| Message ID | a07922bd-4550-41d8-a7cd-8943baf6f8fb@siemens.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | USB: serial: pl2303: account for deficits of clones | expand |
On Sun, Sep 01, 2024 at 11:11:29PM +0200, Jan Kiszka wrote: > From: Jan Kiszka <jan.kiszka@siemens.com> > > There are apparently incomplete clones of the HXD type chip in use. > Those return -EPIPE on GET_LINE_REQUEST and BREAK_REQUEST. Avoid > flooding the kernel log with those errors. Rather use the > line_settings cache for GET_LINE_REQUEST and signal missing support by > returning -ENOTTY from pl2303_set_break. > > Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> > --- > drivers/usb/serial/pl2303.c | 13 +++++++------ > 1 file changed, 7 insertions(+), 6 deletions(-) > > diff --git a/drivers/usb/serial/pl2303.c b/drivers/usb/serial/pl2303.c > index d93f5d584557..04cafa819390 100644 > --- a/drivers/usb/serial/pl2303.c > +++ b/drivers/usb/serial/pl2303.c > @@ -731,12 +731,13 @@ static int pl2303_get_line_request(struct usb_serial_port *port, > GET_LINE_REQUEST, GET_LINE_REQUEST_TYPE, > 0, 0, buf, 7, 100); > if (ret != 7) { > - dev_err(&port->dev, "%s - failed: %d\n", __func__, ret); > + struct pl2303_private *priv = usb_get_serial_port_data(port); > > - if (ret >= 0) > - ret = -EIO; > + dev_dbg(&port->dev, "%s - failed, falling back on cache: %d\n", > + __func__, ret); > + memcpy(buf, priv->line_settings, 7); Ugh, how is this device working in other operating systems? > > - return ret; > + return 0; > } > > dev_dbg(&port->dev, "%s - %7ph\n", __func__, buf); > @@ -1078,8 +1079,8 @@ static int pl2303_set_break(struct usb_serial_port *port, bool enable) > BREAK_REQUEST, BREAK_REQUEST_TYPE, state, > 0, NULL, 0, 100); > if (result) { > - dev_err(&port->dev, "error sending break = %d\n", result); > - return result; > + dev_dbg(&port->dev, "error sending break = %d\n", result); > + return -ENOTTY; Are you sure that ENOTTY is correct here? Why not just send back -EINVAL or something like that telling userspace that this is not allowed for this device? thanks, greg k-h
On 02.09.24 08:06, Greg Kroah-Hartman wrote: > On Sun, Sep 01, 2024 at 11:11:29PM +0200, Jan Kiszka wrote: >> From: Jan Kiszka <jan.kiszka@siemens.com> >> >> There are apparently incomplete clones of the HXD type chip in use. >> Those return -EPIPE on GET_LINE_REQUEST and BREAK_REQUEST. Avoid >> flooding the kernel log with those errors. Rather use the >> line_settings cache for GET_LINE_REQUEST and signal missing support by >> returning -ENOTTY from pl2303_set_break. >> >> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> >> --- >> drivers/usb/serial/pl2303.c | 13 +++++++------ >> 1 file changed, 7 insertions(+), 6 deletions(-) >> >> diff --git a/drivers/usb/serial/pl2303.c b/drivers/usb/serial/pl2303.c >> index d93f5d584557..04cafa819390 100644 >> --- a/drivers/usb/serial/pl2303.c >> +++ b/drivers/usb/serial/pl2303.c >> @@ -731,12 +731,13 @@ static int pl2303_get_line_request(struct usb_serial_port *port, >> GET_LINE_REQUEST, GET_LINE_REQUEST_TYPE, >> 0, 0, buf, 7, 100); >> if (ret != 7) { >> - dev_err(&port->dev, "%s - failed: %d\n", __func__, ret); >> + struct pl2303_private *priv = usb_get_serial_port_data(port); >> >> - if (ret >= 0) >> - ret = -EIO; >> + dev_dbg(&port->dev, "%s - failed, falling back on cache: %d\n", >> + __func__, ret); >> + memcpy(buf, priv->line_settings, 7); > > Ugh, how is this device working in other operating systems? > I don't know. Also, the last downstream driver Prolific posted [1] didn't point to this specialty. I opened a few of those adapter cables (some received from [2], others from a different source), and the chips were not labeled (in contrast to the picture from [2]). So I'm suspecting clones to be in play. Reminds me of the quest for sane WIFI USB adapters. >> >> - return ret; >> + return 0; >> } >> >> dev_dbg(&port->dev, "%s - %7ph\n", __func__, buf); >> @@ -1078,8 +1079,8 @@ static int pl2303_set_break(struct usb_serial_port *port, bool enable) >> BREAK_REQUEST, BREAK_REQUEST_TYPE, state, >> 0, NULL, 0, 100); >> if (result) { >> - dev_err(&port->dev, "error sending break = %d\n", result); >> - return result; >> + dev_dbg(&port->dev, "error sending break = %d\n", result); >> + return -ENOTTY; > > Are you sure that ENOTTY is correct here? Why not just send back > -EINVAL or something like that telling userspace that this is not > allowed for this device? I was copying from serial_break() which now returns -ENOTTY if the handler is NULL. If you prefer -EINVAL, I can change. Just looking for a consistent code. Jan [1] https://prolificusa.com/wp-content/uploads/2021/01/PL2303G_Linux_Driver_v1.0.6.zip [2] https://www.berrybase.de/en/usb-ttl/uart/rs232-adapter-mit-pl2303hx-chipsatz
On Mon, Sep 02, 2024 at 08:28:42AM +0200, Jan Kiszka wrote: > On 02.09.24 08:06, Greg Kroah-Hartman wrote: > > On Sun, Sep 01, 2024 at 11:11:29PM +0200, Jan Kiszka wrote: > >> From: Jan Kiszka <jan.kiszka@siemens.com> > >> > >> There are apparently incomplete clones of the HXD type chip in use. > >> Those return -EPIPE on GET_LINE_REQUEST and BREAK_REQUEST. Avoid > >> flooding the kernel log with those errors. Rather use the > >> line_settings cache for GET_LINE_REQUEST and signal missing support by > >> returning -ENOTTY from pl2303_set_break. > >> > >> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> > >> --- > >> drivers/usb/serial/pl2303.c | 13 +++++++------ > >> 1 file changed, 7 insertions(+), 6 deletions(-) > >> > >> diff --git a/drivers/usb/serial/pl2303.c b/drivers/usb/serial/pl2303.c > >> index d93f5d584557..04cafa819390 100644 > >> --- a/drivers/usb/serial/pl2303.c > >> +++ b/drivers/usb/serial/pl2303.c > >> @@ -731,12 +731,13 @@ static int pl2303_get_line_request(struct usb_serial_port *port, > >> GET_LINE_REQUEST, GET_LINE_REQUEST_TYPE, > >> 0, 0, buf, 7, 100); > >> if (ret != 7) { > >> - dev_err(&port->dev, "%s - failed: %d\n", __func__, ret); > >> + struct pl2303_private *priv = usb_get_serial_port_data(port); > >> > >> - if (ret >= 0) > >> - ret = -EIO; > >> + dev_dbg(&port->dev, "%s - failed, falling back on cache: %d\n", > >> + __func__, ret); > >> + memcpy(buf, priv->line_settings, 7); > > > > Ugh, how is this device working in other operating systems? > > > > I don't know. Also, the last downstream driver Prolific posted [1] > didn't point to this specialty. Given that prolific's windows driver is known to explicitly attempt to disable clone devices, I would be amazed if it supported it at all :) > I opened a few of those adapter cables (some received from [2], others > from a different source), and the chips were not labeled (in contrast to > the picture from [2]). So I'm suspecting clones to be in play. Reminds > me of the quest for sane WIFI USB adapters. This chip has had lots of different cloned devices over the years, which is pretty odd given that the thing is so cheap to start with. Anyway, we live with it and move on... > >> > >> - return ret; > >> + return 0; > >> } > >> > >> dev_dbg(&port->dev, "%s - %7ph\n", __func__, buf); > >> @@ -1078,8 +1079,8 @@ static int pl2303_set_break(struct usb_serial_port *port, bool enable) > >> BREAK_REQUEST, BREAK_REQUEST_TYPE, state, > >> 0, NULL, 0, 100); > >> if (result) { > >> - dev_err(&port->dev, "error sending break = %d\n", result); > >> - return result; > >> + dev_dbg(&port->dev, "error sending break = %d\n", result); > >> + return -ENOTTY; > > > > Are you sure that ENOTTY is correct here? Why not just send back > > -EINVAL or something like that telling userspace that this is not > > allowed for this device? > > I was copying from serial_break() which now returns -ENOTTY if the > handler is NULL. If you prefer -EINVAL, I can change. Just looking for a > consistent code. Ah, yeah, that's tricky. I'll let Johan pick one :) thanks, greg k-h
On Sun, Sep 01, 2024 at 11:11:29PM +0200, Jan Kiszka wrote: > From: Jan Kiszka <jan.kiszka@siemens.com> > > There are apparently incomplete clones of the HXD type chip in use. > Those return -EPIPE on GET_LINE_REQUEST and BREAK_REQUEST. Avoid > flooding the kernel log with those errors. Rather use the > line_settings cache for GET_LINE_REQUEST and signal missing support by > returning -ENOTTY from pl2303_set_break. This looks mostly fine to me, but could you please try to make these changes only apply to the clones or, since those probably cannot be detected reliably, HXD? What is the 'lsusb -v' for these devices? > Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> > --- > drivers/usb/serial/pl2303.c | 13 +++++++------ > 1 file changed, 7 insertions(+), 6 deletions(-) > > diff --git a/drivers/usb/serial/pl2303.c b/drivers/usb/serial/pl2303.c > index d93f5d584557..04cafa819390 100644 > --- a/drivers/usb/serial/pl2303.c > +++ b/drivers/usb/serial/pl2303.c > @@ -731,12 +731,13 @@ static int pl2303_get_line_request(struct usb_serial_port *port, > GET_LINE_REQUEST, GET_LINE_REQUEST_TYPE, > 0, 0, buf, 7, 100); > if (ret != 7) { > - dev_err(&port->dev, "%s - failed: %d\n", __func__, ret); > + struct pl2303_private *priv = usb_get_serial_port_data(port); > > - if (ret >= 0) > - ret = -EIO; > + dev_dbg(&port->dev, "%s - failed, falling back on cache: %d\n", > + __func__, ret); > + memcpy(buf, priv->line_settings, 7); > > - return ret; > + return 0; > } Looking at the driver, it seems like we could move the only call to this function to probe, and perhaps then an error message for cloned devices is not such a big deal. But even that could be suppressed based on the type. Or we could use this call failing to flag the device as a likely clone and use that flag to also skip break signalling completely. > dev_dbg(&port->dev, "%s - %7ph\n", __func__, buf); > @@ -1078,8 +1079,8 @@ static int pl2303_set_break(struct usb_serial_port *port, bool enable) > BREAK_REQUEST, BREAK_REQUEST_TYPE, state, > 0, NULL, 0, 100); > if (result) { > - dev_err(&port->dev, "error sending break = %d\n", result); > - return result; > + dev_dbg(&port->dev, "error sending break = %d\n", result); > + return -ENOTTY; > } And similar here, the log level could depend on the type, unless the function should just bail out early for clones. > > return 0; Johan
On 09.09.24 14:32, Johan Hovold wrote: > On Sun, Sep 01, 2024 at 11:11:29PM +0200, Jan Kiszka wrote: >> From: Jan Kiszka <jan.kiszka@siemens.com> >> >> There are apparently incomplete clones of the HXD type chip in use. >> Those return -EPIPE on GET_LINE_REQUEST and BREAK_REQUEST. Avoid >> flooding the kernel log with those errors. Rather use the >> line_settings cache for GET_LINE_REQUEST and signal missing support by >> returning -ENOTTY from pl2303_set_break. > > This looks mostly fine to me, but could you please try to make these > changes only apply to the clones or, since those probably cannot be > detected reliably, HXD? > OK. Is there a way to switch between dev_err and dev_dbg without duplicating the line? > What is the 'lsusb -v' for these devices? Bus 001 Device 019: ID 067b:2303 Prolific Technology, Inc. PL2303 Serial Port / Mobile Action MA-8910P Couldn't open device, some information will be missing Device Descriptor: bLength 18 bDescriptorType 1 bcdUSB 1.10 bDeviceClass 0 bDeviceSubClass 0 bDeviceProtocol 0 bMaxPacketSize0 64 idVendor 0x067b Prolific Technology, Inc. idProduct 0x2303 PL2303 Serial Port / Mobile Action MA-8910P bcdDevice 4.00 iManufacturer 1 Prolific Technology Inc. iProduct 2 USB-Serial Controller D iSerial 0 bNumConfigurations 1 Configuration Descriptor: bLength 9 bDescriptorType 2 wTotalLength 0x0027 bNumInterfaces 1 bConfigurationValue 1 iConfiguration 0 bmAttributes 0x80 (Bus Powered) MaxPower 100mA Interface Descriptor: bLength 9 bDescriptorType 4 bInterfaceNumber 0 bAlternateSetting 0 bNumEndpoints 3 bInterfaceClass 255 Vendor Specific Class bInterfaceSubClass 0 bInterfaceProtocol 0 iInterface 0 Endpoint Descriptor: bLength 7 bDescriptorType 5 bEndpointAddress 0x81 EP 1 IN bmAttributes 3 Transfer Type Interrupt Synch Type None Usage Type Data wMaxPacketSize 0x000a 1x 10 bytes bInterval 1 Endpoint Descriptor: bLength 7 bDescriptorType 5 bEndpointAddress 0x02 EP 2 OUT bmAttributes 2 Transfer Type Bulk Synch Type None Usage Type Data wMaxPacketSize 0x0040 1x 64 bytes bInterval 0 Endpoint Descriptor: bLength 7 bDescriptorType 5 bEndpointAddress 0x83 EP 3 IN bmAttributes 2 Transfer Type Bulk Synch Type None Usage Type Data wMaxPacketSize 0x0040 1x 64 bytes bInterval 0 Unfortunately, I don't have a HXD device around that behaves and is likely no clone, thus have no reference. > >> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> >> --- >> drivers/usb/serial/pl2303.c | 13 +++++++------ >> 1 file changed, 7 insertions(+), 6 deletions(-) >> >> diff --git a/drivers/usb/serial/pl2303.c b/drivers/usb/serial/pl2303.c >> index d93f5d584557..04cafa819390 100644 >> --- a/drivers/usb/serial/pl2303.c >> +++ b/drivers/usb/serial/pl2303.c >> @@ -731,12 +731,13 @@ static int pl2303_get_line_request(struct usb_serial_port *port, >> GET_LINE_REQUEST, GET_LINE_REQUEST_TYPE, >> 0, 0, buf, 7, 100); >> if (ret != 7) { >> - dev_err(&port->dev, "%s - failed: %d\n", __func__, ret); >> + struct pl2303_private *priv = usb_get_serial_port_data(port); >> >> - if (ret >= 0) >> - ret = -EIO; >> + dev_dbg(&port->dev, "%s - failed, falling back on cache: %d\n", >> + __func__, ret); >> + memcpy(buf, priv->line_settings, 7); >> >> - return ret; >> + return 0; >> } > > Looking at the driver, it seems like we could move the only call to this > function to probe, and perhaps then an error message for cloned devices Nope, this is called on every pl2303_set_termios, thus is even under user control. > is not such a big deal. But even that could be suppressed based on the > type. > > Or we could use this call failing to flag the device as a likely clone > and use that flag to also skip break signalling completely. Oh, you meant the function below? But that is also in user hands (as well). Flagging after the first call is theoretically possible - but is it worth the related effort? I doubt that a bit. Jan > >> dev_dbg(&port->dev, "%s - %7ph\n", __func__, buf); >> @@ -1078,8 +1079,8 @@ static int pl2303_set_break(struct usb_serial_port *port, bool enable) >> BREAK_REQUEST, BREAK_REQUEST_TYPE, state, >> 0, NULL, 0, 100); >> if (result) { >> - dev_err(&port->dev, "error sending break = %d\n", result); >> - return result; >> + dev_dbg(&port->dev, "error sending break = %d\n", result); >> + return -ENOTTY; >> } > > And similar here, the log level could depend on the type, unless the > function should just bail out early for clones. > >> >> return 0; > > Johan
On Mon, Sep 09, 2024 at 02:52:25PM +0200, Jan Kiszka wrote: > On 09.09.24 14:32, Johan Hovold wrote: > > On Sun, Sep 01, 2024 at 11:11:29PM +0200, Jan Kiszka wrote: > >> From: Jan Kiszka <jan.kiszka@siemens.com> > >> > >> There are apparently incomplete clones of the HXD type chip in use. > >> Those return -EPIPE on GET_LINE_REQUEST and BREAK_REQUEST. Avoid > >> flooding the kernel log with those errors. Rather use the > >> line_settings cache for GET_LINE_REQUEST and signal missing support by > >> returning -ENOTTY from pl2303_set_break. > > > > This looks mostly fine to me, but could you please try to make these > > changes only apply to the clones or, since those probably cannot be > > detected reliably, HXD? > > > > OK. Is there a way to switch between dev_err and dev_dbg without > duplicating the line? Perhaps, did you check? But I think we should go with adding a flag and suppressing the known broken calls completely post probe. > > What is the 'lsusb -v' for these devices? > > Bus 001 Device 019: ID 067b:2303 Prolific Technology, Inc. PL2303 Serial > Port / Mobile Action MA-8910P > Couldn't open device, some information will be missing > Device Descriptor: > bLength 18 > bDescriptorType 1 > bcdUSB 1.10 > bDeviceClass 0 > bDeviceSubClass 0 > bDeviceProtocol 0 > bMaxPacketSize0 64 > idVendor 0x067b Prolific Technology, Inc. > idProduct 0x2303 PL2303 Serial Port / Mobile Action MA-8910P > bcdDevice 4.00 So this would be detected as an HXD by the current driver. Not sure what else could be used except possibly the product string and the fact that these requests fail to determine if it's a legit HXD. > >> @@ -731,12 +731,13 @@ static int pl2303_get_line_request(struct usb_serial_port *port, > >> GET_LINE_REQUEST, GET_LINE_REQUEST_TYPE, > >> 0, 0, buf, 7, 100); > >> if (ret != 7) { > >> - dev_err(&port->dev, "%s - failed: %d\n", __func__, ret); > >> + struct pl2303_private *priv = usb_get_serial_port_data(port); > >> > >> - if (ret >= 0) > >> - ret = -EIO; > >> + dev_dbg(&port->dev, "%s - failed, falling back on cache: %d\n", > >> + __func__, ret); > >> + memcpy(buf, priv->line_settings, 7); > >> > >> - return ret; > >> + return 0; > >> } > > > > Looking at the driver, it seems like we could move the only call to this > > function to probe, and perhaps then an error message for cloned devices > > Nope, this is called on every pl2303_set_termios, thus is even under > user control. What do you mean by "nope"? I'm saying that it looks like it may be possible to move this call to probe() given how it is used currently. Or you can just add an additional call to probe() without the dev_err() and use that for clone detection. > > is not such a big deal. But even that could be suppressed based on the > > type. > > > > Or we could use this call failing to flag the device as a likely clone > > and use that flag to also skip break signalling completely. > > Oh, you meant the function below? But that is also in user hands (as well). > > Flagging after the first call is theoretically possible - but is it > worth the related effort? I doubt that a bit. I'm saying that we can issue the get_line_settings request during probe() (for HXD) and if it fails, we treat the device as a clone and skip the requests that are not supported completely. Johan
On 09.09.24 15:43, Johan Hovold wrote: > On Mon, Sep 09, 2024 at 02:52:25PM +0200, Jan Kiszka wrote: >> On 09.09.24 14:32, Johan Hovold wrote: >>> On Sun, Sep 01, 2024 at 11:11:29PM +0200, Jan Kiszka wrote: >>>> From: Jan Kiszka <jan.kiszka@siemens.com> >>>> >>>> There are apparently incomplete clones of the HXD type chip in use. >>>> Those return -EPIPE on GET_LINE_REQUEST and BREAK_REQUEST. Avoid >>>> flooding the kernel log with those errors. Rather use the >>>> line_settings cache for GET_LINE_REQUEST and signal missing support by >>>> returning -ENOTTY from pl2303_set_break. >>> >>> This looks mostly fine to me, but could you please try to make these >>> changes only apply to the clones or, since those probably cannot be >>> detected reliably, HXD? >>> >> >> OK. Is there a way to switch between dev_err and dev_dbg without >> duplicating the line? > > Perhaps, did you check? But I think we should go with adding a flag and > suppressing the known broken calls completely post probe. > >>> What is the 'lsusb -v' for these devices? >> >> Bus 001 Device 019: ID 067b:2303 Prolific Technology, Inc. PL2303 Serial >> Port / Mobile Action MA-8910P >> Couldn't open device, some information will be missing >> Device Descriptor: >> bLength 18 >> bDescriptorType 1 >> bcdUSB 1.10 >> bDeviceClass 0 >> bDeviceSubClass 0 >> bDeviceProtocol 0 >> bMaxPacketSize0 64 >> idVendor 0x067b Prolific Technology, Inc. >> idProduct 0x2303 PL2303 Serial Port / Mobile Action MA-8910P >> bcdDevice 4.00 > > So this would be detected as an HXD by the current driver. Not sure what > else could be used except possibly the product string and the fact that > these requests fail to determine if it's a legit HXD. > >>>> @@ -731,12 +731,13 @@ static int pl2303_get_line_request(struct usb_serial_port *port, >>>> GET_LINE_REQUEST, GET_LINE_REQUEST_TYPE, >>>> 0, 0, buf, 7, 100); >>>> if (ret != 7) { >>>> - dev_err(&port->dev, "%s - failed: %d\n", __func__, ret); >>>> + struct pl2303_private *priv = usb_get_serial_port_data(port); >>>> >>>> - if (ret >= 0) >>>> - ret = -EIO; >>>> + dev_dbg(&port->dev, "%s - failed, falling back on cache: %d\n", >>>> + __func__, ret); >>>> + memcpy(buf, priv->line_settings, 7); >>>> >>>> - return ret; >>>> + return 0; >>>> } >>> >>> Looking at the driver, it seems like we could move the only call to this >>> function to probe, and perhaps then an error message for cloned devices >> >> Nope, this is called on every pl2303_set_termios, thus is even under >> user control. > > What do you mean by "nope"? I'm saying that it looks like it may be > possible to move this call to probe() given how it is used currently. > > Or you can just add an additional call to probe() without the dev_err() > and use that for clone detection. > >>> is not such a big deal. But even that could be suppressed based on the >>> type. >>> >>> Or we could use this call failing to flag the device as a likely clone >>> and use that flag to also skip break signalling completely. >> >> Oh, you meant the function below? But that is also in user hands (as well). >> >> Flagging after the first call is theoretically possible - but is it >> worth the related effort? I doubt that a bit. > > I'm saying that we can issue the get_line_settings request during > probe() (for HXD) and if it fails, we treat the device as a clone and > skip the requests that are not supported completely. OK, now I get the plan. Let me see... Jan
diff --git a/drivers/usb/serial/pl2303.c b/drivers/usb/serial/pl2303.c index d93f5d584557..04cafa819390 100644 --- a/drivers/usb/serial/pl2303.c +++ b/drivers/usb/serial/pl2303.c @@ -731,12 +731,13 @@ static int pl2303_get_line_request(struct usb_serial_port *port, GET_LINE_REQUEST, GET_LINE_REQUEST_TYPE, 0, 0, buf, 7, 100); if (ret != 7) { - dev_err(&port->dev, "%s - failed: %d\n", __func__, ret); + struct pl2303_private *priv = usb_get_serial_port_data(port); - if (ret >= 0) - ret = -EIO; + dev_dbg(&port->dev, "%s - failed, falling back on cache: %d\n", + __func__, ret); + memcpy(buf, priv->line_settings, 7); - return ret; + return 0; } dev_dbg(&port->dev, "%s - %7ph\n", __func__, buf); @@ -1078,8 +1079,8 @@ static int pl2303_set_break(struct usb_serial_port *port, bool enable) BREAK_REQUEST, BREAK_REQUEST_TYPE, state, 0, NULL, 0, 100); if (result) { - dev_err(&port->dev, "error sending break = %d\n", result); - return result; + dev_dbg(&port->dev, "error sending break = %d\n", result); + return -ENOTTY; } return 0;