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
make isa_dmacascade, isa_dmastart, isa_dmadone, and find_isadev MUCH
easier to be found by starting them at the beginging of the line...
remove braces inside of ifdef RESOURCE_CHECK... found by % in vi...
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.)
Here are the remanding changes required to support the Ensoniq
Soundscape using FreeBSD 3.0-current.
Notes:
1) ad1848_init already has code to detect if DMA_DUPLEX should
be set so it is not necessary (and is in fact a mistake) to
hard code setting it. Not all soundcards (i.e. the current
sscape driver) are capable of using DMA_DUPLEX.
2) The other changes are hopefully self explanatory. Feel free
to let me know if you need additional information.
Submitted by: john@feith.com (John Wehle)
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
Sorted the functions into the same order as in ufs_vnops.c so that this
can be compared with the latter without getting 2627 lines of diffs.
Now we get only 1920 lines of diffs.
"high resolution" profiling. The available clocks are:
- the i8254 clock
- on non-SMP i586's and i686's: the TSC
- on systems with I586_PMC_GUPROF configured, and PERFMON configured
and available: all the performance counters.
This is unfinshed (there are problems with locking out the PERFMON
device driver, and with losing calibration after switching the clock),
but better than static configuration or writing to kmem.
Changed ifdefs to avoid generating code for non-working option
combinations.
triple indirect blocks only worked for block sizes of 4K, since
MNINDIR(ump)**3 overflows for larger block sizes (e.g.,
(8192/4)**3 = 2**33 > INT_MAX). This fix is not the obvious one of
changing some types to 64 bits. It rearranges the code to avoid some
unnecessary 64-bit calculations.
Reviewed by: Kirk McKusick <mckusick@McKusick.COM>
a mistake (since softnet interrupts may occur if malloc() waits),
and doing it harmlessly but unnecessarily here interfered with
detection of the mistaken cases.
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
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()).
break for the usual sector size. dscheck() adjusts b_bcount to
handle EOF, so we must scale it and not preserve it. i/o beyond
the end of partitions has been broken since rev.1.96.
Not fixed in: od driver
are used in the `#ifdef notyet' case :-). This case is used except in
the `#if !defined (not_yes)' case :-|. This has something to do with
the `#ifdef notyet_block_reallocation_enabled' case in vfs_cluster.c :-(.
Use OID_AUTO instead of a magic number for the debug.syncprt sysctl.
(This sysctl doesn't actually work. FreeBSD nuked it, but parts
of it were mismerged from Lite2. It is not very good, but better
than nothing.)
`mount -u'. This only matters for `mount -u' competing with unmounts.
If I understand the locking correctly: if mount() blocks, then unmount()
may run and set mp->kern_flag for the same mp. Then unmount() blocks
waiting for mount() to finish. When unmount() continues, its MNTK flags
(MNTK_UNMOUNT and MNTK_MWAIT) may have been clobbered.
Didn't fix old bugs:
- restoring mp->mnt_kern_flag is wrong for the same reasons in the error
case.
- the error case of unmount() seems to be broken too:
(a) MNTK_UNMOUNT gets clobbered, although another unmount() may have
set it. Perhaps it shouldn't be set until after the full lock is
aquired.
(b) MNTK_MWAIT isn't honoured.
Fixed a nearby style bug.
time, but was left at 0. This caused the "can't happen" case in
splz_swi to happen for panics when tsleep() calls splx(safepri)
and there is a SWI_AST pending. This was harmless because the
the error handling happens to be right. Debugging this was tricky
because debugger traps force SWI_AST_MASK on in `cpl'.
there is a natural place to initialize `safepri' in a future commit.
Spinoffs:
- spl0() gets called in the unlikely event that isa is not configured.
- configure() has better control over enabling interrupts.
- it is now less unclear that interrupts aren't actually enabled early.
Rev.1.48 of autoconf.c seems to have done the opposite of what was
intended - moving the isa_configure() call delayed the spl0() side
effect.
Added some comments about the bogons. Removed the splhigh() call since
it is a no-op.
checking the BIOS video mode paramter table. Now syscons uses the
parameter table even if some bits in the table are different from the
current VGA register settings.
Even if comp_vgaregs() finds that the BIOS video parameter table looks
totally unfamiliar to it, syscons allows the user to change the
current video mode to some modes which are based on the VGA 80x25
mode. They are VGA 80x30, VGA 80x50, VGA 80x60. In this case the user
will be warned, during boot, that video mode switching is only
paritally supported on his machine.
PR: bin/4477
this mouse can correctly operate only in the high resolution mode.
If the mouse pointer jumps to the top or left edge or the top-left
corner of the screen, try defining PSM_MSCKLUDGE in the kernel
configuration file. This option will put the mouse in the high
resolution mode during device initialization.
Disallow wait options that are not a combination of the standard POSIX
options WUNTRACED and WNOHANG, as is required by POSIX. BSD doesn't
have any extensions here, but the code was `#ifdef notyet' for some
reason.
interval [VM_MIN_KERNEL_ADDRESS, etext] was used instead of
[btext, etext). Added a comment about this being completely
wrong for LKMs. This only affects interpreting the instructions
after the return to attempt decide the number of args. The
attempt usually fails anyway.
checking was mostly wrong at the boundaries. For the lower limit,
VM_MIN_KERNEL_ADDRESS was used instead of btext and there was an
off-by-(`size' - 1) error. For the upper limit, &etext was used
instead of etext and there was an off-by-1 error. The bugs were
harmless because `size' is not too large and some memory is mapped
just beyond the ends. We still depend on the former to avoid
having to handle the case where the memory range covers the whole
text section, and on the latter to prevent problems when we map
just beyond an end to allow writing an address range that overlaps
the end.
Fixed placement of a nearby comment.
one traditionally reserved for swap devices. The restrictions
should now be the same as the ones for dumpsys(). The restriction
on the partition should be removed someday, and dumpsys() shouldn't
repeat all the checks.
it if the file system is mounted noatime). Not fixed: the access
time is marked at the start of a read() and not marked on successful
completion. I think this should be handled at the vfs level.
Print a better panic message for missing vops. Don't use printf()
before panic(), since the printf()ed part isn't shown by gdb.
This actually loses a little with the current gdb, since gdb just
prints the fmt arg to panic, so %'s aren't expanded. gdb should
fetch the full message from the message buffer if possible.
Fixed default vop function for vop_getpages_desc. It needs to
just return EOPNOTSUPP so that the vnode pager can get the pages
in using a general method. Panicing broke exec'ing of files on
ext2fs file systems. ffs works because it doesn't use the default.
Fixed nearby style bugs.
Fixed nonblocking mode. It was per-device instead of per-file. This
also fixes clobbering of bd_rtout by overloading it to hold a wrong
version of the blocking flag. I hope nothing depends on the bugs.
Fixed nonblocking mode. It was per-device instead of per-file.
Don't depend on gcc's misfeature of rewriting char args in old-style
function definitions to match wrong prototypes. Break K&R1 support
to fix this quickly.
Testing in if_ppp.c is good enough.
Added comments about bogus #includes and #defines.
Removed unused #includes.
Don't depend on gcc's misfeature of rewriting short args in old-style
function definitions to match wrong prototypes. I just changed the
function definition to match the prototype, since this is easy to
verify automatically (it causes no changes in the object code), but
it breaks K&R1 support and doesn't fix the pessimal type.
Use gettime() instead of microtime() to set if_lastchange for i/o's.
microtime() is probably too expensive. However, setting if_lastchange
for i/o's may be wrong.
