Skip to contentSkip to conversation

spi: Add a timeout when waiting for transfers

PATCH In review Mark Brown 2 participants 7 messages Jan 30, 2014 – Feb 4, 2014 6 messages not in the archive SPI

lore
Share & tools
lore

Mark Brown

author
#0 lore raw

PATCH 1/1 spi: Add a timeout when waiting for transfers

From: Mark Brown <[email protected]>

Don't wait indefinitely for transfers to complete but time out after 10ms more than we expect the transfer to take on the wire.

---
 drivers/spi/spi.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/spi/spi.c +11 -1
index 7f23cf9afa79..1826a50c2aaf 100644
@@ -710,6 +710,7 @@ static int spi_transfer_one_message(struct spi_master *master,
 	bool cur_cs = true;
 	bool keep_cs = false;
 	int ret = 0;
+	int ms = 1;
 
 	spi_set_cs(msg->spi, true);
 
@@ -727,7 +728,16 @@ static int spi_transfer_one_message(struct spi_master *master,
 
 		if (ret > 0) {
 			ret = 0;
-			wait_for_completion(&master->xfer_completion);
+			ms = xfer->len * 8 * 1000 / xfer->speed_hz;
+			ms += 10; /* some tolerance */
+
+			ms = wait_for_completion_timeout(&master->xfer_completion,
+							 msecs_to_jiffies(ms));
+		}
+
+		if (ms == 0) {
+			dev_err(&msg->spi->dev, "SPI transfer timed out\n");
+			msg->status = -ETIMEDOUT;
 		}
 
 		trace_spi_transfer_stop(msg, xfer);
Signature · 6 lines
1.9.rc1

--
To unsubscribe from this list: send the line "unsubscribe linux-spi" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 1391121536-10937-1-git-send-email-broonie-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org

#2 lore raw

Hi Mark,

Quoted 30 lines · Mark Brown

From: Mark Brown <[email protected]>

Don't wait indefinitely for transfers to complete but time out after 10ms more than we expect the transfer to take on the wire.

---
 drivers/spi/spi.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/spi/spi.c +4 -1
index 7f23cf9afa79..1826a50c2aaf 100644
@@ -710,6 +710,7 @@ static int spi_transfer_one_message(struct spi_master *master,
        bool cur_cs = true;
        bool keep_cs = false;
        int ret = 0;
+       int ms = 1;
 
        spi_set_cs(msg->spi, true);
 
@@ -727,7 +728,16 @@ static int spi_transfer_one_message(struct spi_master *master,
 
                if (ret > 0) {
                        ret = 0;
-                       wait_for_completion(&master->xfer_completion);
+                       ms = xfer->len * 8 * 1000 / xfer->speed_hz;
+                       ms += 10; /* some tolerance */
+

drivers/spi/spi.c static int spi_transfer_one_message(struct spi_master *master, · line 743

Quoted 10 lines
drivers/spi/spi.c +7 -0
+                       ms = wait_for_completion_timeout(&master->xfer_completion,
+                                                        msecs_to_jiffies(ms));
+               }
+
+               if (ms == 0) {
+                       dev_err(&msg->spi->dev, "SPI transfer timed out\n");
+                       msg->status = -ETIMEDOUT;
                }
 
                trace_spi_transfer_stop(msg, xfer);

What if it still completes in the driver a little bit later? The driver will also call
spi_finalize_current_message(), and we'll get a NULL pointer dereference as master->cur_msg is now NULL.

So I think we need a check like (whitespace damaged) below:

drivers/spi/spi.c +3 -0
@@ -829,6 +829,9 @@ void spi_finalize_current_message(struct spi_master *master)
          queue_kthread_work(&master->kworker, &master->pump_messages);
          spin_unlock_irqrestore(&master->queue_lock, flags);
 
+         if (!mesg)
+                 return NULL;
+
          if (master->cur_msg_prepared && master->unprepare_message) {
                  ret = master->unprepare_message(master, mesg);
          if (ret) {

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
--
To unsubscribe from this list: send the line "unsubscribe linux-spi" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): CAMuHMdUVmL5Zq8AYPVmUyWEBTzokSMV-Xg4uFzpxr7c2uSmU7g-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

author
#4 lore raw

drivers/spi/spi.c void spi_finalize_current_message(struct spi_master *master) · line 834

Quoted 9 lines · Geert Uytterhoeven
drivers/spi/spi.c +3 -0
@@ -829,6 +829,9 @@ void spi_finalize_current_message(struct spi_master *master)
          queue_kthread_work(&master->kworker, &master->pump_messages);
          spin_unlock_irqrestore(&master->queue_lock, flags);
 
+         if (!mesg)
+                 return NULL;
+

Note that this is for transfers, not for messages, and we don't change the completion path for the message here so I'm not sure why the message would be affected? That said there is definitely an issue here - we should probably be adding some sort of error teardown callback to try to stop the hardware if we do hit a timeout.

A message that is not in the archive (sent off-list or not yet received): [email protected]

#6 lore raw
Quoted 7 lines · Mark Brown
Quoted 6 lines · Geert Uytterhoeven
drivers/spi/spi.c +0 -0
@@ -829,6 +829,9 @@ void spi_finalize_current_message(struct spi_master *master)
          queue_kthread_work(&master->kworker, &master->pump_messages);
          spin_unlock_irqrestore(&master->queue_lock, flags);
 
Quoted 9 lines

drivers/spi/spi.c void spi_finalize_current_message(struct spi_master *master) · line 834

Quoted 3 lines
drivers/spi/spi.c +3 -0
+         if (!mesg)
+                 return NULL;
+

Note that this is for transfers, not for messages, and we don't change the completion path for the message here so I'm not sure why the message would be affected? That said there is definitely an issue here - we should probably be adding some sort of error teardown callback to try to stop the hardware if we do hit a timeout.

RIght. Sorry, I mixed them up.

One other thing: I haven't tried your patch yet, but I'm afraid the 10 ms may be too small.

E.g. with PIO-based RSPI I don't get more than 2 Mbps, even though spi-max-frequency = <30000000>, due to the PIO and interrupt overhead. Hence a 1 MiB read would take ca. 4s, while your timeout would be 300 ms.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
--
To unsubscribe from this list: send the line "unsubscribe linux-spi" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): CAMuHMdXLe+Ot5ntOtUC2Qynid4mv1h+UoitJuDW=Z-iyRKXCdg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

author
#8 lore raw
Quoted 6 lines · Geert Uytterhoeven

One other thing: I haven't tried your patch yet, but I'm afraid the 10 ms may be too small.

E.g. with PIO-based RSPI I don't get more than 2 Mbps, even though spi-max-frequency = <30000000>, due to the PIO and interrupt overhead. Hence a 1 MiB read would take ca. 4s, while your timeout would be 300 ms.

Hrm, I wouldn't have expected something doing PIO in more than one burst to be letting the transfer run in the background. Though I suppose that might make sense in some situations...

I was wondering if that was cutting it a bit fine but more for scheduler reasons, it's what the s3c64xx driver has been using for a while without complaints but may not translate so well with greater exposure.

A message that is not in the archive (sent off-list or not yet received): [email protected]

#10 lore raw

Hi Mark,

Quoted 11 lines · Mark Brown
Quoted 6 lines · Geert Uytterhoeven

One other thing: I haven't tried your patch yet, but I'm afraid the 10 ms may be too small.

E.g. with PIO-based RSPI I don't get more than 2 Mbps, even though spi-max-frequency = <30000000>, due to the PIO and interrupt overhead. Hence a 1 MiB read would take ca. 4s, while your timeout would be 300 ms.

Hrm, I wouldn't have expected something doing PIO in more than one burst to be letting the transfer run in the background. Though I suppose that might make sense in some situations...

While I've been thinking about moving more code to the interrupt handler, the current RSPI transfer_one() is indeed still synchronous, so the timeout doesn't apply yet.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
--
To unsubscribe from this list: send the line "unsubscribe linux-spi" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): CAMuHMdXgDAPy4aezxULQsjiRUfyE3F2_NSjuWhBiy119DSXR5A-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

author
#12 lore raw
Quoted 8 lines · Geert Uytterhoeven
Quoted 2 lines · Mark Brown

to be letting the transfer run in the background. Though I suppose that might make sense in some situations...

While I've been thinking about moving more code to the interrupt handler, the current RSPI transfer_one() is indeed still synchronous, so the timeout doesn't apply yet.

OK, that's interesting. Actually one of the things that this refactoring into the core is driving at is that hopefully we'll be able to deploy some of the optimisation techniques like that in the core rather than having individual drivers open coding them - this should on average increase performance.