Re: [PATCH -next] mm: delete oversized WARN_ON() in kvmalloc() calls

[Leon Romanovsky]

On Wed, Dec 01, 2021 at 07:26:43PM -0800, Andrew Morton wrote:
> On Thu,  2 Dec 2021 10:06:24 +0800 Bixuan Cui <cuibixuan@xxxxxxxxxxxxxxxxx> wrote:
> 
> > Delete the WARN_ON() and return NULL directly for oversized parameter
> > in kvmalloc() calls.
> > Also add unlikely().
> > 
> > Fixes: 7661809d493b ("mm: don't allow oversized kvmalloc() calls")
> > Signed-off-by: Bixuan Cui <cuibixuan@xxxxxxxxxxxxxxxxx>
> > ---
> > There are a lot of oversize warnings and patches about kvmalloc() calls
> > recently. Maybe these warnings are not very necessary.
> 
> Or maybe they are.  Please let's take a look at these warnings, one at
> a time.  If a large number of them are bogus then sure, let's disable
> the runtime test.  But perhaps it's the case that calling code has
> genuine issues and should be repaired.

Andrew,

The problem is that this WARN_ON() is triggered by the users.

At least in the RDMA world, users can provide huge sizes and they expect
to get plain -ENOMEM and not dump stack, because it happens indirectly
to them.

In our case, these two kvcalloc() generates WARN_ON().

		umem_odp->pfn_list = kvcalloc(
			npfns, sizeof(*umem_odp->pfn_list), GFP_KERNEL);
		if (!umem_odp->pfn_list)
			return -ENOMEM;

		umem_odp->dma_list = kvcalloc(
			ndmas, sizeof(*umem_odp->dma_list), GFP_KERNEL);
https://elixir.bootlin.com/linux/v5.16-rc3/source/drivers/infiniband/core/umem_odp.c#L80

It is not a kernel programmer error to allow "oversized kvmalloc call" .

Thanks