Re: [x86/mm/gup] 2947ba054a [ 71.329069] kernel BUG at include/linux/pagemap.h:151!

["Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>]

On Fri, Mar 24, 2017 at 11:51:53AM +0100, Peter Zijlstra wrote:
> On Fri, Mar 24, 2017 at 01:24:36PM +0300, Kirill A. Shutemov wrote:
> 
> > I'm not sure what is the best way to fix this.
> > Few options:
> >  - Drop the VM_BUG();
> >  - Bump preempt count during __get_user_pages_fast();
> >  - Use get_page() instead of page_cache_get_speculative() on x86.
> > 
> > Any opinions?
> 
> I think I'm in favour of the first; either remove or amend to include
> irqs_disabled() or so.
> 
> This in favour of keeping the variants of GUP down.

Something like this?

-------------------8<-----------------------

>From d2f416a3ee3e5dbb10e59d0b374d382fdc4ba082 Mon Sep 17 00:00:00 2001
From: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Date: Fri, 24 Mar 2017 14:13:05 +0300
Subject: [PATCH] mm: Fix false-positive VM_BUG_ON() in
 page_cache_{get,add}_speculative

0day triggered this:

  kernel BUG at include/linux/pagemap.h:151!
  invalid opcode: 0000 [#1]
  CPU: 0 PID: 458 Comm: trinity-c0 Not tainted 4.11.0-rc2-00251-g2947ba0 #1
  task: ffff88001f19ab00 task.stack: ffff88001f084000
  RIP: 0010:gup_pud_range+0x56f/0x63d
  RSP: 0018:ffff88001f087ba8 EFLAGS: 00010046
  RAX: 0000000080000000 RBX: 000000000164e000 RCX: ffff88001e0badc0
  RDX: dead000000000100 RSI: 0000000000000001 RDI: ffff88001e0badc0
  RBP: ffff88001f087c38 R08: ffff88001f087cf8 R09: ffff88001f087c6c
  R10: 0000000000000000 R11: ffff88001f19b0f0 R12: ffff88001f087c6c
  R13: ffff88001e0badc0 R14: 800000001e7b7867 R15: 0000000000000000
  FS:  00007f7ea7b60700(0000) GS:ffffffffae02f000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00000000013eb130 CR3: 0000000017ddb000 CR4: 00000000000006f0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 00000000000d0602
  Call Trace:
   __get_user_pages_fast+0x107/0x136
   get_user_pages_fast+0x78/0x89
   get_futex_key+0xfd/0x350
   ? simple_write_end+0x83/0xbe
   futex_requeue+0x1a3/0x585
   do_futex+0x834/0x86f
   ? kvm_clock_read+0x16/0x1e
   ? paravirt_sched_clock+0x9/0xd
   ? lock_release+0x11e/0x328
   SyS_futex+0x125/0x135
   ? write_seqcount_end+0x1a/0x1f
   ? vtime_account_user+0x4b/0x50
   do_syscall_64+0x61/0x74
   entry_SYSCALL64_slow_path+0x25/0x25

It' VM_BUG_ON() due to false-negative in_atomic(). We call
page_cache_get_speculative() with disabled local interrupts.
It should be atomic enough.

Let's check for disabled interrupts in the VM_BUG_ON() too.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
---
 include/linux/pagemap.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index f84cf5f76366..e7bbd9d4dc6c 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -148,7 +148,7 @@ static inline int page_cache_get_speculative(struct page *page)
 
 #ifdef CONFIG_TINY_RCU
 # ifdef CONFIG_PREEMPT_COUNT
-	VM_BUG_ON(!in_atomic());
+	VM_BUG_ON(!in_atomic() && !irqs_disabled());
 # endif
 	/*
 	 * Preempt must be disabled here - we rely on rcu_read_lock doing
@@ -186,7 +186,7 @@ static inline int page_cache_add_speculative(struct page *page, int count)
 
 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
 # ifdef CONFIG_PREEMPT_COUNT
-	VM_BUG_ON(!in_atomic());
+	VM_BUG_ON(!in_atomic() && !irqs_disabled());
 # endif
 	VM_BUG_ON_PAGE(page_count(page) == 0, page);
 	page_ref_add(page, count);
-- 
 Kirill A. Shutemov