Re: [PATCH bpf-next v3] selftests/bpf: trace_helpers.c: optimize kallsyms cache

[Daniel Borkmann]

On 8/12/23 7:57 AM, Rong Tao wrote:
> From: Rong Tao <rongtao@xxxxxxxx>
>
> Static ksyms often have problems because the number of symbols exceeds the
> MAX_SYMS limit. Like changing the MAX_SYMS from 300000 to 400000 in
> commit e76a014334a6("selftests/bpf: Bump and validate MAX_SYMS") solves
> the problem somewhat, but it's not the perfect way.
>
> This commit uses dynamic memory allocation, which completely solves the
> problem caused by the limitation of the number of kallsyms.
>
> Signed-off-by: Rong Tao <rongtao@xxxxxxxx>
> ---
> v3: Do not use structs and judge ksyms__add_symbol function return value.
> v2: https://lore.kernel.org/lkml/tencent_B655EE5E5D463110D70CD2846AB3262EED09@xxxxxx/
>      Do the usual len/capacity scheme here to amortize the cost of realloc, and
>      don't free symbols.
> v1: https://lore.kernel.org/lkml/tencent_AB461510B10CD484E0B2F62E3754165F2909@xxxxxx/
> ---
>   tools/testing/selftests/bpf/trace_helpers.c | 42 ++++++++++++++++-----
>   1 file changed, 32 insertions(+), 10 deletions(-)
>
> diff --git a/tools/testing/selftests/bpf/trace_helpers.c b/tools/testing/selftests/bpf/trace_helpers.c
> index f83d9f65c65b..d8391a2122b4 100644
> --- a/tools/testing/selftests/bpf/trace_helpers.c
> +++ b/tools/testing/selftests/bpf/trace_helpers.c
> @@ -18,10 +18,32 @@
>   #define TRACEFS_PIPE	"/sys/kernel/tracing/trace_pipe"
>   #define DEBUGFS_PIPE	"/sys/kernel/debug/tracing/trace_pipe"
>   
> -#define MAX_SYMS 400000
> -static struct ksym syms[MAX_SYMS];
> +static struct ksym *syms;
> +static int sym_cap;
>   static int sym_cnt;
>   
> +static int ksyms__add_symbol(const char *name, unsigned long addr)
> +{
> +	void *tmp;
> +	unsigned int new_cap;
> +
> +	if (sym_cnt + 1 > sym_cap) {
> +		new_cap = sym_cap * 4 / 3;
> +		tmp = realloc(syms, sizeof(struct ksym) * new_cap);
> +		if (!tmp)
> +			return -ENOMEM;
> +		syms = tmp;
> +		sym_cap = new_cap;
> +	}
> +
> +	syms[sym_cnt].addr = addr;
> +	syms[sym_cnt].name = strdup(name);

Fwiw, strdup() should error check too.. and for teardown in the test suite, lets
also have the counterpart where we release all the allocated mem.

> +	sym_cnt++;
> +
> +	return 0;
> +}
> +
>   static int ksym_cmp(const void *p1, const void *p2)
>   {
>   	return ((struct ksym *)p1)->addr - ((struct ksym *)p2)->addr;
> @@ -33,9 +55,13 @@ int load_kallsyms_refresh(void)
>   	char func[256], buf[256];
>   	char symbol;
>   	void *addr;
> -	int i = 0;
> +	int ret;
>   
> +	sym_cap = 1024;

On my dev node, I have:

  # cat /proc/kallsyms | wc -l
  242586

Why starting out so low with 1k? I would have expected that for most cases we
don't need the realloc() path to begin with, but just in corner cases like in
e76a014334a6.

>   	sym_cnt = 0;
> +	syms = malloc(sizeof(struct ksym) * sym_cap);
> +	if (!syms)
> +		return -ENOMEM;
>   
>   	f = fopen("/proc/kallsyms", "r");
>   	if (!f)
> @@ -46,15 +72,11 @@ int load_kallsyms_refresh(void)
>   			break;
>   		if (!addr)
>   			continue;
> -		if (i >= MAX_SYMS)
> -			return -EFBIG;
> -
> -		syms[i].addr = (long) addr;
> -		syms[i].name = strdup(func);
> -		i++;
> +		ret = ksyms__add_symbol(func, (unsigned long)addr);
> +		if (ret)
> +			return ret;
>   	}
>   	fclose(f);
> -	sym_cnt = i;
>   	qsort(syms, sym_cnt, sizeof(struct ksym), ksym_cmp);
>   	return 0;
>   }