Re: [PATCH] net: usb: Correct PHY handling of smsc95xx

[Ferry Toth <fntoth@gmail.com>]

Op 01-12-2021 om 11:56 schreef Martyn Welch:
> On Tue, 2021-11-30 at 22:47 +0100, Ferry Toth wrote:
>> Op 30-11-2021 om 19:52 schreef Martyn Welch:
>>> On Mon, 2021-11-29 at 15:16 +0100, Ferry Toth wrote:
>>>> Hi,
>>>> Op 29-11-2021 om 13:08 schreef Martyn Welch:
>>>>    
>>>>> On Sat, 2021-11-27 at 18:48 +0100, Ferry Toth wrote:
>>>>>    
>>>>>>
>>>>>> The patch introduces a new error on each unplug:
>>>>>>
>>>>>> usb 1-1: USB disconnect, device number 2
>>>>>> usb 1-1.1: USB disconnect, device number 3
>>>>>> smsc95xx 1-1.1:1.0 eth0: unregister 'smsc95xx' usb-xhci-
>>>>>> hcd.1.auto-
>>>>>> 1.1,
>>>>>> smsc95xx USB 2.0 Ethernet
>>>>>> smsc95xx 1-1.1:1.0 eth0: Link is Down
>>>>>> smsc95xx 1-1.1:1.0 eth0: Failed to read reg index 0x00000114: -
>>>>>> 19
>>>>>> smsc95xx 1-1.1:1.0 eth0: Error reading MII_ACCESS
>>>>>> smsc95xx 1-1.1:1.0 eth0: __smsc95xx_mdio_read: MII is busy
>>>>>> smsc95xx 1-1.1:1.0 eth0: Failed to read reg index 0x00000114: -
>>>>>> 19
>>>>>> smsc95xx 1-1.1:1.0 eth0: Error reading MII_ACCESS
>>>>>> smsc95xx 1-1.1:1.0 eth0: __smsc95xx_mdio_read: MII is busy
>>>>>> smsc95xx 1-1.1:1.0 eth0: hardware isn't capable of remote
>>>>>> wakeup
>>>>>>
>>>>> Agh! Somehow missed that. I'm looking into it...
>>>>    That would be great!
>>>>    
>>>
>>> They appear as a result of phy_disconnect() being called in unbind()
>>> when the hardware has already been disconnected (which is kinda
>>> likely
>>> to physically be the case with USB devices...). The PHY is not going
>>> to
>>> be accessible, but such calls *are* needed for instances where we are
>>> unbinding without the device having been physically removed and want
>>> to
>>> put the device in a suitable state. Failing with -ENODEV (as it
>>> currently does) seems to be the right thing to do.
>>>
>>> I wonder whether removing some of these error messages might be an
>>> option? It appears that some of them are present in other drivers,
>>> but
>>> I don't know whether such messages get displayed when that hardware
>>> is
>>> unplugged too or whether I'm missing something that protects against
>>> that.
>> I just retried 5.10.63 and found it did not yet have the crash that you
>> fix here. But it did have the above Error reading MII_ACCESS. So
>> apparently this is more of an older annoyance since smsc95xx v2.00?
> 
> That's an interesting observation. Trying the v4.19 stable branch I see
> that these errors aren't generated, however it seems that the driver at
> that point may not be making any attempt to stop the PHY when the
> driver is unbound/disconnected, which explains the lack of errors.

If I'm not mistaken v2.0.0 introduced by "smsc95xx: add phylib support" 
in 5.10.0. Version before that was 1.0.6.

>>>>>
>>>>>    
>>>>>> Also as reported earlier, (only) on first plug the more
>>>>>> worrying
>>>>>> but
>>>>>> possibly unrelated crash still happens:
>>>>>> ------------[ cut here ]------------
>>>>>> DMA-API: xhci-hcd xhci-hcd.1.auto: cacheline tracking EEXIST,
>>>>>> overlapping mappings aren't supported
>>>>>> WARNING: CPU: 0 PID: 23 at kernel/dma/debug.c:570
>>>>>> add_dma_entry+0x1d9/0x270
>>>>>> Modules linked in: rfcomm iptable_nat bnep usb_f_uac2 u_audio
>>>>>> snd_sof_nocodec usb_f_mass_storage usb_f_eem u_ether
>>>>>> spi_pxa2xx_platform
>>>>>> dw_dmac usb_f_serial u_serial libcomposite intel_mrfld_pwrbtn
>>>>>> snd_sof_pci_intel_tng pwm_lpss_pci intel_mrfld_adc pwm_lpss
>>>>>> snd_sof_pci
>>>>>> snd_sof_intel_ipc snd_sof_intel_atom dw_dmac_pci dw_dmac_core
>>>>>> snd_sof
>>>>>> snd_sof_xtensa_dsp snd_soc_acpi spi_pxa2xx_pci brcmfmac
>>>>>> brcmutil
>>>>>> hci_uart leds_gpio btbcm ti_ads7950
>>>>>> industrialio_triggered_buffer
>>>>>> kfifo_buf tun ledtrig_timer ledtrig_heartbeat mmc_block
>>>>>> extcon_intel_mrfld sdhci_pci cqhci sdhci led_class mmc_core
>>>>>> intel_soc_pmic_mrfld btrfs libcrc32c xor zstd_compress
>>>>>> zlib_deflate
>>>>>> raid6_pq
>>>>>> CPU: 0 PID: 23 Comm: kworker/0:1 Not tainted 5.15.1-edison-
>>>>>> acpi-
>>>>>> standard #1
>>>>>> Hardware name: Intel Corporation Merrifield/BODEGA BAY, BIOS
>>>>>> 542
>>>>>> 2015.01.21:18.19.48
>>>>>> Workqueue: usb_hub_wq hub_event
>>>>>> RIP: 0010:add_dma_entry+0x1d9/0x270
>>>>>> Code: ff 0f 84 97 00 00 00 4c 8b 67 50 4d 85 e4 75 03 4c 8b
>>>>>> 27 e8
>>>>>> 39
>>>>>> ff
>>>>>> 52 00 48 89 c6 4c 89 e2 48 c7 c7 f0 ab e7 a7 e8 c7 5c b5 00
>>>>>> <0f> 0b
>>>>>> 48
>>>>>> 85 ed 0f 85 a4 b3 b5 00 8b 05 46 38 9f 01 85 c0 0f 85 df
>>>>>> RSP: 0000:ffffb047400cfac8 EFLAGS: 00010282
>>>>>> RAX: 0000000000000000 RBX: 00000000ffffffff RCX:
>>>>>> ffff9e25fe217478
>>>>>> RDX: 00000000ffffffd8 RSI: 0000000000000027 RDI:
>>>>>> ffff9e25fe217470
>>>>>> RBP: ffff9e25c12ff780 R08: ffffffffa83359a8 R09:
>>>>>> 0000000000009ffb
>>>>>> R10: 00000000ffffe000 R11: 3fffffffffffffff R12:
>>>>>> ffff9e25c8a24040
>>>>>> R13: 0000000000000001 R14: 0000000000000206 R15:
>>>>>> 00000000000f8be4
>>>>>> FS:  0000000000000000(0000) GS:ffff9e25fe200000(0000)
>>>>>> knlGS:0000000000000000
>>>>>> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>>>> CR2: 00007f62c6d586d8 CR3: 0000000002f3c000 CR4:
>>>>>> 00000000001006f0
>>>>>> Call Trace:
>>>>>>     dma_map_page_attrs+0xfb/0x250
>>>>>>     ? usb_hcd_link_urb_to_ep+0x14/0xa0
>>>>>>     usb_hcd_map_urb_for_dma+0x3b1/0x4e0
>>>>>>     usb_hcd_submit_urb+0x93/0xbb0
>>>>>>     ? create_prof_cpu_mask+0x20/0x20
>>>>>>     ? arch_stack_walk+0x73/0xf0
>>>>>>     ? usb_hcd_link_urb_to_ep+0x14/0xa0
>>>>>>     ? prepare_transfer+0xff/0x140
>>>>>>     usb_start_wait_urb+0x60/0x160
>>>>>>     usb_control_msg+0xda/0x140
>>>>>>     hub_ext_port_status+0x82/0x100
>>>>>>     hub_event+0x1b1/0x1830
>>>>>>     ? hub_activate+0x58c/0x860
>>>>>>     process_one_work+0x1d4/0x370
>>>>>>     worker_thread+0x48/0x3d0
>>>>>>     ? rescuer_thread+0x360/0x360
>>>>>>     kthread+0x122/0x140
>>>>>>     ? set_kthread_struct+0x40/0x40
>>>>>>     ret_from_fork+0x22/0x30
>>>>>> ---[ end trace da6ffcd9fad23a74 ]---
>>>>>> DMA-API: Mapped at:
>>>>>>     debug_dma_map_page+0x60/0xf0
>>>>>>     dma_map_page_attrs+0xfb/0x250
>>>>>>     usb_hcd_map_urb_for_dma+0x3b1/0x4e0
>>>>>>     usb_hcd_submit_urb+0x93/0xbb0
>>>>>>     usb_start_wait_urb+0x60/0x160
>>>>>> usb 1-1.1: new high-speed USB device number 3 using xhci-hcd
>>>>>> usb 1-1.1: New USB device found, idVendor=0424,
>>>>>> idProduct=ec00,
>>>>>> bcdDevice= 2.00
>>>>>> usb 1-1.1: New USB device strings: Mfr=1, Product=2,
>>>>>> SerialNumber=3
>>>>>> usb 1-1.1: Product: LAN9514
>>>>>> usb 1-1.1: Manufacturer: SMSC
>>>>>> usb 1-1.1: SerialNumber: 00951d0d
>>>>>>
>>>>>>
>>>>> I'm not seeing that at all, but I've also been developing and
>>>>> testing
>>>>> on an ARM based device that has the LAN9500A I'm using built
>>>>> into it.
>>>>>
>>>>> I have also recently got a LAN9500A devkit which I've tried on
>>>>> my
>>>>> Ryzen
>>>>> laptop and that's not throwing that error either.
>>>>>
>>>>> Martyn
>>>> Yes I am using the LAN9514 Evaluation board (EVB9514). Plugging
>>>> the
>>>> board into my desktop with 5.15 Ubuntu PPA kernel I also don't
>>>> see this
>>>> crash.
>>>> OTOH I'm building vanilla kernel so no idea why this happens. It
>>>> might
>>>> be that I have CONFIG_DMA_API_DEBUG and Ubuntu not.
>>>>    
>>>
>>> I suspect this is probably unrelated (directly) to the driver then.
>>
>> Maybe not related to this driver but possibly an old bug in hub.c
>> which
>> only surfaces now due to patch "[PATCH] dma debug: report -EEXIST
>> errors
>> in add_dma_entry"? And only shows when CONFIG_DMA_API_DEBUG is set?
>>
>> Ferry
>>
>>> Martyn
>>>
>>>>
>>
>