From 155ce502e9adf9001d6fa35ca85843b7f0e538c7 Mon Sep 17 00:00:00 2001 From: Philip Yang Date: Fri, 5 Apr 2024 15:46:50 -0400 Subject: drm/amdgpu: Support contiguous VRAM allocation RDMA device with limited scatter-gather ability requires contiguous VRAM buffer allocation for RDMA peer direct support. Add a new KFD alloc memory flag and store as bo alloc flag AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS. When pin this bo to export for RDMA peerdirect access, this will set TTM_PL_FLAG_CONTIFUOUS flag, and ask VRAM buddy allocator to get contiguous VRAM. Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- include/uapi/linux/kfd_ioctl.h | 1 + 1 file changed, 1 insertion(+) (limited to 'include/uapi/linux') diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h index 2040a470ddb4..d09c4a18e571 100644 --- a/include/uapi/linux/kfd_ioctl.h +++ b/include/uapi/linux/kfd_ioctl.h @@ -407,6 +407,7 @@ struct kfd_ioctl_acquire_vm_args { #define KFD_IOC_ALLOC_MEM_FLAGS_COHERENT (1 << 26) #define KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED (1 << 25) #define KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT (1 << 24) +#define KFD_IOC_ALLOC_MEM_FLAGS_CONTIGUOUS (1 << 23) /* Allocate memory for later SVM (shared virtual memory) mapping. * -- cgit v1.2.3 From 52666d8cc7f9c7810d25d5ef41f1a0b62153f368 Mon Sep 17 00:00:00 2001 From: Philip Yang Date: Mon, 8 Apr 2024 09:26:56 -0400 Subject: drm/amdkfd: Bump kfd version for contiguous VRAM allocation Bump the kfd ioctl minor version to delcare the contiguous VRAM allocation flag support. Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- include/uapi/linux/kfd_ioctl.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'include/uapi/linux') diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h index d09c4a18e571..f8e9d3c1d117 100644 --- a/include/uapi/linux/kfd_ioctl.h +++ b/include/uapi/linux/kfd_ioctl.h @@ -41,9 +41,10 @@ * - 1.13 - Add debugger API * - 1.14 - Update kfd_event_data * - 1.15 - Enable managing mappings in compute VMs with GEM_VA ioctl + * - 1.16 - Add contiguous VRAM allocation flag */ #define KFD_IOCTL_MAJOR_VERSION 1 -#define KFD_IOCTL_MINOR_VERSION 15 +#define KFD_IOCTL_MINOR_VERSION 16 struct kfd_ioctl_get_version_args { __u32 major_version; /* from KFD */ -- cgit v1.2.3 From 9243240bed3859fba2d15c016902a4c73a186249 Mon Sep 17 00:00:00 2001 From: Jonathan Kim Date: Mon, 21 Aug 2023 11:47:47 -0400 Subject: drm/amdkfd: enable single alu ops for gfx12 GFX12 debugging requires setting up precise ALU operation for catching ALU exceptions. Signed-off-by: Jonathan Kim Tested-by: Lancelot Six Reviewed-by: Eric Huang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 15 +++++++++++++-- drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 4 ++++ include/uapi/linux/kfd_ioctl.h | 1 + include/uapi/linux/kfd_sysfs.h | 19 ++++++++++--------- 4 files changed, 28 insertions(+), 11 deletions(-) (limited to 'include/uapi/linux') diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c index d889e3545120..45b1975b149a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c @@ -25,6 +25,7 @@ #include "kfd_topology.h" #include #include +#include #define MAX_WATCH_ADDRESSES 4 @@ -497,14 +498,24 @@ int kfd_dbg_trap_set_flags(struct kfd_process *target, uint32_t *flags) int i, r = 0, rewind_count = 0; for (i = 0; i < target->n_pdds; i++) { - if (!kfd_dbg_is_per_vmid_supported(target->pdds[i]->dev) && + struct kfd_topology_device *topo_dev = + kfd_topology_device_by_id(target->pdds[i]->dev->id); + uint32_t caps = topo_dev->node_props.capability; + + if (!(caps | HSA_CAP_TRAP_DEBUG_PRECISE_MEMORY_OPERATIONS_SUPPORTED) && (*flags & KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP)) { *flags = prev_flags; return -EACCES; } + + if (!(caps | HSA_CAP_TRAP_DEBUG_PRECISE_ALU_OPERATIONS_SUPPORTED) && + (*flags & KFD_DBG_TRAP_FLAG_SINGLE_ALU_OP)) { + *flags = prev_flags; + return -EACCES; + } } - target->dbg_flags = *flags & KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP; + target->dbg_flags = *flags; *flags = prev_flags; for (i = 0; i < target->n_pdds; i++) { struct kfd_process_device *pdd = target->pdds[i]; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 71bea1fcb838..ba326b43bec5 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -1929,6 +1929,10 @@ static void kfd_topology_set_capabilities(struct kfd_topology_device *dev) if (KFD_GC_VERSION(dev->gpu) >= IP_VERSION(11, 0, 0)) dev->node_props.capability |= HSA_CAP_TRAP_DEBUG_PRECISE_MEMORY_OPERATIONS_SUPPORTED; + + if (KFD_GC_VERSION(dev->gpu) >= IP_VERSION(12, 0, 0)) + dev->node_props.capability |= + HSA_CAP_TRAP_DEBUG_PRECISE_ALU_OPERATIONS_SUPPORTED; } kfd_topology_set_dbg_firmware_support(dev); diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h index f8e9d3c1d117..285a36601dc9 100644 --- a/include/uapi/linux/kfd_ioctl.h +++ b/include/uapi/linux/kfd_ioctl.h @@ -854,6 +854,7 @@ enum kfd_dbg_trap_address_watch_mode { /* Additional wave settings */ enum kfd_dbg_trap_flags { KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP = 1, + KFD_DBG_TRAP_FLAG_SINGLE_ALU_OP = 2, }; /* Trap exceptions */ diff --git a/include/uapi/linux/kfd_sysfs.h b/include/uapi/linux/kfd_sysfs.h index a51b7331e0b4..5e8d28617efa 100644 --- a/include/uapi/linux/kfd_sysfs.h +++ b/include/uapi/linux/kfd_sysfs.h @@ -51,15 +51,16 @@ /* Old buggy user mode depends on this being 0 */ #define HSA_CAP_RESERVED_WAS_SRAM_EDCSUPPORTED 0x00080000 -#define HSA_CAP_MEM_EDCSUPPORTED 0x00100000 -#define HSA_CAP_RASEVENTNOTIFY 0x00200000 -#define HSA_CAP_ASIC_REVISION_MASK 0x03c00000 -#define HSA_CAP_ASIC_REVISION_SHIFT 22 -#define HSA_CAP_SRAM_EDCSUPPORTED 0x04000000 -#define HSA_CAP_SVMAPI_SUPPORTED 0x08000000 -#define HSA_CAP_FLAGS_COHERENTHOSTACCESS 0x10000000 -#define HSA_CAP_TRAP_DEBUG_FIRMWARE_SUPPORTED 0x20000000 -#define HSA_CAP_RESERVED 0xe00f8000 +#define HSA_CAP_MEM_EDCSUPPORTED 0x00100000 +#define HSA_CAP_RASEVENTNOTIFY 0x00200000 +#define HSA_CAP_ASIC_REVISION_MASK 0x03c00000 +#define HSA_CAP_ASIC_REVISION_SHIFT 22 +#define HSA_CAP_SRAM_EDCSUPPORTED 0x04000000 +#define HSA_CAP_SVMAPI_SUPPORTED 0x08000000 +#define HSA_CAP_FLAGS_COHERENTHOSTACCESS 0x10000000 +#define HSA_CAP_TRAP_DEBUG_FIRMWARE_SUPPORTED 0x20000000 +#define HSA_CAP_TRAP_DEBUG_PRECISE_ALU_OPERATIONS_SUPPORTED 0x40000000 +#define HSA_CAP_RESERVED 0x800f8000 /* debug_prop bits in node properties */ #define HSA_DBG_WATCH_ADDR_MASK_LO_BIT_MASK 0x0000000f -- cgit v1.2.3 From 310ec03841a36e3f45fb528f0dfdfe5b9e84b037 Mon Sep 17 00:00:00 2001 From: Barry Song Date: Wed, 5 Jun 2024 13:26:05 +1200 Subject: dma-buf: align fd_flags and heap_flags with dma_heap_allocation_data dma_heap_allocation_data defines the UAPI as follows: struct dma_heap_allocation_data { __u64 len; __u32 fd; __u32 fd_flags; __u64 heap_flags; }; However, dma_heap_buffer_alloc() casts both fd_flags and heap_flags into unsigned int. We're inconsistent with types in the non UAPI arguments. This patch fixes it. Signed-off-by: Barry Song Acked-by: John Stultz Signed-off-by: Sumit Semwal Link: https://patchwork.freedesktop.org/patch/msgid/20240605012605.5341-1-21cnbao@gmail.com --- drivers/dma-buf/dma-heap.c | 4 ++-- include/uapi/linux/dma-heap.h | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) (limited to 'include/uapi/linux') diff --git a/drivers/dma-buf/dma-heap.c b/drivers/dma-buf/dma-heap.c index 84ae708fafe7..2298ca5e112e 100644 --- a/drivers/dma-buf/dma-heap.c +++ b/drivers/dma-buf/dma-heap.c @@ -50,8 +50,8 @@ static struct class *dma_heap_class; static DEFINE_XARRAY_ALLOC(dma_heap_minors); static int dma_heap_buffer_alloc(struct dma_heap *heap, size_t len, - unsigned int fd_flags, - unsigned int heap_flags) + u32 fd_flags, + u64 heap_flags) { struct dma_buf *dmabuf; int fd; diff --git a/include/uapi/linux/dma-heap.h b/include/uapi/linux/dma-heap.h index 6f84fa08e074..a4cf716a49fa 100644 --- a/include/uapi/linux/dma-heap.h +++ b/include/uapi/linux/dma-heap.h @@ -19,7 +19,7 @@ #define DMA_HEAP_VALID_FD_FLAGS (O_CLOEXEC | O_ACCMODE) /* Currently no heap flags */ -#define DMA_HEAP_VALID_HEAP_FLAGS (0) +#define DMA_HEAP_VALID_HEAP_FLAGS (0ULL) /** * struct dma_heap_allocation_data - metadata passed from userspace for -- cgit v1.2.3