Message-ID: <0f9a0d5a-d51c-45f4-8f1c-050e7c609baa@kylinos.cn>
Date: Fri, 19 Sep 2025 16:00:48 +0800
From: 李佳怡 <lijiayi@...inos.cn>
To: Mika Westerberg <mika.westerberg@...ux.intel.com>
Cc: linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH RFC] thunderbolt: Re-add DP resources on resume



在 2025/9/19 15:30, Mika Westerberg 写道:
> On Fri, Sep 19, 2025 at 03:24:20PM +0800, 李佳怡 wrote:
>>
>>
>> 在 2025/9/19 00:43, Mika Westerberg 写道:
>>> On Thu, Sep 18, 2025 at 06:20:59PM +0200, Mika Westerberg wrote:
>>>> Hi,
>>>>
>>>> On Thu, Sep 18, 2025 at 04:34:31PM +0800, 李佳怡 wrote:
>>>>>
>>>>>
>>>>> 在 2025/9/17 20:50, Mika Westerberg 写道:
>>>>>> On Wed, Sep 17, 2025 at 06:12:31PM +0800, 李佳怡 wrote:
>>>>>>>
>>>>>>> As requested, I've attached the complete dmesg output (from boot to after
>>>>>>> resume) reproducing the issue.
>>>>>>>
>>>>>>> Testing Methodology:
>>>>>>> 1. Start with the Targus Thunderbolt dock already connected to the system
>>>>>>> 2. Enter S3 suspend (sleep) with no DP monitor connected to the dock
>>>>>>> 3. Resume from S3
>>>>>>> 4. After the system has fully resumed, connect the DP monitor to the dock
>>>>>>
>>>>>> Thanks! It is badly line wrapped. I wonder if you can just attach it?
>>>>>> Anyways I found some unexpected things there:
>>>>>>
>>>>>>> [    8.647850] With USB4 patch v1.0.0
>>>>>>
>>>>>> What is this? ;-)
>>>>>
>>>>> Thanks for your help!
>>>>>
>>>>> This is a self-compiled kernel based on version 5.4 with backported
>>>>> Thunderbolt drivers. I will also attach the kernel log from a build using
>>>>> the linux-6.6.y branch of the community linux-stable repository.
>>>>
>>>> Okay but I really suggest using more recent kernel. 6.16 is current stable
>>>> so that would be good.
>>>>
>>>>>>> [    8.647860] ACPI: bus type thunderbolt registered
>>>>>>> [    8.664660] [7] nhi_probe:1326: thunderbolt 0000:2c:00.0: total paths: 21
>>>>>>> [    8.665209] [7] tb_ring_alloc:586: thunderbolt 0000:2c:00.0: allocating
>>>>>>> TX ring 0 of size 10
>>>>>>> [    8.665243] [7] tb_ring_alloc:586: thunderbolt 0000:2c:00.0: allocating
>>>>>>> RX ring 0 of size 10
>>>>>>> [    8.665267] [7] tb_ctl_alloc:665: thunderbolt 0000:2c:00.0: control
>>>>>>> channel created
>>>>>>> [    8.665272] [7] icm_probe:2549: thunderbolt 0000:2c:00.0: ICM not
>>>>>>> supported on this controller
>>>>>>> [    8.665285] [7] tb_ring_free:840: thunderbolt 0000:2c:00.0: freeing RX
>>>>>>> ring 0
>>>>>>> [    8.665294] [7] tb_ring_free:840: thunderbolt 0000:2c:00.0: freeing TX
>>>>>>> ring 0
>>>>>>
>>>>>> What is this?
>>>>>>
>>>>>> Is this Intel TB/USB4 controller or something else? All USB4 compliant
>>>>>> controllers should go directly to tb.c as that's the part dealing with
>>>>>> software connection manager. The above looks like it tries first with the
>>>>>> firmware connection manager and that should not happen outside of Intel
>>>>>> Thunderbolt 3 hosts.
>>>>>
>>>>> Yes, there is a mistake. I discovered that during the USB4_NATIVE_CONTROL
>>>>> negotiation in the firmware, an OSC_CAPABILITIES_MASK_ERROR bit was being
>>>>> set incorrectly, which should not have happened.
>>>>>
>>>>> The log I will attach next has been modified to fix this issue.
>>>>
>>>> [..]
>>>>
>>>>> [    0.498976] [1] tb_switch_reset:1666: thunderbolt 0000:2c:00.0: 0: resetting
>>>>> [    0.533329] [1] tb_add_dp_resources:217: thunderbolt 0000:2c:00.0: 0:10: DP IN resource available
>>>>> [    0.533959] [1] tb_add_dp_resources:217: thunderbolt 0000:2c:00.0: 0:11: DP IN resource available
>>>>
>>>> The DP IN resources were added here.
>>>>
>>>> [..]
>>>>
>>>>> [   19.035726] [171] tb_switch_set_wake:3445: thunderbolt 0000:2c:00.0: 0: enabling wakeup: 0x3f
>>>>> [   19.037401] [171] tb_ring_stop:768: thunderbolt 0000:2c:00.0: stopping RX ring 0
>>>>> [   19.037412] [171] ring_interrupt_active:141: thunderbolt 0000:2c:00.0: disabling interrupt at register 0x38200 bit 21 (0x200001 -> 0x1)
>>>>> [   19.037439] [171] tb_ring_stop:768: thunderbolt 0000:2c:00.0: stopping TX ring 0
>>>>> [   19.037449] [171] ring_interrupt_active:141: thunderbolt 0000:2c:00.0: disabling interrupt at register 0x38200 bit 0 (0x1 -> 0x0)
>>>>> [   19.037463] [171] tb_ctl_stop:733: thunderbolt 0000:2c:00.0: control channel stopped
>>>>
>>>> Runtime suspend.
>>>>
>>>> (and a couple more)
>>>>
>>>>> [  266.399800] [3870] tb_ctl_start:703: thunderbolt 0000:2c:00.0: control channel starting...
>>>>> [  266.399808] [3870] tb_ring_start:693: thunderbolt 0000:2c:00.0: starting TX ring 0
>>>>> [  266.399821] [3870] ring_interrupt_active:141: thunderbolt 0000:2c:00.0: enabling interrupt at register 0x38200 bit 0 (0x0 -> 0x1)
>>>>> [  266.399826] [3870] tb_ring_start:693: thunderbolt 0000:2c:00.0: starting RX ring 0
>>>>> [  266.399837] [3870] ring_interrupt_active:141: thunderbolt 0000:2c:00.0: enabling interrupt at register 0x38200 bit 21 (0x1 -> 0x200001)
>>>>> [  266.399848] [3870] tb_switch_resume:3478: thunderbolt 0000:2c:00.0: 0: resuming switch
>>>>> [  266.399852] [3870] tb_switch_configure:2590: thunderbolt 0000:2c:00.0: restoring Switch at 0x0 (depth: 0, up port: 5)
>>>>> [  266.400032] [3870] tb_switch_set_wake:3447: thunderbolt 0000:2c:00.0: 0: disabling wakeup
>>>>
>>>> Around this time you should see the hotplug events coming to the DP IN
>>>> adapters. But there is none.
>>>>
>>>> There is a bit in the DP adapter config space (DHP) that can be used to
>>>> disable this but the spec says it defaults to 0. Anyways I suggest to check
>>>> that (and also the Plugged bit). You can dump these using tbtools [1]. Let
>>>> me know if you want help with that.
>>>
>>> Nevermind about the Plugged bit, that's not applicable here.
>>>
>>> Here's how to dump the DHP bit. This is from Intel host:
>>>
>>> # tbdump -vv -r 0 -N 1 -a 5 ADP_CS_5
>>> 0x0005 0x00004809 0b00000000 00000000 01001000 00001001 ..H. ADP_CS_5
>>>     [00:06]        0x9 Max Input HopID
>>>     [11:17]        0x9 Max Output HopID
>>>     [22:28]        0x0 Link Credits Allocated
>>>     [29:29]        0x0 HEC Error Enable (HEE)
>>>     [30:30]        0x0 Flow Control Error Enable (FCEE)
>>>     [31:31]        0x0 Disable Hot Plug Events (DHP)
>>
>> Thank you for your help. As you suggested, I tried kernel 6.16, but the
>> issue still persists.
>>
>> The logs and device vendor information are included in the attachment. I
>> also ran the tbdump command, and the result shows the DHP bit is set to 0.
>> Does this indicate there might still be other issues?
>>
>> # tbdump -vv -r 0 -N 1  -a 5 ADP_CS_5
> 
> Please run it against the DP IN adapters. That's 10 and 11. The 5 above was
> just an example from Intel HW.
> 
> # tbdump -vv -r 0 -N 1  -a 10 ADP_CS_5
> 
> and
> 
> # tbdump -vv -r 0 -N 1  -a 11 ADP_CS_5
> 
> The lspci dump indicates this is ASMedia host controller. I did not even
> know that they have such thing. I thought they only have device (which
> should be working in Linux). The host side may require some additional
> enablement.

Apologies, Please find the latest results.

# tbdump -vv -r 0 -N 1  -a 10 ADP_CS_5
0x0005 0x0000d809 0b00000000 00000000 11011000 00001001 .... ADP_CS_5
   [00:06]        0x9 Max Input HopID
   [11:17]       0x1b Max Output HopID
   [22:28]        0x0 Link Credits Allocated
   [29:29]        0x0 HEC Error Enable (HEE)
   [30:30]        0x0 Flow Control Error Enable (FCEE)
   [31:31]        0x0 Disable Hot Plug Events (DHP)

# tbdump -vv -r 0 -N 1  -a 11 ADP_CS_5
0x0005 0x0000d809 0b00000000 00000000 11011000 00001001 .... ADP_CS_5
   [00:06]        0x9 Max Input HopID
   [11:17]       0x1b Max Output HopID
   [22:28]        0x0 Link Credits Allocated
   [29:29]        0x0 HEC Error Enable (HEE)
   [30:30]        0x0 Flow Control Error Enable (FCEE)
   [31:31]        0x0 Disable Hot Plug Events (DHP)

Regarding the ASMedia host controller: if hotplug events for the DP IN 
adapter are not being generated, is it feasible to manually add the DP 
IN resource as I initially thought, or is that approach incorrect for 
this scenario?

Powered by blists - more mailing lists