From: Jason Gunthorpe <jgg@ziepe.ca>
To: "Xiong, Jianxin" <jianxin.xiong@intel.com>
Cc: Leon Romanovsky <leon@kernel.org>,
"linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>,
"dri-devel@lists.freedesktop.org"
<dri-devel@lists.freedesktop.org>,
Doug Ledford <dledford@redhat.com>,
"Vetter, Daniel" <daniel.vetter@intel.com>,
Christian Koenig <christian.koenig@amd.com>
Subject: Re: [RFC PATCH v3 1/4] RDMA/umem: Support importing dma-buf as user memory region
Date: Mon, 5 Oct 2020 13:33:04 -0300 [thread overview]
Message-ID: <20201005163304.GU9916@ziepe.ca> (raw)
In-Reply-To: <MW3PR11MB455572267489B3F6B1C5F8C5E50C0@MW3PR11MB4555.namprd11.prod.outlook.com>
On Mon, Oct 05, 2020 at 04:18:11PM +0000, Xiong, Jianxin wrote:
> > The implementation in mlx5 will be much more understandable, it would just do dma_buf_dynamic_attach() and program the XLT exactly
> > the same as a normal umem.
> >
> > The move_notify() simply zap's the XLT and triggers a work to reload it after the move. Locking is provided by the dma_resv_lock. Only a
> > small disruption to the page fault handler is needed.
>
> We considered such scheme but didn't go that way due to the lack of
> notification when the move is done and thus the work wouldn't know
> when it can reload.
Well, the work would block on the reservation lock and that indicates
the move is done
It would be nicer if the dma_buf could provide an op that things are
ready to go though
> Now I think it again, we could probably signal the reload in the page fault handler.
This also works, with a performance cost
> > > + dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL);
> > > + sgt = dma_buf_map_attachment(umem_dmabuf->attach,
> > > + DMA_BIDIRECTIONAL);
> > > + dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
> >
> > This doesn't look right, this lock has to be held up until the HW is programmed
>
> The mapping remains valid until being invalidated again. There is a
> sequence number checking before programming the HW.
It races, we could immediately trigger invalidation and then
re-program the HW with this stale data.
> > The use of atomic looks probably wrong as well.
>
> Do you mean umem_dmabuf->notifier_seq? Could you elaborate the concern?
It only increments once per invalidation, that usually is racy.
> > > + total_pages = ib_umem_odp_num_pages(umem_odp);
> > > + for_each_sg(umem->sg_head.sgl, sg, umem->sg_head.nents, j) {
> > > + addr = sg_dma_address(sg);
> > > + pages = sg_dma_len(sg) >> page_shift;
> > > + while (pages > 0 && k < total_pages) {
> > > + umem_odp->dma_list[k++] = addr | access_mask;
> > > + umem_odp->npages++;
> > > + addr += page_size;
> > > + pages--;
> >
> > This isn't fragmenting the sg into a page list properly, won't work for unaligned things
>
> I thought the addresses are aligned, but will add explicit alignment here.
I have no idea what comes out of dma_buf, I wouldn't make too many
assumptions since it does have to pass through the IOMMU layer too
> > And really we don't need the dma_list for this case, with a fixed
> > whole mapping DMA SGL a normal umem sgl is OK and the normal umem
> > XLT programming in mlx5 is fine.
>
> The dma_list is used by both "polulate_mtt()" and
> "mlx5_ib_invalidate_range", which are used for XLT programming and
> invalidating (zapping), respectively.
Don't use those functions for the dma_buf flow.
Jason
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
next prev parent reply other threads:[~2020-10-06 7:32 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-10-04 19:12 [RFC PATCH v3 0/4] RDMA: Add dma-buf support Jianxin Xiong
2020-10-04 19:12 ` [RFC PATCH v3 1/4] RDMA/umem: Support importing dma-buf as user memory region Jianxin Xiong
2020-10-05 10:54 ` Christian König
2020-10-05 16:19 ` Xiong, Jianxin
2020-10-05 13:13 ` Jason Gunthorpe
2020-10-05 16:18 ` Xiong, Jianxin
2020-10-05 16:33 ` Jason Gunthorpe [this message]
2020-10-05 19:41 ` Xiong, Jianxin
2020-10-06 9:22 ` Daniel Vetter
2020-10-06 15:26 ` Xiong, Jianxin
2020-10-06 15:49 ` Jason Gunthorpe
2020-10-06 16:34 ` Daniel Vetter
2020-10-06 17:24 ` Daniel Vetter
2020-10-06 18:02 ` Jason Gunthorpe
2020-10-06 18:17 ` Daniel Vetter
2020-10-06 18:38 ` Jason Gunthorpe
2020-10-06 19:12 ` Daniel Vetter
2020-10-07 7:13 ` Christian König
2020-10-06 16:40 ` Daniel Vetter
2020-10-04 19:12 ` [RFC PATCH v3 2/4] RDMA: Expand driver memory registration methods to support dma-buf Jianxin Xiong
2020-10-04 19:12 ` [RFC PATCH v3 3/4] RDMA/mlx5: Support dma-buf based userspace memory region Jianxin Xiong
2020-10-04 19:12 ` [RFC PATCH v3 4/4] RDMA/uverbs: Add uverbs command for dma-buf based MR registration Jianxin Xiong
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20201005163304.GU9916@ziepe.ca \
--to=jgg@ziepe.ca \
--cc=christian.koenig@amd.com \
--cc=daniel.vetter@intel.com \
--cc=dledford@redhat.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=jianxin.xiong@intel.com \
--cc=leon@kernel.org \
--cc=linux-rdma@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).