From: Philip Yang Philip.Yang@amd.com
Get the intersection of attributes over all memory in the given range
Signed-off-by: Philip Yang Philip.Yang@amd.com Signed-off-by: Alex Sierra alex.sierra@amd.com Signed-off-by: Felix Kuehling Felix.Kuehling@amd.com --- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 175 ++++++++++++++++++++++++++- 1 file changed, 173 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 0b0410837be9..017e77e9ae1e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -75,8 +75,8 @@ static void svm_range_set_default_attributes(int32_t *location, int32_t *prefetch_loc, uint8_t *granularity, uint32_t *flags) { - *location = 0; - *prefetch_loc = 0; + *location = KFD_IOCTL_SVM_LOCATION_UNDEFINED; + *prefetch_loc = KFD_IOCTL_SVM_LOCATION_UNDEFINED; *granularity = 9; *flags = KFD_IOCTL_SVM_FLAG_HOST_ACCESS | KFD_IOCTL_SVM_FLAG_COHERENT; @@ -581,6 +581,174 @@ svm_range_set_attr(struct kfd_process *p, uint64_t start, uint64_t size, return r; }
+static int +svm_range_get_attr(struct kfd_process *p, uint64_t start, uint64_t size, + uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs) +{ + DECLARE_BITMAP(bitmap_access, MAX_GPU_INSTANCE); + DECLARE_BITMAP(bitmap_aip, MAX_GPU_INSTANCE); + bool get_preferred_loc = false; + bool get_prefetch_loc = false; + bool get_granularity = false; + bool get_accessible = false; + bool get_flags = false; + uint64_t last = start + size - 1UL; + struct mm_struct *mm = current->mm; + uint8_t granularity = 0xff; + struct interval_tree_node *node; + struct svm_range_list *svms; + struct svm_range *prange; + uint32_t prefetch_loc = KFD_IOCTL_SVM_LOCATION_UNDEFINED; + uint32_t location = KFD_IOCTL_SVM_LOCATION_UNDEFINED; + uint32_t flags = 0xffffffff; + int gpuidx; + uint32_t i; + + pr_debug("svms 0x%p [0x%llx 0x%llx] nattr 0x%x\n", &p->svms, start, + start + size - 1, nattr); + + mmap_read_lock(mm); + if (!svm_range_is_valid(mm, start, size)) { + pr_debug("invalid range\n"); + mmap_read_unlock(mm); + return -EINVAL; + } + mmap_read_unlock(mm); + + for (i = 0; i < nattr; i++) { + switch (attrs[i].type) { + case KFD_IOCTL_SVM_ATTR_PREFERRED_LOC: + get_preferred_loc = true; + break; + case KFD_IOCTL_SVM_ATTR_PREFETCH_LOC: + get_prefetch_loc = true; + break; + case KFD_IOCTL_SVM_ATTR_ACCESS: + if (!svm_get_supported_dev_by_id( + p, attrs[i].value, NULL)) + return -EINVAL; + get_accessible = true; + break; + case KFD_IOCTL_SVM_ATTR_SET_FLAGS: + get_flags = true; + break; + case KFD_IOCTL_SVM_ATTR_GRANULARITY: + get_granularity = true; + break; + case KFD_IOCTL_SVM_ATTR_CLR_FLAGS: + case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE: + case KFD_IOCTL_SVM_ATTR_NO_ACCESS: + fallthrough; + default: + pr_debug("get invalid attr type 0x%x\n", attrs[i].type); + return -EINVAL; + } + } + + svms = &p->svms; + + svms_lock(svms); + + node = interval_tree_iter_first(&svms->objects, start, last); + if (!node) { + pr_debug("range attrs not found return default values\n"); + svm_range_set_default_attributes(&location, &prefetch_loc, + &granularity, &flags); + /* TODO: Automatically create SVM ranges and map them on + * GPU page faults + if (p->xnack_enabled) + bitmap_fill(bitmap_access, MAX_GPU_INSTANCE); + FIXME: Only set bits for supported GPUs + FIXME: I think this should be done inside + svm_range_set_default_attributes, so that it will + apply to all newly created ranges + */ + + goto fill_values; + } + bitmap_fill(bitmap_access, MAX_GPU_INSTANCE); + bitmap_fill(bitmap_aip, MAX_GPU_INSTANCE); + + while (node) { + struct interval_tree_node *next; + + prange = container_of(node, struct svm_range, it_node); + next = interval_tree_iter_next(node, start, last); + + if (get_preferred_loc) { + if (prange->preferred_loc == + KFD_IOCTL_SVM_LOCATION_UNDEFINED || + (location != KFD_IOCTL_SVM_LOCATION_UNDEFINED && + location != prange->preferred_loc)) { + location = KFD_IOCTL_SVM_LOCATION_UNDEFINED; + get_preferred_loc = false; + } else { + location = prange->preferred_loc; + } + } + if (get_prefetch_loc) { + if (prange->prefetch_loc == + KFD_IOCTL_SVM_LOCATION_UNDEFINED || + (prefetch_loc != KFD_IOCTL_SVM_LOCATION_UNDEFINED && + prefetch_loc != prange->prefetch_loc)) { + prefetch_loc = KFD_IOCTL_SVM_LOCATION_UNDEFINED; + get_prefetch_loc = false; + } else { + prefetch_loc = prange->prefetch_loc; + } + } + if (get_accessible) { + bitmap_and(bitmap_access, bitmap_access, + prange->bitmap_access, MAX_GPU_INSTANCE); + bitmap_and(bitmap_aip, bitmap_aip, + prange->bitmap_aip, MAX_GPU_INSTANCE); + } + if (get_flags) + flags &= prange->flags; + + if (get_granularity && prange->granularity < granularity) + granularity = prange->granularity; + + node = next; + } +fill_values: + svms_unlock(svms); + + for (i = 0; i < nattr; i++) { + switch (attrs[i].type) { + case KFD_IOCTL_SVM_ATTR_PREFERRED_LOC: + attrs[i].value = location; + break; + case KFD_IOCTL_SVM_ATTR_PREFETCH_LOC: + attrs[i].value = prefetch_loc; + break; + case KFD_IOCTL_SVM_ATTR_ACCESS: + gpuidx = kfd_process_gpuidx_from_gpuid(p, + attrs[i].value); + if (gpuidx < 0) { + pr_debug("invalid gpuid %x\n", attrs[i].value); + return -EINVAL; + } + if (test_bit(gpuidx, bitmap_access)) + attrs[i].type = KFD_IOCTL_SVM_ATTR_ACCESS; + else if (test_bit(gpuidx, bitmap_aip)) + attrs[i].type = + KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE; + else + attrs[i].type = KFD_IOCTL_SVM_ATTR_NO_ACCESS; + break; + case KFD_IOCTL_SVM_ATTR_SET_FLAGS: + attrs[i].value = flags; + break; + case KFD_IOCTL_SVM_ATTR_GRANULARITY: + attrs[i].value = (uint32_t)granularity; + break; + } + } + + return 0; +} + int svm_ioctl(struct kfd_process *p, enum kfd_ioctl_svm_op op, uint64_t start, uint64_t size, uint32_t nattrs, struct kfd_ioctl_svm_attribute *attrs) @@ -594,6 +762,9 @@ svm_ioctl(struct kfd_process *p, enum kfd_ioctl_svm_op op, uint64_t start, case KFD_IOCTL_SVM_OP_SET_ATTR: r = svm_range_set_attr(p, start, size, nattrs, attrs); break; + case KFD_IOCTL_SVM_OP_GET_ATTR: + r = svm_range_get_attr(p, start, size, nattrs, attrs); + break; default: r = EINVAL; break;