agp/uninorth: Fix lockups with radeon KMS and >1x.
authorMichel Dänzer <daenzer@vmware.com>
Thu, 19 May 2011 14:08:39 +0000 (16:08 +0200)
committerDave Airlie <airlied@gmail.com>
Sun, 22 May 2011 10:23:09 +0000 (20:23 +1000)
This was based on a description by Ben Herrenschmidt:

> I've removed that SBA reset from the normal TLB invalidation path and
> left it only once after turning AGP on.

About six months ago, he said:

> I did it a bit differently, but yeah, you get the idea. I'm doing a
> patch series so don't bother pushing things too hard yet.

But I haven't seen anything from him about this since then, and people are
regularly hitting these lockups, so here we are...

Signed-off-by: Michel Dänzer <daenzer@vmware.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Dave Airlie <airlied@gmail.com>
drivers/char/agp/uninorth-agp.c

index f845a8f718b39af2bd1277e331d50143e24ae9fc..a32c492baf5cc15001ad6882dc25a14d1be86419 100644 (file)
@@ -80,7 +80,7 @@ static void uninorth_tlbflush(struct agp_memory *mem)
                               ctrl | UNI_N_CFG_GART_INVAL);
        pci_write_config_dword(agp_bridge->dev, UNI_N_CFG_GART_CTRL, ctrl);
 
-       if (uninorth_rev <= 0x30) {
+       if (!mem && uninorth_rev <= 0x30) {
                pci_write_config_dword(agp_bridge->dev, UNI_N_CFG_GART_CTRL,
                                       ctrl | UNI_N_CFG_GART_2xRESET);
                pci_write_config_dword(agp_bridge->dev, UNI_N_CFG_GART_CTRL,