On 11.03.2022 10:40, Lucas De Marchi wrote:
On Tue, Mar 08, 2022 at 10:17:42PM +0530, Balasubramani Vivekanandan wrote:
This patch is continuation of the effort to move all pointers in i915, which at any point may be pointing to device memory or system memory, to iosys_map interface. More details about the need of this change is explained in the patch series which initiated this task https://patchwork.freedesktop.org/series/99711/
This patch converts all access to the lrc_desc through iosys_map interfaces.
Cc: Lucas De Marchi lucas.demarchi@intel.com Cc: John Harrison John.C.Harrison@Intel.com Cc: Matthew Brost matthew.brost@intel.com Cc: Umesh Nerlige Ramappa umesh.nerlige.ramappa@intel.com Signed-off-by: Balasubramani Vivekanandan balasubramani.vivekanandan@intel.com
...
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc.h @@ -2245,13 +2256,13 @@ static void prepare_context_registration_info(struct intel_context *ce) GEM_BUG_ON(i915_gem_object_is_lmem(guc->ct.vma->obj) != i915_gem_object_is_lmem(ce->ring->vma->obj));
- desc = __get_lrc_desc(guc, ctx_id);
- desc->engine_class = engine_class_to_guc_class(engine->class);
- desc->engine_submit_mask = engine->logical_mask;
- desc->hw_context_desc = ce->lrc.lrca;
- desc->priority = ce->guc_state.prio;
- desc->context_flags = CONTEXT_REGISTRATION_FLAG_KMD;
- guc_context_policy_init(engine, desc);
- memset(&desc, 0, sizeof(desc));
previously we would re-use whatever was left in guc->lrc_desc_pool_vaddr. Here we are changing it to always zero everything and set the fields we are interested in.
As I'm not too familiar with this part and I see us traversing child guc_process_desc which may point to the same id, it doesn't _feel_ safe. Did you check if this is not zero'ing what it shouldn't?
Matt Brost / John / Daniele, could you clarify?
thanks Lucas De Marchi
I verified that struct guc_lrc_desc is not updated anywhere else in the driver other than in prepare_context_registration_info. So I went ahead with clearing it before updating the fields. But I will still wait for comments from Matt Brost/ John / Daniele for their confirmation.
Thanks Bala