host1x_emu: SyncpointIncre per-improvement

1. Add mechanism(s) to trigger syncpoint fence scan when
syncpoint increment UMD called. Following methods are added.
 - METHOD-1: Check fence expire in user context when syncpoint
   increment  UMD API is  called.
 - METHOD-2: Add tasklet based mechanism that schedule tasklet to scan for
   syncpoint fence expiry. This also improve signaling latency.

   METHOD-1 is enabled by default, to enable METHOD-2 define MACRO
   "HOST1X_EMU_SYNC_INC_TASKLET".

2. Add interface "host1x_syncpt_fence_scan()" that can be called from
   client interrupt handler to initiate syncpoint fence scan.

Jira HOSTX-5527

Change-Id: I4d5a0ba9fd67042d824a1df2794b316831001dc4
Signed-off-by: amitabhd <amitabhd@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3267144
Reviewed-by: Raghavendra Vishnu Kumar <rvk@nvidia.com>
Reviewed-by: Sanif Veeras <sveeras@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Leslin Varghese <lvarghese@nvidia.com>

drivers/gpu/host1x-emu/Makefile

@@ -1,11 +1,14 @@
 # SPDX-License-Identifier: GPL-2.0-only
-# Copyright (c) 2024 NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2024-2025 NVIDIA CORPORATION.  All rights reserved.
 
 # Enable below flag for hypervisor build configuration
 ifeq ($(NV_BUILD_CONFIGURATION_EMU_SYNCPT),1)
 ccflags-y += -DHOST1X_EMU_HYPERVISOR
 endif
 
+#Enable Tasklet based fence scanning in SyncpointIncr
+#ccflags-y += -DHOST1X_EMU_SYNC_INC_TASKLET
+
 ifeq ($(NV_BUILD_CONFIGURATION_EXPOSING_T26X), 1)
 LINUXINCLUDE += -I$(srctree.nvidia-oot)/drivers/gpu/host1x-emu/include
 # Enable for verification in VDK. Below allow Emulated Syncpoint driver

drivers/gpu/host1x-emu/fence.c

@@ -1,7 +1,5 @@
-/*
+// SPDX-License-Identifier: GPL-2.0-only
- * SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
- * SPDX-License-Identifier: GPL-2.0-only
- */
 #include <linux/file.h>
 #include <linux/fs.h>
 #include <linux/slab.h>
@@ -188,3 +186,9 @@ HOST1X_EMU_EXPORT_DECL(void, host1x_fence_cancel(struct dma_fence *dfence))
     flush_delayed_work(&sf->timeout_work);
 }
 HOST1X_EMU_EXPORT_SYMBOL(host1x_fence_cancel);
+
+HOST1X_EMU_EXPORT_DECL(void, host1x_syncpt_fence_scan(struct host1x_syncpt *sp))
+{
+	host1x_poll_irq_check_syncpt_fence(sp);
+}
+HOST1X_EMU_EXPORT_SYMBOL(host1x_syncpt_fence_scan);

drivers/gpu/host1x-emu/include/linux/host1x-emu.h

@@ -1,7 +1,5 @@
-/*
+// SPDX-License-Identifier: GPL-2.0-only
- * SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
- * SPDX-License-Identifier: GPL-2.0-only
- */
 #ifndef __LINUX_HOST1X_H
 #define __LINUX_HOST1X_H
 
@@ -61,6 +59,8 @@ HOST1X_EMU_EXPORT_DECL(int, host1x_syncpt_incr(struct host1x_syncpt *sp));
 HOST1X_EMU_EXPORT_DECL(u32, host1x_syncpt_incr_max(struct host1x_syncpt *sp,
                 u32 incrs));
 
+HOST1X_EMU_EXPORT_DECL(void, host1x_syncpt_fence_scan(struct host1x_syncpt *sp));
+
 HOST1X_EMU_EXPORT_DECL(int, host1x_syncpt_wait_ts(struct host1x_syncpt *sp,
                 u32 thresh, long timeout, u32 *value, ktime_t *ts));
 

drivers/gpu/host1x-emu/include/linux/symbol-emu.h

@@ -1,7 +1,5 @@
-/*
+// SPDX-License-Identifier: GPL-2.0-only
- * SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
- * SPDX-License-Identifier: GPL-2.0-only
- */
 #ifndef __HOST1X_EMU_SYMMBOL_H
 #define __HOST1X_EMU_SYMMBOL_H
 
@@ -14,4 +12,9 @@
 #define HOST1X_EMU_EXPORT_DECL(ret, ...)    ret __VA_ARGS__
 #define HOST1X_EMU_EXPORT_SYMBOL(f)         EXPORT_SYMBOL(f)
 #endif
+
+#ifndef CONFIG_TEGRA_HOST1X_EMU_DBG_SYMBL
+HOST1X_EMU_EXPORT_DECL(void, host1x_syncpt_fence_scan(struct host1x_syncpt *sp));
+#endif /*CONFIG_TEGRA_HOST1X_EMU_DBG_SYMBL*/
+
 #endif /*__HOST1X_EMU_SYMMBOL_H*/

drivers/gpu/host1x-emu/poll.c

@@ -1,7 +1,5 @@
-/*
+// SPDX-License-Identifier: GPL-2.0-only
- * SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
- * SPDX-License-Identifier: GPL-2.0-only
- */
 #include "dev.h"
 #include "fence.h"
 #include "poll.h"
@@ -56,29 +54,11 @@ static void host1x_pool_timeout_handler(struct work_struct *work)
     struct host1x_syncpt      *sp;
     struct host1x_syncpt      *tmp_spt;
     struct host1x             *host = pool->host;
-    ktime_t ts = ktime_get();
 
-    spin_lock(&pool->syncpt_list.lock);
 	list_for_each_entry_safe(sp, tmp_spt, &pool->syncpt_list.list, list) {
-        struct host1x_syncpt_fence *fence, *tmp;
+		host1x_poll_irq_check_syncpt_fence(sp);
-        unsigned int value;
-
-        value = host1x_syncpt_load(sp);
-
-        spin_lock(&sp->fences.lock);
-        list_for_each_entry_safe(fence, tmp, &sp->fences.list, list) {
-            if (((value - fence->threshold) & 0x80000000U) != 0U) {
-                /* Fence is not yet expired, we are done */
-                break;
 	}
 
-            list_del_init(&fence->list);
-            host1x_fence_signal(fence ,ts);
-        }
-        spin_unlock(&sp->fences.lock);
-    }
-    spin_unlock(&pool->syncpt_list.lock);
-
     /**
      * TODO: Optimize pool polling mechanism
      */
@@ -116,12 +96,33 @@ int host1x_poll_init(struct host1x *host)
         INIT_LIST_HEAD(&syncpt->list);
 
         /* Add syncpoint to pool list*/
-        list_add(&syncpt->list, &pool->syncpt_list.list);
+	list_add_tail(&syncpt->list, &pool->syncpt_list.list);
     }
 
     return 0;
 }
 
+void host1x_poll_irq_check_syncpt_fence(struct host1x_syncpt  *sp)
+{
+	unsigned int value;
+	unsigned long irqflags;
+	struct host1x_syncpt_fence *fence, *tmp;
+	ktime_t ts = ktime_get();
+
+	value = host1x_syncpt_load(sp);
+	spin_lock_irqsave(&sp->fences.lock, irqflags);
+	list_for_each_entry_safe(fence, tmp, &sp->fences.list, list) {
+		if (((value - fence->threshold) & 0x80000000U) != 0U) {
+			/* Fence is not yet expired, we are done */
+			break;
+		}
+
+		list_del_init(&fence->list);
+		host1x_fence_signal(fence, ts);
+	}
+	spin_unlock_irqrestore(&sp->fences.lock, irqflags);
+}
+
 void host1x_poll_start(struct host1x *host)
 {
     int id;

drivers/gpu/host1x-emu/poll.h

@@ -1,7 +1,5 @@
-/*
+// SPDX-License-Identifier: GPL-2.0-only
- * SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
- * SPDX-License-Identifier: GPL-2.0-only
- */
 #ifndef __HOST1X_EMU_POLL_H
 #define __HOST1X_EMU_POLL_H
 
@@ -10,6 +8,7 @@
 #include <linux/timekeeping.h>
 
 struct host1x;
+struct host1x_syncpt;
 struct host1x_syncpt_fence;
 
 struct host1x_fence_list {
@@ -58,4 +57,8 @@ void host1x_poll_add_fence_locked(struct host1x *host, struct host1x_syncpt_fenc
  */
 bool host1x_poll_remove_fence(struct host1x *host, struct host1x_syncpt_fence *fence);
 
+/**
+ * Description: Check if syncpoint fence expired
+ */
+void host1x_poll_irq_check_syncpt_fence(struct host1x_syncpt  *sp);
 #endif

drivers/gpu/host1x-emu/syncpt.c

@@ -1,17 +1,33 @@
-/*
+// SPDX-License-Identifier: GPL-2.0-only
- * SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
- * SPDX-License-Identifier: GPL-2.0-only
- */
 #include <linux/module.h>
 #include <linux/delay.h>
 #include <linux/device.h>
 #include <linux/dma-fence.h>
 #include <linux/slab.h>
 #include <linux/timekeeping.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
 
 #include "dev.h"
 #include "syncpt.h"
 
+#ifdef HOST1X_EMU_SYNC_INC_TASKLET
+static void tasklet_fn(struct tasklet_struct *unused);
+static DEFINE_PER_CPU(struct host1x_syncpt *, tasklet_sp);
+
+DECLARE_TASKLET(syncpt_tasklet, tasklet_fn);
+
+static void tasklet_fn(struct tasklet_struct *unused)
+{
+	struct host1x_syncpt *sp = NULL;
+
+	sp = this_cpu_read(tasklet_sp);
+	if (sp != NULL)
+		host1x_poll_irq_check_syncpt_fence(sp);
+}
+#endif
+
 static void syncpt_release(struct kref *ref)
 {
     struct host1x_syncpt *sp = container_of(ref, struct host1x_syncpt, ref);
@@ -241,7 +257,17 @@ HOST1X_EMU_EXPORT_SYMBOL(host1x_syncpt_read_max);
  */
 HOST1X_EMU_EXPORT_DECL(int, host1x_syncpt_incr(struct host1x_syncpt *sp))
 {
-    return host1x_hw_syncpt_cpu_incr(sp->host, sp);
+	int err;
+
+	err = host1x_hw_syncpt_cpu_incr(sp->host, sp);
+#ifdef HOST1X_EMU_SYNC_INC_TASKLET
+	/*Improve Signaling performance*/
+	this_cpu_write(tasklet_sp, sp);
+	tasklet_schedule(&syncpt_tasklet);
+#else
+	host1x_poll_irq_check_syncpt_fence(sp);
+#endif
+	return err;
 }
 HOST1X_EMU_EXPORT_SYMBOL(host1x_syncpt_incr);