2001-08-11 Tim Waugh * serial.c (get_pci_port): Deal with awkward Titan cards. 1998-08-26 Theodore Ts'o * serial.c (rs_open): Correctly decrement the module in-use count on errors. Thu Feb 19 14:24:08 1998 Theodore Ts'o * tty_io.c (tty_name): Remove the non-reentrant (and non-SMP safe) version of tty_name, and rename the reentrant _tty_name function to be tty_name. (tty_open): Add a warning message stating callout devices are deprecated. Mon Dec 1 08:24:15 1997 Theodore Ts'o * tty_io.c (tty_get_baud_rate): Print a warning syslog if the tty->alt_speed kludge is used; this means the system is using the deprecated SPD_HI ioctls. Mon Nov 24 10:37:49 1997 Theodore Ts'o * serial.c, esp.c, rocket.c: Change drivers to take advantage of tty_get_baud_rate(). * tty_io.c (tty_get_baud_rate): New function which computes the correct baud rate for the tty. More factoring out of common code out of the serial driver to the high-level tty functions.... Sat Nov 22 07:53:36 1997 Theodore Ts'o * serial.c, esp.c, rocket.c: Add tty->driver.break() routine, and allow high-level tty code to handle the break and soft carrier ioctls. * tty_ioctl.c (n_tty_ioctl): Support TIOCGSOFTCAR and TIOCSSOFTCAR, so that device drivers don't have to support it. * serial.c (autoconfig): Change 16750 test to hopefully eliminate false results by people with strange 16550As being detected as 16750s. Hopefully 16750s will still be detected as 16750, and other weird UARTs won't get poorly autodetected. If this doesn't work, I'll have to disable the auto identification for the 16750. * tty_io.c (tty_hangup): Now actually do the tty hangup processing during the timer processing, and disable interrupts while doing the hangup processing. This avoids several nasty race conditions which happened when the hangup processing was done asynchronously. (tty_ioctl): Do break handling in the tty driver if driver's break function is supported. (tty_flip_buffer_push): New exported function which should be used by drivers to push characters in the flip buffer to the tty handler. This may either be done using a task queue function for better CPU efficiency, or directly for low latency operation. * serial.c (rs_set_termios): Fix bug rs_set_termios when transitioning away from B0, submitted by Stanislav Voronyi. Thu Jun 19 20:05:58 1997 Theodore Ts'o * serial.c (begin_break, end_break, rs_ioctl): Applied patch to support BSD ioctls to set and clear the break condition explicitly. * console.c (scrup, scrdown, insert_line, delete_line): Applied fix suggested by Aaron Tiensivu to speed up block scrolls up and down. * n_tty.c (opost_block, write_chan): Added a modified "fast console" patch which processes a block of text via "cooking" efficiently. Wed Jun 18 15:25:50 1997 Theodore Ts'o * tty_io.c (init_dev, release_dev): Applied fix suggested by Bill Hawes to prevent race conditions in the tty code. * n_tty.c (n_tty_chars_in_buffer): Applied fix suggested by Bill Hawes so that n_tty_chars_in_buffer returns the correct value in the case when the tty is in cannonical mode. (To avoid a pty deadlock with telnetd.) Thu Feb 27 01:53:08 1997 Theodore Ts'o * serial.c (change_speed): Add support for the termios flag CMSPAR, which allows the user to select stick parity. (i.e, if PARODD is set, the parity bit is always 1; if PARRODD is not set, then the parity bit is always 0). Wed Feb 26 19:03:10 1997 Theodore Ts'o * serial.c (cleanup_module): Fix memory leak when using the serial driver as a module; make sure tmp_buf gets freed! Tue Feb 25 11:01:59 1997 Theodore Ts'o * serial.c (set_modem_info): Add support for setting and clearing the OUT1 and OUT2 bits. (For special case UART's, usually for half-duplex.) (autoconfig, change_speed): Fix TI 16750 support. Sun Feb 16 00:14:43 1997 Theodore Ts'o * tty_io.c (release_dev): Add sanity check to make sure there are no waiters on tty->read_wait or tty->write_wait. * serial.c (rs_init): Don't autoconfig a device if the I/O region is already reserved. * serial.c (serial_proc_info): Add support for /proc/serial. Thu Feb 13 00:49:10 1997 Theodore Ts'o * serial.c (receive_chars): When the UART repotrs an overrun condition, it does so with a valid character. Changed to not throw away the valid character, but instead report the overrun after the valid character. * serial.c: Added new #ifdef's for some of the advanced serial driver features. A minimal driver that only supports COM 1/2/3/4 without sharing serial interrupts only takes 17k; the full driver takes 32k. Wed Feb 12 14:50:44 1997 Theodore Ts'o * vt.c: * pty.c: * tty_ioctl.c: * serial.c: Update routines to use the new 2.1 memory access routines. Wed Dec 4 07:51:52 1996 Theodore Ts'o * serial.c (change_speed): Use save_flags(); cli() and restore_flags() in order to ensure we don't accidentally turn on interrupts when starting up the port. (startup): Move the insertion of serial structure into the IRQ chain earlier into the startup processing. Interrupts should be off this whole time, but we eventually will want to reduce this window. Thu Nov 21 10:05:22 1996 Theodore Ts'o * tty_ioctl.c (tty_wait_until_sent): Always check the driver wait_until_ready routine, even if there are no characters in the xmit buffer. (There may be charactes in the device FIFO.) (n_tty_ioctl): Add new flag tty->flow_stopped which indicates whether the tty is stopped due to a request by the TCXONC ioctl (used by tcflow). If so, don't let an incoming XOFF character restart the tty. The tty can only be restarted by another TCXONC request. * tty_io.c (start_tty): Don't allow the tty to be restarted if tty->flow_stopped is true. * n_tty.c (n_tty_receive_char): If tty->flow_stopped is true, and IXANY is set, don't eat a character trying to restart the tty. * serial.c (startup): Remove need for MCR_noint from the async_struct structure. Only turn on DTR and RTS if the baud rate is not zero. (change_speed): More accurately calculate the timeout value based on the word size. Move responsibility of hangup when speed becomes B0 to rs_set_termios() (set_serial_info): When changing the UART type set the current xmit_fifo_size as well as the permanent xmit_fifo_size. (rs_ioctl): Fix TCSBRK (used by tcdrain) and TCSBRKP ioctls to return EINTR if interrupted by a signal. (rs_set_termios): If the baud rate changes to or from B0, this function is now responsible for setting or clearing DTR and RTS. DTR and RTS are only be changed on the transition to or from the B0 state. (rs_close): Wait for the characters to drain based on info->timeout. At low baud rates (50 bps), it may take a long time for the FIFO to completely drain out! (rs_wait_until_sent): Fixed timeout handling. Now releases control to the scheduler, but checks frequently enough so that the function is sensitive enough to pass the timing requirements of the NIST-PCTS. (block_til_ready): When opening the device, don't turn on DTR and RTS if the baud rate is B0. Thu Nov 14 00:06:09 1996 Theodore Ts'o * serial.c (autoconfig): Fix autoconfiguration problems; info->flags wasn't getting initialized from the state structure. Put in more paranoid test for the 16750. Fri Nov 8 20:19:50 1996 Theodore Ts'o * n_tty.c (n_tty_flush_buffer): Only call driver->unthrottle() if the tty was previous throttled. (n_tty_set_termios, write_chan): Add changes suggested by Simon P. Allen to allow hardware cooking. * tty_ioctl.c (set_termios): If we get a signal while waiting for the tty to drain, return -EINTR. * serial.c (change_speed): Add support for CREAD, as required by POSIX. Sat Nov 2 20:43:10 1996 Theodore Ts'o * serial.c: Wholesale changes. Added support for the Startech 16650 and 16650V2 chips. (WARNING: the new startech 16650A may or may not work!) Added support for the TI16750 (not yet tested). Split async_struct into a transient part (async_struct) and a permanent part (serial_state) which contains the configuration information for the ports. Added new driver routines wait_until_sent() and send_xchar() to help with POSIX compliance. Added support for radio clocks which waggle the carrier detect line (CONFIG_HARD_PPS). * tty_ioctl.c (tty_wait_until_sent): Added call to new driver function tty->driver.wait_until_sent(), which returns when the tty's device xmit buffers are drained. Needed for full POSIX compliance. (send_prio_char): New function, called by the ioctl's TCIOFF and TCION; uses the new driver call send_xchar(), which will send the XON or XOFF character at high priority (and even if tty output is stopped). Wed Jun 5 18:52:04 1996 Theodore Ts'o * pty.c (pty_close): When closing a pty, make sure packet mode is cleared. Sun May 26 09:33:52 1996 Theodore Ts'o * vesa_blank.c (set_vesa_blanking): Add missing verify_area() call. * selection.c (set_selection): Add missing verify_area() call. * tty_io.c (tty_ioctl): Add missing verify_area() calls. * serial.c (rs_ioctl): Add missing verify_area() calls. (rs_init): Allow initialization of serial driver configuration from a module. * random.c (extract_entropy): Add missing verify_area call. Don't limit number of characters returned to 32,768. Extract entropy is now no longer a inlined function. (random_read): Check return value in case extract_entropy returns an error. (secure_tcp_sequence_number): New function which returns a secure TCP sequence number. This is needed to prevent some nasty TCP hijacking attacks. (init_std_data): Initialize using gettimeofday() instead of struct timeval xtime. (fast_add_entropy_word, add_entropy_word): Rename the inline function add_entropy_word() to fast_add_entropy_word(). Make add_entropy_word() be the non-inlined function which is used in non-timing critical places, in order to save space. (initialize_benchmark, begin_benchmark, end_benchmark): New functions defined when RANDOM_BENCHMARK is defined. They allow us to benchmark the speed of the add_timer_randomness() call. (int_ln, rotate_left): Add two new inline functions with i386 optimized asm instructions. This speeds up the critical add_entropy_word() and add_timer_randomness() functions, which are called from interrupt handlers. Tue May 7 22:51:11 1996 * random.c (add_timer_randomness): Limit the amount randomness that we estimate to 12 bits. (An arbitrary amount). (extract_entropy): To make it harder to analyze the hash function, fold the hash function in half using XOR, and use the folded result as the value to emit to the user. Also, add timer randomness each pass through the exact_entropy call, to increase the amount of unknown values during the extraction process. (random_ioctl): Use IOR/IOW definitions to define the ioctl values used by the /dev/random driver. Allow the old ioctl values to be used for backwards compatibility (for a limited amount of time). Wed Apr 24 14:02:04 1996 Theodore Ts'o * random.c (add_timer_randomness): Use 2nd derivative as well to better estimate entropy. (rand_initialize): Explicitly initialize all the pointers to NULL. (Clearing pointers using memset isn't portable.) Initialize the random pool with OS-dependent data. (random_write): Add sanity checking to the arguments to random_write(), so that bad arguments won't cause a kernel SEGV. (random_read): Update the access time of the device inode when you return data to the user. (random_ioctl): Wake up the random_wait channel when there are only WAIT_INPUT_BITS available. Add more paranoia checks to make sure entropy_count doesn't go beyond the bounds of (0, POOLSIZE). Add a few missing verify_area checks. Add support for the RNDCLEARPOOL ioctl, which zaps the random pool. (add_timer_randomness): Wake up the random_wait channel only when there are WAIT_INPUT_BITS available. (random_select): Allow a random refresh daemon process to select on /dev/random for writing; wake up the daemon when there are less than WAIT_OUTPUT_BITS bits of randomness available. Tue Apr 23 22:56:07 1996 * tty_io.c (init_dev): Change return code when user attempts to open master pty which is already open from EAGAIN to EIO, to match with BSD expectations. EIO is more correct anyway, since EAGAIN implies that retrying will be successful --- which it might be.... Eventually!! * pty.c (pty_open, pty_close): Fix wait loop so that we don't busy loop while waiting for the master side to open. Fix tty opening/closing logic. TTY_SLAVE_CLOSED was renamed to TTY_OTHER_CLOSED, so that the name is more descriptive. Also fixed code so that the tty flag actually works correctly now.... Mon Apr 1 10:22:01 1996 * serial.c (rs_close): Cleaned up modularization changes. Remove code which forced line discipline back to N_TTY this is done in the tty upper layers, and there's no reason to do it here. (Making this change also removed the requirement that the serial module access the internal kernel symbol "ldiscs".) * tty_io.c (tty_init): Formally register a tty_driver entry for /dev/tty (device 4, 0) and /dev/console (device 5, 0). This guarantees that major device numbers 4 and 5 will be reserved for the tty subsystem (as they have to be because of /dev/tty and /dev/console). Removed tty_regdev, as this interface is no longer necessary. Sun Mar 17 20:42:47 GMT 1996 * serial.c : modularisation (changes in linux/fs/device.c allow kerneld to automatically load the serial module). * Makefile, Config.in : serial modularisation adds. * tty_io.c : tty_init_ctty used by to register "cua" driver just for the /dev/tty device (5,0). Added tty_regdev. * serial.c (shutdown, rs_ioctl) : when port shuts down wakeup processes waiting on delta_msr_wait. The TIOCMIWAIT ioctl returns EIO if no change was done since the time of call. Sat Mar 16 14:33:13 1996 * tty_io.c (disassociate_ctty): If disassociate_ctty is called by exit, do not perform an implicit vhangup on a pty. Fri Feb 9 14:15:47 1996 * serial.c (block_til_ready): Fixed another race condition which happens if a hangup happens during the open. Wed Jan 10 10:08:00 1996 * serial.c (block_til_ready): Remove race condition which happened if a hangup condition happened during the setup of the UART, before rs_open() called block_til_ready(). This caused the info->count counter to be erroneously decremented. * serial.c (startup, rs_open): Remove race condition that could cause a memory leak of one page. (Fortunately, both race conditions were relatively rare in practice.) Tue Dec 5 13:21:27 1995 * serial.c (check_modem_status, rs_ioctl): Support the new ioctl()'s TIOCGICOUNT, TIOCMIWAIT. These allow an application program to wait on a modem serial register status bit change, and to find out how many changes have taken place for the MSR bits. (rs_write): Eliminate a race condition which is introduced if it is necessary to wait for the semaphore. Sat Nov 4 17:14:45 1995 * tty_io.c (tty_init): Move registration of TTY_MAJOR and TTY_AUX_MAJOR to the end, so that /proc/devices looks prettier. * pty.c (pty_init): Use new major numbers for PTY master and slave devices. This allow us to have more than 64 pty's. We register the old pty devices for backwards compatibility. Note that a system should either be using the old pty devices or the new pty devices --- in general, it should try to use both, since they map into the same pty table. The old pty devices are strictly for backwards compatibility. Wed Oct 11 12:45:24 1995 * tty_io.c (disassociate_ctty): If disassociate_ctty is called by exit, perform an implicit vhangup on the tty. * pty.c (pty_close): When the master pty is closed, send a hangup to the slave pty. (pty_open): Use the flag TTY_SLAVE_CLOSED to test to see if there are any open slave ptys, instead of using tty->link->count. The old method got confused if there were processes that had hung-up file descriptors on the slave tty. Tue May 2 00:53:25 1995 * tty_io.c (tty_set_ldisc): Wait until the output buffer is drained before closing the old line discipline --- needed in only one case: XON/XOFF processing. * n_tty.c (n_tty_close): Don't bother waiting until the output driver is closed; in general, the line discipline shouldn't care if the hardware is finished transmitting before the line discipline terminates. * tty_io.c (release_dev): Shutdown the line discipline after decrementing the tty count variable; but set the TTY_CLOSING flag so that we know that this tty structure isn't long for this world. * tty_io.c (init_dev): Add sanity code to check to see if TTY_CLOSING is set on a tty structure; if so, something bad has happened (probably a line discipline close blocked when it shouldn't have; so do a kernel printk and then return an error). Wed Apr 26 10:23:44 1995 Theodore Y. Ts'o * tty_io.c (release_dev): Try to shutdown the line discipline *before* decrementing the tty count variable; this removes a potential race condition which occurs when the line discipline close blocks, and another process then tries open the same serial port. * serial.c (rs_hangup): When hanging up, flush the output buffer before shutting down the UART. Otherwise the line discipline close blocks waiting for the characters to get flushed, which never happens until the serial port gets reused. Wed Apr 12 08:06:16 1995 Theodore Y. Ts'o * serial.c (do_serial_hangup, do_softint, check_modem_status, rs_init): Hangups are now scheduled via a separate tqueue structure in the async_struct structure, tqueue_hangup. This task is pushed on to the tq_schedule queue, so that it is processed synchronously by the scheduler. Sat Feb 18 12:13:51 1995 Theodore Y. Ts'o (tytso@rt-11) * tty_io.c (disassociate_ctty, tty_open, tty_ioctl): Clear current->tty_old_pgrp field when a session leader acquires a controlling tty, and after a session leader has disassociated from a controlling tty. Fri Feb 17 09:34:09 1995 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_interrupt_single, rs_interrupt, rs_interrupt_multi): Change the number of passes made from 64 to be 256, configurable with the #define RS_ISR_PASS_LIMIT. * serial.c (rs_init, set_serial_info, get_serial_info, rs_close): Remove support for closing_wait2. Instead, set tty->closing and rely on the line discipline to prevent echo wars. * n_tty.c (n_tty_receive_char): IEXTEN does not need to be enabled in order for IXANY to be active. If tty->closing is set, then only process XON and XOFF characters. Sun Feb 12 23:57:48 1995 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_timer): Change the interrupt poll time from 60 seconds to 10 seconds, configurable with the #define RS_STROBE_TIME. * serial.c (rs_interrupt_multi, startup, shutdown, rs_ioctl, set_multiport_struct, get_multiport_struct): Add provisions for a new type of interrupt service routine, which better supports multiple serial ports on a single IRQ. Sun Feb 5 19:35:11 1995 Theodore Y. Ts'o (tytso@rt-11) * tty_ioctl.c (n_tty_ioctl, set_termios, tty_wait_until_sent): * serial.c (rs_ioctl, rs_close): * cyclades.c (cy_ioctl, cy_close): * n_tty.c (n_tty_close): Rename wait_until_sent to tty_wait_until_sent, so that it's a better name to export in ksyms.c. Sat Feb 4 23:36:20 1995 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_close): Added missing check for closing_wait2 being ASYNC_CLOSING_WAIT_NONE. Thu Jan 26 09:02:49 1995 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_init, set_serial_info, get_serial_info, rs_close): Support close_wait in the serial driver. This is helpful for slow devices (like serial plotters) so that their outputs don't get flushed upon device close. This has to be configurable because normally we don't want ports to be hung up for long periods of time during a close when they are not connected to a device, or the device is powered off. The default is to wait 30 seconds; in the case of a very slow device, the close_wait timeout should be lengthened. If it is set to 0, the kernel will wait forever for all of the data to be transmitted. Thu Jan 17 01:17:20 1995 Theodore Y. Ts'o (tytso@rt-11) * serial.c (startup, change_speed, rs_init): Add support to detect the StarTech 16650 chip. Treat it as a 16450 for now, because of its FIFO bugs. Thu Jan 5 21:21:57 1995 * serial.c: (receive_char): Added counter to prevent infinite loop when a PCMCIA serial device is ejected. Thu Dec 29 17:53:48 1994 * tty_io.c (check_tty_count): New procedure which checks tty->count to make sure that it matches with the number of open file descriptors which point at the structure. If the number doesn't match, it prints a warning message. Wed Dec 28 15:41:51 1994 * tty_io.c (do_tty_hangup, disassociate_ctty): At hangup time, save the tty's current foreground process group in the session leader's task structure. When the session leader terminates, send a SIGHUP, SIGCONT to that process group. This is not required by POSIX, but it's not prohibited either, and it appears to be the least intrusive way to fix a problem that dialup servers have with orphaned process groups caused by modem hangups. Thu Dec 8 14:52:11 1994 * serial.c (rs_ioctl): Don't allow most ioctl's if the serial port isn't initialized. * serial.c (rs_close): Don't clear the IER if the serial port isn't initialized. * serial.c (block_til_ready): Don't try to block on the dialin port if the serial port isn't initialized. Wed Dec 7 10:48:30 1994 Si Park (si@wimpol.demon.co.uk) * tty_io.c (tty_register_driver): Fix bug when linking onto the tty_drivers list. We now test that there are elements already on the list before setting the back link from the first element to the new driver. * tty_io.c (tty_unregister_driver): Fix bug in unlinking the specified driver from the tty_drivers list. We were not setting the back link correctly. This used to result in a dangling back link pointer and cause panics on the next call to get_tty_driver(). Tue Nov 29 10:21:09 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_io.c (tty_unregister_driver): Fix bug in tty_unregister_driver where the pointer to the refcount is tested, instead of the refcount itself. This caused tty_unregister_driver to always return EBUSY. Sat Nov 26 11:59:24 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_io.c (tty_ioctl): Add support for the new ioctl TIOCTTYGSTRUCT, which allow a kernel debugging program direct read access to the tty and tty_driver structures. Fri Nov 25 17:26:22 1994 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_set_termios): Don't wake up processes blocked in open when the CLOCAL flag changes, since a blocking open only samples the CLOCAL flag once when it blocks, and doesn't check it again. (n.b. FreeBSD has a different behavior for blocking opens; it's not clear whether Linux or FreeBSD's interpretation is correct. POSIX doesn't give clear guidance on this issue, so this may change in the future....) * serial.c (block_til_ready): Use the correct termios structure to check the CLOCAL flag. If the cuaXX device is active, then check the saved termios for the ttySXX device. Otherwise, use the currently active termios structure. Sun Nov 6 21:05:44 1994 Theodore Y. Ts'o (tytso@rt-11) * serial.c (change_speed): Add support for direct access of 57,600 and 115,200 bps. Wed Nov 2 10:32:36 1994 Theodore Y. Ts'o (tytso@rt-11) * n_tty.c (n_tty_receive_room): Only allow excess characters through if we are in ICANON mode *and* there are other no pending lines in the buffer. Otherwise cut and paste over 4k breaks. Sat Oct 29 18:17:34 1994 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_ioctl, get_lsr_info): Added patch suggested by Arne Riiber so that user mode programs can tell when the transmitter shift register is empty. Thu Oct 27 23:14:29 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_ioctl.c (wait_until_sent): Added debugging printk statements (under the #ifdef TTY_DEBUG_WAIT_UNTIL_SENT) * serial.c (rs_interrupt, rs_interrupt_single, receive_chars, change_speed, rs_close): rs_close now disables receiver interrupts when closing the serial port. This allows the serial port to close quickly when Linux and a modem (or a mouse) are engaged in an echo war; when closing the serial port, we now first stop listening to incoming characters, and *then* wait for the transmit buffer to drain. In order to make this change, the info->read_status_mask is now used to control what bits of the line status register are looked at in the interrupt routine in all cases; previously it was only used in receive_chars to select a few of the status bits. Mon Oct 24 23:36:21 1994 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_close): Add a timeout to the transmitter flush loop; this is just a sanity check in case we have flaky (or non-existent-but-configured-by-the-user) hardware. Fri Oct 21 09:37:23 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_io.c (tty_fasync): When asynchronous I/O is enabled, if the process or process group has not be specified yet, set it to be the tty's process group, or if that is not yet set, to the current process's pid. Thu Oct 20 23:17:28 1994 Theodore Y. Ts'o (tytso@rt-11) * n_tty.c (n_tty_receive_room): If we are doing input canonicalization, let as many characters through as possible, so that the excess characters can be "beeped". Tue Oct 18 10:02:43 1994 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_start): Removed an incorrect '!' that was preventing transmit interrupts from being re-enabled in rs_start(). Fortunately in most cases it would be re-enabled elsewhere, but this still should be fixed correctly. Sun Oct 9 23:46:03 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_io.c (do_tty_hangup): If the tty driver flags TTY_DRIVER_RESET_TERMIOS is set, then reset the termios settings back to the driver's initial configuration. This allows the termios settings to be reset even if a process has hung up file descriptors keeping a pty's termios from being freed and reset. * tty_io.c (release_dev): Fix memory leak. The pty's other termios structure should also be freed. * serial.c (rs_close, shutdown): Change how we wait for the transmitter to completely drain before shutting down the serial port. We now do it by scheduling in another process instead of busy looping with the interrupts turned on. This may eliminate some race condition problems that some people seem to be reporting. Sun Sep 25 14:18:14 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_io.c (release_dev): When freeing a tty make sure that both the tty and the o_tty (if present) aren't a process's controlling tty. (Previously, we only checked the tty.) * serial.c (change_speed): Only enable the Modem Status Interrupt for a port if CLOCAL is not set or CRTSCTS is set. If we're not checking the carrier detect and CTS line, there's no point in enabling the modem status interrupt. This will save spurious interrupts from slowing down systems who have terminals that don't support either line. (Of course, if you want only one of CD and CTS support, you will need a properly wired serial cable.) Thu Sep 22 08:32:48 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_io.c (do_SAK): Return if tty is null. * tty_io.c (_tty_name): Return "NULL tty" if the passed in tty is NULL. Sat Sep 17 13:19:25 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_ioctl.c (n_tty_ioctl): Fix TIOCGLCKTRMIOS and TIOCSLCKTRMIOS, which were totally broken. Remove extra indirection from argument; it should be a struct termios *, not a struct termios **. &real_tty->termios_locked should have been real_tty->termios_locked. This caused us to be reading and writing the termios_locked structure to random places in kernel memory. * tty_io.c (release_dev): Oops! Forgot to delete a critical kfree of the locked_termios. This leaves the locked_termios structure pointed at a freed object. Fri Sep 16 08:13:25 1994 Theodore Y. Ts'o (tytso@rt-11) * tty_io.c (tty_open): Don't check for an exclusive open until after the device specific open routine has been called. Otherwise, the serial device ref counting will be screwed up. * serial.c (rs_open, block_til_ready): Don't set termios structure until after block_til_ready has returned successfully. Modify block_til_ready to check the normal_termios structure directly, so it doesn't rely on termios being set before it's called. Thu Sep 15 23:34:01 1994 Theodore Y. Ts'o (tytso@rt-11) * serial.c (rs_close): Turn off interrupts during rs_close() to prevent a race condition with the hangup code (which runs during a software interrupt). * tty_io.c (release_dev): Don't free the locked_termios structure; its state must be retained across device opens. * tty_io.c (tty_unregister_driver): Added function to unregister a tty driver. (For loadable device drivers.)