linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Yang Jihong <yangjihong1@huawei.com>
To: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Andrii Nakryiko <andrii@kernel.org>,
	Martin KaFai Lau <martin.lau@linux.dev>,
	Song Liu <song@kernel.org>, Yonghong Song <yhs@fb.com>,
	John Fastabend <john.fastabend@gmail.com>,
	KP Singh <kpsingh@kernel.org>,
	Stanislav Fomichev <sdf@google.com>, Hao Luo <haoluo@google.com>,
	Jiri Olsa <jolsa@kernel.org>,
	Shubham Bansal <illusionist.neo@gmail.com>,
	Russell King <linux@armlinux.org.uk>,
	"David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>,
	Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
	Mykola Lysenko <mykolal@fb.com>, Shuah Khan <shuah@kernel.org>,
	Benjamin Tissoires <benjamin.tissoires@redhat.com>,
	Kumar Kartikeya Dwivedi <memxor@gmail.com>,
	<colin.i.king@gmail.com>, Artem Savkov <asavkov@redhat.com>,
	Delyan Kratunov <delyank@fb.com>, bpf <bpf@vger.kernel.org>,
	linux-arm-kernel <linux-arm-kernel@lists.infradead.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Network Development <netdev@vger.kernel.org>,
	"open list:KERNEL SELFTEST FRAMEWORK" 
	<linux-kselftest@vger.kernel.org>
Subject: Re: [PATCH bpf-next v3 1/4] bpf: Adapt 32-bit return value kfunc for 32-bit ARM when zext extension
Date: Mon, 5 Dec 2022 09:19:40 +0800	[thread overview]
Message-ID: <4a2b8cd5-78c4-360a-6eb0-33fcf689d26a@huawei.com> (raw)
In-Reply-To: <CAADnVQJXr6XxpG2E-AkO7__qg-sujrhyO+JWWa1iwYmAO4S0Pw@mail.gmail.com>



On 2022/12/4 0:40, Alexei Starovoitov wrote:
> On Fri, Dec 2, 2022 at 6:58 PM Yang Jihong <yangjihong1@huawei.com> wrote:
>>
>>
>>
>> On 2022/11/29 0:41, Alexei Starovoitov wrote:
>>> On Mon, Nov 28, 2022 at 4:40 AM Yang Jihong <yangjihong1@huawei.com> wrote:
>>>>
>>>>
>>>>
>>>> On 2022/11/28 9:57, Alexei Starovoitov wrote:
>>>>> On Sat, Nov 26, 2022 at 05:45:27PM +0800, Yang Jihong wrote:
>>>>>> For ARM32 architecture, if data width of kfunc return value is 32 bits,
>>>>>> need to do explicit zero extension for high 32-bit, insn_def_regno should
>>>>>> return dst_reg for BPF_JMP type of BPF_PSEUDO_KFUNC_CALL. Otherwise,
>>>>>> opt_subreg_zext_lo32_rnd_hi32 returns -EFAULT, resulting in BPF failure.
>>>>>>
>>>>>> Signed-off-by: Yang Jihong <yangjihong1@huawei.com>
>>>>>> ---
>>>>>>     kernel/bpf/verifier.c | 44 ++++++++++++++++++++++++++++++++++++++++---
>>>>>>     1 file changed, 41 insertions(+), 3 deletions(-)
>>>>>>
>>>>>> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>>>>>> index 264b3dc714cc..193ea927aa69 100644
>>>>>> --- a/kernel/bpf/verifier.c
>>>>>> +++ b/kernel/bpf/verifier.c
>>>>>> @@ -1927,6 +1927,21 @@ find_kfunc_desc(const struct bpf_prog *prog, u32 func_id, u16 offset)
>>>>>>                       sizeof(tab->descs[0]), kfunc_desc_cmp_by_id_off);
>>>>>>     }
>>>>>>
>>>>>> +static int kfunc_desc_cmp_by_imm(const void *a, const void *b);
>>>>>> +
>>>>>> +static const struct bpf_kfunc_desc *
>>>>>> +find_kfunc_desc_by_imm(const struct bpf_prog *prog, s32 imm)
>>>>>> +{
>>>>>> +    struct bpf_kfunc_desc desc = {
>>>>>> +            .imm = imm,
>>>>>> +    };
>>>>>> +    struct bpf_kfunc_desc_tab *tab;
>>>>>> +
>>>>>> +    tab = prog->aux->kfunc_tab;
>>>>>> +    return bsearch(&desc, tab->descs, tab->nr_descs,
>>>>>> +                   sizeof(tab->descs[0]), kfunc_desc_cmp_by_imm);
>>>>>> +}
>>>>>> +
>>>>>>     static struct btf *__find_kfunc_desc_btf(struct bpf_verifier_env *env,
>>>>>>                                         s16 offset)
>>>>>>     {
>>>>>> @@ -2342,6 +2357,13 @@ static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn,
>>>>>>                         */
>>>>>>                        if (insn->src_reg == BPF_PSEUDO_CALL)
>>>>>>                                return false;
>>>>>> +
>>>>>> +                    /* Kfunc call will reach here because of insn_has_def32,
>>>>>> +                     * conservatively return TRUE.
>>>>>> +                     */
>>>>>> +                    if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL)
>>>>>> +                            return true;
>>>>>> +
>>>>>>                        /* Helper call will reach here because of arg type
>>>>>>                         * check, conservatively return TRUE.
>>>>>>                         */
>>>>>> @@ -2405,10 +2427,26 @@ static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn,
>>>>>>     }
>>>>>>
>>>>>>     /* Return the regno defined by the insn, or -1. */
>>>>>> -static int insn_def_regno(const struct bpf_insn *insn)
>>>>>> +static int insn_def_regno(struct bpf_verifier_env *env, const struct bpf_insn *insn)
>>>>>>     {
>>>>>>        switch (BPF_CLASS(insn->code)) {
>>>>>>        case BPF_JMP:
>>>>>> +            if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) {
>>>>>> +                    const struct bpf_kfunc_desc *desc;
>>>>>> +
>>>>>> +                    /* The value of desc cannot be NULL */
>>>>>> +                    desc = find_kfunc_desc_by_imm(env->prog, insn->imm);
>>>>>> +
>>>>>> +                    /* A kfunc can return void.
>>>>>> +                     * The btf type of the kfunc's return value needs
>>>>>> +                     * to be checked against "void" first
>>>>>> +                     */
>>>>>> +                    if (desc->func_model.ret_size == 0)
>>>>>> +                            return -1;
>>>>>> +                    else
>>>>>> +                            return insn->dst_reg;
>>>>>> +            }
>>>>>> +            fallthrough;
>>>>>
>>>>> I cannot make any sense of this patch.
>>>>> insn->dst_reg above is 0.
>>>>> The kfunc call doesn't define a register from insn_def_regno() pov.
>>>>>
>>>>> Are you hacking insn_def_regno() to return 0 so that
>>>>> if (WARN_ON(load_reg == -1)) {
>>>>>      verbose(env, "verifier bug. zext_dst is set, but no reg is defined\n");
>>>>>      return -EFAULT;
>>>>> }
>>>>> in opt_subreg_zext_lo32_rnd_hi32() doesn't trigger ?
>>>>>
>>>>> But this verifier message should have been a hint that you need
>>>>> to analyze why zext_dst is set on this kfunc call.
>>>>> Maybe it shouldn't ?
>>>>> Did you analyze the logic of mark_btf_func_reg_size() ?
>>>> make r0 zext is not caused by mark_btf_func_reg_size.
>>>>
>>>> This problem occurs when running the kfunc_call_test_ref_btf_id test
>>>> case in the 32-bit ARM environment.
>>>
>>> Why is it not failing on x86-32 ?
>> Use the latest mainline kernel code to test on the x86_32 machine. The
>> test also fails:
>>
>>     # ./test_progs -t kfunc_call/kfunc_call_test_ref_btf_id
>>     Failed to load bpf_testmod.ko into the kernel: -8
>>     WARNING! Selftests relying on bpf_testmod.ko will be skipped.
>>     libbpf: prog 'kfunc_call_test_ref_btf_id': BPF program load failed:
>> Bad address
>>     libbpf: prog 'kfunc_call_test_ref_btf_id': -- BEGIN PROG LOAD LOG --
>>     processed 25 insns (limit 1000000) max_states_per_insn 0 total_states
>> 2 peak_states 2 mark_read 1
>>     -- END PROG LOAD LOG --
>>     libbpf: prog 'kfunc_call_test_ref_btf_id': failed to load: -14
>>     libbpf: failed to load object 'kfunc_call_test'
>>     libbpf: failed to load BPF skeleton 'kfunc_call_test': -14
>>     verify_success:FAIL:skel unexpected error: -14
>>
>> Therefore, this problem also exists on x86_32:
>> "verifier bug. zext_dst is set, but no reg is defined"
> 
> The kernel returns -14 == EFAULT.
> That's a completely different issue.
It's the same problem. The opt_subreg_zext_lo32_rnd_hi32 function fails 
to check here and returns -EFAULT

opt_subreg_zext_lo32_rnd_hi32 {
   ...
    if (WARN_ON(load_reg == -1)) { 

            verbose(env, "verifier bug. zext_dst is set, but no reg is 
defined\n");
            return -EFAULT; 

    }
   ...
} 

> .
> 

  reply	other threads:[~2022-12-05  1:19 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-11-26  9:45 [PATCH bpf-next v3 0/4] bpf: Support kernel function call in 32-bit ARM Yang Jihong
2022-11-26  9:45 ` [PATCH bpf-next v3 1/4] bpf: Adapt 32-bit return value kfunc for 32-bit ARM when zext extension Yang Jihong
2022-11-28  1:57   ` Alexei Starovoitov
2022-11-28 12:40     ` Yang Jihong
2022-11-28 16:41       ` Alexei Starovoitov
2022-12-03  2:58         ` Yang Jihong
2022-12-03 16:40           ` Alexei Starovoitov
2022-12-05  1:19             ` Yang Jihong [this message]
2022-12-07  8:49               ` Yang Jihong
2022-12-02  4:11   ` kernel test robot
2022-11-26  9:45 ` [PATCH bpf-next v3 2/4] bpf: Add kernel function call support in 32-bit ARM for EABI Yang Jihong
2022-11-26  9:45 ` [PATCH bpf-next v3 3/4] bpf:selftests: Add kfunc_call test for mixing 32-bit and 64-bit parameters Yang Jihong
2022-11-26  9:45 ` [PATCH bpf-next v3 4/4] bpf: Fix comment error in fixup_kfunc_call function Yang Jihong

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=4a2b8cd5-78c4-360a-6eb0-33fcf689d26a@huawei.com \
    --to=yangjihong1@huawei.com \
    --cc=alexei.starovoitov@gmail.com \
    --cc=andrii@kernel.org \
    --cc=asavkov@redhat.com \
    --cc=ast@kernel.org \
    --cc=benjamin.tissoires@redhat.com \
    --cc=bpf@vger.kernel.org \
    --cc=colin.i.king@gmail.com \
    --cc=daniel@iogearbox.net \
    --cc=davem@davemloft.net \
    --cc=delyank@fb.com \
    --cc=edumazet@google.com \
    --cc=haoluo@google.com \
    --cc=illusionist.neo@gmail.com \
    --cc=john.fastabend@gmail.com \
    --cc=jolsa@kernel.org \
    --cc=kpsingh@kernel.org \
    --cc=kuba@kernel.org \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-kselftest@vger.kernel.org \
    --cc=linux@armlinux.org.uk \
    --cc=martin.lau@linux.dev \
    --cc=memxor@gmail.com \
    --cc=mykolal@fb.com \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    --cc=sdf@google.com \
    --cc=shuah@kernel.org \
    --cc=song@kernel.org \
    --cc=yhs@fb.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).