Re: Modem Manager gets into a funny state with CDMA modems

[Dan Williams <dcbw redhat com>]

On Fri, 2010-03-12 at 12:50 -0500, Jason Glasgow wrote:
> Dan,
> 
> 
> Novatel U760.  The modem appears to be fine (not sure what you mean
> drop off the bus.  If I reopen the device all is fine).  I hacked up
> my version of the code (mm-generic-cdma.c) to always call
> mm_serial_port_open() before using the file descriptor, and it works
> great.  But I don't think this is a good solution.

It looks like this device actually cares about DCD state, or at least
it's sending messages that make 'option' think it does.  Can you try
adding CLOCAL to the c_cflags in mm-serial-port.c's config_fd()
function?  Just logical OR it together with the rest of the flags there
and let us know if that helps things.

I've been meaning to add CLOCAL for a while, since for most 3G modems
the control lines are completely ignored by either the driver or the
device anyway, but I've shied away from it because it's a change that
could potentially break stuff that I have no way to test.  There's a lot
of odd hardware out there.  Now that we're on the track to MM 0.4
though, we may want to revisit that.

Dan

> 
> -Jason
> 
> 
> diff --git a/src/mm-generic-cdma.c b/src/mm-generic-cdma.c
> index 50cd86c..7f44775 100644
> --- a/src/mm-generic-cdma.c
> +++ b/src/mm-generic-cdma.c
> @@ -598,6 +598,10 @@ connect (MMModem *modem,
>  
>      info = mm_callback_info_new (modem, callback, user_data);
>      command = g_strconcat ("DT", number, NULL);
> +    if (!mm_serial_port_open (priv->primary, &info->error)) {
> +        g_assert (info->error);
> +        /* TODO(jglasgow): what should we do? */
> +    }
>      mm_serial_port_queue_command (priv->primary, command, 90,
> dial_done, info);
>      g_free (command);
>  }
> @@ -620,7 +624,9 @@ disconnect_flash_done (MMSerialPort *port,
>                                  MM_MODEM_STATE_REASON_NONE);
>          }
>      } else {
> -        mm_port_set_connected (MM_GENERIC_CDMA_GET_PRIVATE
> (info->modem)->data, FALSE);
> +        MMGenericCdmaPrivate *priv = MM_GENERIC_CDMA_GET_PRIVATE
> (info->modem);
> +
> +        mm_port_set_connected (priv->data, FALSE);
>          update_enabled_state (MM_GENERIC_CDMA (info->modem), FALSE,
> MM_MODEM_STATE_REASON_NONE);
>      }
>  
> @@ -753,6 +759,10 @@ get_card_info (MMModem *modem,
>          port = priv->secondary;
>      }
>  
> +    if (!mm_serial_port_open (port, &info->error)) {
> +        g_assert (info->error);
> +        /* TODO(jglasgow): what should we do? */
> +    }
>      mm_serial_port_queue_command_cached (port, "+GMI", 3,
> get_manufacturer_done, info);
>      mm_serial_port_queue_command_cached (port, "+GMM", 3,
> get_model_done, info);
>      mm_serial_port_queue_command_cached (port, "+GMR", 3,
> get_version_done, info);
> @@ -867,6 +877,10 @@ get_signal_quality (MMModemCdma *modem,
>      }
>  
>      info = mm_callback_info_uint_new (MM_MODEM (modem), callback,
> user_data);
> +    if (!mm_serial_port_open (port, &info->error)) {
> +        g_assert (info->error);
> +        /* TODO(jglasgow): what should we do? */
> +    }
>      mm_serial_port_queue_command (port, "+CSQ", 3,
> get_signal_quality_done, info);
>  }
>  
> @@ -913,6 +927,10 @@ get_esn (MMModemCdma *modem,
>      }
>  
>      info = mm_callback_info_string_new (MM_MODEM (modem), callback,
> user_data);
> +    if (!mm_serial_port_open (port, &info->error)) {
> +        g_assert (info->error);
> +        /* TODO(jglasgow): what should we do? */
> +    }
>      mm_serial_port_queue_command_cached (port, "+GSN", 3,
> get_string_done, info);
>  }
>  
> @@ -1136,6 +1154,11 @@ get_serving_system (MMModemCdma *modem,
>                                        G_CALLBACK (callback),
>                                        user_data);
>  
> +    if (!mm_serial_port_open (port, &info->error)) {
> +        g_assert (info->error);
> +        /* TODO(jglasgow): what should we do? */
> +    }
> +
>      mm_serial_port_queue_command (port, "+CSS?", 3,
> serving_system_done, info);
>  }
>  
> @@ -1380,6 +1403,12 @@ get_registration_state (MMModemCdma *modem,
>      }
>  
>      info = mm_generic_cdma_query_reg_state_callback_info_new
> (MM_GENERIC_CDMA (modem), callback, user_data);
> +
> +    if (!mm_serial_port_open (port, &info->error)) {
> +        g_assert (info->error);
> +        /* TODO(jglasgow): what should we do? */
> +    }
> +
>      mm_serial_port_queue_command (port, "+CAD?", 3,
> get_analog_digital_done, info);
>  }
> 
> On Thu, Mar 11, 2010 at 6:46 PM, Dan Williams <dcbw redhat com> wrote:
>         On Wed, 2010-03-10 at 18:03 -0500, Jason Glasgow wrote:
>         > I've been working with a Novatel Modem on a 2.6.31 kernel
>         using the
>         > generic CDMA driver.   If I enable the modem via
>         ModemManager and then
>         > connect, everything works fine.  But if I disconnect the
>         modem using
>         > the DBUS API, and shut down ppp, ModemManager gets
>         confused.  The
>         > internal modem state indicates that the modem is now
>         REGISTERED, but
>         > the kernel appears to have sent a hangup (G_IO_HUP) to the
>         usb port,
>         > so mm-serial-port.c has closed the port (->fd == -1).
>         
>         
>         Does the modem crash at that point, or drop off the bus?  This
>         may
>         indicate a USB driver problem or just a bug in ModemManager.
>          But
>         normally with serial ports, when you get a hangup, you're
>         done.
>         
>         Also, what specific novatel device?
>         
>         Dan
>         
>         >
>         > Other than Disable(), no command works because the ->fd is
>         now -1.
>         >
>         >
>         > Do you have any thoughts on the best way to fix this?  And
>         what the
>         > semantics of receiving a HUP on the modem port should be.
>          My thought
>         > is that disconnect() should leave the modem in a registered
>         (or
>         > enabled) state, and that fp should still be valid.
>         >       * Might it be reasonable to ignore the G_IO_HUP?
>         >       * Should we automatically reopen then port for all
>         commands that
>         >         are issued?  Seems like a bad idea.
>         >       * Should we automatically reopen the port if
>         >         mm_modem_get_state() > MM_MODEM_STATE_ENABLED?
>         >       * Maybe the most logical is to have G_IO_HUP invoke a
>         callback
>         >         to notify the code that originally opened the serial
>         port that
>         >         it has been closed.  It then can update the state of
>         the modem
>         >         appropriately.
>         > Thanks for your thoughts,
>         > Jason
>         
>         > _______________________________________________
>         > NetworkManager-list mailing list
>         > NetworkManager-list gnome org
>         > http://mail.gnome.org/mailman/listinfo/networkmanager-list
>         
>         
> 
>