diff mbox

[v22,2/4] mailbox: mediatek: Add Mediatek CMDQ driver

Message ID 1530098172-31385-3-git-send-email-houlong.wei@mediatek.com (mailing list archive)
State New, archived
Headers show

Commit Message

houlong.wei June 27, 2018, 11:16 a.m. UTC
This patch is first version of Mediatek Command Queue(CMDQ) driver. The
CMDQ is used to help 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: Houlong Wei <houlong.wei@mediatek.com>
Signed-off-by: HS Liao <hs.liao@mediatek.com>
Signed-off-by: CK Hu <ck.hu@mediatek.com>
---
 drivers/mailbox/Kconfig                  |   10 +
 drivers/mailbox/Makefile                 |    2 +
 drivers/mailbox/mtk-cmdq-mailbox.c       |  634 ++++++++++++++++++++++++++++++
 include/linux/mailbox/mtk-cmdq-mailbox.h |   70 ++++
 4 files changed, 716 insertions(+)
 create mode 100644 drivers/mailbox/mtk-cmdq-mailbox.c
 create mode 100644 include/linux/mailbox/mtk-cmdq-mailbox.h

Comments

CK Hu (胡俊光) June 29, 2018, 7:08 a.m. UTC | #1
Hi, Houlong:

Some inline comment.

On Wed, 2018-06-27 at 19:16 +0800, Houlong Wei wrote:
> This patch is first version of Mediatek Command Queue(CMDQ) driver. The
> CMDQ is used to help 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: Houlong Wei <houlong.wei@mediatek.com>
> Signed-off-by: HS Liao <hs.liao@mediatek.com>
> Signed-off-by: CK Hu <ck.hu@mediatek.com>
> ---
>  drivers/mailbox/Kconfig                  |   10 +
>  drivers/mailbox/Makefile                 |    2 +
>  drivers/mailbox/mtk-cmdq-mailbox.c       |  634 ++++++++++++++++++++++++++++++
>  include/linux/mailbox/mtk-cmdq-mailbox.h |   70 ++++
>  4 files changed, 716 insertions(+)
>  create mode 100644 drivers/mailbox/mtk-cmdq-mailbox.c
>  create mode 100644 include/linux/mailbox/mtk-cmdq-mailbox.h
> 

[...]

> +
> +static int cmdq_thread_reset(struct cmdq *cmdq, struct cmdq_thread *thread)
> +{
> +	u32 warm_reset;
> +
> +	writel(CMDQ_THR_DO_WARM_RESET, thread->base + CMDQ_THR_WARM_RESET);
> +	if (readl_poll_timeout_atomic(thread->base + CMDQ_THR_WARM_RESET,
> +			warm_reset, !(warm_reset & CMDQ_THR_DO_WARM_RESET),
> +			0, 10)) {
> +		dev_err(cmdq->mbox.dev, "reset GCE thread 0x%x failed\n",
> +			(u32)(thread->base - cmdq->base));
> +		return -EFAULT;
> +	}
> +	writel(CMDQ_THR_ACTIVE_SLOT_CYCLES, cmdq->base + CMDQ_THR_SLOT_CYCLES);

The CMDQ_THR_SLOT_CYCLES looks like not relevant to reset. Maybe you
just need to set this value when startup.

> +
> +	return 0;
> +}
> +

[...]

> +
> +static void cmdq_task_exec(struct cmdq_pkt *pkt, struct cmdq_thread *thread)
> +{
> +	struct cmdq *cmdq;
> +	struct cmdq_task *task;
> +	unsigned long curr_pa, end_pa;
> +
> +	cmdq = dev_get_drvdata(thread->chan->mbox->dev);
> +
> +	/* Client should not flush new tasks if suspended. */
> +	WARN_ON(cmdq->suspended);
> +
> +	task = kzalloc(sizeof(*task), GFP_ATOMIC);
> +	task->cmdq = cmdq;
> +	INIT_LIST_HEAD(&task->list_entry);
> +	task->pa_base = pkt->pa_base;
> +	task->thread = thread;
> +	task->pkt = pkt;
> +
> +	if (list_empty(&thread->task_busy_list)) {
> +		WARN_ON(clk_enable(cmdq->clock) < 0);
> +		WARN_ON(cmdq_thread_reset(cmdq, thread) < 0);
> +
> +		writel(task->pa_base, thread->base + CMDQ_THR_CURR_ADDR);
> +		writel(task->pa_base + pkt->cmd_buf_size,
> +		       thread->base + CMDQ_THR_END_ADDR);
> +		writel(thread->priority, thread->base + CMDQ_THR_PRIORITY);
> +		writel(CMDQ_THR_IRQ_EN, thread->base + CMDQ_THR_IRQ_ENABLE);
> +		writel(CMDQ_THR_ENABLED, thread->base + CMDQ_THR_ENABLE_TASK);
> +
> +		if (thread->timeout_ms != CMDQ_NO_TIMEOUT)
> +			mod_timer(&thread->timeout,
> +				jiffies + msecs_to_jiffies(thread->timeout_ms));

I think the timeout processing should be done by client driver. The
total time to execute a command buffer does not depend on GCE HW speed
because the WFE (wait for event) command would wait for client HW event,
so the total time depend on how long a client HW send this event to GCE
and the timeout processing should be client driver's job. Each client
may have different timeout processing mechanism, for example, if display
could dynamic change panel frame rate between 120Hz and 60Hz, and the
timeout time is 2 frame, so it may dynamically change timeout time
between 17ms and 33ms. Another reason is that display have interrupt
every vblank, and it could check timeout in that interrupt, so the timer
in cmdq driver looks redundant. Because each client would define its own
timeout processing mechanism, so it's not wise to put timeout processing
in cmdq driver.

> +	} else {
> +		WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> +		curr_pa = readl(thread->base + CMDQ_THR_CURR_ADDR);
> +		end_pa = readl(thread->base + CMDQ_THR_END_ADDR);
> +
> +		/*
> +		 * Atomic execution should remove the following wfe, i.e. only
> +		 * wait event at first task, and prevent to pause when running.
> +		 */
> +		if (thread->atomic_exec) {
> +			/* GCE is executing if command is not WFE */
> +			if (!cmdq_thread_is_in_wfe(thread)) {
> +				cmdq_thread_resume(thread);
> +				cmdq_thread_wait_end(thread, end_pa);
> +				WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> +				/* set to this task directly */
> +				writel(task->pa_base,
> +				       thread->base + CMDQ_THR_CURR_ADDR);
> +			} else {
> +				cmdq_task_insert_into_thread(task);
> +				cmdq_task_remove_wfe(task);
> +				smp_mb(); /* modify jump before enable thread */
> +			}
> +		} else {
> +			/* check boundary */
> +			if (curr_pa == end_pa - CMDQ_INST_SIZE ||
> +			    curr_pa == end_pa) {
> +				/* set to this task directly */
> +				writel(task->pa_base,
> +				       thread->base + CMDQ_THR_CURR_ADDR);
> +			} else {
> +				cmdq_task_insert_into_thread(task);
> +				smp_mb(); /* modify jump before enable thread */
> +			}
> +		}
> +		writel(task->pa_base + pkt->cmd_buf_size,
> +		       thread->base + CMDQ_THR_END_ADDR);
> +		cmdq_thread_resume(thread);
> +	}
> +	list_move_tail(&task->list_entry, &thread->task_busy_list);
> +}
> +
> +static void cmdq_task_exec_done(struct cmdq_task *task, bool err)
> +{
> +	struct device *dev = task->cmdq->mbox.dev;
> +	struct cmdq_cb_data cmdq_cb_data;
> +
> +	dma_unmap_single(dev, task->pa_base, task->pkt->cmd_buf_size,
> +			 DMA_TO_DEVICE);

Move this to client driver.

> +	if (task->pkt->cb.cb) {
> +		cmdq_cb_data.err = err;
> +		cmdq_cb_data.data = task->pkt->cb.data;
> +		task->pkt->cb.cb(cmdq_cb_data);
> +	}
> +	list_del(&task->list_entry);
> +}
> +

[...]

> +
> +static bool cmdq_mbox_last_tx_done(struct mbox_chan *chan)
> +{
> +	return true;
> +}
> +
> +static const struct mbox_chan_ops cmdq_mbox_chan_ops = {
> +	.send_data = cmdq_mbox_send_data,
> +	.startup = cmdq_mbox_startup,
> +	.shutdown = cmdq_mbox_shutdown,
> +	.last_tx_done = cmdq_mbox_last_tx_done,

Because mbox->txdone_poll is false, so you need not to implement
last_tx_done.

Regards,
CK

> +};
> +
> +static struct mbox_chan *cmdq_xlate(struct mbox_controller *mbox,
> +		const struct of_phandle_args *sp)
> +{
> +	int ind = sp->args[0];
> +	struct cmdq_thread *thread;
> +
> +	if (ind >= mbox->num_chans)
> +		return ERR_PTR(-EINVAL);
> +
> +	thread = mbox->chans[ind].con_priv;
> +	thread->timeout_ms = sp->args[1];
> +	thread->priority = sp->args[2];
> +	thread->atomic_exec = (sp->args[3] != 0);
> +	thread->chan = &mbox->chans[ind];
> +
> +	return &mbox->chans[ind];
> +}
> +
[...]
houlong.wei July 4, 2018, 12:10 a.m. UTC | #2
On Fri, 2018-06-29 at 15:08 +0800, CK Hu wrote:
> Hi, Houlong:
> 
> Some inline comment.
> 
> On Wed, 2018-06-27 at 19:16 +0800, Houlong Wei wrote:
> > This patch is first version of Mediatek Command Queue(CMDQ) driver. The
> > CMDQ is used to help 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: Houlong Wei <houlong.wei@mediatek.com>
> > Signed-off-by: HS Liao <hs.liao@mediatek.com>
> > Signed-off-by: CK Hu <ck.hu@mediatek.com>
> > ---
> >  drivers/mailbox/Kconfig                  |   10 +
> >  drivers/mailbox/Makefile                 |    2 +
> >  drivers/mailbox/mtk-cmdq-mailbox.c       |  634 ++++++++++++++++++++++++++++++
> >  include/linux/mailbox/mtk-cmdq-mailbox.h |   70 ++++
> >  4 files changed, 716 insertions(+)
> >  create mode 100644 drivers/mailbox/mtk-cmdq-mailbox.c
> >  create mode 100644 include/linux/mailbox/mtk-cmdq-mailbox.h
> > 
> 
> [...]
> 
> > +
> > +static int cmdq_thread_reset(struct cmdq *cmdq, struct cmdq_thread *thread)
> > +{
> > +	u32 warm_reset;
> > +
> > +	writel(CMDQ_THR_DO_WARM_RESET, thread->base + CMDQ_THR_WARM_RESET);
> > +	if (readl_poll_timeout_atomic(thread->base + CMDQ_THR_WARM_RESET,
> > +			warm_reset, !(warm_reset & CMDQ_THR_DO_WARM_RESET),
> > +			0, 10)) {
> > +		dev_err(cmdq->mbox.dev, "reset GCE thread 0x%x failed\n",
> > +			(u32)(thread->base - cmdq->base));
> > +		return -EFAULT;
> > +	}
> > +	writel(CMDQ_THR_ACTIVE_SLOT_CYCLES, cmdq->base + CMDQ_THR_SLOT_CYCLES);
> 
> The CMDQ_THR_SLOT_CYCLES looks like not relevant to reset. Maybe you
> just need to set this value when startup.

Will move configuration of CMDQ_THR_SLOT_CYCLES to cmdq_xlate() where is
startup of a GCE thread.

> 
> > +
> > +	return 0;
> > +}
> > +
> 
> [...]
> 
> > +
> > +static void cmdq_task_exec(struct cmdq_pkt *pkt, struct cmdq_thread *thread)
> > +{
> > +	struct cmdq *cmdq;
> > +	struct cmdq_task *task;
> > +	unsigned long curr_pa, end_pa;
> > +
> > +	cmdq = dev_get_drvdata(thread->chan->mbox->dev);
> > +
> > +	/* Client should not flush new tasks if suspended. */
> > +	WARN_ON(cmdq->suspended);
> > +
> > +	task = kzalloc(sizeof(*task), GFP_ATOMIC);
> > +	task->cmdq = cmdq;
> > +	INIT_LIST_HEAD(&task->list_entry);
> > +	task->pa_base = pkt->pa_base;
> > +	task->thread = thread;
> > +	task->pkt = pkt;
> > +
> > +	if (list_empty(&thread->task_busy_list)) {
> > +		WARN_ON(clk_enable(cmdq->clock) < 0);
> > +		WARN_ON(cmdq_thread_reset(cmdq, thread) < 0);
> > +
> > +		writel(task->pa_base, thread->base + CMDQ_THR_CURR_ADDR);
> > +		writel(task->pa_base + pkt->cmd_buf_size,
> > +		       thread->base + CMDQ_THR_END_ADDR);
> > +		writel(thread->priority, thread->base + CMDQ_THR_PRIORITY);
> > +		writel(CMDQ_THR_IRQ_EN, thread->base + CMDQ_THR_IRQ_ENABLE);
> > +		writel(CMDQ_THR_ENABLED, thread->base + CMDQ_THR_ENABLE_TASK);
> > +
> > +		if (thread->timeout_ms != CMDQ_NO_TIMEOUT)
> > +			mod_timer(&thread->timeout,
> > +				jiffies + msecs_to_jiffies(thread->timeout_ms));
> 
> I think the timeout processing should be done by client driver. The
> total time to execute a command buffer does not depend on GCE HW speed
> because the WFE (wait for event) command would wait for client HW event,
> so the total time depend on how long a client HW send this event to GCE
> and the timeout processing should be client driver's job. Each client
> may have different timeout processing mechanism, for example, if display
> could dynamic change panel frame rate between 120Hz and 60Hz, and the
> timeout time is 2 frame, so it may dynamically change timeout time
> between 17ms and 33ms. Another reason is that display have interrupt
> every vblank, and it could check timeout in that interrupt, so the timer
> in cmdq driver looks redundant. Because each client would define its own
> timeout processing mechanism, so it's not wise to put timeout processing
> in cmdq driver.

The client drivers' owners strongly hope to keep the current timeout
mechanism, the reasons are below.
1. If remove, all clients should add timeout mechanism and the code will
be redundant.
2. If timeout happens, only GCE driver can do reset and continue to
execute next packet.

> 
> > +	} else {
> > +		WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> > +		curr_pa = readl(thread->base + CMDQ_THR_CURR_ADDR);
> > +		end_pa = readl(thread->base + CMDQ_THR_END_ADDR);
> > +
> > +		/*
> > +		 * Atomic execution should remove the following wfe, i.e. only
> > +		 * wait event at first task, and prevent to pause when running.
> > +		 */
> > +		if (thread->atomic_exec) {
> > +			/* GCE is executing if command is not WFE */
> > +			if (!cmdq_thread_is_in_wfe(thread)) {
> > +				cmdq_thread_resume(thread);
> > +				cmdq_thread_wait_end(thread, end_pa);
> > +				WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> > +				/* set to this task directly */
> > +				writel(task->pa_base,
> > +				       thread->base + CMDQ_THR_CURR_ADDR);
> > +			} else {
> > +				cmdq_task_insert_into_thread(task);
> > +				cmdq_task_remove_wfe(task);
> > +				smp_mb(); /* modify jump before enable thread */
> > +			}
> > +		} else {
> > +			/* check boundary */
> > +			if (curr_pa == end_pa - CMDQ_INST_SIZE ||
> > +			    curr_pa == end_pa) {
> > +				/* set to this task directly */
> > +				writel(task->pa_base,
> > +				       thread->base + CMDQ_THR_CURR_ADDR);
> > +			} else {
> > +				cmdq_task_insert_into_thread(task);
> > +				smp_mb(); /* modify jump before enable thread */
> > +			}
> > +		}
> > +		writel(task->pa_base + pkt->cmd_buf_size,
> > +		       thread->base + CMDQ_THR_END_ADDR);
> > +		cmdq_thread_resume(thread);
> > +	}
> > +	list_move_tail(&task->list_entry, &thread->task_busy_list);
> > +}
> > +
> > +static void cmdq_task_exec_done(struct cmdq_task *task, bool err)
> > +{
> > +	struct device *dev = task->cmdq->mbox.dev;
> > +	struct cmdq_cb_data cmdq_cb_data;
> > +
> > +	dma_unmap_single(dev, task->pa_base, task->pkt->cmd_buf_size,
> > +			 DMA_TO_DEVICE);
> 
> Move this to client driver.

map/unmap are common code for clients driver, could we move it to cmdq
helper?

> 
> > +	if (task->pkt->cb.cb) {
> > +		cmdq_cb_data.err = err;
> > +		cmdq_cb_data.data = task->pkt->cb.data;
> > +		task->pkt->cb.cb(cmdq_cb_data);
> > +	}
> > +	list_del(&task->list_entry);
> > +}
> > +
> 
> [...]
> 
> > +
> > +static bool cmdq_mbox_last_tx_done(struct mbox_chan *chan)
> > +{
> > +	return true;
> > +}
> > +
> > +static const struct mbox_chan_ops cmdq_mbox_chan_ops = {
> > +	.send_data = cmdq_mbox_send_data,
> > +	.startup = cmdq_mbox_startup,
> > +	.shutdown = cmdq_mbox_shutdown,
> > +	.last_tx_done = cmdq_mbox_last_tx_done,
> 
> Because mbox->txdone_poll is false, so you need not to implement
> last_tx_done.
> 
> Regards,
> CK

Will remove cmdq_mbox_last_tx_done().

> 
> > +};
> > +
> > +static struct mbox_chan *cmdq_xlate(struct mbox_controller *mbox,
> > +		const struct of_phandle_args *sp)
> > +{
> > +	int ind = sp->args[0];
> > +	struct cmdq_thread *thread;
> > +
> > +	if (ind >= mbox->num_chans)
> > +		return ERR_PTR(-EINVAL);
> > +
> > +	thread = mbox->chans[ind].con_priv;
> > +	thread->timeout_ms = sp->args[1];
> > +	thread->priority = sp->args[2];
> > +	thread->atomic_exec = (sp->args[3] != 0);
> > +	thread->chan = &mbox->chans[ind];
> > +
> > +	return &mbox->chans[ind];
> > +}
> > +
> [...]
> 
>
CK Hu (胡俊光) July 4, 2018, 9:03 a.m. UTC | #3
Hi, Houlong:

On Wed, 2018-07-04 at 08:10 +0800, houlong wei wrote:
> On Fri, 2018-06-29 at 15:08 +0800, CK Hu wrote:
> > Hi, Houlong:
> > 
> > Some inline comment.
> > 
> > On Wed, 2018-06-27 at 19:16 +0800, Houlong Wei wrote:
> > > This patch is first version of Mediatek Command Queue(CMDQ) driver. The
> > > CMDQ is used to help 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: Houlong Wei <houlong.wei@mediatek.com>
> > > Signed-off-by: HS Liao <hs.liao@mediatek.com>
> > > Signed-off-by: CK Hu <ck.hu@mediatek.com>
> > > ---
> > >  drivers/mailbox/Kconfig                  |   10 +
> > >  drivers/mailbox/Makefile                 |    2 +
> > >  drivers/mailbox/mtk-cmdq-mailbox.c       |  634 ++++++++++++++++++++++++++++++
> > >  include/linux/mailbox/mtk-cmdq-mailbox.h |   70 ++++
> > >  4 files changed, 716 insertions(+)
> > >  create mode 100644 drivers/mailbox/mtk-cmdq-mailbox.c
> > >  create mode 100644 include/linux/mailbox/mtk-cmdq-mailbox.h
> > > 
> > 
> > [...]
> > 
> > > +
> > > +static int cmdq_thread_reset(struct cmdq *cmdq, struct cmdq_thread *thread)
> > > +{
> > > +	u32 warm_reset;
> > > +
> > > +	writel(CMDQ_THR_DO_WARM_RESET, thread->base + CMDQ_THR_WARM_RESET);
> > > +	if (readl_poll_timeout_atomic(thread->base + CMDQ_THR_WARM_RESET,
> > > +			warm_reset, !(warm_reset & CMDQ_THR_DO_WARM_RESET),
> > > +			0, 10)) {
> > > +		dev_err(cmdq->mbox.dev, "reset GCE thread 0x%x failed\n",
> > > +			(u32)(thread->base - cmdq->base));
> > > +		return -EFAULT;
> > > +	}
> > > +	writel(CMDQ_THR_ACTIVE_SLOT_CYCLES, cmdq->base + CMDQ_THR_SLOT_CYCLES);
> > 
> > The CMDQ_THR_SLOT_CYCLES looks like not relevant to reset. Maybe you
> > just need to set this value when startup.
> 
> Will move configuration of CMDQ_THR_SLOT_CYCLES to cmdq_xlate() where is
> startup of a GCE thread.
> 
> > 
> > > +
> > > +	return 0;
> > > +}
> > > +
> > 
> > [...]
> > 
> > > +
> > > +static void cmdq_task_exec(struct cmdq_pkt *pkt, struct cmdq_thread *thread)
> > > +{
> > > +	struct cmdq *cmdq;
> > > +	struct cmdq_task *task;
> > > +	unsigned long curr_pa, end_pa;
> > > +
> > > +	cmdq = dev_get_drvdata(thread->chan->mbox->dev);
> > > +
> > > +	/* Client should not flush new tasks if suspended. */
> > > +	WARN_ON(cmdq->suspended);
> > > +
> > > +	task = kzalloc(sizeof(*task), GFP_ATOMIC);
> > > +	task->cmdq = cmdq;
> > > +	INIT_LIST_HEAD(&task->list_entry);
> > > +	task->pa_base = pkt->pa_base;
> > > +	task->thread = thread;
> > > +	task->pkt = pkt;
> > > +
> > > +	if (list_empty(&thread->task_busy_list)) {
> > > +		WARN_ON(clk_enable(cmdq->clock) < 0);
> > > +		WARN_ON(cmdq_thread_reset(cmdq, thread) < 0);
> > > +
> > > +		writel(task->pa_base, thread->base + CMDQ_THR_CURR_ADDR);
> > > +		writel(task->pa_base + pkt->cmd_buf_size,
> > > +		       thread->base + CMDQ_THR_END_ADDR);
> > > +		writel(thread->priority, thread->base + CMDQ_THR_PRIORITY);
> > > +		writel(CMDQ_THR_IRQ_EN, thread->base + CMDQ_THR_IRQ_ENABLE);
> > > +		writel(CMDQ_THR_ENABLED, thread->base + CMDQ_THR_ENABLE_TASK);
> > > +
> > > +		if (thread->timeout_ms != CMDQ_NO_TIMEOUT)
> > > +			mod_timer(&thread->timeout,
> > > +				jiffies + msecs_to_jiffies(thread->timeout_ms));
> > 
> > I think the timeout processing should be done by client driver. The
> > total time to execute a command buffer does not depend on GCE HW speed
> > because the WFE (wait for event) command would wait for client HW event,
> > so the total time depend on how long a client HW send this event to GCE
> > and the timeout processing should be client driver's job. Each client
> > may have different timeout processing mechanism, for example, if display
> > could dynamic change panel frame rate between 120Hz and 60Hz, and the
> > timeout time is 2 frame, so it may dynamically change timeout time
> > between 17ms and 33ms. Another reason is that display have interrupt
> > every vblank, and it could check timeout in that interrupt, so the timer
> > in cmdq driver looks redundant. Because each client would define its own
> > timeout processing mechanism, so it's not wise to put timeout processing
> > in cmdq driver.
> 
> The client drivers' owners strongly hope to keep the current timeout
> mechanism, the reasons are below.
> 1. If remove, all clients should add timeout mechanism and the code will
> be redundant.
> 2. If timeout happens, only GCE driver can do reset and continue to
> execute next packet.

For the reason 2, GCE should not continue execute next packet because
the packets may have dependency. So GCE driver could only drop all
packet (this is what you do in cmdq_thread_handle_timeout()). For reason
1, you have a assumption that all client have the same request for
timeout: constant timeout value or no timeout. If it's so, that's ok to
put timeout mechanism in cmdq driver. But if one day, a new request for
timeout, for example, dynamic timeout value, that means not all client
driver implement timeout in the same way, putting timeout mechanism in
cmdq driver does not reduce any thing but just move multiple client
driver's code into cmdq driver. I would accept to put here only if all
client driver use the same timeout mechanism.

Regards,
CK

> 
> > 
> > > +	} else {
> > > +		WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> > > +		curr_pa = readl(thread->base + CMDQ_THR_CURR_ADDR);
> > > +		end_pa = readl(thread->base + CMDQ_THR_END_ADDR);
> > > +
> > > +		/*
> > > +		 * Atomic execution should remove the following wfe, i.e. only
> > > +		 * wait event at first task, and prevent to pause when running.
> > > +		 */
> > > +		if (thread->atomic_exec) {
> > > +			/* GCE is executing if command is not WFE */
> > > +			if (!cmdq_thread_is_in_wfe(thread)) {
> > > +				cmdq_thread_resume(thread);
> > > +				cmdq_thread_wait_end(thread, end_pa);
> > > +				WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> > > +				/* set to this task directly */
> > > +				writel(task->pa_base,
> > > +				       thread->base + CMDQ_THR_CURR_ADDR);
> > > +			} else {
> > > +				cmdq_task_insert_into_thread(task);
> > > +				cmdq_task_remove_wfe(task);
> > > +				smp_mb(); /* modify jump before enable thread */
> > > +			}
> > > +		} else {
> > > +			/* check boundary */
> > > +			if (curr_pa == end_pa - CMDQ_INST_SIZE ||
> > > +			    curr_pa == end_pa) {
> > > +				/* set to this task directly */
> > > +				writel(task->pa_base,
> > > +				       thread->base + CMDQ_THR_CURR_ADDR);
> > > +			} else {
> > > +				cmdq_task_insert_into_thread(task);
> > > +				smp_mb(); /* modify jump before enable thread */
> > > +			}
> > > +		}
> > > +		writel(task->pa_base + pkt->cmd_buf_size,
> > > +		       thread->base + CMDQ_THR_END_ADDR);
> > > +		cmdq_thread_resume(thread);
> > > +	}
> > > +	list_move_tail(&task->list_entry, &thread->task_busy_list);
> > > +}
> > > +
> > > +static void cmdq_task_exec_done(struct cmdq_task *task, bool err)
> > > +{
> > > +	struct device *dev = task->cmdq->mbox.dev;
> > > +	struct cmdq_cb_data cmdq_cb_data;
> > > +
> > > +	dma_unmap_single(dev, task->pa_base, task->pkt->cmd_buf_size,
> > > +			 DMA_TO_DEVICE);
> > 
> > Move this to client driver.
> 
> map/unmap are common code for clients driver, could we move it to cmdq
> helper?
> 
> > 
> > > +	if (task->pkt->cb.cb) {
> > > +		cmdq_cb_data.err = err;
> > > +		cmdq_cb_data.data = task->pkt->cb.data;
> > > +		task->pkt->cb.cb(cmdq_cb_data);
> > > +	}
> > > +	list_del(&task->list_entry);
> > > +}
> > > +
> > 
> > [...]
> > 
> > > +
> > > +static bool cmdq_mbox_last_tx_done(struct mbox_chan *chan)
> > > +{
> > > +	return true;
> > > +}
> > > +
> > > +static const struct mbox_chan_ops cmdq_mbox_chan_ops = {
> > > +	.send_data = cmdq_mbox_send_data,
> > > +	.startup = cmdq_mbox_startup,
> > > +	.shutdown = cmdq_mbox_shutdown,
> > > +	.last_tx_done = cmdq_mbox_last_tx_done,
> > 
> > Because mbox->txdone_poll is false, so you need not to implement
> > last_tx_done.
> > 
> > Regards,
> > CK
> 
> Will remove cmdq_mbox_last_tx_done().
> 
> > 
> > > +};
> > > +
> > > +static struct mbox_chan *cmdq_xlate(struct mbox_controller *mbox,
> > > +		const struct of_phandle_args *sp)
> > > +{
> > > +	int ind = sp->args[0];
> > > +	struct cmdq_thread *thread;
> > > +
> > > +	if (ind >= mbox->num_chans)
> > > +		return ERR_PTR(-EINVAL);
> > > +
> > > +	thread = mbox->chans[ind].con_priv;
> > > +	thread->timeout_ms = sp->args[1];
> > > +	thread->priority = sp->args[2];
> > > +	thread->atomic_exec = (sp->args[3] != 0);
> > > +	thread->chan = &mbox->chans[ind];
> > > +
> > > +	return &mbox->chans[ind];
> > > +}
> > > +
> > [...]
> > 
> > 
> 
>
houlong.wei July 6, 2018, 2:04 a.m. UTC | #4
On Wed, 2018-07-04 at 17:03 +0800, CK Hu wrote:
> Hi, Houlong:
> 
> On Wed, 2018-07-04 at 08:10 +0800, houlong wei wrote:
> > On Fri, 2018-06-29 at 15:08 +0800, CK Hu wrote:
> > > Hi, Houlong:
> > > 
> > > Some inline comment.
> > > 
> > > On Wed, 2018-06-27 at 19:16 +0800, Houlong Wei wrote:
> > > > This patch is first version of Mediatek Command Queue(CMDQ) driver. The
> > > > CMDQ is used to help 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: Houlong Wei <houlong.wei@mediatek.com>
> > > > Signed-off-by: HS Liao <hs.liao@mediatek.com>
> > > > Signed-off-by: CK Hu <ck.hu@mediatek.com>
> > > > ---
> > > >  drivers/mailbox/Kconfig                  |   10 +
> > > >  drivers/mailbox/Makefile                 |    2 +
> > > >  drivers/mailbox/mtk-cmdq-mailbox.c       |  634 ++++++++++++++++++++++++++++++
> > > >  include/linux/mailbox/mtk-cmdq-mailbox.h |   70 ++++
> > > >  4 files changed, 716 insertions(+)
> > > >  create mode 100644 drivers/mailbox/mtk-cmdq-mailbox.c
> > > >  create mode 100644 include/linux/mailbox/mtk-cmdq-mailbox.h
> > > > 
> > > 
> > > [...]
> > > 
> > > > +
> > > > +static int cmdq_thread_reset(struct cmdq *cmdq, struct cmdq_thread *thread)
> > > > +{
> > > > +	u32 warm_reset;
> > > > +
> > > > +	writel(CMDQ_THR_DO_WARM_RESET, thread->base + CMDQ_THR_WARM_RESET);
> > > > +	if (readl_poll_timeout_atomic(thread->base + CMDQ_THR_WARM_RESET,
> > > > +			warm_reset, !(warm_reset & CMDQ_THR_DO_WARM_RESET),
> > > > +			0, 10)) {
> > > > +		dev_err(cmdq->mbox.dev, "reset GCE thread 0x%x failed\n",
> > > > +			(u32)(thread->base - cmdq->base));
> > > > +		return -EFAULT;
> > > > +	}
> > > > +	writel(CMDQ_THR_ACTIVE_SLOT_CYCLES, cmdq->base + CMDQ_THR_SLOT_CYCLES);
> > > 
> > > The CMDQ_THR_SLOT_CYCLES looks like not relevant to reset. Maybe you
> > > just need to set this value when startup.
> > 
> > Will move configuration of CMDQ_THR_SLOT_CYCLES to cmdq_xlate() where is
> > startup of a GCE thread.
> > 

Since cmdq_xlate() is called when a client requests a channel, it may be
called more than once. Will move it to cmdq_probe().

> > > 
> > > > +
> > > > +	return 0;
> > > > +}
> > > > +
> > > 
> > > [...]
> > > 
> > > > +
> > > > +static void cmdq_task_exec(struct cmdq_pkt *pkt, struct cmdq_thread *thread)
> > > > +{
> > > > +	struct cmdq *cmdq;
> > > > +	struct cmdq_task *task;
> > > > +	unsigned long curr_pa, end_pa;
> > > > +
> > > > +	cmdq = dev_get_drvdata(thread->chan->mbox->dev);
> > > > +
> > > > +	/* Client should not flush new tasks if suspended. */
> > > > +	WARN_ON(cmdq->suspended);
> > > > +
> > > > +	task = kzalloc(sizeof(*task), GFP_ATOMIC);
> > > > +	task->cmdq = cmdq;
> > > > +	INIT_LIST_HEAD(&task->list_entry);
> > > > +	task->pa_base = pkt->pa_base;
> > > > +	task->thread = thread;
> > > > +	task->pkt = pkt;
> > > > +
> > > > +	if (list_empty(&thread->task_busy_list)) {
> > > > +		WARN_ON(clk_enable(cmdq->clock) < 0);
> > > > +		WARN_ON(cmdq_thread_reset(cmdq, thread) < 0);
> > > > +
> > > > +		writel(task->pa_base, thread->base + CMDQ_THR_CURR_ADDR);
> > > > +		writel(task->pa_base + pkt->cmd_buf_size,
> > > > +		       thread->base + CMDQ_THR_END_ADDR);
> > > > +		writel(thread->priority, thread->base + CMDQ_THR_PRIORITY);
> > > > +		writel(CMDQ_THR_IRQ_EN, thread->base + CMDQ_THR_IRQ_ENABLE);
> > > > +		writel(CMDQ_THR_ENABLED, thread->base + CMDQ_THR_ENABLE_TASK);
> > > > +
> > > > +		if (thread->timeout_ms != CMDQ_NO_TIMEOUT)
> > > > +			mod_timer(&thread->timeout,
> > > > +				jiffies + msecs_to_jiffies(thread->timeout_ms));
> > > 
> > > I think the timeout processing should be done by client driver. The
> > > total time to execute a command buffer does not depend on GCE HW speed
> > > because the WFE (wait for event) command would wait for client HW event,
> > > so the total time depend on how long a client HW send this event to GCE
> > > and the timeout processing should be client driver's job. Each client
> > > may have different timeout processing mechanism, for example, if display
> > > could dynamic change panel frame rate between 120Hz and 60Hz, and the
> > > timeout time is 2 frame, so it may dynamically change timeout time
> > > between 17ms and 33ms. Another reason is that display have interrupt
> > > every vblank, and it could check timeout in that interrupt, so the timer
> > > in cmdq driver looks redundant. Because each client would define its own
> > > timeout processing mechanism, so it's not wise to put timeout processing
> > > in cmdq driver.
> > 
> > The client drivers' owners strongly hope to keep the current timeout
> > mechanism, the reasons are below.
> > 1. If remove, all clients should add timeout mechanism and the code will
> > be redundant.
> > 2. If timeout happens, only GCE driver can do reset and continue to
> > execute next packet.
> 
> For the reason 2, GCE should not continue execute next packet because
> the packets may have dependency. So GCE driver could only drop all
> packet (this is what you do in cmdq_thread_handle_timeout()). For reason
> 1, you have a assumption that all client have the same request for
> timeout: constant timeout value or no timeout. If it's so, that's ok to
> put timeout mechanism in cmdq driver. But if one day, a new request for
> timeout, for example, dynamic timeout value, that means not all client
> driver implement timeout in the same way, putting timeout mechanism in
> cmdq driver does not reduce any thing but just move multiple client
> driver's code into cmdq driver. I would accept to put here only if all
> client driver use the same timeout mechanism.

The client drivers configure their timeout values via 'mboxes' properity
in device tree. So far, they handle timeout in same way.

> 
> Regards,
> CK
> 
> > 
> > > 
> > > > +	} else {
> > > > +		WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> > > > +		curr_pa = readl(thread->base + CMDQ_THR_CURR_ADDR);
> > > > +		end_pa = readl(thread->base + CMDQ_THR_END_ADDR);
> > > > +
> > > > +		/*
> > > > +		 * Atomic execution should remove the following wfe, i.e. only
> > > > +		 * wait event at first task, and prevent to pause when running.
> > > > +		 */
> > > > +		if (thread->atomic_exec) {
> > > > +			/* GCE is executing if command is not WFE */
> > > > +			if (!cmdq_thread_is_in_wfe(thread)) {
> > > > +				cmdq_thread_resume(thread);
> > > > +				cmdq_thread_wait_end(thread, end_pa);
> > > > +				WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
> > > > +				/* set to this task directly */
> > > > +				writel(task->pa_base,
> > > > +				       thread->base + CMDQ_THR_CURR_ADDR);
> > > > +			} else {
> > > > +				cmdq_task_insert_into_thread(task);
> > > > +				cmdq_task_remove_wfe(task);
> > > > +				smp_mb(); /* modify jump before enable thread */
> > > > +			}
> > > > +		} else {
> > > > +			/* check boundary */
> > > > +			if (curr_pa == end_pa - CMDQ_INST_SIZE ||
> > > > +			    curr_pa == end_pa) {
> > > > +				/* set to this task directly */
> > > > +				writel(task->pa_base,
> > > > +				       thread->base + CMDQ_THR_CURR_ADDR);
> > > > +			} else {
> > > > +				cmdq_task_insert_into_thread(task);
> > > > +				smp_mb(); /* modify jump before enable thread */
> > > > +			}
> > > > +		}
> > > > +		writel(task->pa_base + pkt->cmd_buf_size,
> > > > +		       thread->base + CMDQ_THR_END_ADDR);
> > > > +		cmdq_thread_resume(thread);
> > > > +	}
> > > > +	list_move_tail(&task->list_entry, &thread->task_busy_list);
> > > > +}
> > > > +
> > > > +static void cmdq_task_exec_done(struct cmdq_task *task, bool err)
> > > > +{
> > > > +	struct device *dev = task->cmdq->mbox.dev;
> > > > +	struct cmdq_cb_data cmdq_cb_data;
> > > > +
> > > > +	dma_unmap_single(dev, task->pa_base, task->pkt->cmd_buf_size,
> > > > +			 DMA_TO_DEVICE);
> > > 
> > > Move this to client driver.
> > 
> > map/unmap are common code for clients driver, could we move it to cmdq
> > helper?
> > 
> > > 
> > > > +	if (task->pkt->cb.cb) {
> > > > +		cmdq_cb_data.err = err;
> > > > +		cmdq_cb_data.data = task->pkt->cb.data;
> > > > +		task->pkt->cb.cb(cmdq_cb_data);
> > > > +	}
> > > > +	list_del(&task->list_entry);
> > > > +}
> > > > +
> > > 
> > > [...]
> > > 
> > > > +
> > > > +static bool cmdq_mbox_last_tx_done(struct mbox_chan *chan)
> > > > +{
> > > > +	return true;
> > > > +}
> > > > +
> > > > +static const struct mbox_chan_ops cmdq_mbox_chan_ops = {
> > > > +	.send_data = cmdq_mbox_send_data,
> > > > +	.startup = cmdq_mbox_startup,
> > > > +	.shutdown = cmdq_mbox_shutdown,
> > > > +	.last_tx_done = cmdq_mbox_last_tx_done,
> > > 
> > > Because mbox->txdone_poll is false, so you need not to implement
> > > last_tx_done.
> > > 
> > > Regards,
> > > CK
> > 
> > Will remove cmdq_mbox_last_tx_done().
> > 
> > > 
> > > > +};
> > > > +
> > > > +static struct mbox_chan *cmdq_xlate(struct mbox_controller *mbox,
> > > > +		const struct of_phandle_args *sp)
> > > > +{
> > > > +	int ind = sp->args[0];
> > > > +	struct cmdq_thread *thread;
> > > > +
> > > > +	if (ind >= mbox->num_chans)
> > > > +		return ERR_PTR(-EINVAL);
> > > > +
> > > > +	thread = mbox->chans[ind].con_priv;
> > > > +	thread->timeout_ms = sp->args[1];
> > > > +	thread->priority = sp->args[2];
> > > > +	thread->atomic_exec = (sp->args[3] != 0);
> > > > +	thread->chan = &mbox->chans[ind];
> > > > +
> > > > +	return &mbox->chans[ind];
> > > > +}
> > > > +
> > > [...]
> > > 
> > > 
> > 
> > 
> 
>
diff mbox

Patch

diff --git a/drivers/mailbox/Kconfig b/drivers/mailbox/Kconfig
index e63d29a..2bbabc9 100644
--- a/drivers/mailbox/Kconfig
+++ b/drivers/mailbox/Kconfig
@@ -189,4 +189,14 @@  config STM32_IPCC
 	  Mailbox implementation for STMicroelectonics STM32 family chips
 	  with hardware for Inter-Processor Communication Controller (IPCC)
 	  between processors. Say Y here if you want to have this support.
+
+config MTK_CMDQ_MBOX
+	tristate "MediaTek CMDQ Mailbox Support"
+	depends on ARCH_MEDIATEK || COMPILE_TEST
+	select MTK_INFRACFG
+	help
+	  Say yes here to add support for the MediaTek Command Queue (CMDQ)
+	  mailbox driver. The CMDQ is used to help read/write registers with
+	  critical time limitation, such as updating display configuration
+	  during the vblank.
 endif
diff --git a/drivers/mailbox/Makefile b/drivers/mailbox/Makefile
index 4d501be..4b00804 100644
--- a/drivers/mailbox/Makefile
+++ b/drivers/mailbox/Makefile
@@ -40,3 +40,5 @@  obj-$(CONFIG_QCOM_APCS_IPC)	+= qcom-apcs-ipc-mailbox.o
 obj-$(CONFIG_TEGRA_HSP_MBOX)	+= tegra-hsp.o
 
 obj-$(CONFIG_STM32_IPCC) 	+= stm32-ipcc.o
+
+obj-$(CONFIG_MTK_CMDQ_MBOX)	+= mtk-cmdq-mailbox.o
diff --git a/drivers/mailbox/mtk-cmdq-mailbox.c b/drivers/mailbox/mtk-cmdq-mailbox.c
new file mode 100644
index 0000000..93d87cb
--- /dev/null
+++ b/drivers/mailbox/mtk-cmdq-mailbox.c
@@ -0,0 +1,634 @@ 
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2018 MediaTek Inc.
+ *
+ */
+
+#include <linux/bitops.h>
+#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/module.h>
+#include <linux/platform_device.h>
+#include <linux/mailbox_controller.h>
+#include <linux/mailbox/mtk-cmdq-mailbox.h>
+#include <linux/timer.h>
+
+#define CMDQ_OP_CODE_MASK		(0xff << CMDQ_OP_CODE_SHIFT)
+#define CMDQ_TIMEOUT_MS			1000
+#define CMDQ_IRQ_MASK			0xffff
+#define CMDQ_NUM_CMD(t)			(t->cmd_buf_size / CMDQ_INST_SIZE)
+
+#define CMDQ_CURR_IRQ_STATUS		0x10
+#define CMDQ_THR_SLOT_CYCLES		0x30
+#define CMDQ_THR_BASE			0x100
+#define CMDQ_THR_SIZE			0x80
+#define CMDQ_THR_WARM_RESET		0x00
+#define CMDQ_THR_ENABLE_TASK		0x04
+#define CMDQ_THR_SUSPEND_TASK		0x08
+#define CMDQ_THR_CURR_STATUS		0x0c
+#define CMDQ_THR_IRQ_STATUS		0x10
+#define CMDQ_THR_IRQ_ENABLE		0x14
+#define CMDQ_THR_CURR_ADDR		0x20
+#define CMDQ_THR_END_ADDR		0x24
+#define CMDQ_THR_WAIT_TOKEN		0x30
+#define CMDQ_THR_PRIORITY		0x40
+
+#define CMDQ_NO_TIMEOUT			0xffffffffu
+#define CMDQ_THR_ACTIVE_SLOT_CYCLES	0x3200
+#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_DO_WARM_RESET		BIT(0)
+#define CMDQ_THR_IRQ_DONE		0x1
+#define CMDQ_THR_IRQ_ERROR		0x12
+#define CMDQ_THR_IRQ_EN			(CMDQ_THR_IRQ_ERROR | CMDQ_THR_IRQ_DONE)
+#define CMDQ_THR_IS_WAITING		BIT(31)
+
+#define CMDQ_JUMP_BY_OFFSET		0x10000000
+#define CMDQ_JUMP_BY_PA			0x10000001
+
+struct cmdq_thread {
+	struct mbox_chan	*chan;
+	void __iomem		*base;
+	struct list_head	task_busy_list;
+	struct timer_list	timeout;
+	u32			timeout_ms;
+	u32			priority;
+	bool			atomic_exec;
+};
+
+struct cmdq_task {
+	struct cmdq		*cmdq;
+	struct list_head	list_entry;
+	dma_addr_t		pa_base;
+	struct cmdq_thread	*thread;
+	struct cmdq_pkt		*pkt; /* the packet sent from mailbox client */
+};
+
+struct cmdq {
+	struct mbox_controller	mbox;
+	void __iomem		*base;
+	u32			irq;
+	u32			thread_nr;
+	struct cmdq_thread	*thread;
+	struct clk		*clock;
+	bool			suspended;
+};
+
+static int cmdq_thread_suspend(struct cmdq *cmdq, struct cmdq_thread *thread)
+{
+	u32 status;
+
+	writel(CMDQ_THR_SUSPEND, thread->base + CMDQ_THR_SUSPEND_TASK);
+
+	/* If already disabled, treat as suspended successful. */
+	if (!(readl(thread->base + CMDQ_THR_ENABLE_TASK) & CMDQ_THR_ENABLED))
+		return 0;
+
+	if (readl_poll_timeout_atomic(thread->base + CMDQ_THR_CURR_STATUS,
+			status, status & CMDQ_THR_STATUS_SUSPENDED, 0, 10)) {
+		dev_err(cmdq->mbox.dev, "suspend GCE thread 0x%x failed\n",
+			(u32)(thread->base - cmdq->base));
+		return -EFAULT;
+	}
+
+	return 0;
+}
+
+static void cmdq_thread_resume(struct cmdq_thread *thread)
+{
+	writel(CMDQ_THR_RESUME, thread->base + CMDQ_THR_SUSPEND_TASK);
+}
+
+static int cmdq_thread_reset(struct cmdq *cmdq, struct cmdq_thread *thread)
+{
+	u32 warm_reset;
+
+	writel(CMDQ_THR_DO_WARM_RESET, thread->base + CMDQ_THR_WARM_RESET);
+	if (readl_poll_timeout_atomic(thread->base + CMDQ_THR_WARM_RESET,
+			warm_reset, !(warm_reset & CMDQ_THR_DO_WARM_RESET),
+			0, 10)) {
+		dev_err(cmdq->mbox.dev, "reset GCE thread 0x%x failed\n",
+			(u32)(thread->base - cmdq->base));
+		return -EFAULT;
+	}
+	writel(CMDQ_THR_ACTIVE_SLOT_CYCLES, cmdq->base + CMDQ_THR_SLOT_CYCLES);
+
+	return 0;
+}
+
+static void cmdq_thread_disable(struct cmdq *cmdq, struct cmdq_thread *thread)
+{
+	cmdq_thread_reset(cmdq, thread);
+	writel(CMDQ_THR_DISABLED, thread->base + CMDQ_THR_ENABLE_TASK);
+}
+
+/* notify GCE to re-fetch commands by setting GCE thread PC */
+static void cmdq_thread_invalidate_fetched_data(struct cmdq_thread *thread)
+{
+	writel(readl(thread->base + CMDQ_THR_CURR_ADDR),
+	       thread->base + CMDQ_THR_CURR_ADDR);
+}
+
+static void cmdq_task_insert_into_thread(struct cmdq_task *task)
+{
+	struct device *dev = task->cmdq->mbox.dev;
+	struct cmdq_thread *thread = task->thread;
+	struct cmdq_task *prev_task = list_last_entry(
+			&thread->task_busy_list, typeof(*task), list_entry);
+	u64 *prev_task_base = prev_task->pkt->va_base;
+
+	/* let previous task jump to this task */
+	dma_sync_single_for_cpu(dev, prev_task->pa_base,
+				prev_task->pkt->cmd_buf_size, DMA_TO_DEVICE);
+	prev_task_base[CMDQ_NUM_CMD(prev_task->pkt) - 1] =
+		(u64)CMDQ_JUMP_BY_PA << 32 | task->pa_base;
+	dma_sync_single_for_device(dev, prev_task->pa_base,
+				   prev_task->pkt->cmd_buf_size, DMA_TO_DEVICE);
+
+	cmdq_thread_invalidate_fetched_data(thread);
+}
+
+static bool cmdq_command_is_wfe(u64 cmd)
+{
+	u64 wfe_option = CMDQ_WFE_UPDATE | CMDQ_WFE_WAIT | CMDQ_WFE_WAIT_VALUE;
+	u64 wfe_op = (u64)(CMDQ_CODE_WFE << CMDQ_OP_CODE_SHIFT) << 32;
+	u64 wfe_mask = (u64)CMDQ_OP_CODE_MASK << 32 | 0xffffffff;
+
+	return ((cmd & wfe_mask) == (wfe_op | wfe_option));
+}
+
+/* we assume tasks in the same display GCE thread are waiting the same event. */
+static void cmdq_task_remove_wfe(struct cmdq_task *task)
+{
+	struct device *dev = task->cmdq->mbox.dev;
+	u64 *base = task->pkt->va_base;
+	int i;
+
+	dma_sync_single_for_cpu(dev, task->pa_base, task->pkt->cmd_buf_size,
+				DMA_TO_DEVICE);
+	for (i = 0; i < CMDQ_NUM_CMD(task->pkt); i++)
+		if (cmdq_command_is_wfe(base[i]))
+			base[i] = (u64)CMDQ_JUMP_BY_OFFSET << 32 |
+				  CMDQ_JUMP_PASS;
+	dma_sync_single_for_device(dev, task->pa_base, task->pkt->cmd_buf_size,
+				   DMA_TO_DEVICE);
+}
+
+static bool cmdq_thread_is_in_wfe(struct cmdq_thread *thread)
+{
+	return readl(thread->base + CMDQ_THR_WAIT_TOKEN) & CMDQ_THR_IS_WAITING;
+}
+
+static void cmdq_thread_wait_end(struct cmdq_thread *thread,
+				 unsigned long end_pa)
+{
+	struct device *dev = thread->chan->mbox->dev;
+	unsigned long curr_pa;
+
+	if (readl_poll_timeout_atomic(thread->base + CMDQ_THR_CURR_ADDR,
+			curr_pa, curr_pa == end_pa, 1, 20))
+		dev_err(dev, "GCE thread cannot run to end.\n");
+}
+
+static void cmdq_task_exec(struct cmdq_pkt *pkt, struct cmdq_thread *thread)
+{
+	struct cmdq *cmdq;
+	struct cmdq_task *task;
+	unsigned long curr_pa, end_pa;
+
+	cmdq = dev_get_drvdata(thread->chan->mbox->dev);
+
+	/* Client should not flush new tasks if suspended. */
+	WARN_ON(cmdq->suspended);
+
+	task = kzalloc(sizeof(*task), GFP_ATOMIC);
+	task->cmdq = cmdq;
+	INIT_LIST_HEAD(&task->list_entry);
+	task->pa_base = pkt->pa_base;
+	task->thread = thread;
+	task->pkt = pkt;
+
+	if (list_empty(&thread->task_busy_list)) {
+		WARN_ON(clk_enable(cmdq->clock) < 0);
+		WARN_ON(cmdq_thread_reset(cmdq, thread) < 0);
+
+		writel(task->pa_base, thread->base + CMDQ_THR_CURR_ADDR);
+		writel(task->pa_base + pkt->cmd_buf_size,
+		       thread->base + CMDQ_THR_END_ADDR);
+		writel(thread->priority, thread->base + CMDQ_THR_PRIORITY);
+		writel(CMDQ_THR_IRQ_EN, thread->base + CMDQ_THR_IRQ_ENABLE);
+		writel(CMDQ_THR_ENABLED, thread->base + CMDQ_THR_ENABLE_TASK);
+
+		if (thread->timeout_ms != CMDQ_NO_TIMEOUT)
+			mod_timer(&thread->timeout,
+				jiffies + msecs_to_jiffies(thread->timeout_ms));
+	} else {
+		WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
+		curr_pa = readl(thread->base + CMDQ_THR_CURR_ADDR);
+		end_pa = readl(thread->base + CMDQ_THR_END_ADDR);
+
+		/*
+		 * Atomic execution should remove the following wfe, i.e. only
+		 * wait event at first task, and prevent to pause when running.
+		 */
+		if (thread->atomic_exec) {
+			/* GCE is executing if command is not WFE */
+			if (!cmdq_thread_is_in_wfe(thread)) {
+				cmdq_thread_resume(thread);
+				cmdq_thread_wait_end(thread, end_pa);
+				WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
+				/* set to this task directly */
+				writel(task->pa_base,
+				       thread->base + CMDQ_THR_CURR_ADDR);
+			} else {
+				cmdq_task_insert_into_thread(task);
+				cmdq_task_remove_wfe(task);
+				smp_mb(); /* modify jump before enable thread */
+			}
+		} else {
+			/* check boundary */
+			if (curr_pa == end_pa - CMDQ_INST_SIZE ||
+			    curr_pa == end_pa) {
+				/* set to this task directly */
+				writel(task->pa_base,
+				       thread->base + CMDQ_THR_CURR_ADDR);
+			} else {
+				cmdq_task_insert_into_thread(task);
+				smp_mb(); /* modify jump before enable thread */
+			}
+		}
+		writel(task->pa_base + pkt->cmd_buf_size,
+		       thread->base + CMDQ_THR_END_ADDR);
+		cmdq_thread_resume(thread);
+	}
+	list_move_tail(&task->list_entry, &thread->task_busy_list);
+}
+
+static void cmdq_task_exec_done(struct cmdq_task *task, bool err)
+{
+	struct device *dev = task->cmdq->mbox.dev;
+	struct cmdq_cb_data cmdq_cb_data;
+
+	dma_unmap_single(dev, task->pa_base, task->pkt->cmd_buf_size,
+			 DMA_TO_DEVICE);
+	if (task->pkt->cb.cb) {
+		cmdq_cb_data.err = err;
+		cmdq_cb_data.data = task->pkt->cb.data;
+		task->pkt->cb.cb(cmdq_cb_data);
+	}
+	list_del(&task->list_entry);
+}
+
+static void cmdq_task_handle_error(struct cmdq_task *task)
+{
+	struct cmdq_thread *thread = task->thread;
+	struct cmdq_task *next_task;
+
+	dev_err(task->cmdq->mbox.dev, "task 0x%p error\n", task);
+	WARN_ON(cmdq_thread_suspend(task->cmdq, thread) < 0);
+	next_task = list_first_entry_or_null(&thread->task_busy_list,
+			struct cmdq_task, list_entry);
+	if (next_task)
+		writel(next_task->pa_base, thread->base + CMDQ_THR_CURR_ADDR);
+	cmdq_thread_resume(thread);
+}
+
+static void cmdq_thread_irq_handler(struct cmdq *cmdq,
+				    struct cmdq_thread *thread)
+{
+	struct cmdq_task *task, *tmp, *curr_task = NULL;
+	u32 curr_pa, irq_flag, task_end_pa;
+	bool err;
+
+	irq_flag = readl(thread->base + CMDQ_THR_IRQ_STATUS);
+	writel(~irq_flag, thread->base + CMDQ_THR_IRQ_STATUS);
+
+	/*
+	 * When ISR call this function, another CPU core could run
+	 * "release task" right before we acquire the spin lock, and thus
+	 * reset / disable this GCE thread, so we need to check the enable
+	 * bit of this GCE thread.
+	 */
+	if (!(readl(thread->base + CMDQ_THR_ENABLE_TASK) & CMDQ_THR_ENABLED))
+		return;
+
+	if (irq_flag & CMDQ_THR_IRQ_ERROR)
+		err = true;
+	else if (irq_flag & CMDQ_THR_IRQ_DONE)
+		err = false;
+	else
+		return;
+
+	curr_pa = readl(thread->base + CMDQ_THR_CURR_ADDR);
+
+	list_for_each_entry_safe(task, tmp, &thread->task_busy_list,
+				 list_entry) {
+		task_end_pa = task->pa_base + task->pkt->cmd_buf_size;
+		if (curr_pa >= task->pa_base && curr_pa < task_end_pa)
+			curr_task = task;
+
+		if (!curr_task || curr_pa == task_end_pa - CMDQ_INST_SIZE) {
+			cmdq_task_exec_done(task, false);
+			kfree(task);
+		} else if (err) {
+			cmdq_task_exec_done(task, true);
+			cmdq_task_handle_error(curr_task);
+			kfree(task);
+		}
+
+		if (curr_task)
+			break;
+	}
+
+	if (list_empty(&thread->task_busy_list)) {
+		cmdq_thread_disable(cmdq, thread);
+		clk_disable(cmdq->clock);
+	} else {
+		if (thread->timeout_ms != CMDQ_NO_TIMEOUT)
+			mod_timer(&thread->timeout,
+				jiffies + msecs_to_jiffies(thread->timeout_ms));
+	}
+}
+
+static irqreturn_t cmdq_irq_handler(int irq, void *dev)
+{
+	struct cmdq *cmdq = dev;
+	unsigned long irq_status, flags = 0L;
+	int bit;
+
+	irq_status = readl(cmdq->base + CMDQ_CURR_IRQ_STATUS) & CMDQ_IRQ_MASK;
+	if (!(irq_status ^ CMDQ_IRQ_MASK))
+		return IRQ_NONE;
+
+	for_each_clear_bit(bit, &irq_status, fls(CMDQ_IRQ_MASK)) {
+		struct cmdq_thread *thread = &cmdq->thread[bit];
+
+		spin_lock_irqsave(&thread->chan->lock, flags);
+		cmdq_thread_irq_handler(cmdq, thread);
+		spin_unlock_irqrestore(&thread->chan->lock, flags);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static void cmdq_thread_handle_timeout(struct timer_list *t)
+{
+	struct cmdq_thread *thread = from_timer(thread, t, timeout);
+	struct cmdq *cmdq = container_of(thread->chan->mbox, struct cmdq, mbox);
+	struct cmdq_task *task, *tmp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&thread->chan->lock, flags);
+	WARN_ON(cmdq_thread_suspend(cmdq, thread) < 0);
+
+	/*
+	 * Although IRQ is disabled, GCE continues to execute.
+	 * It may have pending IRQ before GCE thread is suspended,
+	 * so check this condition again.
+	 */
+	cmdq_thread_irq_handler(cmdq, thread);
+
+	if (list_empty(&thread->task_busy_list)) {
+		cmdq_thread_resume(thread);
+		spin_unlock_irqrestore(&thread->chan->lock, flags);
+		return;
+	}
+
+	dev_err(cmdq->mbox.dev, "timeout\n");
+	list_for_each_entry_safe(task, tmp, &thread->task_busy_list,
+				 list_entry) {
+		cmdq_task_exec_done(task, true);
+		kfree(task);
+	}
+
+	cmdq_thread_resume(thread);
+	cmdq_thread_disable(cmdq, thread);
+	clk_disable(cmdq->clock);
+	spin_unlock_irqrestore(&thread->chan->lock, flags);
+}
+
+static int cmdq_suspend(struct device *dev)
+{
+	struct cmdq *cmdq = dev_get_drvdata(dev);
+	struct cmdq_thread *thread;
+	int i;
+	bool task_running = false;
+
+	cmdq->suspended = true;
+
+	for (i = 0; i < cmdq->thread_nr; i++) {
+		thread = &cmdq->thread[i];
+		if (!list_empty(&thread->task_busy_list)) {
+			task_running = true;
+			break;
+		}
+	}
+
+	if (task_running)
+		dev_warn(dev, "exist running task(s) in suspend\n");
+
+	clk_unprepare(cmdq->clock);
+
+	return 0;
+}
+
+static int cmdq_resume(struct device *dev)
+{
+	struct cmdq *cmdq = dev_get_drvdata(dev);
+
+	WARN_ON(clk_prepare(cmdq->clock) < 0);
+	cmdq->suspended = false;
+	return 0;
+}
+
+static int cmdq_remove(struct platform_device *pdev)
+{
+	struct cmdq *cmdq = platform_get_drvdata(pdev);
+
+	mbox_controller_unregister(&cmdq->mbox);
+	clk_unprepare(cmdq->clock);
+
+	if (cmdq->mbox.chans)
+		devm_kfree(&pdev->dev, cmdq->mbox.chans);
+
+	if (cmdq->thread)
+		devm_kfree(&pdev->dev, cmdq->thread);
+
+	devm_kfree(&pdev->dev, cmdq);
+
+	return 0;
+}
+
+static int cmdq_mbox_send_data(struct mbox_chan *chan, void *data)
+{
+	cmdq_task_exec(data, chan->con_priv);
+
+	return 0;
+}
+
+static int cmdq_mbox_startup(struct mbox_chan *chan)
+{
+	return 0;
+}
+
+static void cmdq_mbox_shutdown(struct mbox_chan *chan)
+{
+}
+
+static bool cmdq_mbox_last_tx_done(struct mbox_chan *chan)
+{
+	return true;
+}
+
+static const struct mbox_chan_ops cmdq_mbox_chan_ops = {
+	.send_data = cmdq_mbox_send_data,
+	.startup = cmdq_mbox_startup,
+	.shutdown = cmdq_mbox_shutdown,
+	.last_tx_done = cmdq_mbox_last_tx_done,
+};
+
+static struct mbox_chan *cmdq_xlate(struct mbox_controller *mbox,
+		const struct of_phandle_args *sp)
+{
+	int ind = sp->args[0];
+	struct cmdq_thread *thread;
+
+	if (ind >= mbox->num_chans)
+		return ERR_PTR(-EINVAL);
+
+	thread = mbox->chans[ind].con_priv;
+	thread->timeout_ms = sp->args[1];
+	thread->priority = sp->args[2];
+	thread->atomic_exec = (sp->args[3] != 0);
+	thread->chan = &mbox->chans[ind];
+
+	return &mbox->chans[ind];
+}
+
+static int cmdq_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct resource *res;
+	struct cmdq *cmdq;
+	int err, i;
+
+	cmdq = devm_kzalloc(dev, sizeof(*cmdq), GFP_KERNEL);
+	if (!cmdq)
+		return -ENOMEM;
+
+	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 = platform_get_irq(pdev, 0);
+	if (!cmdq->irq) {
+		dev_err(dev, "failed to get irq\n");
+		return -EINVAL;
+	}
+	err = devm_request_irq(dev, cmdq->irq, cmdq_irq_handler, IRQF_SHARED,
+			       "mtk_cmdq", cmdq);
+	if (err < 0) {
+		dev_err(dev, "failed to register ISR (%d)\n", err);
+		return err;
+	}
+
+	dev_dbg(dev, "cmdq device: addr:0x%p, va:0x%p, irq:%d\n",
+		dev, cmdq->base, cmdq->irq);
+
+	cmdq->clock = devm_clk_get(dev, "gce");
+	if (IS_ERR(cmdq->clock)) {
+		dev_err(dev, "failed to get gce clk\n");
+		return PTR_ERR(cmdq->clock);
+	}
+
+	err = of_property_read_u32(dev->of_node, "thread-num",
+				   &cmdq->thread_nr);
+	if (err < 0) {
+		dev_err(dev, "failed to read thread number.\n");
+		return err;
+	}
+
+	cmdq->mbox.dev = dev;
+	cmdq->mbox.chans = devm_kcalloc(dev, cmdq->thread_nr,
+					sizeof(*cmdq->mbox.chans), GFP_KERNEL);
+	if (!cmdq->mbox.chans)
+		return -ENOMEM;
+
+	cmdq->mbox.num_chans = cmdq->thread_nr;
+	cmdq->mbox.ops = &cmdq_mbox_chan_ops;
+	cmdq->mbox.of_xlate = cmdq_xlate;
+
+	/* make use of TXDONE_BY_ACK */
+	cmdq->mbox.txdone_irq = false;
+	cmdq->mbox.txdone_poll = false;
+
+	cmdq->thread = devm_kcalloc(dev, cmdq->thread_nr,
+					sizeof(*cmdq->thread), GFP_KERNEL);
+	if (!cmdq->thread)
+		return -ENOMEM;
+
+	for (i = 0; i < cmdq->thread_nr; i++) {
+		cmdq->thread[i].base = cmdq->base + CMDQ_THR_BASE +
+				CMDQ_THR_SIZE * i;
+		INIT_LIST_HEAD(&cmdq->thread[i].task_busy_list);
+		if (cmdq->thread[i].timeout_ms != CMDQ_NO_TIMEOUT)
+			timer_setup(&cmdq->thread[i].timeout,
+				cmdq_thread_handle_timeout, 0);
+		cmdq->mbox.chans[i].con_priv = &cmdq->thread[i];
+	}
+
+	err = mbox_controller_register(&cmdq->mbox);
+	if (err < 0) {
+		dev_err(dev, "failed to register mailbox: %d\n", err);
+		return err;
+	}
+
+	platform_set_drvdata(pdev, cmdq);
+	WARN_ON(clk_prepare(cmdq->clock) < 0);
+
+	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"},
+	{}
+};
+
+static struct platform_driver cmdq_drv = {
+	.probe = cmdq_probe,
+	.remove = cmdq_remove,
+	.driver = {
+		.name = "mtk_cmdq",
+		.pm = &cmdq_pm_ops,
+		.of_match_table = cmdq_of_ids,
+	}
+};
+
+static int __init cmdq_drv_init(void)
+{
+	return platform_driver_register(&cmdq_drv);
+}
+
+static void __exit cmdq_drv_exit(void)
+{
+	platform_driver_unregister(&cmdq_drv);
+}
+
+subsys_initcall(cmdq_drv_init);
+module_exit(cmdq_drv_exit);
diff --git a/include/linux/mailbox/mtk-cmdq-mailbox.h b/include/linux/mailbox/mtk-cmdq-mailbox.h
new file mode 100644
index 0000000..1e4a891
--- /dev/null
+++ b/include/linux/mailbox/mtk-cmdq-mailbox.h
@@ -0,0 +1,70 @@ 
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (c) 2018 MediaTek Inc.
+ *
+ */
+
+#ifndef __MTK_CMDQ_MAILBOX_H__
+#define __MTK_CMDQ_MAILBOX_H__
+
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+
+#define CMDQ_INST_SIZE			8 /* instruction is 64-bit */
+#define CMDQ_SUBSYS_SHIFT		16
+#define CMDQ_OP_CODE_SHIFT		24
+#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
+
+/*
+ * CMDQ_CODE_MASK:
+ *   set write 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_MASK = 0x02,
+	CMDQ_CODE_WRITE = 0x04,
+	CMDQ_CODE_JUMP = 0x10,
+	CMDQ_CODE_WFE = 0x20,
+	CMDQ_CODE_EOC = 0x40,
+};
+
+struct cmdq_cb_data {
+	bool	err;
+	void	*data;
+};
+
+typedef void (*cmdq_async_flush_cb)(struct cmdq_cb_data data);
+
+struct cmdq_task_cb {
+	cmdq_async_flush_cb	cb;
+	void			*data;
+};
+
+struct cmdq_pkt {
+	void			*va_base;
+	dma_addr_t		pa_base;
+	size_t			cmd_buf_size; /* command occupied size */
+	size_t			buf_size; /* real buffer size */
+	struct cmdq_task_cb	cb;
+};
+
+#endif /* __MTK_CMDQ_MAILBOX_H__ */