[Intel-gfx] [RFC PATCH v3 0/1] Splitting up platform-specific calls

[Intel-gfx] [RFC PATCH v3 0/1] Splitting up platform-specific calls

[Casey Bowman]

In this RFC I would like to ask the community their thoughts
on how we can best handle splitting architecture-specific
calls.

I would like to address the following:

1. How do we want to split architecture calls? Different object files
per platform? Separate function calls within the same object file?

2. How do we address dummy functions? If we have a function call that is
used for one or more platforms, but is not used in another, what should
we do for this case?

I've given an example of splitting an architecture call
in my patch with run_as_guest() being split into different
implementations for x86 and arm64 in separate object files, sharing
a single header.

Another suggestion from Michael (michael.cheng@intel.com) involved
using a single object file, a single header, and splitting various
functions calls via ifdefs in the header file.

I would appreciate any input on how we can avoid scaling issues when
including multiple architectures and multiple functions (as the number
of function calls will inevitably increase with more architectures).

v2: Revised to use kernel's platform-splitting scheme.
v3: Revised to use simple if-else structure.

Casey Bowman (1):
  i915/drm: Split out x86/arm64 for run_as_guest

 drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++
 1 file changed, 8 insertions(+)

-- 
2.25.1

[Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Casey Bowman]

Splitting out run_as_guest into platform-specific functions
as arm64 does not support this functionality.

Signed-off-by: Casey Bowman <casey.g.bowman@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 1bca510a946d..fdec2b025540 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1381,10 +1381,18 @@ IS_SUBPLATFORM(const struct drm_i915_private *i915,
 #define INTEL_DISPLAY_ENABLED(dev_priv) \
 	(drm_WARN_ON(&(dev_priv)->drm, !HAS_DISPLAY(dev_priv)), !(dev_priv)->params.disable_display)
 
+#if IS_ENABLED(CONFIG_X86)
 static inline bool run_as_guest(void)
 {
 	return !hypervisor_is_type(X86_HYPER_NATIVE);
 }
+#elif IS_ENABLED(CONFIG_ARM64)
+static inline bool run_as_guest(void)
+{
+	/* Not supported for arm64, so we return false  */
+	return false;
+}
+#endif
 
 #define HAS_D12_PLANE_MINIMIZATION(dev_priv) (IS_ROCKETLAKE(dev_priv) || \
 					      IS_ALDERLAKE_S(dev_priv))
-- 
2.25.1

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Casey Bowman]

Wanted to ping this older thread to find out where we stand with this patch,
Are we OK with the current state of these changes?

With more recent information gathered from feedback on other patches, would
we prefer changing this to a more arch-neutral control flow?

e.g.
#if IS_ENABLED(CONFIG_X86)
...
#else
...
#endif

Would we also prefer this RFC series be merged or would it be preferred to
create a new series instead?

Regards,
Casey

On 2/15/22 15:41, Casey Bowman wrote:
> Splitting out run_as_guest into platform-specific functions
> as arm64 does not support this functionality.
>
> Signed-off-by: Casey Bowman <casey.g.bowman@intel.com>
> ---
>   drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++
>   1 file changed, 8 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 1bca510a946d..fdec2b025540 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1381,10 +1381,18 @@ IS_SUBPLATFORM(const struct drm_i915_private *i915,
>   #define INTEL_DISPLAY_ENABLED(dev_priv) \
>   	(drm_WARN_ON(&(dev_priv)->drm, !HAS_DISPLAY(dev_priv)), !(dev_priv)->params.disable_display)
>   
> +#if IS_ENABLED(CONFIG_X86)
>   static inline bool run_as_guest(void)
>   {
>   	return !hypervisor_is_type(X86_HYPER_NATIVE);
>   }
> +#elif IS_ENABLED(CONFIG_ARM64)
> +static inline bool run_as_guest(void)
> +{
> +	/* Not supported for arm64, so we return false  */
> +	return false;
> +}
> +#endif
>   
>   #define HAS_D12_PLANE_MINIMIZATION(dev_priv) (IS_ROCKETLAKE(dev_priv) || \
>   					      IS_ALDERLAKE_S(dev_priv))

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Lucas De Marchi]

On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>Wanted to ping this older thread to find out where we stand with this patch,
>Are we OK with the current state of these changes?
>
>With more recent information gathered from feedback on other patches, would
>we prefer changing this to a more arch-neutral control flow?
>
>e.g.
>#if IS_ENABLED(CONFIG_X86)
>...
>#else
>...
>#endif
>
>Would we also prefer this RFC series be merged or would it be preferred to
>create a new series instead?

for this specific function, that is used in only 2 places I think it's
ok to do:

	static inline bool run_as_guest(void)
	{
	#if IS_ENABLED(CONFIG_X86)
		return !hypervisor_is_type(X86_HYPER_NATIVE);
	#else	
		/* Not supported yet */
		return false;	
	#endif
	}

For PCH it doesn't really matter as we don't execute that function
for discrete. For intel_vtd_active() I figure anything other than
x86 would be fine with false here.

Jani, that this look good to you?

Lucas De Marchi

>
>Regards,
>Casey
>
>On 2/15/22 15:41, Casey Bowman wrote:
>>Splitting out run_as_guest into platform-specific functions
>>as arm64 does not support this functionality.
>>
>>Signed-off-by: Casey Bowman <casey.g.bowman@intel.com>
>>---
>>  drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++
>>  1 file changed, 8 insertions(+)
>>
>>diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
>>index 1bca510a946d..fdec2b025540 100644
>>--- a/drivers/gpu/drm/i915/i915_drv.h
>>+++ b/drivers/gpu/drm/i915/i915_drv.h
>>@@ -1381,10 +1381,18 @@ IS_SUBPLATFORM(const struct drm_i915_private *i915,
>>  #define INTEL_DISPLAY_ENABLED(dev_priv) \
>>  	(drm_WARN_ON(&(dev_priv)->drm, !HAS_DISPLAY(dev_priv)), !(dev_priv)->params.disable_display)
>>+#if IS_ENABLED(CONFIG_X86)
>>  static inline bool run_as_guest(void)
>>  {
>>  	return !hypervisor_is_type(X86_HYPER_NATIVE);
>>  }
>>+#elif IS_ENABLED(CONFIG_ARM64)
>>+static inline bool run_as_guest(void)
>>+{
>>+	/* Not supported for arm64, so we return false  */
>>+	return false;
>>+}
>>+#endif
>>  #define HAS_D12_PLANE_MINIMIZATION(dev_priv) (IS_ROCKETLAKE(dev_priv) || \
>>  					      IS_ALDERLAKE_S(dev_priv))
>

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Jani Nikula]

On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>Wanted to ping this older thread to find out where we stand with this patch,
>>Are we OK with the current state of these changes?
>>
>>With more recent information gathered from feedback on other patches, would
>>we prefer changing this to a more arch-neutral control flow?
>>
>>e.g.
>>#if IS_ENABLED(CONFIG_X86)
>>...
>>#else
>>...
>>#endif
>>
>>Would we also prefer this RFC series be merged or would it be preferred to
>>create a new series instead?
>
> for this specific function, that is used in only 2 places I think it's
> ok to do:
>
> 	static inline bool run_as_guest(void)
> 	{
> 	#if IS_ENABLED(CONFIG_X86)
> 		return !hypervisor_is_type(X86_HYPER_NATIVE);
> 	#else	
> 		/* Not supported yet */
> 		return false;	
> 	#endif
> 	}
>
> For PCH it doesn't really matter as we don't execute that function
> for discrete. For intel_vtd_active() I figure anything other than
> x86 would be fine with false here.
>
> Jani, that this look good to you?

It's more important to me to get this out of i915_drv.h, which is not
supposed to be a collection of random stuff anymore. I've sent patches
to this effect but they've stalled a bit.

In general I like the preprocessor control flow outside of functions,
i.e. completely separate function definitions, but for one-line function
implementations I guess this is fine. This is less important for me than
the first point.


BR,
Jani.


>
> Lucas De Marchi
>
>>
>>Regards,
>>Casey
>>
>>On 2/15/22 15:41, Casey Bowman wrote:
>>>Splitting out run_as_guest into platform-specific functions
>>>as arm64 does not support this functionality.
>>>
>>>Signed-off-by: Casey Bowman <casey.g.bowman@intel.com>
>>>---
>>>  drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++
>>>  1 file changed, 8 insertions(+)
>>>
>>>diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
>>>index 1bca510a946d..fdec2b025540 100644
>>>--- a/drivers/gpu/drm/i915/i915_drv.h
>>>+++ b/drivers/gpu/drm/i915/i915_drv.h
>>>@@ -1381,10 +1381,18 @@ IS_SUBPLATFORM(const struct drm_i915_private *i915,
>>>  #define INTEL_DISPLAY_ENABLED(dev_priv) \
>>>  	(drm_WARN_ON(&(dev_priv)->drm, !HAS_DISPLAY(dev_priv)), !(dev_priv)->params.disable_display)
>>>+#if IS_ENABLED(CONFIG_X86)
>>>  static inline bool run_as_guest(void)
>>>  {
>>>  	return !hypervisor_is_type(X86_HYPER_NATIVE);
>>>  }
>>>+#elif IS_ENABLED(CONFIG_ARM64)
>>>+static inline bool run_as_guest(void)
>>>+{
>>>+	/* Not supported for arm64, so we return false  */
>>>+	return false;
>>>+}
>>>+#endif
>>>  #define HAS_D12_PLANE_MINIMIZATION(dev_priv) (IS_ROCKETLAKE(dev_priv) || \
>>>  					      IS_ALDERLAKE_S(dev_priv))
>>

-- 
Jani Nikula, Intel Open Source Graphics Center

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Lucas De Marchi]

On Tue, Mar 22, 2022 at 12:21:59PM +0200, Jani Nikula wrote:
>On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>>Wanted to ping this older thread to find out where we stand with this patch,
>>>Are we OK with the current state of these changes?
>>>
>>>With more recent information gathered from feedback on other patches, would
>>>we prefer changing this to a more arch-neutral control flow?
>>>
>>>e.g.
>>>#if IS_ENABLED(CONFIG_X86)
>>>...
>>>#else
>>>...
>>>#endif
>>>
>>>Would we also prefer this RFC series be merged or would it be preferred to
>>>create a new series instead?
>>
>> for this specific function, that is used in only 2 places I think it's
>> ok to do:
>>
>> 	static inline bool run_as_guest(void)
>> 	{
>> 	#if IS_ENABLED(CONFIG_X86)
>> 		return !hypervisor_is_type(X86_HYPER_NATIVE);
>> 	#else	
>> 		/* Not supported yet */
>> 		return false;	
>> 	#endif
>> 	}
>>
>> For PCH it doesn't really matter as we don't execute that function
>> for discrete. For intel_vtd_active() I figure anything other than
>> x86 would be fine with false here.
>>
>> Jani, that this look good to you?
>
>It's more important to me to get this out of i915_drv.h, which is not
>supposed to be a collection of random stuff anymore. I've sent patches
>to this effect but they've stalled a bit.

do you have a patch moving this particular one? got a link?

Lucas De Marchi

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Jani Nikula]

On Tue, 22 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
> On Tue, Mar 22, 2022 at 12:21:59PM +0200, Jani Nikula wrote:
>>On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>>>Wanted to ping this older thread to find out where we stand with this patch,
>>>>Are we OK with the current state of these changes?
>>>>
>>>>With more recent information gathered from feedback on other patches, would
>>>>we prefer changing this to a more arch-neutral control flow?
>>>>
>>>>e.g.
>>>>#if IS_ENABLED(CONFIG_X86)
>>>>...
>>>>#else
>>>>...
>>>>#endif
>>>>
>>>>Would we also prefer this RFC series be merged or would it be preferred to
>>>>create a new series instead?
>>>
>>> for this specific function, that is used in only 2 places I think it's
>>> ok to do:
>>>
>>> 	static inline bool run_as_guest(void)
>>> 	{
>>> 	#if IS_ENABLED(CONFIG_X86)
>>> 		return !hypervisor_is_type(X86_HYPER_NATIVE);
>>> 	#else	
>>> 		/* Not supported yet */
>>> 		return false;	
>>> 	#endif
>>> 	}
>>>
>>> For PCH it doesn't really matter as we don't execute that function
>>> for discrete. For intel_vtd_active() I figure anything other than
>>> x86 would be fine with false here.
>>>
>>> Jani, that this look good to you?
>>
>>It's more important to me to get this out of i915_drv.h, which is not
>>supposed to be a collection of random stuff anymore. I've sent patches
>>to this effect but they've stalled a bit.
>
> do you have a patch moving this particular one? got a link?

Yeah, but it was basically shot down by Tvrtko [1], and I stalled there.

I'd just like to get all this cruft out of i915_drv.h. Whenever we have
a file where the name isn't super specific, we seem to have a tendency
of turning it into a dumping ground for random crap. So I'd really like
to move this out of there *before* expanding on it. 

BR,
Jani.


[1] https://patchwork.freedesktop.org/series/99852/


-- 
Jani Nikula, Intel Open Source Graphics Center

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Tvrtko Ursulin]

On 22/03/2022 14:49, Jani Nikula wrote:
> On Tue, 22 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>> On Tue, Mar 22, 2022 at 12:21:59PM +0200, Jani Nikula wrote:
>>> On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>>> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>>>> Wanted to ping this older thread to find out where we stand with this patch,
>>>>> Are we OK with the current state of these changes?
>>>>>
>>>>> With more recent information gathered from feedback on other patches, would
>>>>> we prefer changing this to a more arch-neutral control flow?
>>>>>
>>>>> e.g.
>>>>> #if IS_ENABLED(CONFIG_X86)
>>>>> ...
>>>>> #else
>>>>> ...
>>>>> #endif
>>>>>
>>>>> Would we also prefer this RFC series be merged or would it be preferred to
>>>>> create a new series instead?
>>>>
>>>> for this specific function, that is used in only 2 places I think it's
>>>> ok to do:
>>>>
>>>> 	static inline bool run_as_guest(void)
>>>> 	{
>>>> 	#if IS_ENABLED(CONFIG_X86)
>>>> 		return !hypervisor_is_type(X86_HYPER_NATIVE);
>>>> 	#else	
>>>> 		/* Not supported yet */
>>>> 		return false;	
>>>> 	#endif
>>>> 	}
>>>>
>>>> For PCH it doesn't really matter as we don't execute that function
>>>> for discrete. For intel_vtd_active() I figure anything other than
>>>> x86 would be fine with false here.
>>>>
>>>> Jani, that this look good to you?
>>>
>>> It's more important to me to get this out of i915_drv.h, which is not
>>> supposed to be a collection of random stuff anymore. I've sent patches
>>> to this effect but they've stalled a bit.
>>
>> do you have a patch moving this particular one? got a link?
> 
> Yeah, but it was basically shot down by Tvrtko [1], and I stalled there.
> 
> I'd just like to get all this cruft out of i915_drv.h. Whenever we have
> a file where the name isn't super specific, we seem to have a tendency
> of turning it into a dumping ground for random crap. So I'd really like
> to move this out of there *before* expanding on it.

Sounds like we had agreement on what tweaks to make and I conceded to 
live for now with the IMO wrongly named intel_vtd_run_as_guest.

(I mean I really disagree with file name being trumps, which I think 
this example illustrates - this is i915 asking whether the kernel is 
running as guest so intel_vtd_ prefix is just wrong. Intel VT-d is the 
iommu thingy so it makes no sense when called from PCH detection. But I 
have no better ideas at the moment. We can call it i915_run_as_guest, to 
signify function belongs to i915, but then we lose the first parameter 
names the function rule.)

But in any case I don't see that I created any blockers in this thread. 
AFAICS just a respin with intel_vtd_active taking struct device is 
needed and job done.

Regards,

Tvrtko

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Tvrtko Ursulin]

On 22/03/2022 15:18, Tvrtko Ursulin wrote:
> 
> On 22/03/2022 14:49, Jani Nikula wrote:
>> On Tue, 22 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>> On Tue, Mar 22, 2022 at 12:21:59PM +0200, Jani Nikula wrote:
>>>> On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>>>> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>>>>> Wanted to ping this older thread to find out where we stand with 
>>>>>> this patch,
>>>>>> Are we OK with the current state of these changes?
>>>>>>
>>>>>> With more recent information gathered from feedback on other 
>>>>>> patches, would
>>>>>> we prefer changing this to a more arch-neutral control flow?
>>>>>>
>>>>>> e.g.
>>>>>> #if IS_ENABLED(CONFIG_X86)
>>>>>> ...
>>>>>> #else
>>>>>> ...
>>>>>> #endif
>>>>>>
>>>>>> Would we also prefer this RFC series be merged or would it be 
>>>>>> preferred to
>>>>>> create a new series instead?
>>>>>
>>>>> for this specific function, that is used in only 2 places I think it's
>>>>> ok to do:
>>>>>
>>>>>     static inline bool run_as_guest(void)
>>>>>     {
>>>>>     #if IS_ENABLED(CONFIG_X86)
>>>>>         return !hypervisor_is_type(X86_HYPER_NATIVE);
>>>>>     #else
>>>>>         /* Not supported yet */
>>>>>         return false;
>>>>>     #endif
>>>>>     }
>>>>>
>>>>> For PCH it doesn't really matter as we don't execute that function
>>>>> for discrete. For intel_vtd_active() I figure anything other than
>>>>> x86 would be fine with false here.
>>>>>
>>>>> Jani, that this look good to you?
>>>>
>>>> It's more important to me to get this out of i915_drv.h, which is not
>>>> supposed to be a collection of random stuff anymore. I've sent patches
>>>> to this effect but they've stalled a bit.
>>>
>>> do you have a patch moving this particular one? got a link?
>>
>> Yeah, but it was basically shot down by Tvrtko [1], and I stalled there.
>>
>> I'd just like to get all this cruft out of i915_drv.h. Whenever we have
>> a file where the name isn't super specific, we seem to have a tendency
>> of turning it into a dumping ground for random crap. So I'd really like
>> to move this out of there *before* expanding on it.
> 
> Sounds like we had agreement on what tweaks to make and I conceded to 
> live for now with the IMO wrongly named intel_vtd_run_as_guest.
> 
> (I mean I really disagree with file name being trumps, which I think 
> this example illustrates - this is i915 asking whether the kernel is 
> running as guest so intel_vtd_ prefix is just wrong. Intel VT-d is the 
> iommu thingy so it makes no sense when called from PCH detection. But I 
> have no better ideas at the moment. We can call it i915_run_as_guest, to 
> signify function belongs to i915, but then we lose the first parameter 
> names the function rule.)
> 
> But in any case I don't see that I created any blockers in this thread. 
> AFAICS just a respin with intel_vtd_active taking struct device is 
> needed and job done.

Sorry now I see I also suggested moving intel_scanout_needs_vtd_wa, 
intel_ggtt_update_needs_vtd_wa and intel_vm_no_concurrent_access_wa all 
to their respective files. Which I think is also correct. They are all 
higher components which are asking intel_vtd a question and basing a 
decision upon the answer. I don't think intel_vtd.h should contain 
knowledge about a mix of other driver components.

Regards,

Tvrtko

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Jani Nikula]

On Tue, 22 Mar 2022, Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> wrote:
> On 22/03/2022 14:49, Jani Nikula wrote:
>> On Tue, 22 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>> On Tue, Mar 22, 2022 at 12:21:59PM +0200, Jani Nikula wrote:
>>>> On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>>>> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>>>>> Wanted to ping this older thread to find out where we stand with this patch,
>>>>>> Are we OK with the current state of these changes?
>>>>>>
>>>>>> With more recent information gathered from feedback on other patches, would
>>>>>> we prefer changing this to a more arch-neutral control flow?
>>>>>>
>>>>>> e.g.
>>>>>> #if IS_ENABLED(CONFIG_X86)
>>>>>> ...
>>>>>> #else
>>>>>> ...
>>>>>> #endif
>>>>>>
>>>>>> Would we also prefer this RFC series be merged or would it be preferred to
>>>>>> create a new series instead?
>>>>>
>>>>> for this specific function, that is used in only 2 places I think it's
>>>>> ok to do:
>>>>>
>>>>> 	static inline bool run_as_guest(void)
>>>>> 	{
>>>>> 	#if IS_ENABLED(CONFIG_X86)
>>>>> 		return !hypervisor_is_type(X86_HYPER_NATIVE);
>>>>> 	#else	
>>>>> 		/* Not supported yet */
>>>>> 		return false;	
>>>>> 	#endif
>>>>> 	}
>>>>>
>>>>> For PCH it doesn't really matter as we don't execute that function
>>>>> for discrete. For intel_vtd_active() I figure anything other than
>>>>> x86 would be fine with false here.
>>>>>
>>>>> Jani, that this look good to you?
>>>>
>>>> It's more important to me to get this out of i915_drv.h, which is not
>>>> supposed to be a collection of random stuff anymore. I've sent patches
>>>> to this effect but they've stalled a bit.
>>>
>>> do you have a patch moving this particular one? got a link?
>> 
>> Yeah, but it was basically shot down by Tvrtko [1], and I stalled there.
>> 
>> I'd just like to get all this cruft out of i915_drv.h. Whenever we have
>> a file where the name isn't super specific, we seem to have a tendency
>> of turning it into a dumping ground for random crap. So I'd really like
>> to move this out of there *before* expanding on it.
>
> Sounds like we had agreement on what tweaks to make and I conceded to 
> live for now with the IMO wrongly named intel_vtd_run_as_guest.
>
> (I mean I really disagree with file name being trumps, which I think 
> this example illustrates - this is i915 asking whether the kernel is 
> running as guest so intel_vtd_ prefix is just wrong. Intel VT-d is the 
> iommu thingy so it makes no sense when called from PCH detection. But I 
> have no better ideas at the moment. We can call it i915_run_as_guest, to 
> signify function belongs to i915, but then we lose the first parameter 
> names the function rule.)

I think the "first parameter names the function" rule has backfired in
gem/gt land, because it's pretty difficult to figure out *where* you'd
expect to find or place functions.

BR,
Jani.

>
> But in any case I don't see that I created any blockers in this thread. 
> AFAICS just a respin with intel_vtd_active taking struct device is 
> needed and job done.
>
> Regards,
>
> Tvrtko

-- 
Jani Nikula, Intel Open Source Graphics Center

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Tvrtko Ursulin]

On 22/03/2022 15:26, Jani Nikula wrote:
> On Tue, 22 Mar 2022, Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> wrote:
>> On 22/03/2022 14:49, Jani Nikula wrote:
>>> On Tue, 22 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>>> On Tue, Mar 22, 2022 at 12:21:59PM +0200, Jani Nikula wrote:
>>>>> On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>>>>> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>>>>>> Wanted to ping this older thread to find out where we stand with this patch,
>>>>>>> Are we OK with the current state of these changes?
>>>>>>>
>>>>>>> With more recent information gathered from feedback on other patches, would
>>>>>>> we prefer changing this to a more arch-neutral control flow?
>>>>>>>
>>>>>>> e.g.
>>>>>>> #if IS_ENABLED(CONFIG_X86)
>>>>>>> ...
>>>>>>> #else
>>>>>>> ...
>>>>>>> #endif
>>>>>>>
>>>>>>> Would we also prefer this RFC series be merged or would it be preferred to
>>>>>>> create a new series instead?
>>>>>>
>>>>>> for this specific function, that is used in only 2 places I think it's
>>>>>> ok to do:
>>>>>>
>>>>>> 	static inline bool run_as_guest(void)
>>>>>> 	{
>>>>>> 	#if IS_ENABLED(CONFIG_X86)
>>>>>> 		return !hypervisor_is_type(X86_HYPER_NATIVE);
>>>>>> 	#else	
>>>>>> 		/* Not supported yet */
>>>>>> 		return false;	
>>>>>> 	#endif
>>>>>> 	}
>>>>>>
>>>>>> For PCH it doesn't really matter as we don't execute that function
>>>>>> for discrete. For intel_vtd_active() I figure anything other than
>>>>>> x86 would be fine with false here.
>>>>>>
>>>>>> Jani, that this look good to you?
>>>>>
>>>>> It's more important to me to get this out of i915_drv.h, which is not
>>>>> supposed to be a collection of random stuff anymore. I've sent patches
>>>>> to this effect but they've stalled a bit.
>>>>
>>>> do you have a patch moving this particular one? got a link?
>>>
>>> Yeah, but it was basically shot down by Tvrtko [1], and I stalled there.
>>>
>>> I'd just like to get all this cruft out of i915_drv.h. Whenever we have
>>> a file where the name isn't super specific, we seem to have a tendency
>>> of turning it into a dumping ground for random crap. So I'd really like
>>> to move this out of there *before* expanding on it.
>>
>> Sounds like we had agreement on what tweaks to make and I conceded to
>> live for now with the IMO wrongly named intel_vtd_run_as_guest.
>>
>> (I mean I really disagree with file name being trumps, which I think
>> this example illustrates - this is i915 asking whether the kernel is
>> running as guest so intel_vtd_ prefix is just wrong. Intel VT-d is the
>> iommu thingy so it makes no sense when called from PCH detection. But I
>> have no better ideas at the moment. We can call it i915_run_as_guest, to
>> signify function belongs to i915, but then we lose the first parameter
>> names the function rule.)
> 
> I think the "first parameter names the function" rule has backfired in
> gem/gt land, because it's pretty difficult to figure out *where* you'd
> expect to find or place functions.

Hey surely is not that bad. And I am sure if I tried to add some display 
feature that there's a chance I'd manage to misplace something. :))

No scheme is perfect and there are always edge cases, ambiguities and 
always work to cleanup further because it all evolved rather than 
started from scratch. If you know what the function you wrote is about, 
surely you can place it into a file whose name suggests it is the right 
area. If you want the example of GT, there is a nice collection of files 
per functional area.. intel_gt_<suffix>.. interrupts, power management, 
requests, debugfs, etc.

And if you look for functions you did not write, I certainly suggest 
using ctags rather than try opening random files. I think driver is just 
too big for the latter approach.

Regards,

Tvrtko


> BR,
> Jani.
> 
>>
>> But in any case I don't see that I created any blockers in this thread.
>> AFAICS just a respin with intel_vtd_active taking struct device is
>> needed and job done.
>>
>> Regards,
>>
>> Tvrtko
> 

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Jani Nikula]

On Tue, 22 Mar 2022, Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> wrote:
> On 22/03/2022 15:26, Jani Nikula wrote:
>> On Tue, 22 Mar 2022, Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> wrote:
>>> On 22/03/2022 14:49, Jani Nikula wrote:
>>>> On Tue, 22 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>>>> On Tue, Mar 22, 2022 at 12:21:59PM +0200, Jani Nikula wrote:
>>>>>> On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>>>>>> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>>>>>>> Wanted to ping this older thread to find out where we stand with this patch,
>>>>>>>> Are we OK with the current state of these changes?
>>>>>>>>
>>>>>>>> With more recent information gathered from feedback on other patches, would
>>>>>>>> we prefer changing this to a more arch-neutral control flow?
>>>>>>>>
>>>>>>>> e.g.
>>>>>>>> #if IS_ENABLED(CONFIG_X86)
>>>>>>>> ...
>>>>>>>> #else
>>>>>>>> ...
>>>>>>>> #endif
>>>>>>>>
>>>>>>>> Would we also prefer this RFC series be merged or would it be preferred to
>>>>>>>> create a new series instead?
>>>>>>>
>>>>>>> for this specific function, that is used in only 2 places I think it's
>>>>>>> ok to do:
>>>>>>>
>>>>>>> 	static inline bool run_as_guest(void)
>>>>>>> 	{
>>>>>>> 	#if IS_ENABLED(CONFIG_X86)
>>>>>>> 		return !hypervisor_is_type(X86_HYPER_NATIVE);
>>>>>>> 	#else	
>>>>>>> 		/* Not supported yet */
>>>>>>> 		return false;	
>>>>>>> 	#endif
>>>>>>> 	}
>>>>>>>
>>>>>>> For PCH it doesn't really matter as we don't execute that function
>>>>>>> for discrete. For intel_vtd_active() I figure anything other than
>>>>>>> x86 would be fine with false here.
>>>>>>>
>>>>>>> Jani, that this look good to you?
>>>>>>
>>>>>> It's more important to me to get this out of i915_drv.h, which is not
>>>>>> supposed to be a collection of random stuff anymore. I've sent patches
>>>>>> to this effect but they've stalled a bit.
>>>>>
>>>>> do you have a patch moving this particular one? got a link?
>>>>
>>>> Yeah, but it was basically shot down by Tvrtko [1], and I stalled there.
>>>>
>>>> I'd just like to get all this cruft out of i915_drv.h. Whenever we have
>>>> a file where the name isn't super specific, we seem to have a tendency
>>>> of turning it into a dumping ground for random crap. So I'd really like
>>>> to move this out of there *before* expanding on it.
>>>
>>> Sounds like we had agreement on what tweaks to make and I conceded to
>>> live for now with the IMO wrongly named intel_vtd_run_as_guest.
>>>
>>> (I mean I really disagree with file name being trumps, which I think
>>> this example illustrates - this is i915 asking whether the kernel is
>>> running as guest so intel_vtd_ prefix is just wrong. Intel VT-d is the
>>> iommu thingy so it makes no sense when called from PCH detection. But I
>>> have no better ideas at the moment. We can call it i915_run_as_guest, to
>>> signify function belongs to i915, but then we lose the first parameter
>>> names the function rule.)
>> 
>> I think the "first parameter names the function" rule has backfired in
>> gem/gt land, because it's pretty difficult to figure out *where* you'd
>> expect to find or place functions.
>
> Hey surely is not that bad. And I am sure if I tried to add some display 
> feature that there's a chance I'd manage to misplace something. :))
>
> No scheme is perfect and there are always edge cases, ambiguities and 
> always work to cleanup further because it all evolved rather than 
> started from scratch. If you know what the function you wrote is about, 
> surely you can place it into a file whose name suggests it is the right 
> area. If you want the example of GT, there is a nice collection of files 
> per functional area.. intel_gt_<suffix>.. interrupts, power management, 
> requests, debugfs, etc.
>
> And if you look for functions you did not write, I certainly suggest 
> using ctags rather than try opening random files. I think driver is just 
> too big for the latter approach.

Obviously I use code tagging and search. The point was more about the
surprise in expecting to find a function in some place, and finding it
in a totally different place. And obviously for finding the right place
for new functions.

BR,
Jani.



>
> Regards,
>
> Tvrtko
>
>
>> BR,
>> Jani.
>> 
>>>
>>> But in any case I don't see that I created any blockers in this thread.
>>> AFAICS just a respin with intel_vtd_active taking struct device is
>>> needed and job done.
>>>
>>> Regards,
>>>
>>> Tvrtko
>> 

-- 
Jani Nikula, Intel Open Source Graphics Center

Re: [Intel-gfx] [RFC PATCH v3 1/1] i915/drm: Split out x86/arm64 for run_as_guest

[Lucas De Marchi]

On Tue, Mar 22, 2022 at 04:49:53PM +0200, Jani Nikula wrote:
>On Tue, 22 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>> On Tue, Mar 22, 2022 at 12:21:59PM +0200, Jani Nikula wrote:
>>>On Mon, 21 Mar 2022, Lucas De Marchi <lucas.demarchi@intel.com> wrote:
>>>> On Mon, Mar 21, 2022 at 04:34:49PM -0700, Casey Bowman wrote:
>>>>>Wanted to ping this older thread to find out where we stand with this patch,
>>>>>Are we OK with the current state of these changes?
>>>>>
>>>>>With more recent information gathered from feedback on other patches, would
>>>>>we prefer changing this to a more arch-neutral control flow?
>>>>>
>>>>>e.g.
>>>>>#if IS_ENABLED(CONFIG_X86)
>>>>>...
>>>>>#else
>>>>>...
>>>>>#endif
>>>>>
>>>>>Would we also prefer this RFC series be merged or would it be preferred to
>>>>>create a new series instead?
>>>>
>>>> for this specific function, that is used in only 2 places I think it's
>>>> ok to do:
>>>>
>>>> 	static inline bool run_as_guest(void)
>>>> 	{
>>>> 	#if IS_ENABLED(CONFIG_X86)
>>>> 		return !hypervisor_is_type(X86_HYPER_NATIVE);
>>>> 	#else	
>>>> 		/* Not supported yet */
>>>> 		return false;	
>>>> 	#endif
>>>> 	}
>>>>
>>>> For PCH it doesn't really matter as we don't execute that function
>>>> for discrete. For intel_vtd_active() I figure anything other than
>>>> x86 would be fine with false here.
>>>>
>>>> Jani, that this look good to you?
>>>
>>>It's more important to me to get this out of i915_drv.h, which is not
>>>supposed to be a collection of random stuff anymore. I've sent patches
>>>to this effect but they've stalled a bit.
>>
>> do you have a patch moving this particular one? got a link?
>
>Yeah, but it was basically shot down by Tvrtko [1], and I stalled there.
>
>I'd just like to get all this cruft out of i915_drv.h. Whenever we have
>a file where the name isn't super specific, we seem to have a tendency
>of turning it into a dumping ground for random crap. So I'd really like
>to move this out of there *before* expanding on it.

ok, I took a look there and it seems there is still some discussion on
where to move it. The place you moved it to is not ideal as
`run_as_guest()` has nothing to do with vt-d, even if it's used by one
of those functions. It's also used by the PCH detection/fallback code.

So, I think this is very much orthogonal to moving it and we are not
really extending: this just doesn't make much sense for other archs.
So my proposal is that we continue with this as is and move it when
we have a rough agreement on where to move it to.

As much as I hate the i915_utils.[ch] and I've been trying to reduce it,
I can't think of a best place. Other than of course come up with an
arch-neutral API to add to the kernel

 From some quick grep, include/linux/hypervisor.h might be a good place.
But again, I think it should be orthogonal to what this is doing.

thanks
Lucas De Marchi

>
>BR,
>Jani.
>
>
>[1] https://patchwork.freedesktop.org/series/99852/
>
>
>-- 
>Jani Nikula, Intel Open Source Graphics Center

[Intel-gfx] ✗ Fi.CI.BUILD: failure for Splitting up platform-specific calls (rev3)

[Patchwork]

== Series Details ==

Series: Splitting up platform-specific calls (rev3)
URL   : https://patchwork.freedesktop.org/series/99126/
State : failure

== Summary ==

Applying: i915/drm: Split out x86/arm64 for run_as_guest
Using index info to reconstruct a base tree...
M	drivers/gpu/drm/i915/i915_drv.h
Falling back to patching base and 3-way merge...
Auto-merging drivers/gpu/drm/i915/i915_drv.h
CONFLICT (content): Merge conflict in drivers/gpu/drm/i915/i915_drv.h
error: Failed to merge in the changes.
hint: Use 'git am --show-current-patch=diff' to see the failed patch
Patch failed at 0001 i915/drm: Split out x86/arm64 for run_as_guest
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".