Re: CVS commit: src/sys/dev/rasops

[Valery Ushakov <uwe%stderr.spb.ru@localhost>]

On Sun, Jul 28, 2019 at 02:18:14 -0400, Michael wrote:

> On Sat, 27 Jul 2019 21:35:04 +0200
> Joerg Sonnenberger <joerg%bec.de@localhost> wrote:
> 
> > On Fri, Jul 26, 2019 at 07:54:15AM -0300, Jared McNeill wrote:
> > > On Fri, 26 Jul 2019, Rin Okuyama wrote:
> > >   
> > > > Also, convert loop of byte-wise copy into memset.  
> > > 
> > > I am a bit concerned about this change because IIRC there are platforms
> > > where memset etc. cannot be used on device memory.  
> > 
> > memset can access memory twice or alignment depending on the platform.
> > The question is whether we care about that for the frame buffers though.
> 
> One hazard I can think of is if there's endian-twiddling involved. Some
> graphics chip / host bridge combinations will only convert accesses up
> to 32bit properly, resulting in 64bit accesses having their upper/lower
> 32bit parts swapped on delivery. That should only bite with memcpy()
> though.

I remember running into something like this with igsfb(4) on krups.

-uwe