Message-ID: <4e5a3a4d-9b6b-443b-b3c2-eac1b44e96e0@molgen.mpg.de>
Date: Mon, 28 Jul 2025 09:39:05 +0200
From: Paul Menzel <pmenzel@...gen.mpg.de>
To: Baochen Qiang <baochen.qiang@....qualcomm.com>
Cc: Baochen Qiang <quic_bqiang@...cinc.com>,
 Jeff Johnson <jjohnson@...nel.org>, ath10k@...ts.infradead.org,
 James Prestwood <prestwoj@...il.com>, Ingo Molnar <mingo@...hat.com>,
 Peter Zijlstra <peterz@...radead.org>, Juri Lelli <juri.lelli@...hat.com>,
 Vincent Guittot <vincent.guittot@...aro.org>,
 LKML <linux-kernel@...r.kernel.org>
Subject: Re: athk10: Poll service ready completion by default to avoid warning
 `failed to receive service ready completion, polling..`?

[CC: +scheduler folks for input on the wait_for_completion_timeout() part]

Dear Baochen,


Thank you for your reply.

Am 28.07.25 um 04:18 schrieb Baochen Qiang:
> On 7/25/2025 8:15 PM, Paul Menzel wrote:

>> Am 22.07.25 um 11:38 schrieb Baochen Qiang:
>>
>>> On 7/22/2025 4:37 PM, Paul Menzel wrote:
>>
>>>> Today, on the Intel Kaby Lake laptop Dell XPS 13 9360 with
>>>>
>>>>       $ lspci -nn -s 3a:
>>>>       3a:00.0 Network controller [0280]: Qualcomm Atheros QCA6174 802.11ac Wireless Network Adapter [168c:003e] (rev 32)
>>>>
>>>> resuming from ACPI S3 took longer, as it sometimes does, and looking into this, I see
>>>> `failed to receive service ready completion, polling..` after a delay of five seconds:
>>>>
>>>> ```
>>>> [    0.000000] Linux version 6.16.0-rc6-00253-g4871b7cb27f4 (build@...emianrhapsody.molgen.mpg.de) (gcc (Debian 14.2.0-19) 14.2.0, GNU ld (GNU Binutils for Debian) 2.44) #90 SMP PREEMPT_DYNAMIC Sat Jul 19 08:53:39 CEST 2025
>>>> […]
>>>> [    8.588020] abreu kernel: ath10k_pci 0000:3a:00.0: qca6174 hw3.2 target 0x05030000 chip_id 0x00340aff sub 1a56:1535
>>>> [    8.588372] abreu kernel: ath10k_pci 0000:3a:00.0: kconfig debug 0 debugfs 0 tracing 0 dfs 0 testmode 0
>>>> [    8.588603] abreu kernel: ath10k_pci 0000:3a:00.0: firmware ver WLAN.RM.4.4.1-00309- api 6 features wowlan,ignore-otp,mfp crc32 0793bcf2
>>>> […]
>>>> [    9.113550] Bluetooth: hci0: QCA: patch rome 0x302 build 0x3e8, firmware rome 0x302 build 0x111
>>>> […]
>>>> [41804.953487] PM: suspend entry (deep)
>>>> [41804.988361] Filesystems sync: 0.034 seconds
>>>> [41805.007216] Freezing user space processes
>>>> [41805.009650] Freezing user space processes completed (elapsed 0.002 seconds)
>>>> [41805.009663] OOM killer disabled.
>>>> [41805.009666] Freezing remaining freezable tasks
>>>> [41805.011383] Freezing remaining freezable tasks completed (elapsed 0.001 seconds)
>>>> [41805.011502] printk: Suspending console(s) (use no_console_suspend to debug)
>>>> [41805.523883] ACPI: EC: interrupt blocked
>>>> [41805.545779] ACPI: PM: Preparing to enter system sleep state S3
>>>> [41805.556040] ACPI: EC: event blocked
>>>> [41805.556045] ACPI: EC: EC stopped
>>>> [41805.556046] ACPI: PM: Saving platform NVS memory
>>>> [41805.559408] Disabling non-boot CPUs ...
>>>> [41805.562480] smpboot: CPU 3 is now offline
>>>> [41805.567105] smpboot: CPU 2 is now offline
>>>> [41805.572122] smpboot: CPU 1 is now offline
>>>> [41805.582034] ACPI: PM: Low-level resume complete
>>>> [41805.582079] ACPI: EC: EC started
>>>> [41805.582080] ACPI: PM: Restoring platform NVS memory
>>>> [41805.583986] Enabling non-boot CPUs ...
>>>> [41805.584009] smpboot: Booting Node 0 Processor 1 APIC 0x2
>>>> [41805.584734] CPU1 is up
>>>> [41805.584749] smpboot: Booting Node 0 Processor 2 APIC 0x1
>>>> [41805.585514] CPU2 is up
>>>> [41805.585530] smpboot: Booting Node 0 Processor 3 APIC 0x3
>>>> [41805.586216] CPU3 is up
>>>> [41805.589070] ACPI: PM: Waking up from system sleep state S3
>>>> [41805.623652] ACPI: EC: interrupt unblocked
>>>> [41805.640074] ACPI: EC: event unblocked
>>>> [41805.651951] nvme nvme0: 4/0/0 default/read/poll queues
>>>> [41805.865391] atkbd serio0: Failed to deactivate keyboard on isa0060/serio0
>>>> [41810.933639] ath10k_pci 0000:3a:00.0: failed to receive service ready completion, polling..
>>>> [41810.933769] ath10k_pci 0000:3a:00.0: service ready completion received, continuing normally
>>>> [41810.986330] OOM killer enabled.
>>>> [41810.986332] Restarting tasks: Starting
>>>> […]
>>>> ```
>>>>
>>>> Commit e57b7d62a1b2 (wifi: ath10k: poll service ready message before failing) [1][2],
>>>> present since Linux v6.10-rc1, added this to avoid the hardware not being initialized:
>>>>
>>>>           time_left = wait_for_completion_timeout(&ar->wmi.service_ready,
>>>>                                                   WMI_SERVICE_READY_TIMEOUT_HZ);
>>>>           if (!time_left) {
>>>>                   /* Sometimes the PCI HIF doesn't receive interrupt
>>>>                    * for the service ready message even if the buffer
>>>>                    * was completed. PCIe sniffer shows that it's
>>>>                    * because the corresponding CE ring doesn't fires
>>>>                    * it. Workaround here by polling CE rings once.
>>>>                    */
>>>>                   ath10k_warn(ar, "failed to receive service ready completion, polling..\n");
>>>>
>>>>                   for (i = 0; i < CE_COUNT; i++)
>>>>                           ath10k_hif_send_complete_check(ar, i, 1);
>>>>
>>>>                   time_left = wait_for_completion_timeout(&ar->wmi.service_ready,
>>>>                                                           WMI_SERVICE_READY_TIMEOUT_HZ);
>>>>                   if (!time_left) {
>>>>                           ath10k_warn(ar, "polling timed out\n");
>>>>                           return -ETIMEDOUT;
>>>>                   }
>>>>
>>>>                   ath10k_warn(ar, "service ready completion received, continuing normally\n");
>>>>           }
>>>>
>>>> The comment says, it’s a hardware issue. I guess from the Qualcomm device and not the
>>>> board design, as it happens with several devices like James’?
>>>>
>>>> Anyway, should polling be used by default then to avoid the delay?
>>>
>>> Adding additional polling before wait seems OK to me
>>
>> With the attached diff, I didn’t notice any issue on the Dell XPS 13 9360 with QCA6174.
> 
> In the diff you are moving polling ahead of wait, IMO this might introduce some race: what
> if hardware/firmware send the event right after polling is done?
> 
> So how about, instead of moving, just adding a new polling before wait:
> 
> 1. polling
> 2. wait
> 3. poling again if wait fail

I do not know the hardware behavior/design and the error, so cannot 
judge, if a race would be possible.

Could Qualcomm take over to cook up a patch
I’d appreciated if Qualcomm could take over to cook up a patch, as you 
have the datasheets, erratas and a line to the hardware designers.

>> Unrelated: The only thing I noticed is, that during boot (not resume) the function seems
>> to be called twice. It looks like once for Wi-Fi and once for Bluetooth:
>>
>> ```
>> [   35.507604] ath10k_pci 0000:3a:00.0: board_file api 2 bmi_id N/A crc32 d2863f91
>> [   35.516010] usb 1-5: New USB device found, idVendor=0c45, idProduct=670c, bcdDevice=56.26
>> [   35.516022] usb 1-5: New USB device strings: Mfr=2, Product=1, SerialNumber=0
>> [   35.516026] usb 1-5: Product: Integrated_Webcam_HD
>> [   35.516029] usb 1-5: Manufacturer: CN09GTFMLOG008C8B7FWA01
>> [   35.587852] ath10k_pci 0000:3a:00.0: service ready completion received, continuing normally
>> [   35.606632] ath10k_pci 0000:3a:00.0: htt-ver 3.87 wmi-op 4 htt-op 3 cal otp max-sta 32 raw 0 hwcrypto 1
>> [   35.628744] mc: Linux media interface: v0.10
>> [   35.651301] nvme nvme0: using unchecked data buffer
>> [   35.687466] Bluetooth: Core ver 2.22
>> [   35.687493] NET: Registered PF_BLUETOOTH protocol family
>> [   35.687495] Bluetooth: HCI device and connection manager initialized
>> [   35.687499] Bluetooth: HCI socket layer initialized
>> [   35.687501] Bluetooth: L2CAP socket layer initialized
>> [   35.687505] Bluetooth: SCO socket layer initialized
>> [   35.696050] ath: EEPROM regdomain: 0x6c
>> [   35.696055] ath: EEPROM indicates we should expect a direct regpair map
>> [   35.696057] ath: Country alpha2 being used: 00
>> [   35.696058] ath: Regpair used: 0x6c
>> [   35.712821] ath10k_pci 0000:3a:00.0 wlp58s0: renamed from wlan0
>> [   35.716790] input: ELAN Touchscreen as /devices/pci0000:00/0000:00:14.0/usb1/1-4/1-4:1.0/0003:04F3:2234.0002/input/input40
>> [   35.718912] videodev: Linux video capture interface: v2.00
>> [   35.719492] input: ELAN Touchscreen UNKNOWN as /devices/pci0000:00/0000:00:14.0/usb1/1-4/1-4:1.0/0003:04F3:2234.0002/input/input41
>> [   35.719595] input: ELAN Touchscreen UNKNOWN as /devices/pci0000:00/0000:00:14.0/usb1/1-4/1-4:1.0/0003:04F3:2234.0002/input/input42
>> [   35.720899] hid-multitouch 0003:04F3:2234.0002: input,hiddev0,hidraw1: USB HID v1.10 Device [ELAN Touchscreen] on usb-0000:00:14.0-4/input0
>> [   35.720947] usbcore: registered new interface driver usbhid
>> [   35.720949] usbhid: USB HID core driver
>> [   35.812081] usbcore: registered new interface driver btusb
>> [   35.815263] Bluetooth: hci0: using rampatch file: qca/rampatch_usb_00000302.bin
>> [   35.815270] Bluetooth: hci0: QCA: patch rome 0x302 build 0x3e8, firmware rome 0x302 build 0x111
>> [   36.174345] Bluetooth: hci0: using NVM file: qca/nvm_usb_00000302.bin
>> [   36.199643] Bluetooth: hci0: HCI Enhanced Setup Synchronous Connection command is advertised, but not supported.
>> [   36.398657] ath10k_pci 0000:3a:00.0: service ready completion received, continuing normally
> 
> Hmm, I don't think this is for BT as ath10k is not a BT driver. Something must be wrong
> here ...
> 
>> ```

Can you reproduce it?

How would I get a call graph for both function calls?

>>>> Additionally I have two questions regarding the code:
>>>>
>>>> 1.  Is `WMI_SERVICE_READY_TIMEOUT_HZ` the right value to pass to
>>>> `wait_for_completion_timeout(struct completion *done, unsigned long timeout)`?
>>>>
>>>> The macro is defined as:
>>>>
>>>>       drivers/net/wireless/ath/ath10k/wmi.h:#define WMI_SERVICE_READY_TIMEOUT_HZ (5 * HZ)
>>>>
>>>> `timeout` is supposed to be in jiffies, and `CONFIG_HZ_250=y` on my system. I wonder how
>>>> that amounts to five seconds on my system.
>>>
>>> HZ is defined as jiffies per second, so 5 * HZ equals 5 seconds.

Sorry, I missed to comment here in my previous reply. HZ can be defined 
differently – like 1000 HZ –, so the timeout would very, and then not 
match the actual timeout required by the hardware? 
`Documentation/scheduler/completion.rst` contains:

> Timeouts are preferably calculated with msecs_to_jiffies() or usecs_to_jiffies(),
> to make the code largely HZ-invariant.


>>>> The timeout should probably be defined in seconds? Does the WMI specification say
>>>> something about this?
>>>>
>>>> 2.  Is the task interruptable and should `wait_for_completion_interruptible_timeout(struct
>>>> completion *done, unsigned long timeout)` be used?
>>>
>>> While I am not sure for now, may I ask why the question?
>>
>> I was just reading up on `wait_for_completion_*()`, and so the different variants.
> 
> If there is no obvious benefits I don't think the change is necessary.

Thinking about it, the driver initialization is in the boot path (hot 
patch) so would block one thread(?) – or is that a wrong assumption –, 
which is unwanted?


Kind regards,

Paul


>>>> [1]: https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e57b7d62a1b2f496caf0beba81cec3c90fad80d5
>>>> [2]: https://lore.kernel.org/all/20240227030409.89702-1-quic_bqiang@quicinc.com/

Powered by blists - more mailing lists