Luis Chamberlain [off-list ref] writes:

On Mon, May 09, 2022 at 09:13:03AM -0700, Luis Chamberlain wrote:

quoted

On Mon, May 09, 2022 at 09:23:39PM +1000, Michael Ellerman wrote:

quoted

Herbert Xu [off-list ref] writes:

quoted

Hi:

There are some code paths in the kernel where you can reliably
trigger a request_module of a non-existant module.  For example,
if you attempt to load a non-existent crypto algorithm, or create
a socket of a non-existent network family, it will result in a
request_module call that is guaranteed to fail.

As user-space can do this repeatedly, it can quickly overwhelm
the concurrency limit in kmod.  This in itself is expected,
however, at least on some platforms this appears to result in
a live-lock.  Here is an example triggered by stress-ng on ppc64:

[  529.853264] request_module: kmod_concurrent_max (0) close to 0 (max_modprobes: 50), for module crypto-aegis128l, throttling...

...

quoted

[  580.414590] __request_module: 25 callbacks suppressed
[  580.414597] request_module: kmod_concurrent_max (0) close to 0 (max_modprobes: 50), for module crypto-aegis256-all, throttling...
[  580.423082] watchdog: CPU 784 self-detected hard LOCKUP @ plpar_hcall_norets_notrace+0x18/0x2c
[  580.423097] watchdog: CPU 784 TB:1297691958559475, last heartbeat TB:1297686321743840 (11009ms ago)
[  580.423099] Modules linked in: cast6_generic cast5_generic cast_common camellia_generic blowfish_generic blowfish_common tun nft_fib_inet nft_fib_ipv4 nft_fib_ipv6 nft_fib nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 rfkill bonding tls ip_set nf_tables nfnetlink pseries_rng binfmt_misc drm drm_panel_orientation_quirks xfs libcrc32c sd_mod t10_pi sg ibmvscsi ibmveth scsi_transport_srp vmx_crypto dm_mirror dm_region_hash dm_log dm_mod fuse
[  580.423136] CPU: 784 PID: 77071 Comm: stress-ng Kdump: loaded Not tainted 5.14.0-55.el9.ppc64le #1
[  580.423139] NIP:  c0000000000f8ff4 LR: c0000000001f7c38 CTR: 0000000000000000
[  580.423140] REGS: c0000043fdd7bd60 TRAP: 0900   Not tainted  (5.14.0-55.el9.ppc64le)
[  580.423142] MSR:  800000000280b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE>  CR: 28008202  XER: 20040000
[  580.423148] CFAR: 0000000000000c00 IRQMASK: 1
               GPR00: 0000000028008202 c0000044c46b3850 c000000002a46f00 0000000000000000
               GPR04: ffffffffffffffff 0000000000000000 0000000000000010 c000000002a83060
               GPR08: 0000000000000000 0000000000000001 0000000000000001 0000000000000000
               GPR12: c0000000001b9530 c0000043ffe16700 0000000200000117 0000000010185ea8
               GPR16: 0000000010212150 0000000010186198 00000000101863a0 000000001021b3c0
               GPR20: 0000000000000001 0000000000000000 0000000000000001 00000000000000ff
               GPR24: c0000043f4a00e14 c0000043fafe0e00 000000000c440000 0000000000000000
               GPR28: c0000043f4a00e00 c0000043f4a00e00 c0000000021e0e00 c000000002561aa0
[  580.423166] NIP [c0000000000f8ff4] plpar_hcall_norets_notrace+0x18/0x2c
[  580.423168] LR [c0000000001f7c38] __pv_queued_spin_lock_slowpath+0x528/0x530
[  580.423173] Call Trace:
[  580.423174] [c0000044c46b3850] [0000000100006b60] 0x100006b60 (unreliable)
[  580.423177] [c0000044c46b3910] [c000000000ea6948] _raw_spin_lock_irqsave+0xa8/0xc0
[  580.423182] [c0000044c46b3940] [c0000000001dd7c0] prepare_to_wait_event+0x40/0x200
[  580.423185] [c0000044c46b39a0] [c00000000019e9e0] __request_module+0x320/0x510
[  580.423188] [c0000044c46b3ac0] [c0000000006f1a14] crypto_alg_mod_lookup+0x1e4/0x2e0
[  580.423192] [c0000044c46b3b60] [c0000000006f2178] crypto_alloc_tfm_node+0xa8/0x1a0
[  580.423194] [c0000044c46b3be0] [c0000000006f84f8] crypto_alloc_aead+0x38/0x50
[  580.423196] [c0000044c46b3c00] [c00000000072cba0] aead_bind+0x70/0x140
[  580.423199] [c0000044c46b3c40] [c000000000727824] alg_bind+0xb4/0x210
[  580.423201] [c0000044c46b3cc0] [c000000000bc2ad4] __sys_bind+0x114/0x160
[  580.423205] [c0000044c46b3d90] [c000000000bc2b48] sys_bind+0x28/0x40
[  580.423207] [c0000044c46b3db0] [c000000000030880] system_call_exception+0x160/0x300
[  580.423209] [c0000044c46b3e10] [c00000000000c168] system_call_vectored_common+0xe8/0x278
[  580.423213] --- interrupt: 3000 at 0x7fff9b824464
[  580.423214] NIP:  00007fff9b824464 LR: 0000000000000000 CTR: 0000000000000000
[  580.423215] REGS: c0000044c46b3e80 TRAP: 3000   Not tainted  (5.14.0-55.el9.ppc64le)
[  580.423216] MSR:  800000000280f033 <SF,VEC,VSX,EE,PR,FP,ME,IR,DR,RI,LE>  CR: 42004802  XER: 00000000
[  580.423221] IRQMASK: 0
               GPR00: 0000000000000147 00007fffdcff2780 00007fff9b917100 0000000000000004
               GPR04: 00007fffdcff27e0 0000000000000058 0000000000000000 0000000000000000
               GPR08: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
               GPR12: 0000000000000000 00007fff9bc9efe0 0000000200000117 0000000010185ea8
               GPR16: 0000000010212150 0000000010186198 00000000101863a0 000000001021b3c0
               GPR20: 0000000000000004 00007fffdcff2a00 0000000300000117 00000000101862b8
               GPR24: 0000000000000004 0000000046401570 0000000046401120 0000000046404650
               GPR28: 0000000000000020 0000000000000020 0000000000000060 0000000046404bf0
[  580.423236] NIP [00007fff9b824464] 0x7fff9b824464
[  580.423237] LR [0000000000000000] 0x0
[  580.423238] --- interrupt: 3000
[  580.423239] Instruction dump:
[  580.423241] e8690000 7c0803a6 3884fff8 78630100 78840020 4bfffeb8 3c4c0295 3842df24
[  580.423244] 7c421378 7c000026 90010008 44000022 <38800000> 988d0931 80010008 7c0ff120

Would it be possible to modify kmod so that in such cases that
request_module calls fail more quickly rather than repeatedly
obtaining a spinlock that appears to be under high contention?

If you run stress-ng with a timeout does the system eventually recover?

OK the respective stress-ng test should be something like:

./stress-ng --af-alg 8192

I had left this running overnight on x86_64 without issues:

sudo ./tools/testing/selftests/kmod/kmod.sh -t 0009

Going to leave the above stress-ng call running in a loop to see
if I can reproduce the live lock on x86_64.

The following loop has been running on 5.18.0-rc5-next-20220506 since
May 9 without any issues on x86_64:

while true; do sudo ./stress-ng --af-alg 8192; done

I ran the above on a ppc64le system here, no issues. But it's only a 32
CPU machine.

Can someone try this on ppc64le system? At this point I am not convinced
this issue is generic.

Does your x86 system have at least 784 CPUs?

I don't know where the original report came from, but the trace shows
"CPU 784", which would usually indicate a system with at least that many
CPUs.

I would hope we can handle that many CPUs banging on a spin lock without
throwing traces, but maybe not. It could also be that the system is
under load and the hypervisor has scheduled us off for too long.

cheers