Re: [sched] 10e7071b2f: BUG:kernel_NULL_pointer_dereference,address

[Quentin Perret <qperret@google.com>]

On Thursday 07 Nov 2019 at 11:41:38 (+0000), Valentin Schneider wrote:
> On 07/11/2019 09:08, kernel test robot wrote:
> > [   84.432464] BUG: kernel NULL pointer dereference, address: 0000000000000064
> > [   84.433700] #PF: supervisor read access in kernel mode
> > [   84.434589] #PF: error_code(0x0000) - not-present page
> > [   84.435499] PGD 0 P4D 0 
> > [   84.435933] Oops: 0000 [#1] SMP PTI
> > [   84.436581] CPU: 1 PID: 15 Comm: migration/1 Not tainted 5.3.0-rc1-00086-g10e7071b2f491 #1
> > [   84.438004] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
> > [   84.439461] RIP: 0010:pick_next_task_dl+0xe/0xf0
> > [   84.440266] Code: ed bd 70 01 01 e8 42 2d fb ff 0f 0b e9 6b ff ff ff 66 66 2e 0f 1f 84 00 00 00 00 00 66 66 66 66 90 55 53 48 89 fb 48 83 ec 10 <8b> 46 64 85 c0 78 73 48 81 7e 78 a0 3f e2 a7 74 57 48 83 bb 10 09
> > [   84.443485] RSP: 0000:ffffa5518008bd40 EFLAGS: 00010082
> > [   84.444423] RAX: ffffffffa6eeeae0 RBX: ffff98ebbfd2b0c0 RCX: ffff98ebbfd2d040
> > [   84.445641] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff98ebbfd2b0c0
> > [   84.446877] RBP: ffffa5518008bdc0 R08: 0000001ac1016512 R09: 0000000000000001
> > [   84.448128] R10: ffffffffa863e640 R11: 0000000000000003 R12: ffff98ebbfd2b0c0
> > [   84.449349] R13: ffffffffa7e23fa0 R14: ffffffffa7e24060 R15: 0000000000000000
> > [   84.450603] FS:  0000000000000000(0000) GS:ffff98ebbfd00000(0000) knlGS:0000000000000000
> > [   84.452007] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [   84.453022] CR2: 0000000000000064 CR3: 00000001aab84000 CR4: 00000000000406e0
> > [   84.454244] Call Trace:
> > [   84.455263]  ? update_rq_clock+0x6d/0xe0
> > [   84.456081]  sched_cpu_dying+0x104/0x380
> > [   84.456777]  ? sched_cpu_starting+0xf0/0xf0
> > [   84.457510]  cpuhp_invoke_callback+0x86/0x5d0
> > [   84.458279]  ? cpu_disable_common+0x292/0x2b0
> > [   84.459047]  take_cpu_down+0x60/0xb0
> > [   84.459649]  multi_cpu_stop+0x6b/0x100
> > [   84.460339]  ? stop_machine_yield+0x10/0x10
> > [   84.461078]  cpu_stopper_thread+0x9e/0x110
> > [   84.461809]  ? smpboot_thread_fn+0x2f/0x1e0
> > [   84.462539]  ? smpboot_thread_fn+0x74/0x1e0
> > [   84.463280]  ? smpboot_thread_fn+0x14e/0x1e0
> > [   84.464024]  smpboot_thread_fn+0x149/0x1e0
> > [   84.464768]  ? sort_range+0x20/0x20
> > [   84.465389]  kthread+0x11e/0x140
> > [   84.465961]  ? kthread_park+0xa0/0xa0
> > [   84.466605]  ret_from_fork+0x35/0x40
> > [   84.467236] Modules linked in: rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver binfmt_misc intel_rapl_msr intel_rapl_common sr_mod crct10dif_pclmul cdrom crc32_pclmul sg crc32c_intel ghash_clmulni_intel ata_generic pata_acpi ppdev bochs_drm drm_vram_helper ttm drm_kms_helper syscopyarea sysfillrect snd_pcm sysimgblt fb_sys_fops drm aesni_intel snd_timer ata_piix crypto_simd snd cryptd glue_helper libata soundcore pcspkr joydev serio_raw parport_pc i2c_piix4 parport floppy ip_tables
> > [   84.474471] CR2: 0000000000000064
> > [   84.475066] ---[ end trace af8f1919a81ca744 ]---
> > 
> 
> It would be really nice if the lkpbot could feed this through
> scripts/decode_stracktrace.sh too.
> 
> 
> AFAICT this is dying in __pick_migrate_task(), where we have a 
> pick_next_task()/put_prev_task() dance. Sounds oddly familiar :-)

Argh, indeed :( For context, this has been reported too:

  https://lore.kernel.org/lkml/20191028174603.GA246917@google.com/

> However I don't think we have an issue with the lock being released when
> we reach pick_next_task_fair() - if we ever reach the newidle_balance()
> in there it means we didn't have any CFS task to play with in the first
> place.
>
> We *may* e.g. go through sched_move_task() on a task we'll end up picking
> via active balance, but AFAICT that would be harmless:

sched_move_task() follows what Peter called the 'change' pattern, so I'm
thinking this is most likely the same issue. Dropping the lock causes an
unmitigated race between sched_move_task() and pick_next_task_dl(), so
hilarity ensues (set_next_task() being called twice for instance).

> __pick_migrate_task()
>   ...
>   pick_next_task_fair()
>     newidle_balance()
>       drop lock
>                                   sched_move_task()
>                                     put_prev_task()
>                                     set_next_task()
>   active balance
>   ...
>   put_prev_task()
>
> On top of that, the deathpoint seems to be select_task_rq_dl() which would
> happen way before we reach fair. I don't see anything obvious by staring
> at the old fake task pick loop vs __pick_migrate_task(). Lemme try to
> reproduce.


> 
> > 
> > To reproduce:
> > 
> >         # build kernel
> > 	cd linux
> > 	cp config-5.3.0-rc1-00086-g10e7071b2f491 .config
> > 	make HOSTCC=gcc-7 CC=gcc-7 ARCH=x86_64 olddefconfig prepare modules_prepare bzImage
> > 
> >         git clone https://github.com/intel/lkp-tests.git
> >         cd lkp-tests
> >         bin/lkp qemu -k <bzImage> job-script # job-script is attached in this email
> > 
> > 
> > 
> > Thanks,
> > lkp
> >