[PATCH net-next] stmmac: align RX buffers

[Marc Zyngier]

On Fri, 20 Aug 2021 19:14:22 +0100,
Matteo Croce <mcroce at linux.microsoft.com> wrote:
> 
> On Fri, Aug 20, 2021 at 8:09 PM Marc Zyngier <maz at kernel.org> wrote:
> >
> > On Fri, 20 Aug 2021 18:56:33 +0100,
> > Matteo Croce <mcroce at linux.microsoft.com> wrote:
> > >
> > > On Fri, Aug 20, 2021 at 7:51 PM Marc Zyngier <maz at kernel.org> wrote:
> > > >
> > > > On Fri, 20 Aug 2021 18:35:45 +0100,
> > > > Matteo Croce <mcroce at linux.microsoft.com> wrote:
> > > > >
> > > > > > > I think it's wrong. The original offset was 0, and to align it to the
> > > > > > > boundary we need to add just NET_IP_ALIGN, which is two.
> > > > > > > NET_SKB_PAD is a much bigger value, (I think 64), which is used to
> > > > > > > reserve space to prepend an header, e.g. with tunnels.
> > > > > >
> > > > > > How about the other adjustments that Eric mentioned regarding the size
> > > > > > of the buffer? Aren't they required?
> > > > > >
> > > > >
> > > > > I guess that if stmmac_rx_buf1_len() needed such adjustment, it would
> > > > > be already broken when XDP is in use.
> > > > > When you use XDP, stmmac_rx_offset() adds a pretty big headroom of 256
> > > > > byte, which would easily trigger an overflow if not accounted.
> > > > > Did you try attaching a simple XDP program on a stock 5.13 kernel?
> > > >
> > > > Yes, as mentioned in [1], to which you replied...
> > > >
> > > >         M.
> > > >
> > > > [1] https://lore.kernel.org/r/87wnohqty1.wl-maz@kernel.org
> > > >
> > >
> > > Great.
> > > So I doubt that the adjustment is needed.
> > > Does it work with all the frame size?
> >
> > I have no idea. Honestly, you are the one who should be able to answer
> > these questions, given that you should have worked out how the buffer
> > allocations work in this particular driver.
> >
> > This whole "let's try another random set of values until something
> > sticks" is not how things ought to be done, and doesn't fill me with
> > the utmost confidence that 5.14 (which apparently may well be cut in
> > *two days*) is going to have a solid stmmac driver.
> >
> > I re-re-request that this patch gets reverted until you figure out
> > what is wrong with the initial patch.
> >
> > Thanks,
> >
> 
> I would have done it, but I'll not have the hardware until next week at least,
> otherwise I'd have tried all these tests myself.
> 
> I'm sure that NET_SKB_PAD doesn't need to be there, if just removing
> it fixes the problem, consider applying it and put a Fixes tag.

No, I don't think that's the right thing to do. A patch breaks a
driver, and the author of the patch is not in a position to fix it.
That's OK, these things happen, it's just bad timing.

But I don't understand this part of the kernel well enough to submit a
patch based on a sample of *one*, at the last minute, just because "it
works for me", and have the confidence that it doesn't break anything
else.

I have now posted a revert of the original patch[1]. I'll be happy to
work with you, with a less pressure, in order to have something that
works for everyone in the next cycle.

Thanks,

	M.

[1] https://lore.kernel.org/netdev/20210820183002.457226-1-maz@kernel.org

-- 
Without deviation from the norm, progress is not possible.