Re: [PATCH 3/3] gk20a: use NOUVEAU_BO_GART as VRAM domain

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 10/30/2014 12:29 AM, Ilia Mirkin wrote:
On Mon, Oct 27, 2014 at 6:34 AM, Alexandre Courbot <acourbot@xxxxxxxxxx> wrote:
GK20A does not have dedicated VRAM, therefore allocating in VRAM can be
sub-optimal and sometimes even harmful. Set its VRAM domain to
NOUVEAU_BO_GART so all objects are allocated in system memory.

Signed-off-by: Alexandre Courbot <acourbot@xxxxxxxxxx>
---
  src/gallium/drivers/nouveau/nvc0/nvc0_screen.c | 10 ++++++++++
  1 file changed, 10 insertions(+)

diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
index ac5823e4a8d5..ad143cd9a140 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
@@ -620,6 +620,16 @@ nvc0_screen_create(struct nouveau_device *dev)
        return NULL;
     pscreen = &screen->base.base;

+   /* Recognize chipsets with no VRAM */
+   switch (dev->chipset) {
+   /* GK20A */
+   case 0xea:
+      screen->base.vram_domain = NOUVEAU_BO_GART;

I think you also want to set vidmem_bindings = 0... although
potentially after the |= that's done below. Although I guess that
constbuf + command args buf need to be |='d into the sysmem_bindings
for this to work out well. That said, we don't really handle explicit
migration well right now, and those PIPE_BIND_* are *incredibly*
misleading and don't actually necessarily reflect the current usage.
[I have some patches to improve the situation, but you don't really
have to worry about that.]

In the light of that it could be that the vram_domain member I am introducing is completely useless - if we set NV_VRAM_DOMAIN to be the following:

#define NV_VRAM_DOMAIN(screen) ((screen)->vidmem_bindings == 0 ? NOUVEAU_BO_GART : NOUVEAU_BO_VRAM)

then I suspect we can just live without it. I tested quickly and it seems to work. Ilia, do you agree? Or could we imagine having GPUs with VRAM for which none of the PIPE_BIND_* targets should reside in VRAM?

Also thinking, prior to setting vidmem_bindings to 0, shouldn't we also do a "sysmem_bindings |= vidmem_bindings" to make sure all the set bindings are tracked somewhere?
--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html




[Index of Archives]     [ARM Kernel]     [Linux ARM]     [Linux ARM MSM]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux