summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2022-10-05 11:24:12 -0700
committerLinus Torvalds <torvalds@linux-foundation.org>2022-10-05 11:24:12 -0700
commit7e6739b9336e61fe23ca4e2c8d1fda8f19f979bf (patch)
treeda3dc30b7ba682edbb7392028e20dcce194b958b /drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
parenta47e60729d9624e931f988709ab76e043e2ee8b9 (diff)
parent65898687cf7392c372ea8d04a88617e2cb794465 (diff)
downloadlwn-7e6739b9336e61fe23ca4e2c8d1fda8f19f979bf.tar.gz
lwn-7e6739b9336e61fe23ca4e2c8d1fda8f19f979bf.zip
Merge tag 'drm-next-2022-10-05' of git://anongit.freedesktop.org/drm/drm
Pull drm updates from Dave Airlie: "Lots of stuff all over, some new AMD IP support and gang submit support. i915 has further DG2 and Meteorlake pieces, and a bunch of i915 display refactoring. msm has a shrinker rework. There are also a bunch of conversions to use kunit. This has two external pieces, some MEI changes needed for future Intel discrete GPUs. These should be acked by Greg. There is also a cross maintainer shared tree with some backlight rework from Hans in here. Core: - convert selftests to kunit - managed init for more objects - move to idr_init_base - rename fb and gem cma helpers to dma - hide unregistered connectors from getconnector ioctl - DSC passthrough aux support - backlight handling improvements - add dma_resv_assert_held to vmap/vunmap edid: - move luminance calculation to core fbdev: - fix aperture helper usage fourcc: - add more format helpers - add DRM_FORMAT_Cxx, DRM_FORMAT_Rxx, DRM_FORMAT_Dxx - add packed AYUV8888, XYUV8888 - add some kunit tests ttm: - allow bos without backing store - rewrite placement to use intersect/compatible functions dma-buf: - docs update - improve signalling when debugging udmabuf: - fix failure path GPF dp: - drop dp/mst legacy code - atomic mst state support - audio infoframe packing panel: - Samsung LTL101AL01 - B120XAN01.0 - R140NWF5 RH - DMT028VGHMCMI-1A T - AUO B133UAN02.1 - IVO M133NW4J-R3 - Innolux N120ACA-EA1 amdgpu: - Gang submit support - Mode2 reset for RDNA2 - New IP support: DCN 3.1.4, 3.2 SMU 13.x NBIO 7.7 GC 11.x PSP 13.x SDMA 6.x GMC 11.x - DSC passthrough support - PSP fixes for TA support - vangogh GFXOFF stats - clang fixes - gang submit CS cleanup prep work - fix VRAM eviction issues amdkfd: - GC 10.3 IP ISA fixes - fix CRIU regression - CPU fault on COW mapping fixes i915: - align fw versioning with kernel practices - add display substruct to i915 private - add initial runtime info to driver info - split out HDCP and backlight registers - MEI XeHP SDV GSC support - add per-gt sysfs defaults - TLB invalidation improvements - Disable PCI BAR resize on 32-bit - GuC firmware updates and compat changes - GuC log timestamp translation - DG2 preemption workaround changes - DG2 improved HDMI pixel clocks support - PCI BAR sanity checks - Enable DC5 on DG2 - DG2 DMC fw bumped - ADL-S PCI ID added - Meteorlake enablement - Rename ggtt_view to gtt_view - host RPS fixes - release mmaps on rpm suspend on discrete - clocking and dpll refactoring - VBT definitions and parsing updates - SKL watermark code extracted to separate file - allow seamless M/N changes on eDP panels - BUG_ON removal and cleanups msm: - DPU: simplified VBIF configuration cleanup CTL interfaces - DSI: removed unused msm_display_dsc_config struct switch regulator calls to new API switched to PANEL_BRIDGE for direct attached panels - DSI_PHY: convert drivers to parent_hws - DP: cleanup pixel_rate handling - HDMI: turned hdmi-phy-8996 into OF clk provider - misc dt-bindings fixes - choose eDP as primary display if it's available - support getting interconnects from either the mdss or the mdp5/dpu device nodes - gem: Shrinker + LRU re-work: - adds a shared GEM LRU+shrinker helper and moves msm over to that - reduce lock contention between retire and submit by avoiding the need to acquire obj lock in retire path (and instead using resv seeing obj's busyness in the shrinker - fix reclaim vs submit issues - GEM fault injection for triggering userspace error paths - Map/unmap optimization - Improved robustness for a6xx GPU recovery virtio: - improve error and edge conditions handling - convert to use managed helpers - stop exposing LINEAR modifier mgag200: - split modeset handling per model udl: - suspend/disconnect handling improvements vc4: - rework HDMI power up - depend on PM - better unplugging support ast: - resolution handling improvements ingenic: - add JZ4760(B) support - avoid a modeset when sharpness property is unchanged - use the new PM ops it6505: - power seq and clock updates ssd130x: - regmap bulk write - use atomic helpers instead of simple helpers via: - rename via_drv to via_dri1, consolidate all code. radeon: - drop DP MST experimental support - delayed work flush fix - use time_after ti-sn65dsi86: - DP support mediatek: - MT8195 DP support - drop of_gpio header - remove unneeded result - small DP code improvements vkms: - RGB565, XRGB64 and ARGB64 support sun4i: - tv: convert to atomic rcar-du: - Synopsys DW HDMI bridge DT bindings update exynos: - use drm_display_info.is_hdmi - correct return of mixer_mode_valid and hdmi_mode_valid omap: - refcounting fix rockchip: - RK3568 support - RK3399 gamma support" * tag 'drm-next-2022-10-05' of git://anongit.freedesktop.org/drm/drm: (1374 commits) drm/amdkfd: Fix UBSAN shift-out-of-bounds warning drm/amdkfd: Track unified memory when switching xnack mode drm/amdgpu: Enable sram on vcn_4_0_2 drm/amdgpu: Enable VCN DPG for GC11_0_1 drm/msm: Fix build break with recent mm tree drm/panel: simple: Use dev_err_probe() to simplify code drm/panel: panel-edp: Use dev_err_probe() to simplify code drm/panel: simple: Add Multi-Inno Technology MI0800FT-9 dt-bindings: display: simple: Add Multi-Inno Technology MI0800FT-9 panel drm/amdgpu: correct the memcpy size for ip discovery firmware drm/amdgpu: Skip put_reset_domain if it doesn't exist drm/amdgpu: remove switch from amdgpu_gmc_noretry_set drm/amdgpu: Fix mc_umc_status used uninitialized warning drm/amd/display: Prevent OTG shutdown during PSR SU drm/amdgpu: add page retirement handling for CPU RAS drm/amdgpu: use RAS error address convert api in mca notifier drm/amdgpu: support to convert dedicated umc mca address drm/amdgpu: export umc error address convert interface drm/amdgpu: fix sdma v4 init microcode error drm/amd/display: fix array-bounds error in dc_stream_remove_writeback() ...
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c')
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c30
1 files changed, 19 insertions, 11 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index 2170db83e41d..978d3970b5cc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1,3 +1,4 @@
+// SPDX-License-Identifier: MIT
/*
* Copyright 2014-2018 Advanced Micro Devices, Inc.
*
@@ -297,7 +298,7 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo,
*/
replacement = dma_fence_get_stub();
dma_resv_replace_fences(bo->tbo.base.resv, ef->base.context,
- replacement, DMA_RESV_USAGE_READ);
+ replacement, DMA_RESV_USAGE_BOOKKEEP);
dma_fence_put(replacement);
return 0;
}
@@ -1390,8 +1391,9 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,
ret = dma_resv_reserve_fences(vm->root.bo->tbo.base.resv, 1);
if (ret)
goto reserve_shared_fail;
- amdgpu_bo_fence(vm->root.bo,
- &vm->process_info->eviction_fence->base, true);
+ dma_resv_add_fence(vm->root.bo->tbo.base.resv,
+ &vm->process_info->eviction_fence->base,
+ DMA_RESV_USAGE_BOOKKEEP);
amdgpu_bo_unreserve(vm->root.bo);
/* Update process info */
@@ -1612,6 +1614,7 @@ size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev)
uint64_t reserved_for_pt =
ESTIMATE_PT_SIZE(amdgpu_amdkfd_total_mem_size);
size_t available;
+
spin_lock(&kfd_mem_limit.mem_limit_lock);
available = adev->gmc.real_vram_size
- adev->kfd.vram_used_aligned
@@ -1987,9 +1990,9 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
}
if (!amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) && !bo->tbo.pin_count)
- amdgpu_bo_fence(bo,
- &avm->process_info->eviction_fence->base,
- true);
+ dma_resv_add_fence(bo->tbo.base.resv,
+ &avm->process_info->eviction_fence->base,
+ DMA_RESV_USAGE_BOOKKEEP);
ret = unreserve_bo_and_vms(&ctx, false, false);
goto out;
@@ -2216,7 +2219,7 @@ int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,
{
if (atomic_read(&adev->gmc.vm_fault_info_updated) == 1) {
*mem = *adev->gmc.vm_fault_info;
- mb();
+ mb(); /* make sure read happened */
atomic_set(&adev->gmc.vm_fault_info_updated, 0);
}
return 0;
@@ -2758,15 +2761,18 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
if (mem->bo->tbo.pin_count)
continue;
- amdgpu_bo_fence(mem->bo,
- &process_info->eviction_fence->base, true);
+ dma_resv_add_fence(mem->bo->tbo.base.resv,
+ &process_info->eviction_fence->base,
+ DMA_RESV_USAGE_BOOKKEEP);
}
/* Attach eviction fence to PD / PT BOs */
list_for_each_entry(peer_vm, &process_info->vm_list_head,
vm_list_node) {
struct amdgpu_bo *bo = peer_vm->root.bo;
- amdgpu_bo_fence(bo, &process_info->eviction_fence->base, true);
+ dma_resv_add_fence(bo->tbo.base.resv,
+ &process_info->eviction_fence->base,
+ DMA_RESV_USAGE_BOOKKEEP);
}
validate_map_fail:
@@ -2820,7 +2826,9 @@ int amdgpu_amdkfd_add_gws_to_process(void *info, void *gws, struct kgd_mem **mem
ret = dma_resv_reserve_fences(gws_bo->tbo.base.resv, 1);
if (ret)
goto reserve_shared_fail;
- amdgpu_bo_fence(gws_bo, &process_info->eviction_fence->base, true);
+ dma_resv_add_fence(gws_bo->tbo.base.resv,
+ &process_info->eviction_fence->base,
+ DMA_RESV_USAGE_BOOKKEEP);
amdgpu_bo_unreserve(gws_bo);
mutex_unlock(&(*mem)->process_info->lock);