[Openchrome-users] [openChrome] #417: Unable to map mmio BAR. Argument list too long (7)

[Luc Verhaegen]

On Tue, Oct 25, 2011 at 01:19:34AM +0200, Xavier Bachelot wrote:
> On 10/24/2011 03:18 PM, Luc Verhaegen wrote:
> > On Mon, Oct 24, 2011 at 12:18:20PM -0000, OpenChrome Trac wrote:
> >> #417: Unable to map mmio BAR. Argument list too long (7)
> >> -----------------------+----------------------------------------------------
> >>   Reporter:  brian     |       Owner:  somebody
> >>       Type:  defect    |      Status:  closed  
> >>   Priority:  critical  |   Milestone:          
> >>  Component:  xdriver   |     Version:  0.2.904 
> >> Resolution:  wontfix   |    Keywords:          
> >>   Blocking:            |   Blockedby:          
> >> -----------------------+----------------------------------------------------
> >> Changes (by brian):
> >>
> >>  * cc: brian@? (added)
> >>
> >>
> >> Comment:
> >>
> >>  Replying to [comment:3 schlobinux]:
> >>  >
> >>  > No, it was not a known issue.
> >>
> >>  Perhaps adding a hint in the Xorg log where the reported error is printed
> >>  that lowering memory might help would be useful to future victims.
> >>
> >>  > This is a limitation of PCI, and it will need TTM support to work
> >>  properly. There's nothing that can be done about that currently.
> >>
> >>  OK.  Fair enough.  But shouldn't that mean leaving this ticket open until
> >>  that work is done and this issue is properly resolved?
> >>
> >>  > You shouldn't notice any performance regression.
> >>
> >>  Great news, thanx!
> >>
> >> -- 
> >> Ticket URL: <http://www.openchrome.org/trac/ticket/417#comment:4>
> >> openChrome <http://www.openchrome.org/>
> >> The openChrome project
> > 
> > Here's an idea.. Why doesn't openchrome try to only map 256MB?
> > 
> > if (size > (256 << 10)) {
> > 	xf86DrvMsg(..., "Only mapping the accessible first 256MB of %dMB 
> > available videoram.\n", size >> 10);
> > 	size = 256 << 10;
> > }
> > 
> > This worked fine for radeonhd late july 2007, where i for the first time 
> > dealt with 1GB cards.
> > 
> > You'd think that users would be happier with a working X, and a small 
> > note stating "we can only use 256MB", over a failing X, and someone 
> > stating: "somewhere, over the rainbow, we might have some infrastructure 
> > implemented which can actually use that much memory".
> > 
> > Luc Verhaegen.
> > 
> This probably could have been written more gently, but after all this
> years, I can't say I'm surprised ;-) Thanks for the suggestion.
> 
> Regards,
> Xavier

And with a message like "Cannot use more than 256 MB of VRAM.\n", you 
open the door to a ton of "WAAH! I cannot use all my vram!!!!!" reports.

Also, the directly mappable ram limitation does not go away with a 
memory manager, this is either a hardware limitation (without the 
resizable bar cap), or broken/missing pci(-e) support for resizable 
bars in X/system infrastructure. With a mm, you do however get to play 
tricks with mappings and translation tables.

Amazingly, the remainder of your patch does seem correct.

Luc Verhaegen.