Message-ID: <aA8HImKeUutpFoeD@google.com>
Date: Sun, 27 Apr 2025 21:42:10 -0700
From: Namhyung Kim <namhyung@...nel.org>
To: Arnaldo Carvalho de Melo <acme@...nel.org>,
	Ian Rogers <irogers@...gle.com>,
	Kan Liang <kan.liang@...ux.intel.com>
Cc: Jiri Olsa <jolsa@...nel.org>, Adrian Hunter <adrian.hunter@...el.com>,
	Peter Zijlstra <peterz@...radead.org>,
	Ingo Molnar <mingo@...nel.org>, LKML <linux-kernel@...r.kernel.org>,
	linux-perf-users@...r.kernel.org, Song Liu <song@...nel.org>,
	bpf@...r.kernel.org, Stephane Eranian <eranian@...gle.com>,
	linux-mm@...ck.org
Subject: Re: [PATCH] perf lock contention: Symbolize zone->lock using BTF

Ping!

On Mon, Mar 31, 2025 at 11:30:55PM -0700, Namhyung Kim wrote:
> The struct zone is embedded in struct pglist_data which can be allocated
> for each NUMA node early in the boot process.  As it's not a slab object
> nor a global lock, this was not symbolized.
> 
> Since the zone->lock is often contended, it'd be nice if we can
> symbolize it.  On NUMA systems, node_data array will have pointers for
> struct pglist_data.  By following the pointer, it can calculate the
> address of each zone and its lock using BTF.  On UMA, it can just use
> contig_page_data and its zones.
> 
> The following example shows the zone lock contention at the end.
> 
>   $ sudo ./perf lock con -abl -E 5 -- ./perf bench sched messaging
>   # Running 'sched/messaging' benchmark:
>   # 20 sender and receiver processes per group
>   # 10 groups == 400 processes run
> 
>        Total time: 0.038 [sec]
>    contended   total wait     max wait     avg wait            address   symbol
> 
>         5167     18.17 ms     10.27 us      3.52 us   ffff953340052d00   &kmem_cache_node (spinlock)
>           38     11.75 ms    465.49 us    309.13 us   ffff95334060c480   &sock_inode_cache (spinlock)
>         3916     10.13 ms     10.43 us      2.59 us   ffff953342aecb40   &kmem_cache_node (spinlock)
>         2963     10.02 ms     13.75 us      3.38 us   ffff9533d2344098   &kmalloc-rnd-08-2k (spinlock)
>          216      5.05 ms     99.49 us     23.39 us   ffff9542bf7d65d0   zone_lock (spinlock)
> 
> Cc: linux-mm@...ck.org
> Signed-off-by: Namhyung Kim <namhyung@...nel.org>
> ---
>  tools/perf/util/bpf_lock_contention.c         | 88 +++++++++++++++++--
>  .../perf/util/bpf_skel/lock_contention.bpf.c  | 64 ++++++++++++++
>  tools/perf/util/bpf_skel/lock_data.h          |  1 +
>  tools/perf/util/bpf_skel/vmlinux/vmlinux.h    |  9 ++
>  tools/perf/util/lock-contention.h             |  1 +
>  5 files changed, 157 insertions(+), 6 deletions(-)
> 
> diff --git a/tools/perf/util/bpf_lock_contention.c b/tools/perf/util/bpf_lock_contention.c
> index 5af8f6d1bc952613..98395667220e58ee 100644
> --- a/tools/perf/util/bpf_lock_contention.c
> +++ b/tools/perf/util/bpf_lock_contention.c
> @@ -12,6 +12,7 @@
>  #include "util/lock-contention.h"
>  #include <linux/zalloc.h>
>  #include <linux/string.h>
> +#include <api/fs/fs.h>
>  #include <bpf/bpf.h>
>  #include <bpf/btf.h>
>  #include <inttypes.h>
> @@ -35,28 +36,26 @@ static bool slab_cache_equal(long key1, long key2, void *ctx __maybe_unused)
>  
>  static void check_slab_cache_iter(struct lock_contention *con)
>  {
> -	struct btf *btf = btf__load_vmlinux_btf();
>  	s32 ret;
>  
>  	hashmap__init(&slab_hash, slab_cache_hash, slab_cache_equal, /*ctx=*/NULL);
>  
> -	if (btf == NULL) {
> +	con->btf = btf__load_vmlinux_btf();
> +	if (con->btf == NULL) {
>  		pr_debug("BTF loading failed: %s\n", strerror(errno));
>  		return;
>  	}
>  
> -	ret = btf__find_by_name_kind(btf, "bpf_iter__kmem_cache", BTF_KIND_STRUCT);
> +	ret = btf__find_by_name_kind(con->btf, "bpf_iter__kmem_cache", BTF_KIND_STRUCT);
>  	if (ret < 0) {
>  		bpf_program__set_autoload(skel->progs.slab_cache_iter, false);
>  		pr_debug("slab cache iterator is not available: %d\n", ret);
> -		goto out;
> +		return;
>  	}
>  
>  	has_slab_iter = true;
>  
>  	bpf_map__set_max_entries(skel->maps.slab_caches, con->map_nr_entries);
> -out:
> -	btf__free(btf);
>  }
>  
>  static void run_slab_cache_iter(void)
> @@ -109,6 +108,75 @@ static void exit_slab_cache_iter(void)
>  	hashmap__clear(&slab_hash);
>  }
>  
> +static void init_numa_data(struct lock_contention *con)
> +{
> +	struct symbol *sym;
> +	struct map *kmap;
> +	char *buf = NULL, *p;
> +	size_t len;
> +	long last = -1;
> +	int ret;
> +
> +	/*
> +	 * 'struct zone' is embedded in 'struct pglist_data' as an array.
> +	 * As we may not have full information of the struct zone in the
> +	 * (fake) vmlinux.h, let's get the actual size from BTF.
> +	 */
> +	ret = btf__find_by_name_kind(con->btf, "zone", BTF_KIND_STRUCT);
> +	if (ret < 0) {
> +		pr_debug("cannot get type of struct zone: %d\n", ret);
> +		return;
> +	}
> +
> +	ret = btf__resolve_size(con->btf, ret);
> +	if (ret < 0) {
> +		pr_debug("cannot get size of struct zone: %d\n", ret);
> +		return;
> +	}
> +	skel->rodata->sizeof_zone = ret;
> +
> +	/* UMA system doesn't have 'node_data[]' - just use contig_page_data. */
> +	sym = machine__find_kernel_symbol_by_name(con->machine,
> +						  "contig_page_data",
> +						  &kmap);
> +	if (sym) {
> +		skel->rodata->contig_page_data_addr = map__unmap_ip(kmap, sym->start);
> +		map__put(kmap);
> +		return;
> +	}
> +
> +	/*
> +	 * The 'node_data' is an array of pointers to struct pglist_data.
> +	 * It needs to follow the pointer for each node in BPF to get the
> +	 * address of struct pglist_data and its zones.
> +	 */
> +	sym = machine__find_kernel_symbol_by_name(con->machine,
> +						  "node_data",
> +						  &kmap);
> +	if (sym == NULL)
> +		return;
> +
> +	skel->rodata->node_data_addr = map__unmap_ip(kmap, sym->start);
> +	map__put(kmap);
> +
> +	/* get the number of online nodes using the last node number + 1 */
> +	ret = sysfs__read_str("devices/system/node/online", &buf, &len);
> +	if (ret < 0) {
> +		pr_debug("failed to read online node: %d\n", ret);
> +		return;
> +	}
> +
> +	p = buf;
> +	while (p && *p) {
> +		last = strtol(p, &p, 0);
> +
> +		if (p && (*p == ',' || *p == '-' || *p == '\n'))
> +			p++;
> +	}
> +	skel->rodata->nr_nodes = last + 1;
> +	free(buf);
> +}
> +
>  int lock_contention_prepare(struct lock_contention *con)
>  {
>  	int i, fd;
> @@ -218,6 +286,8 @@ int lock_contention_prepare(struct lock_contention *con)
>  
>  	bpf_map__set_max_entries(skel->maps.slab_filter, nslabs);
>  
> +	init_numa_data(con);
> +
>  	if (lock_contention_bpf__load(skel) < 0) {
>  		pr_err("Failed to load lock-contention BPF skeleton\n");
>  		return -1;
> @@ -505,6 +575,11 @@ static const char *lock_contention_get_name(struct lock_contention *con,
>  				return "rq_lock";
>  		}
>  
> +		if (!bpf_map_lookup_elem(lock_fd, &key->lock_addr_or_cgroup, &flags)) {
> +			if (flags == LOCK_CLASS_ZONE_LOCK)
> +				return "zone_lock";
> +		}
> +
>  		/* look slab_hash for dynamic locks in a slab object */
>  		if (hashmap__find(&slab_hash, flags & LCB_F_SLAB_ID_MASK, &slab_data)) {
>  			snprintf(name_buf, sizeof(name_buf), "&%s", slab_data->name);
> @@ -743,6 +818,7 @@ int lock_contention_finish(struct lock_contention *con)
>  	}
>  
>  	exit_slab_cache_iter();
> +	btf__free(con->btf);
>  
>  	return 0;
>  }
> diff --git a/tools/perf/util/bpf_skel/lock_contention.bpf.c b/tools/perf/util/bpf_skel/lock_contention.bpf.c
> index 69be7a4234e076e8..6f12c7d978a2e015 100644
> --- a/tools/perf/util/bpf_skel/lock_contention.bpf.c
> +++ b/tools/perf/util/bpf_skel/lock_contention.bpf.c
> @@ -11,6 +11,9 @@
>  /* for collect_lock_syms().  4096 was rejected by the verifier */
>  #define MAX_CPUS  1024
>  
> +/* for collect_zone_lock().  It should be more than the actual zones. */
> +#define MAX_ZONES  10
> +
>  /* lock contention flags from include/trace/events/lock.h */
>  #define LCB_F_SPIN	(1U << 0)
>  #define LCB_F_READ	(1U << 1)
> @@ -801,6 +804,11 @@ int contention_end(u64 *ctx)
>  
>  extern struct rq runqueues __ksym;
>  
> +const volatile __u64 contig_page_data_addr;
> +const volatile __u64 node_data_addr;
> +const volatile int nr_nodes;
> +const volatile int sizeof_zone;
> +
>  struct rq___old {
>  	raw_spinlock_t lock;
>  } __attribute__((preserve_access_index));
> @@ -809,6 +817,59 @@ struct rq___new {
>  	raw_spinlock_t __lock;
>  } __attribute__((preserve_access_index));
>  
> +static void collect_zone_lock(void)
> +{
> +	__u64 nr_zones, zone_off;
> +	__u64 lock_addr, lock_off;
> +	__u32 lock_flag = LOCK_CLASS_ZONE_LOCK;
> +
> +	zone_off = offsetof(struct pglist_data, node_zones);
> +	lock_off = offsetof(struct zone, lock);
> +
> +	if (contig_page_data_addr) {
> +		struct pglist_data *contig_page_data;
> +
> +		contig_page_data = (void *)(long)contig_page_data_addr;
> +		nr_zones = BPF_CORE_READ(contig_page_data, nr_zones);
> +
> +		for (int i = 0; i < MAX_ZONES; i++) {
> +			__u64 zone_addr;
> +
> +			if (i >= nr_zones)
> +				break;
> +
> +			zone_addr = contig_page_data_addr + (sizeof_zone * i) + zone_off;
> +			lock_addr = zone_addr + lock_off;
> +
> +			bpf_map_update_elem(&lock_syms, &lock_addr, &lock_flag, BPF_ANY);
> +		}
> +	} else if (nr_nodes > 0) {
> +		struct pglist_data **node_data = (void *)(long)node_data_addr;
> +
> +		for (int i = 0; i < nr_nodes; i++) {
> +			struct pglist_data *pgdat = NULL;
> +			int err;
> +
> +			err = bpf_core_read(&pgdat, sizeof(pgdat), &node_data[i]);
> +			if (err < 0 || pgdat == NULL)
> +				break;
> +
> +			nr_zones = BPF_CORE_READ(pgdat, nr_zones);
> +			for (int k = 0; k < MAX_ZONES; k++) {
> +				__u64 zone_addr;
> +
> +				if (k >= nr_zones)
> +					break;
> +
> +				zone_addr = (__u64)(void *)pgdat + (sizeof_zone * k) + zone_off;
> +				lock_addr = zone_addr + lock_off;
> +
> +				bpf_map_update_elem(&lock_syms, &lock_addr, &lock_flag, BPF_ANY);
> +			}
> +		}
> +	}
> +}
> +
>  SEC("raw_tp/bpf_test_finish")
>  int BPF_PROG(collect_lock_syms)
>  {
> @@ -830,6 +891,9 @@ int BPF_PROG(collect_lock_syms)
>  		lock_flag = LOCK_CLASS_RQLOCK;
>  		bpf_map_update_elem(&lock_syms, &lock_addr, &lock_flag, BPF_ANY);
>  	}
> +
> +	collect_zone_lock();
> +
>  	return 0;
>  }
>  
> diff --git a/tools/perf/util/bpf_skel/lock_data.h b/tools/perf/util/bpf_skel/lock_data.h
> index 15f5743bd409f2f9..28c5e5aced7fcc91 100644
> --- a/tools/perf/util/bpf_skel/lock_data.h
> +++ b/tools/perf/util/bpf_skel/lock_data.h
> @@ -67,6 +67,7 @@ enum lock_aggr_mode {
>  enum lock_class_sym {
>  	LOCK_CLASS_NONE,
>  	LOCK_CLASS_RQLOCK,
> +	LOCK_CLASS_ZONE_LOCK,
>  };
>  
>  struct slab_cache_data {
> diff --git a/tools/perf/util/bpf_skel/vmlinux/vmlinux.h b/tools/perf/util/bpf_skel/vmlinux/vmlinux.h
> index 7b81d3173917fdb5..a59ce912be18cd0f 100644
> --- a/tools/perf/util/bpf_skel/vmlinux/vmlinux.h
> +++ b/tools/perf/util/bpf_skel/vmlinux/vmlinux.h
> @@ -203,4 +203,13 @@ struct bpf_iter__kmem_cache {
>  	struct kmem_cache *s;
>  } __attribute__((preserve_access_index));
>  
> +struct zone {
> +	spinlock_t lock;
> +} __attribute__((preserve_access_index));
> +
> +struct pglist_data {
> +	struct zone node_zones[6]; /* value for all possible config */
> +	int nr_zones;
> +} __attribute__((preserve_access_index));
> +
>  #endif // __VMLINUX_H
> diff --git a/tools/perf/util/lock-contention.h b/tools/perf/util/lock-contention.h
> index b5d916aa49df6424..d331ce8e3caad4cb 100644
> --- a/tools/perf/util/lock-contention.h
> +++ b/tools/perf/util/lock-contention.h
> @@ -142,6 +142,7 @@ struct lock_contention {
>  	struct lock_filter *filters;
>  	struct lock_contention_fails fails;
>  	struct rb_root cgroups;
> +	void *btf;
>  	unsigned long map_nr_entries;
>  	int max_stack;
>  	int stack_skip;
> -- 
> 2.49.0
> 

Powered by blists - more mailing lists