Re: [tip:x86/alternatives] x86, alternatives: Use 16-bit numbers for cpufeature index

["H. Peter Anvin" <hpa@zytor.com>]

On 06/29/2010 02:15 AM, Ingo Molnar wrote:
> 
> Hm, this patch is causing trouble in -tip testing again - it's triggering a 
> colorful boot crash:
> 
> [    2.220002] calling  inet_init+0x0/0x23d @ 1
> [    2.223343] NET: Registered protocol family 2
> [    2.226727] IP route cache hash table entries: 32768 (order: 6, 262144 bytes)
> [    2.233492] ------------[ cut here ]------------
> [    2.236671] WARNING: at mm/vmalloc.c:107 vmap_page_range_noflush+0x309/0x3a0()
> [    2.240001] Modules linked in:
> ...
> [    3.090002] Kernel panic - not syncing: Failed to allocate TCP established hash table
> 
> So i've zapped them again. We really need to get to the bottom of this. Config 
> and bootlog attached.
> 
> The crash looks very weird - and it's consistent with possible effects of some 
> sort of code patching failure/mismatch.
> 
> It goes away if i revert these two:
> 
>  a3d2d12: x86, alternatives: correct obsolete use of "u8" in static_cpu_has()
>  5dc71d4: x86, alternatives: Use 16-bit numbers for cpufeature index
> 
> I reproduced the crash twice before testing the revert.
> 

Hi Ingo,

I'm pretty sure that these are related to gcc and/or binutils
differences, so it would be nice to get the .o and .s files of the
failing locations (in this case mm/vmalloc.[so]) *as built on the
failing machines*.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.