On Fri, Dec 16, 2022 at 12:10 PM Yonghong Song <yhs@xxxxxxxx> wrote:
>
>
>
> On 12/14/22 2:38 AM, xiangxia.m.yue@xxxxxxxxx wrote:
> > From: Tonghao Zhang <xiangxia.m.yue@xxxxxxxxx>
> >
> > This testing show how to reproduce deadlock in special case.
> >
> > Signed-off-by: Tonghao Zhang <xiangxia.m.yue@xxxxxxxxx>
> > Cc: Alexei Starovoitov <ast@xxxxxxxxxx>
> > Cc: Daniel Borkmann <daniel@xxxxxxxxxxxxx>
> > Cc: Andrii Nakryiko <andrii@xxxxxxxxxx>
> > Cc: Martin KaFai Lau <martin.lau@xxxxxxxxx>
> > Cc: Song Liu <song@xxxxxxxxxx>
> > Cc: Yonghong Song <yhs@xxxxxx>
> > Cc: John Fastabend <john.fastabend@xxxxxxxxx>
> > Cc: KP Singh <kpsingh@xxxxxxxxxx>
> > Cc: Stanislav Fomichev <sdf@xxxxxxxxxx>
> > Cc: Hao Luo <haoluo@xxxxxxxxxx>
> > Cc: Jiri Olsa <jolsa@xxxxxxxxxx>
> > Cc: Hou Tao <houtao1@xxxxxxxxxx>
> > ---
> > .../selftests/bpf/prog_tests/htab_deadlock.c | 74 +++++++++++++++++++
> > .../selftests/bpf/progs/htab_deadlock.c | 30 ++++++++
> > 2 files changed, 104 insertions(+)
> > create mode 100644 tools/testing/selftests/bpf/prog_tests/htab_deadlock.c
> > create mode 100644 tools/testing/selftests/bpf/progs/htab_deadlock.c
> >
> > diff --git a/tools/testing/selftests/bpf/prog_tests/htab_deadlock.c b/tools/testing/selftests/bpf/prog_tests/htab_deadlock.c
> > new file mode 100644
> > index 000000000000..7dce4c2fe4f5
> > --- /dev/null
> > +++ b/tools/testing/selftests/bpf/prog_tests/htab_deadlock.c
> > @@ -0,0 +1,74 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/* Copyright (c) 2022 DiDi Global Inc. */
> > +#define _GNU_SOURCE
> > +#include <pthread.h>
> > +#include <sched.h>
> > +#include <test_progs.h>
> > +
> > +#include "htab_deadlock.skel.h"
> > +
> > +static int perf_event_open(void)
> > +{
> > + struct perf_event_attr attr = {0};
> > + int pfd;
> > +
> > + /* create perf event */
> > + attr.size = sizeof(attr);
> > + attr.type = PERF_TYPE_HARDWARE;
> > + attr.config = PERF_COUNT_HW_CPU_CYCLES;
> > + attr.freq = 1;
> > + attr.sample_freq = 1000;
> > + pfd = syscall(__NR_perf_event_open, &attr, -1, 0, -1, PERF_FLAG_FD_CLOEXEC);
> > +
> > + return pfd >= 0 ? pfd : -errno;
> > +}
> > +
> > +void test_htab_deadlock(void)
> > +{
> > + unsigned int val = 0, key = 20;
> > + struct bpf_link *link = NULL;
> > + struct htab_deadlock *skel;
> > + cpu_set_t cpus;
> > + int err;
> > + int pfd;
> > + int i;
>
> No need to have three lines for type 'int' variables. One line
> is enough to hold all three variables.
>
> > +
> > + skel = htab_deadlock__open_and_load();
> > + if (!ASSERT_OK_PTR(skel, "skel_open_and_load"))
> > + return;
> > +
> > + err = htab_deadlock__attach(skel);
> > + if (!ASSERT_OK(err, "skel_attach"))
> > + goto clean_skel;
> > +
> > + /* NMI events. */
> > + pfd = perf_event_open();
> > + if (pfd < 0) {
> > + if (pfd == -ENOENT || pfd == -EOPNOTSUPP) {
> > + printf("%s:SKIP:no PERF_COUNT_HW_CPU_CYCLES\n", __func__);
> > + test__skip();
> > + goto clean_skel;
> > + }
> > + if (!ASSERT_GE(pfd, 0, "perf_event_open"))
> > + goto clean_skel;
> > + }
> > +
> > + link = bpf_program__attach_perf_event(skel->progs.bpf_perf_event, pfd);
> > + if (!ASSERT_OK_PTR(link, "attach_perf_event"))
> > + goto clean_pfd;
> > +
> > + /* Pinned on CPU 0 */
> > + CPU_ZERO(&cpus);
> > + CPU_SET(0, &cpus);
> > + pthread_setaffinity_np(pthread_self(), sizeof(cpus), &cpus);
> > +
> > + for (i = 0; i < 100000; i++)
>
> Please add some comments in the above loop to mention the test
> expects (hopefully) duriing one of bpf_map_update_elem(), one
> perf event might kick to trigger prog bpf_nmi_handle run.
>
> > + bpf_map_update_elem(bpf_map__fd(skel->maps.htab),
> > + &key, &val, BPF_ANY);
> > +
> > + bpf_link__destroy(link);
> > +clean_pfd:
> > + close(pfd);
> > +clean_skel:
> > + htab_deadlock__destroy(skel);
> > +}
> > diff --git a/tools/testing/selftests/bpf/progs/htab_deadlock.c b/tools/testing/selftests/bpf/progs/htab_deadlock.c
> > new file mode 100644
> > index 000000000000..c4bd1567f882
> > --- /dev/null
> > +++ b/tools/testing/selftests/bpf/progs/htab_deadlock.c
> > @@ -0,0 +1,30 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/* Copyright (c) 2022 DiDi Global Inc. */
> > +#include <linux/bpf.h>
> > +#include <bpf/bpf_helpers.h>
> > +#include <bpf/bpf_tracing.h>
> > +
> > +char _license[] SEC("license") = "GPL";
> > +
> > +struct {
> > + __uint(type, BPF_MAP_TYPE_HASH);
> > + __uint(max_entries, 2);
> > + __uint(map_flags, BPF_F_ZERO_SEED);
> > + __uint(key_size, sizeof(unsigned int));
> > + __uint(value_size, sizeof(unsigned int));
> > +} htab SEC(".maps");
>
> You can use
> __type(key, unsigned int);
> __type(value, unsigned int);
> This is more expressive.
>
> > +
> > +SEC("fentry/nmi_handle")
> > +int bpf_nmi_handle(struct pt_regs *regs)
>
> Do we need this fentry function? Can be just put
> bpf_map_update_elem() into bpf_perf_event program?
Hi
bpf_overflow_handler will check the bpf_prog_active, and
bpf_map_update_value invokes bpf_disable_instrumentation,
so the deadlock will not occur. In fentry/nmi_handle, bpf does not
check the bpf_prog_active.
Other comments look good to me, I will send v2 soon.
> Also s390x and aarch64 failed the test due to none/incomplete trampoline
> support. See bpf ci https://github.com/kernel-patches/bpf/pull/4211.
> You need to add them in their corresponding deny list if this fentry
> bpf program is used.
>
> > +{
> > + unsigned int val = 0, key = 4;
> > +
> > + bpf_map_update_elem(&htab, &key, &val, BPF_ANY);
> > + return 0;
> > +}
> > +
> > +SEC("perf_event")
> > +int bpf_perf_event(struct pt_regs *regs)
> > +{
> > + return 0;
> > +}
--
Best regards, Tonghao
