Message ID | 20230510192731.300786-1-jefferymiller@google.com |
---|---|
State | New |
Headers | show |
Series | Input: synaptics-rmi4 - retry reading SMBus version on resume | expand |
> From: Jeffery Miller <jefferymiller@google.com> > Sent: Friday, May 12, 2023 12:37 > To: Dmitry Torokhov <dmitry.torokhov@gmail.com> > Cc: Andrew Duggan <andrew@duggan.us>; Jonathan Denose > <jdenose@chromium.org>; jdenose@google.com; Lyude Paul > <lyude@redhat.com>; loic.poulain@linaro.org; > benjamin.tissoires@redhat.com; Andrew Duggan > <aduggan@synaptics.com>; Jonathan Cameron > <Jonathan.Cameron@huawei.com>; Maximilian Luz > <luzmaximilian@gmail.com>; Miguel Ojeda <ojeda@kernel.org>; Uwe Kleine- > König <u.kleine-koenig@pengutronix.de>; linux-input@vger.kernel.org; > linux-kernel@vger.kernel.org > Subject: Re: [PATCH] Input: synaptics-rmi4 - retry reading SMBus version on > resume > > CAUTION: Email originated externally, do not click links or open attachments > unless you recognize the sender and know the content is safe. > > > On Wed, May 10, 2023 at 3:06 PM Dmitry Torokhov > <dmitry.torokhov@gmail.com> wrote: > > > > > > I am not really fond of adding random repeats in the code base. > > Andrew, do you know if the Synaptics device needs certain delay when > > switching to SMbus mode? > > > That's reasonable. It's true this is a sleep to workaround rmi_smbus' > expectation that > it is able to successfully read from the i801 i2c smbus addr 0x2c on resume > when it cannot. > > I do not know why the i801 i2c addr 0x2c is not responding on resume. > Infrequently it won't respond on boot during the initial > psmouse_smbus_create_companion but that is less noticeable since it will > stay in ps/2 mode and just lack features. > > I do know adding a sleep in-between the psmouse deactivate call and > rmi_smbus's attempts to read from the 0x2c addr allow it to succeed on this > machine. I do not know the details as to why however. Our Windows driver does a 30 ms delay between sending the disable command and trying to communicate over SMBus. I don’t know specifically why this delay is needed, but this is how we handle this case. > > I can apply workarounds until the underlying issue can be identified and > properly resolved. > > I can apply patches and provide any sort of debugging or extra information > that might be useful to anyone familiar with these devices. > > The smbus controller in this particular machine is: > 00:1f.3 SMBus [0c05]: Intel Corporation 8 Series/C220 Series Chipset Family > SMBus Controller [8086:8c22] (rev 04) > > Perhaps there's some way to detect when the addr is available later and have > it trigger a new probe similar to how psmouse_smbus_notifier is triggered > when the i801 bus becomes available? > > Thank You, > Jeff Andrew
diff --git a/drivers/input/rmi4/rmi_smbus.c b/drivers/input/rmi4/rmi_smbus.c index 4bf0e1df6a4a..386e80ae141b 100644 --- a/drivers/input/rmi4/rmi_smbus.c +++ b/drivers/input/rmi4/rmi_smbus.c @@ -43,15 +43,26 @@ static int rmi_smb_get_version(struct rmi_smb_xport *rmi_smb) { struct i2c_client *client = rmi_smb->client; int retval; + int tries = 0; /* 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) { - dev_err(&client->dev, "failed to get SMBus version number!\n"); - return retval; - } + do { + if (tries > 0) { + dev_warn(&client->dev, "sleeping to retry getting the SMBus version number\n"); + fsleep(5000); + } + retval = i2c_smbus_read_byte_data(client, + SMB_PROTOCOL_VERSION_ADDRESS); + if (retval >= 0) + return retval + 1; - return retval + 1; + dev_err(&client->dev, "failed to get SMBus version number!\n"); + /* On resume the read of the version can + * momentarily return -ENXIO. + * Retry to allow additional time for it to succeed. + */ + } while (retval == -ENXIO && tries++ < 5); + return retval; } /* SMB block write - wrapper over ic2_smb_write_block */
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. Adding a retry loop ensures that after rmi_smb_reset returns 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 retry limit of 5 is chosen to be larger than this observed delay. 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 21351 usecs ``` Signed-off-by: Jeffery Miller <jefferymiller@google.com> --- 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/89456fcd-a113-4c82-4b10-a9bcaefac68f@google.com/ 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. drivers/input/rmi4/rmi_smbus.c | 23 +++++++++++++++++------ 1 file changed, 17 insertions(+), 6 deletions(-)