4.17.0-10146-gf0dc7f9c6dd9: hw csum failure on powerpc+sungem

[Mathieu Malaterre]

Hi all,

On Tue, Jun 12, 2018 at 10:15 AM Mathieu Malaterre <malat at debian.org> wrote:
>
> Hi Balbir,
>
> On Tue, Jun 12, 2018 at 9:39 AM Balbir Singh <bsingharora at gmail.com> wrote:
> >
> >
> > On 12/06/18 06:20, Mathieu Malaterre wrote:
> >
> > > Hi Meelis,
> > >
> > > On Mon, Jun 11, 2018 at 1:21 PM Meelis Roos <mroos at linux.ee> wrote:
> > >> I am seeing this on PowerMac G4 with sungem ethernet driver. 4.17 was
> > >> OK, 4.17.0-10146-gf0dc7f9c6dd9 is problematic.
> > > Same here.
> > >
> > >> [  140.518664] eth0: hw csum failure
> > >> [  140.518699] CPU: 0 PID: 1237 Comm: postconf Not tainted 4.17.0-10146-gf0dc7f9c6dd9 #83
> > >> [  140.518707] Call Trace:
> > >> [  140.518734] [effefd90] [c03d6db8] __skb_checksum_complete+0xd8/0xdc (unreliable)
> > >> [  140.518759] [effefdb0] [c04c1284] icmpv6_rcv+0x248/0x4ec
> > >> [  140.518775] [effefdd0] [c049a448] ip6_input_finish.constprop.0+0x11c/0x5f4
> > >> [  140.518786] [effefe10] [c049b1c0] ip6_mc_input+0xcc/0x100
> > >> [  140.518807] [effefe20] [c03e110c] __netif_receive_skb_core+0x310/0x944
> > >> [  140.518820] [effefe70] [c03e76ec] napi_gro_receive+0xd0/0xe8
> > >> [  140.518845] [effefe80] [f3e1f66c] gem_poll+0x618/0x1274 [sungem]
> > >> [  140.518856] [effeff30] [c03e6f0c] net_rx_action+0x198/0x374
> > >> [  140.518872] [effeff90] [c0501a88] __do_softirq+0x120/0x278
> > >> [  140.518890] [effeffe0] [c0036188] irq_exit+0xd8/0xdc
> > >> [  140.518908] [effefff0] [c000f478] call_do_irq+0x24/0x3c
> > >> [  140.518925] [d05a5d30] [c0007120] do_IRQ+0x74/0xf0
> > >> [  140.518941] [d05a5d50] [c0012474] ret_from_except+0x0/0x14
> > >> [  140.518960] --- interrupt: 501 at copy_page+0x40/0x90
> > >>                    LR = copy_user_page+0x18/0x30
> > >> [  140.518973] [d05a5e10] [d058cd80] 0xd058cd80 (unreliable)
> > >> [  140.518989] [d05a5e20] [c00fa2bc] wp_page_copy+0xec/0x654
> > >> [  140.519002] [d05a5e60] [c00fd3a4] do_wp_page+0xa8/0x5b4
> > >> [  140.519013] [d05a5e90] [c00fe934] handle_mm_fault+0x564/0xa84
> > >> [  140.519025] [d05a5f00] [c0016230] do_page_fault+0x1bc/0x7e8
> > >> [  140.519037] [d05a5f40] [c0012300] handle_page_fault+0x14/0x40
> > >> [  140.519048] --- interrupt: 301 at 0xb78b6864
> > >>                    LR = 0xb78b6c54
> > >>
> > > For some reason if I do a git bisect it returns that:
> > >
> > > $ git bisect good
> > > 3036bc45364f98515a2c446d7fac2c34dcfbeff4 is the first bad commit
> > >
> > > Could you also check on your side please.
> >
> > Don't have a G4, but the related commits look like
> >
> > 373e098e1e788d7b89ec0f31765a6c08e2ea0f7c and e9c4943a107b56696e4872cdffdba6b0c7277c77 Balbir
> >
>
> Indeed that makes more sense. I must have messed up during my
> git-bisect operation. Will try a simple git-revert on those and report
> back.

Here is what I get.

Current git HEAD is: 0c14e43a42e4e44f70963f8ccf89461290c4e4da if I do:

$ git revert e9c4943a107b56696e4872cdffdba6b0c7277c77.
$ git revert 373e098e1e788d7b89ec0f31765a6c08e2ea0f7c
$ rm -rf g4
$ make O=g4 ARCH=powerpc g4_defconfig
$ make -j8 O=g4 ARCH=powerpc CROSS_COMPILE=powerpc-linux-gnu- bindeb-pkg

-> I can still see the error in dmesg.

> Thanks