svn commit: r268624 - head/sys/dev/vt/hw/efifb

[Nathan Whitehorn]

On 07/14/14 11:47, Konstantin Belousov wrote:
> On Mon, Jul 14, 2014 at 11:43:03AM -0700, Nathan Whitehorn wrote:
>> On 07/14/14 10:53, Konstantin Belousov wrote:
>>> On Mon, Jul 14, 2014 at 05:42:22PM +0000, Nathan Whitehorn wrote:
>>>> +	info->fb_vbase = (intptr_t)pmap_mapdev_attr(info->fb_pbase,
>>>> +	    info->fb_size, VM_MEMATTR_WRITE_COMBINING);
>>>> +}
>>>> +
>>> Could you use pmap_change_attr() ? This would save some KVA.
>> Does that work with the direct map? I'm pretty hesitant to muck with the
>> direct map region this way...
> Yes, it works on direct map.
>
> Note that Intel explicitely states that the behaviour is undefined
> if the same physical page is mapped with differrent caching attributes.
> At least some revisions of Pentium IV hang in this situation.
>
> We keep direct map mapping attributes consistent with other mappings.

Ah, interesting. I'll try to correct it tomorrow. I'm at a conference in 
Chicago at the moment, which means it might take a little longer than usual.
-Nathan