From: Alexander Bluhm <bluhm@openbsd.org>
Subject: Re: igc(4): fix recv. jumbo frames
To: David Gwynne <david@gwynne.id.au>
Cc: Jan Klemkow <jan@openbsd.org>, tech@openbsd.org, Theo de Raadt <deraadt@openbsd.org>
Date: Sun, 11 Aug 2024 13:39:32 +0200

On Sat, Aug 10, 2024 at 02:53:58PM +1000, David Gwynne wrote:
> On Fri, Aug 09, 2024 at 08:38:37PM +0200, Alexander Bluhm wrote:
> > 
> > On sparc64 it does not panic.  Jumbo frames are not well supported
> > on this machine, I guess some bugs in the IOMMU implementation.
> > Other drivers also have problems with jumbo on sparc64.  Only part
> > of my tests pass.  This is not a regression.
> 
> does this help?

Maybe.  A lot of my network tests on sparc64 fail.  Others are not
stable and results flip around.

http://bluhm.genua.de/netlink/results/2024-08-04T11:37:11Z/netlink.html

This diff might fix some corner cases.  Or it is just coincidence
that some tests passed in this run.  Jumbo in gneneral looks quite
broken, igc is a bit worse than other drivers on sparc64.

bluhm

> Index: machdep.c
> ===================================================================
> RCS file: /cvs/src/sys/arch/sparc64/sparc64/machdep.c,v
> diff -u -p -r1.218 machdep.c
> --- machdep.c	22 May 2024 05:51:49 -0000	1.218
> +++ machdep.c	10 Aug 2024 04:51:22 -0000
> @@ -993,7 +993,7 @@ _bus_dmamap_load_mbuf(bus_dma_tag_t t, b
>  
>  			if (i > 0 && pa == (segs[i - 1].ds_addr +
>  			    segs[i - 1].ds_len) && ((segs[i - 1].ds_len + incr)
> -			    < map->_dm_maxsegsz)) {
> +			    <= map->_dm_maxsegsz)) {
>  				/* Hey, waddyaknow, they're contiguous */
>  				segs[i - 1].ds_len += incr;
>  				continue;