From 240c811ccde49ad336b23c582018c6001690f19d Mon Sep 17 00:00:00 2001 From: "Tianci.Yin" Date: Mon, 20 Jan 2020 15:08:35 +0800 Subject: [PATCH] drm/amdgpu: fix VRAM partially encroached issue in GDDR6 memory training(V2) MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit [why] In GDDR6 BIST training, a certain mount of bottom VRAM will be encroached by UMC, that causes problems(like GTT corrupted and page fault observed). [how] Saving the content of this bottom VRAM to system memory before training, and restoring it after training to avoid VRAM corruption. Acked-by: Christian König Signed-off-by: Tianci.Yin Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h | 2 ++ drivers/gpu/drm/amd/amdgpu/psp_v11_0.c | 36 ++++++++++++++++++++++--- 2 files changed, 35 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h index 3265487b859f..611021514c52 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h @@ -172,6 +172,8 @@ struct psp_dtm_context { #define MEM_TRAIN_SYSTEM_SIGNATURE 0x54534942 #define GDDR6_MEM_TRAINING_DATA_SIZE_IN_BYTES 0x1000 #define GDDR6_MEM_TRAINING_OFFSET 0x8000 +/*Define the VRAM size that will be encroached by BIST training.*/ +#define GDDR6_MEM_TRAINING_ENCROACHED_SIZE 0x2000000 enum psp_memory_training_init_flag { PSP_MEM_TRAIN_NOT_SUPPORT = 0x0, diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c index 685dd9754c67..ac173d2eb809 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c @@ -971,10 +971,13 @@ Err_out: */ static int psp_v11_0_memory_training(struct psp_context *psp, uint32_t ops) { - int ret; - uint32_t p2c_header[4]; struct psp_memory_training_context *ctx = &psp->mem_train_ctx; uint32_t *pcache = (uint32_t*)ctx->sys_cache; + struct amdgpu_device *adev = psp->adev; + uint32_t p2c_header[4]; + uint32_t sz; + void *buf; + int ret; if (ctx->init == PSP_MEM_TRAIN_NOT_SUPPORT) { DRM_DEBUG("Memory training is not supported.\n"); @@ -989,7 +992,7 @@ static int psp_v11_0_memory_training(struct psp_context *psp, uint32_t ops) return 0; } - amdgpu_device_vram_access(psp->adev, ctx->p2c_train_data_offset, p2c_header, sizeof(p2c_header), false); + amdgpu_device_vram_access(adev, ctx->p2c_train_data_offset, p2c_header, sizeof(p2c_header), false); DRM_DEBUG("sys_cache[%08x,%08x,%08x,%08x] p2c_header[%08x,%08x,%08x,%08x]\n", pcache[0], pcache[1], pcache[2], pcache[3], p2c_header[0], p2c_header[1], p2c_header[2], p2c_header[3]); @@ -1026,11 +1029,38 @@ static int psp_v11_0_memory_training(struct psp_context *psp, uint32_t ops) DRM_DEBUG("Memory training ops:%x.\n", ops); if (ops & PSP_MEM_TRAIN_SEND_LONG_MSG) { + /* + * Long traing will encroach certain mount of bottom VRAM, + * saving the content of this bottom VRAM to system memory + * before training, and restoring it after training to avoid + * VRAM corruption. + */ + sz = GDDR6_MEM_TRAINING_ENCROACHED_SIZE; + + if (adev->gmc.visible_vram_size < sz || !adev->mman.aper_base_kaddr) { + DRM_ERROR("visible_vram_size %llx or aper_base_kaddr %p is not initialized.\n", + adev->gmc.visible_vram_size, + adev->mman.aper_base_kaddr); + return -EINVAL; + } + + buf = vmalloc(sz); + if (!buf) { + DRM_ERROR("failed to allocate system memory.\n"); + return -ENOMEM; + } + + memcpy_fromio(buf, adev->mman.aper_base_kaddr, sz); ret = psp_v11_0_memory_training_send_msg(psp, PSP_BL__DRAM_LONG_TRAIN); if (ret) { DRM_ERROR("Send long training msg failed.\n"); + vfree(buf); return ret; } + + memcpy_toio(adev->mman.aper_base_kaddr, buf, sz); + adev->nbio.funcs->hdp_flush(adev, NULL); + vfree(buf); } if (ops & PSP_MEM_TRAIN_SAVE) { -- 2.30.2