Date:   Thu, 8 Aug 2019 22:08:08 +0200
From:   Heiner Kallweit <hkallweit1@...il.com>
To:     Holger Hoffstätte <holger@...lied-asynchrony.com>,
        Realtek linux nic maintainers <nic_swsd@...ltek.com>,
        David Miller <davem@...emloft.net>
Cc:     "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        Sander Eikelenboom <linux@...elenboom.it>,
        Eric Dumazet <edumazet@...gle.com>
Subject: Re: [PATCH net-next] r8169: make use of xmit_more

On 08.08.2019 21:52, Holger Hoffstätte wrote:
> On 8/8/19 8:17 PM, Heiner Kallweit wrote:
>> On 08.08.2019 17:53, Holger Hoffstätte wrote:
>>> On 8/8/19 4:37 PM, Holger Hoffstätte wrote:
>>>>
>>>> Hello Heiner -
>>>>
>>>> On 7/28/19 11:25 AM, Heiner Kallweit wrote:
>>>>> There was a previous attempt to use xmit_more, but the change had to be
>>>>> reverted because under load sometimes a transmit timeout occurred [0].
>>>>> Maybe this was caused by a missing memory barrier, the new attempt
>>>>> keeps the memory barrier before the call to netif_stop_queue like it
>>>>> is used by the driver as of today. The new attempt also changes the
>>>>> order of some calls as suggested by Eric.
>>>>>
>>>>> [0] https://lkml.org/lkml/2019/2/10/39
>>>>>
>>>>> Signed-off-by: Heiner Kallweit <hkallweit1@...il.com>
>>>>
>>>> I decided to take one for the team and merged this into my 5.2.x tree (just
>>>> fixing up the path) and it has been working fine for the last 2 weeks in two
>>>> machines..until today, when for the first time in forever some random NFS traffic
>>>> made this old friend come out from under the couch:
>>>>
>>>> [Aug 8 14:13] ------------[ cut here ]------------
>>>> [  +0.000006] NETDEV WATCHDOG: eth0 (r8169): transmit queue 0 timed out
>>>> [  +0.000021] WARNING: CPU: 3 PID: 0 at net/sched/sch_generic.c:442 dev_watchdog+0x21f/0x230
>>>> [  +0.000001] Modules linked in: lz4 lz4_compress lz4_decompress nfsd auth_rpcgss oid_registry lockd grace sunrpc sch_fq_codel btrfs xor zstd_compress raid6_pq zstd_decompress bfq jitterentropy_rng nct6775 hwmon_vid coretemp hwmon x86_pkg_temp_thermal aesni_intel aes_x86_64 i915 glue_helper crypto_simd cryptd i2c_i801 intel_gtt i2c_algo_bit iosf_mbi drm_kms_helper syscopyarea usbhid sysfillrect r8169 sysimgblt fb_sys_fops realtek drm libphy drm_panel_orientation_quirks i2c_core video backlight mq_deadline
>>>> [  +0.000026] CPU: 3 PID: 0 Comm: swapper/3 Not tainted 5.2.7 #1
>>>> [  +0.000001] Hardware name: System manufacturer System Product Name/P8Z68-V LX, BIOS 4105 07/01/2013
>>>> [  +0.000004] RIP: 0010:dev_watchdog+0x21f/0x230
>>>> [  +0.000002] Code: 3b 00 75 ea eb ad 4c 89 ef c6 05 1c 45 bd 00 01 e8 66 35 fc ff 44 89 e1 4c 89 ee 48 c7 c7 e8 5e fc 81 48 89 c2 e8 90 df 92 ff <0f> 0b eb 8e 66 66 2e 0f 1f 84 00 00 00 00 00 66 90 66 66 66 66 90
>>>> [  +0.000002] RSP: 0018:ffffc90000118e68 EFLAGS: 00010286
>>>> [  +0.000002] RAX: 0000000000000000 RBX: ffff8887f7837600 RCX: 0000000000000303
>>>> [  +0.000001] RDX: 0000000000000001 RSI: 0000000000000092 RDI: ffffffff827a488c
>>>> [  +0.000001] RBP: ffff8887f9fbc440 R08: 0000000000000303 R09: 0000000000000003
>>>> [  +0.000001] R10: 000000000001004c R11: 0000000000000001 R12: 0000000000000000
>>>> [  +0.000009] R13: ffff8887f9fbc000 R14: ffffffff8173aa20 R15: dead000000000200
>>>> [  +0.000001] FS:  0000000000000000(0000) GS:ffff8887ff580000(0000) knlGS:0000000000000000
>>>> [  +0.000000] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>> [  +0.000001] CR2: 00007f8d1c04d000 CR3: 0000000002209001 CR4: 00000000000606e0
>>>> [  +0.000000] Call Trace:
>>>> [  +0.000002]  <IRQ>
>>>> [  +0.000005]  call_timer_fn+0x2b/0x120
>>>> [  +0.000002]  expire_timers+0xa4/0x100
>>>> [  +0.000001]  run_timer_softirq+0x8c/0x170
>>>> [  +0.000002]  ? __hrtimer_run_queues+0x13a/0x290
>>>> [  +0.000003]  ? sched_clock_cpu+0xe/0x130
>>>> [  +0.000003]  __do_softirq+0xeb/0x2de
>>>> [  +0.000003]  irq_exit+0x9d/0xe0
>>>> [  +0.000002]  smp_apic_timer_interrupt+0x60/0x110
>>>> [  +0.000003]  apic_timer_interrupt+0xf/0x20
>>>> [  +0.000001]  </IRQ>
>>>> [  +0.000003] RIP: 0010:cpuidle_enter_state+0xad/0x930
>>>> [  +0.000001] Code: c5 66 66 66 66 90 31 ff e8 90 99 9e ff 80 7c 24 0b 00 74 12 9c 58 f6 c4 02 0f 85 39 08 00 00 31 ff e8 e7 26 a2 ff fb 45 85 e4 <0f> 88 34 02 00 00 49 63 cc 4c 2b 2c 24 48 8d 04 49 48 c1 e0 05 8b
>>>> [  +0.000000] RSP: 0018:ffffc9000008be50 EFLAGS: 00000202 ORIG_RAX: ffffffffffffff13
>>>> [  +0.000001] RAX: ffff8887ff5a9180 RBX: ffffffff822b6c40 RCX: 000000000000001f
>>>> [  +0.000001] RDX: 0000000000000000 RSI: 0000000033087154 RDI: 0000000000000000
>>>> [  +0.000001] RBP: ffff8887ff5b1310 R08: 000030d021fae397 R09: ffff8887ff59c8c0
>>>> [  +0.000000] R10: ffff8887ff59c8c0 R11: 0000000000000006 R12: 0000000000000004
>>>> [  +0.000001] R13: 000030d021fae397 R14: 0000000000000004 R15: ffff8887fc281600
>>>> [  +0.000001]  cpuidle_enter+0x29/0x40
>>>> [  +0.000002]  do_idle+0x1e5/0x280
>>>> [  +0.000001]  cpu_startup_entry+0x19/0x20
>>>> [  +0.000002]  start_secondary+0x186/0x1c0
>>>> [  +0.000001]  secondary_startup_64+0xa4/0xb0
>>>> [  +0.000001] ---[ end trace 99493c768580f4fd ]---
>>>>
>>>> The device is:
>>>>
>>>> Aug  7 23:19:09 tux kernel: libphy: r8169: probed
>>>> Aug  7 23:19:09 tux kernel: r8169 0000:04:00.0 eth0: RTL8168evl/8111evl, c8:60:00:68:33:cc, XID 2c9, IRQ 36
>>>> Aug  7 23:19:09 tux kernel: r8169 0000:04:00.0 eth0: jumbo features [frames: 9200 bytes, tx checksumming: ko]
>>>> Aug  7 23:19:12 tux kernel: RTL8211E Gigabit Ethernet r8169-400:00: attached PHY driver [RTL8211E Gigabit Ethernet] (mii_bus:phy_addr=r8169-400:00, irq=IGNORE)
>>>> Aug  7 23:19:13 tux kernel: r8169 0000:04:00.0 eth0: No native access to PCI extended config space, falling back to CSI
>>>>
>>>> and using fq_codel, of course.
>>>>
>>>> This cpuidle hiccup used to be completely gone without xmit_more and this was
>>>> the first (and so far only) time since merging it (regardless of load).
>>>> Also, while I'm using BMQ as CPU scheduler, that hasn't made a difference for
>>>> this particular problem in the past (with MuQSS/PDS) either; way back when I had
>>>> Eric's previous attempt(s) it also hiccupped with CFS.
>>>>
>>>> Revert or wait for more reports when -next is merged in 5.4?
>>>
>>> Another question/data point: I've had the whole basket of offloads activated:
>>>
>>>    ethtool --offload eth0 rx on tx on gro on gso on sg on tso on
>>>
>>> and this caused zero problems without the xmit_more patch. However I just saw
>>> that net-next has a patch where TSO is disabled due to a known HW defect in
>>> RTL8168evl, which is of course what I have. Could this be the reason for the
>>> stall/hiccup when xmit_more has its fingers in the pie? I kind of know what
>>> xmit_more does, just not how it could interact with a possibly broken TSO that
>>> nevertheless seems to work fine otherwise..
>>>
>>
>> I was about to ask exactly that, whether you have TSO enabled. I don't know what
>> can trigger the HW issue, it was just confirmed by Realtek that this chip version
>> has a problem with TSO. So the logical conclusion is: test w/o TSO, ideally the
>> linux-next version.
> 
> So disabling TSO alone didn't work - it leads to reduced throughout (~70 MB/s in iperf).
> Instead I decided to backport 93681cd7d94f ("r8169: enable HW csum and TSO"), which
> wasn't easy due to cleanups/renamings of dependencies, but I managed to backport
> it and .. got the same problem of reduced throughout. wat?!
> 
> After lots of trial & error I started disabling all offloads and finally found
> that sg (Scatter-Gather) enabled alone - without TSO - will lead to the throughput
> drop. So the culprit seems 93681cd7d94f, which disabled TSO on my NIC, but left
> sg on by default. This weas repeatable - switch on sg, throughput drop; turn it
> off - smooth sailing, now with reduced buffers.
> 
> I modified the relevant bits to disable tso & sg like this:
> 
>     /* RTL8168e-vl has a HW issue with TSO */
>     if (tp->mac_version == RTL_GIGA_MAC_VER_34) {
> +        dev->vlan_features &= ~(NETIF_F_ALL_TSO|NETIF_F_SG);
> +        dev->hw_features &= ~(NETIF_F_ALL_TSO|NETIF_F_SG);
> +        dev->features &= ~(NETIF_F_ALL_TSO|NETIF_F_SG);
>     }
> 
> This seems to work since it restores performance without sg/tso by default
> and without any additional offloads, yet with xmit_more in the mix.
> We'll see whether that is stable over the next few days, but I strongly
> suspect it will be good and that the hiccups were due to xmit_more/TSO
> interaction.
> 
Thanks a lot for the analysis and testing. Then I'll submit the disabling
of SG on RTL8168evl (on your behalf), independent of whether it fixes
the timeout issue.

> thanks,
> Holger
> 
Heiner

Powered by blists - more mailing lists