Re: [PATCH] powerpc/mm: tell if a bad page fault on data is read or write.

[Christophe Leroy <christophe.leroy@c-s.fr>]

Le 29/08/2019 à 14:14, Michael Ellerman a écrit :
> Christophe Leroy <christophe.leroy@c-s.fr> writes:
>> DSISR has a bit to tell if the fault is due to a read or a write.
> 
> Except some CPUs don't have a DSISR?
> 
> Which is why we have page_fault_is_write() that's used in
> __do_page_fault().


And that's why I'm also using page_fault_is_write() in my patch.

> 
> Or is that old cruft?
> 
> I see eg. in head_40x.S we pass r5=0 for error code, and we don't set
> regs->dsisr anywhere AFAICS. So it might just contain some junk.


We pass r5=0 in ISI but r5=SPRN_ESR in DSI.
And r5 is also saved into _ESR(r11)

And in asm-offset.c, we have:

	STACK_PT_REGS_OFFSET(_ESR, dsisr);

So regs->dsisr has the expected content.

Christophe


> 
> cheers
> 
>> diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
>> index 8432c281de92..b5047f9b5dec 100644
>> --- a/arch/powerpc/mm/fault.c
>> +++ b/arch/powerpc/mm/fault.c
>> @@ -645,6 +645,7 @@ NOKPROBE_SYMBOL(do_page_fault);
>>   void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
>>   {
>>   	const struct exception_table_entry *entry;
>> +	int is_write = page_fault_is_write(regs->dsisr);
>>   
>>   	/* Are we prepared to handle this fault?  */
>>   	if ((entry = search_exception_tables(regs->nip)) != NULL) {
>> @@ -658,9 +659,10 @@ void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
>>   	case 0x300:
>>   	case 0x380:
>>   	case 0xe00:
>> -		pr_alert("BUG: %s at 0x%08lx\n",
>> +		pr_alert("BUG: %s on %s at 0x%08lx\n",
>>   			 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" :
>> -			 "Unable to handle kernel data access", regs->dar);
>> +			 "Unable to handle kernel data access",
>> +			 is_write ? "write" : "read", regs->dar);
> 
>>   		break;
>>   	case 0x400:
>>   	case 0x480:
>> -- 
>> 2.13.3