Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed startonX61s laptop

[Hugh Dickins]

On Mon, 19 May 2008, Linus Torvalds wrote:
> On Mon, 19 May 2008, Hugh Dickins wrote:
> > 
> > This comes from an assumption in 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
> > mprotect: prevent alteration of the PAT bits, that PTE_MASK is what it's
> > supposed to be: whereas it's been wrong forever with PAE, staying 32-bit
> > where 64-bit is needed.
> 
> Can we *please* just fix PTE_MASK?

That's very much what I'd prefer too.  Jeremy has patches in Ingo's
tree to do that, which have been tested - though perhaps not in
combination with the PAT pte_modify changes.  I did check that they're
not incompatible in theory, but I sure better try them out later today.

> And can we agree to never EVER use that PAGE_MASK thing (which was only 
> ever meant to work on *addresses*) for any pte operations (including the 
> definition of PTE_MASK)? Because PAGE_MASK is very much the word-size, and 
> in 32-bit PAE, the page table entry is bigger.
> 
> IOE, PTE_MASK should be a "pteval_t". And it should have absolutely 
> *nothing* to do with PAGE_MASK. EVER.

Yes, Jeremy makes it a pteval_t.  (My builds and Ingo's builds succeed,
but I've not worked out how that goes down in assembly: there was an
_AT macro in there before, which you've kept too - Jeremy?)

> IOW, maybe something like this?
> 
> And no, I haven't tested this at all. But it should make PTE_MASK have
>  (a) the right type ("pteval_t", not "long" - the latter is pure and utter 
>      crap)
>  (b) the right value (proper mask, not a sign-extended long - again, the 
>      latter is pure and utter crap)
> 
> but for all I know there might be some broken code that depends on the 
> current incorrect and totally broken #defines, so this needs testing and 
> thinking about.

Yes, I'm highly resistant to taking untested patches here.  The two-liner
I sent last night was about my fifth attempt to get it working, and I did
start off from a small PTE_MASK correction which didn't work at all.  It
looked rather like yours, I guess I missed the  __PHYSICAL_LOW_BITS part.
Jeremy's goes a lot further, he'll know the gotchas better.

> It also causes these warnings on 32-bit PAE:
> 
> 	  AS      arch/x86/kernel/head_32.o
> 	arch/x86/kernel/head_32.S: Assembler messages:
> 	arch/x86/kernel/head_32.S:225: Warning: left operand is a bignum; integer 0 assumed
> 	arch/x86/kernel/head_32.S:609: Warning: left operand is a bignum; integer 0 assumed
> 
> and I do not see why (the end result seems to be identical).
> 
> Ingo, comments?
> 
> Oh, and those #define's should be moved from <asm/page.h> to 
> <asm/pgtable.h>, I think. They have nothing to do with pages (despite the 
> name of "physical_page_mask", and really are meaningful only in the 
> context of some kind of page table entry.

Jeremy still has them in asm/page.h.  Like your subsequent pte bit
cleanups, that can be added later: the important thing is to get X
working again on 32-bit NX systems.

Hugh
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/