[PATCH v3] ARM: early fixmap support for earlycon

[stefan@agner.ch (Stefan Agner)]

On 2015-07-10 13:25, Russell King - ARM Linux wrote:
> On Sat, Jun 06, 2015 at 01:48:00PM +0100, Russell King - ARM Linux wrote:
>> On Sat, Jun 06, 2015 at 02:31:28PM +0200, Stefan Agner wrote:
>> > @@ -1231,7 +1272,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
>> >
>> >  	early_trap_init(vectors);
>> >
>> > -	for (addr = VMALLOC_START; addr; addr += PMD_SIZE)
>> > +	for (addr = VMALLOC_START; addr < FIXADDR_START; addr += PMD_SIZE)
>> >  		pmd_clear(pmd_off_k(addr));
>>
>> You introduce a bug here - we no logner clear the very top entry of the
>> page tables, which means it could contain anything - and means that the
>> subsequent creation of the L2 table in early_pte_alloc() can fail.
> 
> You trimmed my reply hard, and seemingly ignored this; the v4 patch
> contains exactly the same code here, and therefore continues to leave
> the top of the page table uninitialised.

Maybe I get something wrong here, but the very top PMD is left out on
purpose, since early_fixmap_shutdown takes care of that.

However v3 also did not clear the second to top PMD. I thought I did
take care of that by using this slightly different for loop in v4:
for (addr = VMALLOC_START; addr < (FIXADDR_TOP & PMD_MASK); addr +=
PMD_SIZE)

However, looking now@the code I'm not sure if this is right either. 

I think what is needed is addr < (FIXADDR_TOP & ~PMD_MASK). This will
make sure that only the top PMD is not cleared, which is taken care of
by early fixmap...

--
Stefan