Re: [Sound-open-firmware] [PATCH 5/7] soundwire: intel: Don't disable interrupt until children are removed
On 9/7/22 12:14, Richard Fitzgerald wrote:
The cadence_master code needs the interrupt to complete message transfers. When the bus driver is being removed child drivers are removed, and their remove actions might need bus transactions.
Use the sdw_master_ops.remove callback to disable the interrupt handling only after the child drivers have been removed.
Signed-off-by: Richard Fitzgerald rf@opensource.cirrus.com
drivers/soundwire/intel.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/soundwire/intel.c b/drivers/soundwire/intel.c index 01be62fa6c83..d5e723a9c80b 100644 --- a/drivers/soundwire/intel.c +++ b/drivers/soundwire/intel.c @@ -1255,6 +1255,13 @@ static int intel_prop_read(struct sdw_bus *bus) return 0; }
+static void intel_bus_remove(struct sdw_bus *bus) +{
- struct sdw_cdns *cdns = bus_to_cdns(bus);
- sdw_cdns_enable_interrupt(cdns, false);
don't you need to check for any on-going transactions on the bus?
I wonder if there could be a corner case where there are no child devices but still a device physically attached to the bus. I am not sure if the 'no devices left' is a good-enough indication of no activity on the bus.
+}
static struct sdw_master_ops sdw_intel_ops = { .read_prop = sdw_master_read_prop, .override_adr = sdw_dmi_override_adr, @@ -1264,6 +1271,7 @@ static struct sdw_master_ops sdw_intel_ops = { .set_bus_conf = cdns_bus_conf, .pre_bank_switch = intel_pre_bank_switch, .post_bank_switch = intel_post_bank_switch,
- .remove = intel_bus_remove,
};
static int intel_init(struct sdw_intel *sdw) @@ -1502,7 +1510,6 @@ static void intel_link_remove(struct auxiliary_device *auxdev) */ if (!bus->prop.hw_disabled) { intel_debugfs_exit(sdw);
snd_soc_unregister_component(dev); } sdw_bus_master_delete(bus);sdw_cdns_enable_interrupt(cdns, false);
On 12/09/2022 11:53, Pierre-Louis Bossart wrote:
On 9/7/22 12:14, Richard Fitzgerald wrote:
The cadence_master code needs the interrupt to complete message transfers. When the bus driver is being removed child drivers are removed, and their remove actions might need bus transactions.
Use the sdw_master_ops.remove callback to disable the interrupt handling only after the child drivers have been removed.
Signed-off-by: Richard Fitzgerald rf@opensource.cirrus.com
drivers/soundwire/intel.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/soundwire/intel.c b/drivers/soundwire/intel.c index 01be62fa6c83..d5e723a9c80b 100644 --- a/drivers/soundwire/intel.c +++ b/drivers/soundwire/intel.c @@ -1255,6 +1255,13 @@ static int intel_prop_read(struct sdw_bus *bus) return 0; }
+static void intel_bus_remove(struct sdw_bus *bus) +{
- struct sdw_cdns *cdns = bus_to_cdns(bus);
- sdw_cdns_enable_interrupt(cdns, false);
don't you need to check for any on-going transactions on the bus?
As all the child drivers have removed, I think the only other place that can generate bus transactions is the PING handler but sdw_cdns_enable_interrupt(false) calls cancel_work_sync() to cancel the cdns->work and it sets a flag so that it will not be re-queued.
I wonder if there could be a corner case where there are no child devices but still a device physically attached to the bus. I am not sure if the 'no devices left' is a good-enough indication of no activity on the bus.
As above - yes there could, but sdw_cdns_enable_interrupt(false) will cancel the work and stop it being re-queued.
+}
- static struct sdw_master_ops sdw_intel_ops = { .read_prop = sdw_master_read_prop, .override_adr = sdw_dmi_override_adr,
@@ -1264,6 +1271,7 @@ static struct sdw_master_ops sdw_intel_ops = { .set_bus_conf = cdns_bus_conf, .pre_bank_switch = intel_pre_bank_switch, .post_bank_switch = intel_post_bank_switch,
.remove = intel_bus_remove, };
static int intel_init(struct sdw_intel *sdw)
@@ -1502,7 +1510,6 @@ static void intel_link_remove(struct auxiliary_device *auxdev) */ if (!bus->prop.hw_disabled) { intel_debugfs_exit(sdw);
snd_soc_unregister_component(dev); } sdw_bus_master_delete(bus);sdw_cdns_enable_interrupt(cdns, false);
On 9/12/22 17:36, Richard Fitzgerald wrote:
On 12/09/2022 11:53, Pierre-Louis Bossart wrote:
On 9/7/22 12:14, Richard Fitzgerald wrote:
The cadence_master code needs the interrupt to complete message transfers. When the bus driver is being removed child drivers are removed, and their remove actions might need bus transactions.
Use the sdw_master_ops.remove callback to disable the interrupt handling only after the child drivers have been removed.
Signed-off-by: Richard Fitzgerald rf@opensource.cirrus.com
drivers/soundwire/intel.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/soundwire/intel.c b/drivers/soundwire/intel.c index 01be62fa6c83..d5e723a9c80b 100644 --- a/drivers/soundwire/intel.c +++ b/drivers/soundwire/intel.c @@ -1255,6 +1255,13 @@ static int intel_prop_read(struct sdw_bus *bus) return 0; } +static void intel_bus_remove(struct sdw_bus *bus) +{ + struct sdw_cdns *cdns = bus_to_cdns(bus);
+ sdw_cdns_enable_interrupt(cdns, false);
don't you need to check for any on-going transactions on the bus?
As all the child drivers have removed, I think the only other place that can generate bus transactions is the PING handler but sdw_cdns_enable_interrupt(false) calls cancel_work_sync() to cancel the cdns->work and it sets a flag so that it will not be re-queued.
I wonder if there could be a corner case where there are no child devices but still a device physically attached to the bus. I am not sure if the 'no devices left' is a good-enough indication of no activity on the bus.
As above - yes there could, but sdw_cdns_enable_interrupt(false) will cancel the work and stop it being re-queued.
Ah yes, I forgot that part, thanks!
Reviewed-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com
On 12/09/2022 18:12, Pierre-Louis Bossart wrote:
On 9/12/22 17:36, Richard Fitzgerald wrote:
On 12/09/2022 11:53, Pierre-Louis Bossart wrote:
On 9/7/22 12:14, Richard Fitzgerald wrote:
The cadence_master code needs the interrupt to complete message transfers. When the bus driver is being removed child drivers are removed, and their remove actions might need bus transactions.
Use the sdw_master_ops.remove callback to disable the interrupt handling only after the child drivers have been removed.
Signed-off-by: Richard Fitzgerald rf@opensource.cirrus.com
drivers/soundwire/intel.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/soundwire/intel.c b/drivers/soundwire/intel.c index 01be62fa6c83..d5e723a9c80b 100644 --- a/drivers/soundwire/intel.c +++ b/drivers/soundwire/intel.c @@ -1255,6 +1255,13 @@ static int intel_prop_read(struct sdw_bus *bus) return 0; } +static void intel_bus_remove(struct sdw_bus *bus) +{ + struct sdw_cdns *cdns = bus_to_cdns(bus);
+ sdw_cdns_enable_interrupt(cdns, false);
don't you need to check for any on-going transactions on the bus?
As all the child drivers have removed, I think the only other place that can generate bus transactions is the PING handler but sdw_cdns_enable_interrupt(false) calls cancel_work_sync() to cancel the cdns->work and it sets a flag so that it will not be re-queued.
I wonder if there could be a corner case where there are no child devices but still a device physically attached to the bus. I am not sure if the 'no devices left' is a good-enough indication of no activity on the bus.
As above - yes there could, but sdw_cdns_enable_interrupt(false) will cancel the work and stop it being re-queued.
Ah yes, I forgot that part, thanks!
... but I have noticed that there is a bug in sdw_cdns_enable_interrupt(). It doesn't ensure that the IRQ thread has seen the cdns->interrupt_enabled = false. I'll add a patch to fix that when I re-push this chain.
participants (2)
-
Pierre-Louis Bossart
-
Richard Fitzgerald