On Sat, Oct 03, 2009 at 02:18:07PM +0100, Alan Cox wrote: > On Sat, 3 Oct 2009 13:42:29 +0200 > Johan Hovold <jhovold@xxxxxxxxx> wrote: > > On Fri, Oct 02, 2009 at 05:33:08PM +0100, Alan Cox wrote: > > And, obviously, this doesn't solve the problem of tty_flip_buffer_push > > being called from interrupt context (but I assume that was never the > > intention). > > Calling tty_flip_buffer_push from an interrupt is perfectly acceptable > providing tty->low_latency isn't set: which it isn't. Of course -- the "with low latency set" part fell out (or, was implicit ;-) ). Your patch, however, still has low_latency set when it calls tty_flip_buffer_push and that's the problem. > > - spin_lock_irqsave(&priv->rx_lock, flags); > > - priv->rx_flags &= ~(THROTTLED | ACTUALLY_THROTTLED); > > - spin_unlock_irqrestore(&priv->rx_lock, flags); > > + spin_lock_irqsave(&port->lock, flags); > > + port->throttled = 0; > > + spin_unlock_irqrestore(&port->lock, flags); > > If you only have a single bit use the set_bit/clear_bit/test_and_xxx_bit > stuff as it's faster on most boxes The generic driver uses two fields in usb_serial_port for throttled/trottle_req, whereas ftdi_sio, whiteheat, aircable, cypress and perhaps a couple more have private a flag field for THROTTLED and ACTUALLY_THROTTLED. How about unifying them to all use a single flag field (with two flags) in usb_serial_port? > > + * The per character mucking around with sysrq path it too slow, so > > + * shortcircuit it in the 99.9999999% of cases where the USB serial is > > + * not a console anyway. > > + */ > > + ch = packet + 2; > > + len -= 2; > > + if (!port->console || !port->sysrq) > > You need && flag == TTY_NORMAL ? You tell me. :-) Are we interested in them unless port->console is set? > Definitely a move in the right direction Thanks, Johan -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
