Re: [PATCH bpf-next v8] selftests/bpf: trace_helpers.c: optimize kallsyms cache

From: Jiri Olsa <olsajiri@gmail.com>
To: Rong Tao <rtoax@foxmail.com>
Cc: olsajiri@gmail.com, alexandre.torgue@foss.st.com,
	andrii@kernel.org, ast@kernel.org, bpf@vger.kernel.org,
	chantr4@gmail.com, daniel@iogearbox.net, deso@posteo.net,
	eddyz87@gmail.com, haoluo@google.com, iii@linux.ibm.com,
	john.fastabend@gmail.com, kpsingh@kernel.org,
	laoar.shao@gmail.com, linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, linux-kselftest@vger.kernel.org,
	linux-stm32@st-md-mailman.stormreply.com, martin.lau@linux.dev,
	mcoquelin.stm32@gmail.com, mykolal@fb.com, rongtao@cestc.cn,
	sdf@google.com, shuah@kernel.org, song@kernel.org,
	xukuohai@huawei.com, yonghong.song@linux.dev, zwisler@google.com
Subject: Re: [PATCH bpf-next v8] selftests/bpf: trace_helpers.c: optimize kallsyms cache
Date: Mon, 28 Aug 2023 15:54:18 +0200	[thread overview]
Message-ID: <ZOynCqBv/JJyZ2Kj@krava> (raw)
In-Reply-To: <tencent_D53295A257B55119C425836EA19E2CE84B07@qq.com>

On Mon, Aug 28, 2023 at 08:57:21AM +0800, Rong Tao wrote:
> Hi, jirka. Thanks for your reply.
> 
> > > @@ -164,13 +165,14 @@ int main(int argc, char **argv)
> > >  	}
> > >  
> > >  	/* initialize kernel symbol translation */
> > > -	if (load_kallsyms()) {
> > > +	ksyms = load_kallsyms();
> > 
> > if we keep the load_kallsyms/ksym_search/ksym_get_addr functions as described
> > in [1] the samples/bpf would stay untouched apart from the Makefile change
> 
> Maybe we should make this modification, wouldn't it be better? After all,
> not modifying the source code of samples/bpf is not really a reason not to
> make modifications to load_kallsyms(), what do you think?

I think we want separate selftest and samples changes and I don't see
other way to do that

> 
> In addition, if we continue to keep the original ksym_search() interface,
> the following problems are very difficult to deal with:
> 
> 	Source code ksym_search [1]
> 
>     struct ksym *ksym_search(long key)
>     {
>     	int start = 0, end = sym_cnt;
>     	int result;
>     
>     	/* kallsyms not loaded. return NULL */
>     	if (sym_cnt <= 0)
>     		return NULL;
>     
>     	while (start < end) {
>     		size_t mid = start + (end - start) / 2;
>     
>     		result = key - syms[mid].addr;
>     		if (result < 0)
>     			end = mid;
>     		else if (result > 0)
>     			start = mid + 1;
>     		else
>     			return &syms[mid];                         <<<
>     	}
>     
>     	if (start >= 1 && syms[start - 1].addr < key &&
>     	    key < syms[start].addr)
>     		/* valid ksym */
>     		return &syms[start - 1];                       <<<
>     
>     	/* out of range. return _stext */
>     	return &syms[0];                                   <<<
>     }
> 
> The original ksym_search() interface directly returns the global syms 
> address, which is also dangerous for multi-threading. If we allocate new
> memory for this, it is not a perfect solution.

the assumption was that the original ksym_search touches the global
syms allocated before running tests.. then tests that actually need
fresh kallsyms data (because of bpf_testmod load/unload) would get
their own copy of kallsyms

jirka

> 
> If we rewrite
> 
> 	struct ksym *ksym_search(long key)
> 
> to
> 	struct ksym ksym_search(long key)
> 
> This also affects the source code in samples/bpf.
> 
> The same problem exists with ksym_get_addr().
> 
> Best wishes,
> Rong Tao
> 
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/tree/tools/testing/selftests/bpf/trace_helpers.c#n100
> 
> 

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel