[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230823092429.643b4f7b.alex.williamson@redhat.com>
Date: Wed, 23 Aug 2023 09:24:29 -0600
From: Alex Williamson <alex.williamson@...hat.com>
To: Jason Gunthorpe <jgg@...dia.com>
Cc: Ankit Agrawal <ankita@...dia.com>,
Yishai Hadas <yishaih@...dia.com>,
"shameerali.kolothum.thodi@...wei.com"
<shameerali.kolothum.thodi@...wei.com>,
"kevin.tian@...el.com" <kevin.tian@...el.com>,
Aniket Agashe <aniketa@...dia.com>, Neo Jia <cjia@...dia.com>,
Kirti Wankhede <kwankhede@...dia.com>,
"Tarun Gupta (SW-GPU)" <targupta@...dia.com>,
Vikram Sethi <vsethi@...dia.com>,
Andy Currid <acurrid@...dia.com>,
Alistair Popple <apopple@...dia.com>,
John Hubbard <jhubbard@...dia.com>,
Dan Williams <danw@...dia.com>,
"kvm@...r.kernel.org" <kvm@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v7 1/1] vfio/nvgpu: Add vfio pci variant module for
grace hopper
On Wed, 23 Aug 2023 12:16:23 -0300
Jason Gunthorpe <jgg@...dia.com> wrote:
> On Wed, Aug 23, 2023 at 09:14:07AM -0600, Alex Williamson wrote:
> > On Wed, 23 Aug 2023 14:50:31 +0000
> > Ankit Agrawal <ankita@...dia.com> wrote:
> >
> > > >> + if (index == VFIO_PCI_BAR2_REGION_INDEX) {
> > > >> + if (!nvdev->opregion) {
> > > >> + nvdev->opregion = memremap(nvdev->hpa, nvdev->mem_length, MEMREMAP_WB);
> > > >> + if (!nvdev->opregion)
> > > >> + return -ENOMEM;
> > > >> + }
> > > >
> > > > [AW] Seems like this would be susceptible to concurrent accesses causing
> > > > duplicate mappings.
> > > >
> > > > [JG] Needs some kind of locking on opregion
> > >
> > > Right, will add a new lock item in nvdev to control the access to opregion/memmap.
> > > Please let me know if it is preferable to do memremap in open_device instead of
> > > read/write.
> >
> > That's a valid option also, certainly avoids the locking and
> > serialization per access. Thanks,
>
> open_device is no good, that would waste large amounts of kernel
> memory for page tables to support something we don't expect to be
> used.
A lock it is then, I guess this is a very large range of memory. The
contention/serialization is also not a priority, we expect access
through the mmap for anything other than debug. Thanks,
Alex
Powered by blists - more mailing lists