From 34c4eb7d4e0cd443399a0f114d467d2b3ff05419 Mon Sep 17 00:00:00 2001 From: Karol Przybylski Date: Sun, 15 Dec 2024 13:28:57 +0100 Subject: [PATCH 01/16] drm/amdgpu: Fix potential integer overflow in scheduler mask calculations The use of 1 << i in scheduler mask calculations can result in an unintentional integer overflow due to the expression being evaluated as a 32-bit signed integer. This patch replaces 1 << i with 1ULL << i to ensure the operation is performed as a 64-bit unsigned integer, preventing overflow Discovered in coverity scan, CID 1636393, 1636175, 1636007, 1635853 Fixes: c5c63d9cb5d3 ("drm/amdgpu: add amdgpu_gfx_sched_mask and amdgpu_compute_sched_mask debugfs") Signed-off-by: Karol Przybylski Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index a4dde54512b1..4a4e40dd25d6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -2124,7 +2124,7 @@ static int amdgpu_debugfs_gfx_sched_mask_set(void *data, u64 val) if (!adev) return -ENODEV; - mask = (1 << adev->gfx.num_gfx_rings) - 1; + mask = (1ULL << adev->gfx.num_gfx_rings) - 1; if ((val & mask) == 0) return -EINVAL; @@ -2152,7 +2152,7 @@ static int amdgpu_debugfs_gfx_sched_mask_get(void *data, u64 *val) for (i = 0; i < adev->gfx.num_gfx_rings; ++i) { ring = &adev->gfx.gfx_ring[i]; if (ring->sched.ready) - mask |= 1 << i; + mask |= 1ULL << i; } *val = mask; @@ -2194,7 +2194,7 @@ static int amdgpu_debugfs_compute_sched_mask_set(void *data, u64 val) if (!adev) return -ENODEV; - mask = (1 << adev->gfx.num_compute_rings) - 1; + mask = (1ULL << adev->gfx.num_compute_rings) - 1; if ((val & mask) == 0) return -EINVAL; @@ -2223,7 +2223,7 @@ static int amdgpu_debugfs_compute_sched_mask_get(void *data, u64 *val) for (i = 0; i < adev->gfx.num_compute_rings; ++i) { ring = &adev->gfx.compute_ring[i]; if (ring->sched.ready) - mask |= 1 << i; + mask |= 1ULL << i; } *val = mask; -- 2.51.0 From b4b7271e5ca95b581f2fcc4ae852c4079215e92d Mon Sep 17 00:00:00 2001 From: Philip Yang Date: Tue, 26 Nov 2024 15:45:32 -0500 Subject: [PATCH 02/16] drm/amdgpu: Don't enable sdma 4.4.5 CTXEMPTY interrupt The sdma context empty interrupt is dropped in amdgpu_irq_dispatch as unregistered interrupt src_id 243, this interrupt accounts to 1/3 of total interrupts and causes IH primary ring overflow when running stressful benchmark application. Disable this interrupt has no side effect found. Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c index 4c8308b2878b..56507ae919b0 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c @@ -973,10 +973,12 @@ static int sdma_v4_4_2_inst_start(struct amdgpu_device *adev, /* set utc l1 enable flag always to 1 */ temp = RREG32_SDMA(i, regSDMA_CNTL); temp = REG_SET_FIELD(temp, SDMA_CNTL, UTC_L1_ENABLE, 1); - /* enable context empty interrupt during initialization */ - temp = REG_SET_FIELD(temp, SDMA_CNTL, CTXEMPTY_INT_ENABLE, 1); - WREG32_SDMA(i, regSDMA_CNTL, temp); + if (amdgpu_ip_version(adev, SDMA0_HWIP, 0) < IP_VERSION(4, 4, 5)) { + /* enable context empty interrupt during initialization */ + temp = REG_SET_FIELD(temp, SDMA_CNTL, CTXEMPTY_INT_ENABLE, 1); + WREG32_SDMA(i, regSDMA_CNTL, temp); + } if (!amdgpu_sriov_vf(adev)) { if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { /* unhalt engine */ -- 2.51.0 From d1ebe307b44bbc9a98578c8f8089bb8789c5ecd7 Mon Sep 17 00:00:00 2001 From: Candice Li Date: Mon, 16 Dec 2024 17:20:12 +0800 Subject: [PATCH 03/16] drm/amdgpu: Enable psp v14_0_3 RAS support for non-SRIOV configurations. Enable psp v14_0_3 RAS support for non-SRIOV configurations. Signed-off-by: Candice Li Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index db081618e85c..01c947066a2e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -3572,7 +3572,6 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev) case IP_VERSION(13, 0, 6): case IP_VERSION(13, 0, 12): case IP_VERSION(13, 0, 14): - case IP_VERSION(14, 0, 3): return true; default: return false; @@ -3586,6 +3585,7 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev) case IP_VERSION(13, 0, 10): case IP_VERSION(13, 0, 12): case IP_VERSION(13, 0, 14): + case IP_VERSION(14, 0, 3): return true; default: return false; -- 2.51.0 From 57f812d171af4ba233d3ed7c94dfa5b8e92dcc04 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Christian=20K=C3=B6nig?= Date: Thu, 12 Dec 2024 16:29:18 +0100 Subject: [PATCH 04/16] drm/amdgpu: fix amdgpu_coredump MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit The VM pointer might already be outdated when that function is called. Use the PASID instead to gather the information instead. Signed-off-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c index 946c48829f19..824f9da5b6ce 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c @@ -343,11 +343,10 @@ void amdgpu_coredump(struct amdgpu_device *adev, bool skip_vram_check, coredump->skip_vram_check = skip_vram_check; coredump->reset_vram_lost = vram_lost; - if (job && job->vm) { - struct amdgpu_vm *vm = job->vm; + if (job && job->pasid) { struct amdgpu_task_info *ti; - ti = amdgpu_vm_get_task_info_vm(vm); + ti = amdgpu_vm_get_task_info_pasid(adev, job->pasid); if (ti) { coredump->reset_task_info = *ti; amdgpu_vm_put_task_info(ti); -- 2.51.0 From 26c95e838e6301b0230430ec2fadeabfcb07aeda Mon Sep 17 00:00:00 2001 From: =?utf8?q?Christian=20K=C3=B6nig?= Date: Thu, 12 Dec 2024 16:43:45 +0100 Subject: [PATCH 05/16] drm/amdgpu: set the VM pointer to NULL in amdgpu_job_prepare MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit As soon as the prepare phase is completed the VM might be released, better set it to NULL. Signed-off-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 2e7e1c7dfe7c..04691753cabf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -361,6 +361,13 @@ amdgpu_job_prepare_job(struct drm_sched_job *sched_job, dev_err(ring->adev->dev, "Error getting VM ID (%d)\n", r); goto error; } + /* + * The VM structure might be released after the VMID is + * assigned, we had multiple problems with people trying to use + * the VM pointer so better set it to NULL. + */ + if (!fence) + job->vm = NULL; } return fence; -- 2.51.0 From 11815bb0e30966321ff4351b55ad7b6f2e0a63bf Mon Sep 17 00:00:00 2001 From: =?utf8?q?Christian=20K=C3=B6nig?= Date: Thu, 12 Dec 2024 16:51:04 +0100 Subject: [PATCH 06/16] drm/amdgpu: partially revert "reduce reset time" MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit This partially reverts commit 194eb174cbe4fe2b3376ac30acca2dc8c8beca00. This commit introduced a new state variable into adev without even remotely worrying about CPU barriers. Since we already have the amdgpu_in_reset() function exactly for this use case partially revert that. Signed-off-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 - drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 3 --- drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 2 +- 5 files changed, 4 insertions(+), 8 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 9f5351c013ff..69895fccb474 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1175,7 +1175,6 @@ struct amdgpu_device { struct work_struct reset_work; - bool job_hang; bool dc_enabled; /* Mask of active clusters */ uint32_t aid_mask; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index de30143ea51b..2e5732dfd425 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -836,7 +836,7 @@ int amdgpu_amdkfd_unmap_hiq(struct amdgpu_device *adev, u32 doorbell_off, if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) return -EINVAL; - if (!kiq_ring->sched.ready || adev->job_hang) + if (!kiq_ring->sched.ready || amdgpu_in_reset(adev)) return 0; ring_funcs = kzalloc(sizeof(*ring_funcs), GFP_KERNEL); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index 4a4e40dd25d6..6d5d81f0dc4e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -515,7 +515,7 @@ int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id) if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) return -EINVAL; - if (!kiq_ring->sched.ready || adev->job_hang || amdgpu_in_reset(adev)) + if (!kiq_ring->sched.ready || amdgpu_in_reset(adev)) return 0; spin_lock(&kiq->ring_lock); @@ -567,7 +567,7 @@ int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id) if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) return -EINVAL; - if (!adev->gfx.kiq[0].ring.sched.ready || adev->job_hang) + if (!adev->gfx.kiq[0].ring.sched.ready || amdgpu_in_reset(adev)) return 0; if (amdgpu_gfx_is_master_xcc(adev, xcc_id)) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 04691753cabf..100f04475943 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -102,8 +102,6 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job) return DRM_GPU_SCHED_STAT_ENODEV; } - adev->job_hang = true; - /* * Do the coredump immediately after a job timeout to get a very * close dump/snapshot/representation of GPU's current error status @@ -181,7 +179,6 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job) } exit: - adev->job_hang = false; drm_dev_exit(idx); return DRM_GPU_SCHED_STAT_NOMINAL; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c index 9484f3b5a9b7..003522c2d902 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c @@ -5957,7 +5957,7 @@ static int gfx_v10_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) else WREG32_SOC15(GC, 0, mmCP_ME_CNTL, tmp); - if (adev->job_hang && !enable) + if (amdgpu_in_reset(adev) && !enable) return 0; for (i = 0; i < adev->usec_timeout; i++) { -- 2.51.0 From f607b2b867bbef8a3a76de8d0eccf7429782bdca Mon Sep 17 00:00:00 2001 From: Philip Yang Date: Fri, 22 Nov 2024 17:36:15 -0500 Subject: [PATCH 07/16] drm/amdkfd: KFD interrupt access ih_fifo data in-place To handle 40000 to 80000 interrupts per second running CPX mode with 4 streams/queues per KFD node, KFD interrupt handler becomes the performance bottleneck. Remove the kfifo_out memcpy overhead by accessing ih_fifo data in-place and updating rptr with kfifo_skip_count. Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c | 35 +++++++++------------- 1 file changed, 14 insertions(+), 21 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c index 9b6b6e882593..e7412de9a0ac 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c @@ -114,50 +114,43 @@ void kfd_interrupt_exit(struct kfd_node *node) */ bool enqueue_ih_ring_entry(struct kfd_node *node, const void *ih_ring_entry) { - int count; - - count = kfifo_in(&node->ih_fifo, ih_ring_entry, - node->kfd->device_info.ih_ring_entry_size); - if (count != node->kfd->device_info.ih_ring_entry_size) { + if (kfifo_is_full(&node->ih_fifo)) { dev_dbg_ratelimited(node->adev->dev, - "Interrupt ring overflow, dropping interrupt %d\n", - count); + "Interrupt ring overflow, dropping interrupt\n"); return false; } + kfifo_in(&node->ih_fifo, ih_ring_entry, node->kfd->device_info.ih_ring_entry_size); return true; } /* * Assumption: single reader/writer. This function is not re-entrant */ -static bool dequeue_ih_ring_entry(struct kfd_node *node, void *ih_ring_entry) +static bool dequeue_ih_ring_entry(struct kfd_node *node, u32 **ih_ring_entry) { int count; - count = kfifo_out(&node->ih_fifo, ih_ring_entry, - node->kfd->device_info.ih_ring_entry_size); - - WARN_ON(count && count != node->kfd->device_info.ih_ring_entry_size); + if (kfifo_is_empty(&node->ih_fifo)) + return false; + count = kfifo_out_linear_ptr(&node->ih_fifo, ih_ring_entry, + node->kfd->device_info.ih_ring_entry_size); + WARN_ON(count != node->kfd->device_info.ih_ring_entry_size); return count == node->kfd->device_info.ih_ring_entry_size; } static void interrupt_wq(struct work_struct *work) { - struct kfd_node *dev = container_of(work, struct kfd_node, - interrupt_work); - uint32_t ih_ring_entry[KFD_MAX_RING_ENTRY_SIZE]; + struct kfd_node *dev = container_of(work, struct kfd_node, interrupt_work); + uint32_t *ih_ring_entry; unsigned long start_jiffies = jiffies; - if (dev->kfd->device_info.ih_ring_entry_size > sizeof(ih_ring_entry)) { - dev_err_once(dev->adev->dev, "Ring entry too small\n"); - return; - } - - while (dequeue_ih_ring_entry(dev, ih_ring_entry)) { + while (dequeue_ih_ring_entry(dev, &ih_ring_entry)) { dev->kfd->device_info.event_interrupt_class->interrupt_wq(dev, ih_ring_entry); + kfifo_skip_count(&dev->ih_fifo, dev->kfd->device_info.ih_ring_entry_size); + if (time_is_before_jiffies(start_jiffies + HZ)) { /* If we spent more than a second processing signals, * reschedule the worker to avoid soft-lockup warnings -- 2.51.0 From 1b00143231d3e6f4b76f88f4edd6bb8a1332ef9b Mon Sep 17 00:00:00 2001 From: Philip Yang Date: Tue, 12 Nov 2024 22:07:33 -0500 Subject: [PATCH 08/16] drm/amdgpu: Optimize gfx v9 GPU page fault handling After GPU page fault, there are lots of page fault interrupts generated at short period even with CAM filter enabled because the fault address is different. Each page fault copy to KFD ih fifo to send event to user space by KFD interrupt worker, this could cause KFD ih fifo overflow while other processes generate events at same time. KFD process is aborted after GPU page fault, we only need one GPU page fault interrupt sent to KFD ih fifo to send memory exception event to user space. Incease KFD ih fifo size to 2 times of IH primary ring size, to handle the burst events case. This patch handle the gfx v9 path, cover retry on/off and CAM filter on/off cases. Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 10 ++++ drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 3 + drivers/gpu/drm/amd/amdkfd/kfd_device.c | 67 ++++++++++++++++++++++ drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c | 2 +- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 + 5 files changed, 84 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index 4b80ad860639..8af67f18500a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -433,6 +433,9 @@ void kgd2kfd_unlock_kfd(void); int kgd2kfd_start_sched(struct kfd_dev *kfd, uint32_t node_id); int kgd2kfd_stop_sched(struct kfd_dev *kfd, uint32_t node_id); bool kgd2kfd_compute_active(struct kfd_dev *kfd, uint32_t node_id); +bool kgd2kfd_vmfault_fast_path(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry, + bool retry_fault); + #else static inline int kgd2kfd_init(void) { @@ -518,5 +521,12 @@ static inline bool kgd2kfd_compute_active(struct kfd_dev *kfd, uint32_t node_id) { return false; } + +static inline bool kgd2kfd_vmfault_fast_path(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry, + bool retry_fault) +{ + return false; +} + #endif #endif /* AMDGPU_AMDKFD_H_INCLUDED */ diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c index 48f1b9cf1f88..291549765c38 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c @@ -623,6 +623,9 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev, } } + if (kgd2kfd_vmfault_fast_path(adev, entry, retry_fault)) + return 1; + if (!printk_ratelimit()) return 0; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index ac0fdaa1ea23..9b49563f2c42 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -1521,6 +1521,73 @@ bool kgd2kfd_compute_active(struct kfd_dev *kfd, uint32_t node_id) return kfd_compute_active(node); } +/** + * kgd2kfd_vmfault_fast_path() - KFD vm page fault interrupt handling fast path for gmc v9 + * @adev: amdgpu device + * @entry: vm fault interrupt vector + * @retry_fault: if this is retry fault + * + * retry fault - + * with CAM enabled, adev primary ring + * | gmc_v9_0_process_interrupt() + * adev soft_ring + * | gmc_v9_0_process_interrupt() worker failed to recover page fault + * KFD node ih_fifo + * | KFD interrupt_wq worker + * kfd_signal_vm_fault_event + * + * without CAM, adev primary ring1 + * | gmc_v9_0_process_interrupt worker failed to recvoer page fault + * KFD node ih_fifo + * | KFD interrupt_wq worker + * kfd_signal_vm_fault_event + * + * no-retry fault - + * adev primary ring + * | gmc_v9_0_process_interrupt() + * KFD node ih_fifo + * | KFD interrupt_wq worker + * kfd_signal_vm_fault_event + * + * fast path - After kfd_signal_vm_fault_event, gmc_v9_0_process_interrupt drop the page fault + * of same process, don't copy interrupt to KFD node ih_fifo. + * With gdb debugger enabled, need convert the retry fault to no-retry fault for + * debugger, cannot use the fast path. + * + * Return: + * true - use the fast path to handle this fault + * false - use normal path to handle it + */ +bool kgd2kfd_vmfault_fast_path(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry, + bool retry_fault) +{ + struct kfd_process *p; + u32 cam_index; + + if (entry->ih == &adev->irq.ih_soft || entry->ih == &adev->irq.ih1) { + p = kfd_lookup_process_by_pasid(entry->pasid); + if (!p) + return true; + + if (p->gpu_page_fault && !p->debug_trap_enabled) { + if (retry_fault && adev->irq.retry_cam_enabled) { + cam_index = entry->src_data[2] & 0x3ff; + WDOORBELL32(adev->irq.retry_cam_doorbell_index, cam_index); + } + + kfd_unref_process(p); + return true; + } + + /* + * This is the first page fault, set flag and then signal user space + */ + p->gpu_page_fault = true; + kfd_unref_process(p); + } + return false; +} + #if defined(CONFIG_DEBUG_FS) /* This function will send a package to HIQ to hang the HWS diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c index e7412de9a0ac..8e00800f3207 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c @@ -46,7 +46,7 @@ #include #include "kfd_priv.h" -#define KFD_IH_NUM_ENTRIES 8192 +#define KFD_IH_NUM_ENTRIES 16384 static void interrupt_wq(struct work_struct *); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index 800e4ae5b573..e529fdc1b422 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -1003,6 +1003,9 @@ struct kfd_process { struct semaphore runtime_enable_sema; bool is_runtime_retry; struct kfd_runtime_info runtime_info; + + /* if gpu page fault sent to KFD */ + bool gpu_page_fault; }; #define KFD_PROCESS_TABLE_SIZE 5 /* bits: 32 entries */ -- 2.51.0 From 34db5a32617d102e8042151bb87590e43c97132e Mon Sep 17 00:00:00 2001 From: Philip Yang Date: Tue, 26 Nov 2024 11:33:15 -0500 Subject: [PATCH 09/16] drm/amdkfd: Queue interrupt work to different CPU For CPX mode, each KFD node has interrupt worker to process ih_fifo to send events to user space. Currently all interrupt workers of same adev queue to same CPU, all workers execution are actually serialized and this cause KFD ih_fifo overflow when CPU usage is high. Use per-GPU unbounded highpri queue with number of workers equals to number of partitions, let queue_work select the next CPU round robin among the local CPUs of same NUMA. Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdkfd/kfd_device.c | 25 ++++++++-------------- drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c | 25 ++++++++-------------- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 ++- 3 files changed, 20 insertions(+), 33 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index 9b49563f2c42..a29374c86405 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -649,6 +649,14 @@ static void kfd_cleanup_nodes(struct kfd_dev *kfd, unsigned int num_nodes) struct kfd_node *knode; unsigned int i; + /* + * flush_work ensures that there are no outstanding + * work-queue items that will access interrupt_ring. New work items + * can't be created because we stopped interrupt handling above. + */ + flush_workqueue(kfd->ih_wq); + destroy_workqueue(kfd->ih_wq); + for (i = 0; i < num_nodes; i++) { knode = kfd->nodes[i]; device_queue_manager_uninit(knode->dqm); @@ -1066,21 +1074,6 @@ static int kfd_resume(struct kfd_node *node) return err; } -static inline void kfd_queue_work(struct workqueue_struct *wq, - struct work_struct *work) -{ - int cpu, new_cpu; - - cpu = new_cpu = smp_processor_id(); - do { - new_cpu = cpumask_next(new_cpu, cpu_online_mask) % nr_cpu_ids; - if (cpu_to_node(new_cpu) == numa_node_id()) - break; - } while (cpu != new_cpu); - - queue_work_on(new_cpu, wq, work); -} - /* This is called directly from KGD at ISR. */ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry) { @@ -1106,7 +1099,7 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry) patched_ihre, &is_patched) && enqueue_ih_ring_entry(node, is_patched ? patched_ihre : ih_ring_entry)) { - kfd_queue_work(node->ih_wq, &node->interrupt_work); + queue_work(node->kfd->ih_wq, &node->interrupt_work); spin_unlock_irqrestore(&node->interrupt_lock, flags); return; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c index 8e00800f3207..6beb786c582a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c @@ -62,11 +62,14 @@ int kfd_interrupt_init(struct kfd_node *node) return r; } - node->ih_wq = alloc_workqueue("KFD IH", WQ_HIGHPRI, 1); - if (unlikely(!node->ih_wq)) { - kfifo_free(&node->ih_fifo); - dev_err(node->adev->dev, "Failed to allocate KFD IH workqueue\n"); - return -ENOMEM; + if (!node->kfd->ih_wq) { + node->kfd->ih_wq = alloc_workqueue("KFD IH", WQ_HIGHPRI | WQ_UNBOUND, + node->kfd->num_nodes); + if (unlikely(!node->kfd->ih_wq)) { + kfifo_free(&node->ih_fifo); + dev_err(node->adev->dev, "Failed to allocate KFD IH workqueue\n"); + return -ENOMEM; + } } spin_lock_init(&node->interrupt_lock); @@ -96,16 +99,6 @@ void kfd_interrupt_exit(struct kfd_node *node) spin_lock_irqsave(&node->interrupt_lock, flags); node->interrupts_active = false; spin_unlock_irqrestore(&node->interrupt_lock, flags); - - /* - * flush_work ensures that there are no outstanding - * work-queue items that will access interrupt_ring. New work items - * can't be created because we stopped interrupt handling above. - */ - flush_workqueue(node->ih_wq); - - destroy_workqueue(node->ih_wq); - kfifo_free(&node->ih_fifo); } @@ -155,7 +148,7 @@ static void interrupt_wq(struct work_struct *work) /* If we spent more than a second processing signals, * reschedule the worker to avoid soft-lockup warnings */ - queue_work(dev->ih_wq, &dev->interrupt_work); + queue_work(dev->kfd->ih_wq, &dev->interrupt_work); break; } } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index e529fdc1b422..c32b255c0eb2 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -274,7 +274,6 @@ struct kfd_node { /* Interrupts */ struct kfifo ih_fifo; - struct workqueue_struct *ih_wq; struct work_struct interrupt_work; spinlock_t interrupt_lock; @@ -367,6 +366,8 @@ struct kfd_dev { struct kfd_node *nodes[MAX_KFD_NODES]; unsigned int num_nodes; + struct workqueue_struct *ih_wq; + /* Kernel doorbells for KFD device */ struct amdgpu_bo *doorbells; -- 2.51.0 From de844846f72b152119faaef1b363448dc8ea368f Mon Sep 17 00:00:00 2001 From: Philip Yang Date: Wed, 4 Dec 2024 17:49:08 -0500 Subject: [PATCH 10/16] drm/amdkfd: Improve signal event slow path If event slot is not signaled, kfd_signal_event_interrupt goes to slow path to scan all event slots to find the signaled event, this is needed for old ASICs that don't have the event ID or the event IDs are incorrect in the IH payload. There is case that GPU signal the same event twice, then driver process the first event interrupt, set_event and event slot is auto-reset, then for the second event interrupt, KFD goes to slow path as event is not signaled, just drop the second event interrupt because the application only need wakeup once. Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdkfd/kfd_events.c | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c index ea3792249209..d075f24e5f9f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c @@ -748,6 +748,16 @@ void kfd_signal_event_interrupt(u32 pasid, uint32_t partial_id, uint64_t *slots = page_slots(p->signal_page); uint32_t id; + /* + * If id is valid but slot is not signaled, GPU may signal the same event twice + * before driver have chance to process the first interrupt, then signal slot is + * auto-reset after set_event wakeup the user space, just drop the second event as + * the application only need wakeup once. + */ + if ((valid_id_bits > 31 || (1U << valid_id_bits) >= KFD_SIGNAL_EVENT_LIMIT) && + partial_id < KFD_SIGNAL_EVENT_LIMIT && slots[partial_id] == UNSIGNALED_EVENT_SLOT) + goto out_unlock; + if (valid_id_bits) pr_debug_ratelimited("Partial ID invalid: %u (%u valid bits)\n", partial_id, valid_id_bits); @@ -776,6 +786,7 @@ void kfd_signal_event_interrupt(u32 pasid, uint32_t partial_id, } } +out_unlock: rcu_read_unlock(); kfd_unref_process(p); } -- 2.51.0 From e37ccf44ace3f58fc9d84de1acd439077b9f7fef Mon Sep 17 00:00:00 2001 From: Philip Yang Date: Tue, 3 Dec 2024 10:00:25 -0500 Subject: [PATCH 11/16] drm/amdgpu: Show warning message if IH ring overflow MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit If IH primary ring and KFD ih fifo overflows, we may miss CP, SDMA interrupts and cause application soft hang. Show warning message with ring name if overflow happens. Add function to get ih ring name to avoid duplicating it. To keep warning message consistent between GPU generations, change all *_ih.c except ASICs older than Vega which has only one ih ring. Signed-off-by: Philip Yang Reviewed-by: Christian König Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c | 6 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h | 1 + drivers/gpu/drm/amd/amdgpu/navi10_ih.c | 5 ++--- drivers/gpu/drm/amd/amdgpu/vega10_ih.c | 5 ++--- drivers/gpu/drm/amd/amdgpu/vega20_ih.c | 5 ++--- drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c | 4 ++-- 6 files changed, 15 insertions(+), 11 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c index f3b0aaf3ebc6..901f8b12c672 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c @@ -298,3 +298,9 @@ uint64_t amdgpu_ih_decode_iv_ts_helper(struct amdgpu_ih_ring *ih, u32 rptr, dw2 = le32_to_cpu(ih->ring[ring_index + 2]); return dw1 | ((u64)(dw2 & 0xffff) << 32); } + +const char *amdgpu_ih_ring_name(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) +{ + return ih == &adev->irq.ih ? "ih" : ih == &adev->irq.ih_soft ? "sw ih" : + ih == &adev->irq.ih1 ? "ih1" : ih == &adev->irq.ih2 ? "ih2" : "unknown"; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h index 508f02eb0cf8..7d4395a5d8ac 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h @@ -110,4 +110,5 @@ void amdgpu_ih_decode_iv_helper(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry); uint64_t amdgpu_ih_decode_iv_ts_helper(struct amdgpu_ih_ring *ih, u32 rptr, signed int offset); +const char *amdgpu_ih_ring_name(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/navi10_ih.c b/drivers/gpu/drm/amd/amdgpu/navi10_ih.c index ebc2ab9c3c5c..62cdfe10e6f4 100644 --- a/drivers/gpu/drm/amd/amdgpu/navi10_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/navi10_ih.c @@ -434,9 +434,8 @@ static u32 navi10_ih_get_wptr(struct amdgpu_device *adev, * this should allow us to catch up. */ tmp = (wptr + 32) & ih->ptr_mask; - dev_warn(adev->dev, "IH ring buffer overflow " - "(0x%08X, 0x%08X, 0x%08X)\n", - wptr, ih->rptr, tmp); + dev_warn(adev->dev, "%s ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", + amdgpu_ih_ring_name(adev, ih), wptr, ih->rptr, tmp); ih->rptr = tmp; tmp = RREG32_NO_KIQ(ih_regs->ih_rb_cntl); diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c index 378da889e075..98fc6941159e 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c @@ -364,9 +364,8 @@ static u32 vega10_ih_get_wptr(struct amdgpu_device *adev, * this should allow us to catchup. */ tmp = (wptr + 32) & ih->ptr_mask; - dev_warn(adev->dev, "IH ring buffer overflow " - "(0x%08X, 0x%08X, 0x%08X)\n", - wptr, ih->rptr, tmp); + dev_warn_ratelimited(adev->dev, "%s ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", + amdgpu_ih_ring_name(adev, ih), wptr, ih->rptr, tmp); ih->rptr = tmp; tmp = RREG32_NO_KIQ(ih_regs->ih_rb_cntl); diff --git a/drivers/gpu/drm/amd/amdgpu/vega20_ih.c b/drivers/gpu/drm/amd/amdgpu/vega20_ih.c index 2c1c4b788b6d..e9e3b2ed4b7b 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega20_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/vega20_ih.c @@ -444,9 +444,8 @@ static u32 vega20_ih_get_wptr(struct amdgpu_device *adev, * this should allow us to catchup. */ tmp = (wptr + 32) & ih->ptr_mask; - dev_warn(adev->dev, "IH ring buffer overflow " - "(0x%08X, 0x%08X, 0x%08X)\n", - wptr, ih->rptr, tmp); + dev_warn_ratelimited(adev->dev, "%s ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", + amdgpu_ih_ring_name(adev, ih), wptr, ih->rptr, tmp); ih->rptr = tmp; tmp = RREG32_NO_KIQ(ih_regs->ih_rb_cntl); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c index 6beb786c582a..783c2f5a04e4 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c @@ -108,8 +108,8 @@ void kfd_interrupt_exit(struct kfd_node *node) bool enqueue_ih_ring_entry(struct kfd_node *node, const void *ih_ring_entry) { if (kfifo_is_full(&node->ih_fifo)) { - dev_dbg_ratelimited(node->adev->dev, - "Interrupt ring overflow, dropping interrupt\n"); + dev_warn_ratelimited(node->adev->dev, "KFD node %d ih_fifo overflow\n", + node->node_id); return false; } -- 2.51.0 From b64f2f3e870d324703246757cb67cec09a64a1c9 Mon Sep 17 00:00:00 2001 From: Srinivasan Shanmugam Date: Thu, 12 Dec 2024 15:03:29 +0530 Subject: [PATCH 12/16] drm/amd/display: Fix NULL pointer dereference in dmub_tracebuffer_show It corrects the issue by checking if 'adev->dm.dmub_srv' is NULL before accessing its 'meta_info' member. This ensures that we do not dereference a NULL pointer. Fixes the below: drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm_debugfs.c:917 dmub_tracebuffer_show() warn: address of 'adev->dm.dmub_srv->meta_info' is non-NULL drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm_debugfs.c 901 static int dmub_tracebuffer_show(struct seq_file *m, void *data) 902 { 903 struct amdgpu_device *adev = m->private; 904 struct dmub_srv_fb_info *fb_info = adev->dm.dmub_fb_info; 905 struct dmub_fw_meta_info *fw_meta_info = &adev->dm.dmub_srv->meta_info; ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Even if adev->dm.dmub_srv is NULL, the address of ->meta_info can't be NULL 906 struct dmub_debugfs_trace_entry *entries; 907 uint8_t *tbuf_base; 908 uint32_t tbuf_size, max_entries, num_entries, first_entry, i; 909 910 if (!fb_info) 911 return 0; 912 913 tbuf_base = (uint8_t *)fb_info->fb[DMUB_WINDOW_5_TRACEBUFF].cpu_addr; 914 if (!tbuf_base) 915 return 0; 916 --> 917 tbuf_size = fw_meta_info ? fw_meta_info->trace_buffer_size : ^^^^^^^^^^^^ Always non-NULL 918 DMUB_TRACE_BUFFER_SIZE; 919 max_entries = (tbuf_size - sizeof(struct dmub_debugfs_trace_header)) / 920 sizeof(struct dmub_debugfs_trace_entry); 921 922 num_entries = v2: Initialize struct dmub_fw_meta_info *fw_meta_info to NULL (Dan Carpenter) Fixes: 5a498172c8d0 ("drm/amd/display: Make DMCUB tracebuffer debugfs chronological") Cc: Leo Li Cc: Tom Chung Cc: Rodrigo Siqueira Cc: Roman Li Cc: Alex Hung Cc: Aurabindo Pillai Cc: Harry Wentland Cc: Hamza Mahfooz Reported-by: Dan Carpenter Signed-off-by: Srinivasan Shanmugam Reviewed-by: Roman Li Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c index 2a3c4ac065de..bc3e7a82b402 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c @@ -902,7 +902,7 @@ static int dmub_tracebuffer_show(struct seq_file *m, void *data) { struct amdgpu_device *adev = m->private; struct dmub_srv_fb_info *fb_info = adev->dm.dmub_fb_info; - struct dmub_fw_meta_info *fw_meta_info = &adev->dm.dmub_srv->meta_info; + struct dmub_fw_meta_info *fw_meta_info = NULL; struct dmub_debugfs_trace_entry *entries; uint8_t *tbuf_base; uint32_t tbuf_size, max_entries, num_entries, first_entry, i; @@ -914,6 +914,9 @@ static int dmub_tracebuffer_show(struct seq_file *m, void *data) if (!tbuf_base) return 0; + if (adev->dm.dmub_srv) + fw_meta_info = &adev->dm.dmub_srv->meta_info; + tbuf_size = fw_meta_info ? fw_meta_info->trace_buffer_size : DMUB_TRACE_BUFFER_SIZE; max_entries = (tbuf_size - sizeof(struct dmub_debugfs_trace_header)) / -- 2.51.0 From a21ab06b8c2d8d25c4a83bdf39542834b1f3beae Mon Sep 17 00:00:00 2001 From: Mirsad Todorovac Date: Tue, 17 Dec 2024 23:58:10 +0100 Subject: [PATCH 13/16] drm/admgpu: replace kmalloc() and memcpy() with kmemdup() MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit The static analyser tool gave the following advice: ./drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c:1266:7-14: WARNING opportunity for kmemdup → 1266 tmp = kmalloc(used_size, GFP_KERNEL); 1267 if (!tmp) 1268 return -ENOMEM; 1269 → 1270 memcpy(tmp, &host_telemetry->body.error_count, used_size); Replacing kmalloc() + memcpy() with kmemdump() doesn't change semantics. Original code works without fault, so this is not a bug fix but proposed improvement. Link: https://lwn.net/Articles/198928/ Fixes: 84a2947ecc85 ("drm/amdgpu: Implement virt req_ras_err_count") Cc: Alex Deucher Cc: "Christian König" Cc: Xinhui Pan Cc: David Airlie Cc: Simona Vetter Cc: Zhigang Luo Cc: Victor Skvortsov Cc: Hawking Zhang Cc: Lijo Lazar Cc: Yunxiang Li Cc: Jack Xiao Cc: Vignesh Chander Cc: Danijel Slivka Cc: amd-gfx@lists.freedesktop.org Cc: dri-devel@lists.freedesktop.org Cc: linux-kernel@vger.kernel.org Reviewed-by: Lijo Lazar Signed-off-by: Mirsad Todorovac Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c index c704e9803e11..0af469ec6fcc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c @@ -1263,12 +1263,10 @@ static int amdgpu_virt_cache_host_error_counts(struct amdgpu_device *adev, if (used_size > (AMD_SRIOV_RAS_TELEMETRY_SIZE_KB << 10)) return 0; - tmp = kmalloc(used_size, GFP_KERNEL); + tmp = kmemdup(&host_telemetry->body.error_count, used_size, GFP_KERNEL); if (!tmp) return -ENOMEM; - memcpy(tmp, &host_telemetry->body.error_count, used_size); - if (checksum != amd_sriov_msg_checksum(tmp, used_size, 0, 0)) goto out; -- 2.51.0 From 695c2c745e5dff201b75da8a1d237ce403600d04 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Michel=20D=C3=A4nzer?= Date: Tue, 17 Dec 2024 18:22:56 +0100 Subject: [PATCH 14/16] drm/amdgpu: Handle NULL bo->tbo.resource (again) in amdgpu_vm_bo_update MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Third time's the charm, I hope? Fixes: d3116756a710 ("drm/ttm: rename bo->mem and make it a pointer") Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3837 Reviewed-by: Christian König Signed-off-by: Michel Dänzer Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index ddd7f05e4db9..c9c48b782ec1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -1266,10 +1266,9 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, * next command submission. */ if (amdgpu_vm_is_bo_always_valid(vm, bo)) { - uint32_t mem_type = bo->tbo.resource->mem_type; - - if (!(bo->preferred_domains & - amdgpu_mem_type_to_domain(mem_type))) + if (bo->tbo.resource && + !(bo->preferred_domains & + amdgpu_mem_type_to_domain(bo->tbo.resource->mem_type))) amdgpu_vm_bo_evicted(&bo_va->base); else amdgpu_vm_bo_idle(&bo_va->base); -- 2.51.0 From 3782305ce5807c18fbf092124b9e8303cf1723ae Mon Sep 17 00:00:00 2001 From: Alex Hung Date: Tue, 17 Dec 2024 14:03:50 -0700 Subject: [PATCH 15/16] drm/amd/display: Remove unnecessary amdgpu_irq_get/put [WHY & HOW] commit 7fb363c57522 ("drm/amd/display: Let drm_crtc_vblank_on/off manage interrupts") lets drm_crtc_vblank_* to manage interrupts in amdgpu_dm_crtc_set_vblank, and amdgpu_irq_get/put do not need to be called here. Part of that patch got lost somehow, so fix it up. Reviewed-by: Alex Deucher Reviewed-by: Leo Li Signed-off-by: Alex Hung Signed-off-by: Alex Deucher --- .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 31 ------------------- 1 file changed, 31 deletions(-) diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c index 4f0267b2d5ad..e9a4750a83d6 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -8398,16 +8398,6 @@ static void manage_dm_interrupts(struct amdgpu_device *adev, struct amdgpu_crtc *acrtc, struct dm_crtc_state *acrtc_state) { - /* - * We have no guarantee that the frontend index maps to the same - * backend index - some even map to more than one. - * - * TODO: Use a different interrupt or check DC itself for the mapping. - */ - int irq_type = - amdgpu_display_crtc_idx_to_irq_type( - adev, - acrtc->crtc_id); struct drm_vblank_crtc_config config = {0}; struct dc_crtc_timing *timing; int offdelay; @@ -8433,28 +8423,7 @@ static void manage_dm_interrupts(struct amdgpu_device *adev, drm_crtc_vblank_on_config(&acrtc->base, &config); - - amdgpu_irq_get( - adev, - &adev->pageflip_irq, - irq_type); -#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY) - amdgpu_irq_get( - adev, - &adev->vline0_irq, - irq_type); -#endif } else { -#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY) - amdgpu_irq_put( - adev, - &adev->vline0_irq, - irq_type); -#endif - amdgpu_irq_put( - adev, - &adev->pageflip_irq, - irq_type); drm_crtc_vblank_off(&acrtc->base); } } -- 2.51.0 From 1c86c81a86c60f9b15d3e3f43af0363cf56063e7 Mon Sep 17 00:00:00 2001 From: Melissa Wen Date: Tue, 17 Dec 2024 17:45:03 -0300 Subject: [PATCH 16/16] drm/amd/display: fix page fault due to max surface definition mismatch DC driver is using two different values to define the maximum number of surfaces: MAX_SURFACES and MAX_SURFACE_NUM. Consolidate MAX_SURFACES as the unique definition for surface updates across DC. It fixes page fault faced by Cosmic users on AMD display versions that support two overlay planes, since the introduction of cursor overlay mode. [Nov26 21:33] BUG: unable to handle page fault for address: 0000000051d0f08b [ +0.000015] #PF: supervisor read access in kernel mode [ +0.000006] #PF: error_code(0x0000) - not-present page [ +0.000005] PGD 0 P4D 0 [ +0.000007] Oops: Oops: 0000 [#1] PREEMPT SMP NOPTI [ +0.000006] CPU: 4 PID: 71 Comm: kworker/u32:6 Not tainted 6.10.0+ #300 [ +0.000006] Hardware name: Valve Jupiter/Jupiter, BIOS F7A0131 01/30/2024 [ +0.000007] Workqueue: events_unbound commit_work [drm_kms_helper] [ +0.000040] RIP: 0010:copy_stream_update_to_stream.isra.0+0x30d/0x750 [amdgpu] [ +0.000847] Code: 8b 10 49 89 94 24 f8 00 00 00 48 8b 50 08 49 89 94 24 00 01 00 00 8b 40 10 41 89 84 24 08 01 00 00 49 8b 45 78 48 85 c0 74 0b <0f> b6 00 41 88 84 24 90 64 00 00 49 8b 45 60 48 85 c0 74 3b 48 8b [ +0.000010] RSP: 0018:ffffc203802f79a0 EFLAGS: 00010206 [ +0.000009] RAX: 0000000051d0f08b RBX: 0000000000000004 RCX: ffff9f964f0a8070 [ +0.000004] RDX: ffff9f9710f90e40 RSI: ffff9f96600c8000 RDI: ffff9f964f000000 [ +0.000004] RBP: ffffc203802f79f8 R08: 0000000000000000 R09: 0000000000000000 [ +0.000005] R10: 0000000000000000 R11: 0000000000000000 R12: ffff9f96600c8000 [ +0.000004] R13: ffff9f9710f90e40 R14: ffff9f964f000000 R15: ffff9f96600c8000 [ +0.000004] FS: 0000000000000000(0000) GS:ffff9f9970000000(0000) knlGS:0000000000000000 [ +0.000005] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ +0.000005] CR2: 0000000051d0f08b CR3: 00000002e6a20000 CR4: 0000000000350ef0 [ +0.000005] Call Trace: [ +0.000011] [ +0.000010] ? __die_body.cold+0x19/0x27 [ +0.000012] ? page_fault_oops+0x15a/0x2d0 [ +0.000014] ? exc_page_fault+0x7e/0x180 [ +0.000009] ? asm_exc_page_fault+0x26/0x30 [ +0.000013] ? copy_stream_update_to_stream.isra.0+0x30d/0x750 [amdgpu] [ +0.000739] ? dc_commit_state_no_check+0xd6c/0xe70 [amdgpu] [ +0.000470] update_planes_and_stream_state+0x49b/0x4f0 [amdgpu] [ +0.000450] ? srso_return_thunk+0x5/0x5f [ +0.000009] ? commit_minimal_transition_state+0x239/0x3d0 [amdgpu] [ +0.000446] update_planes_and_stream_v2+0x24a/0x590 [amdgpu] [ +0.000464] ? srso_return_thunk+0x5/0x5f [ +0.000009] ? sort+0x31/0x50 [ +0.000007] ? amdgpu_dm_atomic_commit_tail+0x159f/0x3a30 [amdgpu] [ +0.000508] ? srso_return_thunk+0x5/0x5f [ +0.000009] ? amdgpu_crtc_get_scanout_position+0x28/0x40 [amdgpu] [ +0.000377] ? srso_return_thunk+0x5/0x5f [ +0.000009] ? drm_crtc_vblank_helper_get_vblank_timestamp_internal+0x160/0x390 [drm] [ +0.000058] ? srso_return_thunk+0x5/0x5f [ +0.000005] ? dma_fence_default_wait+0x8c/0x260 [ +0.000010] ? srso_return_thunk+0x5/0x5f [ +0.000005] ? wait_for_completion_timeout+0x13b/0x170 [ +0.000006] ? srso_return_thunk+0x5/0x5f [ +0.000005] ? dma_fence_wait_timeout+0x108/0x140 [ +0.000010] ? commit_tail+0x94/0x130 [drm_kms_helper] [ +0.000024] ? process_one_work+0x177/0x330 [ +0.000008] ? worker_thread+0x266/0x3a0 [ +0.000006] ? __pfx_worker_thread+0x10/0x10 [ +0.000004] ? kthread+0xd2/0x100 [ +0.000006] ? __pfx_kthread+0x10/0x10 [ +0.000006] ? ret_from_fork+0x34/0x50 [ +0.000004] ? __pfx_kthread+0x10/0x10 [ +0.000005] ? ret_from_fork_asm+0x1a/0x30 [ +0.000011] Fixes: 1b04dcca4fb1 ("drm/amd/display: Introduce overlay cursor mode") Suggested-by: Leo Li Link: https://gitlab.freedesktop.org/drm/amd/-/issues/3693 Signed-off-by: Melissa Wen Reviewed-by: Rodrigo Siqueira Signed-off-by: Rodrigo Siqueira Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/display/dc/core/dc.c | 2 +- drivers/gpu/drm/amd/display/dc/core/dc_state.c | 8 ++++---- drivers/gpu/drm/amd/display/dc/dc.h | 2 +- drivers/gpu/drm/amd/display/dc/dc_stream.h | 2 +- drivers/gpu/drm/amd/display/dc/dc_types.h | 1 - drivers/gpu/drm/amd/display/dc/dml2/dml2_mall_phantom.c | 2 +- 6 files changed, 8 insertions(+), 9 deletions(-) diff --git a/drivers/gpu/drm/amd/display/dc/core/dc.c b/drivers/gpu/drm/amd/display/dc/core/dc.c index dfa36368ae63..91a4c52289ac 100644 --- a/drivers/gpu/drm/amd/display/dc/core/dc.c +++ b/drivers/gpu/drm/amd/display/dc/core/dc.c @@ -4519,7 +4519,7 @@ static bool commit_minimal_transition_based_on_current_context(struct dc *dc, struct pipe_split_policy_backup policy; struct dc_state *intermediate_context; struct dc_state *old_current_state = dc->current_state; - struct dc_surface_update srf_updates[MAX_SURFACE_NUM] = {0}; + struct dc_surface_update srf_updates[MAX_SURFACES] = {0}; int surface_count; /* diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_state.c b/drivers/gpu/drm/amd/display/dc/core/dc_state.c index e006f816ff2f..1b2cce127981 100644 --- a/drivers/gpu/drm/amd/display/dc/core/dc_state.c +++ b/drivers/gpu/drm/amd/display/dc/core/dc_state.c @@ -483,9 +483,9 @@ bool dc_state_add_plane( if (stream_status == NULL) { dm_error("Existing stream not found; failed to attach surface!\n"); goto out; - } else if (stream_status->plane_count == MAX_SURFACE_NUM) { + } else if (stream_status->plane_count == MAX_SURFACES) { dm_error("Surface: can not attach plane_state %p! Maximum is: %d\n", - plane_state, MAX_SURFACE_NUM); + plane_state, MAX_SURFACES); goto out; } else if (!otg_master_pipe) { goto out; @@ -600,7 +600,7 @@ bool dc_state_rem_all_planes_for_stream( { int i, old_plane_count; struct dc_stream_status *stream_status = NULL; - struct dc_plane_state *del_planes[MAX_SURFACE_NUM] = { 0 }; + struct dc_plane_state *del_planes[MAX_SURFACES] = { 0 }; for (i = 0; i < state->stream_count; i++) if (state->streams[i] == stream) { @@ -875,7 +875,7 @@ bool dc_state_rem_all_phantom_planes_for_stream( { int i, old_plane_count; struct dc_stream_status *stream_status = NULL; - struct dc_plane_state *del_planes[MAX_SURFACE_NUM] = { 0 }; + struct dc_plane_state *del_planes[MAX_SURFACES] = { 0 }; for (i = 0; i < state->stream_count; i++) if (state->streams[i] == phantom_stream) { diff --git a/drivers/gpu/drm/amd/display/dc/dc.h b/drivers/gpu/drm/amd/display/dc/dc.h index 8c6347413038..00c040d60238 100644 --- a/drivers/gpu/drm/amd/display/dc/dc.h +++ b/drivers/gpu/drm/amd/display/dc/dc.h @@ -1403,7 +1403,7 @@ struct dc_scratch_space { * store current value in plane states so we can still recover * a valid current state during dc update. */ - struct dc_plane_state plane_states[MAX_SURFACE_NUM]; + struct dc_plane_state plane_states[MAX_SURFACES]; struct dc_stream_state stream_state; }; diff --git a/drivers/gpu/drm/amd/display/dc/dc_stream.h b/drivers/gpu/drm/amd/display/dc/dc_stream.h index 713884103aea..7ccdeb681918 100644 --- a/drivers/gpu/drm/amd/display/dc/dc_stream.h +++ b/drivers/gpu/drm/amd/display/dc/dc_stream.h @@ -56,7 +56,7 @@ struct dc_stream_status { int plane_count; int audio_inst; struct timing_sync_info timing_sync_info; - struct dc_plane_state *plane_states[MAX_SURFACE_NUM]; + struct dc_plane_state *plane_states[MAX_SURFACES]; bool is_abm_supported; struct mall_stream_config mall_stream_config; bool fpo_in_use; diff --git a/drivers/gpu/drm/amd/display/dc/dc_types.h b/drivers/gpu/drm/amd/display/dc/dc_types.h index edf4df1d03b5..9466b63644d5 100644 --- a/drivers/gpu/drm/amd/display/dc/dc_types.h +++ b/drivers/gpu/drm/amd/display/dc/dc_types.h @@ -76,7 +76,6 @@ struct dc_perf_trace { unsigned long last_entry_write; }; -#define MAX_SURFACE_NUM 6 #define NUM_PIXEL_FORMATS 10 enum tiling_mode { diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml2_mall_phantom.c b/drivers/gpu/drm/amd/display/dc/dml2/dml2_mall_phantom.c index 3d29169dd6bb..6b3b8803e0ae 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml2_mall_phantom.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml2_mall_phantom.c @@ -813,7 +813,7 @@ static bool remove_all_phantom_planes_for_stream(struct dml2_context *ctx, struc { int i, old_plane_count; struct dc_stream_status *stream_status = NULL; - struct dc_plane_state *del_planes[MAX_SURFACE_NUM] = { 0 }; + struct dc_plane_state *del_planes[MAX_SURFACES] = { 0 }; for (i = 0; i < context->stream_count; i++) if (context->streams[i] == stream) { -- 2.51.0