Re: [PATCH] mm/damon/tests/vaddr-kunit: don't use mas_lock for MM_MT_FLAGS-initialized maple tree

[Guenter Roeck <linux@roeck-us.net>]

On 9/4/24 12:26, Liam R. Howlett wrote:
> * Guenter Roeck <linux@roeck-us.net> [240904 00:27]:
>> On 9/3/24 20:36, Liam R. Howlett wrote:
>>> * Guenter Roeck <linux@roeck-us.net> [240903 22:38]:
>>>> On 9/3/24 19:31, Liam R. Howlett wrote:
>>>>> * SeongJae Park <sj@kernel.org> [240903 21:18]:
>>>>>> On Tue,  3 Sep 2024 17:58:15 -0700 SeongJae Park <sj@kernel.org> wrote:
>>>>>>
>>>>>>> On Tue, 3 Sep 2024 20:48:53 -0400 "Liam R. Howlett" <Liam.Howlett@oracle.com> wrote:
>>>>>>>
>>>>>>>> * SeongJae Park <sj@kernel.org> [240903 20:45]:
>>>>>>>>> damon_test_three_regions_in_vmas() initializes a maple tree with
>>>>>>>>> MM_MT_FLAGS.  The flags contains MT_FLAGS_LOCK_EXTERN, which means
>>>>>>>>> mt_lock of the maple tree will not be used.  And therefore the maple
>>>>>>>>> tree initialization code skips initialization of the mt_lock.  However,
>>>>>>>>> __link_vmas(), which adds vmas for test to the maple tree, uses the
>>>>>>>>> mt_lock.  In other words, the uninitialized spinlock is used.  The
>>>>>>>>> problem becomes celar when spinlock debugging is turned on, since it
>>>>>>>>> reports spinlock bad magic bug.  Fix the issue by not using the mt_lock
>>>>>>>>> as promised.
>>>>>>>>
>>>>>>>> You can't do this, lockdep will tell you this is wrong.
>>>>>>>
>>>>>>> Hmm, but lockdep was silence on my setup?
>>>>>>>
>>>>>>>> We need a lock and to use the lock for writes.
>>>>>>>
>>>>>>> This code is executed by a single-thread test code.  Do we still need the lock?
>>>>>>>
>>>>>>>>
>>>>>>>> I'd suggest using different flags so the spinlock is used.
>>>>>>>
>>>>>>> The reporter mentioned simply dropping MT_FLAGS_LOCK_EXTERN from the flags
>>>>>>> causes suspicious RCU usage message.  May I ask if you have a suggestion of
>>>>>>> better flags?
>>>>>
>>>>> That would be the lockdep complaining, so that's good.
>>>>>
>>>>>>
>>>>>> I was actually thinking replacing the mt_init_flags() with mt_init(), which
>>>>>> same to mt_init_flags() with zero flag, like below.
>>>>>
>>>>> Yes.  This will use the spinlock which should fix your issue, but it
>>>>> will use a different style of maple tree.
>>>>>
>>>>> Perhaps use MT_FLAGS_ALLOC_RANGE to use the same type of maple tree, if
>>>>> you ever add threading you will want the rcu flag as well
>>>>> (MT_FLAGS_USE_RCU).
>>>>>
>>>>> I would recommend those two and just use the spinlock.
>>>>>
>>>>
>>>> I tried that (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_USE_RCU). it also triggers
>>>> the suspicious RCU usage message.
>>>>
>>>
>>> I am running ./tools/testing/kunit/kunit.py run '*damon*' --arch x86_64 --raw
>>> with:
>>> CONFIG_LOCKDEP=y
>>> CONFIG_DEBUG_SPINLOCK=y
>>>
>>> and I don't have any issue with locking in the existing code.  How do I
>>> recreate this issue?
>>>
>>
>> I tested again, and I still see
>>
>>
>> [    6.233483] ok 4 damon
>> [    6.234190]     KTAP version 1
>> [    6.234263]     # Subtest: damon-operations
>> [    6.234335]     # module: vaddr
>> [    6.234384]     1..6
>> [    6.235726]
>> [    6.235931] =============================
>> [    6.236018] WARNING: suspicious RCU usage
>> [    6.236280] 6.11.0-rc6-00029-gda66250b210f-dirty #1 Tainted: G                 N
>> [    6.236398] -----------------------------
>> [    6.236474] lib/maple_tree.c:832 suspicious rcu_dereference_check() usage!
>> [    6.236579]
>> [    6.236579] other info that might help us debug this:
>> [    6.236579]
>> [    6.236738]
>> [    6.236738] rcu_scheduler_active = 2, debug_locks = 1
>> [    6.237039] no locks held by kunit_try_catch/208.
>> [    6.237166]
>> [    6.237166] stack backtrace:
>> [    6.237385] CPU: 0 UID: 0 PID: 208 Comm: kunit_try_catch Tainted: G                 N 6.11.0-rc6-00029-gda66250b210f-dirty #1
>> [    6.237629] Tainted: [N]=TEST
>> [    6.237714] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
>> [    6.238065] Call Trace:
>> [    6.238233]  <TASK>
>> [    6.238547]  dump_stack_lvl+0x9e/0xe0
>> [    6.239473]  lockdep_rcu_suspicious+0x145/0x1b0
>> [    6.239621]  mas_walk+0x19f/0x1d0
>> [    6.239765]  mas_find+0xb5/0x150
>> [    6.239873]  __damon_va_three_regions+0x7e/0x130
> 
> This function isn't taking the rcu read lock while iterating the tree.
> 
> Try this:
> 
> diff --git a/mm/damon/vaddr.c b/mm/damon/vaddr.c
> index b0e8b361891d..08cfd22b5249 100644
> --- a/mm/damon/vaddr.c
> +++ b/mm/damon/vaddr.c
> @@ -126,6 +126,7 @@ static int __damon_va_three_regions(struct mm_struct *mm,
>           * If this is too slow, it can be optimised to examine the maple
>           * tree gaps.
>           */
> +       rcu_read_lock();
>          for_each_vma(vmi, vma) {
>                  unsigned long gap;
>   
> @@ -146,6 +147,7 @@ static int __damon_va_three_regions(struct mm_struct *mm,
>   next:
>                  prev = vma;
>          }
> +       rcu_read_unlock();
>   
>          if (!sz_range(&second_gap) || !sz_range(&first_gap))
>                  return -EINVAL;
> 


Yes, that fixes the problem for me.

Thanks,
Guenter