d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Device Polling code for -current.

Browse Source 
Non-SMP, i386-only, no polling in the idle loop at the moment.

To use this code you must compile a kernel with

        options DEVICE_POLLING

and at runtime enable polling with

        sysctl kern.polling.enable=1

The percentage of CPU reserved to userland can be set with

        sysctl kern.polling.user_frac=NN (default is 50)

while the remainder is used by polling device drivers and netisr's.
These are the only two variables that you should need to touch. There
are a few more parameters in kern.polling but the default values
are adequate for all purposes. See the code in kern_poll.c for
more details on them.

Polling in the idle loop will be implemented shortly by introducing
a kernel thread which does the job. Until then, the amount of CPU
dedicated to polling will never exceed (100-user_frac).
The equivalent (actually, better) code for -stable is at

	http://info.iet.unipi.it/~luigi/polling/

and also supports polling in the idle loop.

NOTE to Alpha developers:
There is really nothing in this code that is i386-specific.
If you move the 2 lines supporting the new option from
sys/conf/{files,options}.i386 to sys/conf/{files,options} I am
pretty sure that this should work on the Alpha as well, just that
I do not have a suitable test box to try it. If someone feels like
trying it, I would appreciate it.

NOTE to other developers:
sure some things could be done better, and as always I am open to
constructive criticism, which a few of you have already given and
I greatly appreciated.
However, before proposing radical architectural changes, please
take some time to possibly try out this code, or at the very least
read the comments in kern_poll.c, especially re. the reason why I
am using a soft netisr and cannot (I believe) replace it with a
simple timeout.

Quick description of files touched by this commit:

sys/conf/files.i386
        new file kern/kern_poll.c
sys/conf/options.i386
        new option
sys/i386/i386/trap.c
        poll in trap (disabled by default)
sys/kern/kern_clock.c
        initialization and hardclock hooks.
sys/kern/kern_intr.c
        minor swi_net changes
sys/kern/kern_poll.c
        the bulk of the code.
sys/net/if.h
        new flag
sys/net/if_var.h
        declaration for functions used in device drivers.
sys/net/netisr.h
        NETISR_POLL
sys/dev/fxp/if_fxp.c
sys/dev/fxp/if_fxpvar.h
sys/pci/if_dc.c
sys/pci/if_dcreg.h
sys/pci/if_sis.c
sys/pci/if_sisreg.h
        device driver modifications
This commit is contained in:
Luigi Rizzo 2001-12-14 17:56:12 +00:00
parent 99adc698c1
commit e4fc250c15
Notes: svn2git 2020-12-20 02:59:44 +00:00 
svn path=/head/; revision=87902
18 changed files with 816 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
sys/amd64/amd64/trap.c
View File

@ -162,6 +162,11 @@ SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
extern char *syscallnames[];
#endif
#ifdef DEVICE_POLLING
extern u_int32_t poll_in_trap;
extern int ether_poll __P((int count));
#endif /* DEVICE_POLLING */
/*
* Exception, fault, and trap interface to the FreeBSD kernel.
* This common code is called from assembly language IDT gate entry
@ -239,6 +244,11 @@ trap(frame)
trap_fatal(&frame, eva);
}
#ifdef DEVICE_POLLING
if (poll_in_trap)
ether_poll(poll_in_trap);
#endif /* DEVICE_POLLING */
if ((ISPL(frame.tf_cs) == SEL_UPL) ||
((frame.tf_eflags & PSL_VM) && !in_vm86call)) {
/* user trap */

1
sys/conf/files.i386
View File

@ -383,6 +383,7 @@ isa/syscons_isa.c optional sc
isa/vga_isa.c optional vga
kern/imgact_aout.c standard
kern/imgact_gzip.c optional gzip
kern/kern_poll.c optional device_polling
kern/link_aout.c standard
kern/md4c.c optional netsmb
kern/subr_diskmbr.c standard

5
sys/conf/options.i386
View File

@ -209,6 +209,11 @@ NETSMBCRYPTO opt_netsmb.h
# SMB/CIFS filesystem
SMBFS
# -------------------------------
# Polling device handling
# -------------------------------
DEVICE_POLLING opt_global.h
# -------------------------------
# EOF
# -------------------------------

83
sys/dev/dc/if_dc.c
View File

@ -2473,6 +2473,13 @@ static void dc_rxeof(sc)
while(!(sc->dc_ldata->dc_rx_list[i].dc_status & DC_RXSTAT_OWN)) {
#ifdef DEVICE_POLLING
if (ifp->if_ipending & IFF_POLLING) {
if (sc->rxcycles <= 0)
break;
sc->rxcycles--;
}
#endif /* DEVICE_POLLING */
cur_rx = &sc->dc_ldata->dc_rx_list[i];
rxstat = cur_rx->dc_status;
m = sc->dc_cdata.dc_rx_chain[i];
@ -2786,6 +2793,60 @@ static void dc_tx_underrun(sc)
return;
}
#ifdef DEVICE_POLLING
static poll_handler_t dc_poll;
static void
dc_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
{
struct dc_softc *sc = ifp->if_softc;
if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
/* Re-enable interrupts. */
CSR_WRITE_4(sc, DC_IMR, DC_INTRS);
return;
}
sc->rxcycles = count;
dc_rxeof(sc);
dc_txeof(sc);
if (ifp->if_snd.ifq_head != NULL && !(ifp->if_flags & IFF_OACTIVE))
dc_start(ifp);
if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
u_int32_t status;
status = CSR_READ_4(sc, DC_ISR);
status &= (DC_ISR_RX_WATDOGTIMEO|DC_ISR_RX_NOBUF|
DC_ISR_TX_NOBUF|DC_ISR_TX_IDLE|DC_ISR_TX_UNDERRUN|
DC_ISR_BUS_ERR);
if (!status)
return ;
/* ack what we have */
CSR_WRITE_4(sc, DC_ISR, status);
if (status & (DC_ISR_RX_WATDOGTIMEO|DC_ISR_RX_NOBUF) ) {
u_int32_t r = CSR_READ_4(sc, DC_FRAMESDISCARDED);
ifp->if_ierrors += (r & 0xffff) + ((r >> 17) & 0x7ff);
if (dc_rx_resync(sc))
dc_rxeof(sc);
}
/* restart transmit unit if necessary */
if (status & DC_ISR_TX_IDLE && sc->dc_cdata.dc_tx_cnt)
CSR_WRITE_4(sc, DC_TXSTART, 0xFFFFFFFF);
if (status & DC_ISR_TX_UNDERRUN)
dc_tx_underrun(sc);
if (status & DC_ISR_BUS_ERR) {
printf("dc_poll: dc%d bus error\n", sc->dc_unit);
dc_reset(sc);
dc_init(sc);
}
}
}
#endif /* DEVICE_POLLING */
static void dc_intr(arg)
void *arg;
{
@ -2800,6 +2861,14 @@ static void dc_intr(arg)
DC_LOCK(sc);
ifp = &sc->arpcom.ac_if;
#ifdef DEVICE_POLLING
if (ifp->if_ipending & IFF_POLLING)
goto done;
if (ether_poll_register(dc_poll, ifp)) { /* ok, disable interrupts */
CSR_WRITE_4(sc, DC_IMR, 0x00000000);
goto done;
}
#endif /* DEVICE_POLLING */
/* Suppress unwanted interrupts */
if (!(ifp->if_flags & IFF_UP)) {
@ -2864,6 +2933,7 @@ static void dc_intr(arg)
if (ifp->if_snd.ifq_head != NULL)
dc_start(ifp);
done:
DC_UNLOCK(sc);
return;
@ -3163,6 +3233,16 @@ static void dc_init(xsc)
/*
* Enable interrupts.
*/
#ifdef DEVICE_POLLING
/*
* ... but only if we are not polling, and make sure they are off in
* the case of polling. Some cards (e.g. fxp) turn interrupts on
* after a reset.
*/
if (ifp->if_ipending & IFF_POLLING)
CSR_WRITE_4(sc, DC_IMR, 0x00000000);
else
#endif
CSR_WRITE_4(sc, DC_IMR, DC_INTRS);
CSR_WRITE_4(sc, DC_ISR, 0xFFFFFFFF);
@ -3378,6 +3458,9 @@ static void dc_stop(sc)
callout_stop(&sc->dc_stat_ch);
ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
#ifdef DEVICE_POLLING
ether_poll_deregister(ifp);
#endif
DC_CLRBIT(sc, DC_NETCFG, (DC_NETCFG_RX_ON|DC_NETCFG_TX_ON));
CSR_WRITE_4(sc, DC_IMR, 0x00000000);

7
sys/dev/dc/if_dcreg.h
View File

@ -453,7 +453,11 @@ struct dc_desc {
#define DC_FILTER_HASHONLY 0x10400000
#define DC_MAXFRAGS 16
#ifdef DEVICE_POLLING
#define DC_RX_LIST_CNT 192
#else
#define DC_RX_LIST_CNT 64
#endif
#define DC_TX_LIST_CNT 256
#define DC_MIN_FRAMELEN 60
#define DC_RXLEN 1536
@ -717,6 +721,9 @@ struct dc_softc {
int dc_srm_media;
#endif
struct mtx dc_mtx;
#ifdef DEVICE_POLLING
int rxcycles; /* ... when polling */
#endif
};

61
sys/dev/fxp/if_fxp.c
View File

@ -1141,6 +1141,38 @@ fxp_start(struct ifnet *ifp)
}
}
static void fxp_intr_body(struct fxp_softc *sc, u_int8_t statack, int count);
#ifdef DEVICE_POLLING
static poll_handler_t fxp_poll;
static void
fxp_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
{
struct fxp_softc *sc = ifp->if_softc;
u_int8_t statack ;
if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
CSR_WRITE_1(sc, FXP_CSR_SCB_INTRCNTL, 0);
return;
}
statack = FXP_SCB_STATACK_CXTNO | FXP_SCB_STATACK_CNA |
FXP_SCB_STATACK_FR ;
if (cmd == POLL_AND_CHECK_STATUS) {
u_int8_t tmp ;
tmp = CSR_READ_1(sc, FXP_CSR_SCB_STATACK);
if (tmp == 0xff || tmp == 0)
return ; /* nothing to do */
tmp &= ~statack ;
/* ack what we can */
if (tmp != 0)
CSR_WRITE_1(sc, FXP_CSR_SCB_STATACK, tmp);
statack |= tmp ;
}
fxp_intr_body(sc, statack, count);
}
#endif /* DEVICE_POLLING */
/*
* Process interface interrupts.
*/
@ -1148,9 +1180,21 @@ static void
fxp_intr(void *xsc)
{
struct fxp_softc *sc = xsc;
struct ifnet *ifp = &sc->sc_if;
u_int8_t statack;
#ifdef DEVICE_POLLING
struct ifnet *ifp = &sc->sc_if;
if (ifp->if_ipending & IFF_POLLING)
return ;
if (ether_poll_register(fxp_poll, ifp)) {
/* disable interrupts */
CSR_WRITE_1(sc, FXP_CSR_SCB_INTRCNTL, FXP_SCB_INTR_DISABLE);
fxp_poll(ifp, 0, 1);
return ;
}
#endif
if (sc->suspended) {
return;
}
@ -1169,6 +1213,14 @@ fxp_intr(void *xsc)
* First ACK all the interrupts in this pass.
*/
CSR_WRITE_1(sc, FXP_CSR_SCB_STATACK, statack);
fxp_intr_body(sc, statack, -1);
}
}
static void
fxp_intr_body(struct fxp_softc *sc, u_int8_t statack, int count)
{
struct ifnet *ifp = &sc->sc_if;
/*
* Free any finished transmit mbuf chains.
@ -1220,6 +1272,9 @@ fxp_intr(void *xsc)
rfa = (struct fxp_rfa *)(m->m_ext.ext_buf +
RFA_ALIGNMENT_FUDGE);
#ifdef DEVICE_POLLING /* loop at most count times if count >=0 */
if (count < 0 || count-- > 0)
#endif
if (rfa->rfa_status & FXP_RFA_STATUS_C) {
/*
* Remove first packet from the chain.
@ -1276,7 +1331,6 @@ fxp_intr(void *xsc)
fxp_scb_cmd(sc, FXP_SCB_COMMAND_RU_START);
}
}
}
}
/*
@ -1404,6 +1458,9 @@ fxp_stop(struct fxp_softc *sc)
ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
ifp->if_timer = 0;
#ifdef DEVICE_POLLING
ether_poll_deregister(ifp);
#endif
/*
* Cancel stats updater.
*/

4
sys/dev/fxp/if_fxpvar.h
View File

@ -57,7 +57,11 @@
* Number of receive frame area buffers. These are large so chose
* wisely.
*/
#ifdef DEVICE_POLLING
#define FXP_NRFABUFS 192
#else
#define FXP_NRFABUFS 64
#endif
/*
* Maximum number of seconds that the receiver can be idle before we

10
sys/i386/i386/trap.c
View File

@ -162,6 +162,11 @@ SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
extern char *syscallnames[];
#endif
#ifdef DEVICE_POLLING
extern u_int32_t poll_in_trap;
extern int ether_poll __P((int count));
#endif /* DEVICE_POLLING */
/*
* Exception, fault, and trap interface to the FreeBSD kernel.
* This common code is called from assembly language IDT gate entry
@ -239,6 +244,11 @@ trap(frame)
trap_fatal(&frame, eva);
}
#ifdef DEVICE_POLLING
if (poll_in_trap)
ether_poll(poll_in_trap);
#endif /* DEVICE_POLLING */
if ((ISPL(frame.tf_cs) == SEL_UPL) ||
((frame.tf_eflags & PSL_VM) && !in_vm86call)) {
/* user trap */

12
sys/kern/kern_clock.c
View File

@ -69,6 +69,12 @@
#include <sys/gmon.h>
#endif
#ifdef DEVICE_POLLING
#include <net/netisr.h> /* for NETISR_POLL */
extern void ether_poll1(void);
extern void hardclock_device_poll(void);
#endif /* DEVICE_POLLING */
static void initclocks __P((void *dummy));
SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
@ -140,6 +146,9 @@ initclocks(dummy)
psdiv = pscnt = 1;
cpu_initclocks();
#ifdef DEVICE_POLLING
register_netisr(NETISR_POLL, ether_poll1);
#endif
/*
* Compute profhz/stathz, and fix profhz if needed.
*/
@ -212,6 +221,9 @@ hardclock(frame)
statclock(frame);
tc_windup();
#ifdef DEVICE_POLLING
hardclock_device_poll();
#endif /* DEVICE_POLLING */
/*
* Process callouts at a very low cpu priority, so we don't keep the

14
sys/kern/kern_intr.c
View File

@ -623,7 +623,16 @@ swi_net(void *dummy)
u_int bits;
int i;
#ifdef DEVICE_POLLING
for (;;) {
int pollmore;
#endif
bits = atomic_readandclear_int(&netisr);
#ifdef DEVICE_POLLING
if (bits == 0)
return;
pollmore = bits & (1 << NETISR_POLL);
#endif
while ((i = ffs(bits)) != 0) {
i--;
if (netisrs[i] != NULL)
@ -632,6 +641,11 @@ swi_net(void *dummy)
printf("swi_net: unregistered isr number: %d.\n", i);
bits &= ~(1 << i);
}
#ifdef DEVICE_POLLING
if (pollmore)
ether_pollmore();
}
#endif
}
/*

422
sys/kern/kern_poll.c Normal file
View File

@ -0,0 +1,422 @@
/*-
* Copyright (c) 2001 Luigi Rizzo
*
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHORS 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 AUTHORS 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.
*
* $FreeBSD$
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/socket.h> /* needed by net/if.h */
#include <sys/sysctl.h>
#include <net/if.h> /* for IFF_* flags */
#include <net/netisr.h> /* for NETISR_POLL */
#ifdef SMP
#error DEVICE_POLLING is not compatible with SMP
#endif
void ether_poll1(void);
void ether_poll(int); /* polling while in trap */
void ether_pollmore(void);
void hardclock_device_poll(void);
/*
* Polling support for [network] device drivers.
*
* Drivers which support this feature try to register with the
* polling code.
*
* If registration is successful, the driver must disable interrupts,
* and further I/O is performed through the handler, which is invoked
* (at least once per clock tick) with 3 arguments: the "arg" passed at
* register time (a struct ifnet pointer), a command, and a "count" limit.
*
* The command can be one of the following:
* POLL_ONLY: quick move of "count" packets from input/output queues.
* POLL_AND_CHECK_STATUS: as above, plus check status registers or do
* other more expensive operations. This command is issued periodically
* but less frequently than POLL_ONLY.
* POLL_DEREGISTER: deregister and return to interrupt mode.
*
* The first two commands are only issued if the interface is marked as
* 'IFF_UP and IFF_RUNNING', the last one only if IFF_RUNNING is set.
*
* The count limit specifies how much work the handler can do during the
* call -- typically this is the number of packets to be received, or
* transmitted, etc. (drivers are free to interpret this number, as long
* as the max time spent in the function grows roughly linearly with the
* count).
*
* Deregistration can be requested by the driver itself (typically in the
* *_stop() routine), or by the polling code, by invoking the handler.
*
* Polling can be globally enabled or disabled with the sysctl variable
* kern.polling.enable (default is 0, disabled)
*
* A second variable controls the sharing of CPU between polling/kernel
* network processing, and other activities (typically userlevel tasks):
* kern.polling.user_frac (between 0 and 100, default 50) sets the share
* of CPU allocated to user tasks. CPU is allocated proportionally to the
* shares, by dynamically adjusting the "count" (poll_burst).
*
* Other parameters can should be left to their default values.
* The following constraints hold
*
* 1 <= poll_each_burst <= poll_burst <= poll_burst_max
* 0 <= poll_in_trap <= poll_each_burst
* MIN_POLL_BURST_MAX <= poll_burst_max <= MAX_POLL_BURST_MAX
*/
#define MIN_POLL_BURST_MAX 10
#define MAX_POLL_BURST_MAX 1000
SYSCTL_NODE(_kern, OID_AUTO, polling, CTLFLAG_RW, 0,
"Device polling parameters");
static u_int32_t poll_burst = 5;
SYSCTL_ULONG(_kern_polling, OID_AUTO, burst, CTLFLAG_RW,
&poll_burst, 0, "Current polling burst size");
static u_int32_t poll_each_burst = 5;
SYSCTL_ULONG(_kern_polling, OID_AUTO, each_burst, CTLFLAG_RW,
&poll_each_burst, 0, "Max size of each burst");
static u_int32_t poll_burst_max = 150; /* good for 100Mbit net and HZ=1000 */
SYSCTL_ULONG(_kern_polling, OID_AUTO, burst_max, CTLFLAG_RW,
&poll_burst_max, 0, "Max Polling burst size");
u_int32_t poll_in_trap; /* used in trap.c */
SYSCTL_ULONG(_kern_polling, OID_AUTO, poll_in_trap, CTLFLAG_RW,
&poll_in_trap, 0, "Poll burst size during a trap");
static u_int32_t user_frac = 50;
SYSCTL_ULONG(_kern_polling, OID_AUTO, user_frac, CTLFLAG_RW,
&user_frac, 0, "Desired user fraction of cpu time");
static u_int32_t reg_frac = 20 ;
SYSCTL_ULONG(_kern_polling, OID_AUTO, reg_frac, CTLFLAG_RW,
&reg_frac, 0, "Every this many cycles poll register");
static u_int32_t short_ticks;
SYSCTL_ULONG(_kern_polling, OID_AUTO, short_ticks, CTLFLAG_RW,
&short_ticks, 0, "Hardclock ticks shorter than they should be");
static u_int32_t lost_polls;
SYSCTL_ULONG(_kern_polling, OID_AUTO, lost_polls, CTLFLAG_RW,
&lost_polls, 0, "How many times we would have lost a poll tick");
static u_int32_t poll_handlers; /* next free entry in pr[]. */
SYSCTL_ULONG(_kern_polling, OID_AUTO, handlers, CTLFLAG_RD,
&poll_handlers, 0, "Number of registered poll handlers");
static int polling = 0; /* global polling enable */
SYSCTL_ULONG(_kern_polling, OID_AUTO, enable, CTLFLAG_RW,
&polling, 0, "Polling enabled");
static u_int32_t poll1_active;
static u_int32_t need_poll_again;
#define POLL_LIST_LEN 128
struct pollrec {
poll_handler_t *handler;
struct ifnet *ifp;
};
static struct pollrec pr[POLL_LIST_LEN];
/*
* Hook from hardclock. Tries to schedule a netisr, but keeps track
* of lost ticks due to the previous handler taking too long.
* The first part of the code is just for debugging purposes, and tries
* to count how often hardclock ticks are shorter than they should,
* meaning either stray interrupts or delayed events.
*/
void
hardclock_device_poll(void)
{
static struct timeval prev_t, t;
int delta;
microuptime(&t);
delta = (t.tv_usec - prev_t.tv_usec) +
(t.tv_sec - prev_t.tv_sec)*1000000;
if (delta * hz < 500000)
short_ticks++;
else
prev_t = t;
if (poll_handlers > 0) {
if (poll1_active) {
lost_polls++;
need_poll_again++;
} else {
poll1_active = 1;
schednetisr(NETISR_POLL);
}
}
}
/*
* ether_poll is called from the idle loop or from the trap handler.
*/
void
ether_poll(int count)
{
int i;
int s = splimp();
mtx_lock(&Giant);
if (count > poll_each_burst)
count = poll_each_burst;
for (i = 0 ; i < poll_handlers ; i++)
if (pr[i].handler && (IFF_UP|IFF_RUNNING) ==
(pr[i].ifp->if_flags & (IFF_UP|IFF_RUNNING)) )
pr[i].handler(pr[i].ifp, 0, count); /* quick check */
mtx_unlock(&Giant);
splx(s);
}
/*
* ether_pollmore is called after other netisr's, possibly scheduling
* another NETISR_POLL call, or adapting the burst size for the next cycle.
*
* It is very bad to fetch large bursts of packets from a single card at once,
* because the burst could take a long time to be completely processed, or
* could saturate the intermediate queue (ipintrq or similar) leading to
* losses or unfairness. To reduce the problem, and also to account better for
* time spent in network-related processnig, we split the burst in smaller
* chunks of fixed size, giving control to the other netisr's between chunks.
* This helps in improving the fairness, reducing livelock (because we
* emulate more closely the "process to completion" that we have with
* fastforwarding) and accounting for the work performed in low level
* handling and forwarding.
*/
static int residual_burst = 0;
static struct timeval poll_start_t;
void
ether_pollmore()
{
struct timeval t;
int kern_load;
int s = splhigh();
if (residual_burst > 0) {
schednetisr(NETISR_POLL);
/* will run immediately on return, followed by netisrs */
splx(s);
return ;
}
/* here we can account time spent in netisr's in this tick */
microuptime(&t);
kern_load = (t.tv_usec - poll_start_t.tv_usec) +
(t.tv_sec - poll_start_t.tv_sec)*1000000; /* us */
kern_load = (kern_load * hz) / 10000; /* 0..100 */
if (kern_load > (100 - user_frac)) { /* try decrease ticks */
if (poll_burst > 1)
poll_burst--;
} else {
if (poll_burst < poll_burst_max)
poll_burst++;
}
if (need_poll_again) {
/*
* Last cycle was long and caused us to miss one or more
* hardclock ticks. Restart processnig again, but slightly
* reduce the burst size to prevent that this happens again.
*/
need_poll_again--;
poll_burst -= (poll_burst / 8);
if (poll_burst < 1)
poll_burst = 1;
schednetisr(NETISR_POLL);
} else
poll1_active = 0;
splx(s);
}
/*
* ether_poll1 is called by schednetisr when appropriate, typically once
* per tick. It is called at splnet() so first thing to do is to upgrade to
* splimp(), and call all registered handlers.
*/
void
ether_poll1(void)
{
static int reg_frac_count;
int i, cycles;
enum poll_cmd arg = POLL_ONLY;
int s=splimp();
mtx_lock(&Giant);
if (residual_burst == 0) { /* first call in this tick */
microuptime(&poll_start_t);
/*
* Check that paremeters are consistent with runtime
* variables. Some of these tests could be done at sysctl
* time, but the savings would be very limited because we
* still have to check against reg_frac_count and
* poll_each_burst. So, instead of writing separate sysctl
* handlers, we do all here.
*/
if (reg_frac > hz)
reg_frac = hz;
else if (reg_frac < 1)
reg_frac = 1;
if (reg_frac_count > reg_frac)
reg_frac_count = reg_frac - 1;
if (reg_frac_count-- == 0) {
arg = POLL_AND_CHECK_STATUS;
reg_frac_count = reg_frac - 1;
}
if (poll_burst_max < MIN_POLL_BURST_MAX)
poll_burst_max = MIN_POLL_BURST_MAX;
else if (poll_burst_max > MAX_POLL_BURST_MAX)
poll_burst_max = MAX_POLL_BURST_MAX;
if (poll_each_burst < 1)
poll_each_burst = 1;
else if (poll_each_burst > poll_burst_max)
poll_each_burst = poll_burst_max;
residual_burst = poll_burst;
}
cycles = (residual_burst < poll_each_burst) ?
residual_burst : poll_each_burst;
residual_burst -= cycles;
if (polling) {
for (i = 0 ; i < poll_handlers ; i++)
if (pr[i].handler && (IFF_UP|IFF_RUNNING) ==
(pr[i].ifp->if_flags & (IFF_UP|IFF_RUNNING)) )
pr[i].handler(pr[i].ifp, arg, cycles);
} else { /* unregister */
for (i = 0 ; i < poll_handlers ; i++) {
if (pr[i].handler &&
pr[i].ifp->if_flags & IFF_RUNNING) {
pr[i].ifp->if_ipending &= ~IFF_POLLING;
pr[i].handler(pr[i].ifp, POLL_DEREGISTER, 1);
}
pr[i].handler=NULL;
}
residual_burst = 0;
poll_handlers = 0;
}
/* on -stable, schednetisr(NETISR_POLLMORE); */
mtx_unlock(&Giant);
splx(s);
}
/*
* Try to register routine for polling. Returns 1 if successful
* (and polling should be enabled), 0 otherwise.
* A device is not supposed to register itself multiple times.
*
* This is called from within the *_intr() function, so we should
* probably not need further locking. XXX
*/
int
ether_poll_register(poll_handler_t *h, struct ifnet *ifp)
{
int s;
if (polling == 0) /* polling disabled, cannot register */
return 0;
if (h == NULL || ifp == NULL) /* bad arguments */
return 0;
if ( !(ifp->if_flags & IFF_UP) ) /* must be up */
return 0;
if (ifp->if_ipending & IFF_POLLING) /* already polling */
return 0;
s = splhigh();
if (poll_handlers >= POLL_LIST_LEN) {
/*
* List full, cannot register more entries.
* This should never happen; if it does, it is probably a
* broken driver trying to register multiple times. Checking
* this at runtime is expensive, and won't solve the problem
* anyways, so just report a few times and then give up.
*/
static int verbose = 10 ;
splx(s);
if (verbose >0) {
printf("poll handlers list full, "
"maybe a broken driver ?\n");
verbose--;
}
return 0; /* no polling for you */
}
pr[poll_handlers].handler = h;
pr[poll_handlers].ifp = ifp;
poll_handlers++;
ifp->if_ipending |= IFF_POLLING;
splx(s);
return 1; /* polling enabled in next call */
}
/*
* Remove the interface from the list of polling ones.
* Normally run by *_stop().
* We allow it being called with IFF_POLLING clear, the
* call is sufficiently rare so it is preferable to save the
* space for the extra test in each device in exchange of one
* additional function call.
*/
int
ether_poll_deregister(struct ifnet *ifp)
{
int i;
mtx_lock(&Giant);
if ( !ifp || !(ifp->if_ipending & IFF_POLLING) ) {
mtx_unlock(&Giant);
return 0;
}
for (i = 0 ; i < poll_handlers ; i++)
if (pr[i].ifp == ifp) /* found it */
break;
ifp->if_ipending &= ~IFF_POLLING; /* found or not... */
if (i == poll_handlers) {
mtx_unlock(&Giant);
printf("ether_poll_deregister: ifp not found!!!\n");
return 0;
}
poll_handlers--;
if (i < poll_handlers) { /* Last entry replaces this one. */
pr[i].handler = pr[poll_handlers].handler;
pr[i].ifp = pr[poll_handlers].ifp;
}
mtx_unlock(&Giant);
return 1;
}

9
sys/net/if.h
View File

@ -131,6 +131,15 @@ struct if_data {
#define IFF_ALTPHYS IFF_LINK2 /* use alternate physical connection */
#define IFF_MULTICAST 0x8000 /* supports multicast */
/*
* The following flag(s) ought to go in if_flags, but we cannot change
* struct ifnet because of binary compatibility, so we store them in
* if_ipending, which is not used so far.
* If possible, make sure the value is not conflicting with other
* IFF flags, so we have an easier time when we want to merge them.
*/
#define IFF_POLLING 0x10000 /* Interface is in polling mode. */
/* flags set internally only: */
#define IFF_CANTCHANGE \
(IFF_BROADCAST|IFF_POINTOPOINT|IFF_RUNNING|IFF_OACTIVE|\

9
sys/net/if_var.h
View File

@ -459,6 +459,15 @@ int if_clone_destroy __P((const char *));
#define IF_LLADDR(ifp) \
LLADDR((struct sockaddr_dl *) ifaddr_byindex((ifp)->if_index)->ifa_addr)
#ifdef DEVICE_POLLING
enum poll_cmd { POLL_ONLY, POLL_AND_CHECK_STATUS, POLL_DEREGISTER };
typedef void poll_handler_t __P((struct ifnet *ifp,
enum poll_cmd cmd, int count));
int ether_poll_register __P((poll_handler_t *h, struct ifnet *ifp));
int ether_poll_deregister __P((struct ifnet *ifp));
#endif /* DEVICE_POLLING */
#endif /* _KERNEL */
#endif /* !_NET_IF_VAR_H_ */

1
sys/net/netisr.h
View File

@ -52,6 +52,7 @@
* interrupt used for scheduling the network code to calls
* on the lowest level routine of each protocol.
*/
#define NETISR_POLL 0 /* polling callback, must be first */
#define NETISR_IP 2 /* same as AF_INET */
#define NETISR_NS 6 /* same as AF_NS */
#define NETISR_ATALK 16 /* same as AF_APPLETALK */

83
sys/pci/if_dc.c
View File

@ -2473,6 +2473,13 @@ static void dc_rxeof(sc)
while(!(sc->dc_ldata->dc_rx_list[i].dc_status & DC_RXSTAT_OWN)) {
#ifdef DEVICE_POLLING
if (ifp->if_ipending & IFF_POLLING) {
if (sc->rxcycles <= 0)
break;
sc->rxcycles--;
}
#endif /* DEVICE_POLLING */
cur_rx = &sc->dc_ldata->dc_rx_list[i];
rxstat = cur_rx->dc_status;
m = sc->dc_cdata.dc_rx_chain[i];
@ -2786,6 +2793,60 @@ static void dc_tx_underrun(sc)
return;
}
#ifdef DEVICE_POLLING
static poll_handler_t dc_poll;
static void
dc_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
{
struct dc_softc *sc = ifp->if_softc;
if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
/* Re-enable interrupts. */
CSR_WRITE_4(sc, DC_IMR, DC_INTRS);
return;
}
sc->rxcycles = count;
dc_rxeof(sc);
dc_txeof(sc);
if (ifp->if_snd.ifq_head != NULL && !(ifp->if_flags & IFF_OACTIVE))
dc_start(ifp);
if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
u_int32_t status;
status = CSR_READ_4(sc, DC_ISR);
status &= (DC_ISR_RX_WATDOGTIMEO|DC_ISR_RX_NOBUF|
DC_ISR_TX_NOBUF|DC_ISR_TX_IDLE|DC_ISR_TX_UNDERRUN|
DC_ISR_BUS_ERR);
if (!status)
return ;
/* ack what we have */
CSR_WRITE_4(sc, DC_ISR, status);
if (status & (DC_ISR_RX_WATDOGTIMEO|DC_ISR_RX_NOBUF) ) {
u_int32_t r = CSR_READ_4(sc, DC_FRAMESDISCARDED);
ifp->if_ierrors += (r & 0xffff) + ((r >> 17) & 0x7ff);
if (dc_rx_resync(sc))
dc_rxeof(sc);
}
/* restart transmit unit if necessary */
if (status & DC_ISR_TX_IDLE && sc->dc_cdata.dc_tx_cnt)
CSR_WRITE_4(sc, DC_TXSTART, 0xFFFFFFFF);
if (status & DC_ISR_TX_UNDERRUN)
dc_tx_underrun(sc);
if (status & DC_ISR_BUS_ERR) {
printf("dc_poll: dc%d bus error\n", sc->dc_unit);
dc_reset(sc);
dc_init(sc);
}
}
}
#endif /* DEVICE_POLLING */
static void dc_intr(arg)
void *arg;
{
@ -2800,6 +2861,14 @@ static void dc_intr(arg)
DC_LOCK(sc);
ifp = &sc->arpcom.ac_if;
#ifdef DEVICE_POLLING
if (ifp->if_ipending & IFF_POLLING)
goto done;
if (ether_poll_register(dc_poll, ifp)) { /* ok, disable interrupts */
CSR_WRITE_4(sc, DC_IMR, 0x00000000);
goto done;
}
#endif /* DEVICE_POLLING */
/* Suppress unwanted interrupts */
if (!(ifp->if_flags & IFF_UP)) {
@ -2864,6 +2933,7 @@ static void dc_intr(arg)
if (ifp->if_snd.ifq_head != NULL)
dc_start(ifp);
done:
DC_UNLOCK(sc);
return;
@ -3163,6 +3233,16 @@ static void dc_init(xsc)
/*
* Enable interrupts.
*/
#ifdef DEVICE_POLLING
/*
* ... but only if we are not polling, and make sure they are off in
* the case of polling. Some cards (e.g. fxp) turn interrupts on
* after a reset.
*/
if (ifp->if_ipending & IFF_POLLING)
CSR_WRITE_4(sc, DC_IMR, 0x00000000);
else
#endif
CSR_WRITE_4(sc, DC_IMR, DC_INTRS);
CSR_WRITE_4(sc, DC_ISR, 0xFFFFFFFF);
@ -3378,6 +3458,9 @@ static void dc_stop(sc)
callout_stop(&sc->dc_stat_ch);
ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
#ifdef DEVICE_POLLING
ether_poll_deregister(ifp);
#endif
DC_CLRBIT(sc, DC_NETCFG, (DC_NETCFG_RX_ON|DC_NETCFG_TX_ON));
CSR_WRITE_4(sc, DC_IMR, 0x00000000);

7
sys/pci/if_dcreg.h
View File

@ -453,7 +453,11 @@ struct dc_desc {
#define DC_FILTER_HASHONLY 0x10400000
#define DC_MAXFRAGS 16
#ifdef DEVICE_POLLING
#define DC_RX_LIST_CNT 192
#else
#define DC_RX_LIST_CNT 64
#endif
#define DC_TX_LIST_CNT 256
#define DC_MIN_FRAMELEN 60
#define DC_RXLEN 1536
@ -717,6 +721,9 @@ struct dc_softc {
int dc_srm_media;
#endif
struct mtx dc_mtx;
#ifdef DEVICE_POLLING
int rxcycles; /* ... when polling */
#endif
};

78
sys/pci/if_sis.c
View File

@ -1273,6 +1273,13 @@ static void sis_rxeof(sc)
while(SIS_OWNDESC(&sc->sis_ldata.sis_rx_list[i])) {
#ifdef DEVICE_POLLING
if (ifp->if_ipending & IFF_POLLING) {
if (sc->rxcycles <= 0)
break;
sc->rxcycles--;
}
#endif /* DEVICE_POLLING */
cur_rx = &sc->sis_ldata.sis_rx_list[i];
rxstat = cur_rx->sis_rxstat;
bus_dmamap_sync(sc->sis_tag,
@ -1441,6 +1448,55 @@ static void sis_tick(xsc)
return;
}
#ifdef DEVICE_POLLING
static poll_handler_t sis_poll;
static void
sis_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
{
struct sis_softc *sc = ifp->if_softc;
SIS_LOCK(sc);
if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
CSR_WRITE_4(sc, SIS_IER, 1);
goto done;
}
/*
* On the sis, reading the status register also clears it.
* So before returning to intr mode we must make sure that all
* possible pending sources of interrupts have been served.
* In practice this means run to completion the *eof routines,
* and then call the interrupt routine
*/
sc->rxcycles = count;
sis_rxeof(sc);
sis_txeof(sc);
if (ifp->if_snd.ifq_head != NULL)
sis_start(ifp);
if (sc->rxcycles > 0 || cmd == POLL_AND_CHECK_STATUS) {
u_int32_t status;
/* Reading the ISR register clears all interrupts. */
status = CSR_READ_4(sc, SIS_ISR);
if (status & (SIS_ISR_RX_ERR|SIS_ISR_RX_OFLOW))
sis_rxeoc(sc);
if (status & (SIS_ISR_RX_IDLE))
SIS_SETBIT(sc, SIS_CSR, SIS_CSR_RX_ENABLE);
if (status & SIS_ISR_SYSERR) {
sis_reset(sc);
sis_init(sc);
}
}
done:
SIS_UNLOCK(sc);
return;
}
#endif /* DEVICE_POLLING */
static void sis_intr(arg)
void *arg;
{
@ -1452,6 +1508,15 @@ static void sis_intr(arg)
ifp = &sc->arpcom.ac_if;
SIS_LOCK(sc);
#ifdef DEVICE_POLLING
if (ifp->if_ipending & IFF_POLLING)
goto done;
if (ether_poll_register(sis_poll, ifp)) { /* ok, disable interrupts */
CSR_WRITE_4(sc, SIS_IER, 0);
goto done;
}
#endif /* DEVICE_POLLING */
/* Supress unwanted interrupts */
if (!(ifp->if_flags & IFF_UP)) {
sis_stop(sc);
@ -1741,6 +1806,15 @@ static void sis_init(xsc)
* Enable interrupts.
*/
CSR_WRITE_4(sc, SIS_IMR, SIS_INTRS);
#ifdef DEVICE_POLLING
/*
* ... only enable interrupts if we are not polling, make sure
* they are off otherwise.
*/
if (ifp->if_ipending & IFF_POLLING)
CSR_WRITE_4(sc, SIS_IER, 0);
else
#endif /* DEVICE_POLLING */
CSR_WRITE_4(sc, SIS_IER, 1);
/* Enable receiver and transmitter. */
@ -1910,7 +1984,9 @@ static void sis_stop(sc)
untimeout(sis_tick, sc, sc->sis_stat_ch);
ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
#ifdef DEVICE_POLLING
ether_poll_deregister(ifp);
#endif
CSR_WRITE_4(sc, SIS_IER, 0);
CSR_WRITE_4(sc, SIS_IMR, 0);
SIS_SETBIT(sc, SIS_CSR, SIS_CSR_TX_DISABLE|SIS_CSR_RX_DISABLE);

3
sys/pci/if_sisreg.h
View File

@ -418,6 +418,9 @@ struct sis_softc {
bus_dma_tag_t sis_tag;
struct sis_ring_data sis_cdata;
struct callout_handle sis_stat_ch;
#ifdef DEVICE_POLLING
int rxcycles;
#endif
struct mtx sis_mtx;
};