Skip to contentSkip to conversation

radeon issues on MacBook Pro 8,2

DISCUSSION Seth Forshee 3 participants 12 messages Jan 19, 2012 – Feb 1, 2012

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

I'm seeing several issues related to the radeon driver on a MacBook Pro 8,2 with the following graphics card:

ATI Technologies Inc Whistler [AMD Radeon HD 6600M Series] [1002:6741]

All problems were observed when using kernel version 3.2.1. None are seen when using fglrx.

1. Excessive power draw. When using the radeon driver ACPI reports a power draw of about 30W on an idle desktop. Using fglrx brings this number down to 15W.

2. Occasional long delays when suspending. When this happens I see messages like following in dmesg:

[drm:atom_op_jump] *ERROR* atombios stuck in loop for more than 5secs aborting [drm:atom_execute_table_locked] *ERROR* atombios stuck executing D44E (len 62, WS 0, PS 0) @ 0xD46A

Sometimes one of suspend or resume hangs completely, but I can't tell which and am not sure whether or not it's related. I'm also testing a Mac Mini with the exact same card which does not seem to suffer from this issue.

I ran a bisections that identified f8d0edd (drm/radeon/kms: improve DP detect logic) as introducing problems with suspend, and reverting this patch on top of 3.2.1 does seem to eliminate both issues.

3. When the LVDS panel is powered off and back on, the display flickers, as if the backlight is cycling rapidly between low and high brightness. If the panel is left on this effect gradually lessens and is eventually no longer noticable. This is not seen with fglrx.

Thanks,
Seth

#1 lore raw
Quoted 11 lines · Seth Forshee

I'm seeing several issues related to the radeon driver on a MacBook Pro 8,2 with the following graphics card:

 ATI Technologies Inc Whistler [AMD Radeon HD 6600M Series] [1002:6741]

All problems were observed when using kernel version 3.2.1. None are seen when using fglrx.

 1. Excessive power draw. When using the radeon driver ACPI reports a power draw of about 30W on an idle desktop. Using fglrx brings this number down to 15W.

The power saving features of the open source driver are not yet as good as the closed source driver. Please see the power management section of this page (http://wiki.x.org/wiki/RadeonFeature) for more info on the options currently available.

Quoted 7 lines

 2. Occasional long delays when suspending. When this happens I see messages like following in dmesg:

     [drm:atom_op_jump] *ERROR* atombios stuck in loop for more than 5secs aborting [drm:atom_execute_table_locked] *ERROR* atombios stuck executing D44E (len 62, WS 0, PS 0) @ 0xD46A

Quoted 9 lines

   Sometimes one of suspend or resume hangs completely, but I can't tell which and am not sure whether or not it's related. I'm also testing a Mac Mini with the exact same card which does not seem to suffer from this issue.

   I ran a bisections that identified f8d0edd (drm/radeon/kms: improve DP detect logic) as introducing problems with suspend, and reverting this patch on top of 3.2.1 does seem to eliminate both issues.

That patch doesn't really affect the modesetting paths directly; it looks like a red herring to me.

Quoted 6 lines

 3. When the LVDS panel is powered off and back on, the display flickers, as if the backlight is cycling rapidly between low and high brightness. If the panel is left on this effect gradually lessens and is eventually no longer noticable. This is not seen with fglrx.

For the sake of tracking this properly, it would probably be best to file a bug:
https://bugs.freedesktop.org
Product: DRI
Component: DRM/Radeon
What connectors does your card actually have on it? Please attach a copy of your dmesg output and vbios:
(as root)
(use lspci to get the bus id)
cd /sys/bus/pci/devices/<pci bus id>
echo 1 > rom
cat rom > /tmp/vbios.rom
echo 0 > rom

Also, keep in mind that this is a mac so it's likely there may be wonky mac specific problems.

Alex

Quoted 6 lines

Thanks,
Seth
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel

#2 lore raw
Quoted 10 lines · Alex Deucher
Quoted 8 lines · Seth Forshee

I'm seeing several issues related to the radeon driver on a MacBook Pro 8,2 with the following graphics card:

 ATI Technologies Inc Whistler [AMD Radeon HD 6600M Series] [1002:6741]

All problems were observed when using kernel version 3.2.1. None are seen when using fglrx.

Quoted 8 lines
Quoted 3 lines

 1. Excessive power draw. When using the radeon driver ACPI reports a power draw of about 30W on an idle desktop. Using fglrx brings this number down to 15W.

The power saving features of the open source driver are not yet as good as the closed source driver. Please see the power management section of this page (http://wiki.x.org/wiki/RadeonFeature) for more info on the options currently available.

The dynpm option makes a small difference, saving about 2W. I did notice an ocassional flash on the screen with this option, and the same flash each time I changed the power options.

The only thing that improves things significatnly is the "low" profile method, which gets it down to about 18.5W.

Quoted 11 lines
Quoted 11 lines

 2. Occasional long delays when suspending. When this happens I see messages like following in dmesg:

     [drm:atom_op_jump] *ERROR* atombios stuck in loop for more than 5secs aborting [drm:atom_execute_table_locked] *ERROR* atombios stuck executing D44E (len 62, WS 0, PS 0) @ 0xD46A

   Sometimes one of suspend or resume hangs completely, but I can't tell which and am not sure whether or not it's related. I'm also testing a Mac Mini with the exact same card which does not seem to suffer from this issue.

Quoted 7 lines
Quoted 4 lines

   I ran a bisections that identified f8d0edd (drm/radeon/kms: improve DP detect logic) as introducing problems with suspend, and reverting this patch on top of 3.2.1 does seem to eliminate both issues.

That patch doesn't really affect the modesetting paths directly; it looks like a red herring to me.

Perhaps. I just started a run of 200 s3 cycles with the patch reverted to see if I can reproduce the issues. I can usually trigger the problem with 15 or fewer s3 cycles.

Quoted 20 lines
Quoted 6 lines

 3. When the LVDS panel is powered off and back on, the display flickers, as if the backlight is cycling rapidly between low and high brightness. If the panel is left on this effect gradually lessens and is eventually no longer noticable. This is not seen with fglrx.

For the sake of tracking this properly, it would probably be best to file a bug:
https://bugs.freedesktop.org
Product: DRI
Component: DRM/Radeon
What connectors does your card actually have on it? Please attach a copy of your dmesg output and vbios:
(as root)
(use lspci to get the bus id)
cd /sys/bus/pci/devices/<pci bus id>
echo 1 > rom
cat rom > /tmp/vbios.rom
echo 0 > rom

Quoted 2 lines

Also, keep in mind that this is a mac so it's likely there may be wonky mac specific problems.

I filed bug #44955 for the flickering issue.

Thanks,
Seth

#3 lore raw
Quoted 16 lines · Seth Forshee
Quoted 16 lines
Quoted 15 lines

 2. Occasional long delays when suspending. When this happens I see messages like following in dmesg:

     [drm:atom_op_jump] *ERROR* atombios stuck in loop for more than 5secs aborting [drm:atom_execute_table_locked] *ERROR* atombios stuck executing D44E (len 62, WS 0, PS 0) @ 0xD46A

   Sometimes one of suspend or resume hangs completely, but I can't tell which and am not sure whether or not it's related. I'm also testing a Mac Mini with the exact same card which does not seem to suffer from this issue.

   I ran a bisections that identified f8d0edd (drm/radeon/kms: improve DP detect logic) as introducing problems with suspend, and reverting this patch on top of 3.2.1 does seem to eliminate both issues.

Quoted 6 lines
Quoted 2 lines

That patch doesn't really affect the modesetting paths directly; it looks like a red herring to me.

Perhaps. I just started a run of 200 s3 cycles with the patch reverted to see if I can reproduce the issues. I can usually trigger the problem with 15 or fewer s3 cycles.

The machine completed 200 s3 cycles with the patch reverted without long delays, hangs, or any atombios error messages. Without reverting it doesn't get through many at all before experiencing the errors and long delays or hangs.

I added a printout of the connector status resulting from the logic that was changed by f8d0edd. With the logic prior to this commit it consistently returns the status as disconnected, which is correct as I have nothing connected. But with the improved logic the status is sometimes reported as connected, and these coincide with the atombios errors.

Seth

#4 lore raw
Quoted 29 lines · Seth Forshee
Quoted 16 lines · Seth Forshee
Quoted 16 lines
Quoted 15 lines

 2. Occasional long delays when suspending. When this happens I see messages like following in dmesg:

     [drm:atom_op_jump] *ERROR* atombios stuck in loop for more than 5secs aborting [drm:atom_execute_table_locked] *ERROR* atombios stuck executing D44E (len 62, WS 0, PS 0) @ 0xD46A

   Sometimes one of suspend or resume hangs completely, but I can't tell which and am not sure whether or not it's related. I'm also testing a Mac Mini with the exact same card which does not seem to suffer from this issue.

   I ran a bisections that identified f8d0edd (drm/radeon/kms: improve DP detect logic) as introducing problems with suspend, and reverting this patch on top of 3.2.1 does seem to eliminate both issues.

Quoted 6 lines
Quoted 2 lines

That patch doesn't really affect the modesetting paths directly; it looks like a red herring to me.

Perhaps. I just started a run of 200 s3 cycles with the patch reverted to see if I can reproduce the issues. I can usually trigger the problem with 15 or fewer s3 cycles.

The machine completed 200 s3 cycles with the patch reverted without long delays, hangs, or any atombios error messages. Without reverting it doesn't get through many at all before experiencing the errors and long delays or hangs.

Quoted 7 lines

I added a printout of the connector status resulting from the logic that was changed by f8d0edd. With the logic prior to this commit it consistently returns the status as disconnected, which is correct as I have nothing connected. But with the improved logic the status is sometimes reported as connected, and these coincide with the atombios errors.

Do any of the disconnected displayport connectors get misdetected as connected during normal operation or does it only happen during suspend/resume?

Alex

#5 lore raw
Quoted 39 lines · Alex Deucher
Quoted 29 lines · Seth Forshee
Quoted 16 lines · Seth Forshee
Quoted 16 lines
Quoted 15 lines

 2. Occasional long delays when suspending. When this happens I see messages like following in dmesg:

     [drm:atom_op_jump] *ERROR* atombios stuck in loop for more than 5secs aborting [drm:atom_execute_table_locked] *ERROR* atombios stuck executing D44E (len 62, WS 0, PS 0) @ 0xD46A

   Sometimes one of suspend or resume hangs completely, but I can't tell which and am not sure whether or not it's related. I'm also testing a Mac Mini with the exact same card which does not seem to suffer from this issue.

   I ran a bisections that identified f8d0edd (drm/radeon/kms: improve DP detect logic) as introducing problems with suspend, and reverting this patch on top of 3.2.1 does seem to eliminate both issues.

Quoted 6 lines
Quoted 2 lines

That patch doesn't really affect the modesetting paths directly; it looks like a red herring to me.

Perhaps. I just started a run of 200 s3 cycles with the patch reverted to see if I can reproduce the issues. I can usually trigger the problem with 15 or fewer s3 cycles.

The machine completed 200 s3 cycles with the patch reverted without long delays, hangs, or any atombios error messages. Without reverting it doesn't get through many at all before experiencing the errors and long delays or hangs.

Quoted 7 lines

I added a printout of the connector status resulting from the logic that was changed by f8d0edd. With the logic prior to this commit it consistently returns the status as disconnected, which is correct as I have nothing connected. But with the improved logic the status is sometimes reported as connected, and these coincide with the atombios errors.

Quoted 3 lines

Do any of the disconnected displayport connectors get misdetected as connected during normal operation or does it only happen during suspend/resume?

So far I've only seen them during suspend/resume.

#6 lore raw
Quoted 40 lines · Seth Forshee
Quoted 39 lines · Alex Deucher
Quoted 36 lines · Seth Forshee
Quoted 16 lines · Seth Forshee
Quoted 16 lines
Quoted 15 lines

 2. Occasional long delays when suspending. When this happens I see messages like following in dmesg:

     [drm:atom_op_jump] *ERROR* atombios stuck in loop for more than 5secs aborting [drm:atom_execute_table_locked] *ERROR* atombios stuck executing D44E (len 62, WS 0, PS 0) @ 0xD46A

   Sometimes one of suspend or resume hangs completely, but I can't tell which and am not sure whether or not it's related. I'm also testing a Mac Mini with the exact same card which does not seem to suffer from this issue.

   I ran a bisections that identified f8d0edd (drm/radeon/kms: improve DP detect logic) as introducing problems with suspend, and reverting this patch on top of 3.2.1 does seem to eliminate both issues.

Quoted 6 lines
Quoted 2 lines

That patch doesn't really affect the modesetting paths directly; it looks like a red herring to me.

Perhaps. I just started a run of 200 s3 cycles with the patch reverted to see if I can reproduce the issues. I can usually trigger the problem with 15 or fewer s3 cycles.

The machine completed 200 s3 cycles with the patch reverted without long delays, hangs, or any atombios error messages. Without reverting it doesn't get through many at all before experiencing the errors and long delays or hangs.

I added a printout of the connector status resulting from the logic that was changed by f8d0edd. With the logic prior to this commit it consistently returns the status as disconnected, which is correct as I have nothing connected. But with the improved logic the status is sometimes reported as connected, and these coincide with the atombios errors.

Quoted 5 lines
Quoted 3 lines

Do any of the disconnected displayport connectors get misdetected as connected during normal operation or does it only happen during suspend/resume?

So far I've only seen them during suspend/resume.

Can you track down who is calling the connector->detect() callbacks during suspend and resume?

Alex

#7 lore raw
Quoted 42 lines · Alex Deucher
Quoted 40 lines · Seth Forshee
Quoted 39 lines · Alex Deucher
Quoted 36 lines · Seth Forshee
Quoted 16 lines · Seth Forshee
Quoted 16 lines
Quoted 15 lines

 2. Occasional long delays when suspending. When this happens I see messages like following in dmesg:

     [drm:atom_op_jump] *ERROR* atombios stuck in loop for more than 5secs aborting [drm:atom_execute_table_locked] *ERROR* atombios stuck executing D44E (len 62, WS 0, PS 0) @ 0xD46A

   Sometimes one of suspend or resume hangs completely, but I can't tell which and am not sure whether or not it's related. I'm also testing a Mac Mini with the exact same card which does not seem to suffer from this issue.

   I ran a bisections that identified f8d0edd (drm/radeon/kms: improve DP detect logic) as introducing problems with suspend, and reverting this patch on top of 3.2.1 does seem to eliminate both issues.

Quoted 6 lines
Quoted 2 lines

That patch doesn't really affect the modesetting paths directly; it looks like a red herring to me.

Perhaps. I just started a run of 200 s3 cycles with the patch reverted to see if I can reproduce the issues. I can usually trigger the problem with 15 or fewer s3 cycles.

The machine completed 200 s3 cycles with the patch reverted without long delays, hangs, or any atombios error messages. Without reverting it doesn't get through many at all before experiencing the errors and long delays or hangs.

I added a printout of the connector status resulting from the logic that was changed by f8d0edd. With the logic prior to this commit it consistently returns the status as disconnected, which is correct as I have nothing connected. But with the improved logic the status is sometimes reported as connected, and these coincide with the atombios errors.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

Do any of the disconnected displayport connectors get misdetected as connected during normal operation or does it only happen during suspend/resume?

So far I've only seen them during suspend/resume.

Can you track down who is calling the connector->detect() callbacks during suspend and resume?

I got two different stack traces, see below.

And to slightly amend my statement above, I'm only seeing the wrong status returned on the suspend side of things. The status during resume always seems to be correct.

Pid: 49, comm: kworker/0:2 Tainted: G C O 3.2.0-10-generic #17-Ubuntu

 Call Trace:
  [<ffffffffa03f5e1e>] radeon_dp_detect+0x1de/0x230 [radeon]
  [<ffffffffa014ac6d>] output_poll_execute+0xbd/0x1a0 [drm_kms_helper]
  [<ffffffffa014abb0>] ? drm_helper_mode_fill_fb_struct+0x30/0x30 [drm_kms_helper]
  [<ffffffff81083dca>] process_one_work+0x11a/0x480
  [<ffffffff81084b74>] worker_thread+0x164/0x370
  [<ffffffff81084a10>] ? manage_workers.isra.30+0x130/0x130
  [<ffffffff810893cc>] kthread+0x8c/0xa0
  [<ffffffff81660674>] kernel_thread_helper+0x4/0x10
  [<ffffffff81089340>] ? flush_kthread_worker+0xa0/0xa0
  [<ffffffff81660670>] ? gs_change+0x13/0x13

Pid: 49, comm: kworker/0:2 Tainted: G C O 3.2.0-10-generic #17-Ubuntu

 Call Trace:
  [<ffffffffa03f5e1e>] radeon_dp_detect+0x1de/0x230 [radeon]
  [<ffffffffa014b6b3>] drm_helper_probe_single_connector_modes+0x2c3/0x380 [drm_kms_helper]
  [<ffffffffa0149d42>] drm_fb_helper_hotplug_event+0xf2/0x150 [drm_kms_helper]
  [<ffffffffa03fe8d5>] radeon_fb_output_poll_changed+0x15/0x20 [radeon]
  [<ffffffffa03f7b65>] radeon_output_poll_changed+0x15/0x20 [radeon]
  [<ffffffffa014ad40>] output_poll_execute+0x190/0x1a0 [drm_kms_helper]
  [<ffffffffa014abb0>] ? drm_helper_mode_fill_fb_struct+0x30/0x30 [drm_kms_helper]
  [<ffffffff81083dca>] process_one_work+0x11a/0x480
  [<ffffffff81084b74>] worker_thread+0x164/0x370
  [<ffffffff81084a10>] ? manage_workers.isra.30+0x130/0x130
  [<ffffffff810893cc>] kthread+0x8c/0xa0
  [<ffffffff81660674>] kernel_thread_helper+0x4/0x10
  [<ffffffff81089340>] ? flush_kthread_worker+0xa0/0xa0
  [<ffffffff81660670>] ? gs_change+0x13/0x13
#8 lore raw
Quoted 39 lines · Seth Forshee
Quoted 2 lines

Can you track down who is calling the connector->detect() callbacks during suspend and resume?

I got two different stack traces, see below.

And to slightly amend my statement above, I'm only seeing the wrong status returned on the suspend side of things. The status during resume always seems to be correct.

Pid: 49, comm: kworker/0:2 Tainted: G C O 3.2.0-10-generic #17-Ubuntu

 Call Trace:
  [<ffffffffa03f5e1e>] radeon_dp_detect+0x1de/0x230 [radeon]
  [<ffffffffa014ac6d>] output_poll_execute+0xbd/0x1a0 [drm_kms_helper]
  [<ffffffffa014abb0>] ? drm_helper_mode_fill_fb_struct+0x30/0x30 [drm_kms_helper]
  [<ffffffff81083dca>] process_one_work+0x11a/0x480
  [<ffffffff81084b74>] worker_thread+0x164/0x370
  [<ffffffff81084a10>] ? manage_workers.isra.30+0x130/0x130
  [<ffffffff810893cc>] kthread+0x8c/0xa0
  [<ffffffff81660674>] kernel_thread_helper+0x4/0x10
  [<ffffffff81089340>] ? flush_kthread_worker+0xa0/0xa0
  [<ffffffff81660670>] ? gs_change+0x13/0x13

Pid: 49, comm: kworker/0:2 Tainted: G C O 3.2.0-10-generic #17-Ubuntu

 Call Trace:
  [<ffffffffa03f5e1e>] radeon_dp_detect+0x1de/0x230 [radeon]
  [<ffffffffa014b6b3>] drm_helper_probe_single_connector_modes+0x2c3/0x380 [drm_kms_helper]
  [<ffffffffa0149d42>] drm_fb_helper_hotplug_event+0xf2/0x150 [drm_kms_helper]
  [<ffffffffa03fe8d5>] radeon_fb_output_poll_changed+0x15/0x20 [radeon]
  [<ffffffffa03f7b65>] radeon_output_poll_changed+0x15/0x20 [radeon]
  [<ffffffffa014ad40>] output_poll_execute+0x190/0x1a0 [drm_kms_helper]
  [<ffffffffa014abb0>] ? drm_helper_mode_fill_fb_struct+0x30/0x30 [drm_kms_helper]
  [<ffffffff81083dca>] process_one_work+0x11a/0x480
  [<ffffffff81084b74>] worker_thread+0x164/0x370
  [<ffffffff81084a10>] ? manage_workers.isra.30+0x130/0x130
  [<ffffffff810893cc>] kthread+0x8c/0xa0
  [<ffffffff81660674>] kernel_thread_helper+0x4/0x10
  [<ffffffff81089340>] ? flush_kthread_worker+0xa0/0xa0
  [<ffffffff81660670>] ? gs_change+0x13/0x13
Quoted 1 line

From these traces it looks like all that really needs to happen is to

disable the output polling during suspend. The following patch seems to get rid of the problems I'm seeing. Does this look okay to you?

Quoted 1 line

From 0c01950f248c541198b7560793cf57c59b2c11f8 Mon Sep 17 00:00:00 2001

From: Seth Forshee <[email protected]>
Date: Tue, 31 Jan 2012 15:37:02 -0600
Subject: [PATCH 1/1] drm/radeon/kms: disable output polling when suspended

Polling the outputs when the device is suspended can result in erroneous status updates. Disable output polling during suspend to prevent this from happening.

---
 drivers/gpu/drm/radeon/radeon_device.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)
drivers/gpu/drm/radeon/radeon_device.c +4 -0
index cec51a5..49f7cb7 100644
@@ -883,6 +883,8 @@ int radeon_suspend_kms(struct drm_device *dev, pm_message_t state)
 	if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
 		return 0;
 
+	drm_kms_helper_poll_disable(dev);
+
 	/* turn off display hw */
 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 		drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF);
@@ -972,6 +974,8 @@ int radeon_resume_kms(struct drm_device *dev)
 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 		drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON);
 	}
+
+	drm_kms_helper_poll_enable(dev);
 	return 0;
 }
 
Signature · 1 line
1.7.8.3
Reviewed-by
#9 lore raw
Quoted 40 lines · Seth Forshee
Quoted 23 lines · Seth Forshee
Quoted 2 lines

Can you track down who is calling the connector->detect() callbacks during suspend and resume?

I got two different stack traces, see below.

And to slightly amend my statement above, I'm only seeing the wrong status returned on the suspend side of things. The status during resume always seems to be correct.

 Pid: 49, comm: kworker/0:2 Tainted: G         C O 3.2.0-10-generic #17-Ubuntu

 Call Trace:
  [<ffffffffa03f5e1e>] radeon_dp_detect+0x1de/0x230 [radeon]
  [<ffffffffa014ac6d>] output_poll_execute+0xbd/0x1a0 [drm_kms_helper]
  [<ffffffffa014abb0>] ? drm_helper_mode_fill_fb_struct+0x30/0x30 [drm_kms_helper]
  [<ffffffff81083dca>] process_one_work+0x11a/0x480
  [<ffffffff81084b74>] worker_thread+0x164/0x370
  [<ffffffff81084a10>] ? manage_workers.isra.30+0x130/0x130
  [<ffffffff810893cc>] kthread+0x8c/0xa0
  [<ffffffff81660674>] kernel_thread_helper+0x4/0x10
  [<ffffffff81089340>] ? flush_kthread_worker+0xa0/0xa0
  [<ffffffff81660670>] ? gs_change+0x13/0x13

 Pid: 49, comm: kworker/0:2 Tainted: G         C O 3.2.0-10-generic #17-Ubuntu

Quoted 15 lines
 Call Trace:
  [<ffffffffa03f5e1e>] radeon_dp_detect+0x1de/0x230 [radeon]
  [<ffffffffa014b6b3>] drm_helper_probe_single_connector_modes+0x2c3/0x380 [drm_kms_helper]
  [<ffffffffa0149d42>] drm_fb_helper_hotplug_event+0xf2/0x150 [drm_kms_helper]
  [<ffffffffa03fe8d5>] radeon_fb_output_poll_changed+0x15/0x20 [radeon]
  [<ffffffffa03f7b65>] radeon_output_poll_changed+0x15/0x20 [radeon]
  [<ffffffffa014ad40>] output_poll_execute+0x190/0x1a0 [drm_kms_helper]
  [<ffffffffa014abb0>] ? drm_helper_mode_fill_fb_struct+0x30/0x30 [drm_kms_helper]
  [<ffffffff81083dca>] process_one_work+0x11a/0x480
  [<ffffffff81084b74>] worker_thread+0x164/0x370
  [<ffffffff81084a10>] ? manage_workers.isra.30+0x130/0x130
  [<ffffffff810893cc>] kthread+0x8c/0xa0
  [<ffffffff81660674>] kernel_thread_helper+0x4/0x10
  [<ffffffff81089340>] ? flush_kthread_worker+0xa0/0xa0
  [<ffffffff81660670>] ? gs_change+0x13/0x13
Quoted 4 lines

From these traces it looks like all that really needs to happen is to disable the output polling during suspend. The following patch seems to get rid of the problems I'm seeing. Does this look okay to you?

Looks great. Thanks for tracking this down. We should probably cc stable as well when this gets committed.

Quoted 40 lines

From 0c01950f248c541198b7560793cf57c59b2c11f8 Mon Sep 17 00:00:00 2001
From: Seth Forshee <[email protected]>
Date: Tue, 31 Jan 2012 15:37:02 -0600
Subject: [PATCH 1/1] drm/radeon/kms: disable output polling when suspended

Polling the outputs when the device is suspended can result in erroneous status updates. Disable output polling during suspend to prevent this from happening.

---
 drivers/gpu/drm/radeon/radeon_device.c |    4 ++++

 1 files changed, 4 insertions(+), 0 deletions(-)
drivers/gpu/drm/radeon/radeon_device.c +4 -1
index cec51a5..49f7cb7 100644
@@ -883,6 +883,8 @@ int radeon_suspend_kms(struct drm_device *dev, pm_message_t state)
        if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
                return 0;
 
+       drm_kms_helper_poll_disable(dev);
+
        /* turn off display hw */
        list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
                drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF);
@@ -972,6 +974,8 @@ int radeon_resume_kms(struct drm_device *dev)
        list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
                drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON);
        }
+
+       drm_kms_helper_poll_enable(dev);
        return 0;
  }
 
--

1.7.8.3

#10 lore raw
In reply to Seth Forshee
Quoted 15 lines · Seth Forshee
Quoted 14 lines · Alex Deucher
Quoted 11 lines · Seth Forshee

I'm seeing several issues related to the radeon driver on a MacBook Pro 8,2 with the following graphics card:

 ATI Technologies Inc Whistler [AMD Radeon HD 6600M Series] [1002:6741]

All problems were observed when using kernel version 3.2.1. None are seen when using fglrx.

 1. Excessive power draw. When using the radeon driver ACPI reports a power draw of about 30W on an idle desktop. Using fglrx brings this number down to 15W.

Quoted 9 lines
Quoted 4 lines

The power saving features of the open source driver are not yet as good as the closed source driver. Please see the power management section of this page (http://wiki.x.org/wiki/RadeonFeature) for more info on the options currently available.

The dynpm option makes a small difference, saving about 2W. I did notice an ocassional flash on the screen with this option, and the same flash each time I changed the power options.

Btw how do you measure the power draw?

Quoted 3 lines

The only thing that improves things significatnly is the "low" profile method, which gets it down to about 18.5W.

I have older radeon (Mobility HD3650) and only the "low" profile makes it usable.. the "default" profile will overheat the laptop and crash it, or then Linux does automatic emergency thermal shutdown to protect the laptop.

If I boot some Linux livecd the laptop will basicly overheat and shutdown by itself before the desktop is loaded from the livecd.. :(

-- Pasi

#11 lore raw
Quoted 21 lines · Pasi Kärkkäinen
Quoted 20 lines · Seth Forshee
Quoted 18 lines · Alex Deucher
Quoted 11 lines · Seth Forshee

I'm seeing several issues related to the radeon driver on a MacBook Pro 8,2 with the following graphics card:

 ATI Technologies Inc Whistler [AMD Radeon HD 6600M Series] [1002:6741]

All problems were observed when using kernel version 3.2.1. None are seen when using fglrx.

 1. Excessive power draw. When using the radeon driver ACPI reports a power draw of about 30W on an idle desktop. Using fglrx brings this number down to 15W.

The power saving features of the open source driver are not yet as good as the closed source driver. Please see the power management section of this page (http://wiki.x.org/wiki/RadeonFeature) for more info on the options currently available.

Quoted 6 lines
Quoted 4 lines

The dynpm option makes a small difference, saving about 2W. I did notice an ocassional flash on the screen with this option, and the same flash each time I changed the power options.

Btw how do you measure the power draw?

You can get the instantaneous rate from the data under /proc/acpi/battery, but I use a tool called powerstat [1], written by my colleague Colin King. The advantage of powerstat is that it samples the ACPI data over a period of time and reports the average and standard deviation. That way I have a better idea of how much power is really being drawn and the quality of the value reported.

[1] http://kernel.ubuntu.com/git?p=cking/powerstat.git