bpf: Optimize program stats

Move bpf_prog_stats from prog->aux into prog to avoid one extra load
in critical path of program execution.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210210033634.62081-2-alexei.starovoitov@gmail.com

include/linux/bpf.h

@@ -14,7 +14,6 @@
 #include <linux/numa.h>
 #include <linux/mm_types.h>
 #include <linux/wait.h>
-#include <linux/u64_stats_sync.h>
 #include <linux/refcount.h>
 #include <linux/mutex.h>
 #include <linux/module.h>
@@ -507,12 +506,6 @@ enum bpf_cgroup_storage_type {
  */
 #define MAX_BPF_FUNC_ARGS 12
 
-struct bpf_prog_stats {
-	u64 cnt;
-	u64 nsecs;
-	struct u64_stats_sync syncp;
-} __aligned(2 * sizeof(u64));
-
 struct btf_func_model {
 	u8 ret_size;
 	u8 nr_args;
@@ -845,7 +838,6 @@ struct bpf_prog_aux {
 	u32 linfo_idx;
 	u32 num_exentries;
 	struct exception_table_entry *extable;
-	struct bpf_prog_stats __percpu *stats;
 	union {
 		struct work_struct work;
 		struct rcu_head	rcu;

include/linux/filter.h

@@ -22,6 +22,7 @@
 #include <linux/vmalloc.h>
 #include <linux/sockptr.h>
 #include <crypto/sha1.h>
+#include <linux/u64_stats_sync.h>
 
 #include <net/sch_generic.h>
 
@@ -539,6 +540,12 @@ struct bpf_binary_header {
 	u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
 };
 
+struct bpf_prog_stats {
+	u64 cnt;
+	u64 nsecs;
+	struct u64_stats_sync syncp;
+} __aligned(2 * sizeof(u64));
+
 struct bpf_prog {
 	u16			pages;		/* Number of allocated pages */
 	u16			jited:1,	/* Is our filter JIT'ed? */
@@ -557,10 +564,11 @@ struct bpf_prog {
 	u32			len;		/* Number of filter blocks */
 	u32			jited_len;	/* Size of jited insns in bytes */
 	u8			tag[BPF_TAG_SIZE];
-	struct bpf_prog_aux	*aux;		/* Auxiliary fields */
+	struct bpf_prog_stats __percpu *stats;
-	struct sock_fprog_kern	*orig_prog;	/* Original BPF program */
 	unsigned int		(*bpf_func)(const void *ctx,
 					    const struct bpf_insn *insn);
+	struct bpf_prog_aux	*aux;		/* Auxiliary fields */
+	struct sock_fprog_kern	*orig_prog;	/* Original BPF program */
 	/* Instructions for interpreter */
 	struct sock_filter	insns[0];
 	struct bpf_insn		insnsi[];
@@ -581,7 +589,7 @@ DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
 		struct bpf_prog_stats *__stats;				\
 		u64 __start = sched_clock();				\
 		__ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func);	\
-		__stats = this_cpu_ptr(prog->aux->stats);		\
+		__stats = this_cpu_ptr(prog->stats);			\
 		u64_stats_update_begin(&__stats->syncp);		\
 		__stats->cnt++;						\
 		__stats->nsecs += sched_clock() - __start;		\

kernel/bpf/core.c

@@ -114,8 +114,8 @@ struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
 	if (!prog)
 		return NULL;
 
-	prog->aux->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
+	prog->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
-	if (!prog->aux->stats) {
+	if (!prog->stats) {
 		kfree(prog->aux);
 		vfree(prog);
 		return NULL;
@@ -124,7 +124,7 @@ struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
 	for_each_possible_cpu(cpu) {
 		struct bpf_prog_stats *pstats;
 
-		pstats = per_cpu_ptr(prog->aux->stats, cpu);
+		pstats = per_cpu_ptr(prog->stats, cpu);
 		u64_stats_init(&pstats->syncp);
 	}
 	return prog;
@@ -249,10 +249,10 @@ void __bpf_prog_free(struct bpf_prog *fp)
 	if (fp->aux) {
 		mutex_destroy(&fp->aux->used_maps_mutex);
 		mutex_destroy(&fp->aux->dst_mutex);
-		free_percpu(fp->aux->stats);
 		kfree(fp->aux->poke_tab);
 		kfree(fp->aux);
 	}
+	free_percpu(fp->stats);
 	vfree(fp);
 }
 

kernel/bpf/syscall.c

@@ -1739,7 +1739,7 @@ static void bpf_prog_get_stats(const struct bpf_prog *prog,
 		unsigned int start;
 		u64 tnsecs, tcnt;
 
-		st = per_cpu_ptr(prog->aux->stats, cpu);
+		st = per_cpu_ptr(prog->stats, cpu);
 		do {
 			start = u64_stats_fetch_begin_irq(&st->syncp);
 			tnsecs = st->nsecs;

kernel/bpf/trampoline.c

@@ -412,7 +412,7 @@ void notrace __bpf_prog_exit(struct bpf_prog *prog, u64 start)
 	     * Hence check that 'start' is not zero.
 	     */
 	    start) {
-		stats = this_cpu_ptr(prog->aux->stats);
+		stats = this_cpu_ptr(prog->stats);
 		u64_stats_update_begin(&stats->syncp);
 		stats->cnt++;
 		stats->nsecs += sched_clock() - start;

kernel/bpf/verifier.c

@@ -11253,7 +11253,7 @@ static int jit_subprogs(struct bpf_verifier_env *env)
 		/* BPF_PROG_RUN doesn't call subprogs directly,
 		 * hence main prog stats include the runtime of subprogs.
 		 * subprogs don't have IDs and not reachable via prog_get_next_id
-		 * func[i]->aux->stats will never be accessed and stays NULL
+		 * func[i]->stats will never be accessed and stays NULL
 		 */
 		func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER);
 		if (!func[i])