Message-ID: <CA+fCnZcaLBUUEEUNr8uZqW1dJ8fsHcOGCy3mJttfFDKq=A_9OQ@mail.gmail.com>
Date: Mon, 17 Feb 2025 17:28:46 +0100
From: Andrey Konovalov <andreyknvl@...il.com>
To: Waiman Long <longman@...hat.com>
Cc: Andrey Ryabinin <ryabinin.a.a@...il.com>, Alexander Potapenko <glider@...gle.com>, 
	Dmitry Vyukov <dvyukov@...gle.com>, Vincenzo Frascino <vincenzo.frascino@....com>, 
	Andrew Morton <akpm@...ux-foundation.org>, 
	Sebastian Andrzej Siewior <bigeasy@...utronix.de>, Clark Williams <clrkwllms@...nel.org>, 
	Steven Rostedt <rostedt@...dmis.org>, kasan-dev@...glegroups.com, linux-mm@...ck.org, 
	linux-kernel@...r.kernel.org, linux-rt-devel@...ts.linux.dev, 
	Nico Pache <npache@...hat.com>
Subject: Re: [PATCH v3] kasan: Don't call find_vm_area() in RT kernel

On Mon, Feb 17, 2025 at 5:21 AM Waiman Long <longman@...hat.com> wrote:
>
> The following bug report appeared with a test run in a RT debug kernel.
>
> [ 3359.353842] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:48
> [ 3359.353848] in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid: 140605, name: kunit_try_catch
> [ 3359.353853] preempt_count: 1, expected: 0
>   :
> [ 3359.353933] Call trace:
>   :
> [ 3359.353955]  rt_spin_lock+0x70/0x140
> [ 3359.353959]  find_vmap_area+0x84/0x168
> [ 3359.353963]  find_vm_area+0x1c/0x50
> [ 3359.353966]  print_address_description.constprop.0+0x2a0/0x320
> [ 3359.353972]  print_report+0x108/0x1f8
> [ 3359.353976]  kasan_report+0x90/0xc8
> [ 3359.353980]  __asan_load1+0x60/0x70
>
> Commit e30a0361b851 ("kasan: make report_lock a raw spinlock")
> changes report_lock to a raw_spinlock_t to avoid a similar RT problem.
> The print_address_description() function is called with report_lock
> acquired and interrupt disabled.  However, the find_vm_area() function
> still needs to acquire a spinlock_t which becomes a sleeping lock in
> the RT kernel. IOW, we can't call find_vm_area() in a RT kernel and
> changing report_lock to a raw_spinlock_t is not enough to completely
> solve this RT kernel problem.
>
> Fix this bug report by skipping the find_vm_area() call in this case
> and just print out the address as is.
>
> For !RT kernel, follow the example set in commit 0cce06ba859a
> ("debugobjects,locking: Annotate debug_object_fill_pool() wait type
> violation") and use DEFINE_WAIT_OVERRIDE_MAP() to avoid a spinlock_t
> inside raw_spinlock_t warning.
>
> Fixes: e30a0361b851 ("kasan: make report_lock a raw spinlock")
> Signed-off-by: Waiman Long <longman@...hat.com>
> ---
>  mm/kasan/report.c | 43 ++++++++++++++++++++++++++++++-------------
>  1 file changed, 30 insertions(+), 13 deletions(-)
>
>  [v3] Rename helper to print_vmalloc_info_set_page.
>
> diff --git a/mm/kasan/report.c b/mm/kasan/report.c
> index 3fe77a360f1c..7c8c2e173aa4 100644
> --- a/mm/kasan/report.c
> +++ b/mm/kasan/report.c
> @@ -370,6 +370,34 @@ static inline bool init_task_stack_addr(const void *addr)
>                         sizeof(init_thread_union.stack));
>  }
>
> +/*
> + * RT kernel cannot call find_vm_area() in atomic context. For !RT kernel,
> + * prevent spinlock_t inside raw_spinlock_t warning by raising wait-type
> + * to WAIT_SLEEP.

Quoting your response from the other thread:

> Lockdep currently issues warnings for taking spinlock_t inside
> raw_spinlock_t because it is not allowed in RT. Test coverage of RT
> kernels is likely less than !RT kernel and so less bug of this kind will
> be caught. By making !RT doing the same check, we increase coverage.
> However, we do allow override in the !RT case, but it has to be done on
> a case-by-case basis.

Got it.

So let's put this exactly this explanation in the comment, otherwise
it's unclear why we need something special for the !RT case.

> + */
> +static inline void print_vmalloc_info_set_page(void *addr, struct page **ppage)
> +{
> +       if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
> +               static DEFINE_WAIT_OVERRIDE_MAP(vmalloc_map, LD_WAIT_SLEEP);
> +               struct vm_struct *va;
> +
> +               lock_map_acquire_try(&vmalloc_map);
> +               va = find_vm_area(addr);
> +               if (va) {
> +                       pr_err("The buggy address belongs to the virtual mapping at\n"
> +                              " [%px, %px) created by:\n"
> +                              " %pS\n",
> +                              va->addr, va->addr + va->size, va->caller);
> +                       pr_err("\n");
> +
> +                       *ppage = vmalloc_to_page(addr);

Looking at the code again, I actually like the Andrey Ryabinin's
suggestion from the v1 thread: add a separate function that contains
an annotated call of find_vm_area(). And keep vmalloc_to_page()
outside of it, just as done in the upstream version now.

> +               }
> +               lock_map_release(&vmalloc_map);
> +               return;
> +       }
> +       pr_err("The buggy address %px belongs to a vmalloc virtual mapping\n", addr);
> +}
> +
>  static void print_address_description(void *addr, u8 tag,
>                                       struct kasan_report_info *info)
>  {
> @@ -398,19 +426,8 @@ static void print_address_description(void *addr, u8 tag,
>                 pr_err("\n");
>         }
>
> -       if (is_vmalloc_addr(addr)) {
> -               struct vm_struct *va = find_vm_area(addr);
> -
> -               if (va) {
> -                       pr_err("The buggy address belongs to the virtual mapping at\n"
> -                              " [%px, %px) created by:\n"
> -                              " %pS\n",
> -                              va->addr, va->addr + va->size, va->caller);
> -                       pr_err("\n");
> -
> -                       page = vmalloc_to_page(addr);
> -               }
> -       }
> +       if (is_vmalloc_addr(addr))
> +               print_vmalloc_info_set_page(addr, &page);
>
>         if (page) {
>                 pr_err("The buggy address belongs to the physical page:\n");
> --
> 2.48.1
>

Powered by blists - more mailing lists