Current mailbox driver of Hi3660 release the mailbox directly
before sending a new message which may cause last message lost
and next message sending doesn't take effect actually.
This patch fixs this issue by following the right progress below,
each time before sending a message, mailbox driver will check
whether the mailbox is in ready state, if last message has been
acknowledged, the mailbox driver will clear the ack state to turn
the mailbox to ready state again.
Signed-off-by: Kevin Wangtao <[email protected]>
---
drivers/mailbox/hi3660-mailbox.c | 22 ++++++++++------------
1 file changed, 10 insertions(+), 12 deletions(-)
diff --git a/drivers/mailbox/hi3660-mailbox.c b/drivers/mailbox/hi3660-mailbox.c
index 3eea6b6..035b71a 100644
--- a/drivers/mailbox/hi3660-mailbox.c
+++ b/drivers/mailbox/hi3660-mailbox.c
@@ -38,6 +38,7 @@
#define MBOX_AUTOMATIC_ACK 1
#define MBOX_STATE_IDLE BIT(4)
+#define MBOX_STATE_READY BIT(5)
#define MBOX_STATE_ACK BIT(7)
#define MBOX_MSG_LEN 8
@@ -91,8 +92,8 @@ static int hi3660_mbox_check_state(struct mbox_chan *chan)
unsigned long val;
unsigned int ret;
- /* Mailbox is idle so directly bail out */
- if (readl(base + MBOX_MODE_REG) & MBOX_STATE_IDLE)
+ /* Mailbox is ready to use */
+ if (readl(base + MBOX_MODE_REG) & MBOX_STATE_READY)
return 0;
/* Wait for acknowledge from remote */
@@ -103,9 +104,9 @@ static int hi3660_mbox_check_state(struct mbox_chan *chan)
return ret;
}
- /* Ensure channel is released */
- writel(0xffffffff, base + MBOX_IMASK_REG);
- writel(BIT(mchan->ack_irq), base + MBOX_SRC_REG);
+ /* clear ack state, mailbox will get back to ready state */
+ writel(BIT(mchan->ack_irq), base + MBOX_ICLR_REG);
+
return 0;
}
@@ -160,10 +161,6 @@ static int hi3660_mbox_startup(struct mbox_chan *chan)
{
int ret;
- ret = hi3660_mbox_check_state(chan);
- if (ret)
- return ret;
-
ret = hi3660_mbox_unlock(chan);
if (ret)
return ret;
@@ -183,10 +180,11 @@ static int hi3660_mbox_send_data(struct mbox_chan *chan, void *msg)
void __iomem *base = MBOX_BASE(mbox, ch);
u32 *buf = msg;
unsigned int i;
+ int ret;
- /* Ensure channel is released */
- writel_relaxed(0xffffffff, base + MBOX_IMASK_REG);
- writel_relaxed(BIT(mchan->ack_irq), base + MBOX_SRC_REG);
+ ret = hi3660_mbox_check_state(chan);
+ if (ret)
+ return ret;
/* Clear mask for destination interrupt */
writel_relaxed(~BIT(mchan->dst_irq), base + MBOX_IMASK_REG);
--
2.8.1
Current mailbox driver of Hi3660 release the mailbox directly
before sending a new message which may cause last message lost
and next message sending doesn't take effect actually.
This patch fixes this issue by following the right process below,
each time before sending a message, mailbox driver will check
whether the mailbox is in ready state, if last message has been
acknowledged, mailbox driver will clear the ack state to turn
the mailbox to ready state again.
Signed-off-by: Kevin Wangtao <[email protected]>
---
Changes v1 -> v2:
- update commit message
drivers/mailbox/hi3660-mailbox.c | 22 ++++++++++------------
1 file changed, 10 insertions(+), 12 deletions(-)
diff --git a/drivers/mailbox/hi3660-mailbox.c b/drivers/mailbox/hi3660-mailbox.c
index 3eea6b6..035b71a 100644
--- a/drivers/mailbox/hi3660-mailbox.c
+++ b/drivers/mailbox/hi3660-mailbox.c
@@ -38,6 +38,7 @@
#define MBOX_AUTOMATIC_ACK 1
#define MBOX_STATE_IDLE BIT(4)
+#define MBOX_STATE_READY BIT(5)
#define MBOX_STATE_ACK BIT(7)
#define MBOX_MSG_LEN 8
@@ -91,8 +92,8 @@ static int hi3660_mbox_check_state(struct mbox_chan *chan)
unsigned long val;
unsigned int ret;
- /* Mailbox is idle so directly bail out */
- if (readl(base + MBOX_MODE_REG) & MBOX_STATE_IDLE)
+ /* Mailbox is ready to use */
+ if (readl(base + MBOX_MODE_REG) & MBOX_STATE_READY)
return 0;
/* Wait for acknowledge from remote */
@@ -103,9 +104,9 @@ static int hi3660_mbox_check_state(struct mbox_chan *chan)
return ret;
}
- /* Ensure channel is released */
- writel(0xffffffff, base + MBOX_IMASK_REG);
- writel(BIT(mchan->ack_irq), base + MBOX_SRC_REG);
+ /* clear ack state, mailbox will get back to ready state */
+ writel(BIT(mchan->ack_irq), base + MBOX_ICLR_REG);
+
return 0;
}
@@ -160,10 +161,6 @@ static int hi3660_mbox_startup(struct mbox_chan *chan)
{
int ret;
- ret = hi3660_mbox_check_state(chan);
- if (ret)
- return ret;
-
ret = hi3660_mbox_unlock(chan);
if (ret)
return ret;
@@ -183,10 +180,11 @@ static int hi3660_mbox_send_data(struct mbox_chan *chan, void *msg)
void __iomem *base = MBOX_BASE(mbox, ch);
u32 *buf = msg;
unsigned int i;
+ int ret;
- /* Ensure channel is released */
- writel_relaxed(0xffffffff, base + MBOX_IMASK_REG);
- writel_relaxed(BIT(mchan->ack_irq), base + MBOX_SRC_REG);
+ ret = hi3660_mbox_check_state(chan);
+ if (ret)
+ return ret;
/* Clear mask for destination interrupt */
writel_relaxed(~BIT(mchan->dst_irq), base + MBOX_IMASK_REG);
--
2.8.1
On Mon, Dec 03, 2018 at 02:13:08PM +0800, Kevin Wangtao wrote:
> Current mailbox driver of Hi3660 release the mailbox directly
> before sending a new message which may cause last message lost
> and next message sending doesn't take effect actually.
>
> This patch fixes this issue by following the right process below,
> each time before sending a message, mailbox driver will check
> whether the mailbox is in ready state, if last message has been
> acknowledged, mailbox driver will clear the ack state to turn
> the mailbox to ready state again.
>
> Signed-off-by: Kevin Wangtao <[email protected]>
Thanks for sending out this patch, Kevin.
I tested this patch at my side with two cases (one is dhry2 for
performance testing for every OPPs, and another is stress test
two clusters frequency change concurrently), both of them can
pass after applied this patch.
Reviewed-and-tested-by: Leo Yan <[email protected]>
Jassi, please review and pick it up if it's okay for you.
Thanks,
Leo Yan
> ---
> Changes v1 -> v2:
> - update commit message
>
> drivers/mailbox/hi3660-mailbox.c | 22 ++++++++++------------
> 1 file changed, 10 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/mailbox/hi3660-mailbox.c b/drivers/mailbox/hi3660-mailbox.c
> index 3eea6b6..035b71a 100644
> --- a/drivers/mailbox/hi3660-mailbox.c
> +++ b/drivers/mailbox/hi3660-mailbox.c
> @@ -38,6 +38,7 @@
> #define MBOX_AUTOMATIC_ACK 1
>
> #define MBOX_STATE_IDLE BIT(4)
> +#define MBOX_STATE_READY BIT(5)
> #define MBOX_STATE_ACK BIT(7)
>
> #define MBOX_MSG_LEN 8
> @@ -91,8 +92,8 @@ static int hi3660_mbox_check_state(struct mbox_chan *chan)
> unsigned long val;
> unsigned int ret;
>
> - /* Mailbox is idle so directly bail out */
> - if (readl(base + MBOX_MODE_REG) & MBOX_STATE_IDLE)
> + /* Mailbox is ready to use */
> + if (readl(base + MBOX_MODE_REG) & MBOX_STATE_READY)
> return 0;
>
> /* Wait for acknowledge from remote */
> @@ -103,9 +104,9 @@ static int hi3660_mbox_check_state(struct mbox_chan *chan)
> return ret;
> }
>
> - /* Ensure channel is released */
> - writel(0xffffffff, base + MBOX_IMASK_REG);
> - writel(BIT(mchan->ack_irq), base + MBOX_SRC_REG);
> + /* clear ack state, mailbox will get back to ready state */
> + writel(BIT(mchan->ack_irq), base + MBOX_ICLR_REG);
> +
> return 0;
> }
>
> @@ -160,10 +161,6 @@ static int hi3660_mbox_startup(struct mbox_chan *chan)
> {
> int ret;
>
> - ret = hi3660_mbox_check_state(chan);
> - if (ret)
> - return ret;
> -
> ret = hi3660_mbox_unlock(chan);
> if (ret)
> return ret;
> @@ -183,10 +180,11 @@ static int hi3660_mbox_send_data(struct mbox_chan *chan, void *msg)
> void __iomem *base = MBOX_BASE(mbox, ch);
> u32 *buf = msg;
> unsigned int i;
> + int ret;
>
> - /* Ensure channel is released */
> - writel_relaxed(0xffffffff, base + MBOX_IMASK_REG);
> - writel_relaxed(BIT(mchan->ack_irq), base + MBOX_SRC_REG);
> + ret = hi3660_mbox_check_state(chan);
> + if (ret)
> + return ret;
>
> /* Clear mask for destination interrupt */
> writel_relaxed(~BIT(mchan->dst_irq), base + MBOX_IMASK_REG);
> --
> 2.8.1
>
Hi,
On 03/12/2018 06:13, Kevin Wangtao wrote:
> Current mailbox driver of Hi3660 release the mailbox directly
> before sending a new message which may cause last message lost
> and next message sending doesn't take effect actually.
>
> This patch fixes this issue by following the right process below,
> each time before sending a message, mailbox driver will check
> whether the mailbox is in ready state, if last message has been
> acknowledged, mailbox driver will clear the ack state to turn
> the mailbox to ready state again.
>
> Signed-off-by: Kevin Wangtao <[email protected]>
Validated using sysbench at each OPP ([1]). Test passes with patch and fails
without:
| CPU | OPP | Base | Patch |
|------+---------+------+-------|
| CPU0 | 533000 | 104 | 104 |
| CPU0 | 999000 | 104 | 201 |
| CPU0 | 140200 | 285 | 285 |
| CPU0 | 1709000 | 285 | 349 |
| CPU0 | 1844000 | 377 | 377 |
|------+---------+------+-------|
| CPU4 | 903000 | 249 | 248 |
| CPU4 | 1421000 | 249 | 394 |
| CPU4 | 1805000 | 500 | 500 |
| CPU4 | 2112000 | 499 | 583 |
| CPU4 | 2362000 | 653 | 654 |
We need this pretty badly, otherwise frequency setting is a coin toss.
Tested-by: Valentin Schneider <[email protected]>
Thanks,
Valentin
[1]: https://github.com/ARM-software/lisa/blob/next/lisa/tests/kernel/cpufreq/sanity.py#L23
Ran the above with this script on each kernel:
----->8-----
#!/usr/bin/env python3
from lisa.env import TestEnv
from lisa.tests.kernel.cpufreq.sanity import UserspaceSanity
te = TestEnv.from_cli()
print(UserspaceSanity.from_testenv(te).test_performance_sanity())
-----8<-----
> ---
> Changes v1 -> v2:
> - update commit message
>
> drivers/mailbox/hi3660-mailbox.c | 22 ++++++++++------------
> 1 file changed, 10 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/mailbox/hi3660-mailbox.c b/drivers/mailbox/hi3660-mailbox.c
> index 3eea6b6..035b71a 100644
> --- a/drivers/mailbox/hi3660-mailbox.c
> +++ b/drivers/mailbox/hi3660-mailbox.c
> @@ -38,6 +38,7 @@
> #define MBOX_AUTOMATIC_ACK 1
>
> #define MBOX_STATE_IDLE BIT(4)
> +#define MBOX_STATE_READY BIT(5)
> #define MBOX_STATE_ACK BIT(7)
>
> #define MBOX_MSG_LEN 8
> @@ -91,8 +92,8 @@ static int hi3660_mbox_check_state(struct mbox_chan *chan)
> unsigned long val;
> unsigned int ret;
>
> - /* Mailbox is idle so directly bail out */
> - if (readl(base + MBOX_MODE_REG) & MBOX_STATE_IDLE)
> + /* Mailbox is ready to use */
> + if (readl(base + MBOX_MODE_REG) & MBOX_STATE_READY)
> return 0;
>
> /* Wait for acknowledge from remote */
> @@ -103,9 +104,9 @@ static int hi3660_mbox_check_state(struct mbox_chan *chan)
> return ret;
> }
>
> - /* Ensure channel is released */
> - writel(0xffffffff, base + MBOX_IMASK_REG);
> - writel(BIT(mchan->ack_irq), base + MBOX_SRC_REG);
> + /* clear ack state, mailbox will get back to ready state */
> + writel(BIT(mchan->ack_irq), base + MBOX_ICLR_REG);
> +
> return 0;
> }
>
> @@ -160,10 +161,6 @@ static int hi3660_mbox_startup(struct mbox_chan *chan)
> {
> int ret;
>
> - ret = hi3660_mbox_check_state(chan);
> - if (ret)
> - return ret;
> -
> ret = hi3660_mbox_unlock(chan);
> if (ret)
> return ret;
> @@ -183,10 +180,11 @@ static int hi3660_mbox_send_data(struct mbox_chan *chan, void *msg)
> void __iomem *base = MBOX_BASE(mbox, ch);
> u32 *buf = msg;
> unsigned int i;
> + int ret;
>
> - /* Ensure channel is released */
> - writel_relaxed(0xffffffff, base + MBOX_IMASK_REG);
> - writel_relaxed(BIT(mchan->ack_irq), base + MBOX_SRC_REG);
> + ret = hi3660_mbox_check_state(chan);
> + if (ret)
> + return ret;
>
> /* Clear mask for destination interrupt */
> writel_relaxed(~BIT(mchan->dst_irq), base + MBOX_IMASK_REG);
>