Benjamin Herrenschmidt [off-list ref] wrote on 03/10/2009 10:31:18:

On Sat, 2009-10-03 at 10:05 +0200, Joakim Tjernlund wrote:

quoted

Cannot shake the feeling that it this snip of code that causes it
        lwz     r11, 0(r10)     /* Get the level 1 entry */
        rlwinm. r10, r11,0,0,19 /* Extract page descriptor page
address */
        beq     2f              /* If zero, don't try to find a pte */
Perhaps we can do something better? I still feel that we need to
force a TLB Error as the TLBMiss does not set DSISR so we have no way
of
knowing if it is an load or store.

Can't we manufacture a DSISR and branch to the right function ?

Not if we want know if it is a load or store. There is no info to manufacture
a DSISR from. The best we can do here is try getting the RPN physical page
number correct. Perhaps something like this will do:
	/* Copy 20 msb from MD_EPN to r20 to get the correct page
	 * number. Do not rely on DAR since the dcxx instructions fails
	 * to update DAR when they cause a DTLB Miss
	 */
	mfspr	r21, MD_EPN
	li	r20, 0x0
	rlwimi	r20, r21, 0, 0, 19
Then go back and set the RPN accordingly.

The 8xx is different as as it will force a TLB error every time
it needs to deal with a page fault.
I suspect adding
if (!ret)
   _tlbil_va(address);
in do_page_fault() will do the trick too.

So yes, there is a missing _tlbil_va() missing for 8xx somewhere
but there is something more too.
Maybe your new filter functions and my
 powerpc, 8xx: DTLB Error must check for more errors.
will do the trick?

 Jocke