Re: BISECTED: 'alloc_tag: populate memory for module tags as needed' crashes on boot.

[Ben Greear <greearb@candelatech.com>]

On 12/6/24 16:15, Suren Baghdasaryan wrote:
> On Fri, Dec 6, 2024 at 2:55 PM Suren Baghdasaryan <surenb@google.com> wrote:
>>
>> On Fri, Dec 6, 2024 at 2:43 PM Ben Greear <greearb@candelatech.com> wrote:
>>>
>>> On 12/6/24 14:03, Suren Baghdasaryan wrote:
>>>> On Fri, Dec 6, 2024 at 1:50 PM Ben Greear <greearb@candelatech.com> wrote:
>>>>>
>>>>> Hello Suren,
>>>>>
>>>>> My system crashes on bootup, and I bisected to this commit.
>>>>>
>>>>> 0f9b685626daa2f8e19a9788625c9b624c223e45 is the first bad commit
>>>>> commit 0f9b685626daa2f8e19a9788625c9b624c223e45
>>>>> Author: Suren Baghdasaryan <surenb@google.com>
>>>>> Date:   Wed Oct 23 10:07:57 2024 -0700
>>>>>
>>>>>        alloc_tag: populate memory for module tags as needed
>>>>>
>>>>>        The memory reserved for module tags does not need to be backed by physical
>>>>>        pages until there are tags to store there.  Change the way we reserve this
>>>>>        memory to allocate only virtual area for the tags and populate it with
>>>>>        physical pages as needed when we load a module.
>>>>>
>>>>> The crash looks like this:
>>>>>
>>>>> BUG: unable to handle page fault for address: fffffbfff4041000
>>>>> #PF: supervisor read access in kernel mode
>>>>> #PF: error_code(0x0000) - not-present page
>>>>> PGD 44d0e7067 P4D 44d0e7067 PUD 44d0e3067 PMD 10bb38067 PTE 0
>>>>> Oops: Oops: 0000 [#1] PREEMPT SMP KASAN
>>>>> CPU: 0 UID: 0 PID: 319 Comm: systemd-udevd Not tainted 6.12.0-rc6+ #21
>>>>> Hardware name: Default string Default string/SKYBAY, BIOS 5.12 02/15/2023
>>>>> RIP: 0010:kasan_check_range+0xa5/0x190
>>>>> Code: 8d 5a 07 4c 0f 49 da 49 c1 fb 03 45 85 db 0f 84 ce 00 00 00 45 89 db 4a 8d 14 d8 eb 0d 48 83 c0 08 48 39 d0 0f 84 b29
>>>>> RSP: 0018:ffff88812c26f980 EFLAGS: 00010206
>>>>> RAX: fffffbfff4041000 RBX: fffffbfff404101e RCX: ffffffff814ec29b
>>>>> [  OK  DX: fffffbfff4041018 RSI: 00000000000000f0 RDI: ffffffffa0208000
>>>>> 0m] Finished BP: fffffbfff4041000 R08: 0000000000000001 R09: fffffbfff404101d
>>>>> ;1;39msystemd-udR10: ffffffffa02080ef R11: 0000000000000003 R12: ffffffffa0208000
>>>>> ev-trig…e R13: ffffc90000dac7c8 R14: ffffc90000dac7e8 R15: dffffc0000000000
>>>>> - Coldplug All uFS:  00007fe869216b40(0000) GS:ffff88841da00000(0000) knlGS:0000000000000000
>>>>> dev Devices.
>>>>> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>>> CR2: fffffbfff4041000 CR3: 0000000121e86002 CR4: 00000000003706f0
>>>>> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>>>>> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>>>>> Call Trace:
>>>>>     <TASK>
>>>>> [  OK  ? __die+0x1f/0x60
>>>>> 0m] Reached targ ? page_fault_oops+0x258/0x910
>>>>> et sysi ? dump_pagetable+0x690/0x690
>>>>> nit.target - ? search_bpf_extables+0x22/0x250
>>>>>     System Initiali ? trace_page_fault_kernel+0x120/0x120
>>>>> zation.
>>>>>     ? search_bpf_extables+0x164/0x250
>>>>>     ? kasan_check_range+0xa5/0x190
>>>>>     ? fixup_exception+0x4d/0xc70
>>>>>     ? exc_page_fault+0xe1/0xf0
>>>>> [  OK  ? asm_exc_page_fault+0x22/0x30
>>>>> 0m] Reached targ ? load_module+0x3d7b/0x7560
>>>>> et netw ? kasan_check_range+0xa5/0x190
>>>>> ork.target - __asan_memcpy+0x38/0x60
>>>>>     Network.
>>>>>     load_module+0x3d7b/0x7560
>>>>>     ? module_frob_arch_sections+0x30/0x30
>>>>>     ? lockdep_lock+0xbe/0x1b0
>>>>>     ? rw_verify_area+0x18d/0x5e0
>>>>>     ? kernel_read_file+0x246/0x870
>>>>>     ? __x64_sys_fspick+0x290/0x290
>>>>>     ? init_module_from_file+0xd1/0x130
>>>>>     init_module_from_file+0xd1/0x130
>>>>>     ? __ia32_sys_init_module+0xa0/0xa0
>>>>>     ? lock_acquire+0x2d/0xb0
>>>>>     ? idempotent_init_module+0x116/0x790
>>>>>     ? do_raw_spin_unlock+0x54/0x220
>>>>>     idempotent_init_module+0x226/0x790
>>>>>     ? init_module_from_file+0x130/0x130
>>>>>     ? vm_mmap_pgoff+0x203/0x2e0
>>>>>     __x64_sys_finit_module+0xba/0x130
>>>>>     do_syscall_64+0x69/0x160
>>>>>     entry_SYSCALL_64_after_hwframe+0x4b/0x53
>>>>> RIP: 0033:0x7fe869de327d
>>>>> Code: 5d c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 248
>>>>> RSP: 002b:00007ffe34a828d8 EFLAGS: 00000246 ORIG_RAX: 0000000000000139
>>>>> RAX: ffffffffffffffda RBX: 0000557fa8f3f3f0 RCX: 00007fe869de327d
>>>>> RDX: 0000000000000000 RSI: 00007fe869f4943c RDI: 0000000000000006
>>>>> RBP: 00007fe869f4943c R08: 0000000000000000 R09: 0000000000000000
>>>>> R10: 0000000000000006 R11: 0000000000000246 R12: 0000000000020000
>>>>> R13: 0000557fa8f3f030 R14: 0000000000000000 R15: 0000557fa8f3d110
>>>>>     </TASK>
>>>>> Modules linked in:
>>>>> CR2: fffffbfff4041000
>>>>> ---[ end trace 0000000000000000 ]---
>>>>>
>>>>> I suspect you only hit this with an unlucky amount of debugging enabled.  The kernel config I used
>>>>> is found here:
>>>>>
>>>>> http://www.candelatech.com/downloads/cfg-kasan-crash-regression.config
>>>>>
>>>>> I will be happy to test fixes.
>>>>
>>>> Hi Ben,
>>>> Thanks for reporting the issue. Do you have these recent fixes in your tree:
>>>>
>>>> https://lore.kernel.org/all/20241130001423.1114965-1-surenb@google.com/
>>>> https://lore.kernel.org/all/20241205170528.81000-1-hao.ge@linux.dev/
>>>>
>>>> If not, couple you please apply them and see if the issue is still happening?
>>>> Thanks,
>>>> Suren.
>>>
>>> Hello Suren,
>>>
>>> Thanks for the quick response.  The first patch is already in latest Linus tree,
> 
> Hmm. Could you please double-check which tree you are using? I don't
> see the first patch
> (https://lore.kernel.org/all/20241130001423.1114965-1-surenb@google.com/)
> in Linus' tree. Maybe you are using linux-next?

Sorry, you are correct.  I must have mangled something when trying to apply
the patch and I didn't look hard enough when patch said changes were already applied.

I can re-test this next week...and for reference, kernel boots fine when you disable
KASAN and other debugging.

Thanks,
Ben


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com