Re: Bisecting tip/auto-x86-next?

[Yinghai Lu]

On Sat, Jun 21, 2008 at 3:39 AM, Kevin Winchester
<kjwinchester@xxxxxxxxx> wrote:
> Yinghai Lu wrote:
>>
>> On Sat, Jun 21, 2008 at 3:23 AM, Kevin Winchester
>> <kjwinchester@xxxxxxxxx> wrote:
>>>
>>> On Sat, Jun 21, 2008 at 7:18 AM, Yinghai Lu <yhlu.kernel@xxxxxxxxx>
>>> wrote:
>>>>
>>>> On Sat, Jun 21, 2008 at 3:00 AM, Kevin Winchester
>>>> <kjwinchester@xxxxxxxxx> wrote:
>>>>>
>>>>> Yinghai Lu wrote:
>>>>>>
>>>>>> On Fri, Jun 20, 2008 at 5:00 PM, Kevin Winchester
>>>>>> <kjwinchester@xxxxxxxxx> wrote:
>>>>>>>
>>>>>>> Ingo Molnar wrote:
>>>>>>>>
>>>>>>>> * Kevin Winchester <kjwinchester@xxxxxxxxx> wrote:
>>>>>>>>
>>>>>>>>>> hm, could you send me the config that triggered this?
>>>>>>>>>
>>>>>>>>> I will do so tonight when I am home again.  It is a UP AMD64 box
>>>>>>>>> with a
>>>>>>>>> VIA chipset, if that helps.
>>>>>>>>>
>>>>>>>>>> btw., you can probably ignore this one safely. Also please tell me
>>>>>>>>>> at
>>>>>>>>>> which commit ID you were at when you triggered this warning.
>>>>>>>>>
>>>>>>>>> Good to know - I will get the commit ID tonight as well, although
>>>>>>>>> wouldn't following the same bisection sequence that I did give you
>>>>>>>>> the
>>>>>>>>> same
>>>>>>>>> bisection point?  I guess that would assume that linus/master and
>>>>>>>>> auto-x86-next haven't changed much since last night, which might
>>>>>>>>> not be
>>>>>>>>> correct.
>>>>>>>>
>>>>>>>> yeah, you'd probably not hit that warning with the x86/gart
>>>>>>>> bisection
>>>>>>>> sequence. (Assuming the bug is introduced in that branch - so you
>>>>>>>> should
>>>>>>>> first check whether pure x86/gart kernel triggers the problem too.)
>>>>>>>>
>>>>>>>> If you still have the commit ID around then please send it - if you
>>>>>>>> dont,
>>>>>>>> no problem, it's no big issue. I wanted to check how wide the
>>>>>>>> bisection
>>>>>>>> window is where the warning triggers.
>>>>>>>>
>>>>>>> I'm sorry - I accidentally checkout out x86/gart to test it before
>>>>>>> grabbing
>>>>>>> the commit ID.
>>>>>>>
>>>>>>> I went ahead with the bisection and found:
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> 8c9fd91a0dc503f085169d44f4360be025f75224 is first bad commit
>>>>>>> commit 8c9fd91a0dc503f085169d44f4360be025f75224
>>>>>>> Author: Yinghai Lu <yhlu.kernel@xxxxxxxxx>
>>>>>>> Date:   Sun Apr 13 18:42:31 2008 -0700
>>>>>>>
>>>>>>>  x86: checking aperture size order
>>>>>>>
>>>>>>>  some systems are using 32M for gart and agp when memory is less than
>>>>>>> 4G.
>>>>>>>  Kernel will reject and try to allcate another 64M that is not
>>>>>>> needed,
>>>>>>>  and we will waste 64M of perfectly good RAM.
>>>>>>>
>>>>>>>  this patch adds a workaround by checking aper_base/order between NB
>>>>>>> and
>>>>>>>  agp bridge. If they are the same, and memory size is less than 4G,
>>>>>>> it
>>>>>>>  will allow it.
>>>>>>>
>>>>>>>  Signed-off-by: Yinghai Lu <yhlu.kernel@xxxxxxxxx>
>>>>>>>  Signed-off-by: Ingo Molnar <mingo@xxxxxxx>
>>>>>>>  Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
>>>>>>>
>>>>>>> :040000 040000 848d6e4045a14d01fc0a794d4350d8a84f3ceff6
>>>>>>> 4a10a52b41309060cd5dc1bf0c322f6d43b2477b M      arch
>>>>>>> :040000 040000 aa1cee87b1f5b1b30ed03ce6164ad7f404fef2a3
>>>>>>> f9ce0aaa1f7d4fdc7bdc5a43285495db53a6f531 M      drivers
>>>>>>>
>>>>>>>
>>>>>>> as the first bad commit.  I do not have time to look at the patch
>>>>>>> right
>>>>>>> now,
>>>>>>> but in case anyone else does, I figured I would post it.
>>>>>>>
>>>>>> please send out whole boot logs with "debug" in command line.
>>>>>>
>>>>> "debug" in the command line doesn't seem to have any effect on the
>>>>> printout
>>>>> (is there some config option I need to use with it?), but here it is
>>>>> anyway:
>>>>
>>>> are you using tip/master?
>>>>
>>>> http://people.redhat.com/mingo/sched-devel.git/README
>>>>
>>> That log is from tip/x86/gart, which is where the problem patch was
>>> bisected.  Would you get better debugging info from tip/master (which
>>> includes tip/x86/gart, I believe, and thus would show the problem as
>>> well)?
>>
>> tip/master doesn't work?
>>
>
> I will check tip/master now, but I believe it doesn't work.
>
> I will also correct myself, the log I sent you was from:
>
> 8c9fd91... x86: checking aperture size order
>

please try attached patch...

YH
[PATCH] x86: change back to any_mapped

agp check is using request_mem_region. and could fail if e820 reserved...

Signed-off-by: Yinghai Lu <yhlu.kernel@xxxxxxxxx>

diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c
index e6deed1..600470d 100644
--- a/arch/x86/kernel/aperture_64.c
+++ b/arch/x86/kernel/aperture_64.c
@@ -324,8 +324,8 @@ void __init early_gart_iommu_check(void)
 		fix = 1;
 
 	if (gart_fix_e820 && !fix && aper_enabled) {
-		if (!e820_all_mapped(aper_base, aper_base + aper_size,
-				    E820_RESERVED)) {
+		if (e820_any_mapped(aper_base, aper_base + aper_size,
+				    E820_RAM)) {
 			/* reserve it, so we can reuse it in second kernel */
 			printk(KERN_INFO "update e820 for GART\n");
 			e820_add_region(aper_base, aper_size, E820_RESERVED);