freebsd-skq/sys/dev/sio/sio.c

2643 lines
69 KiB
C
Raw Normal View History

/*-
* Copyright (c) 1991 The Regents of the University of California.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* from: @(#)com.c 7.5 (Berkeley) 5/16/91
* from: i386/isa sio.c,v 1.234
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_compat.h"
#include "opt_gdb.h"
#include "opt_kdb.h"
#include "opt_sio.h"
/*
* Serial driver, based on 386BSD-0.1 com driver.
* Mostly rewritten to use pseudo-DMA.
* Works for National Semiconductor NS8250-NS16550AF UARTs.
* COM driver, based on HP dca driver.
*
2005-09-22 06:01:44 +00:00
* Changes for PC Card integration:
* - Added PC Card driver table and handlers
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/conf.h>
#include <sys/fcntl.h>
#include <sys/interrupt.h>
#include <sys/kdb.h>
#include <sys/kernel.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/module.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/reboot.h>
#include <sys/serial.h>
#include <sys/sysctl.h>
#include <sys/syslog.h>
#include <sys/tty.h>
#include <machine/bus.h>
#include <sys/rman.h>
#include <sys/timepps.h>
#include <sys/uio.h>
#include <sys/cons.h>
#include <isa/isavar.h>
#include <machine/resource.h>
#include <dev/sio/sioreg.h>
#include <dev/sio/siovar.h>
#ifdef COM_ESP
#include <dev/ic/esp.h>
#endif
#include <dev/ic/ns16550.h>
#define LOTS_OF_EVENTS 64 /* helps separate urgent events from input */
#ifdef COM_MULTIPORT
/* checks in flags for multiport and which is multiport "master chip"
* for a given card
*/
#define COM_ISMULTIPORT(flags) ((flags) & 0x01)
#define COM_MPMASTER(flags) (((flags) >> 8) & 0x0ff)
#define COM_NOTAST4(flags) ((flags) & 0x04)
#else
#define COM_ISMULTIPORT(flags) (0)
#endif /* COM_MULTIPORT */
#define COM_C_IIR_TXRDYBUG 0x80000
#define COM_CONSOLE(flags) ((flags) & 0x10)
#define COM_DEBUGGER(flags) ((flags) & 0x80)
#define COM_FIFOSIZE(flags) (((flags) & 0xff000000) >> 24)
#define COM_FORCECONSOLE(flags) ((flags) & 0x20)
#define COM_IIR_TXRDYBUG(flags) ((flags) & COM_C_IIR_TXRDYBUG)
#define COM_LLCONSOLE(flags) ((flags) & 0x40)
#define COM_LOSESOUTINTS(flags) ((flags) & 0x08)
#define COM_NOFIFO(flags) ((flags) & 0x02)
#define COM_NOPROBE(flags) ((flags) & 0x40000)
#define COM_NOSCR(flags) ((flags) & 0x100000)
#define COM_PPSCTS(flags) ((flags) & 0x10000)
#define COM_ST16650A(flags) ((flags) & 0x20000)
#define COM_TI16754(flags) ((flags) & 0x200000)
#define sio_getreg(com, off) \
(bus_space_read_1((com)->bst, (com)->bsh, (off)))
#define sio_setreg(com, off, value) \
(bus_space_write_1((com)->bst, (com)->bsh, (off), (value)))
/*
* com state bits.
* (CS_BUSY | CS_TTGO) and (CS_BUSY | CS_TTGO | CS_ODEVREADY) must be higher
* than the other bits so that they can be tested as a group without masking
* off the low bits.
*
* The following com and tty flags correspond closely:
* CS_BUSY = TS_BUSY (maintained by comstart(), siopoll() and
* comstop())
* CS_TTGO = ~TS_TTSTOP (maintained by comparam() and comstart())
* CS_CTS_OFLOW = CCTS_OFLOW (maintained by comparam())
* CS_RTS_IFLOW = CRTS_IFLOW (maintained by comparam())
* TS_FLUSH is not used.
* XXX I think TIOCSETA doesn't clear TS_TTSTOP when it clears IXON.
* XXX CS_*FLOW should be CF_*FLOW in com->flags (control flags not state).
*/
#define CS_BUSY 0x80 /* output in progress */
#define CS_TTGO 0x40 /* output not stopped by XOFF */
#define CS_ODEVREADY 0x20 /* external device h/w ready (CTS) */
#define CS_CHECKMSR 1 /* check of MSR scheduled */
#define CS_CTS_OFLOW 2 /* use CTS output flow control */
#define CS_ODONE 4 /* output completed */
#define CS_RTS_IFLOW 8 /* use RTS input flow control */
#define CSE_BUSYCHECK 1 /* siobusycheck() scheduled */
static char const * const error_desc[] = {
#define CE_OVERRUN 0
"silo overflow",
#define CE_INTERRUPT_BUF_OVERFLOW 1
"interrupt-level buffer overflow",
#define CE_TTY_BUF_OVERFLOW 2
"tty-level buffer overflow",
};
#define CE_NTYPES 3
#define CE_RECORD(com, errnum) (++(com)->delta_error_counts[errnum])
/* types. XXX - should be elsewhere */
typedef u_int Port_t; /* hardware port */
typedef u_char bool_t; /* boolean */
/* queue of linear buffers */
struct lbq {
u_char *l_head; /* next char to process */
u_char *l_tail; /* one past the last char to process */
struct lbq *l_next; /* next in queue */
bool_t l_queued; /* nonzero if queued */
};
/* com device structure */
struct com_s {
u_char state; /* miscellaneous flag bits */
u_char cfcr_image; /* copy of value written to CFCR */
#ifdef COM_ESP
bool_t esp; /* is this unit a hayes esp board? */
#endif
u_char extra_state; /* more flag bits, separate for order trick */
u_char fifo_image; /* copy of value written to FIFO */
bool_t hasfifo; /* nonzero for 16550 UARTs */
bool_t loses_outints; /* nonzero if device loses output interrupts */
u_char mcr_image; /* copy of value written to MCR */
#ifdef COM_MULTIPORT
bool_t multiport; /* is this unit part of a multiport device? */
#endif /* COM_MULTIPORT */
bool_t no_irq; /* nonzero if irq is not attached */
bool_t gone; /* hardware disappeared */
bool_t poll; /* nonzero if polling is required */
bool_t poll_output; /* nonzero if polling for output is required */
bool_t st16650a; /* nonzero if Startech 16650A compatible */
int unit; /* unit number */
u_int flags; /* copy of device flags */
u_int tx_fifo_size;
/*
* The high level of the driver never reads status registers directly
* because there would be too many side effects to handle conveniently.
* Instead, it reads copies of the registers stored here by the
* interrupt handler.
*/
u_char last_modem_status; /* last MSR read by intr handler */
u_char prev_modem_status; /* last MSR handled by high level */
u_char *ibuf; /* start of input buffer */
u_char *ibufend; /* end of input buffer */
u_char *ibufold; /* old input buffer, to be freed */
u_char *ihighwater; /* threshold in input buffer */
u_char *iptr; /* next free spot in input buffer */
int ibufsize; /* size of ibuf (not include error bytes) */
int ierroff; /* offset of error bytes in ibuf */
struct lbq obufq; /* head of queue of output buffers */
struct lbq obufs[2]; /* output buffers */
bus_space_tag_t bst;
bus_space_handle_t bsh;
Port_t data_port; /* i/o ports */
#ifdef COM_ESP
Port_t esp_port;
#endif
Port_t int_ctl_port;
Port_t int_id_port;
Port_t modem_ctl_port;
Port_t line_status_port;
Port_t modem_status_port;
struct tty *tp; /* cross reference */
struct pps_state pps;
int pps_bit;
#ifdef KDB
int alt_brk_state;
#endif
u_long bytes_in; /* statistics */
u_long bytes_out;
u_int delta_error_counts[CE_NTYPES];
u_long error_counts[CE_NTYPES];
u_long rclk;
struct resource *irqres;
struct resource *ioportres;
int ioportrid;
void *cookie;
/*
* Data area for output buffers. Someday we should build the output
* buffer queue without copying data.
*/
u_char obuf1[256];
u_char obuf2[256];
};
#ifdef COM_ESP
static int espattach(struct com_s *com, Port_t esp_port);
#endif
static void combreak(struct tty *tp, int sig);
static timeout_t siobusycheck;
static u_int siodivisor(u_long rclk, speed_t speed);
static void comclose(struct tty *tp);
static int comopen(struct tty *tp, struct cdev *dev);
static void sioinput(struct com_s *com);
static void siointr1(struct com_s *com);
static int siointr(void *arg);
static int commodem(struct tty *tp, int sigon, int sigoff);
static int comparam(struct tty *tp, struct termios *t);
static void siopoll(void *);
static void siosettimeout(void);
static int siosetwater(struct com_s *com, speed_t speed);
static void comstart(struct tty *tp);
static void comstop(struct tty *tp, int rw);
static timeout_t comwakeup;
char sio_driver_name[] = "sio";
static struct mtx sio_lock;
static int sio_inited;
/* table and macro for fast conversion from a unit number to its com struct */
devclass_t sio_devclass;
/*
* XXX Assmues that devclass_get_device, devclass_get_softc and
* device_get_softc are fast interrupt safe. The current implementation
* of these functions are.
*/
#define com_addr(unit) ((struct com_s *) \
devclass_get_softc(sio_devclass, unit)) /* XXX */
int comconsole = -1;
static volatile speed_t comdefaultrate = CONSPEED;
static u_long comdefaultrclk = DEFAULT_RCLK;
SYSCTL_ULONG(_machdep, OID_AUTO, conrclk, CTLFLAG_RW, &comdefaultrclk, 0, "");
static speed_t gdbdefaultrate = GDBSPEED;
SYSCTL_UINT(_machdep, OID_AUTO, gdbspeed, CTLFLAG_RW,
&gdbdefaultrate, GDBSPEED, "");
static u_int com_events; /* input chars + weighted output completions */
static Port_t siocniobase;
static int siocnunit = -1;
static void *sio_slow_ih;
static void *sio_fast_ih;
static int sio_timeout;
static int sio_timeouts_until_log;
static struct callout_handle sio_timeout_handle
= CALLOUT_HANDLE_INITIALIZER(&sio_timeout_handle);
static int sio_numunits;
#ifdef GDB
static Port_t siogdbiobase = 0;
#endif
#ifdef COM_ESP
/* XXX configure this properly. */
/* XXX quite broken for new-bus. */
static Port_t likely_com_ports[] = { 0x3f8, 0x2f8, 0x3e8, 0x2e8, };
static Port_t likely_esp_ports[] = { 0x140, 0x180, 0x280, 0 };
#endif
/*
* handle sysctl read/write requests for console speed
*
* In addition to setting comdefaultrate for I/O through /dev/console,
* also set the initial and lock values for the /dev/ttyXX device
* if there is one associated with the console. Finally, if the /dev/tty
* device has already been open, change the speed on the open running port
* itself.
*/
static int
sysctl_machdep_comdefaultrate(SYSCTL_HANDLER_ARGS)
{
int error, s;
speed_t newspeed;
struct com_s *com;
struct tty *tp;
newspeed = comdefaultrate;
error = sysctl_handle_opaque(oidp, &newspeed, sizeof newspeed, req);
if (error || !req->newptr)
return (error);
comdefaultrate = newspeed;
if (comconsole < 0) /* serial console not selected? */
return (0);
com = com_addr(comconsole);
if (com == NULL)
return (ENXIO);
tp = com->tp;
if (tp == NULL)
return (ENXIO);
/*
* set the initial and lock rates for /dev/ttydXX and /dev/cuaXX
* (note, the lock rates really are boolean -- if non-zero, disallow
* speed changes)
*/
tp->t_init_in.c_ispeed = tp->t_init_in.c_ospeed =
tp->t_lock_in.c_ispeed = tp->t_lock_in.c_ospeed =
tp->t_init_out.c_ispeed = tp->t_init_out.c_ospeed =
tp->t_lock_out.c_ispeed = tp->t_lock_out.c_ospeed = comdefaultrate;
if (tp->t_state & TS_ISOPEN) {
tp->t_termios.c_ispeed =
tp->t_termios.c_ospeed = comdefaultrate;
s = spltty();
error = comparam(tp, &tp->t_termios);
splx(s);
}
return error;
}
SYSCTL_PROC(_machdep, OID_AUTO, conspeed, CTLTYPE_INT | CTLFLAG_RW,
0, 0, sysctl_machdep_comdefaultrate, "I", "");
2005-12-07 07:23:53 +00:00
TUNABLE_INT("machdep.conspeed", __DEVOLATILE(int *, &comdefaultrate));
#define SET_FLAG(dev, bit) device_set_flags(dev, device_get_flags(dev) | (bit))
#define CLR_FLAG(dev, bit) device_set_flags(dev, device_get_flags(dev) & ~(bit))
/*
* Unload the driver and clear the table.
* XXX this is mostly wrong.
* XXX TODO:
* This is usually called when the card is ejected, but
2002-10-11 20:22:20 +00:00
* can be caused by a kldunload of a controller driver.
* The idea is to reset the driver's view of the device
* and ensure that any driver entry points such as
* read and write do not hang.
*/
int
siodetach(device_t dev)
{
struct com_s *com;
com = (struct com_s *) device_get_softc(dev);
if (com == NULL) {
device_printf(dev, "NULL com in siounload\n");
return (0);
}
com->gone = TRUE;
if (com->tp)
ttyfree(com->tp);
if (com->irqres) {
bus_teardown_intr(dev, com->irqres, com->cookie);
bus_release_resource(dev, SYS_RES_IRQ, 0, com->irqres);
}
if (com->ioportres)
bus_release_resource(dev, SYS_RES_IOPORT, com->ioportrid,
com->ioportres);
if (com->ibuf != NULL)
free(com->ibuf, M_DEVBUF);
device_set_softc(dev, NULL);
free(com, M_DEVBUF);
return (0);
}
int
sioprobe(dev, xrid, rclk, noprobe)
device_t dev;
int xrid;
u_long rclk;
int noprobe;
{
#if 0
static bool_t already_init;
device_t xdev;
#endif
struct com_s *com;
u_int divisor;
bool_t failures[10];
int fn;
device_t idev;
Port_t iobase;
intrmask_t irqmap[4];
intrmask_t irqs;
u_char mcr_image;
int result;
u_long xirq;
u_int flags = device_get_flags(dev);
int rid;
struct resource *port;
rid = xrid;
port = bus_alloc_resource(dev, SYS_RES_IOPORT, &rid,
0, ~0, IO_COMSIZE, RF_ACTIVE);
if (!port)
return (ENXIO);
com = malloc(sizeof(*com), M_DEVBUF, M_NOWAIT | M_ZERO);
if (com == NULL) {
bus_release_resource(dev, SYS_RES_IOPORT, rid, port);
return (ENOMEM);
}
device_set_softc(dev, com);
com->bst = rman_get_bustag(port);
com->bsh = rman_get_bushandle(port);
if (rclk == 0)
rclk = DEFAULT_RCLK;
com->rclk = rclk;
while (sio_inited != 2)
if (atomic_cmpset_int(&sio_inited, 0, 1)) {
mtx_init(&sio_lock, sio_driver_name, NULL,
(comconsole != -1) ?
MTX_SPIN | MTX_QUIET : MTX_SPIN);
atomic_store_rel_int(&sio_inited, 2);
}
#if 0
/*
* XXX this is broken - when we are first called, there are no
* previously configured IO ports. We could hard code
* 0x3f8, 0x2f8, 0x3e8, 0x2e8 etc but that's probably worse.
* This code has been doing nothing since the conversion since
* "count" is zero the first time around.
*/
if (!already_init) {
/*
* Turn off MCR_IENABLE for all likely serial ports. An unused
* port with its MCR_IENABLE gate open will inhibit interrupts
* from any used port that shares the interrupt vector.
* XXX the gate enable is elsewhere for some multiports.
*/
device_t *devs;
int count, i, xioport;
devclass_get_devices(sio_devclass, &devs, &count);
for (i = 0; i < count; i++) {
xdev = devs[i];
if (device_is_enabled(xdev) &&
bus_get_resource(xdev, SYS_RES_IOPORT, 0, &xioport,
NULL) == 0)
outb(xioport + com_mcr, 0);
}
free(devs, M_TEMP);
already_init = TRUE;
}
#endif
if (COM_LLCONSOLE(flags)) {
printf("sio%d: reserved for low-level i/o\n",
device_get_unit(dev));
bus_release_resource(dev, SYS_RES_IOPORT, rid, port);
device_set_softc(dev, NULL);
free(com, M_DEVBUF);
return (ENXIO);
}
/*
* If the device is on a multiport card and has an AST/4
* compatible interrupt control register, initialize this
* register and prepare to leave MCR_IENABLE clear in the mcr.
* Otherwise, prepare to set MCR_IENABLE in the mcr.
* Point idev to the device struct giving the correct id_irq.
* This is the struct for the master device if there is one.
*/
idev = dev;
mcr_image = MCR_IENABLE;
#ifdef COM_MULTIPORT
if (COM_ISMULTIPORT(flags)) {
Port_t xiobase;
u_long io;
idev = devclass_get_device(sio_devclass, COM_MPMASTER(flags));
if (idev == NULL) {
printf("sio%d: master device %d not configured\n",
device_get_unit(dev), COM_MPMASTER(flags));
idev = dev;
}
if (!COM_NOTAST4(flags)) {
if (bus_get_resource(idev, SYS_RES_IOPORT, 0, &io,
NULL) == 0) {
xiobase = io;
if (bus_get_resource(idev, SYS_RES_IRQ, 0,
NULL, NULL) == 0)
outb(xiobase + com_scr, 0x80);
else
outb(xiobase + com_scr, 0);
}
mcr_image = 0;
}
}
#endif /* COM_MULTIPORT */
if (bus_get_resource(idev, SYS_RES_IRQ, 0, NULL, NULL) != 0)
mcr_image = 0;
bzero(failures, sizeof failures);
iobase = rman_get_start(port);
/*
* We don't want to get actual interrupts, just masked ones.
* Interrupts from this line should already be masked in the ICU,
* but mask them in the processor as well in case there are some
* (misconfigured) shared interrupts.
*/
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
/* EXTRA DELAY? */
/*
* For the TI16754 chips, set prescaler to 1 (4 is often the
* default after-reset value) as otherwise it's impossible to
* get highest baudrates.
*/
if (COM_TI16754(flags)) {
u_char cfcr, efr;
cfcr = sio_getreg(com, com_cfcr);
sio_setreg(com, com_cfcr, CFCR_EFR_ENABLE);
efr = sio_getreg(com, com_efr);
/* Unlock extended features to turn off prescaler. */
sio_setreg(com, com_efr, efr | EFR_EFE);
/* Disable EFR. */
sio_setreg(com, com_cfcr, (cfcr != CFCR_EFR_ENABLE) ? cfcr : 0);
/* Turn off prescaler. */
sio_setreg(com, com_mcr,
sio_getreg(com, com_mcr) & ~MCR_PRESCALE);
sio_setreg(com, com_cfcr, CFCR_EFR_ENABLE);
sio_setreg(com, com_efr, efr);
sio_setreg(com, com_cfcr, cfcr);
}
/*
* Initialize the speed and the word size and wait long enough to
* drain the maximum of 16 bytes of junk in device output queues.
* The speed is undefined after a master reset and must be set
* before relying on anything related to output. There may be
* junk after a (very fast) soft reboot and (apparently) after
* master reset.
* XXX what about the UART bug avoided by waiting in comparam()?
* We don't want to to wait long enough to drain at 2 bps.
*/
if (iobase == siocniobase)
DELAY((16 + 1) * 1000000 / (comdefaultrate / 10));
else {
sio_setreg(com, com_cfcr, CFCR_DLAB | CFCR_8BITS);
divisor = siodivisor(rclk, SIO_TEST_SPEED);
sio_setreg(com, com_dlbl, divisor & 0xff);
sio_setreg(com, com_dlbh, divisor >> 8);
sio_setreg(com, com_cfcr, CFCR_8BITS);
DELAY((16 + 1) * 1000000 / (SIO_TEST_SPEED / 10));
}
/*
2007-10-12 06:03:46 +00:00
* Enable the interrupt gate and disable device interrupts. This
* should leave the device driving the interrupt line low and
* guarantee an edge trigger if an interrupt can be generated.
*/
/* EXTRA DELAY? */
sio_setreg(com, com_mcr, mcr_image);
sio_setreg(com, com_ier, 0);
DELAY(1000); /* XXX */
irqmap[0] = isa_irq_pending();
/*
* Attempt to set loopback mode so that we can send a null byte
* without annoying any external device.
*/
/* EXTRA DELAY? */
sio_setreg(com, com_mcr, mcr_image | MCR_LOOPBACK);
/*
* Attempt to generate an output interrupt. On 8250's, setting
* IER_ETXRDY generates an interrupt independent of the current
* setting and independent of whether the THR is empty. On 16450's,
* setting IER_ETXRDY generates an interrupt independent of the
* current setting. On 16550A's, setting IER_ETXRDY only
* generates an interrupt when IER_ETXRDY is not already set.
*/
sio_setreg(com, com_ier, IER_ETXRDY);
/*
* On some 16x50 incompatibles, setting IER_ETXRDY doesn't generate
* an interrupt. They'd better generate one for actually doing
* output. Loopback may be broken on the same incompatibles but
* it's unlikely to do more than allow the null byte out.
*/
sio_setreg(com, com_data, 0);
if (iobase == siocniobase)
DELAY((1 + 2) * 1000000 / (comdefaultrate / 10));
else
DELAY((1 + 2) * 1000000 / (SIO_TEST_SPEED / 10));
/*
* Turn off loopback mode so that the interrupt gate works again
* (MCR_IENABLE was hidden). This should leave the device driving
* an interrupt line high. It doesn't matter if the interrupt
* line oscillates while we are not looking at it, since interrupts
* are disabled.
*/
/* EXTRA DELAY? */
sio_setreg(com, com_mcr, mcr_image);
/*
* It seems my Xircom CBEM56G Cardbus modem wants to be reset
* to 8 bits *again*, or else probe test 0 will fail.
* gwk@sgi.com, 4/19/2001
*/
sio_setreg(com, com_cfcr, CFCR_8BITS);
/*
* Some PCMCIA cards (Palido 321s, DC-1S, ...) have the "TXRDY bug",
* so we probe for a buggy IIR_TXRDY implementation even in the
* noprobe case. We don't probe for it in the !noprobe case because
* noprobe is always set for PCMCIA cards and the problem is not
* known to affect any other cards.
*/
if (noprobe) {
/* Read IIR a few times. */
for (fn = 0; fn < 2; fn ++) {
DELAY(10000);
failures[6] = sio_getreg(com, com_iir);
}
/* IIR_TXRDY should be clear. Is it? */
result = 0;
if (failures[6] & IIR_TXRDY) {
/*
* No. We seem to have the bug. Does our fix for
* it work?
*/
sio_setreg(com, com_ier, 0);
if (sio_getreg(com, com_iir) & IIR_NOPEND) {
/* Yes. We discovered the TXRDY bug! */
SET_FLAG(dev, COM_C_IIR_TXRDYBUG);
} else {
/* No. Just fail. XXX */
result = ENXIO;
sio_setreg(com, com_mcr, 0);
}
} else {
/* Yes. No bug. */
CLR_FLAG(dev, COM_C_IIR_TXRDYBUG);
}
sio_setreg(com, com_ier, 0);
sio_setreg(com, com_cfcr, CFCR_8BITS);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
bus_release_resource(dev, SYS_RES_IOPORT, rid, port);
if (iobase == siocniobase)
result = 0;
/*
* XXX: Since we don't return 0, we shouldn't be relying on
* the softc that we set to persist to the call to attach
* since other probe routines may be called, and the malloc
* here causes subr_bus to not allocate anything for the
* other probes. Instead, this softc is preserved and other
* probe routines can corrupt it.
*/
if (result != 0) {
device_set_softc(dev, NULL);
free(com, M_DEVBUF);
}
return (result == 0 ? BUS_PROBE_DEFAULT + 1 : result);
}
/*
* Check that
* o the CFCR, IER and MCR in UART hold the values written to them
* (the values happen to be all distinct - this is good for
* avoiding false positive tests from bus echoes).
* o an output interrupt is generated and its vector is correct.
* o the interrupt goes away when the IIR in the UART is read.
*/
/* EXTRA DELAY? */
failures[0] = sio_getreg(com, com_cfcr) - CFCR_8BITS;
failures[1] = sio_getreg(com, com_ier) - IER_ETXRDY;
failures[2] = sio_getreg(com, com_mcr) - mcr_image;
DELAY(10000); /* Some internal modems need this time */
irqmap[1] = isa_irq_pending();
failures[4] = (sio_getreg(com, com_iir) & IIR_IMASK) - IIR_TXRDY;
DELAY(1000); /* XXX */
irqmap[2] = isa_irq_pending();
failures[6] = (sio_getreg(com, com_iir) & IIR_IMASK) - IIR_NOPEND;
/*
* Turn off all device interrupts and check that they go off properly.
* Leave MCR_IENABLE alone. For ports without a master port, it gates
* the OUT2 output of the UART to
* the ICU input. Closing the gate would give a floating ICU input
* (unless there is another device driving it) and spurious interrupts.
* (On the system that this was first tested on, the input floats high
* and gives a (masked) interrupt as soon as the gate is closed.)
*/
sio_setreg(com, com_ier, 0);
sio_setreg(com, com_cfcr, CFCR_8BITS); /* dummy to avoid bus echo */
failures[7] = sio_getreg(com, com_ier);
DELAY(1000); /* XXX */
irqmap[3] = isa_irq_pending();
failures[9] = (sio_getreg(com, com_iir) & IIR_IMASK) - IIR_NOPEND;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
irqs = irqmap[1] & ~irqmap[0];
if (bus_get_resource(idev, SYS_RES_IRQ, 0, &xirq, NULL) == 0 &&
((1 << xirq) & irqs) == 0) {
printf(
"sio%d: configured irq %ld not in bitmap of probed irqs %#x\n",
device_get_unit(dev), xirq, irqs);
printf(
"sio%d: port may not be enabled\n",
device_get_unit(dev));
}
if (bootverbose)
printf("sio%d: irq maps: %#x %#x %#x %#x\n",
device_get_unit(dev),
irqmap[0], irqmap[1], irqmap[2], irqmap[3]);
result = 0;
for (fn = 0; fn < sizeof failures; ++fn)
if (failures[fn]) {
sio_setreg(com, com_mcr, 0);
result = ENXIO;
if (bootverbose) {
printf("sio%d: probe failed test(s):",
device_get_unit(dev));
for (fn = 0; fn < sizeof failures; ++fn)
if (failures[fn])
printf(" %d", fn);
printf("\n");
}
break;
}
bus_release_resource(dev, SYS_RES_IOPORT, rid, port);
if (iobase == siocniobase)
result = 0;
/*
* XXX: Since we don't return 0, we shouldn't be relying on the softc
* that we set to persist to the call to attach since other probe
* routines may be called, and the malloc here causes subr_bus to not
* allocate anything for the other probes. Instead, this softc is
* preserved and other probe routines can corrupt it.
*/
if (result != 0) {
device_set_softc(dev, NULL);
free(com, M_DEVBUF);
}
return (result == 0 ? BUS_PROBE_DEFAULT + 1 : result);
}
#ifdef COM_ESP
static int
espattach(com, esp_port)
struct com_s *com;
Port_t esp_port;
{
u_char dips;
u_char val;
/*
* Check the ESP-specific I/O port to see if we're an ESP
* card. If not, return failure immediately.
*/
if ((inb(esp_port) & 0xf3) == 0) {
printf(" port 0x%x is not an ESP board?\n", esp_port);
return (0);
}
/*
* We've got something that claims to be a Hayes ESP card.
* Let's hope so.
*/
/* Get the dip-switch configuration */
outb(esp_port + ESP_CMD1, ESP_GETDIPS);
dips = inb(esp_port + ESP_STATUS1);
/*
* Bits 0,1 of dips say which COM port we are.
*/
if (rman_get_start(com->ioportres) == likely_com_ports[dips & 0x03])
printf(" : ESP");
else {
printf(" esp_port has com %d\n", dips & 0x03);
return (0);
}
/*
* Check for ESP version 2.0 or later: bits 4,5,6 = 010.
*/
outb(esp_port + ESP_CMD1, ESP_GETTEST);
val = inb(esp_port + ESP_STATUS1); /* clear reg 1 */
val = inb(esp_port + ESP_STATUS2);
if ((val & 0x70) < 0x20) {
printf("-old (%o)", val & 0x70);
return (0);
}
/*
* Check for ability to emulate 16550: bit 7 == 1
*/
if ((dips & 0x80) == 0) {
printf(" slave");
return (0);
}
/*
* Okay, we seem to be a Hayes ESP card. Whee.
*/
com->esp = TRUE;
com->esp_port = esp_port;
return (1);
}
#endif /* COM_ESP */
int
sioattach(dev, xrid, rclk)
device_t dev;
int xrid;
u_long rclk;
{
struct com_s *com;
#ifdef COM_ESP
Port_t *espp;
#endif
Port_t iobase;
int unit;
u_int flags;
int rid;
struct resource *port;
int ret;
int error;
struct tty *tp;
rid = xrid;
port = bus_alloc_resource(dev, SYS_RES_IOPORT, &rid,
0, ~0, IO_COMSIZE, RF_ACTIVE);
if (!port)
return (ENXIO);
iobase = rman_get_start(port);
unit = device_get_unit(dev);
com = device_get_softc(dev);
flags = device_get_flags(dev);
if (unit >= sio_numunits)
sio_numunits = unit + 1;
/*
* sioprobe() has initialized the device registers as follows:
* o cfcr = CFCR_8BITS.
* It is most important that CFCR_DLAB is off, so that the
* data port is not hidden when we enable interrupts.
* o ier = 0.
* Interrupts are only enabled when the line is open.
* o mcr = MCR_IENABLE, or 0 if the port has AST/4 compatible
* interrupt control register or the config specifies no irq.
* Keeping MCR_DTR and MCR_RTS off might stop the external
* device from sending before we are ready.
*/
bzero(com, sizeof *com);
com->unit = unit;
com->ioportres = port;
com->ioportrid = rid;
com->bst = rman_get_bustag(port);
com->bsh = rman_get_bushandle(port);
com->cfcr_image = CFCR_8BITS;
com->loses_outints = COM_LOSESOUTINTS(flags) != 0;
com->no_irq = bus_get_resource(dev, SYS_RES_IRQ, 0, NULL, NULL) != 0;
com->tx_fifo_size = 1;
com->obufs[0].l_head = com->obuf1;
com->obufs[1].l_head = com->obuf2;
com->data_port = iobase + com_data;
com->int_ctl_port = iobase + com_ier;
com->int_id_port = iobase + com_iir;
com->modem_ctl_port = iobase + com_mcr;
com->mcr_image = inb(com->modem_ctl_port);
com->line_status_port = iobase + com_lsr;
com->modem_status_port = iobase + com_msr;
tp = com->tp = ttyalloc();
tp->t_oproc = comstart;
tp->t_param = comparam;
tp->t_stop = comstop;
tp->t_modem = commodem;
tp->t_break = combreak;
tp->t_close = comclose;
tp->t_open = comopen;
tp->t_sc = com;
if (rclk == 0)
rclk = DEFAULT_RCLK;
com->rclk = rclk;
if (unit == comconsole)
ttyconsolemode(tp, comdefaultrate);
error = siosetwater(com, tp->t_init_in.c_ispeed);
mtx_unlock_spin(&sio_lock);
if (error) {
/*
* Leave i/o resources allocated if this is a `cn'-level
* console, so that other devices can't snarf them.
*/
if (iobase != siocniobase)
bus_release_resource(dev, SYS_RES_IOPORT, rid, port);
return (ENOMEM);
}
/* attempt to determine UART type */
printf("sio%d: type", unit);
if (!COM_ISMULTIPORT(flags) &&
!COM_IIR_TXRDYBUG(flags) && !COM_NOSCR(flags)) {
u_char scr;
u_char scr1;
u_char scr2;
scr = sio_getreg(com, com_scr);
sio_setreg(com, com_scr, 0xa5);
scr1 = sio_getreg(com, com_scr);
sio_setreg(com, com_scr, 0x5a);
scr2 = sio_getreg(com, com_scr);
sio_setreg(com, com_scr, scr);
if (scr1 != 0xa5 || scr2 != 0x5a) {
printf(" 8250 or not responding");
goto determined_type;
}
}
sio_setreg(com, com_fifo, FIFO_ENABLE | FIFO_RX_HIGH);
DELAY(100);
switch (inb(com->int_id_port) & IIR_FIFO_MASK) {
case FIFO_RX_LOW:
printf(" 16450");
break;
case FIFO_RX_MEDL:
printf(" 16450?");
break;
case FIFO_RX_MEDH:
printf(" 16550?");
break;
case FIFO_RX_HIGH:
if (COM_NOFIFO(flags)) {
printf(" 16550A fifo disabled");
break;
}
com->hasfifo = TRUE;
if (COM_ST16650A(flags)) {
printf(" ST16650A");
com->st16650a = TRUE;
com->tx_fifo_size = 32;
break;
}
if (COM_TI16754(flags)) {
printf(" TI16754");
com->tx_fifo_size = 64;
break;
}
printf(" 16550A");
#ifdef COM_ESP
for (espp = likely_esp_ports; *espp != 0; espp++)
if (espattach(com, *espp)) {
com->tx_fifo_size = 1024;
break;
}
if (com->esp)
break;
#endif
com->tx_fifo_size = COM_FIFOSIZE(flags);
if (com->tx_fifo_size == 0)
com->tx_fifo_size = 16;
else
printf(" lookalike with %u bytes FIFO",
com->tx_fifo_size);
break;
}
#ifdef COM_ESP
if (com->esp) {
/*
* Set 16550 compatibility mode.
* We don't use the ESP_MODE_SCALE bit to increase the
* fifo trigger levels because we can't handle large
* bursts of input.
* XXX flow control should be set in comparam(), not here.
*/
outb(com->esp_port + ESP_CMD1, ESP_SETMODE);
outb(com->esp_port + ESP_CMD2, ESP_MODE_RTS | ESP_MODE_FIFO);
/* Set RTS/CTS flow control. */
outb(com->esp_port + ESP_CMD1, ESP_SETFLOWTYPE);
outb(com->esp_port + ESP_CMD2, ESP_FLOW_RTS);
outb(com->esp_port + ESP_CMD2, ESP_FLOW_CTS);
/* Set flow-control levels. */
outb(com->esp_port + ESP_CMD1, ESP_SETRXFLOW);
outb(com->esp_port + ESP_CMD2, HIBYTE(768));
outb(com->esp_port + ESP_CMD2, LOBYTE(768));
outb(com->esp_port + ESP_CMD2, HIBYTE(512));
outb(com->esp_port + ESP_CMD2, LOBYTE(512));
}
#endif /* COM_ESP */
sio_setreg(com, com_fifo, 0);
determined_type: ;
#ifdef COM_MULTIPORT
if (COM_ISMULTIPORT(flags)) {
device_t masterdev;
com->multiport = TRUE;
printf(" (multiport");
if (unit == COM_MPMASTER(flags))
printf(" master");
printf(")");
masterdev = devclass_get_device(sio_devclass,
COM_MPMASTER(flags));
com->no_irq = (masterdev == NULL || bus_get_resource(masterdev,
SYS_RES_IRQ, 0, NULL, NULL) != 0);
}
#endif /* COM_MULTIPORT */
if (unit == comconsole)
printf(", console");
if (COM_IIR_TXRDYBUG(flags))
printf(" with a buggy IIR_TXRDY implementation");
printf("\n");
if (sio_fast_ih == NULL) {
Reorganize the interrupt handling code a bit to make a few things cleaner and increase flexibility to allow various different approaches to be tried in the future. - Split struct ithd up into two pieces. struct intr_event holds the list of interrupt handlers associated with interrupt sources. struct intr_thread contains the data relative to an interrupt thread. Currently we still provide a 1:1 relationship of events to threads with the exception that events only have an associated thread if there is at least one threaded interrupt handler attached to the event. This means that on x86 we no longer have 4 bazillion interrupt threads with no handlers. It also means that interrupt events with only INTR_FAST handlers no longer have an associated thread either. - Renamed struct intrhand to struct intr_handler to follow the struct intr_foo naming convention. This did require renaming the powerpc MD struct intr_handler to struct ppc_intr_handler. - INTR_FAST no longer implies INTR_EXCL on all architectures except for powerpc. This means that multiple INTR_FAST handlers can attach to the same interrupt and that INTR_FAST and non-INTR_FAST handlers can attach to the same interrupt. Sharing INTR_FAST handlers may not always be desirable, but having sio(4) and uhci(4) fight over an IRQ isn't fun either. Drivers can always still use INTR_EXCL to ask for an interrupt exclusively. The way this sharing works is that when an interrupt comes in, all the INTR_FAST handlers are executed first, and if any threaded handlers exist, the interrupt thread is scheduled afterwards. This type of layout also makes it possible to investigate using interrupt filters ala OS X where the filter determines whether or not its companion threaded handler should run. - Aside from the INTR_FAST changes above, the impact on MD interrupt code is mostly just 's/ithread/intr_event/'. - A new MI ddb command 'show intrs' walks the list of interrupt events dumping their state. It also has a '/v' verbose switch which dumps info about all of the handlers attached to each event. - We currently don't destroy an interrupt thread when the last threaded handler is removed because it would suck for things like ppbus(8)'s braindead behavior. The code is present, though, it is just under #if 0 for now. - Move the code to actually execute the threaded handlers for an interrrupt event into a separate function so that ithread_loop() becomes more readable. Previously this code was all in the middle of ithread_loop() and indented halfway across the screen. - Made struct intr_thread private to kern_intr.c and replaced td_ithd with a thread private flag TDP_ITHREAD. - In statclock, check curthread against idlethread directly rather than curthread's proc against idlethread's proc. (Not really related to intr changes) Tested on: alpha, amd64, i386, sparc64 Tested on: arm, ia64 (older version of patch by cognet and marcel)
2005-10-25 19:48:48 +00:00
swi_add(&tty_intr_event, "sio", siopoll, NULL, SWI_TTY, 0,
&sio_fast_ih);
Reorganize the interrupt handling code a bit to make a few things cleaner and increase flexibility to allow various different approaches to be tried in the future. - Split struct ithd up into two pieces. struct intr_event holds the list of interrupt handlers associated with interrupt sources. struct intr_thread contains the data relative to an interrupt thread. Currently we still provide a 1:1 relationship of events to threads with the exception that events only have an associated thread if there is at least one threaded interrupt handler attached to the event. This means that on x86 we no longer have 4 bazillion interrupt threads with no handlers. It also means that interrupt events with only INTR_FAST handlers no longer have an associated thread either. - Renamed struct intrhand to struct intr_handler to follow the struct intr_foo naming convention. This did require renaming the powerpc MD struct intr_handler to struct ppc_intr_handler. - INTR_FAST no longer implies INTR_EXCL on all architectures except for powerpc. This means that multiple INTR_FAST handlers can attach to the same interrupt and that INTR_FAST and non-INTR_FAST handlers can attach to the same interrupt. Sharing INTR_FAST handlers may not always be desirable, but having sio(4) and uhci(4) fight over an IRQ isn't fun either. Drivers can always still use INTR_EXCL to ask for an interrupt exclusively. The way this sharing works is that when an interrupt comes in, all the INTR_FAST handlers are executed first, and if any threaded handlers exist, the interrupt thread is scheduled afterwards. This type of layout also makes it possible to investigate using interrupt filters ala OS X where the filter determines whether or not its companion threaded handler should run. - Aside from the INTR_FAST changes above, the impact on MD interrupt code is mostly just 's/ithread/intr_event/'. - A new MI ddb command 'show intrs' walks the list of interrupt events dumping their state. It also has a '/v' verbose switch which dumps info about all of the handlers attached to each event. - We currently don't destroy an interrupt thread when the last threaded handler is removed because it would suck for things like ppbus(8)'s braindead behavior. The code is present, though, it is just under #if 0 for now. - Move the code to actually execute the threaded handlers for an interrrupt event into a separate function so that ithread_loop() becomes more readable. Previously this code was all in the middle of ithread_loop() and indented halfway across the screen. - Made struct intr_thread private to kern_intr.c and replaced td_ithd with a thread private flag TDP_ITHREAD. - In statclock, check curthread against idlethread directly rather than curthread's proc against idlethread's proc. (Not really related to intr changes) Tested on: alpha, amd64, i386, sparc64 Tested on: arm, ia64 (older version of patch by cognet and marcel)
2005-10-25 19:48:48 +00:00
swi_add(&clk_intr_event, "sio", siopoll, NULL, SWI_CLOCK, 0,
&sio_slow_ih);
}
com->flags = flags;
com->pps.ppscap = PPS_CAPTUREASSERT | PPS_CAPTURECLEAR;
tp->t_pps = &com->pps;
if (COM_PPSCTS(flags))
com->pps_bit = MSR_CTS;
else
com->pps_bit = MSR_DCD;
pps_init(&com->pps);
rid = 0;
com->irqres = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_ACTIVE);
if (com->irqres) {
ret = bus_setup_intr(dev, com->irqres,
INTR_TYPE_TTY,
siointr, NULL, com,
&com->cookie);
if (ret) {
ret = bus_setup_intr(dev,
com->irqres, INTR_TYPE_TTY,
NULL, (driver_intr_t *)siointr, com, &com->cookie);
if (ret == 0)
device_printf(dev, "unable to activate interrupt in fast mode - using normal mode\n");
}
if (ret)
device_printf(dev, "could not activate interrupt\n");
#if defined(KDB)
/*
* Enable interrupts for early break-to-debugger support
* on the console.
*/
if (ret == 0 && unit == comconsole)
outb(siocniobase + com_ier, IER_ERXRDY | IER_ERLS |
IER_EMSC);
#endif
}
/* We're ready, open the doors... */
ttycreate(tp, TS_CALLOUT, "d%r", unit);
return (0);
}
static int
comopen(struct tty *tp, struct cdev *dev)
{
struct com_s *com;
int i;
com = tp->t_sc;
com->poll = com->no_irq;
com->poll_output = com->loses_outints;
if (com->hasfifo) {
/*
* (Re)enable and drain fifos.
*
* Certain SMC chips cause problems if the fifos
* are enabled while input is ready. Turn off the
* fifo if necessary to clear the input. We test
* the input ready bit after enabling the fifos
* since we've already enabled them in comparam()
* and to handle races between enabling and fresh
* input.
*/
for (i = 0; i < 500; i++) {
sio_setreg(com, com_fifo,
FIFO_RCV_RST | FIFO_XMT_RST
| com->fifo_image);
/*
* XXX the delays are for superstitious
* historical reasons. It must be less than
* the character time at the maximum
* supported speed (87 usec at 115200 bps
* 8N1). Otherwise we might loop endlessly
* if data is streaming in. We used to use
* delays of 100. That usually worked
* because DELAY(100) used to usually delay
* for about 85 usec instead of 100.
*/
DELAY(50);
if (!(inb(com->line_status_port) & LSR_RXRDY))
break;
sio_setreg(com, com_fifo, 0);
DELAY(50);
(void) inb(com->data_port);
}
if (i == 500)
return (EIO);
}
mtx_lock_spin(&sio_lock);
(void) inb(com->line_status_port);
(void) inb(com->data_port);
com->prev_modem_status = com->last_modem_status
= inb(com->modem_status_port);
outb(com->int_ctl_port,
IER_ERXRDY | IER_ERLS | IER_EMSC
| (COM_IIR_TXRDYBUG(com->flags) ? 0 : IER_ETXRDY));
mtx_unlock_spin(&sio_lock);
siosettimeout();
/* XXX: should be generic ? */
if (com->prev_modem_status & MSR_DCD || ISCALLOUT(dev))
ttyld_modem(tp, 1);
return (0);
}
static void
comclose(tp)
struct tty *tp;
{
int s;
struct com_s *com;
s = spltty();
com = tp->t_sc;
com->poll = FALSE;
com->poll_output = FALSE;
sio_setreg(com, com_cfcr, com->cfcr_image &= ~CFCR_SBREAK);
#if defined(KDB)
/*
* Leave interrupts enabled and don't clear DTR if this is the
* console. This allows us to detect break-to-debugger events
* while the console device is closed.
*/
if (com->unit != comconsole)
#endif
{
sio_setreg(com, com_ier, 0);
if (tp->t_cflag & HUPCL
/*
* XXX we will miss any carrier drop between here and the
* next open. Perhaps we should watch DCD even when the
* port is closed; it is not sufficient to check it at
* the next open because it might go up and down while
* we're not watching.
*/
|| (!tp->t_actout
&& !(com->prev_modem_status & MSR_DCD)
&& !(tp->t_init_in.c_cflag & CLOCAL))
|| !(tp->t_state & TS_ISOPEN)) {
(void)commodem(tp, 0, SER_DTR);
ttydtrwaitstart(tp);
}
}
if (com->hasfifo) {
/*
* Disable fifos so that they are off after controlled
* reboots. Some BIOSes fail to detect 16550s when the
* fifos are enabled.
*/
sio_setreg(com, com_fifo, 0);
}
tp->t_actout = FALSE;
wakeup(&tp->t_actout);
wakeup(TSA_CARR_ON(tp)); /* restart any wopeners */
siosettimeout();
splx(s);
}
static void
siobusycheck(chan)
void *chan;
{
struct com_s *com;
int s;
com = (struct com_s *)chan;
/*
* Clear TS_BUSY if low-level output is complete.
* spl locking is sufficient because siointr1() does not set CS_BUSY.
* If siointr1() clears CS_BUSY after we look at it, then we'll get
* called again. Reading the line status port outside of siointr1()
* is safe because CS_BUSY is clear so there are no output interrupts
* to lose.
*/
s = spltty();
if (com->state & CS_BUSY)
com->extra_state &= ~CSE_BUSYCHECK; /* False alarm. */
else if ((inb(com->line_status_port) & (LSR_TSRE | LSR_TXRDY))
== (LSR_TSRE | LSR_TXRDY)) {
com->tp->t_state &= ~TS_BUSY;
ttwwakeup(com->tp);
com->extra_state &= ~CSE_BUSYCHECK;
} else
timeout(siobusycheck, com, hz / 100);
splx(s);
}
static u_int
siodivisor(rclk, speed)
u_long rclk;
speed_t speed;
{
long actual_speed;
u_int divisor;
int error;
if (speed == 0)
return (0);
#if UINT_MAX > (ULONG_MAX - 1) / 8
if (speed > (ULONG_MAX - 1) / 8)
return (0);
#endif
divisor = (rclk / (8UL * speed) + 1) / 2;
if (divisor == 0 || divisor >= 65536)
return (0);
actual_speed = rclk / (16UL * divisor);
/* 10 times error in percent: */
error = ((actual_speed - (long)speed) * 2000 / (long)speed + 1) / 2;
/* 3.0% maximum error tolerance: */
if (error < -30 || error > 30)
return (0);
return (divisor);
}
/*
* Call this function with the sio_lock mutex held. It will return with the
* lock still held.
*/
static void
sioinput(com)
struct com_s *com;
{
u_char *buf;
int incc;
u_char line_status;
int recv_data;
struct tty *tp;
buf = com->ibuf;
tp = com->tp;
if (!(tp->t_state & TS_ISOPEN) || !(tp->t_cflag & CREAD)) {
com_events -= (com->iptr - com->ibuf);
com->iptr = com->ibuf;
return;
}
if (tp->t_state & TS_CAN_BYPASS_L_RINT) {
/*
* Avoid the grotesquely inefficient lineswitch routine
* (ttyinput) in "raw" mode. It usually takes about 450
* instructions (that's without canonical processing or echo!).
* slinput is reasonably fast (usually 40 instructions plus
* call overhead).
*/
do {
/*
* This may look odd, but it is using save-and-enable
* semantics instead of the save-and-disable semantics
* that are used everywhere else.
*/
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
incc = com->iptr - buf;
if (tp->t_rawq.c_cc + incc > tp->t_ihiwat
&& (com->state & CS_RTS_IFLOW
|| tp->t_iflag & IXOFF)
&& !(tp->t_state & TS_TBLOCK))
ttyblock(tp);
com->delta_error_counts[CE_TTY_BUF_OVERFLOW]
+= b_to_q((char *)buf, incc, &tp->t_rawq);
buf += incc;
tk_nin += incc;
tk_rawcc += incc;
tp->t_rawcc += incc;
ttwakeup(tp);
if (tp->t_state & TS_TTSTOP
&& (tp->t_iflag & IXANY
|| tp->t_cc[VSTART] == tp->t_cc[VSTOP])) {
tp->t_state &= ~TS_TTSTOP;
tp->t_lflag &= ~FLUSHO;
comstart(tp);
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
} while (buf < com->iptr);
} else {
do {
/*
* This may look odd, but it is using save-and-enable
* semantics instead of the save-and-disable semantics
* that are used everywhere else.
*/
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
line_status = buf[com->ierroff];
recv_data = *buf++;
if (line_status
& (LSR_BI | LSR_FE | LSR_OE | LSR_PE)) {
if (line_status & LSR_BI)
recv_data |= TTY_BI;
if (line_status & LSR_FE)
recv_data |= TTY_FE;
if (line_status & LSR_OE)
recv_data |= TTY_OE;
if (line_status & LSR_PE)
recv_data |= TTY_PE;
}
ttyld_rint(tp, recv_data);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
} while (buf < com->iptr);
}
com_events -= (com->iptr - com->ibuf);
com->iptr = com->ibuf;
/*
* There is now room for another low-level buffer full of input,
* so enable RTS if it is now disabled and there is room in the
* high-level buffer.
*/
if ((com->state & CS_RTS_IFLOW) && !(com->mcr_image & MCR_RTS) &&
!(tp->t_state & TS_TBLOCK))
outb(com->modem_ctl_port, com->mcr_image |= MCR_RTS);
}
static int
siointr(arg)
void *arg;
{
struct com_s *com;
#ifndef COM_MULTIPORT
com = (struct com_s *)arg;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
siointr1(com);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
#else /* COM_MULTIPORT */
bool_t possibly_more_intrs;
int unit;
/*
* Loop until there is no activity on any port. This is necessary
* to get an interrupt edge more than to avoid another interrupt.
* If the IRQ signal is just an OR of the IRQ signals from several
* devices, then the edge from one may be lost because another is
* on.
*/
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
do {
possibly_more_intrs = FALSE;
for (unit = 0; unit < sio_numunits; ++unit) {
com = com_addr(unit);
/*
* XXX COM_LOCK();
* would it work here, or be counter-productive?
*/
if (com != NULL
&& !com->gone
&& (inb(com->int_id_port) & IIR_IMASK)
!= IIR_NOPEND) {
siointr1(com);
possibly_more_intrs = TRUE;
}
/* XXX COM_UNLOCK(); */
}
} while (possibly_more_intrs);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
#endif /* COM_MULTIPORT */
return(FILTER_HANDLED);
}
static struct timespec siots[8];
static int siotso;
static int volatile siotsunit = -1;
static int
sysctl_siots(SYSCTL_HANDLER_ARGS)
{
char buf[128];
long long delta;
size_t len;
int error, i, tso;
for (i = 1, tso = siotso; i < tso; i++) {
delta = (long long)(siots[i].tv_sec - siots[i - 1].tv_sec) *
1000000000 +
(siots[i].tv_nsec - siots[i - 1].tv_nsec);
len = sprintf(buf, "%lld\n", delta);
if (delta >= 110000)
len += sprintf(buf + len - 1, ": *** %ld.%09ld\n",
(long)siots[i].tv_sec, siots[i].tv_nsec) - 1;
if (i == tso - 1)
buf[len - 1] = '\0';
error = SYSCTL_OUT(req, buf, len);
if (error != 0)
return (error);
}
return (0);
}
SYSCTL_PROC(_machdep, OID_AUTO, siots, CTLTYPE_STRING | CTLFLAG_RD,
0, 0, sysctl_siots, "A", "sio timestamps");
static void
siointr1(com)
struct com_s *com;
{
u_char int_ctl;
u_char int_ctl_new;
u_char line_status;
u_char modem_status;
u_char *ioptr;
u_char recv_data;
#ifdef KDB
again:
#endif
if (COM_IIR_TXRDYBUG(com->flags)) {
int_ctl = inb(com->int_ctl_port);
int_ctl_new = int_ctl;
} else {
int_ctl = 0;
int_ctl_new = 0;
}
while (!com->gone) {
if (com->pps.ppsparam.mode & PPS_CAPTUREBOTH) {
modem_status = inb(com->modem_status_port);
if ((modem_status ^ com->last_modem_status) &
com->pps_bit) {
pps_capture(&com->pps);
pps_event(&com->pps,
2003-02-27 12:53:21 +00:00
(modem_status & com->pps_bit) ?
PPS_CAPTUREASSERT : PPS_CAPTURECLEAR);
}
}
line_status = inb(com->line_status_port);
2004-04-07 21:56:20 +00:00
/* input event? (check first to help avoid overruns) */
while (line_status & LSR_RCV_MASK) {
/* break/unnattached error bits or real input? */
if (!(line_status & LSR_RXRDY))
recv_data = 0;
else
recv_data = inb(com->data_port);
#ifdef KDB
if (com->unit == comconsole &&
kdb_alt_break(recv_data, &com->alt_brk_state) != 0)
goto again;
#endif /* KDB */
if (line_status & (LSR_BI | LSR_FE | LSR_PE)) {
/*
* Don't store BI if IGNBRK or FE/PE if IGNPAR.
* Otherwise, push the work to a higher level
* (to handle PARMRK) if we're bypassing.
* Otherwise, convert BI/FE and PE+INPCK to 0.
*
* This makes bypassing work right in the
* usual "raw" case (IGNBRK set, and IGNPAR
* and INPCK clear).
*
* Note: BI together with FE/PE means just BI.
*/
if (line_status & LSR_BI) {
#if defined(KDB)
if (com->unit == comconsole) {
kdb_break();
goto cont;
}
#endif
if (com->tp == NULL
|| com->tp->t_iflag & IGNBRK)
goto cont;
} else {
if (com->tp == NULL
|| com->tp->t_iflag & IGNPAR)
goto cont;
}
if (com->tp->t_state & TS_CAN_BYPASS_L_RINT
&& (line_status & (LSR_BI | LSR_FE)
|| com->tp->t_iflag & INPCK))
recv_data = 0;
}
++com->bytes_in;
if (com->tp != NULL &&
com->tp->t_hotchar != 0 && recv_data == com->tp->t_hotchar)
Change the preemption code for software interrupt thread schedules and mutex releases to not require flags for the cases when preemption is not allowed: The purpose of the MTX_NOSWITCH and SWI_NOSWITCH flags is to prevent switching to a higher priority thread on mutex releease and swi schedule, respectively when that switch is not safe. Now that the critical section API maintains a per-thread nesting count, the kernel can easily check whether or not it should switch without relying on flags from the programmer. This fixes a few bugs in that all current callers of swi_sched() used SWI_NOSWITCH, when in fact, only the ones called from fast interrupt handlers and the swi_sched of softclock needed this flag. Note that to ensure that swi_sched()'s in clock and fast interrupt handlers do not switch, these handlers have to be explicitly wrapped in critical_enter/exit pairs. Presently, just wrapping the handlers is sufficient, but in the future with the fully preemptive kernel, the interrupt must be EOI'd before critical_exit() is called. (critical_exit() can switch due to a deferred preemption in a fully preemptive kernel.) I've tested the changes to the interrupt code on i386 and alpha. I have not tested ia64, but the interrupt code is almost identical to the alpha code, so I expect it will work fine. PowerPC and ARM do not yet have interrupt code in the tree so they shouldn't be broken. Sparc64 is broken, but that's been ok'd by jake and tmm who will be fixing the interrupt code for sparc64 shortly. Reviewed by: peter Tested on: i386, alpha
2002-01-05 08:47:13 +00:00
swi_sched(sio_fast_ih, 0);
ioptr = com->iptr;
if (ioptr >= com->ibufend)
CE_RECORD(com, CE_INTERRUPT_BUF_OVERFLOW);
else {
if (com->tp != NULL && com->tp->t_do_timestamp)
microtime(&com->tp->t_timestamp);
++com_events;
swi_sched(sio_slow_ih, SWI_DELAY);
#if 0 /* for testing input latency vs efficiency */
if (com->iptr - com->ibuf == 8)
Change the preemption code for software interrupt thread schedules and mutex releases to not require flags for the cases when preemption is not allowed: The purpose of the MTX_NOSWITCH and SWI_NOSWITCH flags is to prevent switching to a higher priority thread on mutex releease and swi schedule, respectively when that switch is not safe. Now that the critical section API maintains a per-thread nesting count, the kernel can easily check whether or not it should switch without relying on flags from the programmer. This fixes a few bugs in that all current callers of swi_sched() used SWI_NOSWITCH, when in fact, only the ones called from fast interrupt handlers and the swi_sched of softclock needed this flag. Note that to ensure that swi_sched()'s in clock and fast interrupt handlers do not switch, these handlers have to be explicitly wrapped in critical_enter/exit pairs. Presently, just wrapping the handlers is sufficient, but in the future with the fully preemptive kernel, the interrupt must be EOI'd before critical_exit() is called. (critical_exit() can switch due to a deferred preemption in a fully preemptive kernel.) I've tested the changes to the interrupt code on i386 and alpha. I have not tested ia64, but the interrupt code is almost identical to the alpha code, so I expect it will work fine. PowerPC and ARM do not yet have interrupt code in the tree so they shouldn't be broken. Sparc64 is broken, but that's been ok'd by jake and tmm who will be fixing the interrupt code for sparc64 shortly. Reviewed by: peter Tested on: i386, alpha
2002-01-05 08:47:13 +00:00
swi_sched(sio_fast_ih, 0);
#endif
ioptr[0] = recv_data;
ioptr[com->ierroff] = line_status;
com->iptr = ++ioptr;
if (ioptr == com->ihighwater
&& com->state & CS_RTS_IFLOW)
outb(com->modem_ctl_port,
com->mcr_image &= ~MCR_RTS);
if (line_status & LSR_OE)
CE_RECORD(com, CE_OVERRUN);
}
cont:
if (line_status & LSR_TXRDY
&& com->state >= (CS_BUSY | CS_TTGO | CS_ODEVREADY))
goto txrdy;
/*
* "& 0x7F" is to avoid the gcc-1.40 generating a slow
* jump from the top of the loop to here
*/
line_status = inb(com->line_status_port) & 0x7F;
}
/* modem status change? (always check before doing output) */
modem_status = inb(com->modem_status_port);
if (modem_status != com->last_modem_status) {
/*
* Schedule high level to handle DCD changes. Note
* that we don't use the delta bits anywhere. Some
* UARTs mess them up, and it's easy to remember the
* previous bits and calculate the delta.
*/
com->last_modem_status = modem_status;
if (!(com->state & CS_CHECKMSR)) {
com_events += LOTS_OF_EVENTS;
com->state |= CS_CHECKMSR;
Change the preemption code for software interrupt thread schedules and mutex releases to not require flags for the cases when preemption is not allowed: The purpose of the MTX_NOSWITCH and SWI_NOSWITCH flags is to prevent switching to a higher priority thread on mutex releease and swi schedule, respectively when that switch is not safe. Now that the critical section API maintains a per-thread nesting count, the kernel can easily check whether or not it should switch without relying on flags from the programmer. This fixes a few bugs in that all current callers of swi_sched() used SWI_NOSWITCH, when in fact, only the ones called from fast interrupt handlers and the swi_sched of softclock needed this flag. Note that to ensure that swi_sched()'s in clock and fast interrupt handlers do not switch, these handlers have to be explicitly wrapped in critical_enter/exit pairs. Presently, just wrapping the handlers is sufficient, but in the future with the fully preemptive kernel, the interrupt must be EOI'd before critical_exit() is called. (critical_exit() can switch due to a deferred preemption in a fully preemptive kernel.) I've tested the changes to the interrupt code on i386 and alpha. I have not tested ia64, but the interrupt code is almost identical to the alpha code, so I expect it will work fine. PowerPC and ARM do not yet have interrupt code in the tree so they shouldn't be broken. Sparc64 is broken, but that's been ok'd by jake and tmm who will be fixing the interrupt code for sparc64 shortly. Reviewed by: peter Tested on: i386, alpha
2002-01-05 08:47:13 +00:00
swi_sched(sio_fast_ih, 0);
}
/* handle CTS change immediately for crisp flow ctl */
if (com->state & CS_CTS_OFLOW) {
if (modem_status & MSR_CTS)
com->state |= CS_ODEVREADY;
else
com->state &= ~CS_ODEVREADY;
}
}
txrdy:
/* output queued and everything ready? */
if (line_status & LSR_TXRDY
&& com->state >= (CS_BUSY | CS_TTGO | CS_ODEVREADY)) {
ioptr = com->obufq.l_head;
if (com->tx_fifo_size > 1 && com->unit != siotsunit) {
u_int ocount;
ocount = com->obufq.l_tail - ioptr;
if (ocount > com->tx_fifo_size)
ocount = com->tx_fifo_size;
com->bytes_out += ocount;
do
outb(com->data_port, *ioptr++);
while (--ocount != 0);
} else {
outb(com->data_port, *ioptr++);
++com->bytes_out;
if (com->unit == siotsunit
&& siotso < sizeof siots / sizeof siots[0])
nanouptime(&siots[siotso++]);
}
com->obufq.l_head = ioptr;
if (COM_IIR_TXRDYBUG(com->flags))
int_ctl_new = int_ctl | IER_ETXRDY;
if (ioptr >= com->obufq.l_tail) {
struct lbq *qp;
qp = com->obufq.l_next;
qp->l_queued = FALSE;
qp = qp->l_next;
if (qp != NULL) {
com->obufq.l_head = qp->l_head;
com->obufq.l_tail = qp->l_tail;
com->obufq.l_next = qp;
} else {
/* output just completed */
if (COM_IIR_TXRDYBUG(com->flags))
int_ctl_new = int_ctl
& ~IER_ETXRDY;
com->state &= ~CS_BUSY;
}
if (!(com->state & CS_ODONE)) {
com_events += LOTS_OF_EVENTS;
com->state |= CS_ODONE;
/* handle at high level ASAP */
Change the preemption code for software interrupt thread schedules and mutex releases to not require flags for the cases when preemption is not allowed: The purpose of the MTX_NOSWITCH and SWI_NOSWITCH flags is to prevent switching to a higher priority thread on mutex releease and swi schedule, respectively when that switch is not safe. Now that the critical section API maintains a per-thread nesting count, the kernel can easily check whether or not it should switch without relying on flags from the programmer. This fixes a few bugs in that all current callers of swi_sched() used SWI_NOSWITCH, when in fact, only the ones called from fast interrupt handlers and the swi_sched of softclock needed this flag. Note that to ensure that swi_sched()'s in clock and fast interrupt handlers do not switch, these handlers have to be explicitly wrapped in critical_enter/exit pairs. Presently, just wrapping the handlers is sufficient, but in the future with the fully preemptive kernel, the interrupt must be EOI'd before critical_exit() is called. (critical_exit() can switch due to a deferred preemption in a fully preemptive kernel.) I've tested the changes to the interrupt code on i386 and alpha. I have not tested ia64, but the interrupt code is almost identical to the alpha code, so I expect it will work fine. PowerPC and ARM do not yet have interrupt code in the tree so they shouldn't be broken. Sparc64 is broken, but that's been ok'd by jake and tmm who will be fixing the interrupt code for sparc64 shortly. Reviewed by: peter Tested on: i386, alpha
2002-01-05 08:47:13 +00:00
swi_sched(sio_fast_ih, 0);
}
}
if (COM_IIR_TXRDYBUG(com->flags)
&& int_ctl != int_ctl_new)
outb(com->int_ctl_port, int_ctl_new);
}
/* finished? */
#ifndef COM_MULTIPORT
if ((inb(com->int_id_port) & IIR_IMASK) == IIR_NOPEND)
#endif /* COM_MULTIPORT */
return;
}
}
/* software interrupt handler for SWI_TTY */
static void
siopoll(void *dummy)
{
int unit;
if (com_events == 0)
return;
repeat:
for (unit = 0; unit < sio_numunits; ++unit) {
struct com_s *com;
int incc;
struct tty *tp;
com = com_addr(unit);
if (com == NULL)
continue;
tp = com->tp;
if (tp == NULL || com->gone) {
/*
* Discard any events related to never-opened or
* going-away devices.
*/
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
incc = com->iptr - com->ibuf;
com->iptr = com->ibuf;
if (com->state & CS_CHECKMSR) {
incc += LOTS_OF_EVENTS;
com->state &= ~CS_CHECKMSR;
}
com_events -= incc;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
continue;
}
if (com->iptr != com->ibuf) {
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
sioinput(com);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
}
if (com->state & CS_CHECKMSR) {
u_char delta_modem_status;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
delta_modem_status = com->last_modem_status
^ com->prev_modem_status;
com->prev_modem_status = com->last_modem_status;
com_events -= LOTS_OF_EVENTS;
com->state &= ~CS_CHECKMSR;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
if (delta_modem_status & MSR_DCD)
ttyld_modem(tp,
com->prev_modem_status & MSR_DCD);
}
if (com->state & CS_ODONE) {
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
com_events -= LOTS_OF_EVENTS;
com->state &= ~CS_ODONE;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
if (!(com->state & CS_BUSY)
&& !(com->extra_state & CSE_BUSYCHECK)) {
timeout(siobusycheck, com, hz / 100);
com->extra_state |= CSE_BUSYCHECK;
}
ttyld_start(tp);
}
if (com_events == 0)
break;
}
if (com_events >= LOTS_OF_EVENTS)
goto repeat;
}
static void
combreak(tp, sig)
struct tty *tp;
int sig;
{
struct com_s *com;
com = tp->t_sc;
if (sig)
sio_setreg(com, com_cfcr, com->cfcr_image |= CFCR_SBREAK);
else
sio_setreg(com, com_cfcr, com->cfcr_image &= ~CFCR_SBREAK);
}
static int
comparam(tp, t)
struct tty *tp;
struct termios *t;
{
u_int cfcr;
int cflag;
struct com_s *com;
u_int divisor;
u_char dlbh;
u_char dlbl;
u_char efr_flowbits;
int s;
com = tp->t_sc;
if (com == NULL)
return (ENODEV);
/* check requested parameters */
if (t->c_ispeed != (t->c_ospeed != 0 ? t->c_ospeed : tp->t_ospeed))
return (EINVAL);
divisor = siodivisor(com->rclk, t->c_ispeed);
if (divisor == 0)
return (EINVAL);
/* parameters are OK, convert them to the com struct and the device */
s = spltty();
if (t->c_ospeed == 0)
(void)commodem(tp, 0, SER_DTR); /* hang up line */
else
(void)commodem(tp, SER_DTR, 0);
cflag = t->c_cflag;
switch (cflag & CSIZE) {
case CS5:
cfcr = CFCR_5BITS;
break;
case CS6:
cfcr = CFCR_6BITS;
break;
case CS7:
cfcr = CFCR_7BITS;
break;
default:
cfcr = CFCR_8BITS;
break;
}
if (cflag & PARENB) {
cfcr |= CFCR_PENAB;
if (!(cflag & PARODD))
cfcr |= CFCR_PEVEN;
}
if (cflag & CSTOPB)
cfcr |= CFCR_STOPB;
if (com->hasfifo) {
/*
* Use a fifo trigger level low enough so that the input
* latency from the fifo is less than about 16 msec and
* the total latency is less than about 30 msec. These
* latencies are reasonable for humans. Serial comms
* protocols shouldn't expect anything better since modem
* latencies are larger.
*
* The fifo trigger level cannot be set at RX_HIGH for high
* speed connections without further work on reducing
* interrupt disablement times in other parts of the system,
* without producing silo overflow errors.
*/
com->fifo_image = com->unit == siotsunit ? 0
: t->c_ispeed <= 4800
2001-12-24 01:24:38 +00:00
? FIFO_ENABLE : FIFO_ENABLE | FIFO_RX_MEDH;
#ifdef COM_ESP
/*
* The Hayes ESP card needs the fifo DMA mode bit set
* in compatibility mode. If not, it will interrupt
* for each character received.
*/
if (com->esp)
com->fifo_image |= FIFO_DMA_MODE;
#endif
sio_setreg(com, com_fifo, com->fifo_image);
}
/*
* This returns with interrupts disabled so that we can complete
* the speed change atomically. Keeping interrupts disabled is
* especially important while com_data is hidden.
*/
(void) siosetwater(com, t->c_ispeed);
sio_setreg(com, com_cfcr, cfcr | CFCR_DLAB);
/*
* Only set the divisor registers if they would change, since on
* some 16550 incompatibles (UMC8669F), setting them while input
* is arriving loses sync until data stops arriving.
*/
dlbl = divisor & 0xFF;
if (sio_getreg(com, com_dlbl) != dlbl)
sio_setreg(com, com_dlbl, dlbl);
dlbh = divisor >> 8;
if (sio_getreg(com, com_dlbh) != dlbh)
sio_setreg(com, com_dlbh, dlbh);
efr_flowbits = 0;
if (cflag & CRTS_IFLOW) {
com->state |= CS_RTS_IFLOW;
efr_flowbits |= EFR_AUTORTS;
/*
* If CS_RTS_IFLOW just changed from off to on, the change
* needs to be propagated to MCR_RTS. This isn't urgent,
* so do it later by calling comstart() instead of repeating
* a lot of code from comstart() here.
*/
} else if (com->state & CS_RTS_IFLOW) {
com->state &= ~CS_RTS_IFLOW;
/*
* CS_RTS_IFLOW just changed from on to off. Force MCR_RTS
* on here, since comstart() won't do it later.
*/
outb(com->modem_ctl_port, com->mcr_image |= MCR_RTS);
}
/*
* Set up state to handle output flow control.
* XXX - worth handling MDMBUF (DCD) flow control at the lowest level?
* Now has 10+ msec latency, while CTS flow has 50- usec latency.
*/
com->state |= CS_ODEVREADY;
com->state &= ~CS_CTS_OFLOW;
if (cflag & CCTS_OFLOW) {
com->state |= CS_CTS_OFLOW;
efr_flowbits |= EFR_AUTOCTS;
if (!(com->last_modem_status & MSR_CTS))
com->state &= ~CS_ODEVREADY;
}
if (com->st16650a) {
sio_setreg(com, com_lcr, LCR_EFR_ENABLE);
sio_setreg(com, com_efr,
(sio_getreg(com, com_efr)
& ~(EFR_AUTOCTS | EFR_AUTORTS)) | efr_flowbits);
}
sio_setreg(com, com_cfcr, com->cfcr_image = cfcr);
/* XXX shouldn't call functions while intrs are disabled. */
ttyldoptim(tp);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
splx(s);
comstart(tp);
if (com->ibufold != NULL) {
free(com->ibufold, M_DEVBUF);
com->ibufold = NULL;
}
return (0);
}
/*
* This function must be called with the sio_lock mutex released and will
* return with it obtained.
*/
static int
siosetwater(com, speed)
struct com_s *com;
speed_t speed;
{
int cp4ticks;
u_char *ibuf;
int ibufsize;
struct tty *tp;
/*
* Make the buffer size large enough to handle a softtty interrupt
* latency of about 2 ticks without loss of throughput or data
* (about 3 ticks if input flow control is not used or not honoured,
* but a bit less for CS5-CS7 modes).
*/
cp4ticks = speed / 10 / hz * 4;
for (ibufsize = 128; ibufsize < cp4ticks;)
ibufsize <<= 1;
if (ibufsize == com->ibufsize) {
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
return (0);
}
/*
* Allocate input buffer. The extra factor of 2 in the size is
* to allow for an error byte for each input byte.
*/
ibuf = malloc(2 * ibufsize, M_DEVBUF, M_NOWAIT);
if (ibuf == NULL) {
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
return (ENOMEM);
}
/* Initialize non-critical variables. */
com->ibufold = com->ibuf;
com->ibufsize = ibufsize;
tp = com->tp;
if (tp != NULL) {
tp->t_ififosize = 2 * ibufsize;
tp->t_ispeedwat = (speed_t)-1;
tp->t_ospeedwat = (speed_t)-1;
}
/*
* Read current input buffer, if any. Continue with interrupts
* disabled.
*/
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
if (com->iptr != com->ibuf)
sioinput(com);
/*-
* Initialize critical variables, including input buffer watermarks.
* The external device is asked to stop sending when the buffer
* exactly reaches high water, or when the high level requests it.
* The high level is notified immediately (rather than at a later
* clock tick) when this watermark is reached.
* The buffer size is chosen so the watermark should almost never
* be reached.
* The low watermark is invisibly 0 since the buffer is always
* emptied all at once.
*/
com->iptr = com->ibuf = ibuf;
com->ibufend = ibuf + ibufsize;
com->ierroff = ibufsize;
com->ihighwater = ibuf + 3 * ibufsize / 4;
return (0);
}
static void
comstart(tp)
struct tty *tp;
{
struct com_s *com;
int s;
com = tp->t_sc;
if (com == NULL)
return;
s = spltty();
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
if (tp->t_state & TS_TTSTOP)
com->state &= ~CS_TTGO;
else
com->state |= CS_TTGO;
if (tp->t_state & TS_TBLOCK) {
if (com->mcr_image & MCR_RTS && com->state & CS_RTS_IFLOW)
outb(com->modem_ctl_port, com->mcr_image &= ~MCR_RTS);
} else {
if (!(com->mcr_image & MCR_RTS) && com->iptr < com->ihighwater
&& com->state & CS_RTS_IFLOW)
outb(com->modem_ctl_port, com->mcr_image |= MCR_RTS);
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
if (tp->t_state & (TS_TIMEOUT | TS_TTSTOP)) {
ttwwakeup(tp);
splx(s);
return;
}
if (tp->t_outq.c_cc != 0) {
struct lbq *qp;
struct lbq *next;
if (!com->obufs[0].l_queued) {
com->obufs[0].l_tail
= com->obuf1 + q_to_b(&tp->t_outq, com->obuf1,
sizeof com->obuf1);
com->obufs[0].l_next = NULL;
com->obufs[0].l_queued = TRUE;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
if (com->state & CS_BUSY) {
qp = com->obufq.l_next;
while ((next = qp->l_next) != NULL)
qp = next;
qp->l_next = &com->obufs[0];
} else {
com->obufq.l_head = com->obufs[0].l_head;
com->obufq.l_tail = com->obufs[0].l_tail;
com->obufq.l_next = &com->obufs[0];
com->state |= CS_BUSY;
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
}
if (tp->t_outq.c_cc != 0 && !com->obufs[1].l_queued) {
com->obufs[1].l_tail
= com->obuf2 + q_to_b(&tp->t_outq, com->obuf2,
sizeof com->obuf2);
com->obufs[1].l_next = NULL;
com->obufs[1].l_queued = TRUE;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
if (com->state & CS_BUSY) {
qp = com->obufq.l_next;
while ((next = qp->l_next) != NULL)
qp = next;
qp->l_next = &com->obufs[1];
} else {
com->obufq.l_head = com->obufs[1].l_head;
com->obufq.l_tail = com->obufs[1].l_tail;
com->obufq.l_next = &com->obufs[1];
com->state |= CS_BUSY;
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
}
tp->t_state |= TS_BUSY;
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
if (com->state >= (CS_BUSY | CS_TTGO))
siointr1(com); /* fake interrupt to start output */
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
ttwwakeup(tp);
splx(s);
}
static void
comstop(tp, rw)
struct tty *tp;
int rw;
{
struct com_s *com;
com = tp->t_sc;
if (com == NULL || com->gone)
return;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
if (rw & FWRITE) {
if (com->hasfifo)
#ifdef COM_ESP
/* XXX avoid h/w bug. */
if (!com->esp)
#endif
sio_setreg(com, com_fifo,
FIFO_XMT_RST | com->fifo_image);
com->obufs[0].l_queued = FALSE;
com->obufs[1].l_queued = FALSE;
if (com->state & CS_ODONE)
com_events -= LOTS_OF_EVENTS;
com->state &= ~(CS_ODONE | CS_BUSY);
com->tp->t_state &= ~TS_BUSY;
}
if (rw & FREAD) {
if (com->hasfifo)
#ifdef COM_ESP
/* XXX avoid h/w bug. */
if (!com->esp)
#endif
sio_setreg(com, com_fifo,
FIFO_RCV_RST | com->fifo_image);
com_events -= (com->iptr - com->ibuf);
com->iptr = com->ibuf;
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
comstart(tp);
}
static int
commodem(struct tty *tp, int sigon, int sigoff)
{
struct com_s *com;
int bitand, bitor, msr;
com = tp->t_sc;
if (com->gone)
return(0);
if (sigon != 0 || sigoff != 0) {
bitand = bitor = 0;
if (sigoff & SER_DTR)
bitand |= MCR_DTR;
if (sigoff & SER_RTS)
bitand |= MCR_RTS;
if (sigon & SER_DTR)
bitor |= MCR_DTR;
if (sigon & SER_RTS)
bitor |= MCR_RTS;
bitand = ~bitand;
mtx_lock_spin(&sio_lock);
com->mcr_image &= bitand;
com->mcr_image |= bitor;
outb(com->modem_ctl_port, com->mcr_image);
mtx_unlock_spin(&sio_lock);
return (0);
} else {
bitor = 0;
if (com->mcr_image & MCR_DTR)
bitor |= SER_DTR;
if (com->mcr_image & MCR_RTS)
bitor |= SER_RTS;
msr = com->prev_modem_status;
if (msr & MSR_CTS)
bitor |= SER_CTS;
if (msr & MSR_DCD)
bitor |= SER_DCD;
if (msr & MSR_DSR)
bitor |= SER_DSR;
if (msr & MSR_DSR)
bitor |= SER_DSR;
if (msr & (MSR_RI | MSR_TERI))
bitor |= SER_RI;
return (bitor);
}
}
static void
siosettimeout()
{
struct com_s *com;
bool_t someopen;
int unit;
/*
* Set our timeout period to 1 second if no polled devices are open.
* Otherwise set it to max(1/200, 1/hz).
* Enable timeouts iff some device is open.
*/
untimeout(comwakeup, (void *)NULL, sio_timeout_handle);
sio_timeout = hz;
someopen = FALSE;
for (unit = 0; unit < sio_numunits; ++unit) {
com = com_addr(unit);
if (com != NULL && com->tp != NULL
&& com->tp->t_state & TS_ISOPEN && !com->gone) {
someopen = TRUE;
if (com->poll || com->poll_output) {
sio_timeout = hz > 200 ? hz / 200 : 1;
break;
}
}
}
if (someopen) {
sio_timeouts_until_log = hz / sio_timeout;
sio_timeout_handle = timeout(comwakeup, (void *)NULL,
sio_timeout);
} else {
/* Flush error messages, if any. */
sio_timeouts_until_log = 1;
comwakeup((void *)NULL);
untimeout(comwakeup, (void *)NULL, sio_timeout_handle);
}
}
static void
comwakeup(chan)
void *chan;
{
struct com_s *com;
int unit;
sio_timeout_handle = timeout(comwakeup, (void *)NULL, sio_timeout);
/*
* Recover from lost output interrupts.
* Poll any lines that don't use interrupts.
*/
for (unit = 0; unit < sio_numunits; ++unit) {
com = com_addr(unit);
if (com != NULL && !com->gone
&& (com->state >= (CS_BUSY | CS_TTGO) || com->poll)) {
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
siointr1(com);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
}
}
/*
* Check for and log errors, but not too often.
*/
if (--sio_timeouts_until_log > 0)
return;
sio_timeouts_until_log = hz / sio_timeout;
for (unit = 0; unit < sio_numunits; ++unit) {
int errnum;
com = com_addr(unit);
if (com == NULL)
continue;
if (com->gone)
continue;
for (errnum = 0; errnum < CE_NTYPES; ++errnum) {
u_int delta;
u_long total;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock_spin(&sio_lock);
delta = com->delta_error_counts[errnum];
com->delta_error_counts[errnum] = 0;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock_spin(&sio_lock);
if (delta == 0)
continue;
total = com->error_counts[errnum] += delta;
log(LOG_ERR, "sio%d: %u more %s%s (total %lu)\n",
unit, delta, error_desc[errnum],
delta == 1 ? "" : "s", total);
}
}
}
/*
* Following are all routines needed for SIO to act as console
*/
struct siocnstate {
u_char dlbl;
u_char dlbh;
u_char ier;
u_char cfcr;
u_char mcr;
};
/*
* This is a function in order to not replicate "ttyd%d" more
* places than absolutely necessary.
*/
static void
siocnset(struct consdev *cd, int unit)
{
cd->cn_unit = unit;
sprintf(cd->cn_name, "ttyd%d", unit);
}
2002-03-20 02:08:01 +00:00
static speed_t siocngetspeed(Port_t, u_long rclk);
static void siocnclose(struct siocnstate *sp, Port_t iobase);
static void siocnopen(struct siocnstate *sp, Port_t iobase, int speed);
static void siocntxwait(Port_t iobase);
static cn_probe_t sio_cnprobe;
static cn_init_t sio_cninit;
static cn_term_t sio_cnterm;
static cn_getc_t sio_cngetc;
static cn_putc_t sio_cnputc;
static cn_grab_t sio_cngrab;
static cn_ungrab_t sio_cnungrab;
CONSOLE_DRIVER(sio);
static void
siocntxwait(iobase)
Port_t iobase;
{
int timo;
/*
* Wait for any pending transmission to finish. Required to avoid
* the UART lockup bug when the speed is changed, and for normal
* transmits.
*/
timo = 100000;
while ((inb(iobase + com_lsr) & (LSR_TSRE | LSR_TXRDY))
!= (LSR_TSRE | LSR_TXRDY) && --timo != 0)
;
}
/*
* Read the serial port specified and try to figure out what speed
* it's currently running at. We're assuming the serial port has
* been initialized and is basicly idle. This routine is only intended
* to be run at system startup.
*
* If the value read from the serial port doesn't make sense, return 0.
*/
static speed_t
siocngetspeed(iobase, rclk)
Port_t iobase;
u_long rclk;
{
u_int divisor;
u_char dlbh;
u_char dlbl;
u_char cfcr;
cfcr = inb(iobase + com_cfcr);
outb(iobase + com_cfcr, CFCR_DLAB | cfcr);
dlbl = inb(iobase + com_dlbl);
dlbh = inb(iobase + com_dlbh);
outb(iobase + com_cfcr, cfcr);
divisor = dlbh << 8 | dlbl;
/* XXX there should be more sanity checking. */
if (divisor == 0)
return (CONSPEED);
return (rclk / (16UL * divisor));
}
static void
siocnopen(sp, iobase, speed)
struct siocnstate *sp;
Port_t iobase;
int speed;
{
u_int divisor;
u_char dlbh;
u_char dlbl;
/*
* Save all the device control registers except the fifo register
* and set our default ones (cs8 -parenb speed=comdefaultrate).
* We can't save the fifo register since it is read-only.
*/
sp->ier = inb(iobase + com_ier);
outb(iobase + com_ier, 0); /* spltty() doesn't stop siointr() */
siocntxwait(iobase);
sp->cfcr = inb(iobase + com_cfcr);
outb(iobase + com_cfcr, CFCR_DLAB | CFCR_8BITS);
sp->dlbl = inb(iobase + com_dlbl);
sp->dlbh = inb(iobase + com_dlbh);
/*
* Only set the divisor registers if they would change, since on
* some 16550 incompatibles (Startech), setting them clears the
* data input register. This also reduces the effects of the
* UMC8669F bug.
*/
divisor = siodivisor(comdefaultrclk, speed);
dlbl = divisor & 0xFF;
if (sp->dlbl != dlbl)
outb(iobase + com_dlbl, dlbl);
dlbh = divisor >> 8;
if (sp->dlbh != dlbh)
outb(iobase + com_dlbh, dlbh);
outb(iobase + com_cfcr, CFCR_8BITS);
sp->mcr = inb(iobase + com_mcr);
/*
* We don't want interrupts, but must be careful not to "disable"
* them by clearing the MCR_IENABLE bit, since that might cause
* an interrupt by floating the IRQ line.
*/
outb(iobase + com_mcr, (sp->mcr & MCR_IENABLE) | MCR_DTR | MCR_RTS);
}
static void
siocnclose(sp, iobase)
struct siocnstate *sp;
Port_t iobase;
{
/*
* Restore the device control registers.
*/
siocntxwait(iobase);
outb(iobase + com_cfcr, CFCR_DLAB | CFCR_8BITS);
if (sp->dlbl != inb(iobase + com_dlbl))
outb(iobase + com_dlbl, sp->dlbl);
if (sp->dlbh != inb(iobase + com_dlbh))
outb(iobase + com_dlbh, sp->dlbh);
outb(iobase + com_cfcr, sp->cfcr);
/*
* XXX damp oscillations of MCR_DTR and MCR_RTS by not restoring them.
*/
outb(iobase + com_mcr, sp->mcr | MCR_DTR | MCR_RTS);
outb(iobase + com_ier, sp->ier);
}
static void
sio_cnprobe(cp)
struct consdev *cp;
{
speed_t boot_speed;
u_char cfcr;
u_int divisor;
int s, unit;
struct siocnstate sp;
/*
* Find our first enabled console, if any. If it is a high-level
* console device, then initialize it and return successfully.
* If it is a low-level console device, then initialize it and
* return unsuccessfully. It must be initialized in both cases
* for early use by console drivers and debuggers. Initializing
* the hardware is not necessary in all cases, since the i/o
* routines initialize it on the fly, but it is necessary if
* input might arrive while the hardware is switched back to an
* uninitialized state. We can't handle multiple console devices
* yet because our low-level routines don't take a device arg.
* We trust the user to set the console flags properly so that we
* don't need to probe.
*/
cp->cn_pri = CN_DEAD;
for (unit = 0; unit < 16; unit++) { /* XXX need to know how many */
int flags;
if (resource_disabled("sio", unit))
continue;
if (resource_int_value("sio", unit, "flags", &flags))
continue;
if (COM_CONSOLE(flags) || COM_DEBUGGER(flags)) {
int port;
Port_t iobase;
if (resource_int_value("sio", unit, "port", &port))
continue;
iobase = port;
s = spltty();
if ((boothowto & RB_SERIAL) && COM_CONSOLE(flags)) {
boot_speed =
siocngetspeed(iobase, comdefaultrclk);
if (boot_speed)
comdefaultrate = boot_speed;
}
/*
* Initialize the divisor latch. We can't rely on
* siocnopen() to do this the first time, since it
* avoids writing to the latch if the latch appears
* to have the correct value. Also, if we didn't
* just read the speed from the hardware, then we
* need to set the speed in hardware so that
* switching it later is null.
*/
cfcr = inb(iobase + com_cfcr);
outb(iobase + com_cfcr, CFCR_DLAB | cfcr);
divisor = siodivisor(comdefaultrclk, comdefaultrate);
outb(iobase + com_dlbl, divisor & 0xff);
outb(iobase + com_dlbh, divisor >> 8);
outb(iobase + com_cfcr, cfcr);
siocnopen(&sp, iobase, comdefaultrate);
splx(s);
if (COM_CONSOLE(flags) && !COM_LLCONSOLE(flags)) {
siocnset(cp, unit);
cp->cn_pri = COM_FORCECONSOLE(flags)
|| boothowto & RB_SERIAL
? CN_REMOTE : CN_NORMAL;
siocniobase = iobase;
siocnunit = unit;
}
#ifdef GDB
if (COM_DEBUGGER(flags))
siogdbiobase = iobase;
#endif
}
}
}
static void
sio_cninit(cp)
struct consdev *cp;
{
comconsole = cp->cn_unit;
}
static void
sio_cnterm(cp)
struct consdev *cp;
{
comconsole = -1;
}
static void
sio_cngrab(struct consdev *cp)
{
}
static void
sio_cnungrab(struct consdev *cp)
{
}
static int
sio_cngetc(struct consdev *cd)
{
int c;
Port_t iobase;
int s;
struct siocnstate sp;
speed_t speed;
if (cd != NULL && cd->cn_unit == siocnunit) {
iobase = siocniobase;
speed = comdefaultrate;
} else {
#ifdef GDB
iobase = siogdbiobase;
speed = gdbdefaultrate;
#else
return (-1);
#endif
}
s = spltty();
siocnopen(&sp, iobase, speed);
if (inb(iobase + com_lsr) & LSR_RXRDY)
c = inb(iobase + com_data);
else
c = -1;
siocnclose(&sp, iobase);
splx(s);
return (c);
}
static void
sio_cnputc(struct consdev *cd, int c)
{
int need_unlock;
int s;
struct siocnstate sp;
Port_t iobase;
speed_t speed;
if (cd != NULL && cd->cn_unit == siocnunit) {
iobase = siocniobase;
speed = comdefaultrate;
} else {
#ifdef GDB
iobase = siogdbiobase;
speed = gdbdefaultrate;
#else
return;
#endif
}
s = spltty();
need_unlock = 0;
if (!kdb_active && sio_inited == 2 && !mtx_owned(&sio_lock)) {
mtx_lock_spin(&sio_lock);
need_unlock = 1;
}
siocnopen(&sp, iobase, speed);
siocntxwait(iobase);
outb(iobase + com_data, c);
siocnclose(&sp, iobase);
if (need_unlock)
mtx_unlock_spin(&sio_lock);
splx(s);
}
/*
* Remote gdb(1) support.
*/
#if defined(GDB)
#include <gdb/gdb.h>
static gdb_probe_f siogdbprobe;
static gdb_init_f siogdbinit;
static gdb_term_f siogdbterm;
static gdb_getc_f siogdbgetc;
static gdb_putc_f siogdbputc;
GDB_DBGPORT(sio, siogdbprobe, siogdbinit, siogdbterm, siogdbgetc, siogdbputc);
static int
siogdbprobe(void)
{
return ((siogdbiobase != 0) ? 0 : -1);
}
static void
siogdbinit(void)
{
}
static void
siogdbterm(void)
{
}
static void
siogdbputc(int c)
{
sio_cnputc(NULL, c);
}
static int
siogdbgetc(void)
{
return (sio_cngetc(NULL));
}
#endif