Thread (10 messages) 10 messages, 4 authors, 2021-05-28

Re: [PATCH dwarves] btf_encoder: fix and complete filtering out zero-sized per-CPU variables

From: Andrii Nakryiko <hidden>
Date: 2021-05-27 16:45:22

On Thu, May 27, 2021 at 8:37 AM Michal Suchánek [off-list ref] wrote:
Hello,

On Thu, May 27, 2021 at 11:58:24AM -0300, Arnaldo Carvalho de Melo wrote:
quoted
Em Thu, May 27, 2021 at 11:55:10AM -0300, Arnaldo Carvalho de Melo escreveu:
quoted
Em Mon, May 24, 2021 at 04:42:22PM -0700, Andrii Nakryiko escreveu:
quoted
btf_encoder is ignoring zero-sized per-CPU ELF symbols, but the same has to be
done for DWARF variables when matching them with ELF symbols. This is due to
zero-sized DWARF variables matching unrelated (non-zero-sized) variable that
happens to be allocated at the exact same address, leading to a lot of
confusion in BTF.
quoted
See [0] for when this causes big problems.
quoted
  [0] https://lore.kernel.org/bpf/CAEf4BzZ0-sihSL-UAm21JcaCCY92CqfNxycHRZYXcoj8OYb=wA@mail.gmail.com/ (local)
I also added this:

Reported-by: Michal Suchánek <redacted>

Michal, so you tested this patch and verified it fixed the problem? If
so please let me know so that I also add:
This is the first time I see this patch.
I've posted a link to it in your thread [0].

  [0] https://lore.kernel.org/bpf/CAEf4BzZ9=aLVD7ytgCcSxcbOLqFNK-p1mj14Rv_TGnOyL3aO_g@mail.gmail.com/ (local)
Given that linux-next does not build for me at the moment
I don't think I will test it soon.
This patch applied to pahole master will fix linux-next build.
Thanks

Michal
quoted
Tested-by: Michal Suchánek <redacted>

Thanks,

- Arnaldo
quoted
quoted
+++ b/btf_encoder.c
@@ -550,6 +551,7 @@ int cu__encode_btf(struct cu *cu, int verbose, bool force,

          /* addr has to be recorded before we follow spec */
          addr = var->ip.addr;
+         dwarf_name = variable__name(var, cu);

          /* DWARF takes into account .data..percpu section offset
           * within its segment, which for vmlinux is 0, but for kernel
@@ -582,11 +584,9 @@ int cu__encode_btf(struct cu *cu, int verbose, bool force,
           *  modules per-CPU data section has non-zero offset so all
           *  per-CPU symbols have non-zero values.
           */
-         if (var->ip.addr == 0) {
-                 dwarf_name = variable__name(var, cu);
+         if (var->ip.addr == 0)
                  if (!dwarf_name || strcmp(dwarf_name, name))
                          continue;
-         }

          if (var->spec)
                  var = var->spec;
@@ -600,6 +600,13 @@ int cu__encode_btf(struct cu *cu, int verbose, bool force,
I just changed the above hunk to be:
@@ -583,7 +585,6 @@ int cu__encode_btf(struct cu *cu, int verbose, bool force,
                 *  per-CPU symbols have non-zero values.
                 */
                if (var->ip.addr == 0) {
-                       dwarf_name = variable__name(var, cu);
                        if (!dwarf_name || strcmp(dwarf_name, name))
                                continue;
                }

Which is shorter and keeps the {} around a multi line if block, ok?

Thanks, applied!

- Arnaldo
--

- Arnaldo
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help