[v2] sound: Use -EPROBE_DEFER instead of i915 module loading.

[PATCH v2 2/9] ALSA: hda/i915: Allow override of gpu binding.

Posted by Maarten Lankhorst 2 years, 6 months ago

Selecting CONFIG_DRM selects CONFIG_VIDEO_NOMODESET, which exports
video_firmware_drivers_only(). This can be used as a first
approximation on whether i915 will be available. It's safe to use as
this is only built when CONFIG_SND_HDA_I915 is selected by CONFIG_I915.

It's not completely fool proof, as you can boot with "nomodeset
i915.modeset=1" to make i915 load regardless, or use
"i915.force_probe=!*" to never load i915, but the common case of
booting with nomodeset to disable all GPU drivers this will work as
intended.

Because of this, we add an extra module parameter,
snd_hda_core.gpu_bind that can be used to signal users intent.
-1 follows nomodeset, 0 disables binding, 1 forces wait/-EPROBE_DEFER
on binding.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
 sound/hda/hdac_i915.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/sound/hda/hdac_i915.c b/sound/hda/hdac_i915.c
index 161a9711cd63e..c32709fa4115f 100644
--- a/sound/hda/hdac_i915.c
+++ b/sound/hda/hdac_i915.c
@@ -11,6 +11,13 @@
 #include <sound/hda_i915.h>
 #include <sound/hda_register.h>
 
+#include <video/nomodeset.h>
+
+static int gpu_bind = -1;
+module_param(gpu_bind, int, 0644);
+MODULE_PARM_DESC(gpu_bind, "Whether to bind sound component to GPU "
+			   "(1=always, 0=never, -1=on nomodeset(default))");
+
 #define IS_HSW_CONTROLLER(pci) (((pci)->device == 0x0a0c) || \
 				((pci)->device == 0x0c0c) || \
 				((pci)->device == 0x0d0c) || \
@@ -121,6 +128,9 @@ static int i915_gfx_present(struct pci_dev *hdac_pci)
 {
 	struct pci_dev *display_dev = NULL;
 
+	if (!gpu_bind || (gpu_bind < 0 && video_firmware_drivers_only()))
+		return false;
+
 	for_each_pci_dev(display_dev) {
 		if (display_dev->vendor == PCI_VENDOR_ID_INTEL &&
 		    (display_dev->class >> 16) == PCI_BASE_CLASS_DISPLAY &&
-- 
2.39.2

Re: [PATCH v2 2/9] ALSA: hda/i915: Allow override of gpu binding.

Posted by Péter Ujfalusi 2 years, 6 months ago


On 19/07/2023 19:41, Maarten Lankhorst wrote:
> Selecting CONFIG_DRM selects CONFIG_VIDEO_NOMODESET, which exports
> video_firmware_drivers_only(). This can be used as a first
> approximation on whether i915 will be available. It's safe to use as
> this is only built when CONFIG_SND_HDA_I915 is selected by CONFIG_I915.
> 
> It's not completely fool proof, as you can boot with "nomodeset
> i915.modeset=1" to make i915 load regardless, or use
> "i915.force_probe=!*" to never load i915, but the common case of
> booting with nomodeset to disable all GPU drivers this will work as
> intended.
> 
> Because of this, we add an extra module parameter,
> snd_hda_core.gpu_bind that can be used to signal users intent.
> -1 follows nomodeset, 0 disables binding, 1 forces wait/-EPROBE_DEFER
> on binding.

Reviewed-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> ---
>  sound/hda/hdac_i915.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/sound/hda/hdac_i915.c b/sound/hda/hdac_i915.c
> index 161a9711cd63e..c32709fa4115f 100644
> --- a/sound/hda/hdac_i915.c
> +++ b/sound/hda/hdac_i915.c
> @@ -11,6 +11,13 @@
>  #include <sound/hda_i915.h>
>  #include <sound/hda_register.h>
>  
> +#include <video/nomodeset.h>
> +
> +static int gpu_bind = -1;
> +module_param(gpu_bind, int, 0644);
> +MODULE_PARM_DESC(gpu_bind, "Whether to bind sound component to GPU "
> +			   "(1=always, 0=never, -1=on nomodeset(default))");
> +
>  #define IS_HSW_CONTROLLER(pci) (((pci)->device == 0x0a0c) || \
>  				((pci)->device == 0x0c0c) || \
>  				((pci)->device == 0x0d0c) || \
> @@ -121,6 +128,9 @@ static int i915_gfx_present(struct pci_dev *hdac_pci)
>  {
>  	struct pci_dev *display_dev = NULL;
>  
> +	if (!gpu_bind || (gpu_bind < 0 && video_firmware_drivers_only()))
> +		return false;
> +
>  	for_each_pci_dev(display_dev) {
>  		if (display_dev->vendor == PCI_VENDOR_ID_INTEL &&
>  		    (display_dev->class >> 16) == PCI_BASE_CLASS_DISPLAY &&

-- 
Péter

Re: [PATCH v2 2/9] ALSA: hda/i915: Allow override of gpu binding.

Posted by Pierre-Louis Bossart 2 years, 6 months ago


On 7/21/23 14:19, Péter Ujfalusi wrote:
> 
> 
> On 19/07/2023 19:41, Maarten Lankhorst wrote:
>> Selecting CONFIG_DRM selects CONFIG_VIDEO_NOMODESET, which exports
>> video_firmware_drivers_only(). This can be used as a first
>> approximation on whether i915 will be available. It's safe to use as
>> this is only built when CONFIG_SND_HDA_I915 is selected by CONFIG_I915.
>>
>> It's not completely fool proof, as you can boot with "nomodeset
>> i915.modeset=1" to make i915 load regardless, or use
>> "i915.force_probe=!*" to never load i915, but the common case of
>> booting with nomodeset to disable all GPU drivers this will work as
>> intended.
>>
>> Because of this, we add an extra module parameter,
>> snd_hda_core.gpu_bind that can be used to signal users intent.
>> -1 follows nomodeset, 0 disables binding, 1 forces wait/-EPROBE_DEFER
>> on binding.
> 
> Reviewed-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
> 
>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
>> ---
>>  sound/hda/hdac_i915.c | 10 ++++++++++
>>  1 file changed, 10 insertions(+)
>>
>> diff --git a/sound/hda/hdac_i915.c b/sound/hda/hdac_i915.c
>> index 161a9711cd63e..c32709fa4115f 100644
>> --- a/sound/hda/hdac_i915.c
>> +++ b/sound/hda/hdac_i915.c
>> @@ -11,6 +11,13 @@
>>  #include <sound/hda_i915.h>
>>  #include <sound/hda_register.h>
>>  
>> +#include <video/nomodeset.h>
>> +
>> +static int gpu_bind = -1;
>> +module_param(gpu_bind, int, 0644);
>> +MODULE_PARM_DESC(gpu_bind, "Whether to bind sound component to GPU "
>> +			   "(1=always, 0=never, -1=on nomodeset(default))");
>> +
>>  #define IS_HSW_CONTROLLER(pci) (((pci)->device == 0x0a0c) || \
>>  				((pci)->device == 0x0c0c) || \
>>  				((pci)->device == 0x0d0c) || \
>> @@ -121,6 +128,9 @@ static int i915_gfx_present(struct pci_dev *hdac_pci)
>>  {
>>  	struct pci_dev *display_dev = NULL;
>>  
>> +	if (!gpu_bind || (gpu_bind < 0 && video_firmware_drivers_only()))
>> +		return false;


Bear with me since I am just going back to work mode but I can't figure
out what the second part of the test does

bool video_firmware_drivers_only(void)
{
	return video_nomodeset;
}
EXPORT_SYMBOL(video_firmware_drivers_only);

and video_nomodeset=1 by default, so why would we return false when
gpu_bind = -1?

This seems to be a change of functionality, what am I missing?


>> +
>>  	for_each_pci_dev(display_dev) {
>>  		if (display_dev->vendor == PCI_VENDOR_ID_INTEL &&
>>  		    (display_dev->class >> 16) == PCI_BASE_CLASS_DISPLAY &&
>

Re: [PATCH v2 2/9] ALSA: hda/i915: Allow override of gpu binding.

Posted by Maarten Lankhorst 2 years, 6 months ago

Hey,

On 2023-07-24 12:25, Pierre-Louis Bossart wrote:
> 
> 
> On 7/21/23 14:19, Péter Ujfalusi wrote:
>>
>>
>> On 19/07/2023 19:41, Maarten Lankhorst wrote:
>>> Selecting CONFIG_DRM selects CONFIG_VIDEO_NOMODESET, which exports
>>> video_firmware_drivers_only(). This can be used as a first
>>> approximation on whether i915 will be available. It's safe to use as
>>> this is only built when CONFIG_SND_HDA_I915 is selected by CONFIG_I915.
>>>
>>> It's not completely fool proof, as you can boot with "nomodeset
>>> i915.modeset=1" to make i915 load regardless, or use
>>> "i915.force_probe=!*" to never load i915, but the common case of
>>> booting with nomodeset to disable all GPU drivers this will work as
>>> intended.
>>>
>>> Because of this, we add an extra module parameter,
>>> snd_hda_core.gpu_bind that can be used to signal users intent.
>>> -1 follows nomodeset, 0 disables binding, 1 forces wait/-EPROBE_DEFER
>>> on binding.
>>
>> Reviewed-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
>>
>>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
>>> ---
>>>   sound/hda/hdac_i915.c | 10 ++++++++++
>>>   1 file changed, 10 insertions(+)
>>>
>>> diff --git a/sound/hda/hdac_i915.c b/sound/hda/hdac_i915.c
>>> index 161a9711cd63e..c32709fa4115f 100644
>>> --- a/sound/hda/hdac_i915.c
>>> +++ b/sound/hda/hdac_i915.c
>>> @@ -11,6 +11,13 @@
>>>   #include <sound/hda_i915.h>
>>>   #include <sound/hda_register.h>
>>>   
>>> +#include <video/nomodeset.h>
>>> +
>>> +static int gpu_bind = -1;
>>> +module_param(gpu_bind, int, 0644);
>>> +MODULE_PARM_DESC(gpu_bind, "Whether to bind sound component to GPU "
>>> +			   "(1=always, 0=never, -1=on nomodeset(default))");
>>> +
>>>   #define IS_HSW_CONTROLLER(pci) (((pci)->device == 0x0a0c) || \
>>>   				((pci)->device == 0x0c0c) || \
>>>   				((pci)->device == 0x0d0c) || \
>>> @@ -121,6 +128,9 @@ static int i915_gfx_present(struct pci_dev *hdac_pci)
>>>   {
>>>   	struct pci_dev *display_dev = NULL;
>>>   
>>> +	if (!gpu_bind || (gpu_bind < 0 && video_firmware_drivers_only()))
>>> +		return false;
> 
> 
> Bear with me since I am just going back to work mode but I can't figure
> out what the second part of the test does
> 
> bool video_firmware_drivers_only(void)
> {
> 	return video_nomodeset;
> }
> EXPORT_SYMBOL(video_firmware_drivers_only);
> 
> and video_nomodeset=1 by default, so why would we return false when
> gpu_bind = -1?
> 
> This seems to be a change of functionality, what am I missing?
video_nomodeset is 0 by default, only when nomodeset is given as 
argument is it set to 1. :-)

Cheers,
~Maarten

Re: [PATCH v2 2/9] ALSA: hda/i915: Allow override of gpu binding.

Posted by Pierre-Louis Bossart 2 years, 6 months ago

>>>> @@ -121,6 +128,9 @@ static int i915_gfx_present(struct pci_dev
>>>> *hdac_pci)
>>>>   {
>>>>       struct pci_dev *display_dev = NULL;
>>>>   +    if (!gpu_bind || (gpu_bind < 0 &&
>>>> video_firmware_drivers_only()))
>>>> +        return false;
>>
>>
>> Bear with me since I am just going back to work mode but I can't figure
>> out what the second part of the test does
>>
>> bool video_firmware_drivers_only(void)
>> {
>>     return video_nomodeset;
>> }
>> EXPORT_SYMBOL(video_firmware_drivers_only);
>>
>> and video_nomodeset=1 by default, so why would we return false when
>> gpu_bind = -1?
>>
>> This seems to be a change of functionality, what am I missing?
> video_nomodeset is 0 by default, only when nomodeset is given as
> argument is it set to 1. :-)


I must be missing something on how the default is handled.

bool video_firmware_drivers_only(void)
{
	return video_nomodeset;
}
EXPORT_SYMBOL(video_firmware_drivers_only);

static int __init disable_modeset(char *str)
{
	video_nomodeset = true;

isn't default 'true' then?

Re: [PATCH v2 2/9] ALSA: hda/i915: Allow override of gpu binding.

Posted by Takashi Iwai 2 years, 6 months ago

On Tue, 25 Jul 2023 13:52:32 +0200,
Pierre-Louis Bossart wrote:
> 
> 
> >>>> @@ -121,6 +128,9 @@ static int i915_gfx_present(struct pci_dev
> >>>> *hdac_pci)
> >>>>   {
> >>>>       struct pci_dev *display_dev = NULL;
> >>>>   +    if (!gpu_bind || (gpu_bind < 0 &&
> >>>> video_firmware_drivers_only()))
> >>>> +        return false;
> >>
> >>
> >> Bear with me since I am just going back to work mode but I can't figure
> >> out what the second part of the test does
> >>
> >> bool video_firmware_drivers_only(void)
> >> {
> >>     return video_nomodeset;
> >> }
> >> EXPORT_SYMBOL(video_firmware_drivers_only);
> >>
> >> and video_nomodeset=1 by default, so why would we return false when
> >> gpu_bind = -1?
> >>
> >> This seems to be a change of functionality, what am I missing?
> > video_nomodeset is 0 by default, only when nomodeset is given as
> > argument is it set to 1. :-)
> 
> 
> I must be missing something on how the default is handled.
> 
> bool video_firmware_drivers_only(void)
> {
> 	return video_nomodeset;
> }
> EXPORT_SYMBOL(video_firmware_drivers_only);
> 
> static int __init disable_modeset(char *str)
> {
> 	video_nomodeset = true;
> 
> isn't default 'true' then?

disable_modeset() is called *only* when nomodeset boot option is
given.  Then it overrides the value to true.


Takashi