Re: [PATCH 1/7] drm/nouveau: use bo->base.size instead of mem->num_pages

Am 13.04.21 um 17:54 schrieb Matthew Auld:

On Tue, 13 Apr 2021 at 14:52, Christian König ckoenig.leichtzumerken@gmail.com wrote:

...

Change a couple of cases where it makes more sense to use the base size instead of the number of pages in the resource.

Signed-off-by: Christian König christian.koenig@amd.com

drivers/gpu/drm/nouveau/nouveau_bo.c | 9 ++++----- drivers/gpu/drm/nouveau/nouveau_fbcon.c | 4 ++-- drivers/gpu/drm/nouveau/nouveau_gem.c | 4 ++-- 3 files changed, 8 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index 2d5d68fc15c2..6dbcbe2fa55f 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -302,7 +302,6 @@ nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 domain, int type = sg ? ttm_bo_type_sg : ttm_bo_type_device; int ret;

•   nvbo->bo.mem.num_pages = size >> PAGE_SHIFT;
  

So this was redundant, since ttm_bo_init_reserved() already did this for us?

No, this is redundant since we now use base.size in the functions called by nouveau_bo_placement_set().

Since base.size is already initialized here we should be able to safely drop this.

...

     nouveau_bo_placement_set(nvbo, domain, 0);
     INIT_LIST_HEAD(&nvbo->io_reserve_lru);

@@ -364,12 +363,12 @@ static void set_placement_range(struct nouveau_bo *nvbo, uint32_t domain) { struct nouveau_drm *drm = nouveau_bdev(nvbo->bo.bdev);

•   u32 vram_pages = drm->client.device.info.ram_size >> PAGE_SHIFT;
  

•   u64 vram_size = drm->client.device.info.ram_size;
     unsigned i, fpfn, lpfn;
  
     if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CELSIUS &&
         nvbo->mode && (domain & NOUVEAU_GEM_DOMAIN_VRAM) &&
  

•       nvbo->bo.mem.num_pages < vram_pages / 4) {
  

•       nvbo->bo.base.size < vram_size / 4) {
             /*
              * Make sure that the color and depth buffers are handled
              * by independent memory controller units. Up to a 9x
  

@@ -377,11 +376,11 @@ set_placement_range(struct nouveau_bo *nvbo, uint32_t domain) * at the same time. */ if (nvbo->zeta) {

•                   fpfn = vram_pages / 2;
  

•                   fpfn = (vram_size / 2) >> PAGE_SHIFT;
                     lpfn = ~0;
             } else {
                     fpfn = 0;
  

•                   lpfn = vram_pages / 2;
  

•                   lpfn = (vram_size / 2) >> PAGE_SHIFT;
             }
             for (i = 0; i < nvbo->placement.num_placement; ++i) {
                     nvbo->placements[i].fpfn = fpfn;
  

diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c index 4fc0fa696461..93ac78bda750 100644 --- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c +++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c @@ -379,10 +379,10 @@ nouveau_fbcon_create(struct drm_fb_helper *helper, FBINFO_HWACCEL_IMAGEBLIT; info->fbops = &nouveau_fbcon_sw_ops; info->fix.smem_start = nvbo->bo.mem.bus.offset;

•   info->fix.smem_len = nvbo->bo.mem.num_pages << PAGE_SHIFT;
  

•   info->fix.smem_len = nvbo->bo.base.size;
  

Is byte level granularity a thing in general?

Yes, on AMD GPUs we have resources which must be managed in bytes or even arbitrary units like blocks of registers for example.

I would have assumed that base.size is always aligned to PAGE_SIZE or whatever? At least in ttm_bo_init_reserved() we first align the size and then calculate the num_pages, so not sure. Hopefully this is not a concern, and should be equivalent.

Yes currently the size should always be aligned to a page, but that is exactly what I want to get away from.

Reviewed-by: Matthew Auld matthew.auld@intel.com

Thanks, Christian.