| Message ID | 20231104111743.14668-5-hdegoede@redhat.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Jiri Kosina |
| Headers | show |
| Series | HID: i2c-hid: Rework wait for reset to match Windows | expand |
Hi, On Sat, Nov 4, 2023 at 4:17 AM Hans de Goede <hdegoede@redhat.com> wrote: > > @@ -746,8 +752,6 @@ static int i2c_hid_parse(struct hid_device *hid) > > do { > ret = i2c_hid_start_hwreset(ihid); > - if (ret == 0) > - ret = i2c_hid_finish_hwreset(ihid); The mutex contract (talked about in a previous patch) is a little more confusing now. ;-) Feels like it needs a comment somewhere in here so the reader of the code understands that the reset_mutex might or might not be locked here. ...or would it make more sense for the caller to just handle the mutex to make it more obvious. The "I2C_HID_QUIRK_RESET_ON_RESUME" code would need to grab the mutex too, but that really doesn't seem terrible. In fact, I suspect it cleans up your error handling and makes everything cleaner? > if (ret) > msleep(1000); > } while (tries-- > 0 && ret); > @@ -763,9 +767,8 @@ static int i2c_hid_parse(struct hid_device *hid) > i2c_hid_dbg(ihid, "Using a HID report descriptor override\n"); > } else { > rdesc = kzalloc(rsize, GFP_KERNEL); > - > if (!rdesc) { > - dbg_hid("couldn't allocate rdesc memory\n"); > + i2c_hid_abort_hwreset(ihid); > return -ENOMEM; > } > > @@ -776,10 +779,21 @@ static int i2c_hid_parse(struct hid_device *hid) > rdesc, rsize); > if (ret) { > hid_err(hid, "reading report descriptor failed\n"); > + i2c_hid_abort_hwreset(ihid); > goto out; > } > } > > + /* > + * Windows directly reads the report-descriptor after sending reset > + * and then waits for resets completion afterwards. Some touchpads > + * actually wait for the report-descriptor to be read before signalling > + * reset completion. > + */ > + ret = i2c_hid_finish_hwreset(ihid); > + if (ret) > + goto out; Given your new understanding, I wonder if you should start reading the report descriptor even if "use_override" is set? You'd throw away what you read but maybe it would be important to make the touchscreen properly de-assert reset? I guess this is the subject of the next patch... Also: I guess there's the assumption that the touchscreens needing the other caller of the reset functions (I2C_HID_QUIRK_RESET_ON_RESUME) never need to read the report like this? -Doug
Hi, On 11/6/23 19:53, Doug Anderson wrote: > Hi, > > On Sat, Nov 4, 2023 at 4:17 AM Hans de Goede <hdegoede@redhat.com> wrote: >> >> @@ -746,8 +752,6 @@ static int i2c_hid_parse(struct hid_device *hid) >> >> do { >> ret = i2c_hid_start_hwreset(ihid); >> - if (ret == 0) >> - ret = i2c_hid_finish_hwreset(ihid); > > The mutex contract (talked about in a previous patch) is a little more > confusing now. ;-) Feels like it needs a comment somewhere in here so > the reader of the code understands that the reset_mutex might or might > not be locked here. > > ...or would it make more sense for the caller to just handle the mutex > to make it more obvious. The "I2C_HID_QUIRK_RESET_ON_RESUME" code > would need to grab the mutex too, but that really doesn't seem > terrible. In fact, I suspect it cleans up your error handling and > makes everything cleaner? I agree that moving the mutex to the callers would be better, I've just completed this change for v2 of the series. >> if (ret) >> msleep(1000); >> } while (tries-- > 0 && ret); >> @@ -763,9 +767,8 @@ static int i2c_hid_parse(struct hid_device *hid) >> i2c_hid_dbg(ihid, "Using a HID report descriptor override\n"); >> } else { >> rdesc = kzalloc(rsize, GFP_KERNEL); >> - >> if (!rdesc) { >> - dbg_hid("couldn't allocate rdesc memory\n"); >> + i2c_hid_abort_hwreset(ihid); >> return -ENOMEM; >> } >> >> @@ -776,10 +779,21 @@ static int i2c_hid_parse(struct hid_device *hid) >> rdesc, rsize); >> if (ret) { >> hid_err(hid, "reading report descriptor failed\n"); >> + i2c_hid_abort_hwreset(ihid);>> goto out; >> } >> } >> >> + /* >> + * Windows directly reads the report-descriptor after sending reset >> + * and then waits for resets completion afterwards. Some touchpads >> + * actually wait for the report-descriptor to be read before signalling >> + * reset completion. >> + */ >> + ret = i2c_hid_finish_hwreset(ihid); >> + if (ret) >> + goto out; > > Given your new understanding, I wonder if you should start reading the > report descriptor even if "use_override" is set? You'd throw away what > you read but maybe it would be important to make the touchscreen > properly de-assert reset? I guess this is the subject of the next > patch... Right, for the devices where use_override gets set the I2C_HID_QUIRK_NO_IRQ_AFTER_RESET is also always set, so i2c-hid-core basically does not expect reset-complete to be signalled via IRQ on these devices. Since these devices are all kinda weird I have chosen to just preserve the old behavior (1) there to avoid regressions 1) just doing a msleep(100) instead of waiting for the IRQ > Also: I guess there's the assumption that the touchscreens needing the > other caller of the reset functions (I2C_HID_QUIRK_RESET_ON_RESUME) > never need to read the report like this? That is correct, only a few devices need to read the report like this for the reset complete IRQ to happen and since I2C_HID_QUIRK_RESET_ON_RESUME is only set on a few devices and we know it works there already the assumption indeed is that on those devices the reading of the report after reset is not necessary. Regards, Hans
diff --git a/drivers/hid/i2c-hid/i2c-hid-core.c b/drivers/hid/i2c-hid/i2c-hid-core.c index f029ddce4766..3bd0c3d77d99 100644 --- a/drivers/hid/i2c-hid/i2c-hid-core.c +++ b/drivers/hid/i2c-hid/i2c-hid-core.c @@ -502,6 +502,12 @@ static int i2c_hid_finish_hwreset(struct i2c_hid *ihid) return ret; } +static void i2c_hid_abort_hwreset(struct i2c_hid *ihid) +{ + clear_bit(I2C_HID_RESET_PENDING, &ihid->flags); + mutex_unlock(&ihid->reset_lock); +} + static void i2c_hid_get_input(struct i2c_hid *ihid) { u16 size = le16_to_cpu(ihid->hdesc.wMaxInputLength); @@ -746,8 +752,6 @@ static int i2c_hid_parse(struct hid_device *hid) do { ret = i2c_hid_start_hwreset(ihid); - if (ret == 0) - ret = i2c_hid_finish_hwreset(ihid); if (ret) msleep(1000); } while (tries-- > 0 && ret); @@ -763,9 +767,8 @@ static int i2c_hid_parse(struct hid_device *hid) i2c_hid_dbg(ihid, "Using a HID report descriptor override\n"); } else { rdesc = kzalloc(rsize, GFP_KERNEL); - if (!rdesc) { - dbg_hid("couldn't allocate rdesc memory\n"); + i2c_hid_abort_hwreset(ihid); return -ENOMEM; } @@ -776,10 +779,21 @@ static int i2c_hid_parse(struct hid_device *hid) rdesc, rsize); if (ret) { hid_err(hid, "reading report descriptor failed\n"); + i2c_hid_abort_hwreset(ihid); goto out; } } + /* + * Windows directly reads the report-descriptor after sending reset + * and then waits for resets completion afterwards. Some touchpads + * actually wait for the report-descriptor to be read before signalling + * reset completion. + */ + ret = i2c_hid_finish_hwreset(ihid); + if (ret) + goto out; + i2c_hid_dbg(ihid, "Report Descriptor: %*ph\n", rsize, rdesc); ret = hid_parse_report(hid, rdesc, rsize);
A recent bug made me look at Microsoft's i2c-hid docs again and I noticed the following: """ 4. Issue a RESET (Host Initiated Reset) to the Device. 5. Retrieve report descriptor from the device. Note: Steps 4 and 5 may be done in parallel to optimize for time on I²C. Since report descriptors are (a) static and (b) quite long, Windows 8 may issue a request for 5 while it is waiting for a response from the device on 4. """ Which made me think that maybe on some touchpads the reset ack is delayed till after the report descriptor is read ? Testing a T-BAO Tbook Air 12.5 with a 0911:5288 (SIPODEV SP1064?) touchpad, for which the I2C_HID_QUIRK_NO_IRQ_AFTER_RESET quirk was first introduced, shows that about 1 ms after the report descriptor read finishes the reset indeed does get acked. Move the waiting for the ack to after reading the report-descriptor, so that the I2C_HID_QUIRK_NO_IRQ_AFTER_RESET quirk is no longer necessary (on this model). While at it drop the dbg_hid() for a malloc failure, malloc failures already get logged extensively by malloc itself. Link: https://bugzilla.redhat.com/show_bug.cgi?id=2247751 Signed-off-by: Hans de Goede <hdegoede@redhat.com> --- drivers/hid/i2c-hid/i2c-hid-core.c | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-)