On Thu, Apr 26, 2018 at 10:28:13AM -0600, Jens Axboe wrote:
> during boot. We end up doing the numa_crng_init() from interrupt context,
> which
> makes things unhappy since you do GFP_KERNEL | __GFP_NOFAIL allocations from
> there.
I've already sent a pull request to Linux to fix this; see the
random.git tree.
Apologies for not noticing this earlier. My test environment had
virtio-rng enabled by default, and while I did test the CONFIG_NUMA
case, the call to add randomness doesn't happen from interrupt context
with virtio-rng so I missed it in my testing.
Cheers,
- Ted
