| Message ID | 20240730143748.351651-1-sbinding@opensource.cirrus.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [v1] ALSA: hda: cs35l41: Stop creating ALSA Controls for firmware coefficients | expand |
On 30. 07. 24 16:37, Stefan Binding wrote: > Add a kernel parameter to allow coefficients to be exposed as ALSA controls. > > When the CS35L41 loads its firmware, it has a number of controls to > affect its behaviour. Currently, these controls are exposed as ALSA > Controls by default. > > However, nothing in userspace currently uses them, and is unlikely to > do so in the future, therefore we don't need to create ASLA controls > for them. > > These controls can be useful for debug, so we can add a kernel > parameter to re-enable them if necessary. > > Disabling these controls would prevent userspace from trying to read > these controls when the CS35L41 is hibernating, which ordinarily > would result in an error message. This is probably not a right argument to add this code. The codec should be powered up when those controls are accessed or those controls should be cached by the driver. Although the controls have not been used yet, exposing them in this way is not ideal. Could you fix the driver (no I/O errors)? Jaroslav
On Tue, 30 Jul 2024 16:55:19 +0200, Jaroslav Kysela wrote: > > On 30. 07. 24 16:37, Stefan Binding wrote: > > Add a kernel parameter to allow coefficients to be exposed as ALSA controls. > > > > When the CS35L41 loads its firmware, it has a number of controls to > > affect its behaviour. Currently, these controls are exposed as ALSA > > Controls by default. > > > > However, nothing in userspace currently uses them, and is unlikely to > > do so in the future, therefore we don't need to create ASLA controls > > for them. > > > > These controls can be useful for debug, so we can add a kernel > > parameter to re-enable them if necessary. > > > > Disabling these controls would prevent userspace from trying to read > > these controls when the CS35L41 is hibernating, which ordinarily > > would result in an error message. > > This is probably not a right argument to add this code. The codec > should be powered up when those controls are accessed or those > controls should be cached by the driver. > > Although the controls have not been used yet, exposing them in this > way is not ideal. > > Could you fix the driver (no I/O errors)? While we should fix the potential errors at hibernation, it's not bad to hide those controls, IMO. For the normal use cases, it's nothing but a cause of troubles, after all. thanks, Takashi
On 31. 07. 24 12:30, Takashi Iwai wrote: > On Tue, 30 Jul 2024 16:55:19 +0200, > Jaroslav Kysela wrote: >> >> On 30. 07. 24 16:37, Stefan Binding wrote: >>> Add a kernel parameter to allow coefficients to be exposed as ALSA controls. >>> >>> When the CS35L41 loads its firmware, it has a number of controls to >>> affect its behaviour. Currently, these controls are exposed as ALSA >>> Controls by default. >>> >>> However, nothing in userspace currently uses them, and is unlikely to >>> do so in the future, therefore we don't need to create ASLA controls >>> for them. >>> >>> These controls can be useful for debug, so we can add a kernel >>> parameter to re-enable them if necessary. >>> >>> Disabling these controls would prevent userspace from trying to read >>> these controls when the CS35L41 is hibernating, which ordinarily >>> would result in an error message. >> >> This is probably not a right argument to add this code. The codec >> should be powered up when those controls are accessed or those >> controls should be cached by the driver. >> >> Although the controls have not been used yet, exposing them in this >> way is not ideal. >> >> Could you fix the driver (no I/O errors)? > > While we should fix the potential errors at hibernation, it's not bad > to hide those controls, IMO. For the normal use cases, it's nothing > but a cause of troubles, after all. I do not think that the situation is so obvious. Different coefficients can be used in various UCM profiles for example. But for debugging we have debugfs when the developer thinks that the feature is not useful for users. The module parameter solution is not good in my eyes. Jaroslav
On Wed, 31 Jul 2024 12:36:19 +0200, Jaroslav Kysela wrote: > > On 31. 07. 24 12:30, Takashi Iwai wrote: > > On Tue, 30 Jul 2024 16:55:19 +0200, > > Jaroslav Kysela wrote: > >> > >> On 30. 07. 24 16:37, Stefan Binding wrote: > >>> Add a kernel parameter to allow coefficients to be exposed as ALSA controls. > >>> > >>> When the CS35L41 loads its firmware, it has a number of controls to > >>> affect its behaviour. Currently, these controls are exposed as ALSA > >>> Controls by default. > >>> > >>> However, nothing in userspace currently uses them, and is unlikely to > >>> do so in the future, therefore we don't need to create ASLA controls > >>> for them. > >>> > >>> These controls can be useful for debug, so we can add a kernel > >>> parameter to re-enable them if necessary. > >>> > >>> Disabling these controls would prevent userspace from trying to read > >>> these controls when the CS35L41 is hibernating, which ordinarily > >>> would result in an error message. > >> > >> This is probably not a right argument to add this code. The codec > >> should be powered up when those controls are accessed or those > >> controls should be cached by the driver. > >> > >> Although the controls have not been used yet, exposing them in this > >> way is not ideal. > >> > >> Could you fix the driver (no I/O errors)? > > > > While we should fix the potential errors at hibernation, it's not bad > > to hide those controls, IMO. For the normal use cases, it's nothing > > but a cause of troubles, after all. > > I do not think that the situation is so obvious. Different > coefficients can be used in various UCM profiles for example. If that's the supposed use-case, yes. I doubt it, though, but this needs clarification from Cirrus people. > But for debugging we have debugfs when the developer thinks that the > feature is not useful for users. The module parameter solution is not > good in my eyes. Yeah, I believe we should disable it unconditionally, and provide a different way like debugfs in the firmware driver side, too -- again, if the exposure is only for debugging. Takashi
On 31/07/2024 11:44, Takashi Iwai wrote: > On Wed, 31 Jul 2024 12:36:19 +0200, > Jaroslav Kysela wrote: >> On 31. 07. 24 12:30, Takashi Iwai wrote: >>> On Tue, 30 Jul 2024 16:55:19 +0200, >>> Jaroslav Kysela wrote: >>>> On 30. 07. 24 16:37, Stefan Binding wrote: >>>>> Add a kernel parameter to allow coefficients to be exposed as ALSA controls. >>>>> >>>>> When the CS35L41 loads its firmware, it has a number of controls to >>>>> affect its behaviour. Currently, these controls are exposed as ALSA >>>>> Controls by default. >>>>> >>>>> However, nothing in userspace currently uses them, and is unlikely to >>>>> do so in the future, therefore we don't need to create ASLA controls >>>>> for them. >>>>> >>>>> These controls can be useful for debug, so we can add a kernel >>>>> parameter to re-enable them if necessary. >>>>> >>>>> Disabling these controls would prevent userspace from trying to read >>>>> these controls when the CS35L41 is hibernating, which ordinarily >>>>> would result in an error message. >>>> This is probably not a right argument to add this code. The codec >>>> should be powered up when those controls are accessed or those >>>> controls should be cached by the driver. >>>> >>>> Although the controls have not been used yet, exposing them in this >>>> way is not ideal. >>>> >>>> Could you fix the driver (no I/O errors)? >>> While we should fix the potential errors at hibernation, it's not bad >>> to hide those controls, IMO. For the normal use cases, it's nothing >>> but a cause of troubles, after all. >> I do not think that the situation is so obvious. Different >> coefficients can be used in various UCM profiles for example. > If that's the supposed use-case, yes. > I doubt it, though, but this needs clarification from Cirrus people. > >> But for debugging we have debugfs when the developer thinks that the >> feature is not useful for users. The module parameter solution is not >> good in my eyes. > Yeah, I believe we should disable it unconditionally, and provide a > different way like debugfs in the firmware driver side, too -- again, > if the exposure is only for debugging. I'll update the patch to completely remove the creation of these controls. The tuning files, which are specific for each laptop, that are loaded in conjunction with the firmware files already provide us with the all the control we need over the firmware coefficients, so the ALSA Controls are not needed for anything. These controls were never intended to be exposed to users, anyway, but the firmware doesn't mark them as hidden, so its best for the driver to just ignore them. We think there's very little value in providing an different way (i.e. debugfs) to accessing these controls for debug purposes, so we think we should just remove them. Thanks, Stefan > > > Takashi
diff --git a/sound/pci/hda/cs35l41_hda.c b/sound/pci/hda/cs35l41_hda.c index 4b411ed8c3fe..43d49a69ca14 100644 --- a/sound/pci/hda/cs35l41_hda.c +++ b/sound/pci/hda/cs35l41_hda.c @@ -76,8 +76,13 @@ static const struct cirrus_amp_cal_controls cs35l41_calibration_controls = { static bool firmware_autostart = 1; module_param(firmware_autostart, bool, 0444); -MODULE_PARM_DESC(firmware_autostart, "Allow automatic firmware download on boot" - "(0=Disable, 1=Enable) (default=1); "); +MODULE_PARM_DESC(firmware_autostart, + "Allow automatic firmware download on boot, (0=Disable, 1=Enable) (default=1)"); + +static bool expose_dsp_controls; +module_param(expose_dsp_controls, bool, 0444); +MODULE_PARM_DESC(expose_dsp_controls, + "Expose DSP controls as ALSA controls on boot, (0=Disable, 1=Enable) (default=0)"); static const struct reg_sequence cs35l41_hda_config[] = { { CS35L41_PLL_CLK_CTRL, 0x00000430 }, // 3072000Hz, BCLK Input, PLL_REFCLK_EN = 1 @@ -603,7 +608,8 @@ static int cs35l41_init_dsp(struct cs35l41_hda *cs35l41) if (ret) goto err; - cs35l41_add_controls(cs35l41); + if (expose_dsp_controls) + cs35l41_add_controls(cs35l41); cs35l41_hda_apply_calibration(cs35l41);
Add a kernel parameter to allow coefficients to be exposed as ALSA controls. When the CS35L41 loads its firmware, it has a number of controls to affect its behaviour. Currently, these controls are exposed as ALSA Controls by default. However, nothing in userspace currently uses them, and is unlikely to do so in the future, therefore we don't need to create ASLA controls for them. These controls can be useful for debug, so we can add a kernel parameter to re-enable them if necessary. Disabling these controls would prevent userspace from trying to read these controls when the CS35L41 is hibernating, which ordinarily would result in an error message. Also change the formatting of the other module parameter in cs35l41_hda.c to match the new parameter. Signed-off-by: Stefan Binding <sbinding@opensource.cirrus.com> --- sound/pci/hda/cs35l41_hda.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)