RE: ACPI Linux: acpi_get_firmware fix

RE: ACPI Linux: acpi_get_firmware fix

[Brown, Len]

 

>-----Original Message-----
>From: Yann Droneaud [mailto:ydroneaud-4qZELD6FgxhWk0Htik3J/w@public.gmane.org] 
>Sent: Tuesday, August 30, 2005 10:54 AM
>To: Brown, Len
>Cc: Andrew Morton; Moore, Robert
>Subject: Re: ACPI Linux: acpi_get_firmware fix
>
>On Friday 12 August 2005 06:15, Brown, Len wrote:
>> >Le mercredi 13 juillet 2005 à 16:27 +0200, Yann Droneaud a écrit :
>> >> Hi,
>> >>
>> >> In May, I sent a patch to add a check in acpi_get_firmware_table()
>> >>
>> >> "[PATCH 1/2] IPMI and acpi=off|ht : 
>acpi-get-firmware-failure.patch"
>> >> (Message-ID: <m23bsmzw35.fsf-UvLOT2mcgw93qojSD/UXlMfpnLc0EkqF@public.gmane.org>)
>> >>
>> >> Someone tell me that this change couldn't be merged asis 
>because the
>> >> code touched was managed outside of the Linux tree.
>> >>
>> >> Today, I see in 2.6.13-rc3's changelog that a new ACPI CA 
>was merged.
>> >>
>> >> I look at the code to see if the something changed, but 
>nothing was
>> >> merged. I'm a bit surprised because I check the original 
>ACPI CA code
>> >> from acpica-unix-20050624.tar.gz and found the check added.
>>
>> Yann,
>> 2.6.13 will ship with ACPICA 20050408.
>>
>> Changes that got into the ACPICA core after that will not hit Linux
>> until 2.6.14.
>>
>> The very latest core (20050729) is available in the -mm tree today.
>> If there is a fix missing from that very latest code, please 
>let me know.
>>
>
>It's okay here, the fix applied on ACPICA is present in the 
>-mm (at least in 
>2.6.13-rc6).

thanks for verifying.

>> If the fix is critical, we can always cherry-pick it and
>> back-port it to 2.6.13.1.
>>
>
>I don't think it's critical, but I let you status on it.
>
>The real problem is located in drivers/char/ipmi/ipmi_si_intf.c, where 
>try_init_acpi() call acpi_get_firmware_table() regardless of 
>the status of 
>ACPI. If the ACPI interpretor is disabled, this trigger a bug in 
>acpi_get_firmware_table() trying to unmap rsdt_info->pointer 
>which hasn't 
>been setup after a failure of acpi_tb_get_table().
>
>I propose a patch to check acpi_disabled value, but I'm really 
>not sure it's 
>correct, since, there's really no other places in drivers 
>where the variable 
>is used, so I'm not sure if it's really exposed to the drivers or not 
>restricted to arch/ and drivers/acpi/ .

acpi_disabled is defined in the base kernel and exported for drivers
to check, so like the asus/sony/ibm etc. platform drivers,
ipmi can check it too.

thanks,
-Len

>I also sent the patch to check rsdt_info->pointer before 
>free'ing it, which 
>fix the Oops when acpi= is set to "off" or "ht". This one is 
>applyed in 
>ACPICA and -mm.
>
>References:
>http://lkml.org/lkml/2005/5/16/251
>http://lkml.org/lkml/2005/5/16/261
>http://lkml.org/lkml/2005/5/16/259
>http://lkml.org/lkml/2005/6/7/122
>
>The patch against ipmi_si_intf.c is attached.
>
>Regards.
>


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: ACPI Linux: acpi_get_firmware fix

[Len Brown]

Applied.

thanks,
-Len

-----------
Index: linux-acpi-2.6/drivers/char/ipmi/ipmi_si_intf.c
===================================================================
--- linux-acpi-2.6.orig/drivers/char/ipmi/ipmi_si_intf.c
+++ linux-acpi-2.6/drivers/char/ipmi/ipmi_si_intf.c
@@ -1484,6 +1484,9 @@ static int try_init_acpi(int intf_num, s
        char             *io_type;
        u8               addr_space;

+       if (acpi_disabled)
+               return -ENODEV;
+
        if (acpi_failure)
                return -ENODEV;





-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: ACPI Linux: acpi_get_firmware fix

[Yann Droneaud]

On Wednesday 31 August 2005 22:37, Len Brown wrote:
> Applied.
>
> thanks,


I see the changelog here:
http://lkml.org/lkml/mbox/2005/9/8/30

The two bugs and patches, I sent are fixed in this tree
(event if I'm not credited for this fix:
 
   Added a check for a null RSDT pointer in
    acpi_get_firmware_table (tbxfroot.c) to prevent a fault
    in this error case.

Perhaps you fixed it too. Doesn"t really matter in fact).

Thank you.

(Sorry for this late answer, i was in holiday for a week and half).

Regards.

-- 
Yann Droneaud


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf