mirror/freebsd
1
0
Fork 0
mirror of https://git.FreeBSD.org/src.git synced 2024-12-23 11:18:54 +00:00
Code Issues Releases Activity
7eb884645c
freebsd/sys/kern/subr_devstat.c

579 lines
16 KiB
C
Raw Normal View History

/* -> /*- for copyright notices, minor format tweaks as necessary
2005-01-06 23:35:40 +00:00
/*-
Add a prioritization field to the devstat_add_entry() call so that peripheral drivers can determine where in the devstat(9) list they are inserted. This requires recompilation of libdevstat, systat, vmstat, rpc.rstatd, and any ports that depend on the devstat code, since the size of the devstat structure has changed. The devstat version number has been incremented as well to reflect the change. This sorts devices in the devstat list in "more interesting" to "less interesting" order. So, for instance, da devices are now more important than floppy drives, and so will appear before floppy drives in the default output from systat, iostat, vmstat, etc. The order of devices is, for now, kept in a central table in devicestat.h. If individual drivers were able to make a meaningful decision on what priority they should be at attach time, we could consider splitting the priority information out into the various drivers. For now, though, they have no way of knowing that, so it's easier to put them in an easy to find table. Also, move the checkversion() call in vmstat(8) to a more logical place. Thanks to Bruce and David O'Brien for suggestions, for reviewing this, and for putting up with the long time it has taken me to commit it. Bruce did object somewhat to the central priority table (he would rather the priorities be distributed in each driver), so his objection is duly noted here. Reviewed by: bde, obrien
1999-02-10 00:04:13 +00:00
* Copyright (c) 1997, 1998, 1999 Kenneth D. Merry.
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
* 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.
* 3. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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.
*/
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/systm.h>
Separate the struct bio related stuff out of <sys/buf.h> into <sys/bio.h>. <sys/bio.h> is now a prerequisite for <sys/buf.h> but it shall not be made a nested include according to bdes teachings on the subject of nested includes. Diskdrivers and similar stuff below specfs::strategy() should no longer need to include <sys/buf.> unless they need caching of data. Still a few bogus uses of struct buf to track down. Repocopy by: peter
2000-05-05 09:59:14 +00:00
#include <sys/bio.h>
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
#include <sys/devicestat.h>
Redefine the io provider using the SDT(9) macros instead of doing everything manually. This change has no functional impact. Discussed with: gnn
2013-10-24 02:39:07 +00:00
#include <sys/sdt.h>
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
#include <sys/sysctl.h>
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
#include <sys/malloc.h>
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
#include <sys/lock.h>
#include <sys/mutex.h>
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
#include <sys/conf.h>
#include <vm/vm.h>
#include <vm/pmap.h>
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
#include <machine/atomic.h>
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
Redefine the io provider using the SDT(9) macros instead of doing everything manually. This change has no functional impact. Discussed with: gnn
2013-10-24 02:39:07 +00:00
SDT_PROVIDER_DEFINE(io);
dtrace sdt: remove the ugly sname parameter of SDT_PROBE_DEFINE In its stead use the Solaris / illumos approach of emulating '-' (dash) in probe names with '__' (two consecutive underscores). Reviewed by: markj MFC after: 3 weeks
2013-11-26 08:46:27 +00:00
SDT_PROBE_DEFINE2(io, , , start, "struct bio *", "struct devstat *");
SDT_PROBE_DEFINE2(io, , , done, "struct bio *", "struct devstat *");
SDT_PROBE_DEFINE2(io, , , wait__start, "struct bio *",
Redefine the io provider using the SDT(9) macros instead of doing everything manually. This change has no functional impact. Discussed with: gnn
2013-10-24 02:39:07 +00:00
"struct devstat *");
dtrace sdt: remove the ugly sname parameter of SDT_PROBE_DEFINE In its stead use the Solaris / illumos approach of emulating '-' (dash) in probe names with '__' (two consecutive underscores). Reviewed by: markj MFC after: 3 weeks
2013-11-26 08:46:27 +00:00
SDT_PROBE_DEFINE2(io, , , wait__done, "struct bio *",
Redefine the io provider using the SDT(9) macros instead of doing everything manually. This change has no functional impact. Discussed with: gnn
2013-10-24 02:39:07 +00:00
"struct devstat *");
#define DTRACE_DEVSTAT_START() SDT_PROBE2(io, , , start, NULL, ds)
#define DTRACE_DEVSTAT_BIO_START() SDT_PROBE2(io, , , start, bp, ds)
#define DTRACE_DEVSTAT_DONE() SDT_PROBE2(io, , , done, NULL, ds)
#define DTRACE_DEVSTAT_BIO_DONE() SDT_PROBE2(io, , , done, bp, ds)
dtrace sdt: remove the ugly sname parameter of SDT_PROBE_DEFINE In its stead use the Solaris / illumos approach of emulating '-' (dash) in probe names with '__' (two consecutive underscores). Reviewed by: markj MFC after: 3 weeks
2013-11-26 08:46:27 +00:00
#define DTRACE_DEVSTAT_WAIT_START() SDT_PROBE2(io, , , wait__start, NULL, ds)
#define DTRACE_DEVSTAT_WAIT_DONE() SDT_PROBE2(io, , , wait__done, NULL, ds)
Initial commit of an I/O provider for DTrace on FreeBSD. These probes are most useful when looking into the structures they provide, which are listed in io.d. For example: dtrace -n 'io:genunix::start { printf("%d\n", args[0]->bio_bcount); }' Note that the I/O systems in FreeBSD and Solaris/Illumos are sufficiently different that there is not a 1:1 mapping from scripts that work with one to the other. MFC after: 1 month
2012-07-11 16:27:02 +00:00
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
static int devstat_num_devs;
It is not an error to have no devices in the kernel: Return the generation number and start it from one instead of zero.
2003-08-17 12:06:19 +00:00
static long devstat_generation = 1;
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
static int devstat_version = DEVSTAT_VERSION;
static int devstat_current_devnumber;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
static struct mtx devstat_mutex;
Fix a couple of race conditions in devstat(9) initialization. In devstat_new_entry(), there is no need to initialize the queue and the mutex in this function. There are ways to do static initialization on both, so use STAILQ_HEAD_INITIALIZER and MTX_SYSINIT to initialize the queue and the mutex. In devstat_alloc(), use an atomic test and set routine to guard making our entry in /dev. Using just a plain static variable creates a race condition on multiprocessor machines. If you attempt to create a second entry in devfs, the kernel will panic. Submitted by: kdm Reviewed by: gibbs Sponsored by: Spectra Logic Corporation MFC after: 1 week.
2011-06-13 21:21:02 +00:00
MTX_SYSINIT(devstat_mutex, &devstat_mutex, "devstat", MTX_DEF);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
Fix a couple of race conditions in devstat(9) initialization. In devstat_new_entry(), there is no need to initialize the queue and the mutex in this function. There are ways to do static initialization on both, so use STAILQ_HEAD_INITIALIZER and MTX_SYSINIT to initialize the queue and the mutex. In devstat_alloc(), use an atomic test and set routine to guard making our entry in /dev. Using just a plain static variable creates a race condition on multiprocessor machines. If you attempt to create a second entry in devfs, the kernel will panic. Submitted by: kdm Reviewed by: gibbs Sponsored by: Spectra Logic Corporation MFC after: 1 week.
2011-06-13 21:21:02 +00:00
static struct devstatlist device_statq = STAILQ_HEAD_INITIALIZER(device_statq);
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
static struct devstat *devstat_alloc(void);
static void devstat_free(struct devstat *);
Make devstat_new_entry() take a const void * rather than const char * argument, GEOM nodes are not identified by ascii string.
2003-03-18 07:52:59 +00:00
static void devstat_add_entry(struct devstat *ds, const void *dev_name,
Use ISO C99 integer types in sys/kern where possible. There are only about 100 occurences of the BSD-specific u_int*_t datatypes in sys/kern. The ISO C99 integer types are used here more often.
2010-06-21 09:55:56 +00:00
int unit_number, uint32_t block_size,
Retire devstat_add_entry() as a public function and bump __FreeBSD_version to mark this act.
2003-03-08 21:46:43 +00:00
devstat_support_flags flags,
devstat_type_flags device_type,
devstat_priority priority);
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
/*
* Allocate a devstat and initialize it
*/
struct devstat *
Make devstat_new_entry() take a const void * rather than const char * argument, GEOM nodes are not identified by ascii string.
2003-03-18 07:52:59 +00:00
devstat_new_entry(const void *dev_name,
Use ISO C99 integer types in sys/kern where possible. There are only about 100 occurences of the BSD-specific u_int*_t datatypes in sys/kern. The ISO C99 integer types are used here more often.
2010-06-21 09:55:56 +00:00
int unit_number, uint32_t block_size,
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
devstat_support_flags flags,
devstat_type_flags device_type,
devstat_priority priority)
{
struct devstat *ds;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_assert(&devstat_mutex, MA_NOTOWNED);
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
ds = devstat_alloc();
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_lock(&devstat_mutex);
If devstat_new_entry() is passed a unit number of -1 assume that the devstat is for an "interior" GEOM node and register using the name argument as a geom identity pointer. Do not put these devstat structures on the list returned by the sysctl. This gives us the ability to tell the two kinds of nodes apart and leave the current "strictly physical" view of devstat intact without modifications, yet be able to use devstat for both kinds of devices. It also saves us bloating struct devstat with another 48 bytes of space for the name. At least for now. Reviewed by: ken
2003-03-18 09:30:31 +00:00
if (unit_number == -1) {
Merge GEOM direct dispatch changes from the projects/camlock branch. When safety requirements are met, it allows to avoid passing I/O requests to GEOM g_up/g_down thread, executing them directly in the caller context. That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid several context switches per I/O. The defined now safety requirements are: - caller should not hold any locks and should be reenterable; - callee should not depend on GEOM dual-threaded concurency semantics; - on the way down, if request is unmapped while callee doesn't support it, the context should be sleepable; - kernel thread stack usage should be below 50%. To keep compatibility with GEOM classes not meeting above requirements new provider and consumer flags added: - G_CF_DIRECT_SEND -- consumer code meets caller requirements (request); - G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done); - G_PF_DIRECT_SEND -- provider code meets caller requirements (done); - G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request). Capable GEOM class can set them, allowing direct dispatch in cases where it is safe. If any of requirements are not met, request is queued to g_up or g_down thread same as before. Such GEOM classes were reviewed and updated to support direct dispatch: CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE, VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL, MAP, FLASHMAP, etc). To declare direct completion capability disk(9) KPI got new flag equivalent to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk drivers got it set now thanks to earlier CAM locking work. This change more then twice increases peak block storage performance on systems with manu CPUs, together with earlier CAM locking changes reaching more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to 256 user-level threads). Sponsored by: iXsystems, Inc. MFC after: 2 months
2013-10-22 08:22:19 +00:00
ds->unit_number = unit_number;
If devstat_new_entry() is passed a unit number of -1 assume that the devstat is for an "interior" GEOM node and register using the name argument as a geom identity pointer. Do not put these devstat structures on the list returned by the sysctl. This gives us the ability to tell the two kinds of nodes apart and leave the current "strictly physical" view of devstat intact without modifications, yet be able to use devstat for both kinds of devices. It also saves us bloating struct devstat with another 48 bytes of space for the name. At least for now. Reviewed by: ken
2003-03-18 09:30:31 +00:00
ds->id = dev_name;
binuptime(&ds->creation_time);
devstat_generation++;
} else {
devstat_add_entry(ds, dev_name, unit_number, block_size,
flags, device_type, priority);
}
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_unlock(&devstat_mutex);
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
return (ds);
}
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
/*
* Take a malloced and zeroed devstat structure given to us, fill it in
* and add it to the queue of devices.
*/
Retire devstat_add_entry() as a public function and bump __FreeBSD_version to mark this act.
2003-03-08 21:46:43 +00:00
static void
Make devstat_new_entry() take a const void * rather than const char * argument, GEOM nodes are not identified by ascii string.
2003-03-18 07:52:59 +00:00
devstat_add_entry(struct devstat *ds, const void *dev_name,
Use ISO C99 integer types in sys/kern where possible. There are only about 100 occurences of the BSD-specific u_int*_t datatypes in sys/kern. The ISO C99 integer types are used here more often.
2010-06-21 09:55:56 +00:00
int unit_number, uint32_t block_size,
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
devstat_support_flags flags,
Add a prioritization field to the devstat_add_entry() call so that peripheral drivers can determine where in the devstat(9) list they are inserted. This requires recompilation of libdevstat, systat, vmstat, rpc.rstatd, and any ports that depend on the devstat code, since the size of the devstat structure has changed. The devstat version number has been incremented as well to reflect the change. This sorts devices in the devstat list in "more interesting" to "less interesting" order. So, for instance, da devices are now more important than floppy drives, and so will appear before floppy drives in the default output from systat, iostat, vmstat, etc. The order of devices is, for now, kept in a central table in devicestat.h. If individual drivers were able to make a meaningful decision on what priority they should be at attach time, we could consider splitting the priority information out into the various drivers. For now, though, they have no way of knowing that, so it's easier to put them in an easy to find table. Also, move the checkversion() call in vmstat(8) to a more logical place. Thanks to Bruce and David O'Brien for suggestions, for reviewing this, and for putting up with the long time it has taken me to commit it. Bruce did object somewhat to the central priority table (he would rather the priorities be distributed in each driver), so his objection is duly noted here. Reviewed by: bde, obrien
1999-02-10 00:04:13 +00:00
devstat_type_flags device_type,
devstat_priority priority)
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
{
struct devstatlist *devstat_head;
Add a prioritization field to the devstat_add_entry() call so that peripheral drivers can determine where in the devstat(9) list they are inserted. This requires recompilation of libdevstat, systat, vmstat, rpc.rstatd, and any ports that depend on the devstat code, since the size of the devstat structure has changed. The devstat version number has been incremented as well to reflect the change. This sorts devices in the devstat list in "more interesting" to "less interesting" order. So, for instance, da devices are now more important than floppy drives, and so will appear before floppy drives in the default output from systat, iostat, vmstat, etc. The order of devices is, for now, kept in a central table in devicestat.h. If individual drivers were able to make a meaningful decision on what priority they should be at attach time, we could consider splitting the priority information out into the various drivers. For now, though, they have no way of knowing that, so it's easier to put them in an easy to find table. Also, move the checkversion() call in vmstat(8) to a more logical place. Thanks to Bruce and David O'Brien for suggestions, for reviewing this, and for putting up with the long time it has taken me to commit it. Bruce did object somewhat to the central priority table (he would rather the priorities be distributed in each driver), so his objection is duly noted here. Reviewed by: bde, obrien
1999-02-10 00:04:13 +00:00
struct devstat *ds_tmp;
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_assert(&devstat_mutex, MA_OWNED);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
devstat_num_devs++;
devstat_head = &device_statq;
Add a prioritization field to the devstat_add_entry() call so that peripheral drivers can determine where in the devstat(9) list they are inserted. This requires recompilation of libdevstat, systat, vmstat, rpc.rstatd, and any ports that depend on the devstat code, since the size of the devstat structure has changed. The devstat version number has been incremented as well to reflect the change. This sorts devices in the devstat list in "more interesting" to "less interesting" order. So, for instance, da devices are now more important than floppy drives, and so will appear before floppy drives in the default output from systat, iostat, vmstat, etc. The order of devices is, for now, kept in a central table in devicestat.h. If individual drivers were able to make a meaningful decision on what priority they should be at attach time, we could consider splitting the priority information out into the various drivers. For now, though, they have no way of knowing that, so it's easier to put them in an easy to find table. Also, move the checkversion() call in vmstat(8) to a more logical place. Thanks to Bruce and David O'Brien for suggestions, for reviewing this, and for putting up with the long time it has taken me to commit it. Bruce did object somewhat to the central priority table (he would rather the priorities be distributed in each driver), so his objection is duly noted here. Reviewed by: bde, obrien
1999-02-10 00:04:13 +00:00
/*
* Priority sort. Each driver passes in its priority when it adds
* its devstat entry. Drivers are sorted first by priority, and
* then by probe order.
*
* For the first device, we just insert it, since the priority
* doesn't really matter yet. Subsequent devices are inserted into
* the list using the order outlined above.
*/
if (devstat_num_devs == 1)
STAILQ_INSERT_TAIL(devstat_head, ds, dev_links);
else {
Another round of the <sys/queue.h> FOREACH transmogriffer. Created with: sed(1) Reviewed by: md5(1)
2001-02-04 16:08:18 +00:00
STAILQ_FOREACH(ds_tmp, devstat_head, dev_links) {
Add a prioritization field to the devstat_add_entry() call so that peripheral drivers can determine where in the devstat(9) list they are inserted. This requires recompilation of libdevstat, systat, vmstat, rpc.rstatd, and any ports that depend on the devstat code, since the size of the devstat structure has changed. The devstat version number has been incremented as well to reflect the change. This sorts devices in the devstat list in "more interesting" to "less interesting" order. So, for instance, da devices are now more important than floppy drives, and so will appear before floppy drives in the default output from systat, iostat, vmstat, etc. The order of devices is, for now, kept in a central table in devicestat.h. If individual drivers were able to make a meaningful decision on what priority they should be at attach time, we could consider splitting the priority information out into the various drivers. For now, though, they have no way of knowing that, so it's easier to put them in an easy to find table. Also, move the checkversion() call in vmstat(8) to a more logical place. Thanks to Bruce and David O'Brien for suggestions, for reviewing this, and for putting up with the long time it has taken me to commit it. Bruce did object somewhat to the central priority table (he would rather the priorities be distributed in each driver), so his objection is duly noted here. Reviewed by: bde, obrien
1999-02-10 00:04:13 +00:00
struct devstat *ds_next;
ds_next = STAILQ_NEXT(ds_tmp, dev_links);
/*
* If we find a break between higher and lower
* priority items, and if this item fits in the
* break, insert it. This also applies if the
* "lower priority item" is the end of the list.
*/
if ((priority <= ds_tmp->priority)
&& ((ds_next == NULL)
|| (priority > ds_next->priority))) {
STAILQ_INSERT_AFTER(devstat_head, ds_tmp, ds,
dev_links);
break;
} else if (priority > ds_tmp->priority) {
/*
* If this is the case, we should be able
* to insert ourselves at the head of the
* list. If we can't, something is wrong.
*/
if (ds_tmp == STAILQ_FIRST(devstat_head)) {
STAILQ_INSERT_HEAD(devstat_head,
ds, dev_links);
break;
} else {
STAILQ_INSERT_TAIL(devstat_head,
ds, dev_links);
printf("devstat_add_entry: HELP! "
"sorting problem detected "
Make devstat_new_entry() take a const void * rather than const char * argument, GEOM nodes are not identified by ascii string.
2003-03-18 07:52:59 +00:00
"for name %p unit %d\n",
dev_name, unit_number);
Add a prioritization field to the devstat_add_entry() call so that peripheral drivers can determine where in the devstat(9) list they are inserted. This requires recompilation of libdevstat, systat, vmstat, rpc.rstatd, and any ports that depend on the devstat code, since the size of the devstat structure has changed. The devstat version number has been incremented as well to reflect the change. This sorts devices in the devstat list in "more interesting" to "less interesting" order. So, for instance, da devices are now more important than floppy drives, and so will appear before floppy drives in the default output from systat, iostat, vmstat, etc. The order of devices is, for now, kept in a central table in devicestat.h. If individual drivers were able to make a meaningful decision on what priority they should be at attach time, we could consider splitting the priority information out into the various drivers. For now, though, they have no way of knowing that, so it's easier to put them in an easy to find table. Also, move the checkversion() call in vmstat(8) to a more logical place. Thanks to Bruce and David O'Brien for suggestions, for reviewing this, and for putting up with the long time it has taken me to commit it. Bruce did object somewhat to the central priority table (he would rather the priorities be distributed in each driver), so his objection is duly noted here. Reviewed by: bde, obrien
1999-02-10 00:04:13 +00:00
break;
}
}
}
}
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
ds->device_number = devstat_current_devnumber++;
ds->unit_number = unit_number;
Use strlcpy() instead of strncpy() to copy NUL terminated strings for safety and consistency.
2002-10-17 20:03:38 +00:00
strlcpy(ds->device_name, dev_name, DEVSTAT_NAME_LEN);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
ds->block_size = block_size;
ds->flags = flags;
ds->device_type = device_type;
Add a prioritization field to the devstat_add_entry() call so that peripheral drivers can determine where in the devstat(9) list they are inserted. This requires recompilation of libdevstat, systat, vmstat, rpc.rstatd, and any ports that depend on the devstat code, since the size of the devstat structure has changed. The devstat version number has been incremented as well to reflect the change. This sorts devices in the devstat list in "more interesting" to "less interesting" order. So, for instance, da devices are now more important than floppy drives, and so will appear before floppy drives in the default output from systat, iostat, vmstat, etc. The order of devices is, for now, kept in a central table in devicestat.h. If individual drivers were able to make a meaningful decision on what priority they should be at attach time, we could consider splitting the priority information out into the various drivers. For now, though, they have no way of knowing that, so it's easier to put them in an easy to find table. Also, move the checkversion() call in vmstat(8) to a more logical place. Thanks to Bruce and David O'Brien for suggestions, for reviewing this, and for putting up with the long time it has taken me to commit it. Bruce did object somewhat to the central priority table (he would rather the priorities be distributed in each driver), so his objection is duly noted here. Reviewed by: bde, obrien
1999-02-10 00:04:13 +00:00
ds->priority = priority;
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
binuptime(&ds->creation_time);
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
devstat_generation++;
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
}
/*
* Remove a devstat structure from the list of devices.
*/
void
devstat_remove_entry(struct devstat *ds)
{
struct devstatlist *devstat_head;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_assert(&devstat_mutex, MA_NOTOWNED);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
if (ds == NULL)
return;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_lock(&devstat_mutex);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
devstat_head = &device_statq;
/* Remove this entry from the devstat queue */
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
atomic_add_acq_int(&ds->sequence1, 1);
Merge GEOM direct dispatch changes from the projects/camlock branch. When safety requirements are met, it allows to avoid passing I/O requests to GEOM g_up/g_down thread, executing them directly in the caller context. That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid several context switches per I/O. The defined now safety requirements are: - caller should not hold any locks and should be reenterable; - callee should not depend on GEOM dual-threaded concurency semantics; - on the way down, if request is unmapped while callee doesn't support it, the context should be sleepable; - kernel thread stack usage should be below 50%. To keep compatibility with GEOM classes not meeting above requirements new provider and consumer flags added: - G_CF_DIRECT_SEND -- consumer code meets caller requirements (request); - G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done); - G_PF_DIRECT_SEND -- provider code meets caller requirements (done); - G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request). Capable GEOM class can set them, allowing direct dispatch in cases where it is safe. If any of requirements are not met, request is queued to g_up or g_down thread same as before. Such GEOM classes were reviewed and updated to support direct dispatch: CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE, VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL, MAP, FLASHMAP, etc). To declare direct completion capability disk(9) KPI got new flag equivalent to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk drivers got it set now thanks to earlier CAM locking work. This change more then twice increases peak block storage performance on systems with manu CPUs, together with earlier CAM locking changes reaching more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to 256 user-level threads). Sponsored by: iXsystems, Inc. MFC after: 2 months
2013-10-22 08:22:19 +00:00
if (ds->unit_number != -1) {
If devstat_new_entry() is passed a unit number of -1 assume that the devstat is for an "interior" GEOM node and register using the name argument as a geom identity pointer. Do not put these devstat structures on the list returned by the sysctl. This gives us the ability to tell the two kinds of nodes apart and leave the current "strictly physical" view of devstat intact without modifications, yet be able to use devstat for both kinds of devices. It also saves us bloating struct devstat with another 48 bytes of space for the name. At least for now. Reviewed by: ken
2003-03-18 09:30:31 +00:00
devstat_num_devs--;
STAILQ_REMOVE(devstat_head, ds, devstat, dev_links);
}
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
devstat_free(ds);
devstat_generation++;
mtx_unlock(&devstat_mutex);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
}
/*
* Record a transaction start.
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
*
* See comments for devstat_end_transaction(). Ordering is very important
* here.
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
*/
void
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
devstat_start_transaction(struct devstat *ds, struct bintime *now)
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
{
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_assert(&devstat_mutex, MA_NOTOWNED);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
/* sanity check */
if (ds == NULL)
return;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
atomic_add_acq_int(&ds->sequence1, 1);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
/*
* We only want to set the start time when we are going from idle
* to busy. The start time is really the start of the latest busy
* period.
*/
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
if (ds->start_count == ds->end_count) {
if (now != NULL)
ds->busy_from = *now;
else
binuptime(&ds->busy_from);
}
ds->start_count++;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
atomic_add_rel_int(&ds->sequence0, 1);
Initial commit of an I/O provider for DTrace on FreeBSD. These probes are most useful when looking into the structures they provide, which are listed in io.d. For example: dtrace -n 'io:genunix::start { printf("%d\n", args[0]->bio_bcount); }' Note that the I/O systems in FreeBSD and Solaris/Illumos are sufficiently different that there is not a 1:1 mapping from scripts that work with one to the other. MFC after: 1 month
2012-07-11 16:27:02 +00:00
DTRACE_DEVSTAT_START();
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
}
void
devstat_start_transaction_bio(struct devstat *ds, struct bio *bp)
{
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_assert(&devstat_mutex, MA_NOTOWNED);
/* sanity check */
if (ds == NULL)
return;
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
binuptime(&bp->bio_t0);
devstat_start_transaction(ds, &bp->bio_t0);
Initial commit of an I/O provider for DTrace on FreeBSD. These probes are most useful when looking into the structures they provide, which are listed in io.d. For example: dtrace -n 'io:genunix::start { printf("%d\n", args[0]->bio_bcount); }' Note that the I/O systems in FreeBSD and Solaris/Illumos are sufficiently different that there is not a 1:1 mapping from scripts that work with one to the other. MFC after: 1 month
2012-07-11 16:27:02 +00:00
DTRACE_DEVSTAT_BIO_START();
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
}
/*
* Record the ending of a transaction, and incrment the various counters.
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
*
* Ordering in this function, and in devstat_start_transaction() is VERY
* important. The idea here is to run without locks, so we are very
* careful to only modify some fields on the way "down" (i.e. at
* transaction start) and some fields on the way "up" (i.e. at transaction
* completion). One exception is busy_from, which we only modify in
* devstat_start_transaction() when there are no outstanding transactions,
* and thus it can't be modified in devstat_end_transaction()
* simultaneously.
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
*
* The sequence0 and sequence1 fields are provided to enable an application
* spying on the structures with mmap(2) to tell when a structure is in a
* consistent state or not.
*
* For this to work 100% reliably, it is important that the two fields
* are at opposite ends of the structure and that they are incremented
* in the opposite order of how a memcpy(3) in userland would copy them.
* We assume that the copying happens front to back, but there is actually
* no way short of writing your own memcpy(3) replacement to guarantee
* this will be the case.
*
* In addition to this, being a kind of locks, they must be updated with
* atomic instructions using appropriate memory barriers.
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
*/
void
Use ISO C99 integer types in sys/kern where possible. There are only about 100 occurences of the BSD-specific u_int*_t datatypes in sys/kern. The ISO C99 integer types are used here more often.
2010-06-21 09:55:56 +00:00
devstat_end_transaction(struct devstat *ds, uint32_t bytes,
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
devstat_tag_type tag_type, devstat_trans_flags flags,
struct bintime *now, struct bintime *then)
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
{
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
struct bintime dt, lnow;
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
/* sanity check */
if (ds == NULL)
return;
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
if (now == NULL) {
now = &lnow;
binuptime(now);
}
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
atomic_add_acq_int(&ds->sequence1, 1);
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
/* Update byte and operations counts */
ds->bytes[flags] += bytes;
ds->operations[flags]++;
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
/*
* Keep a count of the various tag types sent.
*/
Move an end-paren to its intended place.
1999-09-19 08:28:49 +00:00
if ((ds->flags & DEVSTAT_NO_ORDERED_TAGS) == 0 &&
Count "free" operations in their own new tranaction type. WARNING: libdevstat, iostat, vmstat, systat etc etc will need a recompile. Add devstat_end_transaction_buf() which pulls all the vital data out of a struct buf which is ready for biodone().
1999-09-18 21:28:09 +00:00
tag_type != DEVSTAT_TAG_NONE)
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
ds->tag_types[tag_type]++;
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
if (then != NULL) {
/* Update duration of operations */
dt = *now;
bintime_sub(&dt, then);
bintime_add(&ds->duration[flags], &dt);
}
/* Accumulate busy time */
dt = *now;
bintime_sub(&dt, &ds->busy_from);
bintime_add(&ds->busy_time, &dt);
ds->busy_from = *now;
ds->end_count++;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
atomic_add_rel_int(&ds->sequence0, 1);
Initial commit of an I/O provider for DTrace on FreeBSD. These probes are most useful when looking into the structures they provide, which are listed in io.d. For example: dtrace -n 'io:genunix::start { printf("%d\n", args[0]->bio_bcount); }' Note that the I/O systems in FreeBSD and Solaris/Illumos are sufficiently different that there is not a 1:1 mapping from scripts that work with one to the other. MFC after: 1 month
2012-07-11 16:27:02 +00:00
DTRACE_DEVSTAT_DONE();
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
}
Clone bio versions of certain bits of infrastructure: devstat_end_transaction_bio() bioq_* versions of bufq_* incl bioqdisksort() the corresponding "buf" versions will disappear when no longer used. Move b_offset, b_data and b_bcount to struct bio. Add BIO_FORMAT as a hack for fd.c etc. We are now largely ready to start converting drivers to use struct bio instead of struct buf.
2000-04-02 19:08:05 +00:00
void
devstat_end_transaction_bio(struct devstat *ds, struct bio *bp)
MFprojects/camlock r254905: Introduce new function devstat_end_transaction_bio_bt(), adding new argument to specify present time. Use this function to move binuptime() out of lock, substantially reducing lock congestion when slow timecounter is used.
2013-10-16 09:12:40 +00:00
{
devstat_end_transaction_bio_bt(ds, bp, NULL);
}
void
devstat_end_transaction_bio_bt(struct devstat *ds, struct bio *bp,
struct bintime *now)
Clone bio versions of certain bits of infrastructure: devstat_end_transaction_bio() bioq_* versions of bufq_* incl bioqdisksort() the corresponding "buf" versions will disappear when no longer used. Move b_offset, b_data and b_bcount to struct bio. Add BIO_FORMAT as a hack for fd.c etc. We are now largely ready to start converting drivers to use struct bio instead of struct buf.
2000-04-02 19:08:05 +00:00
{
devstat_trans_flags flg;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
/* sanity check */
if (ds == NULL)
return;
Clone bio versions of certain bits of infrastructure: devstat_end_transaction_bio() bioq_* versions of bufq_* incl bioqdisksort() the corresponding "buf" versions will disappear when no longer used. Move b_offset, b_data and b_bcount to struct bio. Add BIO_FORMAT as a hack for fd.c etc. We are now largely ready to start converting drivers to use struct bio instead of struct buf.
2000-04-02 19:08:05 +00:00
if (bp->bio_cmd == BIO_DELETE)
flg = DEVSTAT_FREE;
else if (bp->bio_cmd == BIO_READ)
flg = DEVSTAT_READ;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
else if (bp->bio_cmd == BIO_WRITE)
Clone bio versions of certain bits of infrastructure: devstat_end_transaction_bio() bioq_* versions of bufq_* incl bioqdisksort() the corresponding "buf" versions will disappear when no longer used. Move b_offset, b_data and b_bcount to struct bio. Add BIO_FORMAT as a hack for fd.c etc. We are now largely ready to start converting drivers to use struct bio instead of struct buf.
2000-04-02 19:08:05 +00:00
flg = DEVSTAT_WRITE;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
else
flg = DEVSTAT_NO_DATA;
Clone bio versions of certain bits of infrastructure: devstat_end_transaction_bio() bioq_* versions of bufq_* incl bioqdisksort() the corresponding "buf" versions will disappear when no longer used. Move b_offset, b_data and b_bcount to struct bio. Add BIO_FORMAT as a hack for fd.c etc. We are now largely ready to start converting drivers to use struct bio instead of struct buf.
2000-04-02 19:08:05 +00:00
devstat_end_transaction(ds, bp->bio_bcount - bp->bio_resid,
MFprojects/camlock r254905: Introduce new function devstat_end_transaction_bio_bt(), adding new argument to specify present time. Use this function to move binuptime() out of lock, substantially reducing lock congestion when slow timecounter is used.
2013-10-16 09:12:40 +00:00
DEVSTAT_TAG_SIMPLE, flg, now, &bp->bio_t0);
Initial commit of an I/O provider for DTrace on FreeBSD. These probes are most useful when looking into the structures they provide, which are listed in io.d. For example: dtrace -n 'io:genunix::start { printf("%d\n", args[0]->bio_bcount); }' Note that the I/O systems in FreeBSD and Solaris/Illumos are sufficiently different that there is not a 1:1 mapping from scripts that work with one to the other. MFC after: 1 month
2012-07-11 16:27:02 +00:00
DTRACE_DEVSTAT_BIO_DONE();
Clone bio versions of certain bits of infrastructure: devstat_end_transaction_bio() bioq_* versions of bufq_* incl bioqdisksort() the corresponding "buf" versions will disappear when no longer used. Move b_offset, b_data and b_bcount to struct bio. Add BIO_FORMAT as a hack for fd.c etc. We are now largely ready to start converting drivers to use struct bio instead of struct buf.
2000-04-02 19:08:05 +00:00
}
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
/*
* This is the sysctl handler for the devstat package. The data pushed out
* on the kern.devstat.all sysctl variable consists of the current devstat
* generation number, and then an array of devstat structures, one for each
* device in the system.
*
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
* This is more cryptic that obvious, but basically we neither can nor
* want to hold the devstat_mutex for any amount of time, so we grab it
* only when we need to and keep an eye on devstat_generation all the time.
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
*/
static int
Previous commit changing SYSCTL_HANDLER_ARGS violated KNF. Pointed out by: bde
2000-07-04 11:25:35 +00:00
sysctl_devstat(SYSCTL_HANDLER_ARGS)
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
{
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
int error;
Unbreak vinum, iostat and systat on sparc64 by changing the devstat generation number back to a long (sizeof(u_int) != sizeof(long) on sparc64). The alternative would have been to heavily change the libdevstat API. Discussed with: phk, ken
2003-04-17 15:06:28 +00:00
long mygen;
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
struct devstat *nds;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_assert(&devstat_mutex, MA_NOTOWNED);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
/*
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
* XXX devstat_generation should really be "volatile" but that
* XXX freaks out the sysctl macro below. The places where we
* XXX change it and inspect it are bracketed in the mutex which
* XXX guarantees us proper write barriers. I don't belive the
* XXX compiler is allowed to optimize mygen away across calls
* XXX to other functions, so the following is belived to be safe.
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
*/
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mygen = devstat_generation;
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
error = SYSCTL_OUT(req, &mygen, sizeof(mygen));
It is not an error to have no devices in the kernel: Return the generation number and start it from one instead of zero.
2003-08-17 12:06:19 +00:00
if (devstat_num_devs == 0)
return(0);
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
if (error != 0)
return (error);
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_lock(&devstat_mutex);
nds = STAILQ_FIRST(&device_statq);
if (mygen != devstat_generation)
error = EBUSY;
mtx_unlock(&devstat_mutex);
if (error != 0)
return (error);
for (;nds != NULL;) {
error = SYSCTL_OUT(req, nds, sizeof(struct devstat));
if (error != 0)
return (error);
mtx_lock(&devstat_mutex);
if (mygen != devstat_generation)
error = EBUSY;
else
nds = STAILQ_NEXT(nds, dev_links);
mtx_unlock(&devstat_mutex);
if (error != 0)
return (error);
}
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
return(error);
}
/*
* Sysctl entries for devstat. The first one is a node that all the rest
* hang off of.
*/
Mark all SYSCTL_NODEs static that have no corresponding SYSCTL_DECLs. The SYSCTL_NODE macro defines a list that stores all child-elements of that node. If there's no SYSCTL_DECL macro anywhere else, there's no reason why it shouldn't be static.
2011-11-07 15:43:11 +00:00
static SYSCTL_NODE(_kern, OID_AUTO, devstat, CTLFLAG_RD, NULL,
"Device Statistics");
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
SYSCTL_PROC(_kern_devstat, OID_AUTO, all, CTLFLAG_RD|CTLTYPE_OPAQUE,
Use NULL in preference to 0 in pointer contexts.
2009-02-03 07:54:42 +00:00
NULL, 0, sysctl_devstat, "S,devstat", "All devices in the devstat list");
New Kernel device statistics code. Submitted by: "Kenneth D. Merry" <ken@plutotech.com>
1998-09-15 08:16:17 +00:00
/*
* Export the number of devices in the system so that userland utilities
* can determine how much memory to allocate to hold all the devices.
*/
Add sysctl descriptions to many SYSCTL_XXXs PR: kern/11197 Submitted by: Adrian Chadd <adrian@FreeBSD.org> Reviewed by: billf(spelling/style/minor nits) Looked at by: bde(style)
1999-05-03 23:57:32 +00:00
SYSCTL_INT(_kern_devstat, OID_AUTO, numdevs, CTLFLAG_RD,
&devstat_num_devs, 0, "Number of devices in the devstat list");
Unbreak vinum, iostat and systat on sparc64 by changing the devstat generation number back to a long (sizeof(u_int) != sizeof(long) on sparc64). The alternative would have been to heavily change the libdevstat API. Discussed with: phk, ken
2003-04-17 15:06:28 +00:00
SYSCTL_LONG(_kern_devstat, OID_AUTO, generation, CTLFLAG_RD,
Support for unsigned integer and long sysctl variables. Update the SYSCTL_LONG macro to be consistent with other integer sysctl variables and require an initial value instead of assuming 0. Update several sysctl variables to use the unsigned types. PR: 15251 Submitted by: Kelly Yancey <kbyanc@posi.net>
2000-07-05 07:46:41 +00:00
&devstat_generation, 0, "Devstat list generation");
Add sysctl descriptions to many SYSCTL_XXXs PR: kern/11197 Submitted by: Adrian Chadd <adrian@FreeBSD.org> Reviewed by: billf(spelling/style/minor nits) Looked at by: bde(style)
1999-05-03 23:57:32 +00:00
SYSCTL_INT(_kern_devstat, OID_AUTO, version, CTLFLAG_RD,
&devstat_version, 0, "Devstat list version number");
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
/*
* Allocator for struct devstat structures. We sub-allocate these from pages
* which we get from malloc. These pages are exported for mmap(2)'ing through
* a miniature device driver
*/
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
#define statsperpage (PAGE_SIZE / sizeof(struct devstat))
static d_mmap_t devstat_mmap;
static struct cdevsw devstat_cdevsw = {
Device megapatch 4/6: Introduce d_version field in struct cdevsw, this must always be initialized to D_VERSION. Flip sense of D_NOGIANT flag to D_NEEDGIANT, this involves removing four D_NOGIANT flags and adding 145 D_NEEDGIANT flags.
2004-02-21 21:10:55 +00:00
.d_version = D_VERSION,
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
.d_mmap = devstat_mmap,
.d_name = "devstat",
};
struct statspage {
TAILQ_ENTRY(statspage) list;
struct devstat *stat;
u_int nfree;
};
static TAILQ_HEAD(, statspage) pagelist = TAILQ_HEAD_INITIALIZER(pagelist);
static MALLOC_DEFINE(M_DEVSTAT, "devstat", "Device statistics");
static int
Update d_mmap() to accept vm_ooffset_t and vm_memattr_t. This replaces d_mmap() with the d_mmap2() implementation and also changes the type of offset to vm_ooffset_t. Purge d_mmap2(). All driver modules will need to be rebuilt since D_VERSION is also bumped. Reviewed by: jhb@ MFC after: Not in this lifetime...
2009-12-29 21:51:28 +00:00
devstat_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr,
int nprot, vm_memattr_t *memattr)
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
{
struct statspage *spp;
if (nprot != VM_PROT_READ)
return (-1);
change defaule permissions on /dev/devstat. while i'm here remove D_NEEDGIANT flag Submitted by: jhb Reviewed by: jhb, scottl, rwatson, delphij, phk MFC after: 1 week
2014-03-24 18:13:41 +00:00
mtx_lock(&devstat_mutex);
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
TAILQ_FOREACH(spp, &pagelist, list) {
if (offset == 0) {
*paddr = vtophys(spp->stat);
change defaule permissions on /dev/devstat. while i'm here remove D_NEEDGIANT flag Submitted by: jhb Reviewed by: jhb, scottl, rwatson, delphij, phk MFC after: 1 week
2014-03-24 18:13:41 +00:00
mtx_unlock(&devstat_mutex);
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
return (0);
}
offset -= PAGE_SIZE;
}
change defaule permissions on /dev/devstat. while i'm here remove D_NEEDGIANT flag Submitted by: jhb Reviewed by: jhb, scottl, rwatson, delphij, phk MFC after: 1 week
2014-03-24 18:13:41 +00:00
mtx_unlock(&devstat_mutex);
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
return (-1);
}
static struct devstat *
devstat_alloc(void)
{
struct devstat *dsp;
Don't allocate new unnecessary pages when devstat_alloc() looses the run for re-acuiring the lock, but recheck if new pages are allocatable from the pool and free the previously allocated ones. Tested by: pho, Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-18 13:48:38 +00:00
struct statspage *spp, *spp2;
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
u_int u;
static int once;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_assert(&devstat_mutex, MA_NOTOWNED);
Instead of using an atomic operation to determine whether the devstat(9) device node has been created, pass MAKEDEV_CHECKNAME in so that the devfs code will do the check. Use a regular static variable as before, that's good enough to keep us from calling into devfs most of the time. Suggested by: kib MFC after: 1 week Sponsored by: Spectra Logic Corporation
2011-06-13 22:08:24 +00:00
if (!once) {
make_dev_credf(MAKEDEV_ETERNAL | MAKEDEV_CHECKNAME,
change defaule permissions on /dev/devstat. while i'm here remove D_NEEDGIANT flag Submitted by: jhb Reviewed by: jhb, scottl, rwatson, delphij, phk MFC after: 1 week
2014-03-24 18:13:41 +00:00
&devstat_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0444,
Instead of using an atomic operation to determine whether the devstat(9) device node has been created, pass MAKEDEV_CHECKNAME in so that the devfs code will do the check. Use a regular static variable as before, that's good enough to keep us from calling into devfs most of the time. Suggested by: kib MFC after: 1 week Sponsored by: Spectra Logic Corporation
2011-06-13 22:08:24 +00:00
DEVSTAT_DEVICE_NAME);
once = 1;
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
}
Don't allocate new unnecessary pages when devstat_alloc() looses the run for re-acuiring the lock, but recheck if new pages are allocatable from the pool and free the previously allocated ones. Tested by: pho, Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-18 13:48:38 +00:00
spp2 = NULL;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_lock(&devstat_mutex);
for (;;) {
TAILQ_FOREACH(spp, &pagelist, list) {
if (spp->nfree > 0)
break;
}
if (spp != NULL)
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
break;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_unlock(&devstat_mutex);
Don't allocate new unnecessary pages when devstat_alloc() looses the run for re-acuiring the lock, but recheck if new pages are allocatable from the pool and free the previously allocated ones. Tested by: pho, Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-18 13:48:38 +00:00
spp2 = malloc(sizeof *spp, M_DEVSTAT, M_ZERO | M_WAITOK);
spp2->stat = malloc(PAGE_SIZE, M_DEVSTAT, M_ZERO | M_WAITOK);
spp2->nfree = statsperpage;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
/*
Don't allocate new unnecessary pages when devstat_alloc() looses the run for re-acuiring the lock, but recheck if new pages are allocatable from the pool and free the previously allocated ones. Tested by: pho, Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-18 13:48:38 +00:00
* If free statspages were added while the lock was released
* just reuse them.
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
*/
Don't allocate new unnecessary pages when devstat_alloc() looses the run for re-acuiring the lock, but recheck if new pages are allocatable from the pool and free the previously allocated ones. Tested by: pho, Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-18 13:48:38 +00:00
mtx_lock(&devstat_mutex);
TAILQ_FOREACH(spp, &pagelist, list)
if (spp->nfree > 0)
break;
if (spp == NULL) {
spp = spp2;
/*
* It would make more sense to add the new page at the
* head but the order on the list determine the
* sequence of the mapping so we can't do that.
*/
TAILQ_INSERT_TAIL(&pagelist, spp, list);
} else
break;
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
}
dsp = spp->stat;
for (u = 0; u < statsperpage; u++) {
if (dsp->allocated == 0)
break;
dsp++;
}
spp->nfree--;
dsp->allocated = 1;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_unlock(&devstat_mutex);
Don't allocate new unnecessary pages when devstat_alloc() looses the run for re-acuiring the lock, but recheck if new pages are allocatable from the pool and free the previously allocated ones. Tested by: pho, Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-18 13:48:38 +00:00
if (spp2 != NULL && spp2 != spp) {
free(spp2->stat, M_DEVSTAT);
free(spp2, M_DEVSTAT);
}
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
return (dsp);
}
static void
devstat_free(struct devstat *dsp)
{
struct statspage *spp;
Make devstat fully Giant agnostic: Add a mutex and protect the allocation and traversal of the list with it. When we allocate a page for devstat use we drop the mutex and use M_WAITOK this is not nice, but under the given circumstances the best we can do. In the sysctl handler for returning the devstat entries we do not want to hold the mutex across copyout(9) calls, so we keep a very careful eye on the devstat_generation count, and abandon with EBUSY if it changes under our feet. Specifically test for BIO_WRITE, rather than default non-read,non-deletes as write. Make the default be DEVSTAT_NO_DATA. Add atomic increments of the sequence[01] fields so applications using the mmap'ed view stand a chance of detecting updates in progress. Reviewed by: ken
2003-03-18 09:20:20 +00:00
mtx_assert(&devstat_mutex, MA_OWNED);
Introduce a device driver for /dev/devstat, this will allow us to mmap the device statistics structures into userland instead of using sysctl. Introduce new devstat_new_entry() function which allocates the devstat structure an calls devstat_add_entry() on it.
2003-03-08 19:58:57 +00:00
bzero(dsp, sizeof *dsp);
TAILQ_FOREACH(spp, &pagelist, list) {
if (dsp >= spp->stat && dsp < (spp->stat + statsperpage)) {
spp->nfree++;
return;
}
}
}
Run a revision of the devstat interface: Kernel: Change statistics to use the *uptime() timescale (ie: relative to boottime) rather than the UTC aligned timescale. This makes the device statistics code oblivious to clock steps. Change timestamps to bintime format, they are cheaper. Remove the "busy_count", and replace it with two counter fields: "start_count" and "end_count", which are updated in the down and up paths respectively. This removes the locking constraint on devstat. Add a timestamp argument to devstat_start_transaction(), this will normally be a timestamp set by the *_bio() function in bp->bio_t0. Use this field to calculate duration of I/O operations. Add two timestamp arguments to devstat_end_transaction(), one is the current time, a NULL pointer means "take timestamp yourself", the other is the timestamp of when this transaction started (see above). Change calculation of busy_time to operate on "the salami principle": Only when we are idle, which we can determine by the start+end counts being identical, do we update the "busy_from" field in the down path. In the up path we accumulate the timeslice in busy_time and update busy_from. Change the byte_* and num_* fields into two arrays: bytes[] and operations[]. Userland: Change the misleading "busy_time" name to be called "snap_time" and make the time long double since that is what most users need anyway, fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same timescale as the kernel fields. Change devstat_compute_etime() to operate on struct bintime. Remove the version 2 legacy interface: the change to bintime makes compatibility far too expensive. Fix a bug in systat's "vm" page where boot relative busy times would be bogus. Bump __FreeBSD_version to 500107 Review & Collaboration by: ken
2003-03-15 21:59:06 +00:00
SYSCTL_INT(_debug_sizeof, OID_AUTO, devstat, CTLFLAG_RD,
Use NULL in preference to 0 in pointer contexts.
2009-02-03 07:54:42 +00:00
NULL, sizeof(struct devstat), "sizeof(struct devstat)");
Reference in New Issue Copy Permalink