From: Russell King <rmk+kernel@xxxxxxxxxxxxxxxx>
Running the cairo trace "firefox-36-20090609" results in the Xorg server
trying to queue up 319M of buffer objects in a single commit (due to the
number of drawing operations that will fit into a 32K command buffer.)
This causes the commit ioctl to return -ENOSPC as the MMU is unable to
map all these buffer objects simultaneously.
There are three workarounds possible:
- Increase the GPU MMU table size
- Decrease the command buffer size
- Implement reconstruction of the command buffer on -ENOSPC return
All three should be implemented, but the first is only applicable to the
kernel.
Signed-off-by: Russell King <rmk+kernel@xxxxxxxxxxxxxxxx>
---
drivers/staging/etnaviv/etnaviv_iommu.c | 12 +++++-------
1 file changed, 5 insertions(+), 7 deletions(-)
diff --git a/drivers/staging/etnaviv/etnaviv_iommu.c b/drivers/staging/etnaviv/etnaviv_iommu.c
index 73ba65205705..9efb7d6092b4 100644
--- a/drivers/staging/etnaviv/etnaviv_iommu.c
+++ b/drivers/staging/etnaviv/etnaviv_iommu.c
@@ -25,7 +25,7 @@
#include "etnaviv_iommu.h"
#include "state_hi.xml.h"
-#define PT_SIZE SZ_512K
+#define PT_SIZE SZ_2M
#define PT_ENTRIES (PT_SIZE / sizeof(u32))
#define GPU_MEM_START 0x80000000
@@ -88,7 +88,7 @@ static void pgtable_write(struct etnaviv_iommu_domain_pgtable *pgtable,
static int __etnaviv_iommu_init(struct etnaviv_iommu_domain *etnaviv_domain)
{
- u32 iova, *p;
+ u32 *p;
int ret, i;
etnaviv_domain->bad_page_cpu = dma_alloc_coherent(etnaviv_domain->dev,
@@ -110,11 +110,9 @@ static int __etnaviv_iommu_init(struct etnaviv_iommu_domain *etnaviv_domain)
return ret;
}
- for (iova = etnaviv_domain->domain.geometry.aperture_start;
- iova < etnaviv_domain->domain.geometry.aperture_end; iova += SZ_4K) {
- pgtable_write(&etnaviv_domain->pgtable, iova,
- etnaviv_domain->bad_page_dma);
- }
+ for (i = 0; i < PT_ENTRIES; i++)
+ etnaviv_domain->pgtable.pgtable[i] =
+ etnaviv_domain->bad_page_dma;
spin_lock_init(&etnaviv_domain->map_lock);
--
2.5.1
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/dri-devel
