From dda72ac390bfd2f3f69c5c3d8e1113abca99d4af Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Mon, 21 Apr 2025 10:21:43 -0700 Subject: [PATCH] drm/msm/adreno: Drop fictional address_space_size Really the only purpose of this was to limit the address space size to 4GB to avoid 32b rollover problems in 64b pointer math in older sqe fw. So replace the address_space_size with a quirk limiting the address space to 4GB. In all other cases, use the SMMU input address size (IAS) to determine the address space size. v2: Properly account for vm_start Signed-off-by: Rob Clark Reviewed-by: Dmitry Baryshkov Patchwork: https://patchwork.freedesktop.org/patch/649467/ --- drivers/gpu/drm/msm/adreno/a6xx_catalog.c | 33 +++++++++++------------ drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 2 +- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 19 ++++++++++--- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 4 ++- 4 files changed, 36 insertions(+), 22 deletions(-) diff --git a/drivers/gpu/drm/msm/adreno/a6xx_catalog.c b/drivers/gpu/drm/msm/adreno/a6xx_catalog.c index 53e2ff4406d8..f85b7e89bafb 100644 --- a/drivers/gpu/drm/msm/adreno/a6xx_catalog.c +++ b/drivers/gpu/drm/msm/adreno/a6xx_catalog.c @@ -681,6 +681,7 @@ static const struct adreno_info a6xx_gpus[] = { [ADRENO_FW_SQE] = "a630_sqe.fw", }, .gmem = (SZ_128K + SZ_4K), + .quirks = ADRENO_QUIRK_4GB_VA, .inactive_period = DRM_MSM_INACTIVE_PERIOD, .init = a6xx_gpu_init, .zapfw = "a610_zap.mdt", @@ -713,6 +714,7 @@ static const struct adreno_info a6xx_gpus[] = { [ADRENO_FW_GMU] = "a630_gmu.bin", }, .gmem = SZ_512K, + .quirks = ADRENO_QUIRK_4GB_VA, .inactive_period = DRM_MSM_INACTIVE_PERIOD, .init = a6xx_gpu_init, .zapfw = "a615_zap.mdt", @@ -743,7 +745,8 @@ static const struct adreno_info a6xx_gpus[] = { }, .gmem = SZ_512K, .inactive_period = DRM_MSM_INACTIVE_PERIOD, - .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT, + .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT | + ADRENO_QUIRK_4GB_VA, .init = a6xx_gpu_init, .zapfw = "a615_zap.mbn", .a6xx = &(const struct a6xx_info) { @@ -769,7 +772,8 @@ static const struct adreno_info a6xx_gpus[] = { }, .gmem = SZ_512K, .inactive_period = DRM_MSM_INACTIVE_PERIOD, - .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT, + .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT | + ADRENO_QUIRK_4GB_VA, .init = a6xx_gpu_init, .a6xx = &(const struct a6xx_info) { .protect = &a630_protect, @@ -791,6 +795,7 @@ static const struct adreno_info a6xx_gpus[] = { [ADRENO_FW_GMU] = "a619_gmu.bin", }, .gmem = SZ_512K, + .quirks = ADRENO_QUIRK_4GB_VA, .inactive_period = DRM_MSM_INACTIVE_PERIOD, .init = a6xx_gpu_init, .zapfw = "a615_zap.mdt", @@ -815,6 +820,7 @@ static const struct adreno_info a6xx_gpus[] = { [ADRENO_FW_GMU] = "a619_gmu.bin", }, .gmem = SZ_512K, + .quirks = ADRENO_QUIRK_4GB_VA, .inactive_period = DRM_MSM_INACTIVE_PERIOD, .init = a6xx_gpu_init, .zapfw = "a615_zap.mdt", @@ -838,8 +844,9 @@ static const struct adreno_info a6xx_gpus[] = { [ADRENO_FW_GMU] = "a619_gmu.bin", }, .gmem = SZ_512K, + .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT | + ADRENO_QUIRK_4GB_VA, .inactive_period = DRM_MSM_INACTIVE_PERIOD, - .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT, .init = a6xx_gpu_init, .zapfw = "a615_zap.mdt", .a6xx = &(const struct a6xx_info) { @@ -874,7 +881,6 @@ static const struct adreno_info a6xx_gpus[] = { .gmu_cgc_mode = 0x00020200, .prim_fifo_threshold = 0x00010000, }, - .address_space_size = SZ_16G, .speedbins = ADRENO_SPEEDBINS( { 0, 0 }, { 137, 1 }, @@ -907,7 +913,6 @@ static const struct adreno_info a6xx_gpus[] = { { /* sentinel */ }, }, }, - .address_space_size = SZ_16G, }, { .chip_ids = ADRENO_CHIP_IDS( 0x06030001, @@ -920,8 +925,9 @@ static const struct adreno_info a6xx_gpus[] = { [ADRENO_FW_GMU] = "a630_gmu.bin", }, .gmem = SZ_1M, + .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT | + ADRENO_QUIRK_4GB_VA, .inactive_period = DRM_MSM_INACTIVE_PERIOD, - .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT, .init = a6xx_gpu_init, .zapfw = "a630_zap.mdt", .a6xx = &(const struct a6xx_info) { @@ -939,8 +945,9 @@ static const struct adreno_info a6xx_gpus[] = { [ADRENO_FW_GMU] = "a640_gmu.bin", }, .gmem = SZ_1M, + .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT | + ADRENO_QUIRK_4GB_VA, .inactive_period = DRM_MSM_INACTIVE_PERIOD, - .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT, .init = a6xx_gpu_init, .zapfw = "a640_zap.mdt", .a6xx = &(const struct a6xx_info) { @@ -973,7 +980,6 @@ static const struct adreno_info a6xx_gpus[] = { .gmu_cgc_mode = 0x00020202, .prim_fifo_threshold = 0x00300200, }, - .address_space_size = SZ_16G, .speedbins = ADRENO_SPEEDBINS( { 0, 0 }, { 1, 1 }, @@ -1000,7 +1006,6 @@ static const struct adreno_info a6xx_gpus[] = { .gmu_cgc_mode = 0x00020000, .prim_fifo_threshold = 0x00300200, }, - .address_space_size = SZ_16G, }, { .chip_ids = ADRENO_CHIP_IDS(0x06060300), .family = ADRENO_6XX_GEN4, @@ -1019,7 +1024,6 @@ static const struct adreno_info a6xx_gpus[] = { .gmu_cgc_mode = 0x00020200, .prim_fifo_threshold = 0x00300200, }, - .address_space_size = SZ_16G, }, { .chip_ids = ADRENO_CHIP_IDS(0x06030500), .family = ADRENO_6XX_GEN4, @@ -1039,7 +1043,6 @@ static const struct adreno_info a6xx_gpus[] = { .gmu_cgc_mode = 0x00020202, .prim_fifo_threshold = 0x00200200, }, - .address_space_size = SZ_16G, .speedbins = ADRENO_SPEEDBINS( { 0, 0 }, { 117, 0 }, @@ -1056,8 +1059,9 @@ static const struct adreno_info a6xx_gpus[] = { [ADRENO_FW_GMU] = "a640_gmu.bin", }, .gmem = SZ_2M, + .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT | + ADRENO_QUIRK_4GB_VA, .inactive_period = DRM_MSM_INACTIVE_PERIOD, - .quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT, .init = a6xx_gpu_init, .zapfw = "a640_zap.mdt", .a6xx = &(const struct a6xx_info) { @@ -1085,7 +1089,6 @@ static const struct adreno_info a6xx_gpus[] = { .gmu_cgc_mode = 0x00020200, .prim_fifo_threshold = 0x00800200, }, - .address_space_size = SZ_16G, } }; DECLARE_ADRENO_GPULIST(a6xx); @@ -1395,7 +1398,6 @@ static const struct adreno_info a7xx_gpus[] = { .pwrup_reglist = &a7xx_pwrup_reglist, .gmu_cgc_mode = 0x00020000, }, - .address_space_size = SZ_16G, .preempt_record_size = 2860 * SZ_1K, }, { .chip_ids = ADRENO_CHIP_IDS(0x43050a01), /* "C510v2" */ @@ -1429,7 +1431,6 @@ static const struct adreno_info a7xx_gpus[] = { { /* sentinel */ }, }, }, - .address_space_size = SZ_16G, .preempt_record_size = 4192 * SZ_1K, }, { .chip_ids = ADRENO_CHIP_IDS(0x43050c01), /* "C512v2" */ @@ -1451,7 +1452,6 @@ static const struct adreno_info a7xx_gpus[] = { .gmu_chipid = 0x7050001, .gmu_cgc_mode = 0x00020202, }, - .address_space_size = SZ_256G, .preempt_record_size = 4192 * SZ_1K, }, { .chip_ids = ADRENO_CHIP_IDS(0x43051401), /* "C520v2" */ @@ -1484,7 +1484,6 @@ static const struct adreno_info a7xx_gpus[] = { { /* sentinel */ }, }, }, - .address_space_size = SZ_16G, .preempt_record_size = 3572 * SZ_1K, } }; diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c index 242d02d48c0c..2289fecbbbf1 100644 --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c @@ -2268,7 +2268,7 @@ a6xx_create_private_address_space(struct msm_gpu *gpu) return ERR_CAST(mmu); return msm_gem_address_space_create(mmu, - "gpu", 0x100000000ULL, + "gpu", ADRENO_VM_START, adreno_private_address_space_size(gpu)); } diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c index 26db1f4b5fb9..2348ffb35f7e 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c @@ -236,14 +236,27 @@ adreno_iommu_create_address_space(struct msm_gpu *gpu, u64 adreno_private_address_space_size(struct msm_gpu *gpu) { struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct adreno_smmu_priv *adreno_smmu = dev_get_drvdata(&gpu->pdev->dev); + const struct io_pgtable_cfg *ttbr1_cfg; if (address_space_size) return address_space_size; - if (adreno_gpu->info->address_space_size) - return adreno_gpu->info->address_space_size; + if (adreno_gpu->info->quirks & ADRENO_QUIRK_4GB_VA) + return SZ_4G; - return SZ_4G; + if (!adreno_smmu || !adreno_smmu->get_ttbr1_cfg) + return SZ_4G; + + ttbr1_cfg = adreno_smmu->get_ttbr1_cfg(adreno_smmu->cookie); + + /* + * Userspace VM is actually using TTBR0, but both are the same size, + * with b48 (sign bit) selecting which TTBRn to use. So if IAS is + * 48, the total (kernel+user) address space size is effectively + * 49 bits. But what userspace is control of is the lower 48. + */ + return BIT(ttbr1_cfg->ias) - ADRENO_VM_START; } #define ARM_SMMU_FSR_TF BIT(1) diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h index 92caba3584da..a8f4bf416e64 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h @@ -57,6 +57,7 @@ enum adreno_family { #define ADRENO_QUIRK_HAS_HW_APRIV BIT(3) #define ADRENO_QUIRK_HAS_CACHED_COHERENT BIT(4) #define ADRENO_QUIRK_PREEMPTION BIT(5) +#define ADRENO_QUIRK_4GB_VA BIT(6) /* Helper for formating the chip_id in the way that userspace tools like * crashdec expect. @@ -104,7 +105,6 @@ struct adreno_info { union { const struct a6xx_info *a6xx; }; - u64 address_space_size; /** * @speedbins: Optional table of fuse to speedbin mappings * @@ -578,6 +578,8 @@ static inline int adreno_is_a7xx(struct adreno_gpu *gpu) adreno_is_a740_family(gpu); } +/* Put vm_start above 32b to catch issues with not setting xyz_BASE_HI */ +#define ADRENO_VM_START 0x100000000ULL u64 adreno_private_address_space_size(struct msm_gpu *gpu); int adreno_get_param(struct msm_gpu *gpu, struct msm_file_private *ctx, uint32_t param, uint64_t *value, uint32_t *len); -- 2.25.1