From: Yonghong Song <yhs@fb.com>
To: Lorenz Bauer <lmb@cloudflare.com>, Andrii Nakryiko <andrii@kernel.org>
Cc: bpf <bpf@vger.kernel.org>
Subject: Re: CO-RE: Weird immediate for bpf_core_field_exists
Date: Thu, 29 Apr 2021 08:31:55 -0700 [thread overview]
Message-ID: <eb912b37-d64b-a9e9-7010-7f7da3015479@fb.com> (raw)
In-Reply-To: <CACAyw99n-cMEtVst7aK-3BfHb99GMEChmRLCvhrjsRpHhPrtvA@mail.gmail.com>
On 4/29/21 3:29 AM, Lorenz Bauer wrote:
> Hi Andrii and Yonghong,
>
> This is probably a case of me holding it wrong, but I figured I would
> share this nonetheless. Given the following C:
>
> struct s {
> int _1;
> char _2;
> };
>
> typedef struct s s_t;
>
> union u {
> int *_1;
> char *_2;
> };
>
> __section("socket_filter/fields") int fields() {
> struct t {
> union {
> s_t s[10];
> };
> struct {
> union u u;
> };
> } bar;
> return bpf_core_field_exists((&bar)[1]);
> }
>
> clang-12 generates the following instructions:
>
> 0000000000000000 <fields>:
> ; return bpf_core_field_exists((&bar)[1]);
> 0: b7 00 00 00 58 00 00 00 r0 = 88
> 1: 95 00 00 00 00 00 00 00 exit
>
> The weird bit is that the immediate for instruction 0 isn't 1 but 88.
> Coincidentally sizeof(bar) is also 88 bytes.
Thanks for the reporting. This is a compiler issue which didn't handle
invalid case nicely. The following is an explanation.
After macro expansion,
bpf_core_field_exists((&bar)[1]);
is actually
__builtin_preserve_field_info((&bar)[1], BPF_FIELD_EXISTS);
For BPF_FIELD_EXISTS, the first argument should be a field access.
But in the above, we got an array access, (&bar)[1], internally,
the compiler keeps track of the offset from the base address, and
this offset is used for most other builtin kinds like
FIELD_BYPE_OFFSET etc.
For relative to &bar, (&bar)[1] has an offset 88. In your
particular case, the code never went inside the routine to
generate correct "patch_imm" (i.e., r0 = <patch_imm> in the above)
since we didn't get a field access. So we got a wrong result.
I will fix this in the compiler by issuing an error so people
can correct their usage. Thanks for reporting!
>
> $ clang-12 --version
> Ubuntu clang version
> 12.0.0-++20210126113614+510b3d4b3e02-1~exp1~20210126104320.178
>
> I've tried clang-13 as well, same result.
>
> Best,
> Lorenz
>
prev parent reply other threads:[~2021-04-29 15:32 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-04-29 10:29 CO-RE: Weird immediate for bpf_core_field_exists Lorenz Bauer
2021-04-29 15:31 ` Yonghong Song [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=eb912b37-d64b-a9e9-7010-7f7da3015479@fb.com \
--to=yhs@fb.com \
--cc=andrii@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=lmb@cloudflare.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).