[PATCH bpf-next 2/2] bpf: Switch bpf arena to use drm_mm instead of maple_tree
Alexei Starovoitov
alexei.starovoitov at gmail.com
Fri Nov 1 23:54:53 UTC 2024
From: Alexei Starovoitov <ast at kernel.org>
bpf arena is moving towards non-sleepable allocations in tracing
context while maple_tree does kmalloc/kfree deep inside. Hence switch
bpf arena to drm_mm algorithm that works with externally provided
drm_mm_node-s. This patch kmalloc/kfree-s drm_mm_node-s, but the next
patch will switch to bpf_mem_alloc and preallocated drm_mm_node-s.
Signed-off-by: Alexei Starovoitov <ast at kernel.org>
---
kernel/bpf/arena.c | 67 +++++++++++++++++++++++++++++++++++-----------
lib/Makefile | 1 +
2 files changed, 53 insertions(+), 15 deletions(-)
diff --git a/kernel/bpf/arena.c b/kernel/bpf/arena.c
index e52b3ad231b9..ef1de26020f2 100644
--- a/kernel/bpf/arena.c
+++ b/kernel/bpf/arena.c
@@ -6,6 +6,7 @@
#include <linux/btf_ids.h>
#include <linux/vmalloc.h>
#include <linux/pagemap.h>
+#include <drm/drm_mm.h>
/*
* bpf_arena is a sparsely populated shared memory region between bpf program and
@@ -45,7 +46,7 @@ struct bpf_arena {
u64 user_vm_start;
u64 user_vm_end;
struct vm_struct *kern_vm;
- struct maple_tree mt;
+ struct drm_mm mm;
struct list_head vma_list;
struct mutex lock;
};
@@ -132,7 +133,7 @@ static struct bpf_map *arena_map_alloc(union bpf_attr *attr)
INIT_LIST_HEAD(&arena->vma_list);
bpf_map_init_from_attr(&arena->map, attr);
- mt_init_flags(&arena->mt, MT_FLAGS_ALLOC_RANGE);
+ drm_mm_init(&arena->mm, 0, attr->max_entries);
mutex_init(&arena->lock);
return &arena->map;
@@ -164,6 +165,7 @@ static int existing_page_cb(pte_t *ptep, unsigned long addr, void *data)
static void arena_map_free(struct bpf_map *map)
{
struct bpf_arena *arena = container_of(map, struct bpf_arena, map);
+ struct drm_mm_node *node, *next;
/*
* Check that user vma-s are not around when bpf map is freed.
@@ -183,7 +185,11 @@ static void arena_map_free(struct bpf_map *map)
apply_to_existing_page_range(&init_mm, bpf_arena_get_kern_vm_start(arena),
KERN_VM_SZ - GUARD_SZ, existing_page_cb, NULL);
free_vm_area(arena->kern_vm);
- mtree_destroy(&arena->mt);
+ drm_mm_for_each_node_safe(node, next, &arena->mm) {
+ drm_mm_remove_node(node);
+ kfree(node);
+ }
+ drm_mm_takedown(&arena->mm);
bpf_map_area_free(arena);
}
@@ -257,6 +263,7 @@ static vm_fault_t arena_vm_fault(struct vm_fault *vmf)
{
struct bpf_map *map = vmf->vma->vm_file->private_data;
struct bpf_arena *arena = container_of(map, struct bpf_arena, map);
+ struct drm_mm_node *node;
struct page *page;
long kbase, kaddr;
int ret;
@@ -274,20 +281,30 @@ static vm_fault_t arena_vm_fault(struct vm_fault *vmf)
/* User space requested to segfault when page is not allocated by bpf prog */
return VM_FAULT_SIGSEGV;
- ret = mtree_insert(&arena->mt, vmf->pgoff, MT_ENTRY, GFP_KERNEL);
- if (ret)
+ node = kzalloc(sizeof(*node), GFP_KERNEL);
+ if (!node)
+ return VM_FAULT_SIGSEGV;
+
+ node->start = vmf->pgoff;
+ node->size = 1;
+ ret = drm_mm_reserve_node(&arena->mm, node);
+ if (ret) {
+ kfree(node);
return VM_FAULT_SIGSEGV;
+ }
/* Account into memcg of the process that created bpf_arena */
ret = bpf_map_alloc_pages(map, GFP_KERNEL | __GFP_ZERO, NUMA_NO_NODE, 1, &page);
if (ret) {
- mtree_erase(&arena->mt, vmf->pgoff);
+ drm_mm_remove_node(node);
+ kfree(node);
return VM_FAULT_SIGSEGV;
}
ret = vm_area_map_pages(arena->kern_vm, kaddr, kaddr + PAGE_SIZE, &page);
if (ret) {
- mtree_erase(&arena->mt, vmf->pgoff);
+ drm_mm_remove_node(node);
+ kfree(node);
__free_page(page);
return VM_FAULT_SIGSEGV;
}
@@ -420,6 +437,7 @@ static long arena_alloc_pages(struct bpf_arena *arena, long uaddr, long page_cnt
/* user_vm_end/start are fixed before bpf prog runs */
long page_cnt_max = (arena->user_vm_end - arena->user_vm_start) >> PAGE_SHIFT;
u64 kern_vm_start = bpf_arena_get_kern_vm_start(arena);
+ struct drm_mm_node *node;
struct page **pages;
long pgoff = 0;
u32 uaddr32;
@@ -442,14 +460,21 @@ static long arena_alloc_pages(struct bpf_arena *arena, long uaddr, long page_cnt
if (!pages)
return 0;
+ node = kzalloc(sizeof(*node), GFP_KERNEL);
+ if (!node) {
+ kvfree(pages);
+ return 0;
+ }
guard(mutex)(&arena->lock);
- if (uaddr)
- ret = mtree_insert_range(&arena->mt, pgoff, pgoff + page_cnt - 1,
- MT_ENTRY, GFP_KERNEL);
- else
- ret = mtree_alloc_range(&arena->mt, &pgoff, MT_ENTRY,
- page_cnt, 0, page_cnt_max - 1, GFP_KERNEL);
+ if (uaddr) {
+ node->start = pgoff;
+ node->size = page_cnt;
+ ret = drm_mm_reserve_node(&arena->mm, node);
+ } else {
+ ret = drm_mm_insert_node(&arena->mm, node, page_cnt);
+ pgoff = node->start;
+ }
if (ret)
goto out_free_pages;
@@ -476,7 +501,8 @@ static long arena_alloc_pages(struct bpf_arena *arena, long uaddr, long page_cnt
kvfree(pages);
return clear_lo32(arena->user_vm_start) + uaddr32;
out:
- mtree_erase(&arena->mt, pgoff);
+ drm_mm_remove_node(node);
+ kfree(node);
out_free_pages:
kvfree(pages);
return 0;
@@ -499,6 +525,7 @@ static void zap_pages(struct bpf_arena *arena, long uaddr, long page_cnt)
static void arena_free_pages(struct bpf_arena *arena, long uaddr, long page_cnt)
{
u64 full_uaddr, uaddr_end;
+ struct drm_mm_node *node, *to_remove;
long kaddr, pgoff, i;
struct page *page;
@@ -516,7 +543,17 @@ static void arena_free_pages(struct bpf_arena *arena, long uaddr, long page_cnt)
pgoff = compute_pgoff(arena, uaddr);
/* clear range */
- mtree_store_range(&arena->mt, pgoff, pgoff + page_cnt - 1, NULL, GFP_KERNEL);
+ for (;;) {
+ to_remove = NULL;
+ drm_mm_for_each_node_in_range(node, &arena->mm, pgoff, pgoff + page_cnt) {
+ to_remove = node;
+ break;
+ }
+ if (!to_remove)
+ break;
+ drm_mm_remove_node(to_remove);
+ kfree(to_remove);
+ }
if (page_cnt > 1)
/* bulk zap if multiple pages being freed */
diff --git a/lib/Makefile b/lib/Makefile
index 605aa25b71ab..d4787e6b0942 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -58,6 +58,7 @@ obj-$(CONFIG_TEST_HEXDUMP) += test_hexdump.o
obj-y += kstrtox.o
obj-$(CONFIG_FIND_BIT_BENCHMARK) += find_bit_benchmark.o
obj-$(CONFIG_TEST_BPF) += test_bpf.o
+obj-$(CONFIG_BPF_SYSCALL) += drm_mm.o
obj-$(CONFIG_DRM) += drm_mm.o
test_dhry-objs := dhry_1.o dhry_2.o dhry_run.o
obj-$(CONFIG_TEST_DHRY) += test_dhry.o
--
2.43.5
More information about the dri-devel
mailing list