<div dir="ltr">Hi guys,<div><br></div><div>I just recalled, that for the SLES12 SP3 requirements, I agreed to 'downgrade' armada-8040-mcbin.dtb, so that it could use older DT bindings. Maybe this is the reason of the observed inconsistencies? It's enough to take v4.15 dtb, put it in Platforms/Marvell/Armada/armada-8040-mcbin.dtb, rebuild and reflash the image. I'm curious if it can help anything. When I boot the mainline kernel with mainline dtb (or ACPI) everything seems fine to me, although I admit, that apart from iperf3, I didn't stress the network much (however all apt install succeeded).</div><div><br></div><div>Best regards,</div><div>Marcin</div></div><div class="gmail_extra"><br><div class="gmail_quote">2018-01-25 13:44 GMT+01:00 Marc Zyngier <span dir="ltr"><<a href="mailto:marc.zyngier@arm.com" target="_blank">marc.zyngier@arm.com</a>></span>:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5">On 24/01/18 09:43, Marcin Wojtas wrote:<br>
> +Antoine<br>
><br>
> Hi Marc,<br>
><br>
> 2018-01-24 10:25 GMT+01:00 Marc Zyngier <<a href="mailto:marc.zyngier@arm.com">marc.zyngier@arm.com</a>>:<br>
>> On 23/01/18 23:15, Ard Biesheuvel wrote:<br>
>>> On 23 January 2018 at 23:12, Steve McIntyre <<a href="mailto:steve@einval.com">steve@einval.com</a>> wrote:<br>
>>>> On Tue, Jan 23, 2018 at 11:07:53PM +0000, Matthew Bentham wrote:<br>
>>>>> Hi,<br>
>>>>> Thanks. I've got it showing a Linux terminal using an old Radeon 5570 now :-)<br>
>>>>><br>
>>>>> My problem now is that the network device driver is causing a kernel panic when<br>
>>>>> doing "apt install task-gnome-desktop". I get a lot of warning messages first<br>
>>>>> along the lines of "detected wrong cpu at end of Tx processing".<br>
>>>><br>
>>>> Aha! Not just me then! I've been seeing the same 2 things for a while,<br>
>>>> but Leif and Stuart have not.<br>
>>>><br>
>>>> I was wondering if it's memory corruption in the driver, but I've not<br>
>>>> had much chance to dig into it yet.<br>
>>>><br>
>>><br>
>>> I think this may be a known issue. Marc?<br>
>><br>
>> I fixed something along those lines[1]. But this driver is being<br>
>> randomly changed and almost never tested in a production environment, so<br>
>> it it breaks on almost every single effin' change.<br>
>><br>
>> 4.14 should have the fix though... If it still breaks, I'll dust off the<br>
>> board and give it a spin.<br>
>><br>
>> M.<br>
>><br>
>> [1]<br>
>> <a href="https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/net/ethernet/marvell/mvpp2.c?id=13c249a94f525fe4c757d28854049780b25605c4" rel="noreferrer" target="_blank">https://git.kernel.org/pub/<wbr>scm/linux/kernel/git/torvalds/<wbr>linux.git/commit/drivers/net/<wbr>ethernet/marvell/mvpp2.c?id=<wbr>13c249a94f525fe4c757d288540497<wbr>80b25605c4</a><br>
>> --<br>
><br>
> I saw your patch is already a part of v4.14... Two days ago similar<br>
> bug report was sent to the lists and I Antoine said it will be<br>
> investigated.<br>
<br>
</div></div>I've just booted with 4.15-rc9, and I don't see the TX issue any more,<br>
which is what I expected with the above fix.<br>
<br>
What I am seeing is this:<br>
<br>
[ 11.849032] IPv6: ADDRCONF(NETDEV_UP): eth1: link is not ready<br>
[ 12.720977] IPv6: ADDRCONF(NETDEV_UP): eth0: link is not ready<br>
[ 12.809757] IPv6: ADDRCONF(NETDEV_UP): eth2: link is not ready<br>
[ 13.762656] mvpp2 f2000000.ethernet eth0: Link is Up - 1Gbps/Full - flow control rx/tx<br>
[ 13.770787] IPv6: ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready<br>
[ 14.639318] BUG: Bad page state in process swapper/0 pfn:e6804<br>
[ 14.645296] page:ffff7fd8249a0100 count:0 mapcount:0 mapping: (null) index:0xfffff609268049c0<br>
[ 14.654659] flags: 0xfffe00000000100(slab)<br>
[ 14.658781] raw: 0fffe00000000100 0000000000000000 fffff609268049c0 0000000000400000<br>
[ 14.666563] raw: ffff7fd8232e5420 fffff60927bf0610 fffff6092b546000 0000000000000000<br>
[ 14.674348] page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set<br>
[ 14.680818] bad because of flags: 0x100(slab)<br>
[ 14.685191] Modules linked in:<br>
[ 14.688271] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc9-00124-g9f55a2093a18 #8070<br>
[ 14.696568] Hardware name: Marvell 8040 MACHIATOBin (DT)<br>
[ 14.701900] Call trace:<br>
[ 14.704362] dump_backtrace+0x0/0x1c0<br>
[ 14.708039] show_stack+0x24/0x30<br>
[ 14.711367] dump_stack+0x8c/0xb0<br>
[ 14.714695] bad_page+0xf0/0x158<br>
[ 14.717935] free_pages_check_bad+0x80/0xa0<br>
[ 14.722134] __free_pages_ok+0x24c/0x340<br>
[ 14.726071] page_frag_free+0x78/0x88<br>
[ 14.729749] skb_free_head+0x38/0x48<br>
[ 14.733337] skb_release_data+0x140/0x158<br>
[ 14.737362] skb_release_all+0x30/0x40<br>
[ 14.741126] consume_skb+0x38/0x110<br>
[ 14.744630] arp_process.constprop.4+0x158/<wbr>0x778<br>
[ 14.749264] arp_rcv+0xf0/0x118<br>
[ 14.752418] __netif_receive_skb_core+<wbr>0x5ac/0x8a8<br>
[ 14.757140] __netif_receive_skb+0x28/0x78<br>
[ 14.761252] netif_receive_skb_internal+<wbr>0x30/0x120<br>
[ 14.766060] napi_gro_receive+0x15c/0x188<br>
[ 14.770085] mvpp2_poll+0x37c/0x678<br>
[ 14.773586] net_rx_action+0x108/0x390<br>
[ 14.777349] __do_softirq+0x120/0x370<br>
[ 14.781025] irq_exit+0x90/0xb0<br>
[ 14.784179] __handle_domain_irq+0x90/0xf8<br>
[ 14.788291] gic_handle_irq+0x60/0xb0<br>
[ 14.791966] el1_irq+0xb0/0x128<br>
[ 14.795119] arch_cpu_idle+0x28/0x1c8<br>
[ 14.798796] do_idle+0x138/0x1f0<br>
[ 14.802036] cpu_startup_entry+0x28/0x30<br>
[ 14.805973] rest_init+0xd4/0xe0<br>
[ 14.809216] start_kernel+0x3a4/0x3b8<br>
[ 14.812911] Disabling lock debugging due to kernel taint<br>
[ 15.957032] mvpp2 f4000000.ethernet eth2: Link is Up - 1Gbps/Full - flow control rx/tx<br>
[ 15.965006] IPv6: ADDRCONF(NETDEV_CHANGE): eth2: link becomes ready<br>
[ 46.036434] device eth0 entered promiscuous mode<br>
<br>
which happens if you attach a macvtap to the network interface<br>
(in this case eth0). As this is the only system where this happens,<br>
I assume the Ethernet driver is doing something pretty bad when<br>
it comes to managing the skb life cycle. None of the non-Marvell<br>
systems I have access to explode this way...<br>
<br>
Thanks,<br>
<br>
M.<br>
<div class="HOEnZb"><div class="h5">--<br>
Jazz is not dead. It just smells funny...<br>
</div></div></blockquote></div><br></div>