 101 lines
No that wasn't what I meant. The address is constant for all VM's in the system, isn't it?
Just add a define similar to
AMDGPU_VA_RESERVED_SIZEand stop calculating it all together (BTW:AMDGPU_VA_RESERVED_SIZEshould probably be moved toamdgpu_vm.h).Regards,
Christian.Quoted 92 lines
------------------------------------------------------------------------
*发件人:* Christian König <[email protected]> *发送时间:* 2017年1月9日 18:57:06
*收件人:* Liu, Monk; [email protected] *主题:* Re: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 09.01.2017 um 09:02 schrieb Monk Liu:Review
Quoted 8 lines
Change-Id Ife0eff7b13b8b5946f005a39f6ecb8db1cb72c38Signed-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 19 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 ++ 2 files changed, 21 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.hdrivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 28
Quoted 5 lines
index 0d821d9..5aa7f0c 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0@@ -28,9 +28,28 @@#define AMDGPU_SRIOV_CAPS_ENABLE_IOV (1 << 1)/* sr-iov is enabled on this GPU */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 29
Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0#define AMDGPU_SRIOV_CAPS_IS_VF (1 << 2)/* this GPU is a virtual function */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 30
Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0#define AMDGPU_PASSTHROUGH_MODE (1 << 3)/* thw whole GPU is pass through for VM */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 43
Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +10
-0+ +struct amdgpu_vm ;/* GPU virtualization */ struct amdgpu_virt {uint32_t caps; +uint32_t csa_size; +struct amdgpu_bo *csa_obj; +uint64_t csa_vmid0_addr; +uint64_t gds_vmid0_addr; +int (*allocate_csa)(struct amdgpu_device *adev); +void (*deallocate_csa)(struct amdgpu_device *adev); +int (*map_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); + void (*unmap_csa)(struct amdgpu_device *adev, struct
amdgpu_vm*vm);Why callbacks for this?
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;Please put that directly into the
amdgpu_vmstructure.drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;That should be constant, shouldn't it?
Regards,
Christian.Review
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0};#define amdgpu_sriov_enabled (adev) \(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hQuoted 12 lines
index 42a629b..27cbcbc 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -29,6 +29,7 @@#include "gpu_scheduler.h" #include "amdgpu_sync.h" #include "amdgpu_ring.h" +#include "amdgpu_virt.h" struct amdgpu_bo_va ;struct amdgpu_job ;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h struct amdgpu_vm { · line 118
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -111,6 +112,7 @@ struct amdgpu_vm {/* client id */ u64 client_id; +struct amdgpu_vm_virt vm_virt; };struct amdgpu_vm_id {_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
答复: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
In reply to a message not in the archivedrivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;
Please put that directly into the amdgpu_vm structure.
[ML] they are not a tiny used by bare-metal case so I don't think they should be put into common structure at all.
see we have "struct amdgpu_virt" in "struct amdgpu_device", the same style
BR Monk
________________________________
发件人: Christian König <[email protected]> 发送时间: 2017年1月9日 18:57:06
收件人: Liu, Monk; [email protected]
主题: Re: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 09.01.2017 um 09:02 schrieb Monk Liu:
Quoted 18 lines
Change-Id Ife0eff7b13b8b5946f005a39f6ecb8db1cb72c38Signed-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 19 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 ++ 2 files changed, 21 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +2
-0@@ -28,9 +28,28 @@#define AMDGPU_SRIOV_CAPS_ENABLE_IOV (1 << 1)/* sr-iov is enabled on this GPU */ #define AMDGPU_SRIOV_CAPS_IS_VF (1 << 2)/* this GPU is a virtual function */ #define AMDGPU_PASSTHROUGH_MODE (1 << 3)/* thw whole GPU is pass through for VM */ + +struct amdgpu_vm ;/* GPU virtualization */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 43
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +8
-0struct amdgpu_virt {uint32_t caps; +uint32_t csa_size; +struct amdgpu_bo *csa_obj; +uint64_t csa_vmid0_addr; +uint64_t gds_vmid0_addr; +int (*allocate_csa)(struct amdgpu_device *adev); +void (*deallocate_csa)(struct amdgpu_device *adev); +int (*map_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); +void (*unmap_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm);
Why callbacks for this?
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;
Please put that directly into the amdgpu_vm structure.
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;
That should be constant, shouldn't it?
Regards,
Christian.
Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0};#define amdgpu_sriov_enabled (adev) \drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -29,6 +29,7 @@#include "gpu_scheduler.h" #include "amdgpu_sync.h" #include "amdgpu_ring.h" +#include "amdgpu_virt.h" struct amdgpu_bo_va ;struct amdgpu_job ;
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h struct amdgpu_vm { · line 118
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -111,6 +112,7 @@ struct amdgpu_vm {/* client id */ u64 client_id; +struct amdgpu_vm_virt vm_virt; };struct amdgpu_vm_id {
_______________________________________________
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): BY2PR1201MB1110963B2DE72B2892D4AC1F84670-O28G1zQ8oGliQkyLPkmea2rFom/aUZj6nBOFsp37pqbUKgpGm//BTAC/[email protected]
Re: 答复: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
We don't need the validate buffer and even when you need it putting it into the VM structure is illegal, because it can be used by more than one thread at the same time.
This leaves only the csa_bo_va in the amdgpu_vm_virt structure and having a structure with only one field is rather pointless.
Additional to that I really don't see CSA handling as something related to virtualization.
Regards,
Christian.
Am 10.01.2017 um 04:05 schrieb Liu, Monk:
Quoted 98 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;Please put that directly into the
amdgpu_vmstructure.[ML] they are not a tiny used by bare-metal case so I don't think they should be put into common structure at all.
see we have "struct
amdgpu_virt" in "structamdgpu_device", the same styleBR Monk
------------------------------------------------------------------------
*发件人:* Christian König <[email protected]> *发送时间:* 2017年1月9日 18:57:06
*收件人:* Liu, Monk; [email protected] *主题:* Re: [PATCH 2/7] drm/amdgpu:new field members for SRIOV
Am 09.01.2017 um 09:02 schrieb Monk Liu:Review
Quoted 8 lines
Change-Id Ife0eff7b13b8b5946f005a39f6ecb8db1cb72c38Signed-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 19 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 ++ 2 files changed, 21 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.hdrivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 28
Quoted 5 lines
index 0d821d9..5aa7f0c 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0@@ -28,9 +28,28 @@#define AMDGPU_SRIOV_CAPS_ENABLE_IOV (1 << 1)/* sr-iov is enabled on this GPU */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 29
Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0#define AMDGPU_SRIOV_CAPS_IS_VF (1 << 2)/* this GPU is a virtual function */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 30
Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0#define AMDGPU_PASSTHROUGH_MODE (1 << 3)/* thw whole GPU is pass through for VM */
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h
Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +10
-0+ +struct amdgpu_vm ;/* GPU virtualization */ struct amdgpu_virt {uint32_t caps; +uint32_t csa_size; +struct amdgpu_bo *csa_obj; +uint64_t csa_vmid0_addr; +uint64_t gds_vmid0_addr; +int (*allocate_csa)(struct amdgpu_device *adev); +void (*deallocate_csa)(struct amdgpu_device *adev); +int (*map_csa)(struct amdgpu_device *adev,struct amdgpu_vm *vm); + void (*unmap_csa)(struct amdgpu_device *adev, struct amdgpu_vm*vm);
Why callbacks for this?
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 49
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +6
-0+}; + +struct amdgpu_vm_virt { +/* each VM will map on CSA */ +struct ttm_validate_buffer csa_tv; +struct amdgpu_bo_va *csa_bo_va;Please put that directly into the
amdgpu_vmstructure.drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h line 52
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +3
-0+/* virtual MC address of CSA & GDS for each VM */ +uint64_t vm_csa_addr; +uint64_t vm_gds_addr;That should be constant, shouldn't it?
Regards,
Christian.Review
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +0
-0};#define amdgpu_sriov_enabled (adev) \(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hQuoted 13 lines
index 42a629b..27cbcbc 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -29,6 +29,7 @@#include "gpu_scheduler.h" #include "amdgpu_sync.h" #include "amdgpu_ring.h" +#include "amdgpu_virt.h" struct amdgpu_bo_va ;struct amdgpu_job ;@@ -111,6 +112,7 @@ struct amdgpu_vm {
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h struct amdgpu_vm { · line 118
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0/* client id */ u64 client_id; +struct amdgpu_vm_virt vm_virt; };struct amdgpu_vm_id {
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 3/7 drm/amdgpu:impl static CSA alloc/dealloc
In reply to a message not in the archivefor SRIOV usage, CSA is only used per device and each VM will map on it.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 23 ++++++++++++++++++++++- 1 file changed, 22 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +22 -1
@@ -20,4 +20,25 @@* OTHER DEALINGS IN THE SOFTWARE. * */ -#include "amdgpu.h" +#include "amdgpu.h" +static int amdgpu_allocate_static_csa (struct amdgpu_device *adev) +{ +int r; +void *ptr; + + r =amdgpu_bo_create_kernel (adev, adev->virt.csa_size,PAGE_SIZE , +AMDGPU_GEM_DOMAIN_VRAM , &adev->virt.csa_obj, + &adev->virt.csa_vmid0_addr, &ptr); +if (r) +return r; + +/* gds part take the bottom half buffer */ + adev->virt.gds_vmid0_addr = adev->virt.csa_vmid0_addr +PAGE_SIZE ; +memset (ptr,0 , adev->virt.csa_size); +return 0 ; +} + +static void amdgpu_deallocate_static_csa (struct amdgpu_device *adev) +{ +amdgpu_bo_free_kernel (&adev->virt.csa_obj, &adev->virt.csa_vmid0_addr,NULL ); +}
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): 1483948982-27056-4-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 3/7] drm/amdgpu:impl static CSA alloc/dealloc
Am 09.01.2017 um 09:02 schrieb Monk Liu:
Quoted 14 lines
for SRIOV usage, CSA is only used per device and each VM will map on it.
Change-Id I0736b6e51329d643343e26968a47a81c3a27035fSigned-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 23 ++++++++++++++++++++++- 1 file changed, 22 insertions(+), 1 deletion(-)drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c line 23
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +1
-1@@ -20,4 +20,25 @@* OTHER DEALINGS IN THE SOFTWARE. * */ -#include "amdgpu.h" +#include "amdgpu.h"
Please add at least a newline between the includes and the start of the code.
Quoted 11 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +11
-0+static int amdgpu_allocate_static_csa (struct amdgpu_device *adev) +{ +int r; +void *ptr; + + r =amdgpu_bo_create_kernel (adev, adev->virt.csa_size,PAGE_SIZE , +AMDGPU_GEM_DOMAIN_VRAM , &adev->virt.csa_obj, + &adev->virt.csa_vmid0_addr, &ptr); +if (r) +return r; +
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c line 44
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +10
-0+/* gds part take the bottom half buffer */ + adev->virt.gds_vmid0_addr = adev->virt.csa_vmid0_addr +PAGE_SIZE ; +memset (ptr,0 , adev->virt.csa_size); +return 0 ; +} + +static void amdgpu_deallocate_static_csa (struct amdgpu_device *adev) +{ +amdgpu_bo_free_kernel (&adev->virt.csa_obj, &adev->virt.csa_vmid0_addr,NULL ); +}
Just static function which aren't called will raise compiler warning, please squash that into the functionality where it is used.
Christian.
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 4/7 drm/amdgpu:impl static CSA map/unmap routines
In reply to a message not in the archive--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 90 ++++++++++++++++++++++++++++++++ 1 file changed, 90 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +90 -0
@@ -42,3 +42,93 @@ static void amdgpu_deallocate_static_csa(struct amdgpu_device *adev) {amdgpu_bo_free_kernel (&adev->virt.csa_obj, &adev->virt.csa_vmid0_addr,NULL ); } + +/* +* amdgpu_map_static_csa should be called during amdgpu_vm_init +* it maps virtual address "AMDGPU_VA_RESERVED_SIZE - adev->virt.csa_size" +* to this VM, and each command submission of GFX should use this virtual +* address within META_DATA init package to support SRIOV gfx preemption. +*/ + +static int amdgpu_map_static_csa (struct amdgpu_device *adev,struct amdgpu_vm *vm) +{ +int r; +uint64_t vaddr; +struct amdgpu_bo_va *bo_va; +struct ww_acquire_ctx ticket; +struct list_head list, duplicates; +struct amdgpu_bo_list_entry pd; + +INIT_LIST_HEAD (&list); +INIT_LIST_HEAD (&duplicates); +INIT_LIST_HEAD (&vm->vm_virt.csa_tv.head); + vm->vm_virt.csa_tv.bo = &adev->virt.csa_obj->tbo; + vm->vm_virt.csa_tv.shared = true; + +list_add (&vm->vm_virt.csa_tv.head, &list); +amdgpu_vm_get_pd_bo (vm, &list, &pd); + + vaddr =AMDGPU_VA_RESERVED_SIZE - adev->virt.csa_size; + r =ttm_eu_reserve_buffers (&ticket, &list, true, &duplicates); +if (r) { +DRM_ERROR ("failed to reserve CSA,PD BOs: err=%d\n" , r); +return r; + } + + bo_va =amdgpu_vm_bo_add (adev, vm, adev->virt.csa_obj); +if (!bo_va) { +ttm_eu_backoff_reservation (&ticket, &list); +DRM_ERROR ("failed to create bo_va for static CSA\n" ); +return -ENOMEM ; + } + + r =amdgpu_vm_bo_map (adev, bo_va, vaddr,0 , adev->virt.csa_size, +AMDGPU_PTE_READABLE |AMDGPU_PTE_WRITEABLE | +AMDGPU_PTE_EXECUTABLE ); + +if (r) { +DRM_ERROR ("failed to do bo_map on static CSA, err=%d\n" , r); +amdgpu_vm_bo_rmv (adev, bo_va); +ttm_eu_backoff_reservation (&ticket, &list); +kfree (bo_va); +return r; + } + +ttm_eu_backoff_reservation (&ticket, &list); + + vm->vm_virt.csa_bo_va = bo_va; + vm->vm_virt.vm_csa_addr = vaddr; + vm->vm_virt.vm_gds_addr = vaddr +PAGE_SIZE ; +return 0 ; +} + +static void amdgpu_unmap_static_csa (struct amdgpu_device *adev,struct amdgpu_vm *vm) +{ +int r; +struct ww_acquire_ctx ticket; +struct list_head list, duplicate; +struct amdgpu_bo_list_entry pd; + +if (vm->vm_virt.csa_bo_va) { +INIT_LIST_HEAD (&list); +INIT_LIST_HEAD (&duplicate); +list_add (&vm->vm_virt.csa_tv.head, &list); +amdgpu_vm_get_pd_bo (vm, &list, &pd); + + r =ttm_eu_reserve_buffers (&ticket, &list, true, &duplicate); +if (r) { +DRM_ERROR ("failed to reserve static CSA buffer: err=%d\n" , r); +return ; + } + +amdgpu_vm_bo_rmv (adev, vm->vm_virt.csa_bo_va); +/* maybe no need to do real clearing cuz this vm will die soon */ + r =amdgpu_vm_clear_freed (adev, vm); +if (r) +DRM_ERROR ("failed to clear static CSA bo: err=%d" , r); + +ttm_eu_backoff_reservation (&ticket, &list); + vm->vm_virt.csa_bo_va =NULL ; + vm->vm_virt.vm_csa_addr = vm->vm_virt.vm_gds_addr =0 ; + } +}
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): 1483948982-27056-5-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 4/7] drm/amdgpu:impl static CSA map/unmap routines
Am 09.01.2017 um 09:02 schrieb Monk Liu:
Quoted 2 lines
Change-Id I76ddeb3212c96d87a2d15a608ae8c0771e2d94edSigned-off-by Monk Liu <[email protected]>
As discussed before that approach won't work.
Just add the call I suggest to amdgpu_bo_vm_update_pte() in amdgpu_cs.c.
Christian.
Quoted 92 lines
--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 90 ++++++++++++++++++++++++++++++++ 1 file changed, 90 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +80
-0@@ -42,3 +42,93 @@ static void amdgpu_deallocate_static_csa(struct amdgpu_device *adev) {amdgpu_bo_free_kernel (&adev->virt.csa_obj, &adev->virt.csa_vmid0_addr,NULL ); } + +/* +* amdgpu_map_static_csa should be called during amdgpu_vm_init +* it maps virtual address "AMDGPU_VA_RESERVED_SIZE - adev->virt.csa_size" +* to this VM, and each command submission of GFX should use this virtual +* address within META_DATA init package to support SRIOV gfx preemption. +*/ + +static int amdgpu_map_static_csa (struct amdgpu_device *adev,struct amdgpu_vm *vm) +{ +int r; +uint64_t vaddr; +struct amdgpu_bo_va *bo_va; +struct ww_acquire_ctx ticket; +struct list_head list, duplicates; +struct amdgpu_bo_list_entry pd; + +INIT_LIST_HEAD (&list); +INIT_LIST_HEAD (&duplicates); +INIT_LIST_HEAD (&vm->vm_virt.csa_tv.head); + vm->vm_virt.csa_tv.bo = &adev->virt.csa_obj->tbo; + vm->vm_virt.csa_tv.shared = true; + +list_add (&vm->vm_virt.csa_tv.head, &list); +amdgpu_vm_get_pd_bo (vm, &list, &pd); + + vaddr =AMDGPU_VA_RESERVED_SIZE - adev->virt.csa_size; + r =ttm_eu_reserve_buffers (&ticket, &list, true, &duplicates); +if (r) { +DRM_ERROR ("failed to reserve CSA,PD BOs: err=%d\n" , r); +return r; + } + + bo_va =amdgpu_vm_bo_add (adev, vm, adev->virt.csa_obj); +if (!bo_va) { +ttm_eu_backoff_reservation (&ticket, &list); +DRM_ERROR ("failed to create bo_va for static CSA\n" ); +return -ENOMEM ; + } + + r =amdgpu_vm_bo_map (adev, bo_va, vaddr,0 , adev->virt.csa_size, +AMDGPU_PTE_READABLE |AMDGPU_PTE_WRITEABLE | +AMDGPU_PTE_EXECUTABLE ); + +if (r) { +DRM_ERROR ("failed to do bo_map on static CSA, err=%d\n" , r); +amdgpu_vm_bo_rmv (adev, bo_va); +ttm_eu_backoff_reservation (&ticket, &list); +kfree (bo_va); +return r; + } + +ttm_eu_backoff_reservation (&ticket, &list); + + vm->vm_virt.csa_bo_va = bo_va; + vm->vm_virt.vm_csa_addr = vaddr; + vm->vm_virt.vm_gds_addr = vaddr +PAGE_SIZE ; +return 0 ; +} + +static void amdgpu_unmap_static_csa (struct amdgpu_device *adev,struct amdgpu_vm *vm) +{ +int r; +struct ww_acquire_ctx ticket; +struct list_head list, duplicate; +struct amdgpu_bo_list_entry pd; + +if (vm->vm_virt.csa_bo_va) { +INIT_LIST_HEAD (&list); +INIT_LIST_HEAD (&duplicate); +list_add (&vm->vm_virt.csa_tv.head, &list); +amdgpu_vm_get_pd_bo (vm, &list, &pd); + + r =ttm_eu_reserve_buffers (&ticket, &list, true, &duplicate); +if (r) { +DRM_ERROR ("failed to reserve static CSA buffer: err=%d\n" , r); +return ; + } + +amdgpu_vm_bo_rmv (adev, vm->vm_virt.csa_bo_va);
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c static void amdgpu_deallocate_static_csa(struct amdgpu_device *adev) · line 134
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +10
-0+/* maybe no need to do real clearing cuz this vm will die soon */ + r =amdgpu_vm_clear_freed (adev, vm); +if (r) +DRM_ERROR ("failed to clear static CSA bo: err=%d" , r); + +ttm_eu_backoff_reservation (&ticket, &list); + vm->vm_virt.csa_bo_va =NULL ; + vm->vm_virt.vm_csa_addr = vm->vm_virt.vm_gds_addr =0 ; + } +}
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 5/7 drm/amdgpu:set static csa function
In reply to a message not in the archive--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 8 ++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 2 ++ drivers/gpu/drm/amd/amdgpu/vi.c | 3 +++ 3 files changed, 13 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +8 -0
@@ -132,3 +132,11 @@ static void amdgpu_unmap_static_csa(struct amdgpu_device *adev, struct amdgpu_vm vm->vm_virt.vm_csa_addr = vm->vm_virt.vm_gds_addr =0 ; } } + +void amdgpu_virt_set_csa_funcs (struct amdgpu_device *adev) { + adev->virt.allocate_csa = amdgpu_allocate_static_csa; + adev->virt.deallocate_csa = amdgpu_deallocate_static_csa; + adev->virt.map_csa = amdgpu_map_static_csa; + adev->virt.unmap_csa = amdgpu_unmap_static_csa; + adev->virt.csa_size =8 *1024 ;/* two page now for VI and AI */ +}
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +2 -0
@@ -73,4 +73,6 @@ static inline bool is_virtual_machine(void) #endif } +void amdgpu_virt_set_csa_funcs (struct amdgpu_device *adev); +#endif
drivers/gpu/drm/amd/amdgpu/vi.c +3 -0
@@ -455,6 +455,9 @@ static void vi_detect_hw_virtualization(struct amdgpu_device *adev)if (is_virtual_machine ())/* passthrough mode exclus sr-iov mode */ adev->virt.caps |=AMDGPU_PASSTHROUGH_MODE ; } + +if (amdgpu_sriov_vf (adev)) +amdgpu_virt_set_csa_funcs (adev); }static const struct amdgpu_allowed_register_entry tonga_allowed_read_registers[] = {
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): 1483948982-27056-6-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 5/7] drm/amdgpu:set static csa function
Am 09.01.2017 um 09:03 schrieb Monk Liu:
Quoted 14 lines
Change-Id Ib59c9e7f37a10ef1d6335d400ab41fe3cc9e2fb6Signed-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 8 ++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h | 2 ++ drivers/gpu/drm/amd/amdgpu/vi.c | 3 +++ 3 files changed, 13 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +0
-0@@ -132,3 +132,11 @@ static void amdgpu_unmap_static_csa(struct amdgpu_device *adev, struct amdgpu_vm vm->vm_virt.vm_csa_addr = vm->vm_virt.vm_gds_addr =0 ;
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c static void amdgpu_unmap_static_csa(struct amdgpu_device *adev, struct amdgpu_vm · line 142
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +8
-0} } + +void amdgpu_virt_set_csa_funcs (struct amdgpu_device *adev) { + adev->virt.allocate_csa = amdgpu_allocate_static_csa; + adev->virt.deallocate_csa = amdgpu_deallocate_static_csa; + adev->virt.map_csa = amdgpu_map_static_csa; + adev->virt.unmap_csa = amdgpu_unmap_static_csa; + adev->virt.csa_size =8 *1024 ;/* two page now for VI and AI */ +}
You don't abstract anything by using indirect calls, so please avoid them and call the appropriate functions directly.
Christian.
Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h +2
-0@@ -73,4 +73,6 @@ static inline bool is_virtual_machine(void) #endif } +void amdgpu_virt_set_csa_funcs (struct amdgpu_device *adev); +#endif drivers/gpu/drm/amd/amdgpu/vi.c +0
-0
drivers/gpu/drm/amd/amdgpu/vi.c static void vi_detect_hw_virtualization(struct amdgpu_device *adev) · line 463
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/vi.c +3
-0@@ -455,6 +455,9 @@ static void vi_detect_hw_virtualization(struct amdgpu_device *adev)if (is_virtual_machine ())/* passthrough mode exclus sr-iov mode */ adev->virt.caps |=AMDGPU_PASSTHROUGH_MODE ; } + +if (amdgpu_sriov_vf (adev)) +amdgpu_virt_set_csa_funcs (adev); }static const struct amdgpu_allowed_register_entry tonga_allowed_read_registers[] = {
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 6/7 drm/amdgpu:alloc/dealloc csa accordingly
In reply to a message not in the archive--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ 1 file changed, 7 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +7 -0
@@ -1384,6 +1384,10 @@ static int amdgpu_init(struct amdgpu_device *adev)return r; } adev->ip_blocks[i].status.hw = true; + +/* right after GMC hw init, we create CSA */ +if (amdgpu_sriov_vf (adev)) + adev->virt.allocate_csa (adev); } }
@@ -1517,6 +1521,9 @@ static int amdgpu_fini(struct amdgpu_device *adev) adev->ip_blocks[i].status.late_initialized = false; } +if (amdgpu_sriov_vf (adev)) + adev->virt.deallocate_csa (adev); +return 0 ; }
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): 1483948982-27056-7-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 6/7] drm/amdgpu:alloc/dealloc csa accordingly
Hi Monk, a comment below.
Den 9 jan. 2017 9:07 fm skrev "Monk Liu" <[email protected]>:
--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ 1 file changed, 7 insertions(+)
(quoted diff) +0 -0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index f82919d..5253aa2 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +4 -0
@@ -1384,6 +1384,10 @@ static int amdgpu_init(struct amdgpu_device *adev)return r; } adev->ip_blocks[i].status.hw = true; + +/* right after GMC hw init, we create CSA */ +if (amdgpu_sriov_vf (adev)) + adev->virt.allocate_csa (adev);
The amdgpu_allocate_static_csa function returns a value, presumably an error on failure. Shouldn't the return value be checked here and errors handled?
Best regards,
Nils
}
}@@ -1517,6 +1521,9 @@ static int amdgpu_fini(struct amdgpu_device *adev)
adev->ip_blocks[i].status.late_initialized = false;
}+ if (amdgpu_sriov_vf(adev))
+ adev->virt.deallocate_csa(adev);
+
return 0;
}
--
2.7.4
_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
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): CA+nq7DuCviSbC7ty46j7YqYM83c5=oZF19eAfZLT1_ehYNqu+A-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org
答复: [PATCH 6/7] drm/amdgpu:alloc/dealloc csa accordingly
Yeah, make sense
thanks
BR Monk
________________________________
发件人: Nils Wallménius <[email protected]> 发送时间: 2017年1月10日 15:56:36
收件人: Liu, Monk
抄送: [email protected]
主题: Re: [PATCH 6/7] drm/amdgpu:alloc/dealloc csa accordingly
Hi Monk, a comment below.
Den 9 jan. 2017 9:07 fm skrev "Monk Liu" <[email protected]<mailto:[email protected]>>:
--- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ 1 file changed, 7 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +4 -0
@@ -1384,6 +1384,10 @@ static int amdgpu_init(struct amdgpu_device *adev)return r; } adev->ip_blocks[i].status.hw = true; + +/* right after GMC hw init, we create CSA */ +if (amdgpu_sriov_vf (adev)) + adev->virt.allocate_csa (adev);
The amdgpu_allocate_static_csa function returns a value, presumably an error on failure. Shouldn't the return value be checked here and errors handled?
Best regards,
Nils
}
}@@ -1517,6 +1521,9 @@ static int amdgpu_fini(struct amdgpu_device *adev)
adev->ip_blocks[i].status.late_initialized = false;
}+ if (amdgpu_sriov_vf(adev))
+ adev->virt.deallocate_csa(adev);
+
return 0;
}
--
2.7.4
_______________________________________________
amd-gfx mailing list [email protected]<mailto:[email protected]>
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 7/7 drm/amdgpu:map/unmap static csa accordingly
In reply to a message not in the archiveand update CSA bo_va in each submit
--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 14 ++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 11 +++++++++++ 2 files changed, 25 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +14 -0
@@ -771,6 +771,20 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,if (r)return r; +if (amdgpu_sriov_vf (adev)) { +struct fence *f; + bo_va = vm->vm_virt.csa_bo_va; +BUG_ON (!bo_va); + r =amdgpu_vm_bo_update (adev, bo_va, false); +if (r) +return r; + + f = bo_va->last_pt_update; + r =amdgpu_sync_fence (adev, &p->job->sync, f); +if (r) +return r; + } +if (p->bo_list) {for (i =0 ; i < p->bo_list->num_entries; i++) {struct fence *f;
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +11 -0
@@ -1576,6 +1576,14 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) vm->last_eviction_counter =atomic64_read (&adev->num_evictions);amdgpu_bo_unreserve (vm->page_directory); +if (amdgpu_sriov_vf (adev)) { +BUG_ON (!adev->virt.map_csa); +BUG_ON (!adev->virt.unmap_csa); + r = adev->virt.map_csa (adev, vm); +if (r) + goto error_free_page_directory; + } +return 0 ; error_free_page_directory:
@@ -1606,6 +1614,9 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)struct amdgpu_bo_va_mapping *mapping, *tmp;int i; +if (amdgpu_sriov_vf (adev)) + adev->virt.unmap_csa (adev, vm); +amd_sched_entity_fini (vm->entity.sched, &vm->entity);if (!RB_EMPTY_ROOT (&vm->va)) {
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): 1483948982-27056-8-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly
Am 09.01.2017 um 09:03 schrieb Monk Liu:
Quoted 14 lines
and update CSA
bo_vain each submitChange-Id I5ed73e1b7f89743d90298bc814a42a91e166be3bSigned-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 14 ++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 11 +++++++++++ 2 files changed, 25 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, · line 775
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0@@ -771,6 +771,20 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,if (r)return r; +if (amdgpu_sriov_vf (adev)) { +struct fence *f;
A new line is needed between deceleration and code.
Quoted 21 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +12
-0+ bo_va = vm->vm_virt.csa_bo_va; +BUG_ON (!bo_va); + r =amdgpu_vm_bo_update (adev, bo_va, false); +if (r) +return r; + + f = bo_va->last_pt_update; + r =amdgpu_sync_fence (adev, &p->job->sync, f); +if (r) +return r; + } +if (p->bo_list) {for (i =0 ; i < p->bo_list->num_entries; i++) {struct fence *f;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1576,6 +1576,14 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) vm->last_eviction_counter =atomic64_read (&adev->num_evictions);
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 1586
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +8
-0amdgpu_bo_unreserve (vm->page_directory); +if (amdgpu_sriov_vf (adev)) { +BUG_ON (!adev->virt.map_csa); +BUG_ON (!adev->virt.unmap_csa); + r = adev->virt.map_csa (adev, vm); +if (r) + goto error_free_page_directory; + } +
Just completely drop that. Updating the VM on the first command submission should be sufficient.
Christian.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 1622
Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3
-0return 0 ; error_free_page_directory:@@ -1606,6 +1614,9 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)struct amdgpu_bo_va_mapping *mapping, *tmp;int i; +if (amdgpu_sriov_vf (adev)) + adev->virt.unmap_csa (adev, vm); +amd_sched_entity_fini (vm->entity.sched, &vm->entity);if (!RB_EMPTY_ROOT (&vm->va)) {
_______________________________________________
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): 4f78e5dd-f7e8-0976-5abf-305e0b793ba1-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
答复: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly
why you want to drop that "adev->virt.map_csa()" calling ? without that calling you don't have CSA's initialized bo_va and mappings as well, and without bo_va and mappings how you can make "amdgpu_vm_bo_update()" invoke work ?
Monk
________________________________
发件人: Christian König <[email protected]>
发送时间: 2017年1月9日 19:04:44
收件人: Liu, Monk; [email protected]
主题: Re: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly
Am 09.01.2017 um 09:03 schrieb Monk Liu:
Quoted 14 lines
and update CSA
bo_vain each submitChange-Id I5ed73e1b7f89743d90298bc814a42a91e166be3bSigned-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 14 ++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 11 +++++++++++ 2 files changed, 25 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, · line 775
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0@@ -771,6 +771,20 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,if (r)return r; +if (amdgpu_sriov_vf (adev)) { +struct fence *f;
A new line is needed between deceleration and code.
Quoted 21 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +12
-0+ bo_va = vm->vm_virt.csa_bo_va; +BUG_ON (!bo_va); + r =amdgpu_vm_bo_update (adev, bo_va, false); +if (r) +return r; + + f = bo_va->last_pt_update; + r =amdgpu_sync_fence (adev, &p->job->sync, f); +if (r) +return r; + } +if (p->bo_list) {for (i =0 ; i < p->bo_list->num_entries; i++) {struct fence *f;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1576,6 +1576,14 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) vm->last_eviction_counter =atomic64_read (&adev->num_evictions);
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 1586
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +8
-0amdgpu_bo_unreserve (vm->page_directory); +if (amdgpu_sriov_vf (adev)) { +BUG_ON (!adev->virt.map_csa); +BUG_ON (!adev->virt.unmap_csa); + r = adev->virt.map_csa (adev, vm); +if (r) + goto error_free_page_directory; + } +
Just completely drop that. Updating the VM on the first command submission should be sufficient.
Christian.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 1622
Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3
-0return 0 ; error_free_page_directory:@@ -1606,6 +1614,9 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)struct amdgpu_bo_va_mapping *mapping, *tmp;int i; +if (amdgpu_sriov_vf (adev)) + adev->virt.unmap_csa (adev, vm); +amd_sched_entity_fini (vm->entity.sched, &vm->entity);if (!RB_EMPTY_ROOT (&vm->va)) {
_______________________________________________
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): BY2PR1201MB11108A5E8A3EF7F10C0392AE84670-O28G1zQ8oGliQkyLPkmea2rFom/aUZj6nBOFsp37pqbUKgpGm//BTAC/[email protected]
Re: 答复: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly
Adding the BO and it's mapping is VM specific code, we should initialize that directly in amdgpu_vm_init() and not call any helper to delegate the work.
Especially don't call from the VM code into the virt code and back into the VM code.
If you really want to keep that in amdgpu_virt. an alternative would be to put the call into amdgpu_driver_open_kms() directly after calling amdgpu_vm_init().
Regards,
Christian.
Am 10.01.2017 um 03:40 schrieb Liu, Monk:
Quoted 73 lines
why you want to drop that "
adev->virt.map_csa()" calling ? without that calling you don't have CSA's initializedbo_vaand mappings as well, and withoutbo_vaand mappings how you can make "amdgpu_vm_bo_update()" invoke work ?Monk
------------------------------------------------------------------------
*发件人:* Christian König <[email protected]> *发送时间:* 2017年1月9日 19:04:44
*收件人:* Liu, Monk; [email protected]
*主题:* Re: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly Am 09.01.2017 um 09:03 schrieb Monk Liu:Review
Quoted 10 lines
and update CSA
bo_vain each submitChange-Id I5ed73e1b7f89743d90298bc814a42a91e166be3bSigned-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 14 ++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 11 +++++++++++ 2 files changed, 25 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.cdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct
Quoted 4 lines
index 6159afc..083ab73 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0@@ -771,6 +771,20 @@ static int amdgpu_bo_vm_update_pte(struct
amdgpu_cs_parser*p,drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, · line 775
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0if (r)return r; +if (amdgpu_sriov_vf (adev)) { +struct fence *f;A new line is needed between deceleration and code.
Review
Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +12
-0+ bo_va = vm->vm_virt.csa_bo_va; +BUG_ON (!bo_va); + r =amdgpu_vm_bo_update (adev, bo_va, false); +if (r) +return r; + + f = bo_va->last_pt_update; + r =amdgpu_sync_fence (adev, &p->job->sync, f); +if (r) +return r; + } +if (p->bo_list) {for (i =0 ; i < p->bo_list->num_entries; i++) {struct fence *f;(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.cdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device
Quoted 4 lines
index d05546e..b9cd686 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1576,6 +1576,14 @@ int amdgpu_vm_init(struct amdgpu_device*adev, struct
amdgpu_vm*vm)drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 1586
Quoted 11 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +8
-0vm->last_eviction_counter =atomic64_read (&adev->num_evictions);amdgpu_bo_unreserve (vm->page_directory); +if (amdgpu_sriov_vf (adev)) { +BUG_ON (!adev->virt.map_csa); +BUG_ON (!adev->virt.unmap_csa); + r = adev->virt.map_csa (adev, vm); +if (r) + goto error_free_page_directory; + } +Just completely drop that. Updating the VM on the first command submission should be sufficient.
Quoted 18 lines
Christian.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0return 0 ; error_free_page_directory:@@ -1606,6 +1614,9 @@ void amdgpu_vm_fini(struct amdgpu_device*adev, struct
amdgpu_vm*vm)Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3
-0struct amdgpu_bo_va_mapping *mapping, *tmp;int i; +if (amdgpu_sriov_vf (adev)) + adev->virt.unmap_csa (adev, vm); +amd_sched_entity_fini (vm->entity.sched, &vm->entity);if (!RB_EMPTY_ROOT (&vm->va)) {
_______________________________________________
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): cd95519f-f45b-6865-0fab-e75a970a48b0-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
答复: 答复: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly
then what about "amdgpu_vm_bo_map" ? we need call it first before call "amdgpu_vm_bo_update" in "amdgpu_bo_vm_update_pte" , correct ? otherwise we even not create PT bo for the CSA ...
Do you suggest that rip out of amdgpu_map_csa() routine totally, and manually call "amdgpu_vm_bo_map" as well as "amdgpu_vm_bo_update" in sequence in "amdgpu_vm_bo_update" ?
BR Monk
________________________________
发件人: Christian König <[email protected]>
发送时间: 2017年1月10日 17:44:21
收件人: Liu, Monk; [email protected]
主题: Re: 答复: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly
Adding the BO and it's mapping is VM specific code, we should initialize that directly in amdgpu_vm_init() and not call any helper to delegate the work.
Especially don't call from the VM code into the virt code and back into the VM code.
If you really want to keep that in amdgpu_virt. an alternative would be to put the call into amdgpu_driver_open_kms() directly after calling amdgpu_vm_init().
Regards,
Christian.
Am 10.01.2017 um 03:40 schrieb Liu, Monk:
Quoted 76 lines
why you want to drop that "
adev->virt.map_csa()" calling ? without that calling you don't have CSA's initializedbo_vaand mappings as well, and withoutbo_vaand mappings how you can make "amdgpu_vm_bo_update()" invoke work ?Monk
------------------------------------------------------------------------
*发件人:* Christian König <[email protected]> *发送时间:* 2017年1月9日 19:04:44
*收件人:* Liu, Monk; [email protected]
*主题:* Re: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly Am 09.01.2017 um 09:03 schrieb Monk Liu:Review
Quoted 10 lines
and update CSA
bo_vain each submitChange-Id I5ed73e1b7f89743d90298bc814a42a91e166be3bSigned-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 14 ++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 11 +++++++++++ 2 files changed, 25 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.cdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct
Quoted 4 lines
index 6159afc..083ab73 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0@@ -771,6 +771,20 @@ static int amdgpu_bo_vm_update_pte(struct
amdgpu_cs_parser*p,drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct · line 775
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0if (r)return r; +if (amdgpu_sriov_vf (adev)) { +struct fence *f;A new line is needed between deceleration and code.
Review
Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +12
-0+ bo_va = vm->vm_virt.csa_bo_va; +BUG_ON (!bo_va); + r =amdgpu_vm_bo_update (adev, bo_va, false); +if (r) +return r; + + f = bo_va->last_pt_update; + r =amdgpu_sync_fence (adev, &p->job->sync, f); +if (r) +return r; + } +if (p->bo_list) {for (i =0 ; i < p->bo_list->num_entries; i++) {struct fence *f;(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.cdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device
Quoted 4 lines
index d05546e..b9cd686 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1576,6 +1576,14 @@ int amdgpu_vm_init(struct amdgpu_device*adev, struct
amdgpu_vm*vm)drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device · line 1586
Quoted 11 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +8
-0vm->last_eviction_counter =atomic64_read (&adev->num_evictions);amdgpu_bo_unreserve (vm->page_directory); +if (amdgpu_sriov_vf (adev)) { +BUG_ON (!adev->virt.map_csa); +BUG_ON (!adev->virt.unmap_csa); + r = adev->virt.map_csa (adev, vm); +if (r) + goto error_free_page_directory; + } +Just completely drop that. Updating the VM on the first command submission should be sufficient.
Quoted 18 lines
Christian.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0return 0 ; error_free_page_directory:@@ -1606,6 +1614,9 @@ void amdgpu_vm_fini(struct amdgpu_device*adev, struct
amdgpu_vm*vm)Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3
-0struct amdgpu_bo_va_mapping *mapping, *tmp;int i; +if (amdgpu_sriov_vf (adev)) + adev->virt.unmap_csa (adev, vm); +amd_sched_entity_fini (vm->entity.sched, &vm->entity);if (!RB_EMPTY_ROOT (&vm->va)) {
_______________________________________________
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): BY2PR1201MB1110632E66A596FA820250E584670-O28G1zQ8oGliQkyLPkmea2rFom/aUZj6nBOFsp37pqbUKgpGm//BTAC/[email protected]
Re: 答复: 答复: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly
Quoted 2 lines
Do you suggest that rip out of
amdgpu_map_csa()routine totally,
Yes, mapping the CSA is not something specific to virtualization, so we shouldn't put that into amdgpu_virt.c.
Quoted 2 lines
and manually call "
amdgpu_vm_bo_map" as well as "amdgpu_vm_bo_update" in sequence in "amdgpu_vm_bo_update" ?
No, during command submission it is to late to allocate the PT. Better put this a layer up into amdgpu_driver_open_kms() after the VM is initialized.
Regards,
Christian.
Am 10.01.2017 um 10:53 schrieb Liu, Monk:
Quoted 85 lines
then what about "
amdgpu_vm_bo_map" ? we need call it first before call "amdgpu_vm_bo_update" in "amdgpu_bo_vm_update_pte" , correct ? otherwise we even not create PT bo for the CSA ...Do you suggest that rip out of
amdgpu_map_csa()routine totally, and manually call "amdgpu_vm_bo_map" as well as "amdgpu_vm_bo_update" in sequence in "amdgpu_vm_bo_update" ?BR Monk
------------------------------------------------------------------------
*发件人:* Christian König <[email protected]> *发送时间:* 2017年1月10日 17:44:21
*收件人:* Liu, Monk; [email protected] *主题:* Re: 答复: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly Adding the BO and it's mapping is VM specific code, we should initialize that directly inamdgpu_vm_init()and not call any helper to delegate the work.Especially don't call from the VM code into the virt code and back into the VM code.
If you really want to keep that in
amdgpu_virt. an alternative would be to put the call intoamdgpu_driver_open_kms()directly after callingamdgpu_vm_init().Regards,
Christian.Am 10.01.2017 um 03:40 schrieb Liu, Monk:
Quoted 56 lines
why you want to drop that "
adev->virt.map_csa()" calling ? without that calling you don't have CSA's initializedbo_vaand mappings as well, and withoutbo_vaand mappings how you can make "amdgpu_vm_bo_update()" invoke work ?Monk
------------------------------------------------------------------------
*发件人:* Christian König <[email protected]> *发送时间:* 2017年1月9日 19:04:44
*收件人:* Liu, Monk; [email protected] *主题:* Re: [PATCH 7/7] drm/amdgpu:map/unmap static csa accordingly Am 09.01.2017 um 09:03 schrieb Monk Liu:Review
Quoted 10 lines
and update CSA
bo_vain each submitChange-Id I5ed73e1b7f89743d90298bc814a42a91e166be3bSigned-off-by Monk Liu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 14 ++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 11 +++++++++++ 2 files changed, 25 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.cdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct
Quoted 4 lines
index 6159afc..083ab73 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0@@ -771,6 +771,20 @@ static int amdgpu_bo_vm_update_pte(struct
amdgpu_cs_parser*p,drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct · line 775
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0if (r)return r; +if (amdgpu_sriov_vf (adev)) { +struct fence *f;A new line is needed between deceleration and code.
Review
Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +12
-0+ bo_va = vm->vm_virt.csa_bo_va; +BUG_ON (!bo_va); + r =amdgpu_vm_bo_update (adev, bo_va, false); +if (r) +return r; + + f = bo_va->last_pt_update; + r =amdgpu_sync_fence (adev, &p->job->sync, f); +if (r) +return r; + } +if (p->bo_list) {for (i =0 ; i < p->bo_list->num_entries; i++) {struct fence *f;(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
Quoted 40 lines
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device
Quoted 4 lines
index d05546e..b9cd686 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1576,6 +1576,14 @@ int amdgpu_vm_init(struct amdgpu_device*adev, struct
amdgpu_vm*vm)Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0vm->last_eviction_counter =
atomic64_read(&adev->num_evictions);Quoted 32 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device · line 1586
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +8
-0amdgpu_bo_unreserve (vm->page_directory); +if (amdgpu_sriov_vf (adev)) { +BUG_ON (!adev->virt.map_csa); +BUG_ON (!adev->virt.unmap_csa); + r = adev->virt.map_csa (adev, vm); +if (r) + goto error_free_page_directory; + } +Just completely drop that. Updating the VM on the first command submission should be sufficient.
Christian.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0return 0 ; error_free_page_directory:@@ -1606,6 +1614,9 @@ void amdgpu_vm_fini(struct amdgpu_device*adev, struct
amdgpu_vm*vm)Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3
-0struct amdgpu_bo_va_mapping *mapping, *tmp;int i; +if (amdgpu_sriov_vf (adev)) + adev->virt.unmap_csa (adev, vm); +amd_sched_entity_fini (vm->entity.sched, &vm->entity);if (!RB_EMPTY_ROOT (&vm->va)) {
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx