* [RFC v5 2/4] CMDQ: Mediatek CMDQ driver
2016-05-04 12:32 [RFC v5 0/4] Mediatek MT8173 CMDQ support HS Liao
[not found] ` <1462365137-715-1-git-send-email-hs.liao-NuS5LvNUpcJWk0Htik3J/w@public.gmane.org>
@ 2016-05-04 12:32 ` HS Liao
2016-05-04 12:32 ` [RFC v5 3/4] arm64: dts: mt8173: Add GCE node HS Liao
2016-05-04 12:32 ` [RFC v5 4/4] CMDQ: suspend/resume protection HS Liao
3 siblings, 0 replies; 5+ messages in thread
From: HS Liao @ 2016-05-04 12:32 UTC (permalink / raw)
To: Rob Herring, Matthias Brugger
Cc: Daniel Kurtz, Sascha Hauer, devicetree, linux-kernel,
linux-arm-kernel, linux-mediatek, srv_heupstream, Sascha Hauer,
Philipp Zabel, Nicolas Boichat, CK HU, cawa cheng, Bibby Hsieh,
YT Shen, Daoyuan Huang, Damon Chu, Josh-YC Liu, Glory Hung,
Jiaguang Zhang, HS Liao
This patch is first version of Mediatek Command Queue(CMDQ) driver. The
CMDQ is used to help read/write registers with critical time limitation,
such as updating display configuration during the vblank. It controls
Global Command Engine (GCE) hardware to achieve this requirement.
Currently, CMDQ only supports display related hardwares, but we expect
it can be extended to other hardwares for future requirements.
Signed-off-by: HS Liao <hs.liao@mediatek.com>
---
drivers/soc/mediatek/Kconfig | 10 +
drivers/soc/mediatek/Makefile | 1 +
drivers/soc/mediatek/mtk-cmdq.c | 1425 +++++++++++++++++++++++++++++++++++++++
include/soc/mediatek/cmdq.h | 197 ++++++
4 files changed, 1633 insertions(+)
create mode 100644 drivers/soc/mediatek/mtk-cmdq.c
create mode 100644 include/soc/mediatek/cmdq.h
diff --git a/drivers/soc/mediatek/Kconfig b/drivers/soc/mediatek/Kconfig
index 0a4ea80..c4ad75c 100644
--- a/drivers/soc/mediatek/Kconfig
+++ b/drivers/soc/mediatek/Kconfig
@@ -1,6 +1,16 @@
#
# MediaTek SoC drivers
#
+config MTK_CMDQ
+ bool "MediaTek CMDQ Support"
+ depends on ARCH_MEDIATEK || COMPILE_TEST
+ select MTK_INFRACFG
+ help
+ Say yes here to add support for the MediaTek Command Queue (CMDQ)
+ driver. The CMDQ is used to help read/write registers with critical
+ time limitation, such as updating display configuration during the
+ vblank.
+
config MTK_INFRACFG
bool "MediaTek INFRACFG Support"
depends on ARCH_MEDIATEK || COMPILE_TEST
diff --git a/drivers/soc/mediatek/Makefile b/drivers/soc/mediatek/Makefile
index 12998b0..f7397ef 100644
--- a/drivers/soc/mediatek/Makefile
+++ b/drivers/soc/mediatek/Makefile
@@ -1,3 +1,4 @@
+obj-$(CONFIG_MTK_CMDQ) += mtk-cmdq.o
obj-$(CONFIG_MTK_INFRACFG) += mtk-infracfg.o
obj-$(CONFIG_MTK_PMIC_WRAP) += mtk-pmic-wrap.o
obj-$(CONFIG_MTK_SCPSYS) += mtk-scpsys.o
diff --git a/drivers/soc/mediatek/mtk-cmdq.c b/drivers/soc/mediatek/mtk-cmdq.c
new file mode 100644
index 0000000..d3e9a4c
--- /dev/null
+++ b/drivers/soc/mediatek/mtk-cmdq.c
@@ -0,0 +1,1425 @@
+/*
+ * Copyright (c) 2015 MediaTek Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ */
+
+#include <linux/clk.h>
+#include <linux/clk-provider.h>
+#include <linux/dma-mapping.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/iopoll.h>
+#include <linux/kernel.h>
+#include <linux/kthread.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/of_address.h>
+#include <linux/of_irq.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/suspend.h>
+#include <linux/workqueue.h>
+#include <soc/mediatek/cmdq.h>
+
+#define CMDQ_MAX_THREAD_COUNT 3 /* general, main, sub */
+#define CMDQ_MAX_TASK_IN_THREAD 20
+
+#define CMDQ_INITIAL_CMD_BLOCK_SIZE PAGE_SIZE
+#define CMDQ_INST_SIZE 8 /* instruction is 64-bit */
+
+/*
+ * cmdq_thread cookie value is from 0 to CMDQ_MAX_COOKIE_VALUE.
+ * And, this value also be used as MASK.
+ */
+#define CMDQ_MAX_COOKIE_VALUE 0xffff
+#define CMDQ_COOKIE_MASK CMDQ_MAX_COOKIE_VALUE
+
+#define CMDQ_DEFAULT_TIMEOUT_MS 1000
+#define CMDQ_ACQUIRE_THREAD_TIMEOUT_MS 5000
+
+#define CMDQ_DRIVER_DEVICE_NAME "mtk_cmdq"
+#define CMDQ_CLK_NAME "gce"
+
+#define CMDQ_CURR_IRQ_STATUS_OFFSET 0x010
+#define CMDQ_THR_SLOT_CYCLES_OFFSET 0x030
+
+#define CMDQ_THR_BASE 0x100
+#define CMDQ_THR_SHIFT 0x080
+#define CMDQ_THR_WARM_RESET_OFFSET 0x00
+#define CMDQ_THR_ENABLE_TASK_OFFSET 0x04
+#define CMDQ_THR_SUSPEND_TASK_OFFSET 0x08
+#define CMDQ_THR_CURR_STATUS_OFFSET 0x0c
+#define CMDQ_THR_IRQ_STATUS_OFFSET 0x10
+#define CMDQ_THR_IRQ_ENABLE_OFFSET 0x14
+#define CMDQ_THR_CURR_ADDR_OFFSET 0x20
+#define CMDQ_THR_END_ADDR_OFFSET 0x24
+#define CMDQ_THR_EXEC_CNT_OFFSET 0x28
+#define CMDQ_THR_CFG_OFFSET 0x40
+#define CMDQ_THR_INST_CYCLES_OFFSET 0x50
+
+#define CMDQ_SYNC_TOKEN_SET BIT(16)
+#define CMDQ_IRQ_MASK 0xffff
+
+#define CMDQ_THR_ENABLED 0x1
+#define CMDQ_THR_DISABLED 0x0
+#define CMDQ_THR_SUSPEND 0x1
+#define CMDQ_THR_RESUME 0x0
+#define CMDQ_THR_STATUS_SUSPENDED BIT(1)
+#define CMDQ_THR_WARM_RESET BIT(0)
+#define CMDQ_THR_SLOT_CYCLES 0x3200
+#define CMDQ_THR_NO_TIMEOUT 0x0
+#define CMDQ_THR_PRIORITY 3
+#define CMDQ_THR_IRQ_DONE 0x1
+#define CMDQ_THR_IRQ_ERROR 0x12
+#define CMDQ_THR_IRQ_EN 0x13 /* done + error */
+#define CMDQ_THR_IRQ_MASK 0x13
+#define CMDQ_THR_EXECUTING BIT(31)
+
+#define CMDQ_ARG_A_MASK 0xffffff
+#define CMDQ_ARG_A_WRITE_MASK 0xffff
+#define CMDQ_ARG_A_SUBSYS_MASK 0x1f0000
+#define CMDQ_SUBSYS_MASK 0x1f
+
+#define CMDQ_OP_CODE_SHIFT 24
+#define CMDQ_SUBSYS_SHIFT 16
+
+#define CMDQ_JUMP_BY_OFFSET 0x10000000
+#define CMDQ_JUMP_BY_PA 0x10000001
+#define CMDQ_JUMP_PASS CMDQ_INST_SIZE
+
+#define CMDQ_WFE_UPDATE BIT(31)
+#define CMDQ_WFE_WAIT BIT(15)
+#define CMDQ_WFE_WAIT_VALUE 0x1
+
+#define CMDQ_EOC_IRQ_EN BIT(0)
+
+#define CMDQ_ENABLE_MASK BIT(0)
+
+#define CMDQ_OP_CODE_MASK 0xff000000
+
+struct cmdq_command {
+ struct cmdq *cmdq;
+ u64 engine_flag;
+ void *base; /* command buffer pointer */
+ size_t size; /* command buffer size (bytes) */
+};
+
+/*
+ * CMDQ_CODE_MOVE:
+ * move value into internal register as mask
+ * format: op mask
+ * CMDQ_CODE_WRITE:
+ * write value into target register
+ * format: op subsys address value
+ * CMDQ_CODE_JUMP:
+ * jump by offset
+ * format: op offset
+ * CMDQ_CODE_WFE:
+ * wait for event and clear
+ * it is just clear if no wait
+ * format: [wait] op event update:1 to_wait:1 wait:1
+ * [clear] op event update:1 to_wait:0 wait:0
+ * CMDQ_CODE_EOC:
+ * end of command
+ * format: op irq_flag
+ */
+enum cmdq_code {
+ CMDQ_CODE_MOVE = 0x02,
+ CMDQ_CODE_WRITE = 0x04,
+ CMDQ_CODE_JUMP = 0x10,
+ CMDQ_CODE_WFE = 0x20,
+ CMDQ_CODE_EOC = 0x40,
+};
+
+enum cmdq_task_state {
+ TASK_STATE_IDLE, /* free task */
+ TASK_STATE_BUSY, /* task running on a thread */
+ TASK_STATE_ERROR, /* task execution error */
+ TASK_STATE_DONE, /* task finished */
+ TASK_STATE_WAITING, /* allocated but waiting for available thread */
+};
+
+struct cmdq_task_cb {
+ cmdq_async_flush_cb cb;
+ void *data;
+};
+
+struct cmdq_thread;
+
+struct cmdq_task {
+ struct cmdq *cmdq;
+ struct list_head list_entry;
+ enum cmdq_task_state task_state;
+ void *va_base; /* va */
+ dma_addr_t mva_base; /* pa */
+ u64 engine_flag;
+ size_t command_size;
+ u32 num_cmd; /* 2 * number of commands */
+ struct cmdq_thread *thread;
+ struct cmdq_task_cb cb; /* callback */
+ /* work item when auto release is used */
+ struct work_struct auto_release_work;
+};
+
+struct cmdq_thread {
+ int id;
+ void __iomem *base;
+ u32 task_count;
+ u32 wait_cookie;
+ u32 next_cookie;
+ struct cmdq_task *cur_task[CMDQ_MAX_TASK_IN_THREAD];
+ wait_queue_head_t wait_queue; /* wait task done */
+};
+
+struct cmdq {
+ struct device *dev;
+ void __iomem *base;
+ u32 irq;
+
+ /*
+ * task_cache: struct cmdq_task object cache
+ * task_active_list: active tasks
+ * task_consume_wait_queue_item: task consuming work
+ * task_consume_wq: task consuming workqueue
+ * task_auto_release_wq: auto-release workqueue
+ */
+ struct kmem_cache *task_cache;
+ struct list_head task_active_list;
+ struct list_head task_wait_list;
+ struct work_struct task_consume_wait_queue_item;
+ struct workqueue_struct *task_consume_wq;
+ struct workqueue_struct *task_auto_release_wq;
+
+ struct cmdq_thread thread[CMDQ_MAX_THREAD_COUNT];
+ struct mutex task_mutex; /* for task list */
+ spinlock_t exec_lock; /* for exec task */
+ wait_queue_head_t thread_dispatch_queue;
+ struct clk *clock;
+};
+
+struct cmdq_subsys {
+ u32 base_addr;
+ int id;
+};
+
+static const struct cmdq_subsys g_subsys[] = {
+ {0x1400, 1},
+ {0x1401, 2},
+ {0x1402, 3},
+};
+
+static int cmdq_clk_enable(struct cmdq *cmdq)
+{
+ struct device *dev = cmdq->dev;
+ int err;
+
+ err = clk_prepare_enable(cmdq->clock);
+ if (err < 0) {
+ dev_err(dev, "prepare and enable clk:%s fail\n",
+ CMDQ_CLK_NAME);
+ return err;
+ }
+ return 0;
+}
+
+static void cmdq_clk_disable(struct cmdq *cmdq)
+{
+ clk_disable_unprepare(cmdq->clock);
+}
+
+static int cmdq_subsys_base_addr_to_id(u32 base_addr)
+{
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(g_subsys); i++)
+ if (g_subsys[i].base_addr == base_addr)
+ return g_subsys[i].id;
+ return -EFAULT;
+}
+
+static int cmdq_eng_get_thread(u64 flag)
+{
+ if (flag & BIT_ULL(CMDQ_ENG_DISP_DSI0))
+ return 0;
+ else if (flag & BIT_ULL(CMDQ_ENG_DISP_DPI0))
+ return 1;
+ else
+ return 2;
+}
+
+static void cmdq_task_free_command_buffer(struct cmdq_task *task)
+{
+ if (!task->va_base)
+ return;
+
+ dma_free_coherent(task->cmdq->dev, task->command_size, task->va_base,
+ task->mva_base);
+ task->va_base = NULL;
+ task->mva_base = 0;
+ task->command_size = 0;
+ task->num_cmd = 0;
+}
+
+/* allocate and initialize struct cmdq_task and its command buffer */
+static struct cmdq_task *cmdq_task_create(struct cmdq *cmdq, size_t size)
+{
+ struct device *dev = cmdq->dev;
+ struct cmdq_task *task;
+
+ mutex_lock(&cmdq->task_mutex);
+ task = kmem_cache_zalloc(cmdq->task_cache, GFP_KERNEL);
+ INIT_LIST_HEAD(&task->list_entry);
+ task->task_state = TASK_STATE_IDLE;
+ task->cmdq = cmdq;
+ task->command_size = size;
+ task->va_base = dma_alloc_coherent(dev, size, &task->mva_base,
+ GFP_KERNEL);
+ if (!task->va_base) {
+ dev_err(dev, "allocate command buffer failed\n");
+ kmem_cache_free(cmdq->task_cache, task);
+ task = NULL;
+ }
+ mutex_unlock(&cmdq->task_mutex);
+ return task;
+}
+
+static void cmdq_task_release_internal(struct cmdq_task *task)
+{
+ struct cmdq *cmdq = task->cmdq;
+
+ mutex_lock(&cmdq->task_mutex);
+ cmdq_task_free_command_buffer(task);
+ list_del(&task->list_entry);
+ kmem_cache_free(cmdq->task_cache, task);
+ mutex_unlock(&cmdq->task_mutex);
+}
+
+/* After dropping error task, we have to reorder remaining valid tasks. */
+static void cmdq_thread_reorder_task_array(struct cmdq_thread *thread,
+ int prev_id)
+{
+ int to_id, from_id;
+ struct cmdq_task *task;
+ u32 *task_base;
+
+ to_id = (prev_id + 1) % CMDQ_MAX_TASK_IN_THREAD;
+ if (thread->cur_task[to_id])
+ return;
+
+ thread->next_cookie--;
+ from_id = (to_id + 1) % CMDQ_MAX_TASK_IN_THREAD;
+ for (; from_id != prev_id;
+ from_id = (from_id + 1) % CMDQ_MAX_TASK_IN_THREAD) {
+ if (!thread->cur_task[from_id]) {
+ thread->next_cookie--;
+ continue;
+ }
+ thread->cur_task[to_id] = thread->cur_task[from_id];
+ thread->cur_task[from_id] = NULL;
+ task = thread->cur_task[to_id];
+ task_base = task->va_base;
+ if ((task_base[task->num_cmd - 1] == CMDQ_JUMP_BY_OFFSET) &&
+ (task_base[task->num_cmd - 2] == CMDQ_JUMP_PASS)) {
+ break; /* reach the last task */
+ }
+ to_id = (to_id + 1) % CMDQ_MAX_TASK_IN_THREAD;
+ }
+}
+
+static struct cmdq_task *cmdq_task_acquire(struct cmdq_command *command,
+ struct cmdq_task_cb *cb)
+{
+ struct cmdq *cmdq = command->cmdq;
+ struct device *dev = cmdq->dev;
+ struct cmdq_task *task;
+
+ task = cmdq_task_create(cmdq, command->size);
+ if (!task) {
+ dev_err(dev, "can't acquire task info\n");
+ return NULL;
+ }
+ task->engine_flag = command->engine_flag;
+ task->task_state = TASK_STATE_WAITING;
+ if (cb)
+ task->cb = *cb;
+ else
+ memset(&task->cb, 0, sizeof(task->cb));
+
+ /* copy commands */
+ memcpy(task->va_base, command->base, command->size);
+ task->num_cmd = task->command_size / sizeof(u32);
+
+ /* insert into task waiting list */
+ mutex_lock(&cmdq->task_mutex);
+ list_add_tail(&task->list_entry, &cmdq->task_wait_list);
+ mutex_unlock(&cmdq->task_mutex);
+
+ return task;
+}
+
+static void cmdq_thread_writel(struct cmdq_thread *thread, u32 value,
+ u32 offset)
+{
+ writel(value, thread->base + offset);
+}
+
+static u32 cmdq_thread_readl(struct cmdq_thread *thread, u32 offset)
+{
+ return readl(thread->base + offset);
+}
+
+static u32 cmdq_thread_get_cookie(struct cmdq_thread *thread)
+{
+ return cmdq_thread_readl(thread, CMDQ_THR_EXEC_CNT_OFFSET) &
+ CMDQ_COOKIE_MASK;
+}
+
+static struct cmdq_thread *cmdq_thread_get(struct cmdq *cmdq, u64 flag)
+{
+ int tid = cmdq_eng_get_thread(flag);
+ struct cmdq_thread *thread = &cmdq->thread[tid];
+ u32 next_cookie;
+
+ next_cookie = thread->next_cookie % CMDQ_MAX_TASK_IN_THREAD;
+ if (thread->cur_task[next_cookie])
+ return NULL;
+ cmdq_clk_enable(cmdq);
+ return thread;
+}
+
+static void cmdq_thread_put(struct cmdq *cmdq, struct cmdq_thread *thread)
+{
+ if (WARN_ON(thread == NULL))
+ return;
+ cmdq_clk_disable(cmdq);
+}
+
+static int cmdq_thread_suspend(struct cmdq *cmdq, struct cmdq_thread *thread)
+{
+ u32 enabled;
+ u32 status;
+
+ /* write suspend bit */
+ cmdq_thread_writel(thread, CMDQ_THR_SUSPEND,
+ CMDQ_THR_SUSPEND_TASK_OFFSET);
+
+ /* If already disabled, treat as suspended successful. */
+ enabled = cmdq_thread_readl(thread, CMDQ_THR_ENABLE_TASK_OFFSET);
+ if (!(enabled & CMDQ_THR_ENABLED))
+ return 0;
+
+ /* poll suspended status */
+ if (readl_poll_timeout_atomic(thread->base +
+ CMDQ_THR_CURR_STATUS_OFFSET,
+ status,
+ status & CMDQ_THR_STATUS_SUSPENDED,
+ 0, 10)) {
+ dev_err(cmdq->dev, "Suspend HW thread 0x%x failed\n",
+ (u32)(thread->base - cmdq->base));
+ return -EFAULT;
+ }
+
+ return 0;
+}
+
+static void cmdq_thread_resume(struct cmdq_thread *thread)
+{
+ cmdq_thread_writel(thread, CMDQ_THR_RESUME,
+ CMDQ_THR_SUSPEND_TASK_OFFSET);
+}
+
+static int cmdq_thread_reset(struct cmdq *cmdq, struct cmdq_thread *thread)
+{
+ void __iomem *gce_base = cmdq->base;
+ u32 warm_reset;
+
+ cmdq_thread_writel(thread, CMDQ_THR_WARM_RESET,
+ CMDQ_THR_WARM_RESET_OFFSET);
+ if (readl_poll_timeout_atomic(thread->base + CMDQ_THR_WARM_RESET_OFFSET,
+ warm_reset,
+ !(warm_reset & CMDQ_THR_WARM_RESET),
+ 0, 10)) {
+ dev_err(cmdq->dev, "Reset HW thread 0x%x failed\n",
+ (u32)(thread->base - cmdq->base));
+ return -EFAULT;
+ }
+ writel(CMDQ_THR_SLOT_CYCLES, gce_base + CMDQ_THR_SLOT_CYCLES_OFFSET);
+ return 0;
+}
+
+static void cmdq_thread_disable(struct cmdq *cmdq, struct cmdq_thread *thread)
+{
+ cmdq_thread_reset(cmdq, thread);
+ cmdq_thread_writel(thread, CMDQ_THR_DISABLED,
+ CMDQ_THR_ENABLE_TASK_OFFSET);
+}
+
+static void cmdq_thread_insert_task_by_cookie(struct cmdq_thread *thread,
+ struct cmdq_task *task,
+ int cookie)
+{
+ thread->wait_cookie = cookie;
+ thread->next_cookie = cookie + 1;
+ if (thread->next_cookie > CMDQ_MAX_COOKIE_VALUE)
+ thread->next_cookie = 0;
+ thread->task_count = 1; /* fist task */
+ thread->cur_task[cookie % CMDQ_MAX_TASK_IN_THREAD] = task;
+}
+
+static int cmdq_thread_remove_task_by_index(struct cmdq_thread *thread,
+ int index, enum cmdq_task_state new_state)
+{
+ struct cmdq_task *task;
+
+ task = thread->cur_task[index];
+ if (!task) {
+ pr_err("%s: remove fail, task:%d on thread:0x%p is NULL\n",
+ __func__, index, thread);
+ return -EINVAL;
+ }
+ task->task_state = new_state;
+ thread->cur_task[index] = NULL;
+ thread->task_count--;
+ return 0;
+}
+
+static struct cmdq_task *cmdq_thread_search_task_by_pc(
+ const struct cmdq_thread *thread, u32 pc)
+{
+ struct cmdq_task *task;
+ int i;
+
+ for (i = 0; i < CMDQ_MAX_TASK_IN_THREAD; i++) {
+ task = thread->cur_task[i];
+ if (task &&
+ pc >= task->mva_base &&
+ pc < task->mva_base + task->command_size)
+ break;
+ }
+ return task;
+}
+
+/* notify GCE to re-fetch commands by setting HW thread PC */
+static void cmdq_thread_invalidate_fetched_data(struct cmdq_thread *thread)
+{
+ cmdq_thread_writel(thread,
+ cmdq_thread_readl(thread, CMDQ_THR_CURR_ADDR_OFFSET),
+ CMDQ_THR_CURR_ADDR_OFFSET);
+}
+
+static int cmdq_task_insert_into_thread(struct cmdq_task *task)
+{
+ struct cmdq_thread *thread = task->thread;
+ struct cmdq_task *prev_task;
+ int prev, curr;
+ u32 *prev_task_base;
+
+ /* find previous task */
+ curr = thread->next_cookie % CMDQ_MAX_TASK_IN_THREAD;
+ prev = (curr + CMDQ_MAX_TASK_IN_THREAD - 1) % CMDQ_MAX_TASK_IN_THREAD;
+ prev_task = thread->cur_task[prev];
+
+ /* cmdq_thread_reorder_task_array() should prevent this error */
+ if (WARN_ON(prev_task == NULL))
+ return -EFAULT;
+
+ /* insert task, and let previous task jump to this task */
+ thread->cur_task[curr] = task;
+ prev_task_base = prev_task->va_base;
+ prev_task_base[prev_task->num_cmd - 1] = CMDQ_JUMP_BY_PA;
+ prev_task_base[prev_task->num_cmd - 2] = task->mva_base;
+
+ /* re-fetch command buffer */
+ cmdq_thread_invalidate_fetched_data(thread);
+
+ return 0;
+}
+
+/* we assume tasks in the same display thread are waiting the same event. */
+static void cmdq_task_remove_wfe(struct cmdq_task *task)
+{
+ u32 wfe_option = CMDQ_WFE_UPDATE | CMDQ_WFE_WAIT | CMDQ_WFE_WAIT_VALUE;
+ u32 wfe_op = CMDQ_CODE_WFE << CMDQ_OP_CODE_SHIFT;
+ u32 *base = task->va_base;
+ int i;
+
+ for (i = 0; i < task->num_cmd; i += 2) {
+ if (base[i] == wfe_option &&
+ (base[i + 1] & CMDQ_OP_CODE_MASK) == wfe_op) {
+ base[i] = CMDQ_JUMP_PASS;
+ base[i + 1] = CMDQ_JUMP_BY_OFFSET;
+ }
+ }
+}
+
+static int cmdq_task_exec_async(struct cmdq_task *task,
+ struct cmdq_thread *thread)
+{
+ struct cmdq *cmdq = task->cmdq;
+ struct device *dev = cmdq->dev;
+ unsigned long flags;
+ int minimum;
+ int cookie;
+
+ spin_lock_irqsave(&cmdq->exec_lock, flags);
+ task->thread = thread;
+ task->task_state = TASK_STATE_BUSY;
+ if (thread->task_count <= 0) {
+ if (WARN_ON(cmdq_thread_reset(cmdq, thread) < 0)) {
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+ return -EFAULT;
+ }
+
+ cmdq_thread_writel(thread, CMDQ_THR_NO_TIMEOUT,
+ CMDQ_THR_INST_CYCLES_OFFSET);
+ cmdq_thread_writel(thread, task->mva_base,
+ CMDQ_THR_CURR_ADDR_OFFSET);
+ cmdq_thread_writel(thread, task->mva_base + task->command_size,
+ CMDQ_THR_END_ADDR_OFFSET);
+ cmdq_thread_writel(thread, CMDQ_THR_PRIORITY,
+ CMDQ_THR_CFG_OFFSET);
+ cmdq_thread_writel(thread, CMDQ_THR_IRQ_EN,
+ CMDQ_THR_IRQ_ENABLE_OFFSET);
+
+ minimum = cmdq_thread_get_cookie(thread);
+ cmdq_thread_insert_task_by_cookie(
+ thread, task, (minimum + 1));
+
+ /* enable HW thread */
+ cmdq_thread_writel(thread, CMDQ_THR_ENABLED,
+ CMDQ_THR_ENABLE_TASK_OFFSET);
+ } else {
+ unsigned long curr_pa, end_pa;
+ int err;
+
+ err = cmdq_thread_suspend(cmdq, thread);
+ if (WARN_ON(err < 0)) {
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+ return err;
+ }
+
+ cmdq_thread_writel(thread, CMDQ_THR_NO_TIMEOUT,
+ CMDQ_THR_INST_CYCLES_OFFSET);
+ cookie = thread->next_cookie;
+
+ /*
+ * check boundary condition
+ * PC = END - 8, EOC is executed
+ * PC = END - 0, all CMDs are executed
+ */
+ curr_pa = cmdq_thread_readl(thread, CMDQ_THR_CURR_ADDR_OFFSET);
+ end_pa = cmdq_thread_readl(thread, CMDQ_THR_END_ADDR_OFFSET);
+ if ((curr_pa == end_pa - 8) || (curr_pa == end_pa - 0)) {
+ /* set to task directly */
+ cmdq_thread_writel(thread, task->mva_base,
+ CMDQ_THR_CURR_ADDR_OFFSET);
+ cmdq_thread_writel(thread,
+ task->mva_base + task->command_size,
+ CMDQ_THR_END_ADDR_OFFSET);
+ thread->cur_task[cookie % CMDQ_MAX_TASK_IN_THREAD] =
+ task;
+ thread->task_count++;
+ } else {
+ err = cmdq_task_insert_into_thread(task);
+ if (err < 0) {
+ cmdq_thread_resume(thread);
+ spin_unlock_irqrestore(
+ &cmdq->exec_lock, flags);
+ dev_err(dev, "insert task failed\n");
+ return err;
+ }
+
+ if (thread == &cmdq->thread[0] ||
+ thread == &cmdq->thread[1])
+ cmdq_task_remove_wfe(task);
+
+ smp_mb(); /* modify jump before enable thread */
+ cmdq_thread_writel(thread,
+ task->mva_base + task->command_size,
+ CMDQ_THR_END_ADDR_OFFSET);
+ thread->task_count++;
+ }
+
+ thread->next_cookie = (thread->next_cookie + 1) %
+ CMDQ_MAX_COOKIE_VALUE;
+ cmdq_thread_resume(thread);
+ }
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+ return 0;
+}
+
+static void cmdq_handle_done(struct cmdq *cmdq, struct cmdq_thread *thread,
+ int cookie)
+{
+ int i, new_wait_cookie;
+ struct cmdq_task *task;
+ struct cmdq_cb_data cmdq_cb_data;
+
+ new_wait_cookie = (cookie + 1) % (CMDQ_MAX_COOKIE_VALUE + 1);
+ for (i = thread->wait_cookie % CMDQ_MAX_TASK_IN_THREAD;
+ i != new_wait_cookie % CMDQ_MAX_TASK_IN_THREAD;
+ i = (i + 1) % CMDQ_MAX_TASK_IN_THREAD) {
+ if (thread->cur_task[i]) {
+ task = thread->cur_task[i];
+ if (task->cb.cb) {
+ cmdq_cb_data.err = false;
+ cmdq_cb_data.data = task->cb.data;
+ task->cb.cb(cmdq_cb_data);
+ }
+ cmdq_thread_remove_task_by_index(
+ thread, i, TASK_STATE_DONE);
+ }
+ }
+ thread->wait_cookie = new_wait_cookie;
+ wake_up(&thread->wait_queue);
+}
+
+static void cmdq_handle_error(struct cmdq *cmdq, struct cmdq_thread *thread)
+{
+ struct device *dev = cmdq->dev;
+ u32 curr_pa = cmdq_thread_readl(thread, CMDQ_THR_CURR_ADDR_OFFSET);
+ u32 end_pa = cmdq_thread_readl(thread, CMDQ_THR_END_ADDR_OFFSET);
+ int cookie = cmdq_thread_get_cookie(thread);
+
+ dev_err(dev, "IRQ: error thread=0x%x\n",
+ (u32)(thread->base - cmdq->base));
+ dev_err(dev, "IRQ: Thread PC: 0x%08x, End PC:0x%08x\n",
+ curr_pa, end_pa);
+
+ /*
+ * We assume error happens BEFORE EOC
+ * because it wouldn't be error if this interrupt is issued by EOC.
+ * So, we should add 1 to locate "current" task.
+ */
+ cookie++;
+
+ /* set the issued task to error state */
+ if (thread->cur_task[cookie % CMDQ_MAX_TASK_IN_THREAD]) {
+ struct cmdq_task *task;
+ struct cmdq_cb_data cmdq_cb_data;
+
+ task = thread->cur_task[cookie % CMDQ_MAX_TASK_IN_THREAD];
+ /* still call cb to prevent lock */
+ if (task->cb.cb) {
+ cmdq_cb_data.err = true;
+ cmdq_cb_data.data = task->cb.data;
+ task->cb.cb(cmdq_cb_data);
+ }
+ cmdq_thread_remove_task_by_index(
+ thread, cookie % CMDQ_MAX_TASK_IN_THREAD,
+ TASK_STATE_ERROR);
+ } else {
+ dev_err(dev,
+ "IRQ: can not find task in %s, pc:0x%08x, end_pc:0x%08x\n",
+ __func__, curr_pa, end_pa);
+ if (thread->task_count <= 0) {
+ /*
+ * suspend HW thread first,
+ * so that we work in a consistent state
+ * outer function should acquire spinlock:
+ * cmdq->exec_lock
+ */
+ if (cmdq_thread_suspend(cmdq, thread) < 0)
+ dev_err(dev, "IRQ: suspend HW thread failed!");
+
+ cmdq_thread_disable(cmdq, thread);
+ dev_err(dev,
+ "IRQ: there is no task for thread (0x%x)\n",
+ (u32)(thread->base - cmdq->base));
+ }
+ }
+
+ /* set the remained tasks to done state */
+ cmdq_handle_done(cmdq, thread, cookie);
+}
+
+static void cmdq_handle_irq(struct cmdq *cmdq, int tid)
+{
+ struct device *dev = cmdq->dev;
+ struct cmdq_thread *thread = &cmdq->thread[tid];
+ unsigned long flags = 0L;
+ int value;
+ int enabled;
+
+ spin_lock_irqsave(&cmdq->exec_lock, flags);
+
+ /*
+ * it is possible for another CPU core
+ * to run "release task" right before we acquire the spin lock
+ * and thus reset / disable this HW thread
+ * so we check both the IRQ flag and the enable bit of this thread
+ */
+ value = cmdq_thread_readl(thread, CMDQ_THR_IRQ_STATUS_OFFSET);
+ if (!(value & CMDQ_THR_IRQ_MASK)) {
+ dev_err(dev,
+ "IRQ: thread 0x%x got interrupt but IRQ flag=0x%x\n",
+ (u32)(thread->base - cmdq->base), value);
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+ return;
+ }
+
+ enabled = cmdq_thread_readl(thread, CMDQ_THR_ENABLE_TASK_OFFSET);
+ if (!(enabled & CMDQ_THR_ENABLED)) {
+ dev_err(dev,
+ "IRQ: thread 0x%x got interrupt but enabled=0x%x\n",
+ (u32)(thread->base - cmdq->base), enabled);
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+ return;
+ }
+
+ /*
+ * Move the reset IRQ before read HW cookie
+ * to prevent race condition and save the cost of suspend
+ */
+ cmdq_thread_writel(thread, ~value, CMDQ_THR_IRQ_STATUS_OFFSET);
+
+ if (value & CMDQ_THR_IRQ_ERROR)
+ cmdq_handle_error(cmdq, thread);
+ else if (value & CMDQ_THR_IRQ_DONE)
+ cmdq_handle_done(cmdq, thread, cmdq_thread_get_cookie(thread));
+
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+}
+
+static void cmdq_consume_waiting_list(struct work_struct *work)
+{
+ struct cmdq *cmdq = container_of(work, struct cmdq,
+ task_consume_wait_queue_item);
+ struct device *dev = cmdq->dev;
+ struct cmdq_task *task, *tmp;
+
+ mutex_lock(&cmdq->task_mutex);
+
+ if (list_empty(&cmdq->task_wait_list)) {
+ mutex_unlock(&cmdq->task_mutex);
+ return;
+ }
+
+ list_for_each_entry_safe(task, tmp, &cmdq->task_wait_list, list_entry) {
+ struct cmdq_thread *thread = NULL;
+ int err;
+
+ thread = cmdq_thread_get(cmdq, task->engine_flag);
+ if (!thread) {
+ dev_warn(dev, "acquire thread fail. wait\n");
+ continue;
+ }
+ err = cmdq_task_exec_async(task, thread);
+ if (err < 0) {
+ dev_warn(dev, "start task fail. wait\n");
+ continue;
+ } else {
+ list_move_tail(&task->list_entry,
+ &cmdq->task_active_list);
+ }
+ }
+
+ /*
+ * Wake up waiting task(s) whether success or not
+ * because wake up condition will check task's thread.
+ * (cmdq_task_wait_and_release)
+ */
+ wake_up_all(&cmdq->thread_dispatch_queue);
+
+ mutex_unlock(&cmdq->task_mutex);
+}
+
+static int cmdq_task_submit_async(struct cmdq_command *command,
+ struct cmdq_task **task_out,
+ struct cmdq_task_cb *cb)
+{
+ struct cmdq *cmdq = command->cmdq;
+
+ /* creates a new task and put into tail of waiting list */
+ *task_out = cmdq_task_acquire(command, cb);
+ if (!(*task_out))
+ return -EFAULT;
+ /* Do consumption here to gain some time if HW thread is available. */
+ cmdq_consume_waiting_list(&cmdq->task_consume_wait_queue_item);
+ return 0;
+}
+
+static int cmdq_task_handle_error_result(struct cmdq_task *task)
+{
+ struct cmdq *cmdq = task->cmdq;
+ struct device *dev = cmdq->dev;
+ struct cmdq_thread *thread = task->thread;
+ int err, i, cookie;
+ struct cmdq_task *next_task;
+ unsigned long thread_pc;
+ u32 *prev_va, *curr_va;
+ u32 irq_flag;
+
+ dev_err(dev,
+ "task(0x%p) state is not TASK_STATE_DONE, but %d.\n",
+ task, task->task_state);
+
+ /* suspend HW thread to ensure consistency */
+ err = cmdq_thread_suspend(cmdq, thread);
+ if (WARN_ON(err < 0))
+ return err;
+
+ /* The cookie of the task currently being processed */
+ cookie = cmdq_thread_get_cookie(thread) + 1;
+ thread_pc = cmdq_thread_readl(thread, CMDQ_THR_CURR_ADDR_OFFSET);
+
+ /*
+ * Although IRQ is disabled, GCE continues to execute.
+ * It may have pending IRQ before HW thread is suspended,
+ * so check this condition again.
+ */
+ irq_flag = cmdq_thread_readl(thread, CMDQ_THR_IRQ_STATUS_OFFSET);
+ if (irq_flag & CMDQ_THR_IRQ_ERROR)
+ cmdq_handle_error(cmdq, thread);
+ else if (irq_flag & CMDQ_THR_IRQ_DONE)
+ cmdq_handle_done(cmdq, thread, cmdq_thread_get_cookie(thread));
+ cmdq_thread_writel(thread, ~irq_flag, CMDQ_THR_IRQ_STATUS_OFFSET);
+
+ if (task->task_state == TASK_STATE_DONE)
+ return 0; /* success after handling pending irq */
+
+ dev_err(dev, "task 0x%p timeout or killed\n", task);
+
+ /* If task is running, force to remove it. */
+ if (task->task_state == TASK_STATE_BUSY) {
+ for (i = 0; i < ARRAY_SIZE(thread->cur_task); i++) {
+ if (thread->cur_task[i] == task) {
+ cmdq_thread_remove_task_by_index(
+ thread, i, TASK_STATE_ERROR);
+ break;
+ }
+ }
+ }
+
+ /* find task's jump destination or no next task*/
+ next_task = NULL;
+ curr_va = task->va_base;
+ if (curr_va[task->num_cmd - 1] == CMDQ_JUMP_BY_PA)
+ next_task = cmdq_thread_search_task_by_pc(
+ thread, curr_va[task->num_cmd - 2]);
+
+ /* remove task from the chain of thread->cur_task. */
+ if (task->num_cmd && thread_pc >= task->mva_base &&
+ thread_pc < (task->mva_base + task->command_size)) {
+ if (next_task) {
+ /* cookie already +1 */
+ cmdq_thread_writel(thread, cookie,
+ CMDQ_THR_EXEC_CNT_OFFSET);
+ thread->wait_cookie = cookie + 1;
+ cmdq_thread_writel(thread, next_task->mva_base,
+ CMDQ_THR_CURR_ADDR_OFFSET);
+ }
+ } else {
+ for (i = 0; i < CMDQ_MAX_TASK_IN_THREAD; i++) {
+ struct cmdq_task *prev_task;
+ u32 prev_num, curr_num;
+
+ prev_task = thread->cur_task[i];
+ if (!prev_task)
+ continue;
+
+ prev_va = prev_task->va_base;
+ prev_num = prev_task->num_cmd;
+
+ curr_va = task->va_base;
+ curr_num = task->num_cmd;
+
+ /* find which task JUMP into task */
+ if (prev_va[prev_num - 2] == task->mva_base &&
+ prev_va[prev_num - 1] == CMDQ_JUMP_BY_PA) {
+ /* copy JUMP instruction */
+ prev_va[prev_num - 2] = curr_va[curr_num - 2];
+ prev_va[prev_num - 1] = curr_va[curr_num - 1];
+ /* reorder task array to prevent empty slot */
+ if (next_task)
+ cmdq_thread_reorder_task_array(
+ thread, i);
+ /* re-fetch command */
+ cmdq_thread_invalidate_fetched_data(thread);
+ break;
+ }
+ }
+ }
+ return -ECANCELED;
+}
+
+static int cmdq_task_wait_result(struct cmdq_task *task)
+{
+ struct cmdq *cmdq = task->cmdq;
+ struct cmdq_thread *thread = task->thread;
+ int err = 0;
+ unsigned long flags;
+
+ spin_lock_irqsave(&cmdq->exec_lock, flags);
+ if (task->task_state != TASK_STATE_DONE)
+ err = cmdq_task_handle_error_result(task);
+ if (thread->task_count <= 0)
+ cmdq_thread_disable(cmdq, thread);
+ else
+ cmdq_thread_resume(thread);
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+ return err;
+}
+
+static int cmdq_task_wait_done(struct cmdq_task *task)
+{
+ struct cmdq *cmdq = task->cmdq;
+ struct device *dev = cmdq->dev;
+ int wait_q;
+ unsigned long timeout = msecs_to_jiffies(
+ CMDQ_ACQUIRE_THREAD_TIMEOUT_MS);
+
+ /* wait for acquiring thread (cmdq_consume_waiting_list) */
+ wait_q = wait_event_timeout(
+ cmdq->thread_dispatch_queue,
+ task->thread, timeout);
+ if (!wait_q) {
+ mutex_lock(&cmdq->task_mutex);
+ /* Check if task was just consumed. */
+ if (!task->thread) {
+ dev_err(dev,
+ "task(0x%p) timeout with invalid thread\n",
+ task);
+ /*
+ * Remove from waiting list,
+ * so it won't be consumed in the future.
+ */
+ list_del_init(&task->list_entry);
+ mutex_unlock(&cmdq->task_mutex);
+ return -EINVAL;
+ }
+ /* valid thread, so we keep going */
+ mutex_unlock(&cmdq->task_mutex);
+ }
+
+ /* wait for execution */
+ wait_q = wait_event_timeout(task->thread->wait_queue,
+ (task->task_state != TASK_STATE_BUSY &&
+ task->task_state != TASK_STATE_WAITING),
+ msecs_to_jiffies(CMDQ_DEFAULT_TIMEOUT_MS));
+ if (!wait_q)
+ dev_dbg(dev, "timeout!\n");
+
+ /* wake up and continue */
+ return cmdq_task_wait_result(task);
+}
+
+static int cmdq_task_wait_and_release(struct cmdq_task *task)
+{
+ int err = cmdq_task_wait_done(task);
+
+ /* release regardless of success or not */
+ cmdq_thread_put(task->cmdq, task->thread);
+ cmdq_task_release_internal(task);
+ return err;
+}
+
+static void cmdq_auto_release(struct work_struct *work_item)
+{
+ struct cmdq_task *task = container_of(work_item, struct cmdq_task,
+ auto_release_work);
+ struct cmdq *cmdq = task->cmdq;
+ struct cmdq_task_cb cb = task->cb;
+ int err = cmdq_task_wait_and_release(task);
+ struct cmdq_cb_data cmdq_cb_data;
+
+ /* isr fail, so call cb here to prevent lock */
+ if (err < 0 && cb.cb) {
+ cmdq_cb_data.err = true;
+ cmdq_cb_data.data = cb.data;
+ cb.cb(cmdq_cb_data);
+ }
+
+ /* prevent no more flush or interrupt to consume waiting tasks */
+ if (err < 0)
+ queue_work(cmdq->task_consume_wq,
+ &cmdq->task_consume_wait_queue_item);
+}
+
+static void cmdq_task_auto_release(struct cmdq_task *task)
+{
+ struct cmdq *cmdq = task->cmdq;
+
+ /*
+ * the work item is embeded in task already
+ * but we need to initialized it
+ */
+ INIT_WORK(&task->auto_release_work, cmdq_auto_release);
+ queue_work(cmdq->task_auto_release_wq, &task->auto_release_work);
+}
+
+static int cmdq_task_submit(struct cmdq_command *command)
+{
+ struct device *dev = command->cmdq->dev;
+ int err;
+ struct cmdq_task *task;
+
+ err = cmdq_task_submit_async(command, &task, NULL);
+ if (err < 0) {
+ dev_err(dev, "cmdq_task_submit_async failed=%d\n", err);
+ return err;
+ }
+ err = cmdq_task_wait_and_release(task);
+ if (err < 0)
+ dev_err(dev, "task(0x%p) wait fail\n", task);
+ return err;
+}
+
+static int cmdq_remove(struct platform_device *pdev)
+{
+ struct cmdq *cmdq = platform_get_drvdata(pdev);
+ struct cmdq_task *task, *tmp;
+
+ /* Destroy work queue safely, so task_active_list must be empty. */
+ destroy_workqueue(cmdq->task_consume_wq);
+ cmdq->task_consume_wq = NULL;
+ destroy_workqueue(cmdq->task_auto_release_wq);
+ cmdq->task_auto_release_wq = NULL;
+
+ /* release task_wait_list */
+ list_for_each_entry_safe(task, tmp, &cmdq->task_wait_list, list_entry) {
+ cmdq_task_free_command_buffer(task);
+ list_del(&task->list_entry);
+ kmem_cache_free(cmdq->task_cache, task);
+ }
+
+ kmem_cache_destroy(cmdq->task_cache);
+ cmdq->task_cache = NULL;
+ return 0;
+}
+
+static irqreturn_t cmdq_irq_handler(int irq, void *dev)
+{
+ struct cmdq *cmdq = dev;
+ u32 irq_status;
+ int i;
+
+ irq_status = readl(cmdq->base + CMDQ_CURR_IRQ_STATUS_OFFSET);
+ irq_status &= CMDQ_IRQ_MASK;
+ irq_status ^= CMDQ_IRQ_MASK;
+
+ if (!irq_status)
+ return IRQ_NONE;
+
+ while (irq_status) {
+ i = ffs(irq_status) - 1;
+ irq_status &= ~BIT(i);
+ cmdq_handle_irq(cmdq, i);
+ }
+
+ queue_work(cmdq->task_consume_wq, &cmdq->task_consume_wait_queue_item);
+ return IRQ_HANDLED;
+}
+
+static int cmdq_initialize(struct cmdq *cmdq)
+{
+ int i;
+
+ mutex_init(&cmdq->task_mutex);
+ spin_lock_init(&cmdq->exec_lock);
+ init_waitqueue_head(&cmdq->thread_dispatch_queue);
+ cmdq->task_cache = kmem_cache_create(
+ CMDQ_DRIVER_DEVICE_NAME "_task",
+ sizeof(struct cmdq_task), __alignof__(struct cmdq_task),
+ SLAB_POISON | SLAB_HWCACHE_ALIGN | SLAB_RED_ZONE, NULL);
+ INIT_LIST_HEAD(&cmdq->task_active_list);
+ INIT_LIST_HEAD(&cmdq->task_wait_list);
+ INIT_WORK(&cmdq->task_consume_wait_queue_item,
+ cmdq_consume_waiting_list);
+ cmdq->task_auto_release_wq = alloc_ordered_workqueue(
+ "%s", WQ_MEM_RECLAIM | WQ_HIGHPRI, "cmdq_auto_release");
+ cmdq->task_consume_wq = alloc_ordered_workqueue(
+ "%s", WQ_MEM_RECLAIM | WQ_HIGHPRI, "cmdq_task");
+
+ /* initialize cmdq thread */
+ for (i = 0; i < ARRAY_SIZE(cmdq->thread) ; i++) {
+ cmdq->thread[i].base = cmdq->base + CMDQ_THR_BASE +
+ CMDQ_THR_SHIFT * i;
+ init_waitqueue_head(&cmdq->thread[i].wait_queue);
+ }
+
+ return 0;
+}
+
+static int cmdq_rec_realloc_cmd_buffer(struct cmdq_rec *rec, size_t size)
+{
+ void *new_buf = krealloc(rec->buf, size, GFP_KERNEL | __GFP_ZERO);
+
+ if (!new_buf)
+ return -ENOMEM;
+ rec->buf = new_buf;
+ rec->buf_size = size;
+ return 0;
+}
+
+struct cmdq_base *cmdq_register_device(struct device *dev)
+{
+ struct cmdq_base *cmdq_base;
+ struct resource res;
+ int subsys;
+ u32 base;
+
+ if (of_address_to_resource(dev->of_node, 0, &res))
+ return NULL;
+ base = (u32)res.start;
+
+ subsys = cmdq_subsys_base_addr_to_id(base >> 16);
+ if (subsys < 0)
+ return NULL;
+
+ cmdq_base = devm_kmalloc(dev, sizeof(*cmdq_base), GFP_KERNEL);
+ if (!cmdq_base)
+ return NULL;
+ cmdq_base->subsys = subsys;
+ cmdq_base->base = base;
+
+ return cmdq_base;
+}
+EXPORT_SYMBOL(cmdq_register_device);
+
+int cmdq_rec_create(struct device *dev, u64 engine_flag,
+ struct cmdq_rec **rec_ptr)
+{
+ struct cmdq_rec *rec;
+ int err;
+
+ rec = kzalloc(sizeof(*rec), GFP_KERNEL);
+ if (!rec)
+ return -ENOMEM;
+ rec->cmdq = dev_get_drvdata(dev);
+ rec->engine_flag = engine_flag;
+ err = cmdq_rec_realloc_cmd_buffer(rec, CMDQ_INITIAL_CMD_BLOCK_SIZE);
+ if (err < 0) {
+ kfree(rec);
+ return err;
+ }
+ *rec_ptr = rec;
+ return 0;
+}
+EXPORT_SYMBOL(cmdq_rec_create);
+
+static int cmdq_rec_append_command(struct cmdq_rec *rec, enum cmdq_code code,
+ u32 arg_a, u32 arg_b)
+{
+ u32 *cmd_ptr;
+ int err;
+
+ if (WARN_ON(rec->finalized))
+ return -EBUSY;
+ if (code < CMDQ_CODE_MOVE || code > CMDQ_CODE_EOC)
+ return -EINVAL;
+ if (unlikely(rec->command_size + CMDQ_INST_SIZE > rec->buf_size)) {
+ err = cmdq_rec_realloc_cmd_buffer(rec, rec->buf_size * 2);
+ if (err < 0)
+ return err;
+ }
+ cmd_ptr = (u32 *)(rec->buf + rec->command_size);
+ cmd_ptr[0] = arg_b;
+ cmd_ptr[1] = (code << CMDQ_OP_CODE_SHIFT) | arg_a;
+ rec->command_size += CMDQ_INST_SIZE;
+ return 0;
+}
+
+int cmdq_rec_write(struct cmdq_rec *rec, u32 value, struct cmdq_base *base,
+ u32 offset)
+{
+ u32 arg_a = ((base->base + offset) & CMDQ_ARG_A_WRITE_MASK) |
+ ((base->subsys & CMDQ_SUBSYS_MASK) << CMDQ_SUBSYS_SHIFT);
+
+ return cmdq_rec_append_command(rec, CMDQ_CODE_WRITE, arg_a, value);
+}
+EXPORT_SYMBOL(cmdq_rec_write);
+
+int cmdq_rec_write_mask(struct cmdq_rec *rec, u32 value,
+ struct cmdq_base *base, u32 offset, u32 mask)
+{
+ u32 offset_mask = offset;
+ int err;
+
+ if (mask != 0xffffffff) {
+ err = cmdq_rec_append_command(rec, CMDQ_CODE_MOVE, 0, ~mask);
+ if (err < 0)
+ return err;
+ offset_mask |= CMDQ_ENABLE_MASK;
+ }
+ return cmdq_rec_write(rec, value, base, offset_mask);
+}
+EXPORT_SYMBOL(cmdq_rec_write_mask);
+
+int cmdq_rec_wfe(struct cmdq_rec *rec, enum cmdq_event event)
+{
+ u32 arg_b;
+
+ if (event >= CMDQ_MAX_HW_EVENT_COUNT || event < 0)
+ return -EINVAL;
+
+ /*
+ * bit 0-11: wait value
+ * bit 15: 1 - wait, 0 - no wait
+ * bit 16-27: update value
+ * bit 31: 1 - update, 0 - no update
+ */
+ arg_b = CMDQ_WFE_UPDATE | CMDQ_WFE_WAIT | CMDQ_WFE_WAIT_VALUE;
+ return cmdq_rec_append_command(rec, CMDQ_CODE_WFE, event, arg_b);
+}
+EXPORT_SYMBOL(cmdq_rec_wfe);
+
+int cmdq_rec_clear_event(struct cmdq_rec *rec, enum cmdq_event event)
+{
+ if (event >= CMDQ_MAX_HW_EVENT_COUNT || event < 0)
+ return -EINVAL;
+
+ return cmdq_rec_append_command(rec, CMDQ_CODE_WFE, event,
+ CMDQ_WFE_UPDATE);
+}
+EXPORT_SYMBOL(cmdq_rec_clear_event);
+
+static int cmdq_rec_fill_command(struct cmdq_rec *rec,
+ struct cmdq_command *command)
+{
+ if (!rec->finalized) {
+ int err;
+
+ /* insert EOC and generate IRQ for each command iteration */
+ err = cmdq_rec_append_command(rec, CMDQ_CODE_EOC, 0,
+ CMDQ_EOC_IRQ_EN);
+ if (err < 0)
+ return err;
+
+ /* JUMP to end */
+ err = cmdq_rec_append_command(rec, CMDQ_CODE_JUMP, 0,
+ CMDQ_JUMP_PASS);
+ if (err < 0)
+ return err;
+
+ rec->finalized = true;
+ }
+
+ command->cmdq = rec->cmdq;
+ command->engine_flag = rec->engine_flag;
+ command->base = rec->buf;
+ command->size = rec->command_size;
+ return 0;
+}
+
+int cmdq_rec_flush(struct cmdq_rec *rec)
+{
+ int err;
+ struct cmdq_command command;
+
+ err = cmdq_rec_fill_command(rec, &command);
+ if (err < 0)
+ return err;
+ return cmdq_task_submit(&command);
+}
+EXPORT_SYMBOL(cmdq_rec_flush);
+
+int cmdq_rec_flush_async(struct cmdq_rec *rec, cmdq_async_flush_cb cb,
+ void *data)
+{
+ int err;
+ struct cmdq_command command;
+ struct cmdq_task *task;
+ struct cmdq_task_cb task_cb;
+
+ err = cmdq_rec_fill_command(rec, &command);
+ if (err < 0)
+ return err;
+ task_cb.cb = cb;
+ task_cb.data = data;
+ err = cmdq_task_submit_async(&command, &task, &task_cb);
+ if (err < 0)
+ return err;
+ cmdq_task_auto_release(task);
+ return 0;
+}
+EXPORT_SYMBOL(cmdq_rec_flush_async);
+
+void cmdq_rec_destroy(struct cmdq_rec *rec)
+{
+ kfree(rec->buf);
+ kfree(rec);
+}
+EXPORT_SYMBOL(cmdq_rec_destroy);
+
+static int cmdq_probe(struct platform_device *pdev)
+{
+ struct device *dev = &pdev->dev;
+ struct device_node *node = dev->of_node;
+ struct resource *res;
+ struct cmdq *cmdq;
+ int err;
+
+ cmdq = devm_kzalloc(dev, sizeof(*cmdq), GFP_KERNEL);
+ if (!cmdq)
+ return -ENOMEM;
+ cmdq->dev = dev;
+
+ res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ cmdq->base = devm_ioremap_resource(dev, res);
+ if (IS_ERR(cmdq->base)) {
+ dev_err(dev, "failed to ioremap gce\n");
+ return PTR_ERR(cmdq->base);
+ }
+
+ cmdq->irq = irq_of_parse_and_map(node, 0);
+ if (!cmdq->irq) {
+ dev_err(dev, "failed to get irq\n");
+ return -EINVAL;
+ }
+
+ dev_dbg(dev, "cmdq device: addr:0x%p, va:0x%p, irq:%d\n",
+ dev, cmdq->base, cmdq->irq);
+
+ /* init cmdq and save to device private data */
+ err = cmdq_initialize(cmdq);
+ if (err < 0) {
+ dev_err(dev, "failed to init cmdq\n");
+ return err;
+ }
+ platform_set_drvdata(pdev, cmdq);
+
+ err = devm_request_irq(dev, cmdq->irq, cmdq_irq_handler, IRQF_SHARED,
+ CMDQ_DRIVER_DEVICE_NAME, cmdq);
+ if (err < 0) {
+ dev_err(dev, "failed to register ISR (%d)\n", err);
+ goto fail;
+ }
+
+ cmdq->clock = devm_clk_get(dev, CMDQ_CLK_NAME);
+ if (IS_ERR(cmdq->clock)) {
+ dev_err(dev, "failed to get clk:%s\n", CMDQ_CLK_NAME);
+ err = PTR_ERR(cmdq->clock);
+ goto fail;
+ }
+ return 0;
+fail:
+ cmdq_remove(pdev);
+ return err;
+}
+
+static const struct of_device_id cmdq_of_ids[] = {
+ {.compatible = "mediatek,mt8173-gce",},
+ {}
+};
+
+static struct platform_driver cmdq_drv = {
+ .probe = cmdq_probe,
+ .remove = cmdq_remove,
+ .driver = {
+ .name = CMDQ_DRIVER_DEVICE_NAME,
+ .owner = THIS_MODULE,
+ .of_match_table = cmdq_of_ids,
+ }
+};
+
+builtin_platform_driver(cmdq_drv);
diff --git a/include/soc/mediatek/cmdq.h b/include/soc/mediatek/cmdq.h
new file mode 100644
index 0000000..60eef3d
--- /dev/null
+++ b/include/soc/mediatek/cmdq.h
@@ -0,0 +1,197 @@
+/*
+ * Copyright (c) 2015 MediaTek Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ */
+
+#ifndef __MTK_CMDQ_H__
+#define __MTK_CMDQ_H__
+
+#include <linux/platform_device.h>
+#include <linux/types.h>
+
+enum cmdq_eng {
+ CMDQ_ENG_DISP_AAL,
+ CMDQ_ENG_DISP_COLOR0,
+ CMDQ_ENG_DISP_COLOR1,
+ CMDQ_ENG_DISP_DPI0,
+ CMDQ_ENG_DISP_DSI0,
+ CMDQ_ENG_DISP_DSI1,
+ CMDQ_ENG_DISP_GAMMA,
+ CMDQ_ENG_DISP_OD,
+ CMDQ_ENG_DISP_OVL0,
+ CMDQ_ENG_DISP_OVL1,
+ CMDQ_ENG_DISP_PWM0,
+ CMDQ_ENG_DISP_PWM1,
+ CMDQ_ENG_DISP_RDMA0,
+ CMDQ_ENG_DISP_RDMA1,
+ CMDQ_ENG_DISP_RDMA2,
+ CMDQ_ENG_DISP_UFOE,
+ CMDQ_ENG_DISP_WDMA0,
+ CMDQ_ENG_DISP_WDMA1,
+ CMDQ_ENG_MAX,
+};
+
+/* events for CMDQ and display */
+enum cmdq_event {
+ /* Display start of frame(SOF) events */
+ CMDQ_EVENT_DISP_OVL0_SOF = 11,
+ CMDQ_EVENT_DISP_OVL1_SOF = 12,
+ CMDQ_EVENT_DISP_RDMA0_SOF = 13,
+ CMDQ_EVENT_DISP_RDMA1_SOF = 14,
+ CMDQ_EVENT_DISP_RDMA2_SOF = 15,
+ CMDQ_EVENT_DISP_WDMA0_SOF = 16,
+ CMDQ_EVENT_DISP_WDMA1_SOF = 17,
+ /* Display end of frame(EOF) events */
+ CMDQ_EVENT_DISP_OVL0_EOF = 39,
+ CMDQ_EVENT_DISP_OVL1_EOF = 40,
+ CMDQ_EVENT_DISP_RDMA0_EOF = 41,
+ CMDQ_EVENT_DISP_RDMA1_EOF = 42,
+ CMDQ_EVENT_DISP_RDMA2_EOF = 43,
+ CMDQ_EVENT_DISP_WDMA0_EOF = 44,
+ CMDQ_EVENT_DISP_WDMA1_EOF = 45,
+ /* Mutex end of frame(EOF) events */
+ CMDQ_EVENT_MUTEX0_STREAM_EOF = 53,
+ CMDQ_EVENT_MUTEX1_STREAM_EOF = 54,
+ CMDQ_EVENT_MUTEX2_STREAM_EOF = 55,
+ CMDQ_EVENT_MUTEX3_STREAM_EOF = 56,
+ CMDQ_EVENT_MUTEX4_STREAM_EOF = 57,
+ /* Display underrun events */
+ CMDQ_EVENT_DISP_RDMA0_UNDERRUN = 63,
+ CMDQ_EVENT_DISP_RDMA1_UNDERRUN = 64,
+ CMDQ_EVENT_DISP_RDMA2_UNDERRUN = 65,
+ /* Keep this at the end of HW events */
+ CMDQ_MAX_HW_EVENT_COUNT = 260,
+};
+
+struct cmdq_cb_data {
+ bool err;
+ void *data;
+};
+
+typedef int (*cmdq_async_flush_cb)(struct cmdq_cb_data data);
+
+struct cmdq_task;
+struct cmdq;
+
+struct cmdq_rec {
+ struct cmdq *cmdq;
+ u64 engine_flag;
+ size_t command_size;
+ void *buf;
+ size_t buf_size;
+ bool finalized;
+};
+
+struct cmdq_base {
+ int subsys;
+ u32 base;
+};
+
+/**
+ * cmdq_register_device() - register device which needs CMDQ
+ * @dev: device
+ *
+ * Return: cmdq_base pointer or NULL for failed
+ */
+struct cmdq_base *cmdq_register_device(struct device *dev);
+
+/**
+ * cmdq_rec_create() - create command queue record
+ * @dev: device
+ * @engine_flag: command queue engine flag
+ * @rec_ptr: command queue record pointer to retrieve cmdq_rec
+ *
+ * Return: 0 for success; else the error code is returned
+ */
+int cmdq_rec_create(struct device *dev, u64 engine_flag,
+ struct cmdq_rec **rec_ptr);
+
+/**
+ * cmdq_rec_write() - append write command to the command queue record
+ * @rec: the command queue record
+ * @value: the specified target register value
+ * @base: the command queue base
+ * @offset: register offset from module base
+ *
+ * Return: 0 for success; else the error code is returned
+ */
+int cmdq_rec_write(struct cmdq_rec *rec, u32 value,
+ struct cmdq_base *base, u32 offset);
+
+/**
+ * cmdq_rec_write_mask() - append write command with mask to the command
+ * queue record
+ * @rec: the command queue record
+ * @value: the specified target register value
+ * @base: the command queue base
+ * @offset: register offset from module base
+ * @mask: the specified target register mask
+ *
+ * Return: 0 for success; else the error code is returned
+ */
+int cmdq_rec_write_mask(struct cmdq_rec *rec, u32 value,
+ struct cmdq_base *base, u32 offset, u32 mask);
+
+/**
+ * cmdq_rec_wfe() - append wait for event command to the command queue record
+ * @rec: the command queue record
+ * @event: the desired event type to "wait and CLEAR"
+ *
+ * Return: 0 for success; else the error code is returned
+ */
+int cmdq_rec_wfe(struct cmdq_rec *rec, enum cmdq_event event);
+
+/**
+ * cmdq_rec_clear_event() - append clear event command to the command queue
+ * record
+ * @rec: the command queue record
+ * @event: the desired event to be cleared
+ *
+ * Return: 0 for success; else the error code is returned
+ */
+int cmdq_rec_clear_event(struct cmdq_rec *rec, enum cmdq_event event);
+
+/**
+ * cmdq_rec_flush() - trigger CMDQ to execute the recorded commands
+ * @rec: the command queue record
+ *
+ * Return: 0 for success; else the error code is returned
+ *
+ * Trigger CMDQ to execute the recorded commands. Note that this is a
+ * synchronous flush function. When the function returned, the recorded
+ * commands have been done.
+ */
+int cmdq_rec_flush(struct cmdq_rec *rec);
+
+/**
+ * cmdq_rec_flush_async() - trigger CMDQ to asynchronously execute the recorded
+ * commands and call back after ISR is finished
+ * @rec: the command queue record
+ * @cb: called in the end of CMDQ ISR
+ * @data: this data will pass back to cb
+ *
+ * Return: 0 for success; else the error code is returned
+ *
+ * Trigger CMDQ to asynchronously execute the recorded commands and call back
+ * after ISR is finished. Note that this is an ASYNC function. When the function
+ * returned, it may or may not be finished. The ISR callback function is called
+ * in the end of ISR.
+ */
+int cmdq_rec_flush_async(struct cmdq_rec *rec, cmdq_async_flush_cb cb,
+ void *data);
+
+/**
+ * cmdq_rec_destroy() - destroy command queue record
+ * @rec: the command queue record
+ */
+void cmdq_rec_destroy(struct cmdq_rec *rec);
+
+#endif /* __MTK_CMDQ_H__ */
--
1.9.1
^ permalink raw reply related [flat|nested] 5+ messages in thread* [RFC v5 4/4] CMDQ: suspend/resume protection
2016-05-04 12:32 [RFC v5 0/4] Mediatek MT8173 CMDQ support HS Liao
` (2 preceding siblings ...)
2016-05-04 12:32 ` [RFC v5 3/4] arm64: dts: mt8173: Add GCE node HS Liao
@ 2016-05-04 12:32 ` HS Liao
3 siblings, 0 replies; 5+ messages in thread
From: HS Liao @ 2016-05-04 12:32 UTC (permalink / raw)
To: Rob Herring, Matthias Brugger
Cc: Daniel Kurtz, Sascha Hauer, devicetree, linux-kernel,
linux-arm-kernel, linux-mediatek, srv_heupstream, Sascha Hauer,
Philipp Zabel, Nicolas Boichat, CK HU, cawa cheng, Bibby Hsieh,
YT Shen, Daoyuan Huang, Damon Chu, Josh-YC Liu, Glory Hung,
Jiaguang Zhang, HS Liao
Add suspend/resume protection mechanism to prevent active task(s) in
suspend.
Signed-off-by: HS Liao <hs.liao@mediatek.com>
---
drivers/soc/mediatek/mtk-cmdq.c | 295 +++++++++++++++++++++++++++++++++++++++-
1 file changed, 289 insertions(+), 6 deletions(-)
diff --git a/drivers/soc/mediatek/mtk-cmdq.c b/drivers/soc/mediatek/mtk-cmdq.c
index d3e9a4c..7599a5b 100644
--- a/drivers/soc/mediatek/mtk-cmdq.c
+++ b/drivers/soc/mediatek/mtk-cmdq.c
@@ -50,6 +50,7 @@
#define CMDQ_CLK_NAME "gce"
#define CMDQ_CURR_IRQ_STATUS_OFFSET 0x010
+#define CMDQ_CURR_LOADED_THR_OFFSET 0x018
#define CMDQ_THR_SLOT_CYCLES_OFFSET 0x030
#define CMDQ_THR_BASE 0x100
@@ -143,6 +144,7 @@ enum cmdq_code {
enum cmdq_task_state {
TASK_STATE_IDLE, /* free task */
TASK_STATE_BUSY, /* task running on a thread */
+ TASK_STATE_KILLED, /* task process being killed */
TASK_STATE_ERROR, /* task execution error */
TASK_STATE_DONE, /* task finished */
TASK_STATE_WAITING, /* allocated but waiting for available thread */
@@ -182,6 +184,8 @@ struct cmdq_thread {
struct cmdq {
struct device *dev;
+ struct notifier_block pm_notifier;
+
void __iomem *base;
u32 irq;
@@ -201,9 +205,12 @@ struct cmdq {
struct cmdq_thread thread[CMDQ_MAX_THREAD_COUNT];
struct mutex task_mutex; /* for task list */
+ atomic_t thread_usage;
spinlock_t exec_lock; /* for exec task */
wait_queue_head_t thread_dispatch_queue;
struct clk *clock;
+ atomic_t suspending;
+ bool suspended;
};
struct cmdq_subsys {
@@ -292,14 +299,24 @@ static struct cmdq_task *cmdq_task_create(struct cmdq *cmdq, size_t size)
return task;
}
-static void cmdq_task_release_internal(struct cmdq_task *task)
+static void cmdq_task_release_unlocked(struct cmdq_task *task)
{
struct cmdq *cmdq = task->cmdq;
- mutex_lock(&cmdq->task_mutex);
+ /* This func should be inside cmdq->task_mutex mutex */
+ lockdep_assert_held(&cmdq->task_mutex);
+
cmdq_task_free_command_buffer(task);
list_del(&task->list_entry);
kmem_cache_free(cmdq->task_cache, task);
+}
+
+static void cmdq_task_release_internal(struct cmdq_task *task)
+{
+ struct cmdq *cmdq = task->cmdq;
+
+ mutex_lock(&cmdq->task_mutex);
+ cmdq_task_release_unlocked(task);
mutex_unlock(&cmdq->task_mutex);
}
@@ -393,6 +410,7 @@ static struct cmdq_thread *cmdq_thread_get(struct cmdq *cmdq, u64 flag)
if (thread->cur_task[next_cookie])
return NULL;
cmdq_clk_enable(cmdq);
+ atomic_inc(&cmdq->thread_usage);
return thread;
}
@@ -401,6 +419,7 @@ static void cmdq_thread_put(struct cmdq *cmdq, struct cmdq_thread *thread)
if (WARN_ON(thread == NULL))
return;
cmdq_clk_disable(cmdq);
+ atomic_dec(&cmdq->thread_usage);
}
static int cmdq_thread_suspend(struct cmdq *cmdq, struct cmdq_thread *thread)
@@ -492,6 +511,80 @@ static int cmdq_thread_remove_task_by_index(struct cmdq_thread *thread,
return 0;
}
+static int cmdq_thread_force_remove_task(struct cmdq_task *task)
+{
+ struct cmdq *cmdq = task->cmdq;
+ struct cmdq_thread *thread = task->thread;
+ int err;
+ int cookie;
+ struct cmdq_task *exec_task;
+
+ err = cmdq_thread_suspend(cmdq, thread);
+
+ cmdq_thread_writel(thread, CMDQ_THR_NO_TIMEOUT,
+ CMDQ_THR_INST_CYCLES_OFFSET);
+
+ /* The cookie of the task currently being processed */
+ cookie = cmdq_thread_get_cookie(thread) + 1;
+
+ exec_task = thread->cur_task[cookie % CMDQ_MAX_TASK_IN_THREAD];
+ if (exec_task == task) {
+ dma_addr_t eoc_pa = task->mva_base + task->command_size - 16;
+
+ /* The task is executed now, set the PC to EOC for bypass */
+ cmdq_thread_writel(thread, eoc_pa, CMDQ_THR_CURR_ADDR_OFFSET);
+
+ thread->cur_task[cookie % CMDQ_MAX_TASK_IN_THREAD] = NULL;
+ task->task_state = TASK_STATE_KILLED;
+ } else {
+ int i, j;
+ u32 *task_base, *exec_task_base;
+
+ j = thread->task_count;
+ for (i = cookie; j > 0; j--, i++) {
+ i %= CMDQ_MAX_TASK_IN_THREAD;
+
+ exec_task = thread->cur_task[i];
+ if (!exec_task)
+ continue;
+
+ task_base = task->va_base;
+ exec_task_base = exec_task->va_base;
+ if ((exec_task_base[exec_task->num_cmd - 1] ==
+ CMDQ_JUMP_BY_OFFSET) &&
+ (exec_task_base[exec_task->num_cmd - 2] ==
+ CMDQ_JUMP_PASS)) {
+ /* reached the last task */
+ break;
+ }
+
+ if (exec_task_base[exec_task->num_cmd - 2] ==
+ task->mva_base) {
+ /* fake EOC command */
+ exec_task_base[exec_task->num_cmd - 2] =
+ CMDQ_EOC_IRQ_EN;
+ exec_task_base[exec_task->num_cmd - 1] =
+ CMDQ_CODE_EOC << CMDQ_OP_CODE_SHIFT;
+
+ /* bypass the task */
+ exec_task_base[exec_task->num_cmd] =
+ task_base[task->num_cmd - 2];
+ exec_task_base[exec_task->num_cmd + 1] =
+ task_base[task->num_cmd - 1];
+
+ i = (i + 1) % CMDQ_MAX_TASK_IN_THREAD;
+
+ thread->cur_task[i] = NULL;
+ task->task_state = TASK_STATE_KILLED;
+ err = 0;
+ break;
+ }
+ }
+ }
+
+ return err;
+}
+
static struct cmdq_task *cmdq_thread_search_task_by_pc(
const struct cmdq_thread *thread, u32 pc)
{
@@ -790,6 +883,18 @@ static void cmdq_handle_irq(struct cmdq *cmdq, int tid)
spin_unlock_irqrestore(&cmdq->exec_lock, flags);
}
+static int cmdq_resumed_notifier(struct cmdq *cmdq)
+{
+ /*
+ * during suspended, there may be queued tasks.
+ * we should process them if any.
+ */
+ queue_work(cmdq->task_consume_wq,
+ &cmdq->task_consume_wait_queue_item);
+
+ return 0;
+}
+
static void cmdq_consume_waiting_list(struct work_struct *work)
{
struct cmdq *cmdq = container_of(work, struct cmdq,
@@ -797,6 +902,12 @@ static void cmdq_consume_waiting_list(struct work_struct *work)
struct device *dev = cmdq->dev;
struct cmdq_task *task, *tmp;
+ /* do not execute any task after suspended */
+ if (cmdq->suspended) {
+ dev_warn(dev, "task is consumed after suspended\n");
+ return;
+ }
+
mutex_lock(&cmdq->task_mutex);
if (list_empty(&cmdq->task_wait_list)) {
@@ -960,6 +1071,14 @@ static int cmdq_task_wait_result(struct cmdq_task *task)
unsigned long flags;
spin_lock_irqsave(&cmdq->exec_lock, flags);
+
+ /* suspend, so just return */
+ if (atomic_read(&cmdq->suspending) &&
+ task->task_state == TASK_STATE_KILLED) {
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+ return 0;
+ }
+
if (task->task_state != TASK_STATE_DONE)
err = cmdq_task_handle_error_result(task);
if (thread->task_count <= 0)
@@ -1019,7 +1138,8 @@ static int cmdq_task_wait_and_release(struct cmdq_task *task)
/* release regardless of success or not */
cmdq_thread_put(task->cmdq, task->thread);
- cmdq_task_release_internal(task);
+ if (!atomic_read(&task->cmdq->suspending))
+ cmdq_task_release_internal(task);
return err;
}
@@ -1074,7 +1194,7 @@ static int cmdq_task_submit(struct cmdq_command *command)
return err;
}
-static int cmdq_remove(struct platform_device *pdev)
+static void cmdq_deinitialize(struct platform_device *pdev)
{
struct cmdq *cmdq = platform_get_drvdata(pdev);
struct cmdq_task *task, *tmp;
@@ -1094,7 +1214,6 @@ static int cmdq_remove(struct platform_device *pdev)
kmem_cache_destroy(cmdq->task_cache);
cmdq->task_cache = NULL;
- return 0;
}
static irqreturn_t cmdq_irq_handler(int irq, void *dev)
@@ -1316,6 +1435,13 @@ int cmdq_rec_flush(struct cmdq_rec *rec)
int err;
struct cmdq_command command;
+ /* do not allow flush after suspended */
+ if (rec->cmdq->suspended) {
+ dev_err(rec->cmdq->dev,
+ "%s is called after suspended\n", __func__);
+ return -EPERM;
+ }
+
err = cmdq_rec_fill_command(rec, &command);
if (err < 0)
return err;
@@ -1331,6 +1457,13 @@ int cmdq_rec_flush_async(struct cmdq_rec *rec, cmdq_async_flush_cb cb,
struct cmdq_task *task;
struct cmdq_task_cb task_cb;
+ /* do not allow flush after suspended */
+ if (rec->cmdq->suspended) {
+ dev_err(rec->cmdq->dev,
+ "%s is called after suspended\n", __func__);
+ return -EPERM;
+ }
+
err = cmdq_rec_fill_command(rec, &command);
if (err < 0)
return err;
@@ -1351,6 +1484,127 @@ void cmdq_rec_destroy(struct cmdq_rec *rec)
}
EXPORT_SYMBOL(cmdq_rec_destroy);
+static int cmdq_pm_notifier_cb(struct notifier_block *nb, unsigned long event,
+ void *ptr)
+{
+ struct cmdq *cmdq = container_of(nb, struct cmdq, pm_notifier);
+
+ switch (event) {
+ case PM_SUSPEND_PREPARE:
+ /*
+ * Going to suspend the system
+ * The next stage is freeze process.
+ * We will queue all request in suspend callback,
+ * so don't care this stage
+ */
+ return NOTIFY_DONE;
+ case PM_POST_SUSPEND:
+ /*
+ * processes had resumed in previous stage
+ * (system resume callback)
+ * resume CMDQ driver to execute.
+ */
+ cmdq_resumed_notifier(cmdq);
+ return NOTIFY_OK;
+ default:
+ return NOTIFY_DONE;
+ }
+ return NOTIFY_DONE;
+}
+
+static int cmdq_suspend(struct device *dev)
+{
+ struct cmdq *cmdq = dev_get_drvdata(dev);
+ u32 exec_threads = readl(cmdq->base + CMDQ_CURR_LOADED_THR_OFFSET);
+ int ref_count;
+ unsigned long flags;
+
+ /*
+ * lock to prevent cmdq_core_consume_waiting_list() and
+ * cmdq_core_acquire_task(), i.e. no new active tasks
+ */
+ mutex_lock(&cmdq->task_mutex);
+
+ ref_count = atomic_read(&cmdq->thread_usage);
+ atomic_set(&cmdq->suspending, 1);
+
+ if (ref_count > 0 || exec_threads & CMDQ_THR_EXECUTING) {
+ struct cmdq_task *task, *tmp;
+ int i;
+
+ dev_err(dev, "suspend: other running, kill tasks.\n");
+ dev_err(dev, "threads:0x%08x, ref:%d, AL empty:%d, base:0x%p\n",
+ exec_threads, ref_count,
+ list_empty(&cmdq->task_active_list), cmdq->base);
+
+ /*
+ * We need to ensure the system is ready to suspend,
+ * so kill all running CMDQ tasks and release HW engines.
+ */
+
+ /* remove all active task from thread */
+ list_for_each_entry_safe(task, tmp, &cmdq->task_active_list,
+ list_entry) {
+ bool already_done = false;
+
+ if (!task->thread)
+ continue;
+
+ spin_lock_irqsave(&cmdq->exec_lock, flags);
+ if (task->task_state == TASK_STATE_BUSY) {
+ /* still wait_event */
+ cmdq_thread_force_remove_task(task);
+ task->task_state = TASK_STATE_KILLED;
+ } else {
+ /* almost finish its work */
+ already_done = true;
+ }
+ spin_unlock_irqrestore(&cmdq->exec_lock, flags);
+
+ /*
+ * TASK_STATE_KILLED will unlock
+ * wait_event_timeout in cmdq_task_wait_done(),
+ * so flush_work to wait auto release flow.
+ *
+ * We don't know processes running order,
+ * so call cmdq_task_release_unlocked() here to
+ * prevent releasing task before flush_work, and
+ * also to prevent deadlock of task_mutex.
+ */
+ if (!already_done) {
+ flush_work(&task->auto_release_work);
+ cmdq_task_release_unlocked(task);
+ }
+ }
+ dev_err(dev, "suspend: AL empty:%d\n",
+ list_empty(&cmdq->task_active_list));
+
+ /* disable all HW thread */
+ dev_err(dev, "suspend: disable all HW threads\n");
+ for (i = 0; i < CMDQ_MAX_THREAD_COUNT; i++)
+ cmdq_thread_disable(cmdq, &cmdq->thread[i]);
+
+ /* reset all cmdq_thread */
+ memset(&cmdq->thread[0], 0, sizeof(cmdq->thread));
+ }
+
+ cmdq->suspended = true;
+ atomic_set(&cmdq->suspending, 0);
+
+ mutex_unlock(&cmdq->task_mutex);
+
+ /* ALWAYS allow suspend */
+ return 0;
+}
+
+static int cmdq_resume(struct device *dev)
+{
+ struct cmdq *cmdq = dev_get_drvdata(dev);
+
+ cmdq->suspended = false;
+ return 0;
+}
+
static int cmdq_probe(struct platform_device *pdev)
{
struct device *dev = &pdev->dev;
@@ -1401,12 +1655,40 @@ static int cmdq_probe(struct platform_device *pdev)
err = PTR_ERR(cmdq->clock);
goto fail;
}
+
+ /* hibernation and suspend events */
+ cmdq->pm_notifier.notifier_call = cmdq_pm_notifier_cb;
+ cmdq->pm_notifier.priority = 5;
+ err = register_pm_notifier(&cmdq->pm_notifier);
+ if (err < 0) {
+ dev_err(dev, "failed to register cmdq pm notifier\n");
+ goto fail;
+ }
+
return 0;
fail:
- cmdq_remove(pdev);
+ cmdq_deinitialize(pdev);
return err;
}
+static int cmdq_remove(struct platform_device *pdev)
+{
+ struct cmdq *cmdq = platform_get_drvdata(pdev);
+ int err;
+
+ err = unregister_pm_notifier(&cmdq->pm_notifier);
+ if (err < 0)
+ dev_err(&pdev->dev, "unregister pm notifier failed\n");
+
+ cmdq_deinitialize(pdev);
+ return 0;
+}
+
+static const struct dev_pm_ops cmdq_pm_ops = {
+ .suspend = cmdq_suspend,
+ .resume = cmdq_resume,
+};
+
static const struct of_device_id cmdq_of_ids[] = {
{.compatible = "mediatek,mt8173-gce",},
{}
@@ -1418,6 +1700,7 @@ static struct platform_driver cmdq_drv = {
.driver = {
.name = CMDQ_DRIVER_DEVICE_NAME,
.owner = THIS_MODULE,
+ .pm = &cmdq_pm_ops,
.of_match_table = cmdq_of_ids,
}
};
--
1.9.1
^ permalink raw reply related [flat|nested] 5+ messages in thread