Re: [x86/pgtable] d1ec551f87: BUG:Bad_page_map_in_process

[David Hildenbrand <david@redhat.com>]

On 19.04.22 10:22, David Hildenbrand wrote:
> On 19.04.22 09:59, kernel test robot wrote:
>>
>>
>> Greeting,
>>
>> FYI, we noticed the following commit (built with clang-15):
>>
>> commit: d1ec551f874e1663bfe76b994c0010a4566cf936 ("x86/pgtable: support __HAVE_ARCH_PTE_SWP_EXCLUSIVE")
>> https://github.com/hnaz/linux-mm master
>>
>> in testcase: trinity
>> version: trinity-static-i386-x86_64-1c734c75-1_2020-01-06
>> with following parameters:
>>
>> 	runtime: 300s
>> 	group: group-01
>>
>> test-description: Trinity is a linux system call fuzz tester.
>> test-url: http://codemonkey.org.uk/projects/trinity/
>>
>>
>> on test machine: qemu-system-i386 -enable-kvm -cpu SandyBridge -smp 2 -m 4G
>>
>> caused below changes (please refer to attached dmesg/kmsg for entire log/backtrace):
>>
>>
>>
>> If you fix the issue, kindly add following tag
>> Reported-by: kernel test robot <oliver.sang@intel.com>
>>
>>
>> [   40.201103][ T5099] BUG: Bad page map in process trinity-c7  pte:1713003a pmd:7ff71067
>> [   40.201999][ T5099] addr:096e7000 vm_flags:00100073 anon_vma:bff0aa00 mapping:00000000 index:96e7
>> [   40.202718][ T5099] file:(null) fault:0x0 mmap:0x0 readpage:0x0
>> [   40.203229][ T5099] CPU: 0 PID: 5099 Comm: trinity-c7 Not tainted 5.18.0-rc2-mm1-00053-gd1ec551f874e #1
>> [   40.203952][ T5099] Call Trace:
>> [ 40.204195][ T5099] ? dump_stack_lvl (??:?) 
>> [ 40.204581][ T5099] ? dump_stack (??:?) 
>> [ 40.204970][ T5099] ? print_bad_pte (memory.c:?) 
>> [ 40.205384][ T5099] ? unmap_page_range (??:?) 
>> [ 40.205843][ T5099] ? unmap_single_vma (memory.c:?) 
>> [ 40.206271][ T5099] ? unmap_vmas (??:?) 
>> [ 40.206647][ T5099] ? exit_mmap (??:?) 
>> [ 40.207032][ T5099] ? __mmput (fork.c:?) 
>> [ 40.207405][ T5099] ? mmput (??:?) 
>> [ 40.207751][ T5099] ? exit_mm (exit.c:?) 
>> [ 40.208121][ T5099] ? do_exit (??:?) 
>> [ 40.208497][ T5099] ? do_group_exit (??:?) 
>> [ 40.208905][ T5099] ? trace_hardirqs_on (??:?) 
>> [ 40.209345][ T5099] ? get_signal (??:?) 
>> [ 40.209750][ T5099] ? arch_do_signal_or_restart (??:?) 
>> [ 40.210287][ T5099] ? exit_to_user_mode_loop (common.c:?) 
>> [ 40.210778][ T5099] ? exit_to_user_mode_prepare (common.c:?) 
>> [ 40.211302][ T5099] ? syscall_exit_to_user_mode (??:?) 
>> [ 40.211808][ T5099] ? ret_from_fork (??:?) 
>> [   40.212268][ T5099] Disabling lock debugging due to kernel taint
>> [   40.231123][ T5097] BUG: Bad page map in process trinity-c5  pte:171e0a3e pmd:0a8d3067
>> [   40.231770][ T5099] BUG: Bad page map in process trinity-c7  pte:1713023a pmd:7ff71067
>> [   40.231883][ T5097] addr:36ed5000 vm_flags:000000fb anon_vma:00000000 mapping:485d0d80 index:1
>> [   40.232611][ T5099] addr:096e8000 vm_flags:00100073 anon_vma:bff0ab18 mapping:00000000 index:96e8
>> [   40.233429][ T5097] file:dev/zero fault:shmem_fault mmap:shmem_mmap readpage:0x0
>> [   40.234271][ T5099] file:(null) fault:0x0 mmap:0x0 readpage:0x0
>> [   40.234971][ T5097] CPU: 1 PID: 5097 Comm: trinity-c5 Tainted: G    B             5.18.0-rc2-mm1-00053-gd1ec551f874e #1
>> [   40.236510][ T5097] Call Trace:
> 
> This is 32bit (i386) I assume. I wonder if something about the 32bit swp
> layout is special and we have to restrict it to 64bit here.
> 

Staring at arch/x86/include/asm/pgtable-2level.h, I think that's the case:
The swp type effectively starts at "_PAGE_BIT_PRESENT + 1" which should be bit 1.

Let's restrict this to x86-64 for now.

I think this is similar to CONFIG_HAVE_ARCH_SOFT_DIRTY, which depends on 64 bit.


The following should do the trick:

diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
index e42e668153e9..a1c555abed26 100644
--- a/arch/x86/include/asm/pgtable.h
+++ b/arch/x86/include/asm/pgtable.h
@@ -1291,7 +1291,7 @@ static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
                unsigned long addr, pud_t *pud)
 {
 }
-
+#ifdef _PAGE_SWP_EXCLUSIVE
 #define __HAVE_ARCH_PTE_SWP_EXCLUSIVE
 static inline pte_t pte_swp_mkexclusive(pte_t pte)
 {
@@ -1307,6 +1307,7 @@ static inline pte_t pte_swp_clear_exclusive(pte_t pte)
 {
        return pte_clear_flags(pte, _PAGE_SWP_EXCLUSIVE);
 }
+#endif /* _PAGE_SWP_EXCLUSIVE */
 
 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
diff --git a/arch/x86/include/asm/pgtable_types.h b/arch/x86/include/asm/pgtable_types.h
index 54a8f370046d..95403ae1f2e5 100644
--- a/arch/x86/include/asm/pgtable_types.h
+++ b/arch/x86/include/asm/pgtable_types.h
@@ -83,10 +83,12 @@
 #define _PAGE_SOFT_DIRTY       (_AT(pteval_t, 0))
 #endif
 
+#ifdef CONFIG_64BIT
 /*
- * We borrow bit 3 to remember PG_anon_exclusive.
+ * We borrow bit 3 on x86-64 to remember PG_anon_exclusive.
  */
 #define _PAGE_SWP_EXCLUSIVE    _PAGE_PWT
+#endif /* CONFIG_64BIT */
 
 /*
  * Tracking soft dirty bit when a page goes to a swap is tricky.


Thoughts?


-- 
Thanks,

David / dhildenb