Remove promisc_user, ray_reset and ray_reset_timo as they are not used. Incidental remove of a timer too. Remove the runq_abort code.

Get ray_detach working correctly. This is a very simple routine as it
just wakes up sleeping processes. Note that anything woken has NO softc
structure available! runq_add is suitably modified to detect a detach and
return straight away.

Due to ray_detach and its implications use a macro for adding things
to the runq in user land.
This commit is contained in:
Duncan Barclay 2000-06-10 13:50:57 +00:00
parent 877bed4000
commit 8204ff6102
1 changed files with 121 additions and 207 deletions

View File

@ -241,26 +241,36 @@
* stop sequence is done
* others are done
* mcast code resurrection - done
* remove ray_reset - done
* detach needs to drain comq - done
* in fact we don't drain the comq just get the hell out asap
* remember to ccs_free on error in _user routines - done
* not relevant anymore
* macro for gone and check is at head of all externally called routines - done
* not relevant anymore
* probably function/macro to test unload at top of commands - done
* detach checks in all routines that access the card - done
* not relevant anymore as they won't be called by runq
* reset in ray_init_user? - done
* no as I don't want to remove it (people can always cycle power
* from the command line)
*
* ***detach needs to drain comq
* ***detach checks in all routines that access the card
* ***reset in ray_init_user?
* ***PCATCH tsleeps and have something that will clean the runq
* ***priorities for each tsleep
* ***watchdog to catch screwed up removals?
* ***remember to ccs_free on error in _user routines
* ***check and rationalise CM mappings
* use /sys/net/if_ieee80211.h and update it
* remove ray_reset
* write up driver structure in comments above
* macro for gone and check is at head of all externally called routines
* probably function/macro to test unload at top of commands
* UPDATE_PARAMS seems to return via an interrupt - maybe the timeout
* is needed for wrong values?
* remember it must be serialised as it uses the HCF-ECF area
* check all RECERRs and make sure that some are RAY_PRINTF not RAY_DPRINTF
* could do with selectively calling ray_mcast in ray_init but can't figure
* out a way that doesn't violate the runq or introduce a state var.
* otoh we do have a state var for promisc
* also we are only doing this to reset the mcast list why not
* have a mcast_reset runq?
* or just leave it to the n/w layer to tell us via the ioctls
* havenet needs checking again
* error handling of ECF command completions
* proper setting of mib_hop_seq_len with country code for v4 firmware
@ -291,7 +301,6 @@
#define XXX_ASSOC 0
#define XXX_ACTING_AP 0
#define XXX_INFRA 0
#define XXX_RESET 0
#define XXX_IFQ_PEEK 0
#define XXX_8BIT 0
#define RAY_DEBUG ( \
@ -304,11 +313,12 @@
/* RAY_DBG_MBUF | */ \
/* RAY_DBG_RX | */ \
/* RAY_DBG_CM | */ \
RAY_DBG_COM | \
RAY_DBG_STOP | \
/* RAY_DBG_COM | */ \
/* RAY_DBG_STOP | */ \
/* RAY_DBG_CTL | */ \
/* RAY_DBG_MGT | */ \
/* RAY_DBG_TX | */ \
/* RAY_DBG_DCOM | */ \
0 \
)
@ -420,13 +430,10 @@ static void ray_mcast_done (struct ray_softc *sc, size_t ccs);
static int ray_mcast_user (struct ray_softc *sc);
static int ray_probe (device_t);
static void ray_promisc (struct ray_softc *sc, struct ray_comq_entry *com);
static int ray_promisc_user (struct ray_softc *sc);
static void ray_repparams (struct ray_softc *sc, struct ray_comq_entry *com);
static void ray_repparams_done (struct ray_softc *sc, size_t ccs);
static int ray_repparams_user (struct ray_softc *sc, struct ray_param_req *pr);
static int ray_repstats_user (struct ray_softc *sc, struct ray_stats_req *sr);
static void ray_reset (struct ray_softc *sc);
static void ray_reset_timo (void *xsc);
static int ray_res_alloc_am (struct ray_softc *sc);
static int ray_res_alloc_cm (struct ray_softc *sc);
static int ray_res_alloc_irq (struct ray_softc *sc);
@ -653,7 +660,6 @@ ray_attach(device_t dev)
* Initialise the timers and driver
*/
callout_handle_init(&sc->com_timerh);
callout_handle_init(&sc->reset_timerh);
callout_handle_init(&sc->tx_timerh);
TAILQ_INIT(&sc->sc_comq);
bpfattach(ifp, DLT_EN10MB, sizeof(struct ether_header));
@ -710,6 +716,9 @@ ray_detach(device_t dev)
struct ray_softc *sc = device_get_softc(dev);
struct ifnet *ifp = &sc->arpcom.ac_if;
struct ray_comq_entry *com;
int s;
s = splnet();
RAY_DPRINTF(sc, RAY_DBG_SUBR | RAY_DBG_STOP, "");
@ -717,44 +726,40 @@ ray_detach(device_t dev)
return (0);
/*
* Clear out timers and sort out driver state
* Mark as not running and detach the interface.
*
* N.B. if_detach can trigger ioctls!
*/
com = TAILQ_FIRST(&sc->sc_comq);
for (com = TAILQ_FIRST(&sc->sc_comq); com != NULL;
com = TAILQ_NEXT(com, c_chain)) {
com->c_flags |= RAY_COM_FDETACH;
}
untimeout(ray_com_ecf_timo, sc, sc->com_timerh);
untimeout(ray_reset_timo, sc, sc->reset_timerh);
untimeout(ray_tx_timo, sc, sc->tx_timerh);
sc->gone = 1;
sc->sc_havenet = 0;
/* XXX
What to do with the queue:-
mark all entries as invalid and then invoke runq to sort it out
as the state is held in the queue then we should be okay
*/
/*
* Mark as not running
*/
ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
/*
* Cleardown interface
*/
if_detach(ifp);
/*
* Mark card as gone and release resources
* Stop the runq and wake up anyone sleeping for us.
*/
untimeout(ray_com_ecf_timo, sc, sc->com_timerh);
untimeout(ray_tx_timo, sc, sc->tx_timerh);
com = TAILQ_FIRST(&sc->sc_comq);
for (com = TAILQ_FIRST(&sc->sc_comq); com != NULL;
com = TAILQ_NEXT(com, c_chain)) {
com->c_flags |= RAY_COM_FDETACHED;
com->c_retval = 0; /* XXX ENXIO? */
RAY_DPRINTF(sc, RAY_DBG_STOP, "looking at com %p %b",
com, com->c_flags, RAY_COM_FLAGS_PRINTFB);
if (com->c_flags & RAY_COM_FWOK) {
RAY_DPRINTF(sc, RAY_DBG_STOP, "waking com %p", com);
wakeup(com->c_wakeup);
}
}
/*
* Release resources
*/
sc->gone = 1;
ray_res_release(sc);
RAY_PRINTF(sc, "unloading complete");
RAY_DPRINTF(sc, RAY_DBG_STOP, "unloading complete");
splx(s);
return (0);
}
@ -802,7 +807,7 @@ ray_ioctl(register struct ifnet *ifp, u_long command, caddr_t data)
/* FALLTHROUGH */
case SIOCSIFFLAGS:
RAY_DPRINTF(sc, RAY_DBG_IOCTL, "SIFFLAGS");
RAY_DPRINTF(sc, RAY_DBG_IOCTL, "SIFFLAGS 0x%0x", ifp->if_flags);
/*
* If the interface is marked up we call ray_init_user.
* This will deal with mcast and promisc flags as well as
@ -916,7 +921,7 @@ static int
ray_init_user(struct ray_softc *sc)
{
struct ray_comq_entry *com[5];
int i, ncom, error;
int error, ncom;
RAY_DPRINTF(sc, RAY_DBG_SUBR | RAY_DBG_STARTJOIN, "");
@ -939,20 +944,16 @@ ray_init_user(struct ray_softc *sc)
com[ncom++] = RAY_COM_MALLOC(ray_mcast, 0);
com[ncom++] = RAY_COM_MALLOC(ray_promisc, 0);
error = ray_com_runq_add(sc, com, ncom, "rayinit");
RAY_COM_RUNQ(sc, com, ncom, "rayinit", error);
/* XXX no real error processing from anything yet! */
for (i = 0; i < ncom; i++) {
if (com[i]->c_flags & RAY_COM_FCOMPLETED) {
} else {
ray_ccs_free(sc, com[i]->c_ccs);
TAILQ_REMOVE(&sc->sc_comq, com[i], c_chain);
}
FREE(com[i], M_RAYCOM);
}
RAY_COM_FREE(com, ncom);
return (error);
}
/*
XXX
runq_arr may fail:
if sleeping in ccs_alloc with eintr/erestart/enxio/enodev
@ -973,8 +974,6 @@ runq_arr may fail:
longer term need to attach a desired nw params to the runq entry
*/
return (error);
}
/*
* Runq entry for resetting driver and downloading start up structures to card
@ -1375,22 +1374,21 @@ static int
ray_stop_user(struct ray_softc *sc)
{
struct ray_comq_entry *com[1];
int error;
int error, ncom;
RAY_DPRINTF(sc, RAY_DBG_SUBR | RAY_DBG_STOP, "");
/*
* Schedule the real stop routine
*/
com[0] = RAY_COM_MALLOC(ray_stop, 0);
ncom = 0;
com[ncom++] = RAY_COM_MALLOC(ray_stop, 0);
error = ray_com_runq_add(sc, com, 1, "raystop");
RAY_COM_RUNQ(sc, com, ncom, "raystop", error);
/* XXX no real error processing from anything yet! */
if (error)
RAY_PRINTF(sc, "got error from ray_stop 0x%x", error);
FREE(com[0], M_RAYCOM);
RAY_COM_FREE(com, ncom);
return (error);
}
@ -1416,60 +1414,6 @@ ray_stop(struct ray_softc *sc, struct ray_comq_entry *com)
ray_com_runq_done(sc);
}
/*
* Reset the card
*
* I'm using the soft reset command in the COR register. I'm not sure
* if the sequence is right but it does seem to do the right thing. A
* nano second after reset is written the flashing light goes out, and
* a few seconds after the default is written the main card light goes
* out. We wait a while and then re-init the card.
*/
static void
ray_reset(struct ray_softc *sc)
{
#if XXX_RESET
struct ifnet *ifp = &sc->arpcom.ac_if;
#endif /* XXX_RESET */
RAY_DPRINTF(sc, RAY_DBG_SUBR, "");
RAY_MAP_CM(sc);
#if XXX_RESET
if (ifp->if_flags & IFF_RUNNING)
ray_stop(sc);
RAY_PRINTF(sc, "resetting ECF");
ATTR_WRITE_1(sc, RAY_COR, RAY_COR_RESET);
ATTR_WRITE_1(sc, RAY_COR, RAY_COR_DEFAULT);
sc->reset_timerh = timeout(ray_reset_timo, sc, RAY_RESET_TIMEOUT);
#else
RAY_PRINTF(sc, "skip reset card");
#endif /* XXX_RESET */
}
/*
* Finishing resetting and restarting the card
*/
static void
ray_reset_timo(void *xsc)
{
struct ray_softc *sc = (struct ray_softc *)xsc;
RAY_DPRINTF(sc, RAY_DBG_SUBR, "");
RAY_MAP_CM(sc);
if (!RAY_ECF_READY(sc)) {
RAY_DPRINTF(sc, RAY_DBG_RECERR, "ECF busy, re-scheduling self");
sc->reset_timerh = timeout(ray_reset_timo, sc, RAY_RESET_TIMEOUT);
return;
}
RAY_HCS_CLEAR_INTR(sc);
RAY_PRINTF(sc, "XXX need to restart ECF but not in sleepable context");
RAY_PRINTF(sc, "XXX the user routines must restart as required");
}
static void
ray_watchdog(struct ifnet *ifp)
{
@ -1485,7 +1429,6 @@ ray_watchdog(struct ifnet *ifp)
/* XXX may need to have remedial action here
for example
ray_reset
ray_stop
...
ray_init
@ -1723,7 +1666,6 @@ ray_tx_timo(void *xsc)
int s;
RAY_DPRINTF(sc, RAY_DBG_SUBR, "");
RAY_MAP_CM(sc);
if (!(ifp->if_flags & IFF_OACTIVE) && (ifp->if_snd.ifq_head != NULL)) {
s = splimp();
@ -2624,7 +2566,7 @@ static int
ray_mcast_user(struct ray_softc *sc)
{
struct ray_comq_entry *com[2];
int error, ncom, i;
int error, ncom;
RAY_DPRINTF(sc, RAY_DBG_SUBR, "");
@ -2638,12 +2580,11 @@ ray_mcast_user(struct ray_softc *sc)
com[ncom++] = RAY_COM_MALLOC(ray_mcast, 0);
com[ncom++] = RAY_COM_MALLOC(ray_promisc, 0);
error = ray_com_runq_add(sc, com, ncom, "raymcast");
RAY_COM_RUNQ(sc, com, ncom, "raymcast", error);
/* XXX no real error processing from anything yet! */
for (i = 0; i < ncom; i++)
FREE(com[i], M_RAYCOM);
RAY_COM_FREE(com, ncom);
return (error);
}
@ -2762,7 +2703,7 @@ static int
ray_repparams_user(struct ray_softc *sc, struct ray_param_req *pr)
{
struct ray_comq_entry *com[1];
int error, ncom, i;
int error, ncom;
RAY_DPRINTF(sc, RAY_DBG_SUBR, "");
@ -2852,15 +2793,14 @@ ray_repparams_user(struct ray_softc *sc, struct ray_param_req *pr)
com[ncom++] = RAY_COM_MALLOC(ray_repparams, RAY_COM_FWOK);
com[ncom-1]->c_pr = pr;
error = ray_com_runq_add(sc, com, ncom, "rayrepparams");
RAY_COM_RUNQ(sc, com, ncom, "rayrparm", error);
/* XXX no real error processing from anything yet! */
error = com[0]->c_retval;
error = com[0]->c_retval; /*XXX wrong */
if (!error && pr->r_failcause)
error = EINVAL;
for (i = 0; i < ncom; i++)
FREE(com[i], M_RAYCOM);
RAY_COM_FREE(com, ncom);
return (error);
}
@ -2937,7 +2877,7 @@ static int
ray_upparams_user(struct ray_softc *sc, struct ray_param_req *pr)
{
struct ray_comq_entry *com[3];
int i, todo, error, ncom;
int error, ncom, todo;
#define RAY_UPP_SJ 0x1
#define RAY_UPP_PARAMS 0x2
@ -3011,15 +2951,14 @@ ray_upparams_user(struct ray_softc *sc, struct ray_param_req *pr)
#endif /* XXX_ASSOC */
}
error = ray_com_runq_add(sc, com, ncom, "rayupparams");
RAY_COM_RUNQ(sc, com, ncom, "rayuparam", error);
/* XXX no real error processing from anything yet! */
error = com[0]->c_retval;
error = com[0]->c_retval; /* XXX wrong */
if (!error && pr->r_failcause)
error = EINVAL;
for (i = 0; i < ncom; i++)
FREE(com[i], M_RAYCOM);
RAY_COM_FREE(com, ncom);
return (error);
}
@ -3123,10 +3062,10 @@ ray_com_malloc(ray_comqfn_t function, int flags, char *mesg)
*
* We add the commands to the queue first to preserve ioctl ordering.
*
* On any error, this routine simply returns. This ensures that commands
* remain serialised, even though recovery is difficult - but as the
* only failure mechanisms are a signal or detach/stop most callers
* won't bother restarting.
* On recoverable errors, this routine removes the entries from the
* runq. A caller can requeue the commands (and still preserve its own
* processes ioctl ordering) but doesn't have to. When the card is
* detached we get out quickly to prevent panics.
*/
static int
ray_com_runq_add(struct ray_softc *sc, struct ray_comq_entry *com[], int ncom, char *wmesg)
@ -3143,33 +3082,57 @@ ray_com_runq_add(struct ray_softc *sc, struct ray_comq_entry *com[], int ncom, c
com[0]->c_flags |= RAY_COM_FWAIT;
for (i = 0; i < ncom; i++) {
com[i]->c_wakeup = com[ncom-1];
RAY_DCOM(sc, RAY_DBG_COM, com[i], "adding");
RAY_DPRINTF(sc, RAY_DBG_COM, "adding %p", com[i]);
RAY_DCOM(sc, RAY_DBG_DCOM, com[i], "adding");
TAILQ_INSERT_TAIL(&sc->sc_comq, com[i], c_chain);
}
com[ncom-1]->c_flags |= RAY_COM_FWOK;
/*
* Allocate ccs's for each command. If we fail, we bail
* for the caller to sort everything out.
* Allocate ccs's for each command. If we fail, return an error
*/
for (i = 0; i < ncom; i++) {
error = ray_ccs_alloc(sc, &com[i]->c_ccs, wmesg);
if (error)
return (error);
goto cleanup;
}
/*
* Allow the queue to run and if needed sleep
* Allow the queue to run and sleep if needed.
*
* Iff the FDETACHED flag is set in the com entry we waited on
* the driver is in a zombie state! The softc structure has been
* freed by the generic bus detach methods - eek. We tread very
* carefully!
*/
com[0]->c_flags &= ~RAY_COM_FWAIT;
ray_com_runq(sc);
if (TAILQ_FIRST(&sc->sc_comq) != NULL) {
RAY_DPRINTF(sc, RAY_DBG_COM, "sleeping");
error = tsleep(com[ncom-1], PCATCH, wmesg, 0);
RAY_DPRINTF(sc, RAY_DBG_COM, "awakened, tsleep returned 0x%x", error);
if (com[ncom-1]->c_flags & RAY_COM_FDETACHED)
return (ENXIO);
RAY_DPRINTF(sc, RAY_DBG_COM,
"awakened, tsleep returned 0x%x", error);
} else
error = 0;
cleanup:
/*
* Only clean the queue on real errors - we don't care about it
* when we detach.
*/
if (error && (error != ENXIO))
for (i = 0; i < ncom; i++)
if (!(com[i]->c_flags & RAY_COM_FCOMPLETED)) {
RAY_DPRINTF(sc, RAY_DBG_COM, "removing %p",
com[i]);
RAY_DCOM(sc, RAY_DBG_DCOM, com[i], "removing");
TAILQ_REMOVE(&sc->sc_comq, com[i], c_chain);
ray_ccs_free(sc, com[i]->c_ccs);
com[i]->c_ccs = NULL;
}
return (error);
}
@ -3184,73 +3147,18 @@ ray_com_runq(struct ray_softc *sc)
RAY_DPRINTF(sc, RAY_DBG_SUBR | RAY_DBG_COM, "");
com = TAILQ_FIRST(&sc->sc_comq);
#if RAY_DEBUG & RAY_DBG_COM /* XXX this can go later */
if (com == NULL) {
RAY_DPRINTF(sc, RAY_DBG_COM, "empty command queue");
return;
}
if (com->c_flags & RAY_COM_FRUNNING) {
RAY_DPRINTF(sc, RAY_DBG_COM, "command already running");
return;
}
if (com->c_flags & RAY_COM_FWAIT) {
RAY_DPRINTF(sc, RAY_DBG_COM, "command not ready");
return;
}
#else
if ((com == NULL) ||
(com->c_flags & RAY_COM_FRUNNING) ||
(com->c_flags & RAY_COM_FWAIT))
(com->c_flags & RAY_COM_FWAIT) ||
(com->c_flags & RAY_COM_FDETACHED))
return;
#endif /* RAY_DEBUG & RAY_DBG_COM */
com->c_flags |= RAY_COM_FRUNNING;
RAY_DCOM(sc, RAY_DBG_COM, com, "running");
RAY_DPRINTF(sc, RAY_DBG_COM, "running %p", com);
RAY_DCOM(sc, RAY_DBG_DCOM, com, "running");
com->c_function(sc, com);
}
/*
* Abort the execution of a run queue entry and wakeup the
* user level caller.
*
* We do not remove the entry from the runq incase the caller want's to
* retry and to prevent any other commands being run. The user level caller
* must acknowledge the abort.
*/
static void
ray_com_runq_abort(struct ray_softc *sc, struct ray_comq_entry *com, int reason)
{
RAY_DPRINTF(sc, RAY_DBG_SUBR | RAY_DBG_COM, "");
#if RAY_DEBUG & RAY_DBG_COM
if (com != TAILQ_FIRST(&sc->sc_comq))
RAY_PANIC(sc, "com and head of queue");
#endif /* RAY_DEBUG & RAY_DBG_COM */
RAY_DCOM(sc, RAY_DBG_COM, com, "aborting");
com->c_retval = reason;
wakeup(com->c_wakeup);
}
/*
* Remove an aborted command and re-run the queue
*/
static void
ray_com_runq_clrabort(struct ray_softc *sc, struct ray_comq_entry *com)
{
RAY_DPRINTF(sc, RAY_DBG_SUBR | RAY_DBG_COM, "");
#if RAY_DEBUG & RAY_DBG_COM
if (com != TAILQ_FIRST(&sc->sc_comq))
RAY_PANIC(sc, "com and head of queue");
#endif /* RAY_DEBUG & RAY_DBG_COM */
RAY_DCOM(sc, RAY_DBG_COM, com, "removing");
TAILQ_REMOVE(&sc->sc_comq, com, c_chain);
ray_com_runq(sc);
}
/*
* Remove run command, free ccs and wakeup caller.
*
@ -3270,7 +3178,8 @@ ray_com_runq_done(struct ray_softc *sc)
RAY_DPRINTF(sc, RAY_DBG_SUBR | RAY_DBG_COM, "");
com = TAILQ_FIRST(&sc->sc_comq); /* XXX shall we check this as below */
RAY_DCOM(sc, RAY_DBG_COM, com, "removing");
RAY_DPRINTF(sc, RAY_DBG_COM, "removing %p", com);
RAY_DCOM(sc, RAY_DBG_DCOM, com, "removing");
TAILQ_REMOVE(&sc->sc_comq, com, c_chain);
com->c_flags &= ~RAY_COM_FRUNNING;
@ -3323,7 +3232,8 @@ ray_com_ecf(struct ray_softc *sc, struct ray_comq_entry *com)
else if (i == 1)
RAY_PRINTF(sc, "spinning");
RAY_DCOM(sc, RAY_DBG_COM, com, "sending");
RAY_DPRINTF(sc, RAY_DBG_COM, "sending %p", com);
RAY_DCOM(sc, RAY_DBG_DCOM, com, "sending");
SRAM_WRITE_1(sc, RAY_SCB_CCSI, RAY_CCS_INDEX(com->c_ccs));
RAY_ECF_START_CMD(sc);
@ -3436,8 +3346,8 @@ ray_com_ecf_check(struct ray_softc *sc, size_t ccs, char *mesg)
* Obtain a ccs for a commmand
*
* Returns 0 and in `ccsp' the bus offset of the free ccs. Will block
* awaiting free ccs if needed - if the sleep is interrupted EINTR/ERESTART
* is returned.
* awaiting free ccs if needed - if the sleep is interrupted
* EINTR/ERESTART is returned, if the card is ejected we return ENXIO.
*/
static int
ray_ccs_alloc(struct ray_softc *sc, size_t *ccsp, char *wmesg)
@ -3460,6 +3370,9 @@ ray_ccs_alloc(struct ray_softc *sc, size_t *ccsp, char *wmesg)
if (i > RAY_CCS_CMD_LAST) {
RAY_DPRINTF(sc, RAY_DBG_CCS, "sleeping");
error = tsleep(ray_ccs_alloc, PCATCH, wmesg, 0);
/* XXX broken: see runq_add */
if (sc->gone)
error = ENXIO;
RAY_DPRINTF(sc, RAY_DBG_CCS,
"awakened, tsleep returned 0x%x", error);
if (error)
@ -3508,7 +3421,8 @@ ray_ccs_free(struct ray_softc *sc, size_t ccs)
if (!sc->sc_ccsinuse[RAY_CCS_INDEX(ccs)])
RAY_PRINTF(sc, "freeing free ccs 0x%02x", RAY_CCS_INDEX(ccs));
#endif /* RAY_DEBUG & RAY_DBG_CCS */
RAY_CCS_FREE(sc, ccs);
if (!sc->gone)
RAY_CCS_FREE(sc, ccs);
sc->sc_ccsinuse[RAY_CCS_INDEX(ccs)] = 0;
RAY_DPRINTF(sc, RAY_DBG_CCS, "freed 0x%02x", RAY_CCS_INDEX(ccs));
wakeup(ray_ccs_alloc);