drm/amdgpu: extend lock range for race condition when gpu reset
Subsystems (2)
PATCH 1/1 drm/amdgpu: extend lock range for race condition when gpu reset
to cover below case:
1. A task gart bind/unbind but not add to adev->gtt_list yet
2. at this time gpu reset, gtt only recover those gtt in adev->gtt_list
--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 6 ++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 22 ++++++++++++++-------- 3 files changed, 19 insertions(+), 11 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -1
@@ -556,7 +556,7 @@ int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev);void amdgpu_gart_table_vram_unpin (struct amdgpu_device *adev);int amdgpu_gart_init (struct amdgpu_device *adev);void amdgpu_gart_fini (struct amdgpu_device *adev); -void amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset, +int amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset,int pages); int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset, int pages, struct page **pagelist,
drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +4 -2
@@ -221,8 +221,9 @@ void amdgpu_gart_table_vram_free(struct amdgpu_device *adev)* * Unbinds the requested pages from the gart page table and * replaces them with the dummy page (all asics). +* Returns 0 for success, -EINVAL for failure. */ -void amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset, +int amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset,int pages) {unsigned t;
@@ -234,7 +235,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset,if (!adev->gart.ready) {WARN (1 ,"trying to unbind memory from uninitialized GART !\n" ); -return ; +return -EINVAL ; } t = offset /AMDGPU_GPU_PAGE_SIZE ;
@@ -255,6 +256,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, }mb ();amdgpu_gart_flush_gpu_tlb (adev,0 ); +return 0 ; }/**
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +14 -8
@@ -745,6 +745,7 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *bo_mem)return r; } +spin_lock (>t->adev->gtt_list_lock); flags =amdgpu_ttm_tt_pte_flags (gtt->adev, ttm, bo_mem); gtt->offset = (u64 )bo_mem->start <<PAGE_SHIFT ; r = amdgpu_gart_bind(gtt->adev, gtt->offset, ttm->num_pages,
@@ -753,12 +754,13 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *bo_mem)if (r) {DRM_ERROR ("failed to bind %lu pages at 0x%08llX\n" , ttm->num_pages, gtt->offset); -return r; + goto error_gart_bind; } -spin_lock (>t->adev->gtt_list_lock); +list_add_tail (>t->list, >t->adev->gtt_list); +error_gart_bind:spin_unlock (>t->adev->gtt_list_lock); -return 0 ; +return r; }int amdgpu_ttm_recover_gart (struct amdgpu_device *adev)
@@ -789,6 +791,7 @@ int amdgpu_ttm_recover_gart(struct amdgpu_device *adev)static int amdgpu_ttm_backend_unbind (struct ttm_tt *ttm) {struct amdgpu_ttm_tt *gtt = (void *)ttm; +int r;if (gtt->userptr)amdgpu_ttm_tt_unpin_userptr (ttm);
@@ -797,14 +800,17 @@ static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm)return 0 ;/* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */ -if (gtt->adev->gart.ready) -amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); -spin_lock (>t->adev->gtt_list_lock); + r =amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); +if (r) { +DRM_ERROR ("failed to unbind %lu pages at 0x%08llX\n" , + gtt->ttm.ttm.num_pages, gtt->offset); + goto error_unbind; + }list_del_init (>t->list); +error_unbind:spin_unlock (>t->adev->gtt_list_lock); - -return 0 ; +return r; }static void amdgpu_ttm_backend_destroy (struct ttm_tt *ttm)
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): 1493968962-10463-1-git-send-email-Hongbo.He-5C7GfCeVMHo@public.gmane.org
zhoucm1
Reviewed-by Chunming ZhouQuoted 107 lines · Roger.He
to cover below case:
1. A task gart bind/unbind but not add toadev->gtt_listyet
2. at this time gpu reset, gtt only recover those gtt inadev->gtt_listChange-Id Ifb4360e3b68624f2be67fa82100623cf4c451873Signed-off-by Roger.He <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 6 ++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 22 ++++++++++++++-------- 3 files changed, 19 insertions(+), 11 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +1
-1@@ -556,7 +556,7 @@ int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev);void amdgpu_gart_table_vram_unpin (struct amdgpu_device *adev);int amdgpu_gart_init (struct amdgpu_device *adev);void amdgpu_gart_fini (struct amdgpu_device *adev); -void amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset, +int amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset,int pages); int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset, int pages, struct page **pagelist,drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +4
-2@@ -221,8 +221,9 @@ void amdgpu_gart_table_vram_free(struct amdgpu_device *adev)* * Unbinds the requested pages from the gart page table and * replaces them with the dummy page (all asics). +* Returns 0 for success, -EINVAL for failure. */ -void amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset, +int amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset,int pages) {unsigned t;@@ -234,7 +235,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset,if (!adev->gart.ready) {WARN (1 ,"trying to unbind memory from uninitialized GART !\n" ); -return ; +return -EINVAL ; } t = offset /AMDGPU_GPU_PAGE_SIZE ;@@ -255,6 +256,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, }mb ();amdgpu_gart_flush_gpu_tlb (adev,0 ); +return 0 ; }/** drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +11
-6@@ -745,6 +745,7 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *bo_mem)return r; } +spin_lock (>t->adev->gtt_list_lock); flags =amdgpu_ttm_tt_pte_flags (gtt->adev, ttm, bo_mem); gtt->offset = (u64 )bo_mem->start <<PAGE_SHIFT ; r = amdgpu_gart_bind(gtt->adev, gtt->offset, ttm->num_pages,@@ -753,12 +754,13 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *bo_mem)if (r) {DRM_ERROR ("failed to bind %lu pages at 0x%08llX\n" , ttm->num_pages, gtt->offset); -return r; + goto error_gart_bind; } -spin_lock (>t->adev->gtt_list_lock); +list_add_tail (>t->list, >t->adev->gtt_list); +error_gart_bind:spin_unlock (>t->adev->gtt_list_lock); -return 0 ; +return r; }int amdgpu_ttm_recover_gart (struct amdgpu_device *adev)@@ -789,6 +791,7 @@ int amdgpu_ttm_recover_gart(struct amdgpu_device *adev)static int amdgpu_ttm_backend_unbind (struct ttm_tt *ttm) {struct amdgpu_ttm_tt *gtt = (void *)ttm; +int r;if (gtt->userptr)amdgpu_ttm_tt_unpin_userptr (ttm);@@ -797,14 +800,17 @@ static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm)return 0 ;/* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */ -if (gtt->adev->gart.ready) -amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); -spin_lock (>t->adev->gtt_list_lock); + r =amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); +if (r) { +DRM_ERROR ("failed to unbind %lu pages at 0x%08llX\n" , + gtt->ttm.ttm.num_pages, gtt->offset); + goto error_unbind;
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) · line 816
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +3
-2+ }list_del_init (>t->list); +error_unbind:spin_unlock (>t->adev->gtt_list_lock); - -return 0 ; +return r; }static void amdgpu_ttm_backend_destroy (struct ttm_tt *ttm)
_______________________________________________
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): [email protected]
Christian König
Reviewed-byAm 05.05.2017 um 09:33 schrieb zhoucm1:
Quoted 118 lines
Reviewed-by Chunming Zhou <[email protected]>Quoted 115 lines · Roger.He
to cover below case:
1. A task gart bind/unbind but not add toadev->gtt_listyet
2. at this time gpu reset, gtt only recover those gtt inadev->gtt_listChange-Id Ifb4360e3b68624f2be67fa82100623cf4c451873Signed-off-by Roger.He <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 6 ++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 22 ++++++++++++++-------- 3 files changed, 19 insertions(+), 11 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +1
-1@@ -556,7 +556,7 @@ int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev);void amdgpu_gart_table_vram_unpin (struct amdgpu_device *adev);int amdgpu_gart_init (struct amdgpu_device *adev);void amdgpu_gart_fini (struct amdgpu_device *adev); -void amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset, +int amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset,int pages); int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,int pages, struct page **pagelist,
drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +2
-1@@ -221,8 +221,9 @@ void amdgpu_gart_table_vram_free(structamdgpu_device *adev) * * Unbinds the requested pages from the gart page table and * replaces them with the dummy page (all asics). +* Returns 0 for success, -EINVAL for failure. */ -void amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset, +int amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset,int pages) {unsigned t;drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +2
-1@@ -234,7 +235,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, if (!adev->gart.ready) {WARN (1 ,"trying to unbind memory from uninitialized GART !\n"); -return ; +return -EINVAL ; } t = offset /AMDGPU_GPU_PAGE_SIZE ;@@ -255,6 +256,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, }mb ();amdgpu_gart_flush_gpu_tlb (adev,0 ); +return 0 ; }/** drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +12
-8@@ -745,6 +745,7 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *bo_mem)return r; } +spin_lock (>t->adev->gtt_list_lock); flags =amdgpu_ttm_tt_pte_flags (gtt->adev, ttm, bo_mem); gtt->offset = (u64 )bo_mem->start <<PAGE_SHIFT ; r = amdgpu_gart_bind(gtt->adev, gtt->offset, ttm->num_pages,@@ -753,12 +754,13 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *bo_mem)if (r) {DRM_ERROR ("failed to bind %lu pages at 0x%08llX\n" , ttm->num_pages, gtt->offset); -return r; + goto error_gart_bind; } -spin_lock (>t->adev->gtt_list_lock); +list_add_tail (>t->list, >t->adev->gtt_list); +error_gart_bind:spin_unlock (>t->adev->gtt_list_lock); -return 0 ; +return r; }int amdgpu_ttm_recover_gart (struct amdgpu_device *adev)@@ -789,6 +791,7 @@ int amdgpu_ttm_recover_gart(struct amdgpu_device *adev)static int amdgpu_ttm_backend_unbind (struct ttm_tt *ttm) {struct amdgpu_ttm_tt *gtt = (void *)ttm; +int r;if (gtt->userptr)amdgpu_ttm_tt_unpin_userptr (ttm);@@ -797,14 +800,17 @@ static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm)return 0 ;/* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */ -if (gtt->adev->gart.ready) -amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); -spin_lock (>t->adev->gtt_list_lock); + r =amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); +if (r) { +DRM_ERROR ("failed to unbind %lu pages at 0x%08llX\n" , + gtt->ttm.ttm.num_pages, gtt->offset); + goto error_unbind; + }list_del_init (>t->list); +error_unbind:spin_unlock (>t->adev->gtt_list_lock); - -return 0 ;
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) · line 816
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +1
-0+return r; }static void amdgpu_ttm_backend_destroy (struct ttm_tt *ttm)_______________________________________________
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): 50c920fd-2645-8d9c-44d5-34a97f3c6f85-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
Yu, Xiangliang
Reviewed-by Christian KönigHi Hongbo,
I got soft lockup message as below when running valkan test with two VFs, and work fine if reverting this patch. Could you help check it when you feel free?
[ 1920.163455] NMI watchdog: BUG: soft lockup - CPU#1 stuck for 23s! [deqp-vk:2175] [ 1920.163459] Modules linked in: amdkfd amd_iommu_v2 amdgpu(OE) ttm drm_kms_helper drm i2c_algo_bit fb_sys_fops syscopyarea sysfillrect sysimgblt snd_hda_codec_generic snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_pcm snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq crct10dif_pclmul crc32_pclmul ghash_clmulni_intel snd_seq_device snd_timer aesni_intel aes_x86_64 lrw gf128mul glue_helper ablk_helper cryptd serio_raw snd mac_hid soundcore i2c_piix4 binfmt_misc parport_pc ppdev sunrpc lp parport autofs4 psmouse pata_acpi floppy [ 1920.163482] CPU: 1 PID: 2175 Comm: deqp-vk Tainted: G OE 4.9.0-custom #1 [ 1920.163483] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014 [ 1920.163484] task: ffff880139c63a80 task.stack: ffffc9000313c000 [ 1920.163485] RIP: 0010:[<ffffffff810c7c75>] [<ffffffff810c7c75>] native_queued_spin_lock_slowpath+0x15/0x1a0 [ 1920.163491] RSP: 0018:ffffc9000313f6d8 EFLAGS: 00000202 [ 1920.163492] RAX: 0000000000000001 RBX: ffff88012f054d80 RCX: 0000000000000000 [ 1920.163492] RDX: 0000000000000001 RSI: 0000000000000001 RDI: ffff8801334899a8 [ 1920.163493] RBP: ffffc9000313f6d8 R08: 0000000000000000 R09: 0000000000000128 [ 1920.163493] R10: 000001b9269bb35e R11: 0000000000000001 R12: ffff88012f054e28 [ 1920.163494] R13: ffff88012f054d80 R14: ffff8800369ecc00 R15: ffff880133482778 [ 1920.163495] FS: 00007fc0cf438740(0000) GS:ffff88013fc80000(0000) knlGS:0000000000000000 [ 1920.163496] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1920.163496] CR2: 00007fc0c5100fa0 CR3: 000000013870d000 CR4: 00000000001406e0 [ 1920.163500] Stack: [ 1920.163500] ffffc9000313f6e8 ffffffff811816ef ffffc9000313f6f8 ffffffff817e1a10 [ 1920.163502] ffffc9000313f738 ffffffffa03242c0 ffff8800369ecc00 0000000000000400 [ 1920.163503] 0000000000000001 ffff88012f054d80 0000000000000002 ffff88012e831380 [ 1920.163505] Call Trace: [ 1920.163509] [<ffffffff811816ef>] queued_spin_lock_slowpath+0xb/0xf [ 1920.163513] [<ffffffff817e1a10>] _raw_spin_lock+0x20/0x30 [ 1920.163562] [<ffffffffa03242c0>] amdgpu_ttm_backend_unbind+0x50/0x1e0 [amdgpu] [ 1920.163567] [<ffffffffa02ecbee>] ttm_tt_unbind+0x1e/0x30 [ttm] [ 1920.163570] [<ffffffffa02ece27>] ttm_tt_destroy+0x17/0x60 [ttm] [ 1920.163572] [<ffffffffa02ed370>] ttm_bo_cleanup_memtype_use+0x30/0x70 [ttm] [ 1920.163575] [<ffffffffa02ee4fa>] ttm_bo_release+0x1ca/0x2a0 [ttm] [ 1920.163578] [<ffffffffa02ee5f4>] ttm_bo_unref+0x24/0x30 [ttm] [ 1920.163580] [<ffffffffa02f1ce7>] ttm_bo_pipeline_move+0x2a7/0x3a0 [ttm] [ 1920.163599] [<ffffffffa0325fbc>] amdgpu_move_blit+0x1bc/0x260 [amdgpu] [ 1920.163617] [<ffffffffa03263e9>] amdgpu_bo_move+0xb9/0x230 [amdgpu] [ 1920.163620] [<ffffffffa02eec38>] ttm_bo_handle_move_mem+0x268/0x590 [ttm] [ 1920.163623] [<ffffffffa02ef7ed>] ? ttm_bo_mem_space+0x38d/0x440 [ttm] [ 1920.163625] [<ffffffff8103d583>] ? __save_stack_trace+0x73/0xd0 [ 1920.163628] [<ffffffffa02efd74>] ttm_bo_validate+0x114/0x130 [ttm] [ 1920.163657] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163675] [<ffffffffa032e325>] amdgpu_cs_bo_validate.isra.5+0x75/0xb0 [amdgpu] [ 1920.163693] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163710] [<ffffffffa032e3a9>] amdgpu_cs_validate+0x49/0x1b0 [amdgpu] [ 1920.163727] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163743] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163762] [<ffffffffa033d57f>] amdgpu_vm_validate_level.isra.9+0x4f/0x90 [amdgpu] [ 1920.163791] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163807] [<ffffffffa033d596>] amdgpu_vm_validate_level.isra.9+0x66/0x90 [amdgpu] [ 1920.163823] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163840] [<ffffffffa033d596>] amdgpu_vm_validate_level.isra.9+0x66/0x90 [amdgpu] [ 1920.163856] [<ffffffffa033e606>] amdgpu_vm_validate_pt_bos+0x26/0x30 [amdgpu] [ 1920.163872] [<ffffffffa032fef8>] amdgpu_cs_ioctl+0xca8/0x1490 [amdgpu] [ 1920.163889] [<ffffffffa023cbac>] drm_ioctl+0x32c/0x440 [drm] [ 1920.163904] [<ffffffffa032f250>] ? amdgpu_cs_find_mapping+0xb0/0xb0 [amdgpu] [ 1920.163906] [<ffffffff811fe356>] ? mem_cgroup_commit_charge+0x76/0xe0 [ 1920.163908] [<ffffffff811c5529>] ? page_add_new_anon_rmap+0x89/0xc0 [ 1920.163910] [<ffffffff81194249>] ? lru_cache_add_active_or_unevictable+0x39/0xc0 [ 1920.163925] [<ffffffffa031104c>] amdgpu_drm_ioctl+0x4c/0x80 [amdgpu] [ 1920.163927] [<ffffffff81220db6>] do_vfs_ioctl+0x96/0x5b0 [ 1920.163929] [<ffffffff810674d7>] ? __do_page_fault+0x267/0x4d0 [ 1920.163930] [<ffffffff81221349>] SyS_ioctl+0x79/0x90 [ 1920.163932] [<ffffffff8100392e>] do_syscall_64+0x6e/0x180 [ 1920.163933] [<ffffffff817e1d2f>] entry_SYSCALL64_slow_path+0x25/0x25
-----Original Message-----
From: amd-gfx [mailto:[email protected]] On Behalf Of Christian K?nig
Sent: Friday, May 05, 2017 8:04 PM
To: Zhou, David(ChunMing) <[email protected]>; He, Hongbo <[email protected]>; [email protected]
Subject: Re: [PATCH] drm/amdgpu: extend lock range for race condition when gpu reset
Am 05.05.2017 um 09:33 schrieb zhoucm1:
Quoted 119 lines
Reviewed-by Chunming Zhou <[email protected]>Quoted 116 lines · Roger.He
to cover below case:
1. A task gart bind/unbind but not add toadev->gtt_listyet 2. at this time gpu reset, gtt only recover those gtt inadev->gtt_listChange-Id Ifb4360e3b68624f2be67fa82100623cf4c451873Signed-off-by Roger.He <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 6 ++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 22 ++++++++++++++-------- 3 files changed, 19 insertions(+), 11 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +1
-0@@ -556,7 +556,7 @@ int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev);void amdgpu_gart_table_vram_unpin (struct amdgpu_device *adev);int amdgpu_gart_init (struct amdgpu_device *adev); void amdgpu_gart_fini(struct amdgpu_device *adev); -void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, +int amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset,int pages);int amdgpu_gart_bind (struct amdgpu_device *adev,uint64_t offset, int pages, struct page **pagelist, diff --git
a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.cb/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index e7406ce..ccef3cf 100644drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +2
-1@@ -221,8 +221,9 @@ void amdgpu_gart_table_vram_free(structamdgpu_device *adev) * * Unbinds the requested pages from the gart page table and * replaces them with the dummy page (all asics). +* Returns 0 for success, -EINVAL for failure. */ -void amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset, +int amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset,int pages) {unsigned t;drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +1
-1@@ -234,7 +235,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, if (!adev->gart.ready) {WARN (1 ,"trying to unbind memory from uninitialized GART !\n"); -return ; +return -EINVAL ; } t = offset / AMDGPU_GPU_PAGE_SIZE; @@ -255,6 +256,7 @@ voidamdgpu_gart_unbind(struct
amdgpu_device*adev,uint64_toffset,} mb(); amdgpu_gart_flush_gpu_tlb(adev, 0);drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +1
-0+return 0 ; }/** drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0
-1@@ -745,6 +745,7 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *bo_mem)return r; } +spin_lock (>t->adev->gtt_list_lock); flags =amdgpu_ttm_tt_pte_flags (gtt->adev, ttm, bo_mem); gtt->offset = (u64 )bo_mem->start <<PAGE_SHIFT ; r = amdgpu_gart_bind(gtt->adev, gtt->offset, ttm->num_pages, @@ -753,12 +754,13 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo,struct
ttm_mem_reg*bo_mem)if (r) { DRM_ERROR("failed to bind %lu pages at 0x%08llX\n", ttm->num_pages, gtt->offset);drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +5
-3-return r; + goto error_gart_bind; } -spin_lock (>t->adev->gtt_list_lock); +list_add_tail (>t->list, >t->adev->gtt_list); +error_gart_bind:spin_unlock (>t->adev->gtt_list_lock); -return 0 ; +return r; } int amdgpu_ttm_recover_gart(struct amdgpu_device *adev) @@ -789,6 +791,7 @@ int amdgpu_ttm_recover_gart(struct amdgpu_device*adev)
static intamdgpu_ttm_backend_unbind(struct ttm_tt *ttm){ struct amdgpu_ttm_tt *gtt = (void *)ttm; + int r; if (gtt->userptr) amdgpu_ttm_tt_unpin_userptr(ttm); @@ -797,14 +800,17 @@ static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) return 0; /* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +7
-5-if (gtt->adev->gart.ready) -amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); -spin_lock (>t->adev->gtt_list_lock); + r =amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); +if (r) { +DRM_ERROR ("failed to unbind %lu pages at 0x%08llX\n" , + gtt->ttm.ttm.num_pages, gtt->offset); + goto error_unbind; + }list_del_init (>t->list); +error_unbind:spin_unlock (>t->adev->gtt_list_lock); - -return 0 ;
Quoted 8 lines
Quoted 3 lines
+ return r;
}
static voidamdgpu_ttm_backend_destroy(struct ttm_tt *ttm)_______________________________________________
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
_______________________________________________
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): BY2PR1201MB093501712C1A8297EA44A45CEBC10-O28G1zQ8oGkaqtME6NEo1mrFom/aUZj6nBOFsp37pqbUKgpGm//BTAC/[email protected]
He, Roger
Reviewed-by Christian KönigHi Xiangliang:
Could you try to replace spinlock with mutex?
Thanks
Roger(Hongbo.He)
-----Original Message-----
From: Yu, Xiangliang
Sent: Friday, June 16, 2017 2:30 PM
To: Christian König <[email protected]>; Zhou, David(ChunMing) <[email protected]>; He, Roger <[email protected]>; [email protected]
Subject: RE: [PATCH] drm/amdgpu: extend lock range for race condition when gpu reset
Hi Hongbo,
I got soft lockup message as below when running valkan test with two VFs, and work fine if reverting this patch. Could you help check it when you feel free?
[ 1920.163455] NMI watchdog: BUG: soft lockup - CPU#1 stuck for 23s! [deqp-vk:2175] [ 1920.163459] Modules linked in: amdkfd amd_iommu_v2 amdgpu(OE) ttm drm_kms_helper drm i2c_algo_bit fb_sys_fops syscopyarea sysfillrect sysimgblt snd_hda_codec_generic snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_pcm snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq crct10dif_pclmul crc32_pclmul ghash_clmulni_intel snd_seq_device snd_timer aesni_intel aes_x86_64 lrw gf128mul glue_helper ablk_helper cryptd serio_raw snd mac_hid soundcore i2c_piix4 binfmt_misc parport_pc ppdev sunrpc lp parport autofs4 psmouse pata_acpi floppy [ 1920.163482] CPU: 1 PID: 2175 Comm: deqp-vk Tainted: G OE 4.9.0-custom #1 [ 1920.163483] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014 [ 1920.163484] task: ffff880139c63a80 task.stack: ffffc9000313c000 [ 1920.163485] RIP: 0010:[<ffffffff810c7c75>] [<ffffffff810c7c75>] native_queued_spin_lock_slowpath+0x15/0x1a0 [ 1920.163491] RSP: 0018:ffffc9000313f6d8 EFLAGS: 00000202 [ 1920.163492] RAX: 0000000000000001 RBX: ffff88012f054d80 RCX: 0000000000000000 [ 1920.163492] RDX: 0000000000000001 RSI: 0000000000000001 RDI: ffff8801334899a8 [ 1920.163493] RBP: ffffc9000313f6d8 R08: 0000000000000000 R09: 0000000000000128 [ 1920.163493] R10: 000001b9269bb35e R11: 0000000000000001 R12: ffff88012f054e28 [ 1920.163494] R13: ffff88012f054d80 R14: ffff8800369ecc00 R15: ffff880133482778 [ 1920.163495] FS: 00007fc0cf438740(0000) GS:ffff88013fc80000(0000) knlGS:0000000000000000 [ 1920.163496] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1920.163496] CR2: 00007fc0c5100fa0 CR3: 000000013870d000 CR4: 00000000001406e0 [ 1920.163500] Stack: [ 1920.163500] ffffc9000313f6e8 ffffffff811816ef ffffc9000313f6f8 ffffffff817e1a10 [ 1920.163502] ffffc9000313f738 ffffffffa03242c0 ffff8800369ecc00 0000000000000400 [ 1920.163503] 0000000000000001 ffff88012f054d80 0000000000000002 ffff88012e831380 [ 1920.163505] Call Trace: [ 1920.163509] [<ffffffff811816ef>] queued_spin_lock_slowpath+0xb/0xf [ 1920.163513] [<ffffffff817e1a10>] _raw_spin_lock+0x20/0x30 [ 1920.163562] [<ffffffffa03242c0>] amdgpu_ttm_backend_unbind+0x50/0x1e0 [amdgpu] [ 1920.163567] [<ffffffffa02ecbee>] ttm_tt_unbind+0x1e/0x30 [ttm] [ 1920.163570] [<ffffffffa02ece27>] ttm_tt_destroy+0x17/0x60 [ttm] [ 1920.163572] [<ffffffffa02ed370>] ttm_bo_cleanup_memtype_use+0x30/0x70 [ttm] [ 1920.163575] [<ffffffffa02ee4fa>] ttm_bo_release+0x1ca/0x2a0 [ttm] [ 1920.163578] [<ffffffffa02ee5f4>] ttm_bo_unref+0x24/0x30 [ttm] [ 1920.163580] [<ffffffffa02f1ce7>] ttm_bo_pipeline_move+0x2a7/0x3a0 [ttm] [ 1920.163599] [<ffffffffa0325fbc>] amdgpu_move_blit+0x1bc/0x260 [amdgpu] [ 1920.163617] [<ffffffffa03263e9>] amdgpu_bo_move+0xb9/0x230 [amdgpu] [ 1920.163620] [<ffffffffa02eec38>] ttm_bo_handle_move_mem+0x268/0x590 [ttm] [ 1920.163623] [<ffffffffa02ef7ed>] ? ttm_bo_mem_space+0x38d/0x440 [ttm] [ 1920.163625] [<ffffffff8103d583>] ? __save_stack_trace+0x73/0xd0 [ 1920.163628] [<ffffffffa02efd74>] ttm_bo_validate+0x114/0x130 [ttm] [ 1920.163657] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163675] [<ffffffffa032e325>] amdgpu_cs_bo_validate.isra.5+0x75/0xb0 [amdgpu] [ 1920.163693] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163710] [<ffffffffa032e3a9>] amdgpu_cs_validate+0x49/0x1b0 [amdgpu] [ 1920.163727] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163743] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163762] [<ffffffffa033d57f>] amdgpu_vm_validate_level.isra.9+0x4f/0x90 [amdgpu] [ 1920.163791] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163807] [<ffffffffa033d596>] amdgpu_vm_validate_level.isra.9+0x66/0x90 [amdgpu] [ 1920.163823] [<ffffffffa032e360>] ? amdgpu_cs_bo_validate.isra.5+0xb0/0xb0 [amdgpu] [ 1920.163840] [<ffffffffa033d596>] amdgpu_vm_validate_level.isra.9+0x66/0x90 [amdgpu] [ 1920.163856] [<ffffffffa033e606>] amdgpu_vm_validate_pt_bos+0x26/0x30 [amdgpu] [ 1920.163872] [<ffffffffa032fef8>] amdgpu_cs_ioctl+0xca8/0x1490 [amdgpu] [ 1920.163889] [<ffffffffa023cbac>] drm_ioctl+0x32c/0x440 [drm] [ 1920.163904] [<ffffffffa032f250>] ? amdgpu_cs_find_mapping+0xb0/0xb0 [amdgpu] [ 1920.163906] [<ffffffff811fe356>] ? mem_cgroup_commit_charge+0x76/0xe0 [ 1920.163908] [<ffffffff811c5529>] ? page_add_new_anon_rmap+0x89/0xc0 [ 1920.163910] [<ffffffff81194249>] ? lru_cache_add_active_or_unevictable+0x39/0xc0 [ 1920.163925] [<ffffffffa031104c>] amdgpu_drm_ioctl+0x4c/0x80 [amdgpu] [ 1920.163927] [<ffffffff81220db6>] do_vfs_ioctl+0x96/0x5b0 [ 1920.163929] [<ffffffff810674d7>] ? __do_page_fault+0x267/0x4d0 [ 1920.163930] [<ffffffff81221349>] SyS_ioctl+0x79/0x90 [ 1920.163932] [<ffffffff8100392e>] do_syscall_64+0x6e/0x180 [ 1920.163933] [<ffffffff817e1d2f>] entry_SYSCALL64_slow_path+0x25/0x25
-----Original Message-----
From: amd-gfx [mailto:[email protected]] On Behalf Of Christian K?nig
Sent: Friday, May 05, 2017 8:04 PM
To: Zhou, David(ChunMing) <[email protected]>; He, Hongbo <[email protected]>; [email protected]
Subject: Re: [PATCH] drm/amdgpu: extend lock range for race condition when gpu reset
Am 05.05.2017 um 09:33 schrieb zhoucm1:
Quoted 119 lines
Reviewed-by Chunming Zhou <[email protected]>Quoted 116 lines · Roger.He
to cover below case:
1. A task gart bind/unbind but not add toadev->gtt_listyet 2. at this time gpu reset, gtt only recover those gtt inadev->gtt_listChange-Id Ifb4360e3b68624f2be67fa82100623cf4c451873Signed-off-by Roger.He <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 6 ++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 22 ++++++++++++++-------- 3 files changed, 19 insertions(+), 11 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +1
-0@@ -556,7 +556,7 @@ int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev);void amdgpu_gart_table_vram_unpin (struct amdgpu_device *adev);int amdgpu_gart_init (struct amdgpu_device *adev); void amdgpu_gart_fini(struct amdgpu_device *adev); -void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, +int amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset,int pages);int amdgpu_gart_bind (struct amdgpu_device *adev,uint64_t offset, int pages, struct page **pagelist, diff --git
a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.cb/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index e7406ce..ccef3cf 100644drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +2
-1@@ -221,8 +221,9 @@ void amdgpu_gart_table_vram_free(structamdgpu_device *adev) * * Unbinds the requested pages from the gart page table and * replaces them with the dummy page (all asics). +* Returns 0 for success, -EINVAL for failure. */ -void amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset, +int amdgpu_gart_unbind (struct amdgpu_device *adev,uint64_t offset,int pages) {unsigned t;drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +1
-1@@ -234,7 +235,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, if (!adev->gart.ready) {WARN (1 ,"trying to unbind memory from uninitialized GART !\n"); -return ; +return -EINVAL ; } t = offset / AMDGPU_GPU_PAGE_SIZE; @@ -255,6 +256,7 @@ voidamdgpu_gart_unbind(struct
amdgpu_device*adev,uint64_toffset,} mb(); amdgpu_gart_flush_gpu_tlb(adev, 0);drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +1
-0+return 0 ; }/** drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0
-1@@ -745,6 +745,7 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *bo_mem)return r; } +spin_lock (>t->adev->gtt_list_lock); flags =amdgpu_ttm_tt_pte_flags (gtt->adev, ttm, bo_mem); gtt->offset = (u64 )bo_mem->start <<PAGE_SHIFT ; r = amdgpu_gart_bind(gtt->adev, gtt->offset, ttm->num_pages, @@ -753,12 +754,13 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo,struct
ttm_mem_reg*bo_mem)if (r) { DRM_ERROR("failed to bind %lu pages at 0x%08llX\n", ttm->num_pages, gtt->offset);drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +5
-3-return r; + goto error_gart_bind; } -spin_lock (>t->adev->gtt_list_lock); +list_add_tail (>t->list, >t->adev->gtt_list); +error_gart_bind:spin_unlock (>t->adev->gtt_list_lock); -return 0 ; +return r; } int amdgpu_ttm_recover_gart(struct amdgpu_device *adev) @@ -789,6 +791,7 @@ int amdgpu_ttm_recover_gart(struct amdgpu_device*adev)
static intamdgpu_ttm_backend_unbind(struct ttm_tt *ttm){ struct amdgpu_ttm_tt *gtt = (void *)ttm; + int r; if (gtt->userptr) amdgpu_ttm_tt_unpin_userptr(ttm); @@ -797,14 +800,17 @@ static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) return 0; /* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +7
-5-if (gtt->adev->gart.ready) -amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); -spin_lock (>t->adev->gtt_list_lock); + r =amdgpu_gart_unbind (gtt->adev, gtt->offset, ttm->num_pages); +if (r) { +DRM_ERROR ("failed to unbind %lu pages at 0x%08llX\n" , + gtt->ttm.ttm.num_pages, gtt->offset); + goto error_unbind; + }list_del_init (>t->list); +error_unbind:spin_unlock (>t->adev->gtt_list_lock); - -return 0 ;
Quoted 8 lines
Quoted 3 lines
+ return r;
}
static voidamdgpu_ttm_backend_destroy(struct ttm_tt *ttm)_______________________________________________
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
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx