gpu: nvgpu: Remove last Linux kmem usage

Replace the last of the Linux kmem API usage with nvgpu kmem
calls instead. Several places are left alone - allocating the
struct gk20a in particular.

Also one function was updated in the clk code to take a struct
gk20a as an argument so that it could use nvgpu_kmalloc().

Bug 1799159
Bug 1823380

Change-Id: I84fc3f8e19c63d6265bac6098dc727d93e3ff613
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: http://git-master/r/1331702
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/clk/clk_arb.c

@@ -472,12 +472,12 @@ void nvgpu_clk_arb_cleanup_arbiter(struct gk20a *g)
 		destroy_workqueue(arb->update_work_queue);
 		arb->update_work_queue = NULL;
 
-		kfree(arb->gpc2clk_f_points);
-		kfree(arb->mclk_f_points);
+		nvgpu_kfree(g, arb->gpc2clk_f_points);
+		nvgpu_kfree(g, arb->mclk_f_points);
 
 		for (index = 0; index < 2; index++) {
-			kfree(arb->vf_table_pool[index].gpc2clk_points);
-			kfree(arb->vf_table_pool[index].mclk_points);
+			nvgpu_kfree(g, arb->vf_table_pool[index].gpc2clk_points);
+			nvgpu_kfree(g, arb->vf_table_pool[index].mclk_points);
 		}
 	}
 

drivers/gpu/nvgpu/common/linux/ioctl_channel.c

@@ -712,7 +712,7 @@ static int gk20a_channel_event_id_enable(struct channel_gk20a *ch,
 	return 0;
 
 clean_up_free:
-	kfree(event_id_data);
+	nvgpu_kfree(g, event_id_data);
 clean_up_file:
 	fput(file);
 clean_up:

drivers/gpu/nvgpu/common/linux/ioctl_tsg.c

@@ -161,7 +161,7 @@ static int gk20a_tsg_event_id_enable(struct tsg_gk20a *tsg,
 	return 0;
 
 clean_up_free:
-	kfree(event_id_data);
+	nvgpu_kfree(g, event_id_data);
 clean_up_file:
 	fput(file);
 clean_up:

drivers/gpu/nvgpu/common/mm/lockless_allocator.c

@@ -106,7 +106,7 @@ static void nvgpu_lockless_alloc_destroy(struct nvgpu_allocator *a)
 
 	nvgpu_fini_alloc_debug(a);
 
-	vfree(pa->next);
+	nvgpu_vfree(a->g, pa->next);
 	nvgpu_kfree(nvgpu_alloc_to_gpu(a), pa);
 }
 

drivers/gpu/nvgpu/common/semaphore.c

@@ -74,7 +74,7 @@ void gk20a_semaphore_sea_destroy(struct gk20a *g)
 		return;
 
 	nvgpu_mutex_destroy(&g->sema_sea->sea_lock);
-	kfree(g->sema_sea);
+	nvgpu_kfree(g, g->sema_sea);
 	g->sema_sea = NULL;
 }
 

drivers/gpu/nvgpu/gk20a/ce2_gk20a.c

@@ -436,7 +436,7 @@ u32 gk20a_ce_create_context_with_cb(struct device *dev,
 
 	err = nvgpu_mutex_init(&ce_ctx->gpu_ctx_mutex);
 	if (err) {
-		kfree(ce_ctx);
+		nvgpu_kfree(g, ce_ctx);
 		return ctx_id;
 	}
 

drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c

@@ -85,14 +85,15 @@ static int alloc_session(struct gk20a *g, struct dbg_session_gk20a **_dbg_s)
 	return 0;
 }
 
-static int alloc_profiler(struct dbg_profiler_object_data **_prof)
+static int alloc_profiler(struct gk20a *g,
+			  struct dbg_profiler_object_data **_prof)
 {
 	struct dbg_profiler_object_data *prof;
 	*_prof = NULL;
 
 	gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, "");
 
-	prof = kzalloc(sizeof(*prof), GFP_KERNEL);
+	prof = nvgpu_kzalloc(g, sizeof(*prof));
 	if (!prof)
 		return -ENOMEM;
 
@@ -152,7 +153,7 @@ static int gk20a_dbg_gpu_do_dev_open(struct inode *inode,
 err_destroy_lock:
 	nvgpu_mutex_destroy(&dbg_session->ch_list_lock);
 err_free_session:
-	kfree(dbg_session);
+	nvgpu_kfree(g, dbg_session);
 free_ref:
 	gk20a_put(g);
 	return err;
@@ -435,7 +436,7 @@ int dbg_unbind_single_channel_gk20a(struct dbg_session_gk20a *dbg_s,
 				  release_profiler_reservation(dbg_s, prof_obj);
 			}
 			list_del(&prof_obj->prof_obj_entry);
-			kfree(prof_obj);
+			nvgpu_kfree(g, prof_obj);
 		}
 	}
 
@@ -537,7 +538,7 @@ int gk20a_dbg_gpu_dev_release(struct inode *inode, struct file *filp)
 				g->ops.dbg_session_ops.
 				  release_profiler_reservation(dbg_s, prof_obj);
 			list_del(&prof_obj->prof_obj_entry);
-			kfree(prof_obj);
+			nvgpu_kfree(g, prof_obj);
 		}
 	}
 	nvgpu_mutex_release(&g->dbg_sessions_lock);
@@ -1527,7 +1528,7 @@ static int nvgpu_ioctl_allocate_profiler_object(
 
 	nvgpu_mutex_acquire(&g->dbg_sessions_lock);
 
-	err = alloc_profiler(&prof_obj);
+	err = alloc_profiler(g, &prof_obj);
 	if (err)
 		goto clean_up;
 
@@ -1540,7 +1541,7 @@ static int nvgpu_ioctl_allocate_profiler_object(
 		if (prof_obj->ch == NULL) {
 			gk20a_err(dev_from_gk20a(g),
 				"bind a channel for dbg session");
-			kfree(prof_obj);
+			nvgpu_kfree(g, prof_obj);
 			err = -EINVAL;
 			goto clean_up;
 		}
@@ -1586,7 +1587,7 @@ static int nvgpu_ioctl_free_profiler_object(
 				g->ops.dbg_session_ops.
 				  release_profiler_reservation(dbg_s, prof_obj);
 			list_del(&prof_obj->prof_obj_entry);
-			kfree(prof_obj);
+			nvgpu_kfree(g, prof_obj);
 			obj_found = true;
 			break;
 		}

drivers/gpu/nvgpu/gk20a/debug_gk20a.c

@@ -61,8 +61,7 @@ static void gk20a_debug_dump_all_channel_status_ramfc(struct gk20a *g,
 	u32 chid;
 	struct ch_state **ch_state;
 
-	ch_state = kzalloc(sizeof(*ch_state)
-				 * f->num_channels, GFP_KERNEL);
+	ch_state = nvgpu_kzalloc(g, sizeof(*ch_state) * f->num_channels);
 	if (!ch_state) {
 		gk20a_debug_output(o, "cannot alloc memory for channels\n");
 		return;

drivers/gpu/nvgpu/gk20a/fifo_gk20a.c

@@ -3489,8 +3489,8 @@ static void __gk20a_fifo_profile_free(struct kref *ref)
 {
 	struct fifo_gk20a *f = container_of(ref, struct fifo_gk20a,
 						profile.ref);
-	vfree(f->profile.data);
-	vfree(f->profile.sorted);
+	nvgpu_vfree(f->g, f->profile.data);
+	nvgpu_vfree(f->g, f->profile.sorted);
 }
 
 static int gk20a_fifo_profile_enable(void *data, u64 val)
@@ -3518,8 +3518,8 @@ static int gk20a_fifo_profile_enable(void *data, u64 val)
 							FIFO_PROFILING_ENTRIES *
 							sizeof(u64));
 				if (!(f->profile.data && f->profile.sorted)) {
-					vfree(f->profile.data);
-					vfree(f->profile.sorted);
+					nvgpu_vfree(g, f->profile.data);
+					nvgpu_vfree(g, f->profile.sorted);
 					nvgpu_mutex_release(&f->profile.lock);
 					return -ENOMEM;
 				}

drivers/gpu/nvgpu/gk20a/gr_gk20a.c

@@ -3343,13 +3343,16 @@ int gk20a_comptag_allocator_init(struct gk20a_comptag_allocator *allocator,
 
 void gk20a_comptag_allocator_destroy(struct gk20a_comptag_allocator *allocator)
 {
+	struct gr_gk20a *gr = container_of(allocator,
+					   struct gr_gk20a, comp_tags);
+
 	/*
 	 * called only when exiting the driver (gk20a_remove, or unwinding the
 	 * init stage); no users should be active, so taking the mutex is
 	 * unnecessary here.
 	 */
 	allocator->size = 0;
-	vfree(allocator->bitmap);
+	nvgpu_vfree(gr->g, allocator->bitmap);
 }
 
 static void gk20a_remove_gr_support(struct gr_gk20a *gr)
@@ -3419,7 +3422,7 @@ static void gk20a_remove_gr_support(struct gr_gk20a *gr)
 	nvgpu_kfree(g, gr->ctx_vars.ctxsw_regs.pm_ltc.l);
 	nvgpu_kfree(g, gr->ctx_vars.ctxsw_regs.pm_fbpa.l);
 
-	vfree(gr->ctx_vars.local_golden_image);
+	nvgpu_vfree(g, gr->ctx_vars.local_golden_image);
 	gr->ctx_vars.local_golden_image = NULL;
 
 	if (gr->ctx_vars.hwpm_ctxsw_buffer_offset_map)

drivers/gpu/nvgpu/gk20a/mm_gk20a.c

@@ -4188,7 +4188,7 @@ static int init_vm_page_tables(struct vm_gk20a *vm)
 	err = gk20a_zalloc_gmmu_page_table(vm, 0, &vm->mmu_levels[0],
 					   &vm->pdb, NULL);
 	if (err) {
-		vfree(vm->pdb.entries);
+		nvgpu_vfree(vm->mm->g, vm->pdb.entries);
 		return err;
 	}
 

drivers/gpu/nvgpu/gk20a/tsg_gk20a.c

@@ -260,7 +260,7 @@ struct tsg_gk20a *gk20a_tsg_open(struct gk20a *g)
 	if (g->ops.fifo.tsg_open) {
 		err = g->ops.fifo.tsg_open(tsg);
 		if (err) {
-			gk20a_err(dev_from_gk20a(g), 
+			gk20a_err(dev_from_gk20a(g),
 				  "tsg %d fifo open failed %d",
 				  tsg->tsgid, err);
 			goto clean_up;

drivers/gpu/nvgpu/vgpu/gr_vgpu.c

@@ -1126,7 +1126,7 @@ static int vgpu_gr_suspend_resume_contexts(struct gk20a *g,
 
 	size_in = size_out + n * sizeof(u16);
 
-	msg = kmalloc(size_in, GFP_KERNEL);
+	msg = nvgpu_kmalloc(g, size_in);
 	if (!msg)
 		return -ENOMEM;
 
@@ -1159,7 +1159,7 @@ fail:
 	nvgpu_mutex_release(&g->dbg_sessions_lock);
 
 	*ctx_resident_ch_fd = channel_fd;
-	kfree(msg);
+	nvgpu_kfree(g, msg);
 
 	return err;
 }