Skip to contentSkip to conversation

Fix oops in acer_wmi

PATCH NAK Arjan van de Ven 2 participants 4 messages Aug 23, 2008 – Aug 24, 2008 Char and misc drivers

lore
Share & tools
lore
#0 lore raw

In the current kerneloops.org stats, acer_wmi is pretty high on the list for 2.6.27-rc:
http://www.kerneloops.org/search.php?search=acer_wmi_init

this is due to a null pointer due to a misordered initialization...
when the DMI quirks get run, the "interface" variable might not have been initialized. The patch below 1) prevents the oops and 2) runs the quirks again after all the "interface" initialization has completed.

This oops is new in 2.6.27-rc and thus counts as a regression...

drivers/misc/acer-wmi.c renamed from drivers/misc/acer-wmi.c~ +5 -0
@@ -192,6 +192,9 @@ static struct quirk_entry *quirks;
 
 static void set_quirks(void)
 {
+	if (!interface)
+		return;
+
 	if (quirks->mailled)
 		interface->capability |= ACER_CAP_MAILLED;
 
@@ -1237,6 +1240,8 @@ static int __init acer_wmi_init(void)
 		return -ENODEV;
 	}
 
+	set_quirks();
+
 	if (platform_driver_register(&acer_platform_driver)) {
 		printk(ACER_ERR "Unable to register platform driver.\n");
 		goto error_platform_register;
Acked-by NAK
#1 lore raw
Quoted 4 lines · On Saturday 23 August 2008 22:05:09 Arjan van de Ven wrote:

this is due to a null pointer due to a misordered initialization... when the DMI quirks get run, the "interface" variable might not have been initialized. The patch below 1) prevents the oops and 2) runs the quirks again after all the "interface" initialization has completed.

(Although this patch is missing an SOB from yourself).

Quoted 1 line

This oops is new in 2.6.27-rc and thus counts as a regression...

NAK - I disagree. This has been broken since day one, there is nothing special, as far as I am concerned, about 2.6.27 in this respect.

(If this is showing up that much in kerneloops.org, then I would recommend CC'ing it to stable).

-Carlos

Signature · 3 lines
E-Mail: [email protected]
Web: strangeworlds.co.uk
GPG Key ID: 0x23EE722D
author Acked-by Carlos Corbacho
#2 lore raw

On Sun, 24 Aug 2008 02:19:41 +0100

Linus,
below is a patch for a bug that's showing quite hot on kerneloops.org; please consider merging

Quoted 10 lines · Carlos Corbacho
Quoted 5 lines · On Saturday 23 August 2008 22:05:09 Arjan van de Ven wrote:

this is due to a null pointer due to a misordered initialization... when the DMI quirks get run, the "interface" variable might not have been initialized. The patch below 1) prevents the oops and 2) runs the quirks again after all the "interface" initialization has completed.

(Although this patch is missing an SOB from yourself).

oh whoops; updated below

Quoted 6 lines
Quoted 2 lines

This oops is new in 2.6.27-rc and thus counts as a regression...

NAK - I disagree. This has been broken since day one, there is nothing special, as far as I am concerned, about 2.6.27 in this respect.

interesting; I have no reports of this in kerneloops.org (but if this got introduced in 2.6.26 that's possible; there's not that many 2.6.26 users yet); you can see this yourself at
http://www.kerneloops.org/search.php?search=acer_wmi_init
on the left side it shows how often it happens for which versions/

I'll still call it a tentative regression ;=)

Quoted 4 lines

(If this is showing up that much in kerneloops.org, then I would recommend CC'ing it to stable).

--
From: Arjan van de Ven <[email protected]>
Subject: Fix oops in acer_wmi driver (acer_wmi_init)

The acer_wmi driver does a DMI scan for quirks, and then sets flags into the "interface" datastructure for some cases. However, the quirks happen real early before "interface" is per se initialized from NULL.

The patch below 1) adds a NULL pointer check and 2) (re)runs the quirks at the end, when "interface" has it's final value.

drivers/misc/acer-wmi.c renamed from drivers/misc/acer-wmi.c~ +5 -0
@@ -192,6 +192,9 @@ static struct quirk_entry *quirks;
 
 static void set_quirks(void)
 {
+	if (!interface)
+		return;
+
 	if (quirks->mailled)
 		interface->capability |= ACER_CAP_MAILLED;
 
@@ -1237,6 +1240,8 @@ static int __init acer_wmi_init(void)
 		return -ENODEV;
 	}
 
+	set_quirks();
+
 	if (platform_driver_register(&acer_platform_driver)) {
 		printk(ACER_ERR "Unable to register platform driver.\n");
 		goto error_platform_register;
#3 lore raw
Quoted 7 lines · On Sunday 24 August 2008 05:45:21 Arjan van de Ven wrote:

interesting; I have no reports of this in kerneloops.org (but if this got introduced in 2.6.26 that's possible; there's not that many 2.6.26 users yet); you can see this yourself at
http://www.kerneloops.org/search.php?search=acer_wmi_init
on the left side it shows how often it happens for which versions/

I'll still call it a tentative regression ;=)

My bad; this is just a 2.6.27 regression - I thought the offending patch that had caused this (83097aca8567a0bd593534853b71fe0fa9a75d69) went into 2.6.26 as well, it didn't.

-Carlos

Signature · 3 lines
E-Mail: [email protected]
Web: strangeworlds.co.uk
GPG Key ID: 0x23EE722D