On 12/15/23 7:12 PM, Hou Tao wrote:
Hi,
On 12/16/2023 10:30 AM, Yonghong Song wrote:
Commit 41a5db8d8161 ("Add support for non-fix-size percpu mem allocation")
added support for non-fix-size percpu memory allocation.
Such allocation will allocate percpu memory for all buckets on all
cpus and the memory consumption is in the order to quadratic.
For example, let us say, 4 cpus, unit size 16 bytes, so each
cpu has 16 * 4 = 64 bytes, with 4 cpus, total will be 64 * 4 = 256 bytes.
Then let us say, 8 cpus with the same unit size, each cpu
has 16 * 8 = 128 bytes, with 8 cpus, total will be 128 * 8 = 1024 bytes.
So if the number of cpus doubles, the number of memory consumption
will be 4 times. So for a system with large number of cpus, the
memory consumption goes up quickly with quadratic order.
For example, for 4KB percpu allocation, 128 cpus. The total memory
consumption will 4KB * 128 * 128 = 64MB. Things will become
worse if the number of cpus is bigger (e.g., 512, 1024, etc.)
SNIP
+__init int bpf_mem_alloc_percpu_init(struct bpf_mem_alloc *ma)
+{
+ struct bpf_mem_caches __percpu *pcc;
+
+ pcc = __alloc_percpu_gfp(sizeof(struct bpf_mem_caches), 8, GFP_KERNEL);
+ if (!pcc)
+ return -ENOMEM;
+
+ ma->caches = pcc;
+ ma->percpu = true;
+ return 0;
+}
+
+int bpf_mem_alloc_percpu_unit_init(struct bpf_mem_alloc *ma, int size)
+{
+ int cpu, i, err = 0, unit_size, percpu_size;
+ struct bpf_mem_caches *cc, __percpu *pcc;
+ struct obj_cgroup *objcg;
+ struct bpf_mem_cache *c;
+
+ i = bpf_mem_cache_idx(size);
+ if (i < 0)
+ return -EINVAL;
+
+ /* room for llist_node and per-cpu pointer */
+ percpu_size = LLIST_NODE_SZ + sizeof(void *);
+
+ pcc = ma->caches;
+ unit_size = sizes[i];
+
+#ifdef CONFIG_MEMCG_KMEM
+ objcg = get_obj_cgroup_from_current();
+#endif
For bpf_global_percpu_ma, we also need to account the allocated memory
to root memory cgroup just like bpf_global_ma did, do we ? So it seems
that we need to initialize c->objcg early in bpf_mem_alloc_percpu_init ().
Good point. Agree. the original behavior percpu non-fix-size mem
allocation is to do get_obj_cgroup_from_current() at init stage
and charge to root memory cgroup, and we indeed should move
the above bpf_mem_alloc_percpu_init().
+ for_each_possible_cpu(cpu) {
+ cc = per_cpu_ptr(pcc, cpu);
+ c = &cc->cache[i];
+ if (cpu == 0 && c->unit_size)
+ goto out;
+
+ c->unit_size = unit_size;
+ c->objcg = objcg;
+ c->percpu_size = percpu_size;
+ c->tgt = c;
+
+ init_refill_work(c);
+ prefill_mem_cache(c, cpu);
+
+ if (cpu == 0) {
+ err = check_obj_size(c, i);
+ if (err) {
+ drain_mem_cache(c);
+ memset(c, 0, sizeof(*c));
I also forgot about c->objcg. objcg may be leaked if we do memset() here.
The objcg gets a reference at init bpf_mem_alloc_init() stage
and released at bpf_mem_alloc_destroy(). For bpf_global_ma,
if there is a failure, indeed bpf_mem_alloc_destroy() will be
called and the reference c->objcg will be released.
So if we move get_obj_cgroup_from_current() to
bpf_mem_alloc_percpu_init() stage, we should be okay here.
BTW, is check_obj_size() really necessary here? My answer is no
since as you mentioned, the size->cache_index is pretty stable,
so check_obj_size() should not return error in such cases.
What do you think?
+ goto out;
+ }
+ }
+ }
+
+out:
+ return err;
+}
+
.
