Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 28 Jun 2022 11:04:59 -0300
From:   Jason Gunthorpe <jgg@ziepe.ca>
To:     Steven Sistare <steven.sistare@oracle.com>
Cc:     Alex Williamson <alex.williamson@redhat.com>,
        lizhe.67@bytedance.com, cohuck@redhat.com, kvm@vger.kernel.org,
        linux-kernel@vger.kernel.org, lizefan.x@bytedance.com
Subject: Re: [PATCH] vfio: remove useless judgement
Message-ID: <20220628140459.GP23621@ziepe.ca>
References: <20220627035109.73745-1-lizhe.67@bytedance.com>
 <20220627160640.7edca0dd.alex.williamson@redhat.com>
 <7217566f-9c40-ae9d-6fd6-2ef93f13f853@oracle.com>
 <20220628130350.GN23621@ziepe.ca>
 <f7b7a7b0-9404-6b0f-99b5-346af041a479@oracle.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <f7b7a7b0-9404-6b0f-99b5-346af041a479@oracle.com>
Precedence: bulk

On Tue, Jun 28, 2022 at 09:54:19AM -0400, Steven Sistare wrote:

> >> As you and I have discussed, the count is also wrong in the direct
> >> exec model, because exec clears mm->locked_vm.
> > 
> > Really? Yikes, I thought exec would generate a new mm?
> 
> Yes, exec creates a new mm with locked_vm = 0.  The old locked_vm count is dropped
> on the floor.  The existing dma points to the same task, but task->mm has changed,
> and dma->task->mm->locked_vm is 0.  An unmap ioctl drives it
> negative.

Oh.. This is probably a bug, vfio should never use task->mm, the mm
itself should be held using mmgrab instead.

Otherwise exec case is broken as you describe.

> I have prototyped a few possible fixes.  One changes vfio to use user->locked_vm.
> Another changes to mm->pinned_vm and preserves it during exec.  A third preserves
> mm->locked_vm across exec, but that is not practical, because mm->locked_vm mixes
> vfio pins and mlocks.  The mlock component must be cleared during exec, and we don't 
> have a separate count for it.

Lossing locked_vm on exec/fork is the correct and expected behavior
for the core kernel code, the bug is that vfio drives it negative.

Jason