[AMD Official Use Only - AMD Internal Distribution Only] One minor comment b. Either way Reviewed-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@xxxxxxx> -----Original Message----- From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> On Behalf Of Jonathan Kim Sent: Monday, July 22, 2024 2:57 PM To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx Cc: Kim, Jonathan <Jonathan.Kim@xxxxxxx>; Kim, Jonathan <Jonathan.Kim@xxxxxxx> Subject: [PATCH] drm/amdkfd: fix debug watchpoints for logical devices The number of watchpoints should be set and constrained per logical partition device, not by the socket device. Signed-off-by: Jonathan Kim <jonathan.kim@xxxxxxx> --- drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 20 ++++++++++---------- drivers/gpu/drm/amd/amdkfd/kfd_device.c | 4 ++-- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 8 ++++---- 3 files changed, 16 insertions(+), 16 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c index 34a282540c7e..312dfa84f29f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c @@ -365,47 +365,47 @@ static int kfd_dbg_get_dev_watch_id(struct kfd_process_device *pdd, int *watch_i *watch_id = KFD_DEBUGGER_INVALID_WATCH_POINT_ID; - spin_lock(&pdd->dev->kfd->watch_points_lock); + spin_lock(&pdd->dev->watch_points_lock); for (i = 0; i < MAX_WATCH_ADDRESSES; i++) { /* device watchpoint in use so skip */ - if ((pdd->dev->kfd->alloc_watch_ids >> i) & 0x1) + if ((pdd->dev->alloc_watch_ids >> i) & 0x1) continue; pdd->alloc_watch_ids |= 0x1 << i; - pdd->dev->kfd->alloc_watch_ids |= 0x1 << i; + pdd->dev->alloc_watch_ids |= 0x1 << i; *watch_id = i; - spin_unlock(&pdd->dev->kfd->watch_points_lock); + spin_unlock(&pdd->dev->watch_points_lock); return 0; } - spin_unlock(&pdd->dev->kfd->watch_points_lock); + spin_unlock(&pdd->dev->watch_points_lock); return -ENOMEM; } static void kfd_dbg_clear_dev_watch_id(struct kfd_process_device *pdd, int watch_id) { - spin_lock(&pdd->dev->kfd->watch_points_lock); + spin_lock(&pdd->dev->watch_points_lock); /* process owns device watch point so safe to clear */ if ((pdd->alloc_watch_ids >> watch_id) & 0x1) { pdd->alloc_watch_ids &= ~(0x1 << watch_id); - pdd->dev->kfd->alloc_watch_ids &= ~(0x1 << watch_id); + pdd->dev->alloc_watch_ids &= ~(0x1 << watch_id); } - spin_unlock(&pdd->dev->kfd->watch_points_lock); + spin_unlock(&pdd->dev->watch_points_lock); } static bool kfd_dbg_owns_dev_watch_id(struct kfd_process_device *pdd, int watch_id) { bool owns_watch_id = false; - spin_lock(&pdd->dev->kfd->watch_points_lock); + spin_lock(&pdd->dev->watch_points_lock); owns_watch_id = watch_id < MAX_WATCH_ADDRESSES && ((pdd->alloc_watch_ids >> watch_id) & 0x1); - spin_unlock(&pdd->dev->kfd->watch_points_lock); + spin_unlock(&pdd->dev->watch_points_lock); return owns_watch_id; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index f4d20adaa068..f91a9b6ce3fb 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -885,12 +885,12 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, goto node_init_error; } Wouldn't it better to initialize node->watch_points_lock here. spin_lock_init(&node->watch_points_lock); kfd->nodes[i] = node; + + spin_lock_init(&kfd->nodes[i]->watch_points_lock); } svm_range_set_max_pages(kfd->adev); - spin_lock_init(&kfd->watch_points_lock); - kfd->init_complete = true; dev_info(kfd_device, "added device %x:%x\n", kfd->adev->pdev->vendor, kfd->adev->pdev->device); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index 2b3ec92981e8..653e1f934107 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -310,6 +310,10 @@ struct kfd_node { struct kfd_local_mem_info local_mem_info; struct kfd_dev *kfd; + + /* Track per device allocated watch points */ + uint32_t alloc_watch_ids; + spinlock_t watch_points_lock; }; struct kfd_dev { @@ -362,10 +366,6 @@ struct kfd_dev { struct kfd_node *nodes[MAX_KFD_NODES]; unsigned int num_nodes; - /* Track per device allocated watch points */ - uint32_t alloc_watch_ids; - spinlock_t watch_points_lock; - /* Kernel doorbells for KFD device */ struct amdgpu_bo *doorbells; -- 2.34.1