Re: [PATCH net] gve: Fix an edge case for TSO skb validity check

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Jul 19, 2024 at 7:31 AM Praveen Kaligineedi
<pkaligineedi@xxxxxxxxxx> wrote:
>
> On Thu, Jul 18, 2024 at 8:47 PM Willem de Bruijn
> <willemdebruijn.kernel@xxxxxxxxx> wrote:
> >
> > On Thu, Jul 18, 2024 at 9:52 PM Praveen Kaligineedi
> > <pkaligineedi@xxxxxxxxxx> wrote:
> > >
> > > On Thu, Jul 18, 2024 at 4:07 PM Willem de Bruijn
> > > <willemdebruijn.kernel@xxxxxxxxx> wrote:
> > >
> > > > > +                      * segment, then it will count as two descriptors.
> > > > > +                      */
> > > > > +                     if (last_frag_size > GVE_TX_MAX_BUF_SIZE_DQO) {
> > > > > +                             int last_frag_remain = last_frag_size %
> > > > > +                                     GVE_TX_MAX_BUF_SIZE_DQO;
> > > > > +
> > > > > +                             /* If the last frag was evenly divisible by
> > > > > +                              * GVE_TX_MAX_BUF_SIZE_DQO, then it will not be
> > > > > +                              * split in the current segment.
> > > >
> > > > Is this true even if the segment did not start at the start of the frag?
> > > The comment probably is a bit confusing here. The current segment
> > > we are tracking could have a portion in the previous frag. The code
> > > assumed that the portion on the previous frag (if present) mapped to only
> > > one descriptor. However, that portion could have been split across two
> > > descriptors due to the restriction that each descriptor cannot exceed 16KB.
> >
> > >>> /* If the last frag was evenly divisible by
> > >>> +                                * GVE_TX_MAX_BUF_SIZE_DQO, then it will not be
> > >>>  +                              * split in the current segment.
> >
> > This is true because the smallest multiple of 16KB is 32KB, and the
> > largest gso_size at least for Ethernet will be 9K. But I don't think
> > that that is what is used here as the basis for this statement?
> >
> The largest Ethernet gso_size (9K) is less than GVE_TX_MAX_BUF_SIZE_DQO
> is an implicit assumption made in this patch and in that comment. Bailey,
> please correct me if I am wrong..

If last_frag_size is evenly divisible by GVE_TX_MAX_BUF_SIZE_DQO, it
doesn't hit the edge case we're looking for.

- If it's evenly divisible, then we know it will use exactly
(last_frag_size / GVE_TX_MAX_BUF_SIZE_DQO) descriptors
- GVE_TX_MAX_BUF_SIZE_DQO > 9k, so we know each descriptor won't
create a segment which exceeds the limit





[Index of Archives]     [Linux Kernel]     [Kernel Development Newbies]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux