Re: [PATCH] irqchip/gic-v3-its: Don't acquire rt_spin_lock in allocate_vpe_l1_table()

[Waiman Long <llong@redhat.com>]

On 1/12/26 10:09 AM, Thomas Gleixner wrote:
> On Sun, Jan 11 2026 at 18:02, Waiman Long wrote:
>> On 1/11/26 5:38 AM, Marc Zyngier wrote:
>>>> Also that patch seems to be incomplete because there is another
>>>> allocation further down in allocate_vpe_l1_table()....
>>> Yeah, I wondered why page allocation wasn't affected by this issue,
>>> but didn't try to find out.
>> The use of GFP_ATOMIC flag in the page allocation request may help it to
>> dip into the reserved area and avoid taking any spinlock. In my own
>> test, just removing the kzalloc() call is enough to avoid any invalid
>> context warning. In the page allocation code, there is a zone lock and a
>> per_cpu_pages lock. They were not acquired in my particular test case,
>> though further investigation may be needed to make sure it is really safe.
> They might be acquired though. Only alloc_pages_nolock() guarantees that
> no lock is taken IIRC.

Thanks for the suggestion. I will look into using that for page 
allocation. I had actually attempt to use kmalloc_nolock() to replace 
kzalloc() initially. Even though it removed the call to rmqueue(), but 
there were other spinlocks in the slub code that were still being 
acquired like the local_lock() or the spinlock in the get_random() code. 
So I gave up using that. Anyway, kmalloc_nolock() doesn't seem to be 
fully working yet.

Cheers,
Longman