Commit Graph

1584 Commits

Author SHA1 Message Date
dyson
6bd1f74dcf Some performance improvements, and code cleanups (including changing our
expensive OFF_TO_IDX to btoc whenever possible.)
1997-12-19 09:03:37 +00:00
wollman
438677fced Revert poll() for UFS files to traditional behavior where polling for read-
or writability always returns true.  This works around bugs in netscape and
squid, at a minimum.
1997-12-17 14:44:23 +00:00
eivind
9984ce1b48 Regenerate after changing makesyscalls.sh. 1997-12-16 22:27:22 +00:00
eivind
116485ab6e Move around opt_compat include to accomodate Linulator brokenness (for
the time being).
1997-12-16 18:51:45 +00:00
eivind
01dd6091ed Make COMPAT_43 and COMPAT_SUNOS new-style options. 1997-12-16 17:40:42 +00:00
dg
78fabc0578 Fix bug where a struct buf was free()'d back to the system malloc pool.
Quite amazing that the system runs at all with this bug. Also present in
2.2.5. The bug appears to have come in with changes in rev 1.53.

PR:		might fix PR#5313
Submitted by:	bde
1997-12-16 15:40:29 +00:00
wollman
1a06d88098 Add support for poll(2) on files. vop_nopoll() now returns POLLNVAL
if one of the new poll types is requested; hopefully this will not break
any existing code.  (This is done so that programs have a dependable
way of determining whether a filesystem supports the extended poll types
or not.)

The new poll types added are:

	POLLWRITE - file contents may have been modified
	POLLNLINK - file was linked, unlinked, or renamed
	POLLATTRIB - file's attributes may have been changed
	POLLEXTEND - file was extended

Note that the internal operation of poll() means that it is impossible
for two processes to reliably poll for the same event (this could
be fixed but may not be worth it), so it is not possible to rewrite
`tail -f' to use poll at this time.
1997-12-15 03:09:59 +00:00
msmith
22a4475679 Consult sa_len before trampling it with MSG_COMPAT set.
PR:             kern/5291
Submitted by:   pb@fasterix.freenix.org (Pierre Beyssac)
1997-12-15 02:29:11 +00:00
tegge
ba494eff3a Add support for low resolution SMP kernel profiling.
- A nonprofiling version of s_lock (called s_lock_np) is used
    by mcount.

  - When profiling is active, more registers are clobbered in
    seemingly simple assembly routines. This means that some
    callers needed to save/restore extra registers.

  - The stack pointer must have space for a 'fake' return address
    in idle, to avoid stack underflow.
1997-12-15 02:18:35 +00:00
tegge
06b61e24e4 Don't forward hardclock or statclock to stopped cpus. Disable forwarding
when a panic has occured.
1997-12-15 01:14:10 +00:00
jdp
8e5b281441 Make gzipped dynamically linked executables work again. There was
an old bug here that failed to copy the a.out header into memory
properly.  It didn't matter until changes were made recently to
the dynamic linker.
1997-12-14 19:36:24 +00:00
msmith
c10e8d5102 As described by the submitter:
... fix a bug with orecvfrom() or recvfrom() called with
the MSG_COMPAT flag on kernels compiled with the COMPAT_43 option.
The symptom is that the fromaddr is not correctly returned.

This affects the Linux emulator.

Submitted by:	pb@fasterix.freenix.org (Pierre Beyssac)
1997-12-14 03:15:21 +00:00
dyson
738872cad6 After one of my analysis passes to evaluate methods for SMP TLB mgmt, I
noticed some major enhancements available for UP situations.  The number
of UP TLB flushes is decreased much more than significantly with these
changes.  Since a TLB flush appears to cost minimally approx 80 cycles,
this is a "nice" enhancement, equiv to eliminating between 40 and 160
instructions per TLB flush.

Changes include making sure that kernel threads all use the same PTD,
and eliminate unneeded PTD switches at context switch time.
1997-12-14 02:11:23 +00:00
tegge
982ed8caa2 Add needed #include.
Problem found by: Bruce Evans <bde@zeta.org.au>
1997-12-12 21:45:23 +00:00
dyson
1835b7ab0e We have had support for running the kernel daemons as threads for
quite a while, but forgot to do so.  For now, this code supports
most daemons  running as kernel threads in UP kernels, and as
full processes in SMP.  We will soon be able to run them as
threads in SMP, but not yet.
1997-12-12 04:00:59 +00:00
dyson
eb84a1691e Quiet some lint. 1997-12-10 04:14:23 +00:00
fsmp
6dd7dcc53b The improvements to clock statistics by Tor Egge
Wrappered and enabled by the define BETTER_CLOCK (on by default in smpyests.h)

Reviewed by:	smp@csn.net
Submitted by:	Tor Egge <Tor.Egge@idi.ntnu.no>
1997-12-08 23:00:24 +00:00
jmg
525153d523 add process id to tmp files... this prevents two runs from stomping
over each other's tmp files...  (usr.bin/truss uncovered this bug)
1997-12-08 09:00:47 +00:00
dyson
f22f07ffe0 Correct prototypes to match POSIX. Correct return code for aio_cancel.
Submitted by:	Alex Nash <nash@mcs.com>
1997-12-08 02:18:25 +00:00
sef
287c0e3604 Use at_exit() to invoke procfs_exit() instead of calling it directly.
Note that an unload facility should be used to call rm_at_exit() (if
procfs is being loaded as an LKM and is subsequently removed), but it
was non-obvious how to do this in the VFS framework.

Reviewed by:	Julian Elischer
1997-12-08 01:06:36 +00:00
sef
df8959040c Surround the call to procfs_exit() by #ifdef PROCFS/#endif -- much to my
surprise, procfs actually is optional, and some people truly do generate
kernels without it.  Wow.  I built a kernel without 'options PROCFS' and
it compiled and linked.
1997-12-07 18:16:43 +00:00
dyson
331974b30d Slight performance improvement, removal of unneeded SPLs. 1997-12-07 04:06:41 +00:00
bde
f3274407b4 Use ENOIOCTL instead of -1 (= ERESTART) for diskslice ioctls that are
not handled at a particular level.
1997-12-06 14:27:56 +00:00
bde
e868842f11 Use ENOIOCTL instead of -1 (= ERESTART) for tty ioctls that are
not handled at a particular level.  This fixes mainly restarting
of interrupted TIOCDRAINs and TIOCSETA{W,F}s.
1997-12-06 13:25:01 +00:00
sef
c7d273eccb Changes to allow event-based process monitoring and control. 1997-12-06 04:11:14 +00:00
bde
efd51d84cf Don't include <sys/lock.h> in headers when only `struct simplelock' is
required.  Fixed everything that depended on the pollution.
1997-12-05 19:55:52 +00:00
dyson
0f54a3a55f Some fixes from John Hood:
1) Fix the initialization of malloc structure that changed
		due to perf opt.
	2) Remove unneeded include.
	3) An initialization assert added to malloc.
Submitted by:	John Hood <cgull@smoke.marlboro.vt.us>
1997-12-05 05:36:58 +00:00
jmg
81a1ed7d07 document and make the NO_F00F_HACK a proper option...
also, sort some option includes while I'm here..

Forgotten by:	sef
1997-12-04 21:21:26 +00:00
jkh
5409c159ea After consultation with David, change
#ifndef NO_F00F_HACK
to
#if defined(I586_CPU) && !defined(NO_F00F_HACK)
1997-12-04 14:35:40 +00:00
sef
27c596e1ed Work around for the Intel Pentium F00F bug; this is Intel's recommended
workaround.  Note that this currently eats up two pages extra in the system;
this could be alleviated by aligning idt correctly, and then only dealing with
that (as opposed to the current method of allocated two pages and copying the
IDT table to that, and then setting that to be the IDT table).
1997-12-03 02:45:50 +00:00
phk
a1bfb618d9 In all such uses of struct buf: 's/b_un.b_addr/b_data/g' 1997-12-02 21:07:20 +00:00
bde
65d8d62f59 Cleaned up __getcwd(). This should be cosmetic except disabled calls
are now counted.

Reviewed by:	phk
1997-12-02 10:32:21 +00:00
dyson
ded4973e92 Fix a serious problem during resizing buffers where old buffers
address space wasn't being properly reclaimed.
Submitted by:	Bruce Evans <bde@freebsd.org>
1997-12-01 19:04:00 +00:00
dyson
ba00c38faa Fix a problem when creating a new kernel thread. In some cases, aio_read
or aio_write can return the pid of the new thread.  This is due to the
way that return values from system calls being passed by side-effect in
the proc structure now.  This commit fixes the problem with aio_read and
aio_write.
1997-12-01 18:41:08 +00:00
julian
86b96f6423 Cleanup my last patch here
Reviewed by: sef@kthrup.com and phk@freebsd.org
1997-12-01 11:34:41 +00:00
dyson
f6a7ea5e3a Fix error handling for VCHR type I/O. Also, fix another spl problem, and
remove alot of overly verbose debugging statements.
ioproclist {
	int aioprocflags;			/* AIO proc flags */
	TAILQ_ENTRY(aioproclist) list;		/* List of processes */
	struct proc *aioproc;			/* The AIO thread */
	TAILQ_HEAD (,aiocblist) jobtorun;	/* suggested job to run */
};

/*
 * data-structure for lio signal management
 */
struct aio_liojob {
	int lioj_flags;
	int	lioj_buffer_count;
	int	lioj_buffer_finished_count;
	int	lioj_queue_count;
	int	lioj_queue_finished_count;
	struct sigevent lioj_signal;	/* signal on all I/O done */
	TAILQ_ENTRY (aio_liojob) lioj_list;
	struct kaioinfo *lioj_ki;
};
#define	LIOJ_SIGNAL			0x1 /* signal on all done (lio) */
#define	LIOJ_SIGNAL_POSTED	0x2	/* signal has been posted */

/*
 * per process aio data structure
 */
struct kaioinfo {
	int	kaio_flags;			/* per process kaio flags */
	int	kaio_maxactive_count;	/* maximum number of AIOs */
	int	kaio_active_count;	/* number of currently used AIOs */
	int	kaio_qallowed_count;	/* maxiumu size of AIO queue */
	int	kaio_queue_count;	/* size of AIO queue */
	int	kaio_ballowed_count;	/* maximum number of buffers */
	int	kaio_queue_finished_count;	/* number of daemon jobs finished */
	int	kaio_buffer_count;	/* number of physio buffers */
	int	kaio_buffer_finished_count;	/* count of I/O done */
	struct proc *kaio_p;			/* process that uses this kaio block */
	TAILQ_HEAD (,aio_liojob) kaio_liojoblist;	/* list of lio jobs */
	TAILQ_HEAD (,aiocblist)	kaio_jobqueue;	/* job queue for process */
	TAILQ_HEAD (,aiocblist)	kaio_jobdone;	/* done queue for process */
	TAILQ_HEAD (,aiocblist)	kaio_bufqueue;	/* buffer job queue for process */
	TAILQ_HEAD (,aiocblist)	kaio_bufdone;	/* buffer done queue for process */
};

#define KAIO_RUNDOWN 0x1		/* process is being run down */
#define KAIO_WAKEUP 0x2			/* wakeup process when there is a significant
								   event */


TAILQ_HEAD (,aioproclist) aio_freeproc, aio_activeproc;
TAILQ_HEAD(,aiocblist) aio_jobs;			/* Async job list */
TAILQ_HEAD(,aiocblist) aio_bufjobs;			/* Phys I/O job list */
TAILQ_HEAD(,aiocblist) aio_freejobs;		/* Pool of free jobs */

static void aio_init_aioinfo(struct proc *p) ;
static void aio_onceonly(void *) ;
static int aio_free_entry(struct aiocblist *aiocbe);
static void aio_process(struct aiocblist *aiocbe);
static int aio_newproc(void) ;
static int aio_aqueue(struct proc *p, struct aiocb *job, int type) ;
static void aio_physwakeup(struct buf *bp);
static int aio_fphysio(struct proc *p, struct aiocblist *aiocbe, int type);
static int aio_qphysio(struct proc *p, struct aiocblist *iocb);
static void aio_daemon(void *uproc);

SYSINIT(aio, SI_SUB_VFS, SI_ORDER_ANY, aio_onceonly, NULL);

static vm_zone_t kaio_zone=0, aiop_zone=0,
	aiocb_zone=0, aiol_zone=0, aiolio_zone=0;

/*
 * Single AIOD vmspace shared amongst all of them
 */
static struct vmspace *aiovmspace = NULL;

/*
 * Startup initialization
 */
void
aio_onceonly(void *na)
{
	TAILQ_INIT(&aio_freeproc);
	TAILQ_INIT(&aio_activeproc);
	TAILQ_INIT(&aio_jobs);
	TAILQ_INIT(&aio_bufjobs);
	TAILQ_INIT(&aio_freejobs);
	kaio_zone = zinit("AIO", sizeof (struct kaioinfo), 0, 0, 1);
	aiop_zone = zinit("AIOP", sizeof (struct aioproclist), 0, 0, 1);
	aiocb_zone = zinit("AIOCB", sizeof (struct aiocblist), 0, 0, 1);
	aiol_zone = zinit("AIOL", AIO_LISTIO_MAX * sizeof (int), 0, 0, 1);
	aiolio_zone = zinit("AIOLIO",
		AIO_LISTIO_MAX * sizeof (struct aio_liojob), 0, 0, 1);
	aiod_timeout = AIOD_TIMEOUT_DEFAULT;
	aiod_lifetime = AIOD_LIFETIME_DEFAULT;
	jobrefid = 1;
}

/*
 * Init the per-process aioinfo structure.
 * The aioinfo limits are set per-process for user limit (resource) management.
 */
void
aio_init_aioinfo(struct proc *p)
{
	struct kaioinfo *ki;
	if (p->p_aioinfo == NULL) {
		ki = zalloc(kaio_zone);
		p->p_aioinfo = ki
1997-12-01 07:01:45 +00:00
dyson
4f2c953311 Correct a last minute code change. Would have been an infinite loop under
certain error conditions.
Submitted by:	pst@shockwave.com
1997-11-30 23:21:08 +00:00
dyson
f2ef972c8c Fix an spl nit. 1997-11-30 21:47:36 +00:00
dyson
eb130e286e Finish up the vast majority of the AIO/LIO functionality. Proper signal
support was missing in the previous version of the AIO code.  More
tunables added, and very efficient support for VCHR files has been added.
Kernel threads are not used for VCHR files, all work for such files is
done for the requesting process directly.  Some attempt has been made to
charge the requesting process for resource utilization, but more work
is needed.  aio_fsync is still missing (but the original fsync system
call can be used for now.)  aio_cancel is essentially a noop, but that
is okay per POSIX.  More aio_cancel functionality can be added later,
if it is found to be needed.

The functions implemented include:
	aio_read, aio_write, lio_listio, aio_error, aio_return,
	aio_cancel, aio_suspend.

The code has been implemented to support the POSIX spec 1003.1b
(formerly known as POSIX 1003.4 spec) features of the above.  The
async I/O features are truly async, with the VCHR mode of operation
being essentially the same as physio (for appropriate files) for
maximum efficiency.  This code also supports the signal capability,
is highly tunable, allowing management of resource usage, and
has been written to allow a per process usage quota.

Both the O'Reilly POSIX.4 book and the actual POSIX 1003.1b document
were the reference specs used.  Any filedescriptor can be used with
these new system calls.  I know of no exceptions where these
system calls will not work.  (TTY's will also probably work.)
1997-11-30 04:36:31 +00:00
dyson
bdc807a3bb Disable the VCHR optimization for AIO until I have implemented it. Just in
case anyone wants to play with the POSIX AIO/LIO stuff.  (As it is, it should
work with ANY vnode, on UP systems only, for now.)
1997-11-29 02:57:46 +00:00
dyson
6781e003d5 Fix and complete the AIO syscalls. There are some performance enhancements
coming up soon, but the code is functional.  Docs will be forthcoming.
1997-11-29 01:33:10 +00:00
julian
cf4eb29e47 Shift a few SYSINT() calls around.
this results in a few functions becoming static, and
the SYSINITs being close to the code they are related to.
setting up the dump device is with dumpsys() and
kicking off the scheduler is with the scheduler.
Mounting root is with the code that does it.

Reviewed by: phk
1997-11-25 07:07:48 +00:00
bde
67ac522ce5 Fixed multiple definitions of boothowto.
Fixed bitrot in the read-only access to kern.boottime.
1997-11-24 18:35:04 +00:00
bde
17589a0467 Removed all traces of P_IDLEPROC. It was tested but never set. 1997-11-24 15:15:33 +00:00
bde
48ccf62b8e Fixed some #include messes.
Hid the check of the user %cs in syscall() under `#ifdef DIAGNOSTIC'.
1997-11-24 13:25:37 +00:00
dyson
0c60b939c7 Avoid manipulating the buffer map at interrupt time by deferring bfreekva
to getnewbuf, and remove from brelse.
Reviewed by:	dg@root.com
1997-11-24 06:18:27 +00:00
dyson
99ebdb0835 Fix the buffer flag frobbing. Note: It is invalid to gratuitiously modify
b_flags, and this patch removes unneeded modifications.  Only the needed b_flags
bits are modified now.  (Specifically, it is usually wrong to zero b_flags.)
Submitted by:	bde@freebsd.org
1997-11-24 04:14:21 +00:00
bde
608663aedf Fixed a missing conversion of retval to p_retval in disabled code.
Fixed overflow of FFLAGS() in fcntl(F_SETFL, ...).  This was not
a security hole, but gave wrong results for silly flags values.
E.g., it make fcntl(F_SETFL, -1) equivalent to fcntl(F_SETFL, 0).
POSIX requires ignoring the open mode bits in fcntl() (even if
they would be invalid for open()).
1997-11-23 12:24:59 +00:00
bde
6c224bdd6e Fixed duplicate definitions of M_FILE (one static). 1997-11-23 10:43:49 +00:00
bde
4f5d10af43 Fixed some style bugs in the poll() code.
Removed dead code to "Avoid inadvertently sleeping forever".  hzto()
never returns 0.
1997-11-23 10:30:50 +00:00