On 6/21/2023 12:26 PM, Andrew Lunn wrote:
I think what you're asking for is another layer of indirection
like CONFIG_WBRF in addition to CONFIG_ACPI_WBRF.
Producers would call functions like wbrf_supported_producer()
where the source file is not guarded behind CONFIG_ACPI_WBRF,
but instead by CONFIG_WBRF and locally use CONFIG_ACPI_WBRF within
it. So a producer could look like this:
bool wbrf_supported_producer(struct device *dev)
{
#ifdef CONFIG_ACPI_WBRF
struct acpi_device *adev = ACPI_COMPANION(dev);
if (adev)
return check_acpi_wbrf(adev->handle,
WBRF_REVISION,
1ULL << WBRF_RECORD);
#endif
return -ENODEV;
}
EXPORT_SYMBOL_GPL(wbrf_supported_producer);
And then adding/removing could look something like this
int wbrf_add_exclusion(struct device *dev,
struct wbrf_ranges_in *in)
{
#ifdef CONFIG_ACPI_WBRF
struct acpi_device *adev = ACPI_COMPANION(dev);
if (adev)
return wbrf_record(adev, WBRF_RECORD_ADD, in);
#endif
return -ENODEV;
}
EXPORT_SYMBOL_GPL(wbrf_add_exclusion);
int wbrf_remove_exclusion(struct device *dev,
struct wbrf_ranges_in *in)
{
#ifdef CONFIG_ACPI_WBRF
struct acpi_device *adev = ACPI_COMPANION(dev);
if (adev)
return wbrf_record(adev, WBRF_RECORD_REMOVE, in);
#endif
return -ENODEV;
}
EXPORT_SYMBOL_GPL(wbrf_remove_exclusion);
Yes, this looks a lot better.
But what about notifications?
Once you implement this it gets a lot more complex and the driver
consumers would need
to know more about the kernel's implementation. For example consumers
need a
notifier block like:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index e3e2e6e3b485..146fe3c43343 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1066,6 +1066,8 @@ struct amdgpu_device {
bool job_hang;
bool dc_enabled;
+
+ struct notifier_block wbrf_notifier;
};
static inline struct amdgpu_device *drm_to_adev(struct drm_device *ddev)
And then would need matching notifier functions like:
static int amdgpu_wbrf_frequencies_notifier(struct notifier_block *nb,
unsigned long action, void *_arg)
And we'd need to set up a chain to be used in this case in the WBRF code:
static BLOCKING_NOTIFIER_HEAD(wbrf_chain_head);
int wbrf_register_notifier(struct notifier_block *nb)
{
return blocking_notifier_chain_register(&wbrf_chain_head, nb);
}
EXPORT_SYMBOL_GPL(wbrf_register_notifier);
int wbrf_unregister_notifier(struct notifier_block *nb)
{
return blocking_notifier_chain_unregister(&wbrf_chain_head, nb);
}
EXPORT_SYMBOL_GPL(wbrf_unregister_notifier);
And consumer would need to call it, but only if CONFIG_WBRF_ACPI isn't set.
Add/remove functions can easily call something like:
blocking_notifier_call_chain(&wbrf_chain_head, action, data);
With all of this complexity and (effectively) dead code for ACPI vs non-ACPI
path I really have to ask why wouldn't a non-AMD implementation be able to
do this as ACPI?
I don't see why it couldn't be a DT/ACPI hybrid solution for ARM64.