Re: some progress with radeon on C8000

[Thomas Bogendoerfer <tsbogend@alpha.franken.de>]

On Wed, Oct 02, 2019 at 04:37:41PM -0400, John David Anglin wrote:
> On 2019-10-02 10:19 a.m., Thomas Bogendoerfer wrote:
> > +	pa = addr & IOVP_MASK;
> > +	mtsp(0,1);
> > +	asm("lci 0(%%sr1, %1), %0" : "=r" (ci) : "r" (__va(pa)));
> I believe you can remove the mtsp and just use "lci 0(%1), %0" to load the coherence index.  The space
> registers sr4 to sr7 are always 0 in kernel.

ok, good to know.

while reading this I realized what the other hacks were for, which I didn't
include in my first mail. 

diff --git a/drivers/gpu/drm/ttm/ttm_agp_backend.c b/drivers/gpu/drm/ttm/ttm_agp_backend.c
index 028ab6007873..e84c7652eb1b 100644
--- a/drivers/gpu/drm/ttm/ttm_agp_backend.c
+++ b/drivers/gpu/drm/ttm/ttm_agp_backend.c
@@ -66,7 +67,8 @@ static int ttm_agp_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
 		if (!page)
 			page = ttm->dummy_read_page;
 
-		mem->pages[mem->page_count++] = page;
+		mem->pages[(ttm->num_pages - 1) - mem->page_count] = page;
+		mem->page_count++;
 	}
 	agp_be->mem = mem;
 
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index d0459b392e5e..4bb301cab128 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -571,8 +571,14 @@ static int ttm_bo_kmap_ttm(struct ttm_buffer_object *bo,
 		 */
 		prot = ttm_io_prot(mem->placement, PAGE_KERNEL);
 		map->bo_kmap_type = ttm_bo_map_vmap;
+		printk("vmap %p\n", ttm->pages[start_page]);
+#if 0
 		map->virtual = vmap(ttm->pages + start_page, num_pages,
 				    0, prot);
+#else
+		map->virtual = kmap(ttm->pages[start_page]);
+#endif
+		
 	}
 	return (!map->virtual) ? -ENOMEM : 0;
 }

 
This is needed to be able to get the virtual address with __va(pa).

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]