serial.c

linux-2.4.29操作系统的源码

/* $Id: serial.c,v 1.58 2003/08/29 17:32:50 johana Exp $
 *
 * Serial port driver for the ETRAX 100LX chip
 *
 *    Copyright (C) 1998, 1999, 2000, 2001, 2002, 2003  Axis Communications AB
 *
 *    Many, many authors. Based once upon a time on serial.c for 16x50.
 *
 * $Log: serial.c,v $
 * Revision 1.58  2003/08/29 17:32:50  johana
 * Fixed CMSPAR (Mark/Space) support. CMSPAR|PARODD = Mark(1) parity.
 *
 * Revision 1.57  2003/08/26 16:53:06  johana
 * Merged in change_branch--johana to get non DMA support etc.
 *
 * Revision 1.56  2003/07/10 13:18:03  pkj
 * Corrected a copy-paste error.
 *
 * Revision 1.55  2003/07/10 11:00:46  starvik
 * Moved all the latest stuff to a branch until it is stable
 *
 * Revision 1.50.2.2  2003/07/28 09:59:39  johana
 * Clear tr_running so next write really starts transmission.
 *
 * Revision 1.50.2.1  2003/07/10 10:59:54  starvik
 * Moved all the latest stuff to a branch until it is stable
 *
 * Revision 1.54  2003/07/08 12:42:19  johana
 * Removed some test defines within #if 0.
 * Moved a comment to correct place.
 *
 * Revision 1.53  2003/07/08 09:48:28  starvik
 * Removed test code
 *
 * Revision 1.52  2003/07/08 07:24:47  pkj
 * Corrected spelling mistakes originally found in 2.5.x
 *
 * Revision 1.51  2003/07/07 09:17:12  johana
 * Config.in:
 * Don't consider ETRAX_FAST_TIMER experimental.
 * Add CONFIG_ETRAX_EXTERN_PB6CLK_ENABLED
 * Add CONFIG_ETRAX_EXTERN_PB6CLK_FREQ
 * Add CONFIG_ETRAX_SERIAL_PORTx_DMAy_OUT/IN
 *
 * serial.h:
 * Support for with and without DMA.
 * Added baud_base and custom_divisor for flexible baudrate.
 * Added flush_time_usec.
 * Removed unused DMA register pointers.
 *
 * serial.c:
 * Use C99 style initialisation of rs_table, some fields added, some removed.
 * Always disable interrupts when fiddling with shadow variables.
 * Support for flexible baudrate using prescaler or external clock.
 * Only disable the DMA channel if we own it.
 * Support for either using DMA or char-by-char interrupt.
 *
 * DMA changes:
 * o When transmitting with DMA, don't put all in one descriptor, split
 *   it so that we can wake up the application before everything is sent.
 * o When starting a fast timer to flip later, use a longer, fixed time.
 * o Use flush_time_usec and have a minimum value to prevent excessive load
 *   on high baudrates.
 * o Do force_eop_if_needed() after processing to reduce load.
 * o All DMA interrupts uses SA_SHIRQ | SA_INTERRUPT
 *
 * Only have interrupts disabled while processing rx interrupts,
 * unmask just tx interrupts and unblock serial interrupt and keep
 * global interrupts enabled while processing tx interrupts to make
 * char-by-char interrupts usable.
 *
 * Flow control:
 * o Use normal tx interrupt when sending x_char.
 * o Use throttling within the driver when using DMA.
 *
 * Improved debugging with DEBUG_LOG macro.
 *
 * Revision 1.50  2003/06/13 10:04:12  johana
 * Help the user to avoid trouble by:
 * Preventing FAST_TIMER and FAST_DMA_FLUSH at the same time.
 * Forcing mixed mode for status/control lines if not all pins are used.
 *
 * Som minor changes to reduce diff to os/lx25 version.
 *
 * Revision 1.49  2003/05/30 11:31:54  johana
 * Merged in change-branch--serial9bit that adds CMSPAR support for sticky
 * parity (mark/space)
 *
 * Revision 1.48  2003/05/30 11:03:57  johana
 * Implemented rs_send_xchar() by disabling the DMA and writing manually.
 * Added e100_disable_txdma_channel() and e100_enable_txdma_channel().
 * Fixed rs_throttle() and rs_unthrottle() to properly call rs_send_xchar
 * instead of setting info->x_char and check the CRTSCTS flag before
 * controlling the rts pin.
 *
 * Revision 1.47  2003/04/09 08:31:14  pkj
 * Typo correction (taken from Linux 2.5).
 *
 * Revision 1.46  2003/04/01 14:12:06  starvik
 * Added loglevel for lots of printks
 *
 * Revision 1.45  2003/01/22 06:54:46  starvik
 * Fixed warnings issued by GCC 3.2.1
 *
 * Revision 1.44.4.1  2003/01/08 06:47:12  starvik
 * Added mark/space parity (for 9-bit serial)
 *
 * Revision 1.44  2002/11/21 04:28:06  hp
 * Change static inline to extern inline where otherwise outlined with gcc-3.2
 *
 * Revision 1.43  2002/11/15 12:02:25  pkj
 * Make sure we have a tty in flush_to_flip_buffer() before we
 * try to use it...
 *
 * Revision 1.42  2002/11/05 09:08:47  johana
 * Better implementation of rs_stop() and rs_start() that uses the XOFF
 * register to start/stop transmission.
 * change_speed() also initilises XOFF register correctly so that
 * auto_xoff is enabled when IXON flag is set by user.
 * This gives fast XOFF response times.
 *
 * Revision 1.41  2002/11/04 18:40:57  johana
 * Implemented rs_stop() and rs_start().
 * Simple tests using hwtestserial indicates that this should be enough
 * to make it work.
 *
 * Revision 1.40  2002/10/14 05:33:18  starvik
 * RS-485 uses fast timers even if SERIAL_FAST_TIMER is disabled
 *
 * Revision 1.39  2002/09/30 21:00:57  johana
 * Support for CONFIG_ETRAX_SERx_DTR_RI_DSR_CD_MIXED where the status and
 * control pins can be mixed between PA and PB.
 * If no serial port uses MIXED old solution is used
 * (saves a few bytes and cycles).
 * control_pins struct uses masks instead of bit numbers.
 * Corrected dummy values and polarity in line_info() so
 * /proc/tty/driver/serial is now correct.
 * (the E100_xxx_GET() macros is really active low - perhaps not obvious)
 *
 * Revision 1.38  2002/08/23 11:01:36  starvik
 * Check that serial port is enabled in all interrupt handlers to avoid
 * restarts of DMA channels not assigned to serial ports
 *
 * Revision 1.37  2002/08/13 13:02:37  bjornw
 * Removed some warnings because of unused code
 *
 * Revision 1.36  2002/08/08 12:50:01  starvik
 * Serial interrupt is shared with synchronous serial port driver
 *
 * Revision 1.35  2002/06/03 10:40:49  starvik
 * Increased RS-485 RTS toggle timer to 2 characters
 *
 * Revision 1.34  2002/05/28 18:59:36  johana
 * Whitespace and comment fixing to be more like etrax100ser.c 1.71.
 *
 * Revision 1.33  2002/05/28 17:55:43  johana
 * RS-485 uses FAST_TIMER if enabled, and starts a short (one char time)
 * timer from tranismit_chars (interrupt context).
 * The timer toggles RTS in interrupt context when expired giving minimum
 * latencies.
 *
 * Revision 1.32  2002/05/22 13:58:00  johana
 * Renamed rs_write() to raw_write() and made it inline.
 * New rs_write() handles RS-485 if configured and enabled
 * (moved code from e100_write_rs485()).
 * RS-485 ioctl's uses copy_from_user() instead of verify_area().
 *
 * Revision 1.31  2002/04/22 11:20:03  johana
 * Updated copyright years.
 *
 * Revision 1.30  2002/04/22 09:39:12  johana
 * RS-485 support compiles.
 *
 * Revision 1.29  2002/01/14 16:10:01  pkj
 * Allocate the receive buffers dynamically. The static 4kB buffer was
 * too small for the peaks. This means that we can get rid of the extra
 * buffer and the copying to it. It also means we require less memory
 * under normal operations, but can use more when needed (there is a
 * cap at 64kB for safety reasons). If there is no memory available
 * we panic(), and die a horrible death...
 *
 * Revision 1.28  2001/12/18 15:04:53  johana
 * Cleaned up write_rs485() - now it works correctly without padding extra
 * char.
 * Added sane default initialisation of rs485.
 * Added #ifdef around dummy variables.
 *
 * Revision 1.27  2001/11/29 17:00:41  pkj
 * 2kB seems to be too small a buffer when using 921600 bps,
 * so increase it to 4kB (this was already done for the elinux
 * version of the serial driver).
 *
 * Revision 1.26  2001/11/19 14:20:41  pkj
 * Minor changes to comments and unused code.
 *
 * Revision 1.25  2001/11/12 20:03:43  pkj
 * Fixed compiler warnings.
 *
 * Revision 1.24  2001/11/12 15:10:05  pkj
 * Total redesign of the receiving part of the serial driver.
 * Uses eight chained descriptors to write to a 4kB buffer.
 * This data is then serialised into a 2kB buffer. From there it
 * is copied into the TTY's flip buffers when they become available.
 * A lot of copying, and the sizes of the buffers might need to be
 * tweaked, but all in all it should work better than the previous
 * version, without the need to modify the TTY code in any way.
 * Also note that erroneous bytes are now correctly marked in the
 * flag buffers (instead of always marking the first byte).
 *
 * Revision 1.23  2001/10/30 17:53:26  pkj
 * * Set info->uses_dma to 0 when a port is closed.
 * * Mark the timer1 interrupt as a fast one (SA_INTERRUPT).
 * * Call start_flush_timer() in start_receive() if
 *   CONFIG_ETRAX_SERIAL_FLUSH_DMA_FAST is defined.
 *
 * Revision 1.22  2001/10/30 17:44:03  pkj
 * Use %lu for received and transmitted counters in line_info().
 *
 * Revision 1.21  2001/10/30 17:40:34  pkj
 * Clean-up. The only change to functionality is that
 * CONFIG_ETRAX_SERIAL_RX_TIMEOUT_TICKS(=5) is used instead of
 * MAX_FLUSH_TIME(=8).
 *
 * Revision 1.20  2001/10/30 15:24:49  johana
 * Added char_time stuff from 2.0 driver.
 *
 * Revision 1.19  2001/10/30 15:23:03  johana
 * Merged with 1.13.2 branch + fixed indentation
 * and changed CONFIG_ETRAX100_XYS to CONFIG_ETRAX_XYZ
 *
 * Revision 1.18  2001/09/24 09:27:22  pkj
 * Completed ext_baud_table[] in cflag_to_baud() and cflag_to_etrax_baud().
 *
 * Revision 1.17  2001/08/24 11:32:49  ronny
 * More fixes for the CONFIG_ETRAX_SERIAL_PORT0 define.
 *
 * Revision 1.16  2001/08/24 07:56:22  ronny
 * Added config ifdefs around ser0 irq requests.
 *
 * Revision 1.15  2001/08/16 09:10:31  bjarne
 * serial.c - corrected the initialization of rs_table, the wrong defines
 *            where used.
 *            Corrected a test in timed_flush_handler.
 *            Changed configured to enabled.
 * serial.h - Changed configured to enabled.
 *
 * Revision 1.14  2001/08/15 07:31:23  bjarne
 * Introduced two new members to the e100_serial struct.
 * configured - Will be set to 1 if the port has been configured in .config
 * uses_dma   - Should be set to 1 if the port uses DMA. Currently it is set 
 *              to 1
 *              when a port is opened. This is used to limit the DMA interrupt
 *              routines to only manipulate DMA channels actually used by the
 *              serial driver.
 *
 * Revision 1.13.2.2  2001/10/17 13:57:13  starvik
 * Receiver was broken by the break fixes
 *
 * Revision 1.13.2.1  2001/07/20 13:57:39  ronny
 * Merge with new stuff from etrax100ser.c. Works but haven't checked stuff
 * like break handling.
 *
 * Revision 1.13  2001/05/09 12:40:31  johana
 * Use DMA_NBR and IRQ_NBR defines from dma.h and irq.h
 *
 * Revision 1.12  2001/04/19 12:23:07  bjornw
 * CONFIG_RS485 -> CONFIG_ETRAX_RS485
 *
 * Revision 1.11  2001/04/05 14:29:48  markusl
 * Updated according to review remarks i.e.
 * -Use correct types in port structure to avoid compiler warnings
 * -Try to use IO_* macros whenever possible
 * -Open should never return -EBUSY
 *
 * Revision 1.10  2001/03/05 13:14:07  bjornw
 * Another spelling fix
 *
 * Revision 1.9  2001/02/23 13:46:38  bjornw
 * Spellling check
 *
 * Revision 1.8  2001/01/23 14:56:35  markusl
 * Made use of ser1 optional
 * Needed by USB
 *
 * Revision 1.7  2001/01/19 16:14:48  perf
 * Added kernel options for serial ports 234.
 * Changed option names from CONFIG_ETRAX100_XYZ to CONFIG_ETRAX_XYZ.
 *
 * Revision 1.6  2000/11/22 16:36:09  bjornw
 * Please marketing by using the correct case when spelling Etrax.
 *
 * Revision 1.5  2000/11/21 16:43:37  bjornw
 * Fixed so it compiles under CONFIG_SVINTO_SIM
 *
 * Revision 1.4  2000/11/15 17:34:12  bjornw
 * Added a timeout timer for flushing input channels. The interrupt-based
 * fast flush system should be easy to merge with this later (works the same
 * way, only with an irq instead of a system timer_list)
 *
 * Revision 1.3  2000/11/13 17:19:57  bjornw
 * * Incredibly, this almost complete rewrite of serial.c worked (at least
 *   for output) the first time.
 *
 *   Items worth noticing:
 *
 *      No Etrax100 port 1 workarounds (does only compile on 2.4 anyway now)
 *      RS485 is not ported (why can't it be done in userspace as on x86 ?)
 *      Statistics done through async_icount - if any more stats are needed,
 *      that's the place to put them or in an arch-dep version of it.
 *      timeout_interrupt and the other fast timeout stuff not ported yet
 *      There be dragons in this 3k+ line driver
 *
 * Revision 1.2  2000/11/10 16:50:28  bjornw
 * First shot at a 2.4 port, does not compile totally yet
 *
 * Revision 1.1  2000/11/10 16:47:32  bjornw
 * Added verbatim copy of rev 1.49 etrax100ser.c from elinux
 *
 * Revision 1.49  2000/10/30 15:47:14  tobiasa
 * Changed version number.
 *
 * Revision 1.48  2000/10/25 11:02:43  johana
 * Changed %ul to %lu in printf's
 *
 * Revision 1.47  2000/10/18 15:06:53  pkj
 * Compile correctly with CONFIG_ETRAX_SERIAL_FLUSH_DMA_FAST and
 * CONFIG_ETRAX_SERIAL_PROC_ENTRY together.
 * Some clean-up of the /proc/serial file.
 *
 * Revision 1.46  2000/10/16 12:59:40  johana
 * Added CONFIG_ETRAX_SERIAL_PROC_ENTRY for statistics and debug info.
 *
 * Revision 1.45  2000/10/13 17:10:59  pkj
 * Do not flush DMAs while flipping TTY buffers.
 *
 * Revision 1.44  2000/10/13 16:34:29  pkj
 * Added a delay in ser_interrupt() for 2.3ms when an error is detected.
 * We do not know why this delay is required yet, but without it the
 * irmaflash program does not work (this was the program that needed
 * the ser_interrupt() to be needed in the first place). This should not
 * affect normal use of the serial ports.
 *
 * Revision 1.43  2000/10/13 16:30:44  pkj
 * New version of the fast flush of serial buffers code. This time
 * it is localized to the serial driver and uses a fast timer to
 * do the work.
 *
 * Revision 1.42  2000/10/13 14:54:26  bennyo
 * Fix for switching RTS when using rs485
 *
 * Revision 1.41  2000/10/12 11:43:44  pkj
 * Cleaned up a number of comments.
 *
 * Revision 1.40  2000/10/10 11:58:39  johana
 * Made RS485 support generic for all ports.
 * Toggle rts in interrupt if no delay wanted.
 * WARNING: No true transmitter empty check??
 * Set d_wait bit when sending data so interrupt is delayed until
 * fifo flushed. (Fix tcdrain() problem)
 *
 * Revision 1.39  2000/10/04 16:08:02  bjornw
 * * Use virt_to_phys etc. for DMA addresses
 * * Removed CONFIG_FLUSH_DMA_FAST hacks
 * * Indentation fix
 *
 * Revision 1.38  2000/10/02 12:27:10  mattias
 * * added variable used when using fast flush on serial dma.
 *   (CONFIG_FLUSH_DMA_FAST)
 *
 * Revision 1.37  2000/09/27 09:44:24  pkj
 * Uncomment definition of SERIAL_HANDLE_EARLY_ERRORS.
 *
 * Revision 1.36  2000/09/20 13:12:52  johana
 * Support for CONFIG_ETRAX_SERIAL_RX_TIMEOUT_TICKS:
 *   Number of timer ticks between flush of receive fifo (1 tick = 10ms).
 *   Try 0-3 for low latency applications. Approx 5 for high load
 *   applications (e.g. PPP). Maybe this should be more adaptive some day...
 *
 * Revision 1.35  2000/09/20 10:36:08  johana
 * Typo in get_lsr_info()
 *
 * Revision 1.34  2000/09/20 10:29:59  johana
 * Let rs_chars_in_buffer() check fifo content as well.
 * get_lsr_info() might work now (not tested).
 * Easier to change the port to debug.
 *
 * Revision 1.33  2000/09/13 07:52:11  torbjore
 * Support RS485
 *
 * Revision 1.32  2000/08/31 14:45:37  bjornw
 * After sending a break we need to reset the transmit DMA channel
 *
 * Revision 1.31  2000/06/21 12:13:29  johana
 * Fixed wait for all chars sent when closing port.
 * (Used to always take 1 second!)
 * Added shadows for directions of status/ctrl signals.
 *
 * Revision 1.30  2000/05/29 16:27:55  bjornw
 * Simulator ifdef moved a bit
 *
 * Revision 1.29  2000/05/09 09:40:30  mattias
 * * Added description of dma registers used in timeout_interrupt
 * * Removed old code
 *
 * Revision 1.28  2000/05/08 16:38:58  mattias
 * * Bugfix for flushing fifo in timeout_interrupt
 *   Problem occurs when bluetooth stack waits for a small number of bytes
 *   containing an event acknowledging free buffers in bluetooth HW
 *   As before, data was stuck in fifo until more data came on uart and
 *   flushed it up to the stack.
 *
 * Revision 1.27  2000/05/02 09:52:28  jonasd
 * Added fix for peculiar etrax behaviour when eop is forced on an empty