Revert "gpu: nvgpu: correct usage for gk20a_busy_noresume"

This reverts commit c1ea9e3955.

Reason for revert: ap_vulkan, ap_opengles, ap_mods tests failures
Bug 3661058
Bug 3661080 
Bug 3659004 

Change-Id: I929b5675a4fb0ddc8cbf3eeefc982b4ba04ddc59
Signed-off-by: Krishna Reddy <vdumpa@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2718996
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>

drivers/gpu/nvgpu/common/fifo/runlist.c

@@ -658,14 +658,31 @@ static int nvgpu_runlist_do_update(struct gk20a *g, struct nvgpu_runlist *rl,
 static void runlist_select_locked(struct gk20a *g, struct nvgpu_runlist *runlist,
 		struct nvgpu_runlist_domain *next_domain)
 {
+	int err;
+
 	rl_dbg(g, "Runlist[%u]: switching to domain %s",
 	       runlist->id, next_domain->name);
 
 	runlist->domain = next_domain;
 
-	if (!gk20a_busy_noresume(g)) {
+	gk20a_busy_noresume(g);
+	if (nvgpu_is_powered_off(g)) {
 		rl_dbg(g, "Runlist[%u]: power is off, skip submit",
 				runlist->id);
+		gk20a_idle_nosuspend(g);
+		return;
+	}
+
+	err = gk20a_busy(g);
+	gk20a_idle_nosuspend(g);
+
+	if (err != 0) {
+		nvgpu_err(g, "failed to hold power for runlist submit");
+		/*
+		 * probably shutting down though, so don't bother propagating
+		 * the error. Power is already on when the domain scheduler is
+		 * actually in use.
+		 */
 		return;
 	}
 
@@ -678,7 +695,7 @@ static void runlist_select_locked(struct gk20a *g, struct nvgpu_runlist *runlist
 	 */
 	g->ops.runlist.hw_submit(g, runlist);
 
-	gk20a_idle_nosuspend(g);
+	gk20a_idle(g);
 }
 
 static void runlist_switch_domain_locked(struct gk20a *g,

drivers/gpu/nvgpu/common/pmu/perfmon/pmu_perfmon.c

@@ -413,9 +413,10 @@ int nvgpu_pmu_busy_cycles_norm(struct gk20a *g, u32 *norm)
 	u64 busy_cycles, total_cycles;
 	u32 intr_status;
 
-	if (!gk20a_busy_noresume(g)) {
+	gk20a_busy_noresume(g);
+	if (nvgpu_is_powered_off(g)) {
 		*norm = 0;
-		return 0;
+		goto exit;
 	}
 
 	if (g->ops.pmu.pmu_read_idle_counter == NULL ||

drivers/gpu/nvgpu/hal/fb/fb_ga100.c

@@ -1,7 +1,7 @@
 /*
  * GA100 FB
  *
- * Copyright (c) 2020-2022, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2020, NVIDIA CORPORATION.  All rights reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -63,8 +63,9 @@ bool ga100_fb_is_comptagline_mode_enabled(struct gk20a *g)
 	u32 reg = 0U;
 	bool result = true;
 
-	if (!gk20a_busy_noresume(g)) {
-		return result;
+	gk20a_busy_noresume(g);
+	if (nvgpu_is_powered_off(g)) {
+		goto done;
 	}
 
 	reg = nvgpu_readl(g, fb_mmu_hypervisor_ctl_r());
@@ -72,6 +73,7 @@ bool ga100_fb_is_comptagline_mode_enabled(struct gk20a *g)
 	result = (fb_mmu_hypervisor_ctl_force_cbc_raw_mode_v(reg) ==
 		fb_mmu_hypervisor_ctl_force_cbc_raw_mode_disable_v());
 
+done:
 	gk20a_idle_nosuspend(g);
 	return result;
 }

drivers/gpu/nvgpu/hal/fb/fb_gm20b_fusa.c

@@ -88,7 +88,7 @@ int gm20b_fb_tlb_invalidate(struct gk20a *g, struct nvgpu_mem *pdb)
 	   hw. Use the power_on flag to skip tlb invalidation when gpu
 	   power is turned off */
 
-	if (!gk20a_busy_noresume(g)) {
+	if (nvgpu_is_powered_off(g)) {
 		return err;
 	}
 
@@ -150,9 +150,6 @@ int gm20b_fb_tlb_invalidate(struct gk20a *g, struct nvgpu_mem *pdb)
 
 out:
 	nvgpu_mutex_release(&g->mm.tlb_lock);
-
-	gk20a_idle_nosuspend(g);
-
 	return err;
 }
 

drivers/gpu/nvgpu/hal/fb/fb_tu104.c

@@ -56,8 +56,8 @@ int fb_tu104_tlb_invalidate(struct gk20a *g, struct nvgpu_mem *pdb)
 	 * hw. Use the power_on flag to skip tlb invalidation when gpu
 	 * power is turned off
 	 */
-	if (!gk20a_busy_noresume(g)) {
-		return err;
+	if (nvgpu_is_powered_off(g)) {
+		return 0;
 	}
 
 	addr_lo = u64_lo32(nvgpu_mem_get_addr(g, pdb) >> 12);
@@ -101,9 +101,6 @@ int fb_tu104_tlb_invalidate(struct gk20a *g, struct nvgpu_mem *pdb)
 #endif
 
 	nvgpu_mutex_release(&g->mm.tlb_lock);
-
-	gk20a_idle_nosuspend(g);
-
 	return err;
 }
 

drivers/gpu/nvgpu/hal/mm/cache/flush_gk20a_fusa.c

@@ -41,7 +41,9 @@ int gk20a_mm_fb_flush(struct gk20a *g)
 
 	nvgpu_log(g, gpu_dbg_mm, " ");
 
-	if (!gk20a_busy_noresume(g)) {
+	gk20a_busy_noresume(g);
+	if (nvgpu_is_powered_off(g)) {
+		gk20a_idle_nosuspend(g);
 		return 0;
 	}
 
@@ -149,12 +151,13 @@ static void gk20a_mm_l2_invalidate_locked(struct gk20a *g)
 void gk20a_mm_l2_invalidate(struct gk20a *g)
 {
 	struct mm_gk20a *mm = &g->mm;
-	if (gk20a_busy_noresume(g)) {
+	gk20a_busy_noresume(g);
+	if (!nvgpu_is_powered_off(g)) {
 		nvgpu_mutex_acquire(&mm->l2_op_lock);
 		gk20a_mm_l2_invalidate_locked(g);
 		nvgpu_mutex_release(&mm->l2_op_lock);
-		gk20a_idle_nosuspend(g);
 	}
+	gk20a_idle_nosuspend(g);
 }
 
 int gk20a_mm_l2_flush(struct gk20a *g, bool invalidate)
@@ -167,7 +170,9 @@ int gk20a_mm_l2_flush(struct gk20a *g, bool invalidate)
 
 	nvgpu_log(g, gpu_dbg_mm, " ");
 
-	if (!gk20a_busy_noresume(g)) {
+	gk20a_busy_noresume(g);
+	if (nvgpu_is_powered_off(g)) {
+		gk20a_idle_nosuspend(g);
 		return 0;
 	}
 
@@ -229,8 +234,9 @@ void gk20a_mm_cbc_clean(struct gk20a *g)
 
 	nvgpu_log_fn(g, " ");
 
-	if (!gk20a_busy_noresume(g)) {
-		return;
+	gk20a_busy_noresume(g);
+	if (nvgpu_is_powered_off(g)) {
+		goto hw_was_off;
 	}
 
 	if (g->ops.mm.get_flush_retries != NULL) {
@@ -262,6 +268,7 @@ void gk20a_mm_cbc_clean(struct gk20a *g)
 
 	nvgpu_mutex_release(&mm->l2_op_lock);
 
+hw_was_off:
 	gk20a_idle_nosuspend(g);
 }
 #endif

drivers/gpu/nvgpu/include/nvgpu/nvgpu_init.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2022, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2019-2021, NVIDIA CORPORATION.  All rights reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -353,7 +353,7 @@ int nvgpu_init_gpu_characteristics(struct gk20a *g);
  *
  * @param g [in] The GPU
  */
-bool gk20a_busy_noresume(struct gk20a *g);
+void gk20a_busy_noresume(struct gk20a *g);
 
 /**
  * @brief Drops a reference for gpu. Does nothing for safety.

drivers/gpu/nvgpu/os/linux/module.c

@@ -168,63 +168,9 @@ struct device_node *nvgpu_get_node(struct gk20a *g)
 	return dev->of_node;
 }
 
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0)
-static bool gk20a_pm_runtime_get_if_in_use(struct gk20a *g)
+void gk20a_busy_noresume(struct gk20a *g)
 {
-	int ret = pm_runtime_get_if_active(dev_from_gk20a(g), true);
-	if (ret == 1) {
-		return true;
-	} else {
-		return false;
-	}
-}
-#endif
-
-static bool gk20a_busy_noresume_legacy(struct gk20a *g)
-{
-	struct device *dev = dev_from_gk20a(g);;
-
-	pm_runtime_get_noresume(dev);
-	if (nvgpu_is_powered_off(g)) {
-		pm_runtime_put_noidle(dev);
-		return false;
-	} else {
-		return true;
-	}
-}
-
-/*
- * Cases:
- * 1) For older than Kernel 5.8, use legacy
- *    i.e. gk20a_busy_noresume_legacy()
- * 2) else if pm_runtime_disabled (e.g. VGPU, DGPU)
- *    use legacy.
- * 3) Else use gk20a_pm_runtime_get_if_in_use()
- */
-bool gk20a_busy_noresume(struct gk20a *g)
-{
-	struct device *dev;
-
-	if (!g)
-		return false;
-
-	dev = dev_from_gk20a(g);
-
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0)
-	if (pm_runtime_enabled(dev)) {
-		if (gk20a_pm_runtime_get_if_in_use(g)) {
-			atomic_inc(&g->usage_count.atomic_var);
-			return true;
-		} else {
-			return false;
-		}
-	} else {
-		/* VGPU, DGPU */
-		return gk20a_busy_noresume_legacy(g);
-	}
-#else
-	return gk20a_busy_noresume_legacy(g);
-#endif
+	pm_runtime_get_noresume(dev_from_gk20a(g));
 }
 
 int gk20a_busy(struct gk20a *g)
@@ -276,17 +222,7 @@ fail:
 
 void gk20a_idle_nosuspend(struct gk20a *g)
 {
-	struct device *dev = dev_from_gk20a(g);
-
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0)
-	if (pm_runtime_enabled(dev)) {
-		gk20a_idle(g);
-	} else {
-		pm_runtime_put_noidle(dev);
-	}
-#else
-	pm_runtime_put_noidle(dev);
-#endif
+	pm_runtime_put_noidle(dev_from_gk20a(g));
 }
 
 void gk20a_idle(struct gk20a *g)
@@ -2005,14 +1941,6 @@ int nvgpu_remove(struct device *dev)
 	err = nvgpu_quiesce(g);
 	WARN(err, "gpu failed to idle during driver removal");
 
-	/**
-	 * nvgpu_quiesce has been invoked already, disable pm runtime.
-	 * Informs PM domain that its safe to power down the h/w now.
-	 * Anything after this is just software deinit. Any cache/tlb
-	 * flush/invalidate must have already happened before this.
-	 */
-	gk20a_pm_deinit(dev);
-
 	if (nvgpu_mem_is_valid(&g->syncpt_mem))
 		nvgpu_dma_free(g, &g->syncpt_mem);
 
@@ -2073,6 +2001,8 @@ static int __exit gk20a_remove(struct platform_device *pdev)
 
 	nvgpu_put(g);
 
+	gk20a_pm_deinit(dev);
+
 	return err;
 }
 

drivers/gpu/nvgpu/os/posix/nvgpu.c

@@ -116,13 +116,9 @@ void nvgpu_disable_irqs(struct gk20a *g)
 /*
  * We have no runtime PM stuff in userspace so these are really just noops.
  */
-bool gk20a_busy_noresume(struct gk20a *g)
+void gk20a_busy_noresume(struct gk20a *g)
 {
-	if (nvgpu_is_powered_off(g)) {
-		return false;
-	} else {
-		return true;
-	}
+	(void)g;
 }
 
 void gk20a_idle_nosuspend(struct gk20a *g)