On Fri, Aug 25, 2023 at 12:49 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Fri, Aug 25, 2023 at 11:53 AM Alexei Starovoitov
> <alexei.starovoitov@xxxxxxxxx> wrote:
> >
> > On Fri, Aug 25, 2023 at 8:28 AM Yonghong Song <yonghong.song@xxxxxxxxx> wrote:
> > >
> > >
> > >
> > > On 8/25/23 3:32 AM, Björn Töpel wrote:
> > > > I'm chasing a workqueue hang on RISC-V/qemu (TCG), using the bpf
> > > > selftests on bpf-next 9e3b47abeb8f.
> > > >
> > > > I'm able to reproduce the hang by multiple runs of:
> > > > | ./test_progs -a link_api -a linked_list
> > > > I'm currently investigating that.
> > > >
> > > > But! Sometimes (every blue moon) I get a warn_on_once hit:
> > > > | ------------[ cut here ]------------
> > > > | WARNING: CPU: 3 PID: 261 at kernel/bpf/memalloc.c:342 bpf_mem_refill+0x1fc/0x206
> > > > | Modules linked in: bpf_testmod(OE)
> > > > | CPU: 3 PID: 261 Comm: test_progs-cpuv Tainted: G OE N 6.5.0-rc5-01743-gdcb152bb8328 #2
> > > > | Hardware name: riscv-virtio,qemu (DT)
> > > > | epc : bpf_mem_refill+0x1fc/0x206
> > > > | ra : irq_work_single+0x68/0x70
> > > > | epc : ffffffff801b1bc4 ra : ffffffff8015fe84 sp : ff2000000001be20
> > > > | gp : ffffffff82d26138 tp : ff6000008477a800 t0 : 0000000000046600
> > > > | t1 : ffffffff812b6ddc t2 : 0000000000000000 s0 : ff2000000001be70
> > > > | s1 : ff5ffffffffe8998 a0 : ff5ffffffffe8998 a1 : ff600003fef4b000
> > > > | a2 : 000000000000003f a3 : ffffffff80008250 a4 : 0000000000000060
> > > > | a5 : 0000000000000080 a6 : 0000000000000000 a7 : 0000000000735049
> > > > | s2 : ff5ffffffffe8998 s3 : 0000000000000022 s4 : 0000000000001000
> > > > | s5 : 0000000000000007 s6 : ff5ffffffffe8570 s7 : ffffffff82d6bd30
> > > > | s8 : 000000000000003f s9 : ffffffff82d2c5e8 s10: 000000000000ffff
> > > > | s11: ffffffff82d2c5d8 t3 : ffffffff81ea8f28 t4 : 0000000000000000
> > > > | t5 : ff6000008fd28278 t6 : 0000000000040000
> > > > | status: 0000000200000100 badaddr: 0000000000000000 cause: 0000000000000003
> > > > | [<ffffffff801b1bc4>] bpf_mem_refill+0x1fc/0x206
> > > > | [<ffffffff8015fe84>] irq_work_single+0x68/0x70
> > > > | [<ffffffff8015feb4>] irq_work_run_list+0x28/0x36
> > > > | [<ffffffff8015fefa>] irq_work_run+0x38/0x66
> > > > | [<ffffffff8000828a>] handle_IPI+0x3a/0xb4
> > > > | [<ffffffff800a5c3a>] handle_percpu_devid_irq+0xa4/0x1f8
> > > > | [<ffffffff8009fafa>] generic_handle_domain_irq+0x28/0x36
> > > > | [<ffffffff800ae570>] ipi_mux_process+0xac/0xfa
> > > > | [<ffffffff8000a8ea>] sbi_ipi_handle+0x2e/0x88
> > > > | [<ffffffff8009fafa>] generic_handle_domain_irq+0x28/0x36
> > > > | [<ffffffff807ee70e>] riscv_intc_irq+0x36/0x4e
> > > > | [<ffffffff812b5d3a>] handle_riscv_irq+0x54/0x86
> > > > | [<ffffffff812b6904>] do_irq+0x66/0x98
> > > > | ---[ end trace 0000000000000000 ]---
> > > >
> > > > Code:
> > > > | static void free_bulk(struct bpf_mem_cache *c)
> > > > | {
> > > > | struct bpf_mem_cache *tgt = c->tgt;
> > > > | struct llist_node *llnode, *t;
> > > > | unsigned long flags;
> > > > | int cnt;
> > > > |
> > > > | WARN_ON_ONCE(tgt->unit_size != c->unit_size);
> > > > | ...
> > > >
> > > > I'm not well versed in the memory allocator; Before I dive into it --
> > > > has anyone else hit it? Ideas on why the warn_on_once is hit?
> > >
> > > Maybe take a look at the patch
> > > 822fb26bdb55 bpf: Add a hint to allocated objects.
> > >
> > > In the above patch, we have
> > >
> > > + /*
> > > + * Remember bpf_mem_cache that allocated this object.
> > > + * The hint is not accurate.
> > > + */
> > > + c->tgt = *(struct bpf_mem_cache **)llnode;
> > >
> > > I suspect that the warning may be related to the above.
> > > I tried the above ./test_progs command line (running multiple
> > > at the same time) and didn't trigger the issue.
> >
> > Interesting. Thanks for the report.
> > I wasn't able to repo the warn either, but I can confirm that linked_list
> > test is super slow on the kernel with debugs on and may appear to "hang",
> > since it doesn't respond to key press for many seconds.
> >
> > time ./test_progs -a linked_list
> > Summary: 1/132 PASSED, 0 SKIPPED, 0 FAILED
> >
> > real 0m35.897s
> > user 0m1.860s
> > sys 0m32.105s
> >
> > That's not normal.
> >
> > Kumar, Dave,
> >
> > do you have any suggestions?
>
> Andrii,
>
> this issue seems to be a scalability issue with libbpf.
> The perf report looks like:
>
> 9.89% test_progs [kernel.kallsyms] [k] __lock_acquire
> 7.70% test_progs [kernel.kallsyms] [k] check_preemption_disabled
> 2.61% test_progs [kernel.kallsyms] [k] asm_exc_page_fault
> 2.32% test_progs [kernel.kallsyms] [k] rcu_is_watching
> 2.30% test_progs [kernel.kallsyms] [k] memcpy_orig
> 2.26% test_progs test_progs [.] btf_find_by_name_kind
> 2.18% test_progs libc-2.28.so [.] __strcmp_avx2
> 2.14% test_progs [kernel.kallsyms] [k] lock_acquire
> 2.10% test_progs [kernel.kallsyms] [k] mark_lock.part.48
> 2.07% test_progs test_progs [.] btf_kind
> 2.05% test_progs [kernel.kallsyms] [k] clear_page_erms
> 1.89% test_progs test_progs [.] btf_type_by_id
> 1.83% test_progs [kernel.kallsyms] [k] lock_is_held_type
> 1.75% test_progs [kernel.kallsyms] [k] lock_release
> 1.31% test_progs [kernel.kallsyms] [k] unwind_next_frame
> 1.29% test_progs test_progs [.] btf__type_by_id
> 1.28% test_progs [kernel.kallsyms] [k] rep_movs_alternative
> 1.09% test_progs [kernel.kallsyms] [k] __orc_find
> 1.06% swapper [kernel.kallsyms] [k] __lock_acquire
> 1.02% test_progs test_progs [.] btf_type_size
> 0.84% test_progs test_progs [.] btf_parse_type_sec
> 0.78% test_progs [kernel.kallsyms] [k] __create_object
> 0.76% test_progs [kernel.kallsyms] [k] __lookup_object
> 0.75% test_progs test_progs [.] btf__str_by_offset
>
> The top 5 functions coming from kernel due to libbpf
> reading /sys/kernel/btf/vmlinux
> Then libbpf proceeded to do a search in btf.
This test loads 132 skeletons/BPF objects sequentially. Each BPF
object load should cause exactly one /sys/kernel/btf/vmlinux load. Do
you see multiple vmlinux reads per BPF object?
We then do a linear search for each ksym, so depending on the amount
of ksyms, that can scale badly. But linked_list.bpf.o and
linked_list_fail.bpf.o have only a few ksyms:
[72] DATASEC '.ksyms' size=0 vlen=6
type_id=33 offset=0 size=0 (FUNC 'bpf_obj_new_impl')
type_id=37 offset=0 size=0 (FUNC 'bpf_list_pop_front')
type_id=39 offset=0 size=0 (FUNC 'bpf_obj_drop_impl')
type_id=41 offset=0 size=0 (FUNC 'bpf_list_pop_back')
type_id=43 offset=0 size=0 (FUNC 'bpf_list_push_front_impl')
type_id=51 offset=0 size=0 (FUNC 'bpf_list_push_back_impl')
We can improve this part by either pre-building hashmap index for BTF
types or sorting all ksyms and then doing one linear pass over vmlinux
BTF, binary searching. But so far it wasn't a big bottleneck in
practice.
And running this linked_list test on production kernel, it's not that
slow, considering how much work it's doing:
sudo ./test_progs -a linked_list 1.54s user 0.58s system 97% cpu 2.178 total
Also, we are building sefltests+libbpf in debug mode. So I have worse
news, with my BTF validation patch applied, this test becomes
extremely slow, because all those 100+ vmlinux BTF parsings are now
also doing a bunch of validation, and none of the added small static
functions are inlined in debug mode. I need to look into this a bit
more and see if I can mitigate this without switching all selftests to
release mode.
>
> Don't know whether libbpf is doing too many unnecessary reads on vmlinux btf
> or what is actually happening, but something isn't right.
> This test shouldn't be causing libbpf to be the main consumer of cpu.
> Not sure whether other tests are similarly affected.
> This is what I've debugged so far.
