freebsd-skq/sys/kern/subr_rman.c

1125 lines
27 KiB
C
Raw Normal View History

/*-
* Copyright 1998 Massachusetts Institute of Technology
*
* Permission to use, copy, modify, and distribute this software and
* its documentation for any purpose and without fee is hereby
* granted, provided that both the above copyright notice and this
* permission notice appear in all copies, that both the above
* copyright notice and this permission notice appear in all
* supporting documentation, and that the name of M.I.T. not be used
* in advertising or publicity pertaining to distribution of the
* software without specific, written prior permission. M.I.T. makes
* no representations about the suitability of this software for any
* purpose. It is provided "as is" without express or implied
* warranty.
*
* THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS
* ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
* SHALL M.I.T. 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.
*/
/*
* The kernel resource manager. This code is responsible for keeping track
* of hardware resources which are apportioned out to various drivers.
* It does not actually assign those resources, and it is not expected
* that end-device drivers will call into this code directly. Rather,
* the code which implements the buses that those devices are attached to,
* and the code which manages CPU resources, will call this code, and the
* end-device drivers will make upcalls to that code to actually perform
* the allocation.
*
* There are two sorts of resources managed by this code. The first is
* the more familiar array (RMAN_ARRAY) type; resources in this class
* consist of a sequence of individually-allocatable objects which have
* been numbered in some well-defined order. Most of the resources
* are of this type, as it is the most familiar. The second type is
* called a gauge (RMAN_GAUGE), and models fungible resources (i.e.,
* resources in which each instance is indistinguishable from every
* other instance). The principal anticipated application of gauges
* is in the context of power consumption, where a bus may have a specific
* power budget which all attached devices share. RMAN_GAUGE is not
* implemented yet.
*
* For array resources, we make one simplifying assumption: two clients
* sharing the same resource must use the same range of indices. That
* is to say, sharing of overlapping-but-not-identical regions is not
* permitted.
*/
#include "opt_ddb.h"
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/bus.h> /* XXX debugging */
#include <machine/bus.h>
#include <sys/rman.h>
#include <sys/sysctl.h>
#ifdef DDB
#include <ddb/ddb.h>
#endif
/*
* We use a linked list rather than a bitmap because we need to be able to
* represent potentially huge objects (like all of a processor's physical
* address space). That is also why the indices are defined to have type
* `unsigned long' -- that being the largest integral type in ISO C (1990).
* The 1999 version of C allows `long long'; we may need to switch to that
* at some point in the future, particularly if we want to support 36-bit
* addresses on IA32 hardware.
*/
struct resource_i {
struct resource r_r;
TAILQ_ENTRY(resource_i) r_link;
LIST_ENTRY(resource_i) r_sharelink;
LIST_HEAD(, resource_i) *r_sharehead;
rman_res_t r_start; /* index of the first entry in this resource */
rman_res_t r_end; /* index of the last entry (inclusive) */
u_int r_flags;
void *r_virtual; /* virtual address of this resource */
device_t r_dev; /* device which has allocated this resource */
struct rman *r_rm; /* resource manager from whence this came */
int r_rid; /* optional rid for this resource. */
};
static int rman_debug = 0;
SYSCTL_INT(_debug, OID_AUTO, rman_debug, CTLFLAG_RWTUN,
&rman_debug, 0, "rman debug");
#define DPRINTF(params) if (rman_debug) printf params
1999-04-11 02:27:06 +00:00
static MALLOC_DEFINE(M_RMAN, "rman", "Resource manager");
struct rman_head rman_head;
static struct mtx rman_mtx; /* mutex to protect rman_head */
static int int_rman_release_resource(struct rman *rm, struct resource_i *r);
static __inline struct resource_i *
int_alloc_resource(int malloc_flag)
{
struct resource_i *r;
r = malloc(sizeof *r, M_RMAN, malloc_flag | M_ZERO);
if (r != NULL) {
r->r_r.__r_i = r;
}
return (r);
}
int
rman_init(struct rman *rm)
{
static int once = 0;
if (once == 0) {
once = 1;
TAILQ_INIT(&rman_head);
mtx_init(&rman_mtx, "rman head", NULL, MTX_DEF);
}
if (rm->rm_start == 0 && rm->rm_end == 0)
rm->rm_end = ~0;
if (rm->rm_type == RMAN_UNINIT)
panic("rman_init");
if (rm->rm_type == RMAN_GAUGE)
panic("implement RMAN_GAUGE");
TAILQ_INIT(&rm->rm_list);
rm->rm_mtx = malloc(sizeof *rm->rm_mtx, M_RMAN, M_NOWAIT | M_ZERO);
if (rm->rm_mtx == NULL)
return ENOMEM;
mtx_init(rm->rm_mtx, "rman", NULL, MTX_DEF);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(&rman_mtx);
TAILQ_INSERT_TAIL(&rman_head, rm, rm_link);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(&rman_mtx);
return 0;
}
int
rman_manage_region(struct rman *rm, rman_res_t start, rman_res_t end)
{
struct resource_i *r, *s, *t;
int rv = 0;
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
DPRINTF(("rman_manage_region: <%s> request: start %#jx, end %#jx\n",
rm->rm_descr, start, end));
if (start < rm->rm_start || end > rm->rm_end)
return EINVAL;
r = int_alloc_resource(M_NOWAIT);
if (r == NULL)
return ENOMEM;
r->r_start = start;
r->r_end = end;
r->r_rm = rm;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(rm->rm_mtx);
/* Skip entries before us. */
TAILQ_FOREACH(s, &rm->rm_list, r_link) {
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
if (s->r_end == ~0)
break;
if (s->r_end + 1 >= r->r_start)
break;
}
/* If we ran off the end of the list, insert at the tail. */
if (s == NULL) {
TAILQ_INSERT_TAIL(&rm->rm_list, r, r_link);
} else {
/* Check for any overlap with the current region. */
if (r->r_start <= s->r_end && r->r_end >= s->r_start) {
rv = EBUSY;
goto out;
}
/* Check for any overlap with the next region. */
t = TAILQ_NEXT(s, r_link);
if (t && r->r_start <= t->r_end && r->r_end >= t->r_start) {
rv = EBUSY;
goto out;
}
/*
* See if this region can be merged with the next region. If
* not, clear the pointer.
*/
if (t && (r->r_end + 1 != t->r_start || t->r_flags != 0))
t = NULL;
/* See if we can merge with the current region. */
if (s->r_end + 1 == r->r_start && s->r_flags == 0) {
/* Can we merge all 3 regions? */
if (t != NULL) {
s->r_end = t->r_end;
TAILQ_REMOVE(&rm->rm_list, t, r_link);
free(r, M_RMAN);
free(t, M_RMAN);
} else {
s->r_end = r->r_end;
free(r, M_RMAN);
}
} else if (t != NULL) {
/* Can we merge with just the next region? */
t->r_start = r->r_start;
free(r, M_RMAN);
} else if (s->r_end < r->r_start) {
TAILQ_INSERT_AFTER(&rm->rm_list, s, r, r_link);
} else {
TAILQ_INSERT_BEFORE(s, r, r_link);
}
}
out:
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(rm->rm_mtx);
return rv;
}
int
rman_init_from_resource(struct rman *rm, struct resource *r)
{
int rv;
if ((rv = rman_init(rm)) != 0)
return (rv);
return (rman_manage_region(rm, r->__r_i->r_start, r->__r_i->r_end));
}
int
rman_fini(struct rman *rm)
{
struct resource_i *r;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(rm->rm_mtx);
TAILQ_FOREACH(r, &rm->rm_list, r_link) {
if (r->r_flags & RF_ALLOCATED) {
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(rm->rm_mtx);
return EBUSY;
}
}
/*
* There really should only be one of these if we are in this
* state and the code is working properly, but it can't hurt.
*/
while (!TAILQ_EMPTY(&rm->rm_list)) {
r = TAILQ_FIRST(&rm->rm_list);
TAILQ_REMOVE(&rm->rm_list, r, r_link);
free(r, M_RMAN);
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(rm->rm_mtx);
mtx_lock(&rman_mtx);
TAILQ_REMOVE(&rman_head, rm, rm_link);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(&rman_mtx);
mtx_destroy(rm->rm_mtx);
free(rm->rm_mtx, M_RMAN);
return 0;
}
int
rman_first_free_region(struct rman *rm, rman_res_t *start, rman_res_t *end)
{
struct resource_i *r;
mtx_lock(rm->rm_mtx);
TAILQ_FOREACH(r, &rm->rm_list, r_link) {
if (!(r->r_flags & RF_ALLOCATED)) {
*start = r->r_start;
*end = r->r_end;
mtx_unlock(rm->rm_mtx);
return (0);
}
}
mtx_unlock(rm->rm_mtx);
return (ENOENT);
}
int
rman_last_free_region(struct rman *rm, rman_res_t *start, rman_res_t *end)
{
struct resource_i *r;
mtx_lock(rm->rm_mtx);
TAILQ_FOREACH_REVERSE(r, &rm->rm_list, resource_head, r_link) {
if (!(r->r_flags & RF_ALLOCATED)) {
*start = r->r_start;
*end = r->r_end;
mtx_unlock(rm->rm_mtx);
return (0);
}
}
mtx_unlock(rm->rm_mtx);
return (ENOENT);
}
/* Shrink or extend one or both ends of an allocated resource. */
int
rman_adjust_resource(struct resource *rr, rman_res_t start, rman_res_t end)
{
struct resource_i *r, *s, *t, *new;
struct rman *rm;
/* Not supported for shared resources. */
r = rr->__r_i;
if (r->r_flags & RF_SHAREABLE)
return (EINVAL);
/*
* This does not support wholesale moving of a resource. At
* least part of the desired new range must overlap with the
* existing resource.
*/
if (end < r->r_start || r->r_end < start)
return (EINVAL);
/*
* Find the two resource regions immediately adjacent to the
* allocated resource.
*/
rm = r->r_rm;
mtx_lock(rm->rm_mtx);
#ifdef INVARIANTS
TAILQ_FOREACH(s, &rm->rm_list, r_link) {
if (s == r)
break;
}
if (s == NULL)
panic("resource not in list");
#endif
s = TAILQ_PREV(r, resource_head, r_link);
t = TAILQ_NEXT(r, r_link);
KASSERT(s == NULL || s->r_end + 1 == r->r_start,
("prev resource mismatch"));
KASSERT(t == NULL || r->r_end + 1 == t->r_start,
("next resource mismatch"));
/*
* See if the changes are permitted. Shrinking is always allowed,
* but growing requires sufficient room in the adjacent region.
*/
if (start < r->r_start && (s == NULL || (s->r_flags & RF_ALLOCATED) ||
s->r_start > start)) {
mtx_unlock(rm->rm_mtx);
return (EBUSY);
}
if (end > r->r_end && (t == NULL || (t->r_flags & RF_ALLOCATED) ||
t->r_end < end)) {
mtx_unlock(rm->rm_mtx);
return (EBUSY);
}
/*
* While holding the lock, grow either end of the resource as
* needed and shrink either end if the shrinking does not require
* allocating a new resource. We can safely drop the lock and then
* insert a new range to handle the shrinking case afterwards.
*/
if (start < r->r_start ||
(start > r->r_start && s != NULL && !(s->r_flags & RF_ALLOCATED))) {
KASSERT(s->r_flags == 0, ("prev is busy"));
r->r_start = start;
if (s->r_start == start) {
TAILQ_REMOVE(&rm->rm_list, s, r_link);
free(s, M_RMAN);
} else
s->r_end = start - 1;
}
if (end > r->r_end ||
(end < r->r_end && t != NULL && !(t->r_flags & RF_ALLOCATED))) {
KASSERT(t->r_flags == 0, ("next is busy"));
r->r_end = end;
if (t->r_end == end) {
TAILQ_REMOVE(&rm->rm_list, t, r_link);
free(t, M_RMAN);
} else
t->r_start = end + 1;
}
mtx_unlock(rm->rm_mtx);
/*
* Handle the shrinking cases that require allocating a new
* resource to hold the newly-free region. We have to recheck
* if we still need this new region after acquiring the lock.
*/
if (start > r->r_start) {
new = int_alloc_resource(M_WAITOK);
new->r_start = r->r_start;
new->r_end = start - 1;
new->r_rm = rm;
mtx_lock(rm->rm_mtx);
r->r_start = start;
s = TAILQ_PREV(r, resource_head, r_link);
if (s != NULL && !(s->r_flags & RF_ALLOCATED)) {
s->r_end = start - 1;
free(new, M_RMAN);
} else
TAILQ_INSERT_BEFORE(r, new, r_link);
mtx_unlock(rm->rm_mtx);
}
if (end < r->r_end) {
new = int_alloc_resource(M_WAITOK);
new->r_start = end + 1;
new->r_end = r->r_end;
new->r_rm = rm;
mtx_lock(rm->rm_mtx);
r->r_end = end;
t = TAILQ_NEXT(r, r_link);
if (t != NULL && !(t->r_flags & RF_ALLOCATED)) {
t->r_start = end + 1;
free(new, M_RMAN);
} else
TAILQ_INSERT_AFTER(&rm->rm_list, r, new, r_link);
mtx_unlock(rm->rm_mtx);
}
return (0);
}
#define SHARE_TYPE(f) (f & (RF_SHAREABLE | RF_PREFETCHABLE))
struct resource *
rman_reserve_resource_bound(struct rman *rm, rman_res_t start, rman_res_t end,
rman_res_t count, rman_res_t bound, u_int flags,
device_t dev)
{
u_int new_rflags;
struct resource_i *r, *s, *rv;
rman_res_t rstart, rend, amask, bmask;
rv = NULL;
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
DPRINTF(("rman_reserve_resource_bound: <%s> request: [%#jx, %#jx], "
"length %#jx, flags %x, device %s\n", rm->rm_descr, start, end,
count, flags,
dev == NULL ? "<null>" : device_get_nameunit(dev)));
KASSERT((flags & RF_FIRSTSHARE) == 0,
("invalid flags %#x", flags));
new_rflags = (flags & ~RF_FIRSTSHARE) | RF_ALLOCATED;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(rm->rm_mtx);
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
r = TAILQ_FIRST(&rm->rm_list);
if (r == NULL) {
DPRINTF(("NULL list head\n"));
} else {
DPRINTF(("rman_reserve_resource_bound: trying %#jx <%#jx,%#jx>\n",
r->r_end, start, count-1));
}
for (r = TAILQ_FIRST(&rm->rm_list);
r && r->r_end < start + count - 1;
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
r = TAILQ_NEXT(r, r_link)) {
;
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
DPRINTF(("rman_reserve_resource_bound: tried %#jx <%#jx,%#jx>\n",
r->r_end, start, count-1));
}
if (r == NULL) {
DPRINTF(("could not find a region\n"));
goto out;
}
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
amask = (1ull << RF_ALIGNMENT(flags)) - 1;
KASSERT(start <= RM_MAX_END - amask,
("start (%#jx) + amask (%#jx) would wrap around", start, amask));
/* If bound is 0, bmask will also be 0 */
bmask = ~(bound - 1);
/*
* First try to find an acceptable totally-unshared region.
*/
for (s = r; s; s = TAILQ_NEXT(s, r_link)) {
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
DPRINTF(("considering [%#jx, %#jx]\n", s->r_start, s->r_end));
/*
* The resource list is sorted, so there is no point in
* searching further once r_start is too large.
*/
if (s->r_start > end - (count - 1)) {
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
DPRINTF(("s->r_start (%#jx) + count - 1> end (%#jx)\n",
Sometimes, when asked to return region A..C, we'd return A+N..C+N instead of failing. When looking for a region to allocate, we used to check to see if the start address was < end. In the case where A..B is allocated already, and one wants to allocate A..C (B < C), then this test would improperly fail (which means we'd examine that region as a possible one), and we'd return the region B+1..C+(B-A+1) rather than NULL. Since C+(B-A+1) is necessarily larger than C (end argument), this is incorrect behavior for rman_reserve_resource_bound(). The fix is to exclude those regions where r->r_start + count - 1 > end rather than r->r_start > end. This bug has been in this code for a very long time. I believe that all other tests against end are correctly done. This is why sio0 generated a message about interrupts not being enabled properly for the device. When fdc had a bug that allocated from 0x3f7 to 0x3fb, sio0 was then given 0x3fc-0x404 rather than the 0x3f8-0x3ff that it wanted. Now when fdc has the same bug, sio0 fails to allocate its ports, which is the proper behavior. Since the probe failed, we never saw the messed up resources reported. I suspect that there are other places in the tree that have weird looping or other odd work arounds to try to cope with the observed weirdness this bug can introduce. These workarounds should be located and eliminated. Minor debug write fix to match the above test done as well. 'nice' by: mdodd Sponsored by: timing solutions (http://www.timing.com/)
2005-03-15 20:28:51 +00:00
s->r_start, end));
break;
}
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
if (s->r_start > RM_MAX_END - amask) {
DPRINTF(("s->r_start (%#jx) + amask (%#jx) too large\n",
s->r_start, amask));
break;
}
if (s->r_flags & RF_ALLOCATED) {
DPRINTF(("region is allocated\n"));
continue;
}
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
rstart = ummax(s->r_start, start);
/*
* Try to find a region by adjusting to boundary and alignment
* until both conditions are satisfied. This is not an optimal
* algorithm, but in most cases it isn't really bad, either.
*/
do {
rstart = (rstart + amask) & ~amask;
if (((rstart ^ (rstart + count - 1)) & bmask) != 0)
rstart += bound - (rstart & ~bmask);
} while ((rstart & amask) != 0 && rstart < end &&
rstart < s->r_end);
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
rend = ummin(s->r_end, ummax(rstart + count - 1, end));
if (rstart > rend) {
DPRINTF(("adjusted start exceeds end\n"));
continue;
}
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
DPRINTF(("truncated region: [%#jx, %#jx]; size %#jx (requested %#jx)\n",
rstart, rend, (rend - rstart + 1), count));
if ((rend - rstart + 1) >= count) {
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
DPRINTF(("candidate region: [%#jx, %#jx], size %#jx\n",
rstart, rend, (rend - rstart + 1)));
if ((s->r_end - s->r_start + 1) == count) {
DPRINTF(("candidate region is entire chunk\n"));
rv = s;
rv->r_flags = new_rflags;
rv->r_dev = dev;
goto out;
}
/*
* If s->r_start < rstart and
* s->r_end > rstart + count - 1, then
* we need to split the region into three pieces
* (the middle one will get returned to the user).
* Otherwise, we are allocating at either the
* beginning or the end of s, so we only need to
* split it in two. The first case requires
* two new allocations; the second requires but one.
*/
rv = int_alloc_resource(M_NOWAIT);
if (rv == NULL)
goto out;
rv->r_start = rstart;
rv->r_end = rstart + count - 1;
rv->r_flags = new_rflags;
rv->r_dev = dev;
rv->r_rm = rm;
if (s->r_start < rv->r_start && s->r_end > rv->r_end) {
DPRINTF(("splitting region in three parts: "
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
"[%#jx, %#jx]; [%#jx, %#jx]; [%#jx, %#jx]\n",
s->r_start, rv->r_start - 1,
rv->r_start, rv->r_end,
rv->r_end + 1, s->r_end));
/*
* We are allocating in the middle.
*/
r = int_alloc_resource(M_NOWAIT);
if (r == NULL) {
free(rv, M_RMAN);
rv = NULL;
goto out;
}
r->r_start = rv->r_end + 1;
r->r_end = s->r_end;
r->r_flags = s->r_flags;
r->r_rm = rm;
s->r_end = rv->r_start - 1;
TAILQ_INSERT_AFTER(&rm->rm_list, s, rv,
r_link);
TAILQ_INSERT_AFTER(&rm->rm_list, rv, r,
r_link);
} else if (s->r_start == rv->r_start) {
DPRINTF(("allocating from the beginning\n"));
/*
* We are allocating at the beginning.
*/
s->r_start = rv->r_end + 1;
TAILQ_INSERT_BEFORE(s, rv, r_link);
} else {
DPRINTF(("allocating at the end\n"));
/*
* We are allocating at the end.
*/
s->r_end = rv->r_start - 1;
TAILQ_INSERT_AFTER(&rm->rm_list, s, rv,
r_link);
}
goto out;
}
}
/*
* Now find an acceptable shared region, if the client's requirements
* allow sharing. By our implementation restriction, a candidate
* region must match exactly by both size and sharing type in order
* to be considered compatible with the client's request. (The
* former restriction could probably be lifted without too much
* additional work, but this does not seem warranted.)
*/
DPRINTF(("no unshared regions found\n"));
if ((flags & RF_SHAREABLE) == 0)
goto out;
for (s = r; s && s->r_end <= end; s = TAILQ_NEXT(s, r_link)) {
if (SHARE_TYPE(s->r_flags) == SHARE_TYPE(flags) &&
s->r_start >= start &&
(s->r_end - s->r_start + 1) == count &&
(s->r_start & amask) == 0 &&
((s->r_start ^ s->r_end) & bmask) == 0) {
rv = int_alloc_resource(M_NOWAIT);
if (rv == NULL)
goto out;
rv->r_start = s->r_start;
rv->r_end = s->r_end;
rv->r_flags = new_rflags;
rv->r_dev = dev;
rv->r_rm = rm;
if (s->r_sharehead == NULL) {
s->r_sharehead = malloc(sizeof *s->r_sharehead,
M_RMAN, M_NOWAIT | M_ZERO);
if (s->r_sharehead == NULL) {
free(rv, M_RMAN);
rv = NULL;
goto out;
}
LIST_INIT(s->r_sharehead);
LIST_INSERT_HEAD(s->r_sharehead, s,
r_sharelink);
s->r_flags |= RF_FIRSTSHARE;
}
rv->r_sharehead = s->r_sharehead;
LIST_INSERT_HEAD(s->r_sharehead, rv, r_sharelink);
goto out;
}
}
/*
* We couldn't find anything.
*/
out:
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(rm->rm_mtx);
return (rv == NULL ? NULL : &rv->r_r);
}
struct resource *
rman_reserve_resource(struct rman *rm, rman_res_t start, rman_res_t end,
rman_res_t count, u_int flags, device_t dev)
{
return (rman_reserve_resource_bound(rm, start, end, count, 0, flags,
dev));
}
int
rman_activate_resource(struct resource *re)
{
struct resource_i *r;
struct rman *rm;
r = re->__r_i;
rm = r->r_rm;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(rm->rm_mtx);
r->r_flags |= RF_ACTIVE;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(rm->rm_mtx);
return 0;
}
int
rman_deactivate_resource(struct resource *r)
{
struct rman *rm;
rm = r->__r_i->r_rm;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(rm->rm_mtx);
r->__r_i->r_flags &= ~RF_ACTIVE;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(rm->rm_mtx);
return 0;
}
static int
int_rman_release_resource(struct rman *rm, struct resource_i *r)
{
struct resource_i *s, *t;
if (r->r_flags & RF_ACTIVE)
r->r_flags &= ~RF_ACTIVE;
/*
* Check for a sharing list first. If there is one, then we don't
* have to think as hard.
*/
if (r->r_sharehead) {
/*
* If a sharing list exists, then we know there are at
* least two sharers.
*
* If we are in the main circleq, appoint someone else.
*/
LIST_REMOVE(r, r_sharelink);
s = LIST_FIRST(r->r_sharehead);
if (r->r_flags & RF_FIRSTSHARE) {
s->r_flags |= RF_FIRSTSHARE;
TAILQ_INSERT_BEFORE(r, s, r_link);
TAILQ_REMOVE(&rm->rm_list, r, r_link);
}
/*
* Make sure that the sharing list goes away completely
* if the resource is no longer being shared at all.
*/
if (LIST_NEXT(s, r_sharelink) == NULL) {
free(s->r_sharehead, M_RMAN);
s->r_sharehead = NULL;
s->r_flags &= ~RF_FIRSTSHARE;
}
goto out;
}
/*
* Look at the adjacent resources in the list and see if our
* segment can be merged with any of them. If either of the
* resources is allocated or is not exactly adjacent then they
* cannot be merged with our segment.
*/
s = TAILQ_PREV(r, resource_head, r_link);
if (s != NULL && ((s->r_flags & RF_ALLOCATED) != 0 ||
s->r_end + 1 != r->r_start))
s = NULL;
t = TAILQ_NEXT(r, r_link);
if (t != NULL && ((t->r_flags & RF_ALLOCATED) != 0 ||
r->r_end + 1 != t->r_start))
t = NULL;
if (s != NULL && t != NULL) {
/*
* Merge all three segments.
*/
s->r_end = t->r_end;
TAILQ_REMOVE(&rm->rm_list, r, r_link);
TAILQ_REMOVE(&rm->rm_list, t, r_link);
free(t, M_RMAN);
} else if (s != NULL) {
/*
* Merge previous segment with ours.
*/
s->r_end = r->r_end;
TAILQ_REMOVE(&rm->rm_list, r, r_link);
} else if (t != NULL) {
/*
* Merge next segment with ours.
*/
t->r_start = r->r_start;
TAILQ_REMOVE(&rm->rm_list, r, r_link);
} else {
/*
* At this point, we know there is nothing we
* can potentially merge with, because on each
* side, there is either nothing there or what is
* there is still allocated. In that case, we don't
* want to remove r from the list; we simply want to
* change it to an unallocated region and return
* without freeing anything.
*/
r->r_flags &= ~RF_ALLOCATED;
r->r_dev = NULL;
return 0;
}
out:
free(r, M_RMAN);
return 0;
}
int
rman_release_resource(struct resource *re)
{
int rv;
struct resource_i *r;
struct rman *rm;
r = re->__r_i;
rm = r->r_rm;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(rm->rm_mtx);
rv = int_rman_release_resource(rm, r);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(rm->rm_mtx);
return (rv);
}
uint32_t
rman_make_alignment_flags(uint32_t size)
{
int i;
/*
* Find the hightest bit set, and add one if more than one bit
* set. We're effectively computing the ceil(log2(size)) here.
*/
for (i = 31; i > 0; i--)
if ((1 << i) & size)
break;
if (~(1 << i) & size)
i++;
return(RF_ALIGNMENT_LOG2(i));
}
void
rman_set_start(struct resource *r, rman_res_t start)
{
r->__r_i->r_start = start;
}
rman_res_t
rman_get_start(struct resource *r)
{
return (r->__r_i->r_start);
}
void
rman_set_end(struct resource *r, rman_res_t end)
{
r->__r_i->r_end = end;
}
rman_res_t
rman_get_end(struct resource *r)
{
return (r->__r_i->r_end);
}
rman_res_t
rman_get_size(struct resource *r)
{
return (r->__r_i->r_end - r->__r_i->r_start + 1);
}
u_int
rman_get_flags(struct resource *r)
{
return (r->__r_i->r_flags);
}
void
rman_set_virtual(struct resource *r, void *v)
{
r->__r_i->r_virtual = v;
}
void *
rman_get_virtual(struct resource *r)
{
return (r->__r_i->r_virtual);
}
void
rman_set_bustag(struct resource *r, bus_space_tag_t t)
{
r->r_bustag = t;
}
bus_space_tag_t
rman_get_bustag(struct resource *r)
{
return (r->r_bustag);
}
void
rman_set_bushandle(struct resource *r, bus_space_handle_t h)
{
r->r_bushandle = h;
}
bus_space_handle_t
rman_get_bushandle(struct resource *r)
{
return (r->r_bushandle);
}
Add new bus methods for mapping resources. Add a pair of bus methods that can be used to "map" resources for direct CPU access using bus_space(9). bus_map_resource() creates a mapping and bus_unmap_resource() releases a previously created mapping. Mappings are described by 'struct resource_map' object. Pointers to these objects can be passed as the first argument to the bus_space wrapper API used for bus resources. Drivers that wish to map all of a resource using default settings (for example, using uncacheable memory attributes) do not need to change. However, drivers that wish to use non-default settings can now do so without jumping through hoops. First, an RF_UNMAPPED flag is added to request that a resource is not implicitly mapped with the default settings when it is activated. This permits other activation steps (such as enabling I/O or memory decoding in a device's PCI command register) to be taken without creating a mapping. Right now the AGP drivers don't set RF_ACTIVE to avoid using up a large amount of KVA to map the AGP aperture on 32-bit platforms. Once RF_UNMAPPED is supported on all platforms that support AGP this can be changed to using RF_UNMAPPED with RF_ACTIVE instead. Second, bus_map_resource accepts an optional structure that defines additional settings for a given mapping. For example, a driver can now request to map only a subset of a resource instead of the entire range. The AGP driver could also use this to only map the first page of the aperture (IIRC, it calls pmap_mapdev() directly to map the first page currently). I will also eventually change the PCI-PCI bridge driver to request mappings of the subset of the I/O window resource on its parent side to create mappings for child devices rather than passing child resources directly up to nexus to be mapped. This also permits bridges that do address translation to request suitable mappings from a resource on the "upper" side of the bus when mapping resources on the "lower" side of the bus. Another attribute that can be specified is an alternate memory attribute for memory-mapped resources. This can be used to request a Write-Combining mapping of a PCI BAR in an MI fashion. (Currently the drivers that do this call pmap_change_attr() directly for x86 only.) Note that this commit only adds the MI framework. Each platform needs to add support for handling RF_UNMAPPED and thew new bus_map/unmap_resource methods. Generally speaking, any drivers that are calling rman_set_bustag() and rman_set_bushandle() need to be updated. Discussed on: arch Reviewed by: cem Differential Revision: https://reviews.freebsd.org/D5237
2016-05-20 17:57:47 +00:00
void
rman_set_mapping(struct resource *r, struct resource_map *map)
{
KASSERT(rman_get_size(r) == map->r_size,
("rman_set_mapping: size mismatch"));
rman_set_bustag(r, map->r_bustag);
rman_set_bushandle(r, map->r_bushandle);
rman_set_virtual(r, map->r_vaddr);
}
void
rman_get_mapping(struct resource *r, struct resource_map *map)
{
map->r_bustag = rman_get_bustag(r);
map->r_bushandle = rman_get_bushandle(r);
map->r_size = rman_get_size(r);
map->r_vaddr = rman_get_virtual(r);
}
void
rman_set_rid(struct resource *r, int rid)
{
r->__r_i->r_rid = rid;
}
int
rman_get_rid(struct resource *r)
2004-07-01 16:22:10 +00:00
{
return (r->__r_i->r_rid);
2004-07-01 16:22:10 +00:00
}
void
rman_set_device(struct resource *r, device_t dev)
{
r->__r_i->r_dev = dev;
}
device_t
rman_get_device(struct resource *r)
{
return (r->__r_i->r_dev);
}
int
rman_is_region_manager(struct resource *r, struct rman *rm)
{
return (r->__r_i->r_rm == rm);
}
/*
* Sysctl interface for scanning the resource lists.
*
* We take two input parameters; the index into the list of resource
* managers, and the resource offset into the list.
*/
static int
sysctl_rman(SYSCTL_HANDLER_ARGS)
{
int *name = (int *)arg1;
u_int namelen = arg2;
int rman_idx, res_idx;
struct rman *rm;
struct resource_i *res;
struct resource_i *sres;
struct u_rman urm;
struct u_resource ures;
int error;
if (namelen != 3)
return (EINVAL);
if (bus_data_generation_check(name[0]))
return (EINVAL);
rman_idx = name[1];
res_idx = name[2];
/*
* Find the indexed resource manager
*/
mtx_lock(&rman_mtx);
TAILQ_FOREACH(rm, &rman_head, rm_link) {
if (rman_idx-- == 0)
break;
}
mtx_unlock(&rman_mtx);
if (rm == NULL)
return (ENOENT);
/*
* If the resource index is -1, we want details on the
* resource manager.
*/
if (res_idx == -1) {
bzero(&urm, sizeof(urm));
urm.rm_handle = (uintptr_t)rm;
if (rm->rm_descr != NULL)
strlcpy(urm.rm_descr, rm->rm_descr, RM_TEXTLEN);
urm.rm_start = rm->rm_start;
urm.rm_size = rm->rm_end - rm->rm_start + 1;
urm.rm_type = rm->rm_type;
error = SYSCTL_OUT(req, &urm, sizeof(urm));
return (error);
}
/*
* Find the indexed resource and return it.
*/
mtx_lock(rm->rm_mtx);
TAILQ_FOREACH(res, &rm->rm_list, r_link) {
if (res->r_sharehead != NULL) {
LIST_FOREACH(sres, res->r_sharehead, r_sharelink)
if (res_idx-- == 0) {
res = sres;
goto found;
}
}
else if (res_idx-- == 0)
goto found;
}
mtx_unlock(rm->rm_mtx);
return (ENOENT);
found:
bzero(&ures, sizeof(ures));
ures.r_handle = (uintptr_t)res;
ures.r_parent = (uintptr_t)res->r_rm;
ures.r_device = (uintptr_t)res->r_dev;
if (res->r_dev != NULL) {
if (device_get_name(res->r_dev) != NULL) {
snprintf(ures.r_devname, RM_TEXTLEN,
"%s%d",
device_get_name(res->r_dev),
device_get_unit(res->r_dev));
} else {
strlcpy(ures.r_devname, "nomatch",
RM_TEXTLEN);
}
} else {
ures.r_devname[0] = '\0';
}
ures.r_start = res->r_start;
ures.r_size = res->r_end - res->r_start + 1;
ures.r_flags = res->r_flags;
mtx_unlock(rm->rm_mtx);
error = SYSCTL_OUT(req, &ures, sizeof(ures));
return (error);
}
static SYSCTL_NODE(_hw_bus, OID_AUTO, rman, CTLFLAG_RD, sysctl_rman,
"kernel resource manager");
#ifdef DDB
static void
dump_rman_header(struct rman *rm)
{
if (db_pager_quit)
return;
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
db_printf("rman %p: %s (0x%jx-0x%jx full range)\n",
rm, rm->rm_descr, (rman_res_t)rm->rm_start, (rman_res_t)rm->rm_end);
}
static void
dump_rman(struct rman *rm)
{
struct resource_i *r;
const char *devname;
if (db_pager_quit)
return;
TAILQ_FOREACH(r, &rm->rm_list, r_link) {
if (r->r_dev != NULL) {
devname = device_get_nameunit(r->r_dev);
if (devname == NULL)
devname = "nomatch";
} else
devname = NULL;
Use uintmax_t (typedef'd to rman_res_t type) for rman ranges. On some architectures, u_long isn't large enough for resource definitions. Particularly, powerpc and arm allow 36-bit (or larger) physical addresses, but type `long' is only 32-bit. This extends rman's resources to uintmax_t. With this change, any resource can feasibly be placed anywhere in physical memory (within the constraints of the driver). Why uintmax_t and not something machine dependent, or uint64_t? Though it's possible for uintmax_t to grow, it's highly unlikely it will become 128-bit on 32-bit architectures. 64-bit architectures should have plenty of RAM to absorb the increase on resource sizes if and when this occurs, and the number of resources on memory-constrained systems should be sufficiently small as to not pose a drastic overhead. That being said, uintmax_t was chosen for source clarity. If it's specified as uint64_t, all printf()-like calls would either need casts to uintmax_t, or be littered with PRI*64 macros. Casts to uintmax_t aren't horrible, but it would also bake into the API for resource_list_print_type() either a hidden assumption that entries get cast to uintmax_t for printing, or these calls would need the PRI*64 macros. Since source code is meant to be read more often than written, I chose the clearest path of simply using uintmax_t. Tested on a PowerPC p5020-based board, which places all device resources in 0xfxxxxxxxx, and has 8GB RAM. Regression tested on qemu-system-i386 Regression tested on qemu-system-mips (malta profile) Tested PAE and devinfo on virtualbox (live CD) Special thanks to bz for his testing on ARM. Reviewed By: bz, jhb (previous) Relnotes: Yes Sponsored by: Alex Perez/Inertial Computing Differential Revision: https://reviews.freebsd.org/D4544
2016-03-18 01:28:41 +00:00
db_printf(" 0x%jx-0x%jx (RID=%d) ",
r->r_start, r->r_end, r->r_rid);
if (devname != NULL)
db_printf("(%s)\n", devname);
else
db_printf("----\n");
if (db_pager_quit)
return;
}
}
DB_SHOW_COMMAND(rman, db_show_rman)
{
if (have_addr) {
dump_rman_header((struct rman *)addr);
dump_rman((struct rman *)addr);
}
}
DB_SHOW_COMMAND(rmans, db_show_rmans)
{
struct rman *rm;
TAILQ_FOREACH(rm, &rman_head, rm_link) {
dump_rman_header(rm);
}
}
DB_SHOW_ALL_COMMAND(rman, db_show_all_rman)
{
struct rman *rm;
TAILQ_FOREACH(rm, &rman_head, rm_link) {
dump_rman_header(rm);
dump_rman(rm);
}
}
DB_SHOW_ALIAS(allrman, db_show_all_rman);
#endif