ACPI / PM: Traverse video_device_list for backlight restoration

.bind_info in active_list[] is NULL by construction - and used to determine the acpi_video_device pointer during backlight restoration during resume.

Using the list instead fixes backlight restoration for systems where the BIOS doesn't do this by itself already.

---
 drivers/acpi/video.c |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)

index 9865d46..887ac2b 100644
@@ -2346,7 +2346,6 @@ static int acpi_video_resume(struct notifier_block *nb,
 {
 	struct acpi_video_bus *video;
 	struct acpi_video_device *video_device;
-	int i;
 
 	switch (val) {
 	case PM_HIBERNATION_PREPARE:

@@ -2359,11 +2358,12 @@ static int acpi_video_resume(struct notifier_block *nb,
 
 	dev_info(&video->device->dev, "Restoring backlight state\n");
 
-	for (i = 0; i < video->attached_count; i++) {
-		video_device = video->attached_array[i].bind_info;
+	mutex_lock(&video->device_list_lock);
+	list_for_each_entry(video_device, &video->video_device_list, entry) {
 		if (video_device && video_device->backlight)
 			acpi_video_set_brightness(video_device->backlight);
 	}
+	mutex_unlock(&video->device_list_lock);
 
 	return NOTIFY_OK;
 }

1.6.0.2

A bit of background about this patch:

acpi_video_bus_get_devices() explicitly calls acpi_video_device_enumerate() before acpi_video_bus_get_one_device(), which does acpi_video_device_bind().

In _enumerate(), active_list[] is created, with .bind_info explicitly set to NULL, while the related .bind_info entry in video_device_list is set to the acpi_video_device ptr in _bind().

The .bind_info is later used in acpi_video_resume() to re-set the backlight
- but it's only evaluated on the active_list[], on which all .bind_info are NULL by construction. This results in backlight not being restored, if the BIOS doesn't do this by itself already.

The patch resolves this by replacing the active_list[] traversal by a video_device_list traversal, but due to the nature of the issue I'm unsure whether the original issue isn't of more principal quality. I don't understand why the list is (partially) transformed into an array in the first place, especially as both the array *and* the list are used in the code...

Thanks

Matthias

Matthias Hopf <[email protected]>      __        __   __
Maxfeldstr. 5 / 90409 Nuernberg   (_   | |  (_   |__          [email protected]
Phone +49-911-74053-715           __)  |_|  __)  |__  R & D   www.mshopf.de

A bit of background about this patch:

acpi_video_bus_get_devices() explicitly calls acpi_video_device_enumerate() before acpi_video_bus_get_one_device(), which does acpi_video_device_bind().

In _enumerate(), active_list[] is created, with .bind_info explicitly set to NULL, while the related .bind_info entry in video_device_list is set to the acpi_video_device ptr in _bind().

The .bind_info is later used in acpi_video_resume() to re-set the backlight
- but it's only evaluated on the active_list[], on which all .bind_info are NULL by construction. This results in backlight not being restored, if the BIOS doesn't do this by itself already.

The patch resolves this by replacing the active_list[] traversal by a video_device_list traversal, but due to the nature of the issue I'm unsure whether the original issue isn't of more principal quality. I don't understand why the list is (partially) transformed into an array in the first place, especially as both the array *and* the list are used in the code...

Thanks

Matthias

Matthias Hopf <[email protected]>      __        __   __
Maxfeldstr. 5 / 90409 Nuernberg   (_   | |  (_   |__          [email protected]
Phone +49-911-74053-715           __)  |_|  __)  |__  R & D   www.mshopf.de

why? .bind_info in active_list is initialized in acpi_video_device_bind.

no, the active_list is an array in fact, which equals acpi_video_bus->attached_array. Maybe we need to change it to a more proper name.

I'm still wondering why the patch works for you. could you please attach the acpidump output of this laptop?

please rebuild your kernel with CONFIG_ACPI_DEBUG=y, and reboot with kernel parameter "acpi.debug_layer=0x10 acpi.debug_level=0x400". and attach the dmesg output after resume, both w/ and w/o the patch you attached.

thanks,
rui

why? .bind_info in active_list is initialized in acpi_video_device_bind.

no, the active_list is an array in fact, which equals acpi_video_bus->attached_array. Maybe we need to change it to a more proper name.

I'm still wondering why the patch works for you. could you please attach the acpidump output of this laptop?

please rebuild your kernel with CONFIG_ACPI_DEBUG=y, and reboot with kernel parameter "acpi.debug_layer=0x10 acpi.debug_level=0x400". and attach the dmesg output after resume, both w/ and w/o the patch you attached.

thanks,
rui

Hm. Right. acpi_video_bus_get_one_device() is called after _enumerate(). Dunno why I didn't read this correctly.

:-)

Same to me.

Will do.

The most intriguing thing here is that the system starts writing to /var/log/messages continuously, thus dmesg output is sort-of worthless. This might be a reasonable side effect, it might be an indication of the original issue.
I'll try to compile something for you.

Thanks

Matthias

Matthias Hopf <[email protected]>      __        __   __
Maxfeldstr. 5 / 90409 Nuernberg   (_   | |  (_   |__          [email protected]
Phone +49-911-74053-715           __)  |_|  __)  |__  R & D   www.mshopf.de

Hm. Right. acpi_video_bus_get_one_device() is called after _enumerate(). Dunno why I didn't read this correctly.

:-)

Same to me.

Will do.

The most intriguing thing here is that the system starts writing to /var/log/messages continuously, thus dmesg output is sort-of worthless. This might be a reasonable side effect, it might be an indication of the original issue.
I'll try to compile something for you.

Thanks

Matthias

Matthias Hopf <[email protected]>      __        __   __
Maxfeldstr. 5 / 90409 Nuernberg   (_   | |  (_   |__          [email protected]
Phone +49-911-74053-715           __)  |_|  __)  |__  R & D   www.mshopf.de