Date:   Thu, 2 Dec 2021 15:29:47 +0000
From:   Matthew Wilcox <willy@infradead.org>
To:     Leon Romanovsky <leon@kernel.org>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        Bixuan Cui <cuibixuan@linux.alibaba.com>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, torvalds@linux-foundation.org,
        w@1wt.eu, keescook@chromium.org
Subject: Re: [PATCH -next] mm: delete oversized WARN_ON() in kvmalloc() calls
Message-ID: <YajmawzehKqR+j0v@casper.infradead.org>
References: <1638410784-48646-1-git-send-email-cuibixuan@linux.alibaba.com>
 <20211201192643.ecb0586e0d53bf8454c93669@linux-foundation.org>
 <Yajk/oVypyUFTtgd@unreal>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Yajk/oVypyUFTtgd@unreal>
Precedence: bulk

On Thu, Dec 02, 2021 at 05:23:42PM +0200, Leon Romanovsky wrote:
> The problem is that this WARN_ON() is triggered by the users.

... or the problem is that you don't do a sanity check between the user
and the MM system.  I mean, that's what this conversation is about --
is it a bug to be asking for this much memory in the first place?

> 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);

Does it really make sense for the user to specify 2^31 PFNs in a single
call?  I mean, that's 8TB of memory.  Should RDMA put its own limit
in here, or should it rely on kvmalloc returning -ENOMEM?