| Message ID | 20210405202745.16777-1-hdegoede@redhat.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | e479187748a8f151a85116a7091c599b121fdea5 |
| Headers | show |
| Series | [resend] Input: silead - Add workaround for x86 BIOS-es which bring the Silead chip up in a stuck state | expand |
On Mon, Apr 05, 2021 at 10:27:45PM +0200, Hans de Goede wrote: > Some buggy BIOS-es bring up the touchscreen-controller in a stuck > state where it blocks the I2C bus. Specifically this happens on > the Jumper EZpad 7 tablet model. > > After much poking at this problem I have found that the following steps > are necessary to unstuck the chip / bus: > > 1. Turn off the Silead chip. > 2. Try to do an I2C transfer with the chip, this will fail in response to > which the I2C-bus-driver will call: i2c_recover_bus() which will unstuck > the I2C-bus. Note the unstuck-ing of the I2C bus only works if we first > drop the chip of the bus by turning it off. > 3. Turn the chip back on. > > On the x86/ACPI systems were this problem is seen, step 1. and 3. require > making ACPI calls and dealing with ACPI Power Resources. This commit adds > a workaround which runtime-suspends the chip to turn it off, leaving it up > to the ACPI subsystem to deal with all the ACPI specific details. > > There is no good way to detect this bug, so the workaround gets activated > by a new "silead,stuck-controller-bug" boolean device-property. Since this > is only used on x86/ACPI, this will be set by model specific device-props > set by drivers/platform/x86/touchscreen_dmi.c. Therefor this new > device-property is not documented in the DT-bindings. > > Dmesg will contain the following messages on systems where the workaround > is activated: > > [ 54.309029] silead_ts i2c-MSSL1680:00: [Firmware Bug]: Stuck I2C bus: please ignore the next 'controller timed out' error > [ 55.373593] i2c_designware 808622C1:04: controller timed out > [ 55.582186] silead_ts i2c-MSSL1680:00: Silead chip ID: 0x80360000 > > Signed-off-by: Hans de Goede <hdegoede@redhat.com> Applied, thank you.
diff --git a/drivers/input/touchscreen/silead.c b/drivers/input/touchscreen/silead.c index 8fa2f3b7cfd8..4df0f93dbb56 100644 --- a/drivers/input/touchscreen/silead.c +++ b/drivers/input/touchscreen/silead.c @@ -20,6 +20,7 @@ #include <linux/input/mt.h> #include <linux/input/touchscreen.h> #include <linux/pm.h> +#include <linux/pm_runtime.h> #include <linux/irq.h> #include <linux/regulator/consumer.h> @@ -335,10 +336,8 @@ static int silead_ts_get_id(struct i2c_client *client) error = i2c_smbus_read_i2c_block_data(client, SILEAD_REG_ID, sizeof(chip_id), (u8 *)&chip_id); - if (error < 0) { - dev_err(&client->dev, "Chip ID read error %d\n", error); + if (error < 0) return error; - } data->chip_id = le32_to_cpu(chip_id); dev_info(&client->dev, "Silead chip ID: 0x%8X", data->chip_id); @@ -351,12 +350,44 @@ static int silead_ts_setup(struct i2c_client *client) int error; u32 status; + /* + * Some buggy BIOS-es bring up the chip in a stuck state where it blocks the I2C bus. + * The following steps are necessary to unstuck the chip / bus: + * 1. Turn off the Silead chip. + * 2. Try to do an I2C transfer with the chip, this will fail in response to + * which the I2C-bus-driver will call: i2c_recover_bus() which will unstuck + * the I2C-bus. Note the unstuck-ing of the I2C bus only works if we first + * drop the chip of the bus by turning it off. + * 3. Turn the chip back on. + * + * On the x86/ACPI systems were this problem is seen, step 1. and 3. require + * making ACPI calls and dealing with ACPI Power Resources. The workaround below + * runtime-suspends the chip to turn it off, leaving it up to the ACPI subsystem + * to deal with this. + */ + if (device_property_read_bool(&client->dev, "silead,stuck-controller-bug")) { + pm_runtime_set_active(&client->dev); + pm_runtime_enable(&client->dev); + pm_runtime_allow(&client->dev); + + pm_runtime_suspend(&client->dev); + + dev_warn(&client->dev, FW_BUG "Stuck I2C bus: please ignore the next 'controller timed out' error\n"); + silead_ts_get_id(client); + + /* The forbid will also resume the device */ + pm_runtime_forbid(&client->dev); + pm_runtime_disable(&client->dev); + } + silead_ts_set_power(client, SILEAD_POWER_OFF); silead_ts_set_power(client, SILEAD_POWER_ON); error = silead_ts_get_id(client); - if (error) + if (error) { + dev_err(&client->dev, "Chip ID read error %d\n", error); return error; + } error = silead_ts_init(client); if (error)
Some buggy BIOS-es bring up the touchscreen-controller in a stuck state where it blocks the I2C bus. Specifically this happens on the Jumper EZpad 7 tablet model. After much poking at this problem I have found that the following steps are necessary to unstuck the chip / bus: 1. Turn off the Silead chip. 2. Try to do an I2C transfer with the chip, this will fail in response to which the I2C-bus-driver will call: i2c_recover_bus() which will unstuck the I2C-bus. Note the unstuck-ing of the I2C bus only works if we first drop the chip of the bus by turning it off. 3. Turn the chip back on. On the x86/ACPI systems were this problem is seen, step 1. and 3. require making ACPI calls and dealing with ACPI Power Resources. This commit adds a workaround which runtime-suspends the chip to turn it off, leaving it up to the ACPI subsystem to deal with all the ACPI specific details. There is no good way to detect this bug, so the workaround gets activated by a new "silead,stuck-controller-bug" boolean device-property. Since this is only used on x86/ACPI, this will be set by model specific device-props set by drivers/platform/x86/touchscreen_dmi.c. Therefor this new device-property is not documented in the DT-bindings. Dmesg will contain the following messages on systems where the workaround is activated: [ 54.309029] silead_ts i2c-MSSL1680:00: [Firmware Bug]: Stuck I2C bus: please ignore the next 'controller timed out' error [ 55.373593] i2c_designware 808622C1:04: controller timed out [ 55.582186] silead_ts i2c-MSSL1680:00: Silead chip ID: 0x80360000 Signed-off-by: Hans de Goede <hdegoede@redhat.com> --- Changes in v2: - Fix brown paper-bag bug where v1 did not compile --- drivers/input/touchscreen/silead.c | 39 +++++++++++++++++++++++++++--- 1 file changed, 35 insertions(+), 4 deletions(-)