Message-ID: <d50cbda2-7c62-433e-b7a4-c716e6818007@linux.dev>
Date: Sun, 4 Jan 2026 01:29:17 -0800
From: Matthew Schwartz <matthew.schwartz@...ux.dev>
To: "Xu, Baojun" <baojun.xu@...com>, Antheas Kapenekakis <lkml@...heas.dev>
Cc: "Ding, Shenghao" <shenghao-ding@...com>,
 "linux-sound@...r.kernel.org" <linux-sound@...r.kernel.org>,
 "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
 "tiwai@...e.de" <tiwai@...e.de>
Subject: Re: [EXTERNAL] Re: [BUG] hda/tas2781: ASUS ROG Xbox Ally X audio
 issues with default firmware

On 1/4/26 12:58 AM, Xu, Baojun wrote:
> Hi Antheas,
> 
> It may known issue which relative with calibration result.
> Could you try to remove calibration result apply, and check if the problem still exist?
> For example, mask below line (line number is 494) in tasdevice_dspfw_init() from tas2781-hda-i2c.c:
> 
> hda_priv->save_calibration(tas_hda);
> 
> Best Regards
> Jim

Hi Jim,

Thanks for the reply. Antheas does not have hardware affected by this issue, but I can confirm that this fixes the problem on my own device.

Matt

> ________________________________________
> From: Antheas Kapenekakis <lkml@...heas.dev>
> Sent: 03 January 2026 19:48
> To: Matthew Schwartz
> Cc: Xu, Baojun; Ding, Shenghao; linux-sound@...r.kernel.org; linux-kernel@...r.kernel.org; tiwai@...e.de
> Subject: [EXTERNAL] Re: [BUG] hda/tas2781: ASUS ROG Xbox Ally X audio issues with default firmware
> 
> On Sat, 3 Jan 2026 at 02: 31, Matthew Schwartz <matthew. schwartz@ linux. dev> wrote: > > On 1/2/26 3: 13 PM, Antheas Kapenekakis wrote: > > On Fri, 2 Jan 2026 at 21: 17, Matthew Schwartz > > <matthew. schwartz@ linux. dev>
> ZjQcmQRYFpfptBannerStart
> This message was sent from outside of Texas Instruments.
> Do not click links or open attachments unless you recognize the source of this email and know the content is safe.
> <https://us-phishalarm-ewt.proofpoint.com/EWT/v1/G3vK!tDdkczlnnYRTxMpLVz-pN5EZOCpqg7wGjyDRBvMeAZFuWZXBJlfNUQdJ4vR2DLvmO1o91tMvcFY2CkFWEO4UIBOBJ80exLz0o0fjR_qbdsgA4aHVUaeJh-oUWg$>
> Report Suspicious
> 
> ZjQcmQRYFpfptBannerEnd
> 
> On Sat, 3 Jan 2026 at 02:31, Matthew Schwartz
> <matthew.schwartz@...ux.dev> wrote:
>>
>> On 1/2/26 3:13 PM, Antheas Kapenekakis wrote:
>>> On Fri, 2 Jan 2026 at 21:17, Matthew Schwartz
>>> <matthew.schwartz@...ux.dev> wrote:
>>>>
>>>> On 1/2/26 9:12 AM, Antheas Kapenekakis wrote:
>>>>> On Tue, 30 Dec 2025 at 22:44, Matthew Schwartz
>>>>> <matthew.schwartz@...ux.dev> wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>>> On Dec 8, 2025, at 8:00 PM, Matthew Schwartz <matthew.schwartz@...ux.dev> wrote:
>>>>>>>
>>>>>> - snip
>>>>>>> 2.52.0
>>>>>>
>>>>>> After reading the TI E2E forums, it seems these calibration tuning configurations are only meant to be used during a calibration process.
>>>>>
>>>>> A source would be good here, a link or two
>>>>
>>>> Sorry about that, here is where I read about the differences between the two configs:
>>>> https://e2e.ti.com/support/audio-group/audio/f/audio-forum/1558310/tas2563-what-is-the-difference-between-tuning-and-calibration-configuration-in-exported-smartamp-binary
>>>
>>> This link describes "calibration" configurations that are used in the
>>> calibration procedure. It is not clear to me that it refers to the
>>> calibration parameters exported by UEFI or in the configuration itself
>>> to be used alongside a configuration. I would tend toward this being
>>> irrelevant.
>>>
>>>> It's about a different amplifier model, but I assume the same applies to tas2781 given the naming structure is the same for the configurations.
>>>>
>>>>>
>>>>>> Instead, something else I found that works is not overriding the firmware file calibration data with the UEFI calibration data:
>>>
>>> I misunderstood what you meant by this before. I thought you meant
>>> that the firmware overrode the UEFI data, not the other way around.
>>> Surely, using the dummy data in the firmware file is better than using
>>> incorrect data from UEFI. However, the manufacturer calibrated data
>>> from the factory floor for each specific unit is in UEFI, so that is
>>> what should be used.
>>>
>>>>> I did not look into the source code, do you have any reference in the
>>>>> ACPI TAS code that r0_buf is prefilled with UEFI data?
>>>>
>>>> From what I understand, I think the current flow goes like this:
>>>>
>>>> 1. During driver init, tas2781_save_calibration() reads UEFI calibration data into the cali_data memory buffer and sets is_user_space_calidata=true: https://urldefense.com/v3/__https://github.com/torvalds/linux/blob/master/sound/hda/codecs/side-codecs/tas2781_hda.c*L162-L173__;Iw!!G3vK!U3mEJNuRpB9ZxVAGZZCyoMNqMnAl4M1N8YoxbHiMHIsF-13h4Wg_fbVfd008drQfvgstYyA1aQnWHw$
>>>>
>>>> 2. When switching to a DSP config, tasdevice_select_tuningprm_cfg() calls tasdevice_load_data() which writes the firmware configuration, including any calibration values embedded in that config: https://urldefense.com/v3/__https://github.com/torvalds/linux/blob/master/sound/soc/codecs/tas2781-fmwlib.c*L2510__;Iw!!G3vK!U3mEJNuRpB9ZxVAGZZCyoMNqMnAl4M1N8YoxbHiMHIsF-13h4Wg_fbVfd008drQfvgstYyDJTrchbQ$
>>>>
>>>> 3. Immediately after, tasdev_load_calibrated_data() writes the UEFI calibration data from step 1, overwriting the values just set in step 2: https://urldefense.com/v3/__https://github.com/torvalds/linux/blob/9b043680446067358913edc2e9dd71bf8ffae208/sound/soc/codecs/tas2781-fmwlib.c*L2392-L2428__;Iw!!G3vK!U3mEJNuRpB9ZxVAGZZCyoMNqMnAl4M1N8YoxbHiMHIsF-13h4Wg_fbVfd008drQfvgstYyBeYnGr_w$ + https://urldefense.com/v3/__https://github.com/torvalds/linux/blob/master/sound/soc/codecs/tas2781-fmwlib.c*L2519__;Iw!!G3vK!U3mEJNuRpB9ZxVAGZZCyoMNqMnAl4M1N8YoxbHiMHIsF-13h4Wg_fbVfd008drQfvgstYyBz8-7Z6g$
>>>>
>>>> I confirmed this this by inserting some debug logs around tasdev_load_calibrated_data:
>>>
>>> I went through the code. It loads the UEFI data, sets
>>> is_user_space_calidata=1, then if the data is available it loads it.
>>> This is correct.
>>>
>>> To me this seems like the calibration data for amp 1 is written to
>>> both amp 1 and amp 2, and for your firmware this breaks amp2.
>>
>> If this were the case, shouldn't my debug logs have the same data being written to dev 0 and dev 1?
>>
>> [    6.367380] tas2781-hda i2c-TXNW2781:00-tas2781-hda.0: tasdev_load_calibrated_data: dev 0 writing calibration: r0_reg=0x000ce4 data=3c5f7222
>> [    6.371718] tas2781-hda i2c-TXNW2781:00-tas2781-hda.0: tasdev_load_calibrated_data: dev 1 writing calibration: r0_reg=0x000ce4 data=3c90e72d
>>
>>>
>>> In tasdev_load_calibrated_data(), an i index is provided, but
>>> cali_data is nested under priv. So only one calibration set is
>>> supported. This means that amp 2 gets amp 1 calibrations.
>>>
>>> Perhaps there is a "SmartAmpCalibrationData2" that should be read
>>> instead for amp 2 instead, can you dump the EFI variables and check?
>>
>> (128)(deck@...amdeck ~)$ ls /sys/firmware/efi/efivars/
>> AmdAcpiVar-79941ecd-ed36-49d0-8124-e4c31ac75cd4               BugCheckProgress-ba57e015-65b3-4c3c-b274-659192f699e3             MemoryOverwriteRequestControl-e20939be-32d4-41be-a150-897f85d49829
>> <snip>
>> There's only one EFI var that seems to be for calibration data:
>>
>> xxd /sys/firmware/efi/efivars/CALI_DATA-1f52d2a1-bb3a-457d-bc09-43a3f4310a92:
>>
>> 00000000: 0700 0000 dd0a 0000 0300 0000 7e4a 9d68  ............~J.h
>> 00000010: 0000 0000 2272 5f3c d515 f610 ac9a 9b2e  ...."r_<........
>> 00000020: 5d91 9600 0000 8025 0100 0000 2de7 903c  ]......%....-..<
>> 00000030: 283c e810 ffc8 c12e fcce 9600 0000 8025  (<.............%
>> 00000040: 8000 0000 0000 1964 0000 1974 0000 197c  .......d...t...|
>> 00000050: 0000 1560 0000 1a70 eb7f b12b            ...`...p...+
> 
> I reviewed the loader for CALI_DATA, it supports multiple device
> calibration data and indexes it currently. In addition, it has two
> versions with CRC checks that pass, so one of those versions loads the
> data correctly.
> 
> So everything looks technically correct
> 
> Here is your data from the first amp reordered:
>        r0_reg=0x000ce4 408f5c29
>        r0_reg=0x000ce4 3c5f7222
>    invr0_reg=0x000cf4 0fdc788c
>    invr0_reg=0x000cf4 10f615d5
> r0_low_reg=0x000cfc 0f7e9100
> r0_low_reg=0x000cfc 2e9b9aac
>   pow_reg=0x000ae0 009b5281
>   pow_reg=0x000ae0 0096915d
>  tlimit_reg=0x000d70 25800000
>  tlimit_reg=0x000d70 25800000
> 
> And the second amp:
>     r0_reg=0x000ce4 40cccccd
>     r0_reg=0x000ce4 3c90e72d
>  invr0_reg=0x000cf4 0fcd6e9e
>  invr0_reg=0x000cf4 10e83c28
> r0_low_reg=0x000cfc 0f8d4fdf
> r0_low_reg=0x000cfc 2ec1c8ff
>    pow_reg=0x000ae0 009b9c58
>    pow_reg=0x000ae0 0096cefc
> tlimit_reg=0x000d70 25800000
> tlimit_reg=0x000d70 25800000
> 
> There is a large deviation in r0, invr0, and r0_low in both of them.
> Unfortunately, I do not have my Xbox Ally non-X to check to see if
> it's incorrect. This hints to me that the registers get incorrect data
> loaded, so there might be a mistake in the UEFI parser.
> 
> Here is a breakdown of your efivar
> 0700 0000 # This is not supposed to be here / Might have been thrown
> out by kernel if v2 loads
> 
> dd0a 0000 # 2781, is v2 protocol
> 0300 0000 # Data-Group-Sum / How many nodes
> 7e4a 9d68 # Timestamp
> 
> # First amplifier data
> 0000 0000
> 2272 5f3c # r0
> d515 f610 # invr0
> ac9a 9b2e # r0_low
> 5d91 9600 # pow
> 0000 8025 # tlimit
> 
> # Second amplifier data
> 0100 0000
> 2de7 903c # r0
> 283c e810 # invr0
> ffc8 c12e # r0_low
> fcce 9600 # pow
> 0000 8025 # tlimit
> 
> 8000 0000 # Calibration data / Derivation logic in TASDEVICE_REG
> 0000 1964 # r0 addr
> 0000 1974 # invr0 addr
> 0000 197c # r0_low addr
> 0000 1560 # pow addr
> 0000 1a70 # tlimit addr
> 
> eb7f b12b # CRC32
> 
> Specifically for the 0x80, cali_reg addresses get stored into
> r0/invr0... iff p->dspbin_typ == TASDEV_BASIC.
> 
> I suspect the problem here is that the UEFI override for the first
> three values tries to change different registers and not r0, invr0 and
> r0_low. The pow, tlimit look correct.
> 
> But, due to p->dspbin_typ != TASDEV_BASIC, cali_reg ends up being
> thrown out and the UEFI values end up corrupting the r0, invr0 and
> r0_low default addresses. and not 00, 19, 64/74/7c
> 
> Antheas
> 
>>>
>>> In that case, a minor refactor to move cali_data and
>>> is_user_space_calidata from priv to priv->tasdevice[i] and then use
>>> the proper efi var would fix this.
>>>
>>> Antheas
>>>
>>>> [    3.908963] tas2781-hda i2c-TXNW2781:00-tas2781-hda.0: tas2781_apply_calib: dspbin_typ=2, ndev=2, Setting is_user_space_calidata=true
>>>> <snip>
>>>>
>>>> where it loads the UEFI calibration over top of the firmware calibration.
>>>>
>>>>>
>>>>> It could be that there is a specific priority, where UEFI data is
>>>>> supposed to be loaded after the firmware code, replacing the
>>>>> calibration data from the file, or that is what is done in Windows.
>>>>> But here it is done the other way. In that case, it might be more
>>>>> appropriate to set a dummy var such as bool uefi_calib that becomes 1
>>>>> when loading calibration from UEFI, and skip loading from the fw file
>>>>> if available.
>>>>>
>>>>> But links, etc. Here, this would affect all TAS devices too, so it is
>>>>> more major.
>>>>
>>>> Yes, was really hoping to get TI's feedback before potentially sending it out, as it would be a major change and the documentation on this is scarce. I could also be misunderstanding the calibration data load flow, but this is just from poking at this issue from every angle I can think of.
>>>>
>>>> Happy new year,
>>>> Matt
>>>>
>>>>>
>>>>> Happy new year,
>>>>> Antheas
>>>>>
>>>>>> diff --git a/sound/soc/codecs/tas2781-fmwlib.c b/sound/soc/codecs/tas2781-fmwlib.c
>>>>>> index 78fd0a5dc6f2..1e768e6187da 100644
>>>>>> --- a/sound/soc/codecs/tas2781-fmwlib.c
>>>>>> +++ b/sound/soc/codecs/tas2781-fmwlib.c
>>>>>> @@ -2377,6 +2377,7 @@ static void tasdev_load_calibrated_data(struct tasdevice_priv *priv, int i)
>>>>>>         unsigned char *data = cali_data->data;
>>>>>>         struct tasdevice_calibration *cal;
>>>>>>         int k = i * (cali_data->cali_dat_sz_per_dev + 1);
>>>>>> +       unsigned char r0_buf[4];
>>>>>>         int rc;
>>>>>>
>>>>>>         /* Load the calibrated data from cal bin file */
>>>>>> @@ -2389,6 +2390,20 @@ static void tasdev_load_calibrated_data(struct tasdevice_priv *priv, int i)
>>>>>>         }
>>>>>>         if (!priv->is_user_space_calidata)
>>>>>>                 return;
>>>>>> +
>>>>>> +       /*
>>>>>> +        * Check if the DSP config already set the calibration registers.
>>>>>> +        * Some tuning configs contain their own calibration data which should
>>>>>> +        * not be overwritten by user space calibration data.
>>>>>> +        */
>>>>>> +       rc = tasdevice_dev_bulk_read(priv, i, p->r0_reg, r0_buf, 4);
>>>>>> +       if (rc >= 0 && (r0_buf[0] | r0_buf[1] | r0_buf[2] | r0_buf[3])) {
>>>>>> +               dev_dbg(priv->dev,
>>>>>> +                       "%s: dev %d r0_reg already set by config, skipping calibration\n",
>>>>>> +                       __func__, i);
>>>>>> +               return;
>>>>>> +       }
>>>>>> +
>>>>>>         /* load calibrated data from user space */
>>>>>>         if (data[k] != i) {
>>>>>>                 dev_err(priv->dev, "%s: no cal-data for dev %d from usr-spc\n",
>>>>>>
>>>>>> Is this more suitable to be upstreamed?
>>>>>>
>>>>>> Thanks,
>>>>>> Matt
>>>>>> <snip>
>>>>>>
>>>>>
>>>>
>>>>
>>>
>>
>>
> 
> 
> 

Powered by blists - more mailing lists