On Mon, Jul 23, 2018 at 05:19:41PM -0500, Shiraz Saleem wrote:
> On Mon, Jul 23, 2018 at 02:39:56PM -0600, Jason Gunthorpe wrote:
> > On Sat, Jul 21, 2018 at 08:31:30AM -0500, Shiraz Saleem wrote:
> > > On Sun, Jul 01, 2018 at 10:36:24AM -0600, Leon Romanovsky wrote:
> > > > From: Leon Romanovsky <leonro@xxxxxxxxxxxx>
> > > >
> > > > VMA lookup is supposed to be performed while mmap_sem is held.
> > > >
> > > > Fixes: f26c7c83395b ("i40iw: Add 2MB page support")
> > > > Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxxxx>
> > > >
> > > Thanks for fixing this bug.
> > >
> > > Acked-by: Shiraz Saleem <shiraz.saleem@xxxxxxxxx>
> >
> > This code in the driver all looks wrong to me, it should not rely on
> > the huge pages flag and it shouldn't do this VM walking.
> >
> Hi Jason - Why is that we cant use umem->hugetlb to determine if
> MR is composed of huge pages? Isnt that why the flag was defined for
> in the first place. All the VMAs w.r.t the user-space buffer are
> scanned to make this determination in ib_umem_get.
> Isnt it better to use it than implement it in individual driver.
It was a mistake to have this flag and all drivers except i40iw and
bnxt don't use it. I wish to remove it.
Drivers are supposed to upgrade the page order by looking for runs of
suitable physical addresses after DMA translation, not by inspecting
the VMA. Something like mlx5_ib_cont_pages() and related
The core could provde support code for this common algorithm..
Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
