On resume there can be a period of time after the
preceding serio_resume -> psmouse_deactivate call
where calls to rmi_smb_get_version fail with
-ENXIO.
The call path in rmi_smb_resume is rmi_smb_resume -> rmi_smb_reset ->
rmi_smb_enable_smbus_mode -> rmi_smb_get_version where
this failure would occur.
Add a 30ms delay and retry in the ENXIO case to ensure the following
rmi_driver_resume calls in rmi_smbus_resume can succeed.
This behavior was seen on a Lenovo T440p machine that required
a delay of approximately 7-12ms.
The 30ms delay was chosen based on [1].
With this patch the trimmed resume logs look similar to:
```
psmouse serio1: PM: calling serio_resume+0x0/0x8c @ 5399, parent: i8042
[5399] libps2:__ps2_command:316: psmouse serio1: f5 [] - 0/00000000 []
psmouse serio1: PM: serio_resume+0x0/0x8c returned 0 after 3259 usecs
...
rmi4_smbus 0-002c: PM: calling rmi_smb_resume ... @ 5454, parent: i2c-0
...
[5454] i2c_i801:i801_check_post:414: i801_smbus 0000:00:1f.3: No response
smbus_result: i2c-0 a=02c f=0000 c=fd BYTE_DATA rd res=-6
rmi4_smbus 0-002c: failed to get SMBus version number!
rmi4_smbus 0-002c: sleeping to retry getting the SMBus version number
...
rmi4_smbus 0-002c: PM: rmi_smb_resume ... returned 0 after 41351 usecs
```
[1]: https://lore.kernel.org/all/BYAPR03MB47572F2C65E52ED673238D41B2439@BYAPR03MB4757.namprd03.prod.outlook.com/
Signed-off-by: Jeffery Miller <[email protected]>
---
Early boot dmesg include:
```
rmi4_smbus 0-002c: registering SMbus-connected sensor
rmi4_f01 rmi4-00.fn01: found RMI device, manufacturer: Synaptics, product: TM2722-001, fw id: 0
```
The resume order looks correct. The `psmouse serio1` resume returns
before the rmi_smb_resume is called showing the patch from
https://lore.kernel.org/all/[email protected]/
is applied and working for that ordering.
I attempted to try to rule out some interaction between the concurrent
input resume calls for other i8042 devices.
Adding a 7ms delay after psmouse_deactivate which is called in the
preceding psmouse serio1 serio_resume function also allows
this version call to succeed.
If the rmi_smb_probe device_disable_async_suspend patch is applied
it will also avoid this issue. However the time between
the psmouse_deactivate call for serio_resume and rmi_smb_resume
was over 60ms on my test machine. This would naturally be long
enough to avoid this particular delay.
Changes in v2:
- Changed to a single retry of 30ms based on previous feedback.
drivers/input/rmi4/rmi_smbus.c | 22 +++++++++++++++++-----
1 file changed, 17 insertions(+), 5 deletions(-)
diff --git a/drivers/input/rmi4/rmi_smbus.c b/drivers/input/rmi4/rmi_smbus.c
index 4bf0e1df6a4a..b6d90c92e8a2 100644
--- a/drivers/input/rmi4/rmi_smbus.c
+++ b/drivers/input/rmi4/rmi_smbus.c
@@ -45,13 +45,25 @@ static int rmi_smb_get_version(struct rmi_smb_xport *rmi_smb)
int retval;
/* Check if for SMBus new version device by reading version byte. */
- retval = i2c_smbus_read_byte_data(client, SMB_PROTOCOL_VERSION_ADDRESS);
- if (retval < 0) {
+ for (int i = 0; i < 2; i++) {
+ if (i > 0) {
+ dev_warn(&client->dev, "sleeping to retry getting the SMBus version number\n");
+ fsleep(30000);
+ }
+ retval = i2c_smbus_read_byte_data(client,
+ SMB_PROTOCOL_VERSION_ADDRESS);
+ if (retval >= 0)
+ return retval + 1;
+
dev_err(&client->dev, "failed to get SMBus version number!\n");
- return retval;
+ /* There can be a delay on resume where the read returns
+ * -ENXIO. Retry to allow additional time for the read
+ * to become responsive.
+ */
+ if (retval != -ENXIO)
+ break;
}
-
- return retval + 1;
+ return retval;
}
/* SMB block write - wrapper over ic2_smb_write_block */
--
2.41.0.162.gfafddb0af9-goog