linux-kernel - Re: [PATCH] hugetlbfs: Disable IRQ when taking hugetlb_lock in set_max_huge

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <93f8fd73-eb5b-355d-eea9-d74911190296@redhat.com>
Date:   Mon, 9 Dec 2019 11:05:16 -0500
From:   Waiman Long <longman@...hat.com>
To:     Mike Kravetz <mike.kravetz@...cle.com>,
        Andrew Morton <akpm@...ux-foundation.org>
Cc:     linux-kernel@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [PATCH] hugetlbfs: Disable IRQ when taking hugetlb_lock in
 set_max_huge_pages()

On 12/9/19 11:01 AM, Waiman Long wrote:
> The following lockdep splat was observed in some test runs:
>
> [  612.388273] ================================
> [  612.411273] WARNING: inconsistent lock state
> [  612.432273] 4.18.0-159.el8.x86_64+debug #1 Tainted: G        W --------- -  -
> [  612.469273] --------------------------------
> [  612.489273] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
> [  612.517273] swapper/30/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
> [  612.541273] ffffffff9acdc038 (hugetlb_lock){+.?.}, at: free_huge_page+0x36f/0xaa0
> [  612.576273] {SOFTIRQ-ON-W} state was registered at:
> [  612.598273]   lock_acquire+0x14f/0x3b0
> [  612.616273]   _raw_spin_lock+0x30/0x70
> [  612.634273]   __nr_hugepages_store_common+0x11b/0xb30
> [  612.657273]   hugetlb_sysctl_handler_common+0x209/0x2d0
> [  612.681273]   proc_sys_call_handler+0x37f/0x450
> [  612.703273]   vfs_write+0x157/0x460
> [  612.719273]   ksys_write+0xb8/0x170
> [  612.736273]   do_syscall_64+0xa5/0x4d0
> [  612.753273]   entry_SYSCALL_64_after_hwframe+0x6a/0xdf
> [  612.777273] irq event stamp: 691296
> [  612.794273] hardirqs last  enabled at (691296): [<ffffffff99bb034b>] _raw_spin_unlock_irqrestore+0x4b/0x60
> [  612.839273] hardirqs last disabled at (691295): [<ffffffff99bb0ad2>] _raw_spin_lock_irqsave+0x22/0x81
> [  612.882273] softirqs last  enabled at (691284): [<ffffffff97ff0c63>] irq_enter+0xc3/0xe0
> [  612.922273] softirqs last disabled at (691285): [<ffffffff97ff0ebe>] irq_exit+0x23e/0x2b0
> [  612.962273]
> [  612.962273] other info that might help us debug this:
> [  612.993273]  Possible unsafe locking scenario:
> [  612.993273]
> [  613.020273]        CPU0
> [  613.031273]        ----
> [  613.042273]   lock(hugetlb_lock);
> [  613.057273]   <Interrupt>
> [  613.069273]     lock(hugetlb_lock);
> [  613.085273]
> [  613.085273]  *** DEADLOCK ***
>       :
> [  613.245273] Call Trace:
> [  613.256273]  <IRQ>
> [  613.265273]  dump_stack+0x9a/0xf0
> [  613.281273]  mark_lock+0xd0c/0x12f0
> [  613.297273]  ? print_shortest_lock_dependencies+0x80/0x80
> [  613.322273]  ? sched_clock_cpu+0x18/0x1e0
> [  613.341273]  __lock_acquire+0x146b/0x48c0
> [  613.360273]  ? trace_hardirqs_on+0x10/0x10
> [  613.379273]  ? trace_hardirqs_on_caller+0x27b/0x580
> [  613.401273]  lock_acquire+0x14f/0x3b0
> [  613.419273]  ? free_huge_page+0x36f/0xaa0
> [  613.440273]  _raw_spin_lock+0x30/0x70
> [  613.458273]  ? free_huge_page+0x36f/0xaa0
> [  613.477273]  free_huge_page+0x36f/0xaa0
> [  613.495273]  bio_check_pages_dirty+0x2fc/0x5c0
> [  613.516273]  clone_endio+0x17f/0x670 [dm_mod]
> [  613.536273]  ? disable_discard+0x90/0x90 [dm_mod]
> [  613.558273]  ? bio_endio+0x4ba/0x930
> [  613.575273]  ? blk_account_io_completion+0x400/0x530
> [  613.598273]  blk_update_request+0x276/0xe50
> [  613.617273]  scsi_end_request+0x7b/0x6a0
> [  613.636273]  ? lock_downgrade+0x6f0/0x6f0
> [  613.654273]  scsi_io_completion+0x1c6/0x1570
> [  613.674273]  ? sd_completed_bytes+0x3a0/0x3a0 [sd_mod]
> [  613.698273]  ? scsi_mq_requeue_cmd+0xc0/0xc0
> [  613.718273]  blk_done_softirq+0x22e/0x350
> [  613.737273]  ? blk_softirq_cpu_dead+0x230/0x230
> [  613.758273]  __do_softirq+0x23d/0xad8
> [  613.776273]  irq_exit+0x23e/0x2b0
> [  613.792273]  do_IRQ+0x11a/0x200
> [  613.806273]  common_interrupt+0xf/0xf
> [  613.823273]  </IRQ>
>
> Since hugetlb_lock can be taken from both process and IRQ contexts, we
> need to protect the lock from nested locking by disabling IRQ before
> taking it.  So for functions that are only called from the process
> context, the spin_lock_irq()/spin_unlock_irq() pair is used. For
> functions that may be called from both process and IRQ contexts, the
> spin_lock_irqsave()/spin_unlock_irqrestore() pair is used.
>
> For now, I am assuming that only free_huge_page() will be called from
> IRQ context.
>
> Signed-off-by: Waiman Long <longman@...hat.com>
> ---
>  mm/hugetlb.c | 116 +++++++++++++++++++++++++++++++--------------------
>  1 file changed, 71 insertions(+), 45 deletions(-)

Sorry, I forgot to update the patch title. IRQ is disabled in all the
instances where it is acquired. So "in set_max_huge_pages()" should be
removed from the title.

Thanks,
Longman