On Thu, 4 Oct 2018, Jia-Ju Bai wrote:

quoted

Why? Forcing all the report buffer to be limited to be non-sleeping
allocations just because of two drivers, looks like an overkill, and
actually calls for more issues (as GFP_ATOMIC is of course in principle
less likely to succeed).

Okay, I thought that using GFP_ATOMIC is the simplest way to fix these bugs.
But I check the Linux kernel code again, and find that hid_hw_request() are
called at many places.
So changing this function may affect many drivers.
I agree to only change the two drivers, and explicitly anotate __hid_request()
with might_sleep().

Thanks. Are you planning to submit a patch to do that?

-- 
Jiri Kosina
SUSE Labs