gpu: nvgpu: Do not clear PMU state on rail gate

When rail gating, we cleared all PMU status. Clear only the relevant
fields.

Change-Id: I5b4e8d74339aae6f1c6b945f45b8378bb563e8be
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/406843
(cherry-picked from commit 4e35d12b2059140beb58554633c6b3f291e6fec7)
Reviewed-on: http://git-master/r/413226
Reviewed-by: Mandar Padmawar <mpadmawar@nvidia.com>
Tested-by: Mandar Padmawar <mpadmawar@nvidia.com>

3 files changed, 16 insertions, 75 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.c b/drivers/gpu/nvgpu/gk20a/gk20a.c
index 19880da4d82a..aa63f499f85c 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.c
@@ -609,8 +609,9 @@ static void gk20a_remove_support(struct platform_device *dev)
 {
 	struct gk20a *g = get_gk20a(dev);
 
-	/* pmu support should already be removed when driver turns off
-	   gpu power rail in prepapre_poweroff */
+	if (g->pmu.remove_support)
+		g->pmu.remove_support(&g->pmu);
+
 	if (g->gk20a_cdev.gk20a_cooling_dev)
 		thermal_cooling_device_unregister(g->gk20a_cdev.gk20a_cooling_dev);
 
diff --git a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
index 8e2a62ded331..32db00ecef41 100644
--- a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
@@ -1491,57 +1491,11 @@ static int pmu_queue_close(struct pmu_gk20a *pmu,
 	return 0;
 }
 
-static void gk20a_save_pmu_sw_state(struct pmu_gk20a *pmu,
-			struct gk20a_pmu_save_state *save)
-{
-	save->seq = pmu->seq;
-	save->next_seq_desc = pmu->next_seq_desc;
-	save->mutex = pmu->mutex;
-	save->mutex_cnt = pmu->mutex_cnt;
-	save->desc = pmu->desc;
-	save->ucode = pmu->ucode;
-	save->elpg_enable = pmu->elpg_enable;
-	save->pg_wq = pmu->pg_wq;
-	save->seq_buf = pmu->seq_buf;
-	save->pg_buf = pmu->pg_buf;
-	save->sw_ready = pmu->sw_ready;
-	save->pg_init = pmu->pg_init;
-}
-
-static void gk20a_restore_pmu_sw_state(struct pmu_gk20a *pmu,
-			struct gk20a_pmu_save_state *save)
-{
-	pmu->seq = save->seq;
-	pmu->next_seq_desc = save->next_seq_desc;
-	pmu->mutex = save->mutex;
-	pmu->mutex_cnt = save->mutex_cnt;
-	pmu->desc = save->desc;
-	pmu->ucode = save->ucode;
-	pmu->elpg_enable = save->elpg_enable;
-	pmu->pg_wq = save->pg_wq;
-	pmu->seq_buf = save->seq_buf;
-	pmu->pg_buf = save->pg_buf;
-	pmu->sw_ready = save->sw_ready;
-	pmu->pg_init = save->pg_init;
-}
-
 void gk20a_remove_pmu_support(struct pmu_gk20a *pmu)
 {
-	struct gk20a_pmu_save_state save;
-
 	gk20a_dbg_fn("");
 
 	gk20a_allocator_destroy(&pmu->dmem);
-
-	/* Save the stuff you don't want to lose */
-	gk20a_save_pmu_sw_state(pmu, &save);
-
-	/* this function is also called by pmu_destory outside gk20a deinit that
-	   releases gk20a struct so fill up with zeros here. */
-	memset(pmu, 0, sizeof(struct pmu_gk20a));
-
-	/* Restore stuff you want to keep */
-	gk20a_restore_pmu_sw_state(pmu, &save);
 }
 
 int gk20a_init_pmu_reset_enable_hw(struct gk20a *g)
@@ -2197,7 +2151,7 @@ static int pmu_init_perfmon(struct pmu_gk20a *pmu)
 	struct pmu_payload payload;
 	u32 seq;
 	u32 data;
-	int err;
+	int err = 0;
 
 	gk20a_dbg_fn("");
 
@@ -2247,8 +2201,9 @@ static int pmu_init_perfmon(struct pmu_gk20a *pmu)
 			pwr_pmu_idle_ctrl_filter_disabled_f());
 	gk20a_writel(g, pwr_pmu_idle_ctrl_r(2), data);
 
-	pmu->sample_buffer = 0;
-	err = pmu->dmem.alloc(&pmu->dmem, &pmu->sample_buffer, 2 * sizeof(u16));
+	if (!pmu->sample_buffer)
+		err = pmu->dmem.alloc(&pmu->dmem,
+				      &pmu->sample_buffer, 2 * sizeof(u16));
 	if (err) {
 		gk20a_err(dev_from_gk20a(g),
 			"failed to allocate perfmon sample buffer");
@@ -2347,10 +2302,11 @@ static int pmu_process_init_msg(struct pmu_gk20a *pmu,
 	for (i = 0; i < PMU_QUEUE_COUNT; i++)
 		pmu_queue_init(pmu, i, init);
 
-	gk20a_allocator_init(&pmu->dmem, "gk20a_pmu_dmem",
-			pv->get_pmu_init_msg_pmu_sw_mg_off(init),
-			pv->get_pmu_init_msg_pmu_sw_mg_size(init),
-			PMU_DMEM_ALLOC_ALIGNMENT);
+	if (!pmu->dmem.alloc)
+		gk20a_allocator_init(&pmu->dmem, "gk20a_pmu_dmem",
+				pv->get_pmu_init_msg_pmu_sw_mg_off(init),
+				pv->get_pmu_init_msg_pmu_sw_mg_size(init),
+				PMU_DMEM_ALLOC_ALIGNMENT);
 
 	pmu->pmu_ready = true;
 
@@ -3448,11 +3404,10 @@ int gk20a_pmu_destroy(struct gk20a *g)
 	g->pg_gating_cnt += gating_cnt;
 
 	pmu_enable(pmu, false);
-
-	if (pmu->remove_support) {
-		pmu->remove_support(pmu);
-		pmu->remove_support = NULL;
-	}
+	pmu->pmu_ready = false;
+	pmu->perfmon_ready = false;
+	pmu->zbc_ready = false;
+	pmu->elpg_ready = false;
 
 	gk20a_dbg_fn("done");
 	return 0;
diff --git a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
index de519bf67b62..7cc351bf292a 100644
--- a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
@@ -1051,21 +1051,6 @@ struct pmu_gk20a {
 	};
 };
 
-struct gk20a_pmu_save_state {
-	struct pmu_sequence *seq;
-	u32 next_seq_desc;
-	struct pmu_mutex *mutex;
-	u32 mutex_cnt;
-	struct pmu_ucode_desc *desc;
-	struct pmu_mem_desc ucode;
-	struct pmu_mem_desc seq_buf;
-	struct pmu_mem_desc pg_buf;
-	struct delayed_work elpg_enable;
-	wait_queue_head_t pg_wq;
-	bool sw_ready;
-	struct work_struct pg_init;
-};
-
 int gk20a_init_pmu_support(struct gk20a *g);
 int gk20a_init_pmu_setup_hw2(struct gk20a *g);