Excerpts from Christophe Leroy's message of January 3, 2021 3:56 am:

Nicholas Piggin [off-list ref] a écrit :

quoted

The page fault handling still has some complex logic particularly around
hash table handling, in asm. Implement this in C instead.

Hi,

I'm afk at the moment and unable to look at this in details before one  
week but this looks pretty complexe, especially the churn around  
___do_page_fault
Do we really need 3 layers of do_page_fault() ?

Actually it doesn't, patch 10 wants it. I can move it to there at least
which should make this a bit less churn.

It's convenient because lots of return paths in __do_page_fault, but I 
could try convert that to a `goto out` style.

I think it would likely be more straight forward to just move  
handle_page_fault() to C.

The hash fault stuff makes things work better this way. Perhaps if I can 
get to the bottom of the context tracking in the hash fault (I think we
perhaps should avoid it), then it could go back to a common code path.

There also seems to be some unrelated changes, like the (msr & MSR_PR)  
changed to user_mode(regs).

That's part of making it callable from asm and the radix vs hash
prototypes the same so there are no added complexity in the asm:

quoted

@@ -1439,13 +1440,17 @@ EXC_COMMON_BEGIN(data_access_common)
 	GEN_COMMON data_access
 	ld	r4,_DAR(r1)
 	ld	r5,_DSISR(r1)
+	addi	r3,r1,STACK_FRAME_OVERHEAD
 BEGIN_MMU_FTR_SECTION
-	ld	r6,_MSR(r1)
-	li	r3,0x300
-	b	do_hash_page		/* Try to handle as hpte fault */
+	bl	do_hash_fault
 MMU_FTR_SECTION_ELSE
-	b	handle_page_fault
+	bl	do_page_fault
 ALT_MMU_FTR_SECTION_END_IFCLR(MMU_FTR_TYPE_RADIX)

I'll see if anything can be done to move some such changes ahead.

Thanks,
Nick