Re: 2.6.38.2 breaks suspend to disk

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

On 04/01/2011 02:37 PM, Yinghai Lu wrote:
>>
>> Err, yes, that's what I meant.
>>
>> Now, why the heck did you introduce a bunch of CONFIG_HIBERNATION #ifdefs?
> 
> mmu_cr4_features: have two usages:
> 1. for 32bit to control cr4 access in head_32.S
> 2. for hibernation resume.
> 
> include/asm/processor.h:extern unsigned long            mmu_cr4_features;
> include/asm/processor.h:        mmu_cr4_features |= mask;
> include/asm/processor.h:        mmu_cr4_features &= ~mask;
> kernel/head_32.S:#define cr4_bits pa(mmu_cr4_features)
> kernel/setup.c: * mmu_cr4_features two purpose:
> kernel/setup.c:unsigned long mmu_cr4_features;
> kernel/setup.c:unsigned long mmu_cr4_features = X86_CR4_PAE;
> kernel/setup.c:         mmu_cr4_features = read_cr4();
> power/hibernate_asm_32.S:       movl    mmu_cr4_features, %ecx
> power/hibernate_asm_32.S:       movl    mmu_cr4_features, %ecx
> power/hibernate_asm_64.S:       movq    mmu_cr4_features(%rip), %rax
> power/hibernate_asm_64.S:       movq    mmu_cr4_features(%rip), %rax
> 
> So we don't need to read back cr4 and save it if CONFIG_HIBERNATION is
> not defined.
> 

And why on Earth is it worth saving a couple of instructions (and
introducing code ugliness and a more complex testing matrix) in the case
when it is not?

	-hpa