- 09 Jun, 2023 40 commits
-
-
Yifan Zhang authored
gfx 11.0.4 range starts from 0x80. Fixes: 311d5236 ("drm/amdgpu: add soc21 common ip block support for GC 11.0.4") Signed-off-by:
Yifan Zhang <yifan1.zhang@amd.com> Reported-by:
Yogesh Mohan Marimuthu <Yogesh.Mohanmarimuthu@amd.com> Acked-by:
Alex Deucher <alexander.deucher@amd.com> Reviewed-by:
Tim Huang <Tim.Huang@amd.com> Signed-off-by:
-
Srinivasan Shanmugam authored
Fix below warnings reported by checkpatch: WARNING: Prefer 'unsigned int' to bare use of 'unsigned' WARNING: static const char * array should probably be static const char * const WARNING: space prohibited between function name and open parenthesis '(' WARNING: braces {} are not necessary for single statement blocks WARNING: Symbolic permissions 'S_IRUGO' are not preferred. Consider using octal permissions '0444'. Cc: Christian König <christian.koenig@amd.com> Cc: Alex Deucher <alexander.deucher@amd.com> Signed-off-by:Srinivasan Shanmugam <srinivasan.shanmugam@amd.com> Acked-by:
-
Guilherme G. Piccoli authored
(Bas: speculative change to mirror gfx10/gfx9) Signed-off-by:
Guilherme G. Piccoli <gpiccoli@igalia.com> Signed-off-by:
Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl> Cc: Alex Deucher <alexander.deucher@amd.com> Signed-off-by:
-
Bas Nieuwenhuizen authored
Otherwise we get a full system lock (looks like a FW mess). Copied the order from the GFX9 powergating code. Fixes: 366468ff ("drm/amdgpu: Allow GfxOff on Vangogh as default") Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2545Signed-off-by:
-
Dan Carpenter authored
This function was releasing the incorrect lock on the error path. Reported-by:
kernel test robot <lkp@intel.com> Fixes: 1156e1a6 ("drm/amdgpu: add [en/dis]able_kgq() functions") Signed-off-by:
Dan Carpenter <dan.carpenter@linaro.org> Signed-off-by:
-
Huacai Chen authored
LoongArch now provides kernel_fpu_begin() and kernel_fpu_end() that are used like the x86 counterparts in commit 2b3bd32e ("LoongArch: Provide kernel fpu functions"), so we can enable DC_FP on LoongArch for supporting more DCN devices. Signed-off-by:
WANG Xuerui <kernel@xen0n.name> Signed-off-by:
Huacai Chen <chenhuacai@loongson.cn> Signed-off-by:
Hamza Mahfooz <hamza.mahfooz@amd.com> Signed-off-by:
-
Saleemkhan Jamadar authored
Register CC_UVD_HARVESTING is obsolete for JPEG 3.1.2 Signed-off-by:
Saleemkhan Jamadar <saleemkhan.jamadar@amd.com> Reviewed-by:
Veerabadhran Gopalakrishnan <Veerabadhran.Gopalakrishnan@amd.com> Signed-off-by:
-
Guchun Chen authored
gfx9 cp_ecc_error_irq is only enabled when legacy gfx ras is assert. So in gfx_v9_0_hw_fini, interrupt disablement for cp_ecc_error_irq should be executed under such condition, otherwise, an amdgpu_irq_put calltrace will occur. [ 7283.170322] RIP: 0010:amdgpu_irq_put+0x45/0x70 [amdgpu] [ 7283.170964] RSP: 0018:ffff9a5fc3967d00 EFLAGS: 00010246 [ 7283.170967] RAX: ffff98d88afd3040 RBX: ffff98d89da20000 RCX: 0000000000000000 [ 7283.170969] RDX: 0000000000000000 RSI: ffff98d89da2bef8 RDI: ffff98d89da20000 [ 7283.170971] RBP: ffff98d89da20000 R08: ffff98d89da2ca18 R09: 0000000000000006 [ 7283.170973] R10: ffffd5764243c008 R11: 0000000000000000 R12: 0000000000001050 [ 7283.170975] R13: ffff98d89da38978 R14: ffffffff999ae15a R15: ffff98d880130105 [ 7283.170978] FS: 0000000000000000(0000) GS:ffff98d996f00000(0000) knlGS:0000000000000000 [ 7283.170981] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 7283.170983] CR2: 00000000f7a9d178 CR3: 00000001c42ea000 CR4: 00000000003506e0 [ 7283.170986] Call Trace: [ 7283.170988] <TASK> [ 7283.170989] gfx_v9_0_hw_fini+0x1c/0x6d0 [amdgpu] [ 7283.171655] amdgpu_device_ip_suspend_phase2+0x101/0x1a0 [amdgpu] [ 7283.172245] amdgpu_device_suspend+0x103/0x180 [amdgpu] [ 7283.172823] amdgpu_pmops_freeze+0x21/0x60 [amdgpu] [ 7283.173412] pci_pm_freeze+0x54/0xc0 [ 7283.173419] ? __pfx_pci_pm_freeze+0x10/0x10 [ 7283.173425] dpm_run_callback+0x98/0x200 [ 7283.173430] __device_suspend+0x164/0x5f0 v2: drop gfx11 as it's fixed in a different solution by retiring cp_ecc_irq funcs(Hawking) Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2522Signed-off-by:
Guchun Chen <guchun.chen@amd.com> Reviewed-by:
Tao Zhou <tao.zhou1@amd.com> Signed-off-by:
-
Guchun Chen authored
Prevent further dpm casting on legacy asics without od_enabled in amdgpu_dpm_is_overdrive_supported. This can avoid UBSAN complain in init sequence. v2: add a macro to check legacy dpm instead of checking asic family/type v3: refine macro name for naming consistency Suggested-by:
Evan Quan <evan.quan@amd.com> Signed-off-by:
Lijo Lazar <lijo.lazar@amd.com> Signed-off-by:
-
Guchun Chen authored
sdma_v4_0_ip is shared on a few asics, but in sdma_v4_0_hw_fini, driver unconditionally disables ecc_irq which is only enabled on those asics enabling sdma ecc. This will introduce a warning in suspend cycle on those chips with sdma ip v4.0, while without sdma ecc. So this patch correct this. [ 7283.166354] RIP: 0010:amdgpu_irq_put+0x45/0x70 [amdgpu] [ 7283.167001] RSP: 0018:ffff9a5fc3967d08 EFLAGS: 00010246 [ 7283.167019] RAX: ffff98d88afd3770 RBX: 0000000000000001 RCX: 0000000000000000 [ 7283.167023] RDX: 0000000000000000 RSI: ffff98d89da30390 RDI: ffff98d89da20000 [ 7283.167025] RBP: ffff98d89da20000 R08: 0000000000036838 R09: 0000000000000006 [ 7283.167028] R10: ffffd5764243c008 R11: 0000000000000000 R12: ffff98d89da30390 [ 7283.167030] R13: ffff98d89da38978 R14: ffffffff999ae15a R15: ffff98d880130105 [ 7283.167032] FS: 0000000000000000(0000) GS:ffff98d996f00000(0000) knlGS:0000000000000000 [ 7283.167036] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 7283.167039] CR2: 00000000f7a9d178 CR3: 00000001c42ea000 CR4: 00000000003506e0 [ 7283.167041] Call Trace: [ 7283.167046] <TASK> [ 7283.167048] sdma_v4_0_hw_fini+0x38/0xa0 [amdgpu] [ 7283.167704] amdgpu_device_ip_suspend_phase2+0x101/0x1a0 [amdgpu] [ 7283.168296] amdgpu_device_suspend+0x103/0x180 [amdgpu] [ 7283.168875] amdgpu_pmops_freeze+0x21/0x60 [amdgpu] [ 7283.169464] pci_pm_freeze+0x54/0xc0 Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2522Signed-off-by:
-
Aric Cyr authored
This version brings along following fixes: - Block SubVP on displays that have pixclk > 1800Mhz - Block SubVP high refresh when VRR active fixed - Enforce 60us prefetch for 200Mhz DCFCLK modes - Check Vactive for VRR active for FPO + Vactive - Add symclk workaround during disable link output - Show the DCN/DCE version in the log - Add additional pstate registers to HW state query Acked-by:
Alex Hung <alex.hung@amd.com> Signed-off-by:
Aric Cyr <aric.cyr@amd.com> Tested-by:
Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by:
-
Lin.Cao authored
v1: Vmbo->shadow is used to back vram bo up when vram lost. So that we should set shadow as vmbo->shadow to recover vmbo->bo v2: Modify if(vmbo->shadow) shadow = vmbo->shadow as if(!vmbo->shadow) continue; Fixes: e18aaea7 ("drm/amdgpu: move shadow_list to amdgpu_bo_vm") Reviewed-by:
Christian König <christian.koenig@amd.com> Signed-off-by:
Lin.Cao <lincao12@amd.com> Signed-off-by:
-
Horatio Zhang authored
The gfx.cp_ecc_error_irq is retired in gfx11. In gfx_v11_0_hw_fini still use amdgpu_irq_put to disable this interrupt, which caused the call trace in this function. [ 102.873958] Call Trace: [ 102.873959] <TASK> [ 102.873961] gfx_v11_0_hw_fini+0x23/0x1e0 [amdgpu] [ 102.874019] gfx_v11_0_suspend+0xe/0x20 [amdgpu] [ 102.874072] amdgpu_device_ip_suspend_phase2+0x240/0x460 [amdgpu] [ 102.874122] amdgpu_device_ip_suspend+0x3d/0x80 [amdgpu] [ 102.874172] amdgpu_device_pre_asic_reset+0xd9/0x490 [amdgpu] [ 102.874223] amdgpu_device_gpu_recover.cold+0x548/0xce6 [amdgpu] [ 102.874321] amdgpu_debugfs_reset_work+0x4c/0x70 [amdgpu] [ 102.874375] process_one_work+0x21f/0x3f0 [ 102.874377] worker_thread+0x200/0x3e0 [ 102.874378] ? process_one_work+0x3f0/0x3f0 [ 102.874379] kthread+0xfd/0x130 [ 102.874380] ? kthread_complete_and_exit+0x20/0x20 [ 102.874381] ret_from_fork+0x22/0x30 v2: - Handle umc and gfx ras cases in separated patch - Retired the gfx_v11_0_cp_ecc_error_irq_funcs in gfx11 v3: - Improve the subject and code comments - Add judgment on gfx11 in the function of amdgpu_gfx_ras_late_init v4: - Drop the define of CP_ME1_PIPE_INST_ADDR_INTERVAL and SET_ECC_ME_PIPE_STATE which using in gfx_v11_0_set_cp_ecc_error_state - Check cp_ecc_error_irq.funcs rather than ip version for a more sustainable life v5: - Simplify judgment conditions Signed-off-by:
Horatio Zhang <Hongkun.Zhang@amd.com> Reviewed-by:
Hawking Zhang <Hawking.Zhang@amd.com> Acked-by:
Feifei Xu <Feifei.Xu@amd.com> Signed-off-by:
-
Alvin Lee authored
[Description] - Enabling SubVP on high refresh rate displays had a side effect of also enabling on high bandwidth displays such as 8K60 - However, these are not validated and should be blocked for the time being - Block SubVP on displays that have pix rate > 1800Mhz (includes 8K60 displays) Reviewed-by:
Jun Lei <Jun.Lei@amd.com> Reviewed-by:
Nevenko Stupar <Nevenko.Stupar@amd.com> Cc: Mario Limonciello <mario.limonciello@amd.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: stable@vger.kernel.org Acked-by:
Alvin Lee <Alvin.Lee2@amd.com> Tested-by:
-
Alvin Lee authored
[Description] - SubVP high refresh is blocked when VRR is active variable, but we should also block it for when VRR is active fixed (video use case) Reviewed-by:
-
Alvin Lee authored
[Description] - Due to bandwidth / arbitration issues at 200Mhz DCFCLK, we want to enforce minimum 60us of prefetch to avoid intermittent underflow issues - Since 60us prefetch is already enforced for UCLK DPM0, and many DCFCLK's > 200Mhz are mapped to UCLK DPM1, in theory there should not be any UCLK DPM regressions by enforcing greater prefetch Reviewed-by:
-
Alvin Lee authored
[Description] - For FPO + Vactive cases, we rely on the Vactive display to be at it's nominal refresh rate because the Vactive pipe may not necessarily assert P-State allow while it's in VBLANK - For cases where the Vactive display has a stretched VBLANK due to VRR, we could underflow when trying to complete an FPO + Vactive MCLK switch because the FPO display has limited VBLANK time in waiting for the Vactive display to assert P-State allow naturally - Block FPO + Vactive if the Vactive display has VRR active (variable or fixed) Reviewed-by:
-
Leo Chen authored
[Why & How] This is originally a change (9c75891f) in DCN32 because of the lack of interface to set TX while keeping symclk on. Adding this workaround to DCN314 will resolve the current issue. Fixes: 9c75891f ("drm/amd/display: rework recent update PHY state commit") Reviewed-by:
Nicholas Kazlauskas <Nicholas.Kazlauskas@amd.com> Acked-by:
Leo Chen <sancchen@amd.com> Tested-by:
-
YuBiao Wang authored
The number of kernel queues has impact on the latency under sriov usecase. So to reduce the latency we set the default num_kcq = 2 under sriov if not set manually. Signed-off-by:
YuBiao Wang <YuBiao.Wang@amd.com> Reviewed-by:
-
Guchun Chen authored
amdgpu_dpm_is_overdrive_supported is a common API across all asics, so we should cast pp_handle into correct structure under different power frameworks. v2: using return directly to simplify code v3: SI asic does not carry od_enabled member in pp_handle, and update Fixes tag Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2541 Fixes: eb4900aa ("drm/amdgpu: Fix kernel NULL pointer dereference in dpm functions") Suggested-by:
Mario Limonciello <mario.limonciello@amd.com> Signed-off-by:
-
Lang Yu authored
When using cpu to update page tables, vm update fences are unused. Install stub fence into these fence pointers instead of NULL to avoid NULL dereference when calling dma_fence_wait() on them. Suggested-by:
Lang Yu <Lang.Yu@amd.com> Reviewed-by:
-
Rodrigo Siqueira authored
Some times people send their dmesg log for debugging, and one common task is to check the modesetting line to catch which DCN/DCE we need to debug. This commit introduces a simple conversion from the DCN/DCE version to a string shown in the dmesg log. Reviewed-by:
Rodrigo Siqueira <Rodrigo.Siqueira@amd.com> Tested-by:
-
Jiapeng Chong authored
Variable golden_settings_gc_9_4_3 is not effectively used, so delete it. drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c:48:38: warning: ‘golden_settings_gc_9_4_3’ defined but not used. Reported-by:
Abaci Robot <abaci@linux.alibaba.com> Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=4877Signed-off-by:
Jiapeng Chong <jiapeng.chong@linux.alibaba.com> Signed-off-by:
-
Felix Kuehling authored
Don't move DMABuf attachments for PCIe P2P mappings to the SYSTEM domain when unmapping. This avoids triggering eviction fences unnecessarily. Instead do the move to SYSTEM and back to GTT when mapping these attachments to ensure the SG table gets updated after evictions. This may still trigger unnecessary evictions if user mode unmaps and remaps the same BO. However, this is unlikely in real applications. Signed-off-by:
Felix Kuehling <Felix.Kuehling@amd.com> Reviewed-by:
Eric Huang <jinhuieric.huang@amd.com> Signed-off-by:
-
Sung Lee authored
[WHY] These registers would be useful to know when debugging pstate issues. [HOW] Add additional registers to hw state query. Reviewed-by:
Sung Lee <sunglee@amd.com> Tested-by:
-
Jiapeng Chong authored
No functional modification involved. ./drivers/gpu/drm/amd/amdgpu/nbio_v7_9.c:146:2-3: Unneeded semicolon. Reported-by:
-
Le Ma authored
Each XCD needs to do gfxhub init Signed-off-by:
Le Ma <le.ma@amd.com> Reviewed-by:
-
Hamza Mahfooz authored
As made mention of in commit c56edea58c31 ("drm/amdgpu: fix amdgpu_irq_put call trace in gmc_v10_0_hw_fini") and commit aa6ac247ed7d ("drm/amdgpu: fix amdgpu_irq_put call trace in gmc_v11_0_hw_fini"). It is meaningless to call amdgpu_irq_put() for gmc.ecc_irq. So, remove it from gmc_v9_0_hw_fini(). Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2522 Fixes: c8b5a95b ("drm/amdgpu: Fix desktop freezed after gpu-reset") Reviewed-by: -
Dan Carpenter authored
Smatch complains that we need to drop this lock before returning. drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c:1838 gfx_v9_4_3_kiq_resume() warn: inconsistent returns 'ring->mqd_obj->tbo.base.resv'. Fixes: 86301129 ("drm/amdgpu: split gc v9_4_3 functionality from gc v9_0") Signed-off-by: -
Dan Carpenter authored
We changed which lock we are supposed to take but this error path was accidentally over looked so it still drops the old lock. Fixes: def799c6 ("drm/amdgpu: add multi-xcc support to amdgpu_gfx interfaces (v4)") Signed-off-by:
-
Alex Deucher authored
amdgpu_discovery_get_ip_version() has not been used since commit c40bdfb2 ("drm/amdgpu: fix incorrect VCN revision in SRIOV") so drop it. Reviewed-by:
Luben Tuikov <luben.tuikov@amd.com> Reviewed-by:
-
Alex Deucher authored
This was already fixed and dropped in: commit baf3f8f3 ("drm/amdgpu: handle SRIOV VCN revision parsing") commit c40bdfb2 ("drm/amdgpu: fix incorrect VCN revision in SRIOV") But seems to have been accidently been left around in a merge. Reviewed-by:
-
Alex Deucher authored
Reduces preemption latency. Only enable this for gfx10 and 11 for now to avoid changing behavior on gfx 8 and 9. v2: move MES MQDs into VRAM as well (YuBiao) v3: enable on gfx10, 11 only (Alex) v4: minor style changes, document why gfx10/11 only (Alex) Reviewed-by:
-
Guchun Chen authored
Once command submission failed due to userptr invalidation in amdgpu_cs_submit, legacy code will perform cleanup of scheduler job. However, it's not needed at all, as former commit has integrated job cleanup stuff into amdgpu_job_free. Otherwise, because of double free, a NULL pointer dereference will occur in such scenario. Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/2457 Fixes: f7d66fb2 ("drm/amdgpu: cleanup scheduler job initialization v2") Signed-off-by:
-
Srinivasan Shanmugam authored
The following checkpatch errors & warning is removed. ERROR: else should follow close brace '}' ERROR: trailing statements should be on next line WARNING: Prefer 'unsigned int' to bare use of 'unsigned' WARNING: Possible repeated word: 'Fences' WARNING: Missing a blank line after declarations WARNING: braces {} are not necessary for single statement blocks WARNING: Comparisons should place the constant on the right side of the test WARNING: printk() should include KERN_<LEVEL> facility level Cc: Christian König <christian.koenig@amd.com> Cc: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: -
Yifan Zhang authored
APUs w/ gfx9 onwards doesn't reply on PCIe atomics, rather it is internal path w/ native atomic support. Set have_atomics_support to true. Signed-off-by:
-
Alex Deucher authored
Always restore the MQD not just when we do a reset. This allows us to move the MQD to VRAM if we want. v2: always reset ring pointer as well (Christian) Reviewed-by:
-
Thong Thai authored
Update the maximum resolution reported for HEVC encoding on VCN 3 devices to reflect its 8K encoding capability. v2: Also update the max height for H.264 encoding to match spec. (Ruijing) Signed-off-by:
Thong Thai <thong.thai@amd.com> Reviewed-by:
Ruijing Dong <ruijing.dong@amd.com> Acked-by:
-
Arnd Bergmann authored
A global function without a header prototype has made it into linux-next during the merge window: drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm.c:6339:6: error: no previous prototype for 'amdgpu_dm_connector_funcs_force' [-Werror=missing-prototypes] Mark the function static instead, as there are no other callers outside this file. Fixes: 0ba4a784 ("drm/amd/display: implement force function in amdgpu_dm_connector_funcs") Reported-by:
Arnd Bergmann <arnd@arndb.de> Signed-off-by:
-
Alex Deucher authored
Always restore the MQD not just when we do a reset. This allows us to move the MQD to VRAM if we want. v2: always reset ring pointer as well (Christian) Reviewed-by:
-
