Skip to contentSkip to conversation

mailbox/bcm2835: Fix mailbox full detection.

PATCH In review Eric Anholt 3 participants 5 messages May 13, 2015 – May 29, 2015

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mailbox/bcm2835: Fix mailbox full detection.

With the VC reader blocked and the ARM writing, MAIL0_STA reads empty permanently while MAIL1_STA goes from empty (0x40000000) to non-empty (0x00000001-0x00000007) to full (0x80000008).

This bug ended up having no effect on us, because all of our transactions in the client driver were synchronous and under a mutex.

---

This is a port of a patch by Noralf from the downstream Raspberry Pi tree. Noralf said he didn't want attribution on this patch, even if it looks a lot like his.

 drivers/mailbox/bcm2835-mailbox.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/mailbox/bcm2835-mailbox.c +2 -1
index 4b13268..0b47dd4 100644
@@ -49,6 +49,7 @@
 #define MAIL0_STA	(ARM_0_MAIL0 + 0x18)
 #define MAIL0_CNF	(ARM_0_MAIL0 + 0x1C)
 #define MAIL1_WRT	(ARM_0_MAIL1 + 0x00)
+#define MAIL1_STA	(ARM_0_MAIL1 + 0x18)
 
 /* Status register: FIFO state. */
 #define ARM_MS_FULL		BIT(31)
@@ -117,7 +118,7 @@ static bool bcm2835_last_tx_done(struct mbox_chan *link)
 	bool ret;
 
 	spin_lock(&mbox->lock);
-	ret = !(readl(mbox->regs + MAIL0_STA) & ARM_MS_FULL);
+	ret = !(readl(mbox->regs + MAIL1_STA) & ARM_MS_FULL);
 	spin_unlock(&mbox->lock);
 	return ret;
 }
Signature · 1 line
2.1.4
Acked-by on v1 1/1
#1 lore raw
Quoted 6 lines · Eric Anholt

With the VC reader blocked and the ARM writing, MAIL0_STA reads empty permanently while MAIL1_STA goes from empty (0x40000000) to non-empty (0x00000001-0x00000007) to full (0x80000008).

This bug ended up having no effect on us, because all of our transactions in the client driver were synchronous and under a mutex.

If you could get someone at the RPi Foundation or Broadcom to update the register descriptions and example code at the following URLs, that would be rather useful. Otherwise, this code will appear incorrect when compared against the documentation:

https://github.com/raspberrypi/firmware/wiki/Mailboxes
("Mailbox registers" at the bottom)

https://github.com/raspberrypi/firmware/wiki/Accessing-mailboxes
("Sample code")

drivers/mailbox/bcm2835-mailbox.c static bool bcm2835_last_tx_done(struct mbox_chan *link) · line 121

Quoted 8 lines
drivers/mailbox/bcm2835-mailbox.c +1 -1
@@ -117,7 +118,7 @@ static bool bcm2835_last_tx_done(struct mbox_chan *link)
  	bool ret;
 
  	spin_lock(&mbox->lock);
-	ret = !(readl(mbox->regs + MAIL0_STA) & ARM_MS_FULL);
+	ret = !(readl(mbox->regs + MAIL1_STA) & ARM_MS_FULL);

What does "tx done" mean semantically?

If "tx done" means "remote side received all our messages", then surely this should check MAIL1_STA for emptiness, which is different to the "not full" check implemented here?

If "tx done" means "there's space to transmit more messages", then consider this:

... and I guess I'll need to fix U-Boot for the same issue.

#2 lore raw
Quoted 13 lines · Stephen Warren
Quoted 6 lines · Eric Anholt

With the VC reader blocked and the ARM writing, MAIL0_STA reads empty permanently while MAIL1_STA goes from empty (0x40000000) to non-empty (0x00000001-0x00000007) to full (0x80000008).

This bug ended up having no effect on us, because all of our transactions in the client driver were synchronous and under a mutex.

If you could get someone at the RPi Foundation or Broadcom to update the register descriptions and example code at the following URLs, that would be rather useful. Otherwise, this code will appear incorrect when compared against the documentation:

Quoted 5 lines

https://github.com/raspberrypi/firmware/wiki/Mailboxes
("Mailbox registers" at the bottom)

https://github.com/raspberrypi/firmware/wiki/Accessing-mailboxes
("Sample code")

Since it's a wiki, I went ahead and edited the first one. Hopefully that clarifies how the c++ in the other page is supposed to be used.

Quoted 11 lines

drivers/mailbox/bcm2835-mailbox.c static bool bcm2835_last_tx_done(struct mbox_chan *link) · line 121

Quoted 8 lines
drivers/mailbox/bcm2835-mailbox.c +1 -1
@@ -117,7 +118,7 @@ static bool bcm2835_last_tx_done(struct mbox_chan *link)
  	bool ret;
 
  	spin_lock(&mbox->lock);
-	ret = !(readl(mbox->regs + MAIL0_STA) & ARM_MS_FULL);
+	ret = !(readl(mbox->regs + MAIL1_STA) & ARM_MS_FULL);

What does "tx done" mean semantically?

Quoted 6 lines

If "tx done" means "remote side received all our messages", then surely this should check MAIL1_STA for emptiness, which is different to the "not full" check implemented here?

If "tx done" means "there's space to transmit more messages", then consider this:

The mailbox core appears to use this hook as "there's space to transmit more messages." The name does seem really confusing.

maintainer
#3 lore raw
Quoted 15 lines · Eric Anholt
Quoted 13 lines · Stephen Warren
Quoted 6 lines · Eric Anholt

With the VC reader blocked and the ARM writing, MAIL0_STA reads empty permanently while MAIL1_STA goes from empty (0x40000000) to non-empty (0x00000001-0x00000007) to full (0x80000008).

This bug ended up having no effect on us, because all of our transactions in the client driver were synchronous and under a mutex.

If you could get someone at the RPi Foundation or Broadcom to update the register descriptions and example code at the following URLs, that would be rather useful. Otherwise, this code will appear incorrect when compared against the documentation:

Quoted 24 lines
Quoted 5 lines

https://github.com/raspberrypi/firmware/wiki/Mailboxes
("Mailbox registers" at the bottom)

https://github.com/raspberrypi/firmware/wiki/Accessing-mailboxes
("Sample code")

Since it's a wiki, I went ahead and edited the first one. Hopefully that clarifies how the c++ in the other page is supposed to be used.

Quoted 15 lines

drivers/mailbox/bcm2835-mailbox.c static bool bcm2835_last_tx_done(struct mbox_chan *link) · line 121

Quoted 8 lines
drivers/mailbox/bcm2835-mailbox.c +1 -1
@@ -117,7 +118,7 @@ static bool bcm2835_last_tx_done(struct mbox_chan *link)
     bool ret;
 
     spin_lock(&mbox->lock);
-    ret = !(readl(mbox->regs + MAIL0_STA) & ARM_MS_FULL);
+    ret = !(readl(mbox->regs + MAIL1_STA) & ARM_MS_FULL);

What does "tx done" mean semantically?

If "tx done" means "remote side received all our messages", then surely this should check MAIL1_STA for emptiness, which is different to the "not full" check implemented here?

Mailbox core queues messages internally. The controller driver is provided one message at a time.

Quoted 6 lines
Quoted 2 lines

If "tx done" means "there's space to transmit more messages", then consider this:

The mailbox core appears to use this hook as "there's space to transmit more messages." The name does seem really confusing.

'tx_done' => 'is tx done' / 'tx is done'. 'tx' is the last submitted message to be transferred.
So 'tx_done' marks if the last transmitted message was received by the remote. That event could be indicated by some register's bit cleared or even a message(ack) returned by remote (client does mbox_client_txdone). And if last tx was done, we should be able to send next message.

maintainer
#4 lore raw
In reply to Stephen Warren
Quoted 36 lines · Stephen Warren
Quoted 7 lines · Eric Anholt

With the VC reader blocked and the ARM writing, MAIL0_STA reads empty permanently while MAIL1_STA goes from empty (0x40000000) to non-empty (0x00000001-0x00000007) to full (0x80000008).

This bug ended up having no effect on us, because all of our transactions in the client driver were synchronous and under a mutex.

If you could get someone at the RPi Foundation or Broadcom to update the register descriptions and example code at the following URLs, that would be rather useful. Otherwise, this code will appear incorrect when compared against the documentation:

https://github.com/raspberrypi/firmware/wiki/Mailboxes
("Mailbox registers" at the bottom)

https://github.com/raspberrypi/firmware/wiki/Accessing-mailboxes
("Sample code")

drivers/mailbox/bcm2835-mailbox.c static bool bcm2835_last_tx_done(struct mbox_chan · line 122

Quoted 11 lines
drivers/mailbox/bcm2835-mailbox.c +1 -1
@@ -117,7 +118,7 @@ static bool bcm2835_last_tx_done(struct mbox_chan
 *link)
        bool ret;
 
        spin_lock(&mbox->lock);
-       ret = !(readl(mbox->regs + MAIL0_STA) & ARM_MS_FULL);
+       ret = !(readl(mbox->regs + MAIL1_STA) & ARM_MS_FULL);

What does "tx done" mean semantically?

Quoted 4 lines

If "tx done" means "remote side received all our messages", then surely this should check MAIL1_STA for emptiness, which is different to the "not full" check implemented here?

If only one message is active at any time we may have either FULL or EMPTY status(?), so that might explain why this patch works as well. But yes, it should check for emptiness ideally.