NetBSD Problem Report #33185
From yamt@mwd.biglobe.ne.jp Mon Apr 3 22:41:31 2006
Return-Path: <yamt@mwd.biglobe.ne.jp>
Received: from yamt.dyndns.org (FLA1Abg085.kng.mesh.ad.jp [219.107.209.85])
by narn.netbsd.org (Postfix) with ESMTP id BCAE163B8A7
for <gnats-bugs@gnats.NetBSD.org>; Mon, 3 Apr 2006 22:41:30 +0000 (UTC)
Message-Id: <1144104077.282970.785.nullmailer@yamt.dyndns.org>
Date: Tue, 04 Apr 2006 07:41:17 +0900
From: yamt@mwd.biglobe.ne.jp
Reply-To: yamt@mwd.biglobe.ne.jp
To: gnats-bugs@netbsd.org
Subject: kva shortage problems
X-Send-Pr-Version: 3.95
>Number: 33185
>Category: kern
>Synopsis: kva shortage problems
>Confidential: no
>Severity: serious
>Priority: medium
>Responsible: kern-bug-people
>State: closed
>Class: sw-bug
>Submitter-Id: net
>Arrival-Date: Mon Apr 03 22:45:00 +0000 2006
>Closed-Date: Wed Sep 24 10:11:31 +0000 2008
>Last-Modified: Wed Sep 24 10:11:31 +0000 2008
>Originator: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
>Release: NetBSD 3.99.16
>Organization:
>Environment:
System: NetBSD kaeru 3.99.16 NetBSD 3.99.16 (build.kaeru.xen.nodebug.work) #8: Fri Mar 24 18:41:22 JST 2006 takashi@kaeru:/home/takashi/work/kernel/build.kaeru.xen.nodebug.work i386
Architecture: i386
Machine: i386
>Description:
1. lack of kva reclamation mechanism.
http://mail-index.NetBSD.org/tech-kern/2005/12/17/0028.html
2. things like bufmem_hiwater and desiredvnodes doesn't consider
the amount of kva.
PR/31391 has an example of the symptom.
>How-To-Repeat:
>Fix:
>Release-Note:
>Audit-Trail:
From: Jason Thorpe <thorpej@shagadelic.org>
To: gnats-bugs@netbsd.org
Cc: kern-bug-people@netbsd.org, gnats-admin@netbsd.org,
netbsd-bugs@netbsd.org
Subject: Re: kern/33185: kva shortage problems
Date: Mon, 3 Apr 2006 16:26:53 -0700
On Apr 3, 2006, at 3:45 PM, yamt@mwd.biglobe.ne.jp wrote:
> 1. lack of kva reclamation mechanism.
> http://mail-index.NetBSD.org/tech-kern/2005/12/17/0028.html
> 2. things like bufmem_hiwater and desiredvnodes doesn't consider
> the amount of kva.
General lack of available KVA!
Have we thought about doing 4+4 on i386?
-- thorpej
From: christos@zoulas.com (Christos Zoulas)
To: gnats-bugs@netbsd.org, kern-bug-people@netbsd.org,
gnats-admin@netbsd.org, netbsd-bugs@netbsd.org,
yamt@mwd.biglobe.ne.jp
Cc:
Subject: Re: kern/33185: kva shortage problems
Date: Mon, 3 Apr 2006 20:40:49 -0400
On Apr 3, 11:30pm, thorpej@shagadelic.org (Jason Thorpe) wrote:
-- Subject: Re: kern/33185: kva shortage problems
| The following reply was made to PR kern/33185; it has been noted by GNATS.
|
| From: Jason Thorpe <thorpej@shagadelic.org>
| To: gnats-bugs@netbsd.org
| Cc: kern-bug-people@netbsd.org, gnats-admin@netbsd.org,
| netbsd-bugs@netbsd.org
| Subject: Re: kern/33185: kva shortage problems
| Date: Mon, 3 Apr 2006 16:26:53 -0700
|
| On Apr 3, 2006, at 3:45 PM, yamt@mwd.biglobe.ne.jp wrote:
|
| > 1. lack of kva reclamation mechanism.
| > http://mail-index.NetBSD.org/tech-kern/2005/12/17/0028.html
| > 2. things like bufmem_hiwater and desiredvnodes doesn't consider
| > the amount of kva.
|
| General lack of available KVA!
|
| Have we thought about doing 4+4 on i386?
Linux tried it and it gets tricky... Specially dealing with DMA. Plus
there is a ~10% performance penalty.
christos
From: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
To: gnats-bugs@netbsd.org
Cc: tech-kern@netbsd.org
Subject: Re: kern/33185: kva shortage problems
Date: Thu, 06 Apr 2006 20:26:42 +0900
--NextPart-20060406202103-1948600
Content-Type: Text/Plain; charset=us-ascii
> 1. lack of kva reclamation mechanism.
> http://mail-index.NetBSD.org/tech-kern/2005/12/17/0028.html
the attached diff is an attempt to fix this part.
it basically moves wait points for kva to vm_map.
comments?
YAMAMOTO Takashi
--NextPart-20060406202103-1948600
Content-Type: Text/Plain; charset=us-ascii
Content-Disposition: attachment; filename="a.diff"
Index: sys/callback.h
===================================================================
--- sys/callback.h (revision 0)
+++ sys/callback.h (revision 0)
@@ -0,0 +1,57 @@
+/* $NetBSD$ */
+
+/*-
+ * Copyright (c)2006 YAMAMOTO Takashi,
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#ifndef _SYS_CALLBACK_H_
+#define _SYS_CALLBACK_H_
+
+struct callback_entry {
+ TAILQ_ENTRY(callback_entry) ce_q;
+ int (*ce_func)(struct callback_entry *, void *, void *);
+ void *ce_obj;
+};
+
+struct callback_head {
+ struct simplelock ch_lock;
+ TAILQ_HEAD(, callback_entry) ch_q;
+ struct callback_entry *ch_next;
+ int ch_nentries;
+ int ch_running;
+ int ch_flags;
+};
+
+/* return values of ce_func */
+#define CALLBACK_CHAIN_CONTINUE 0
+#define CALLBACK_CHAIN_ABORT 1
+
+int callback_run_roundrobin(struct callback_head *, void *);
+void callback_register(struct callback_head *, struct callback_entry *,
+ void *, int (*)(struct callback_entry *, void *, void *));
+void callback_unregister(struct callback_head *, struct callback_entry *);
+void callback_head_init(struct callback_head *);
+
+#endif /* !_SYS_CALLBACK_H_ */
Index: sys/pool.h
===================================================================
--- sys/pool.h (revision 1574)
+++ sys/pool.h (working copy)
@@ -53,6 +53,9 @@
#include <sys/queue.h>
#include <sys/time.h>
#include <sys/tree.h>
+#if defined(_KERNEL)
+#include <sys/callback.h>
+#endif /* defined(_KERNEL) */
#endif
#define PCG_NOBJECTS 16
@@ -108,9 +111,17 @@ struct pool_allocator {
TAILQ_HEAD(, pool) pa_list; /* list of pools using this allocator */
int pa_flags;
#define PA_INITIALIZED 0x01
-#define PA_WANT 0x02 /* wakeup any sleeping pools on free */
int pa_pagemask;
int pa_pageshift;
+ struct vm_map *pa_backingmap;
+#if defined(_KERNEL)
+ struct {
+ struct vm_map **i_backingmapptr;
+ SLIST_ENTRY(pool_allocator) i_q;
+ } pa_init;
+#define pa_q pa_init.i_q
+#define pa_backingmapptr pa_init.i_backingmapptr
+#endif /* defined(_KERNEL) */
};
LIST_HEAD(pool_pagelist,pool_item_header);
@@ -205,6 +216,8 @@ struct pool {
const char *pr_entered_file; /* reentrancy check */
long pr_entered_line;
+
+ struct callback_entry pr_reclaimerentry;
};
#endif /* __POOL_EXPOSE */
Index: kern/subr_pool.c
===================================================================
--- kern/subr_pool.c (revision 1589)
+++ kern/subr_pool.c (working copy)
@@ -82,12 +82,16 @@ static struct pool phpool[PHPOOL_MAX];
static struct pool psppool;
#endif
+static SLIST_HEAD(, pool_allocator) pa_deferinitq =
+ SLIST_HEAD_INITIALIZER(pa_deferinitq);
+
static void *pool_page_alloc_meta(struct pool *, int);
static void pool_page_free_meta(struct pool *, void *);
/* allocator for pool metadata */
static struct pool_allocator pool_allocator_meta = {
- pool_page_alloc_meta, pool_page_free_meta
+ pool_page_alloc_meta, pool_page_free_meta,
+ .pa_backingmapptr = &kmem_map,
};
/* # of seconds to retain page after last use */
@@ -184,8 +188,8 @@ static void pool_prime_page(struct pool
static void pool_update_curpage(struct pool *);
static int pool_grow(struct pool *, int);
-void *pool_allocator_alloc(struct pool *, int);
-void pool_allocator_free(struct pool *, void *);
+static void *pool_allocator_alloc(struct pool *, int);
+static void pool_allocator_free(struct pool *, void *);
static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
void (*)(const char *, ...));
@@ -443,12 +447,106 @@ pr_rmpage(struct pool *pp, struct pool_i
pool_update_curpage(pp);
}
+static boolean_t
+pa_starved_p(struct pool_allocator *pa)
+{
+
+ if (pa->pa_backingmap != NULL) {
+ return vm_map_starved_p(pa->pa_backingmap);
+ }
+ return FALSE;
+}
+
+static int
+pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
+{
+ struct pool *pp = obj;
+ struct pool_allocator *pa = pp->pr_alloc;
+#if 1
+ unsigned long oidle = pp->pr_nidle;
+ unsigned long nidle;
+#endif
+
+ KASSERT(&pp->pr_reclaimerentry == ce);
+
+ pool_reclaim(pp);
+
+#if 1
+ nidle = pp->pr_nidle;
+ if (nidle != oidle) {
+ printf("%s: '%s' %lu -> %lu\n",
+ __func__, pp->pr_wchan, oidle, nidle);
+ }
+#endif
+
+ if (!pa_starved_p(pa)) {
+ return CALLBACK_CHAIN_ABORT;
+ }
+ return CALLBACK_CHAIN_CONTINUE;
+}
+
+static void
+pool_reclaim_register(struct pool *pp)
+{
+ struct vm_map *map = pp->pr_alloc->pa_backingmap;
+ int s;
+
+ if (map == NULL) {
+#if 1
+ if (pp->pr_alloc->pa_backingmapptr == NULL) {
+ printf("%s: pool %p '%s' doesn't have backing map\n",
+ __func__, pp, pp->pr_wchan);
+ }
+#endif
+ return;
+ }
+
+ s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
+ callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
+ &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
+ splx(s);
+}
+
+static void
+pool_reclaim_unregister(struct pool *pp)
+{
+ struct vm_map *map = pp->pr_alloc->pa_backingmap;
+ int s;
+
+ if (map == NULL) {
+ return;
+ }
+
+ s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
+ callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
+ &pp->pr_reclaimerentry);
+ splx(s);
+}
+
+static void
+pa_reclaim_register(struct pool_allocator *pa)
+{
+ struct vm_map *map = *pa->pa_backingmapptr;
+ struct pool *pp;
+
+ KASSERT(pa->pa_backingmap == NULL);
+ if (map == NULL) {
+ SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
+ return;
+ }
+ pa->pa_backingmap = map;
+ TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
+ pool_reclaim_register(pp);
+ }
+}
+
/*
* Initialize all the pools listed in the "pools" link set.
*/
void
-link_pool_init(void)
+link_pool_init(void) /* XXX rename */
{
+ struct pool_allocator *pa;
__link_set_decl(pools, struct link_pool_init);
struct link_pool_init * const *pi;
@@ -456,6 +554,14 @@ link_pool_init(void)
pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
(*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
(*pi)->palloc);
+
+ /* XXX XXX */
+ while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
+ KASSERT(pa->pa_backingmapptr != NULL);
+ KASSERT(*pa->pa_backingmapptr != NULL);
+ SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
+ pa_reclaim_register(pa);
+ }
}
/*
@@ -502,6 +608,10 @@ pool_init(struct pool *pp, size_t size,
simple_lock_init(&palloc->pa_slock);
palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
+
+ if (palloc->pa_backingmapptr != NULL) {
+ pa_reclaim_register(palloc);
+ }
palloc->pa_flags |= PA_INITIALIZED;
}
@@ -683,6 +793,7 @@ pool_init(struct pool *pp, size_t size,
TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
simple_unlock(&palloc->pa_slock);
splx(s);
+ pool_reclaim_register(pp);
}
/*
@@ -703,6 +814,7 @@ pool_destroy(struct pool *pp)
simple_unlock(&pool_head_slock);
/* Remove this pool from its allocator's list of pools. */
+ pool_reclaim_unregister(pp); /* XXX can sleep */
s = splvm();
simple_lock(&pp->pr_alloc->pa_slock);
TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
@@ -897,24 +1009,10 @@ pool_get(struct pool *pp, int flags)
if (pp->pr_curpage != NULL)
goto startover;
- if ((flags & PR_WAITOK) == 0) {
- pp->pr_nfail++;
- pr_leave(pp);
- simple_unlock(&pp->pr_slock);
- return (NULL);
- }
-
- /*
- * Wait for items to be returned to this pool.
- *
- * wake up once a second and try again,
- * as the check in pool_cache_put_paddr() is racy.
- */
- pp->pr_flags |= PR_WANTED;
- /* PA_WANTED is already set on the allocator. */
+ pp->pr_nfail++;
pr_leave(pp);
- ltsleep(pp, PSWP, pp->pr_wchan, hz, &pp->pr_slock);
- pr_enter(pp, file, line);
+ simple_unlock(&pp->pr_slock);
+ return (NULL);
}
/* Start the allocation process over. */
@@ -1114,7 +1212,7 @@ pool_do_put(struct pool *pp, void *v, st
pp->pr_nidle++;
if (pp->pr_npages > pp->pr_minpages &&
(pp->pr_npages > pp->pr_maxpages ||
- (pp->pr_alloc->pa_flags & PA_WANT) != 0)) {
+ pa_starved_p(pp->pr_alloc))) {
pr_rmpage(pp, ph, pq);
} else {
LIST_REMOVE(ph, ph_pagelist);
@@ -1483,7 +1581,8 @@ pool_reclaim(struct pool *pp)
KASSERT(ph->ph_nmissing == 0);
timersub(&curtime, &ph->ph_time, &diff);
- if (diff.tv_sec < pool_inactive_time)
+ if (diff.tv_sec < pool_inactive_time
+ && !pa_starved_p(pp->pr_alloc))
continue;
/*
@@ -2166,10 +2265,12 @@ void pool_page_free(struct pool *, void
#ifdef POOL_SUBPAGE
struct pool_allocator pool_allocator_kmem_fullpage = {
pool_page_alloc, pool_page_free, 0,
+ .pa_backingmapptr = &kmem_map,
};
#else
struct pool_allocator pool_allocator_kmem = {
pool_page_alloc, pool_page_free, 0,
+ .pa_backingmapptr = &kmem_map,
};
#endif
@@ -2179,10 +2280,12 @@ void pool_page_free_nointr(struct pool *
#ifdef POOL_SUBPAGE
struct pool_allocator pool_allocator_nointr_fullpage = {
pool_page_alloc_nointr, pool_page_free_nointr, 0,
+ .pa_backingmapptr = &kernel_map,
};
#else
struct pool_allocator pool_allocator_nointr = {
pool_page_alloc_nointr, pool_page_free_nointr, 0,
+ .pa_backingmapptr = &kernel_map,
};
#endif
@@ -2192,6 +2295,7 @@ void pool_subpage_free(struct pool *, vo
struct pool_allocator pool_allocator_kmem = {
pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
+ .pa_backingmapptr = &kmem_map,
};
void *pool_subpage_alloc_nointr(struct pool *, int);
@@ -2199,125 +2303,41 @@ void pool_subpage_free_nointr(struct poo
struct pool_allocator pool_allocator_nointr = {
pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
+ .pa_backingmapptr = &kmem_map,
};
#endif /* POOL_SUBPAGE */
-/*
- * We have at least three different resources for the same allocation and
- * each resource can be depleted. First, we have the ready elements in the
- * pool. Then we have the resource (typically a vm_map) for this allocator.
- * Finally, we have physical memory. Waiting for any of these can be
- * unnecessary when any other is freed, but the kernel doesn't support
- * sleeping on multiple wait channels, so we have to employ another strategy.
- *
- * The caller sleeps on the pool (so that it can be awakened when an item
- * is returned to the pool), but we set PA_WANT on the allocator. When a
- * page is returned to the allocator and PA_WANT is set, pool_allocator_free
- * will wake up all sleeping pools belonging to this allocator.
- *
- * XXX Thundering herd.
- */
-void *
-pool_allocator_alloc(struct pool *org, int flags)
+static void *
+pool_allocator_alloc(struct pool *pp, int flags)
{
- struct pool_allocator *pa = org->pr_alloc;
- struct pool *pp, *start;
- int s, freed;
+ struct pool_allocator *pa = pp->pr_alloc;
void *res;
- LOCK_ASSERT(!simple_lock_held(&org->pr_slock));
+ LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
- do {
- if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
- return (res);
- if ((flags & PR_WAITOK) == 0) {
- /*
- * We only run the drain hookhere if PR_NOWAIT.
- * In other cases, the hook will be run in
- * pool_reclaim().
- */
- if (org->pr_drain_hook != NULL) {
- (*org->pr_drain_hook)(org->pr_drain_hook_arg,
- flags);
- if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
- return (res);
- }
- break;
- }
-
+ res = (*pa->pa_alloc)(pp, flags);
+ if (res == NULL && (flags & PR_WAITOK) == 0) {
/*
- * Drain all pools, that use this allocator.
- * We do this to reclaim VA space.
- * pa_alloc is responsible for waiting for
- * physical memory.
- *
- * XXX We risk looping forever if start if someone
- * calls pool_destroy on "start". But there is no
- * other way to have potentially sleeping pool_reclaim,
- * non-sleeping locks on pool_allocator, and some
- * stirring of drained pools in the allocator.
- *
- * XXX Maybe we should use pool_head_slock for locking
- * the allocators?
+ * We only run the drain hook here if PR_NOWAIT.
+ * In other cases, the hook will be run in
+ * pool_reclaim().
*/
- freed = 0;
-
- s = splvm();
- simple_lock(&pa->pa_slock);
- pp = start = TAILQ_FIRST(&pa->pa_list);
- do {
- TAILQ_REMOVE(&pa->pa_list, pp, pr_alloc_list);
- TAILQ_INSERT_TAIL(&pa->pa_list, pp, pr_alloc_list);
- simple_unlock(&pa->pa_slock);
- freed = pool_reclaim(pp);
- simple_lock(&pa->pa_slock);
- } while ((pp = TAILQ_FIRST(&pa->pa_list)) != start &&
- freed == 0);
-
- if (freed == 0) {
- /*
- * We set PA_WANT here, the caller will most likely
- * sleep waiting for pages (if not, this won't hurt
- * that much), and there is no way to set this in
- * the caller without violating locking order.
- */
- pa->pa_flags |= PA_WANT;
+ if (pp->pr_drain_hook != NULL) {
+ (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
+ res = (*pa->pa_alloc)(pp, flags);
}
- simple_unlock(&pa->pa_slock);
- splx(s);
- } while (freed);
- return (NULL);
+ }
+ return res;
}
-void
+static void
pool_allocator_free(struct pool *pp, void *v)
{
struct pool_allocator *pa = pp->pr_alloc;
- int s;
LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
(*pa->pa_free)(pp, v);
-
- s = splvm();
- simple_lock(&pa->pa_slock);
- if ((pa->pa_flags & PA_WANT) == 0) {
- simple_unlock(&pa->pa_slock);
- splx(s);
- return;
- }
-
- TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
- simple_lock(&pp->pr_slock);
- if ((pp->pr_flags & PR_WANTED) != 0) {
- pp->pr_flags &= ~PR_WANTED;
- wakeup(pp);
- }
- simple_unlock(&pp->pr_slock);
- }
- pa->pa_flags &= ~PA_WANT;
- simple_unlock(&pa->pa_slock);
- splx(s);
}
void *
Index: kern/vfs_bio.c
===================================================================
--- kern/vfs_bio.c (revision 1587)
+++ kern/vfs_bio.c (working copy)
@@ -177,9 +177,9 @@ struct simplelock bqueue_slock = SIMPLEL
/*
* Buffer pool for I/O buffers.
- * Access to this pool must be protected with splbio().
*/
-static POOL_INIT(bufpool, sizeof(struct buf), 0, 0, 0, "bufpl", NULL);
+static POOL_INIT(bufpool, sizeof(struct buf), 0, 0, 0, "bufpl",
+ &pool_allocator_nointr);
/* XXX - somewhat gross.. */
@@ -375,8 +375,7 @@ bufinit(void)
if (bufmem_valimit != 0) {
vaddr_t minaddr = 0, maxaddr;
buf_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
- bufmem_valimit, VM_MAP_PAGEABLE,
- FALSE, 0);
+ bufmem_valimit, 0, FALSE, 0);
if (buf_map == NULL)
panic("bufinit: cannot allocate submap");
} else
@@ -393,6 +392,7 @@ bufinit(void)
use_std = 1;
#endif
+ bufmempool_allocator.pa_backingmap = buf_map;
for (i = 0; i < NMEMPOOLS; i++) {
struct pool_allocator *pa;
struct pool *pp = &bmempools[i];
@@ -985,13 +985,13 @@ already_queued:
/* Allow disk interrupts. */
simple_unlock(&bp->b_interlock);
simple_unlock(&bqueue_slock);
+ splx(s);
if (bp->b_bufsize <= 0) {
#ifdef DEBUG
memset((char *)bp, 0, sizeof(*bp));
#endif
pool_put(&bufpool, bp);
}
- splx(s);
}
/*
Index: kern/uipc_socket.c
===================================================================
--- kern/uipc_socket.c (revision 1590)
+++ kern/uipc_socket.c (working copy)
@@ -126,16 +126,8 @@ EVCNT_ATTACH_STATIC(sosend_kvalimit);
#endif /* SOSEND_COUNTERS */
-void
-soinit(void)
-{
+static struct callback_entry sokva_reclaimerentry;
- /* Set the initial adjusted socket buffer size. */
- if (sb_max_set(sb_max))
- panic("bad initial sb_max value: %lu", sb_max);
-
-}
-
#ifdef SOSEND_NO_LOAN
int use_sosend_loan = 0;
#else
@@ -438,6 +430,32 @@ sosend_loan(struct socket *so, struct ui
}
return (space);
+}
+
+static int
+sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
+{
+
+ KASSERT(ce == &sokva_reclaimerentry);
+ KASSERT(obj == NULL);
+
+ sodopendfree();
+ if (!vm_map_starved_p(kernel_map)) {
+ return CALLBACK_CHAIN_ABORT;
+ }
+ return CALLBACK_CHAIN_CONTINUE;
+}
+
+void
+soinit(void)
+{
+
+ /* Set the initial adjusted socket buffer size. */
+ if (sb_max_set(sb_max))
+ panic("bad initial sb_max value: %lu", sb_max);
+
+ callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
+ &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
}
/*
Index: kern/uipc_mbuf.c
===================================================================
--- kern/uipc_mbuf.c (revision 1585)
+++ kern/uipc_mbuf.c (working copy)
@@ -154,6 +154,7 @@ mbinit(void)
KASSERT(sizeof(struct _m_ext) <= MHLEN);
KASSERT(sizeof(struct mbuf) == MSIZE);
+ mclpool_allocator.pa_backingmap = mb_map;
pool_init(&mbpool, msize, 0, 0, 0, "mbpl", NULL);
pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", &mclpool_allocator);
Index: kern/subr_callback.c
===================================================================
--- kern/subr_callback.c (revision 0)
+++ kern/subr_callback.c (revision 0)
@@ -0,0 +1,139 @@
+/* $NetBSD$ */
+
+/*-
+ * Copyright (c)2006 YAMAMOTO Takashi,
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <sys/cdefs.h>
+__KERNEL_RCSID(0, "$NetBSD$");
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/proc.h>
+#include <sys/callback.h>
+
+#define CH_WANT 1
+
+void
+callback_head_init(struct callback_head *ch)
+{
+
+ simple_lock_init(&ch->ch_lock);
+ TAILQ_INIT(&ch->ch_q);
+ ch->ch_next = NULL;
+ ch->ch_nentries = 0;
+}
+
+void
+callback_register(struct callback_head *ch, struct callback_entry *ce,
+ void *obj, int (*fn)(struct callback_entry *, void *, void *))
+{
+
+ ce->ce_func = fn;
+ ce->ce_obj = obj;
+ simple_lock(&ch->ch_lock);
+ TAILQ_INSERT_TAIL(&ch->ch_q, ce, ce_q);
+ ch->ch_nentries++;
+ simple_unlock(&ch->ch_lock);
+}
+
+void
+callback_unregister(struct callback_head *ch, struct callback_entry *ce)
+{
+
+ simple_lock(&ch->ch_lock);
+ while (ch->ch_running > 0) {
+ ch->ch_flags |= CH_WANT;
+ ltsleep(&ch->ch_running, PVM, "recunreg", 0, &ch->ch_lock);
+ }
+ if (__predict_false(ch->ch_next == ce)) {
+ ch->ch_next = TAILQ_NEXT(ce, ce_q);
+ }
+ TAILQ_REMOVE(&ch->ch_q, ce, ce_q);
+ ch->ch_nentries--;
+ simple_unlock(&ch->ch_lock);
+}
+
+static int
+callback_runone(struct callback_head *ch, void *arg)
+{
+ struct callback_entry *ce;
+ int result;
+
+ KASSERT(ch->ch_nentries > 0);
+ KASSERT(ch->ch_running > 0);
+
+ ce = ch->ch_next;
+ if (ce == NULL) {
+ ce = TAILQ_FIRST(&ch->ch_q);
+ }
+ KASSERT(ce != NULL);
+ result = (*ce->ce_func)(ce, ce->ce_obj, arg);
+ ch->ch_next = TAILQ_NEXT(ce, ce_q);
+ return result;
+}
+
+static void
+callback_run_enter(struct callback_head *ch)
+{
+
+ simple_lock(&ch->ch_lock);
+ ch->ch_running++;
+ simple_unlock(&ch->ch_lock);
+}
+
+static void
+callback_run_leave(struct callback_head *ch)
+{
+
+ simple_lock(&ch->ch_lock);
+ KASSERT(ch->ch_running > 0);
+ ch->ch_running--;
+ if (ch->ch_running == 0 && (ch->ch_flags & CH_WANT) != 0) {
+ ch->ch_flags &= ~CH_WANT;
+ wakeup(&ch->ch_running);
+ }
+ simple_unlock(&ch->ch_lock);
+}
+
+int
+callback_run_roundrobin(struct callback_head *ch, void *arg)
+{
+ int i;
+ int n;
+ int result = 0;
+
+ callback_run_enter(ch);
+ n = ch->ch_nentries;
+ for (i = 0; i < n; i++) {
+ result = callback_runone(ch, arg);
+ if (result != CALLBACK_CHAIN_CONTINUE) {
+ break;
+ }
+ }
+ callback_run_leave(ch);
+
+ return result;
+}
Index: uvm/uvm_km.c
===================================================================
--- uvm/uvm_km.c (revision 1591)
+++ uvm/uvm_km.c (working copy)
@@ -188,7 +188,8 @@ km_vacache_alloc(struct pool *pp, int fl
if (uvm_map(map, &va, size, NULL, UVM_UNKNOWN_OFFSET, size,
UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
UVM_ADV_RANDOM, UVM_FLAG_QUANTUM |
- ((flags & PR_WAITOK) ? 0 : UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT))))
+ ((flags & PR_WAITOK) ? UVM_FLAG_WAITVA :
+ UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT))))
return NULL;
return (void *)va;
@@ -226,10 +227,9 @@ km_vacache_init(struct vm_map *map, cons
pa->pa_alloc = km_vacache_alloc;
pa->pa_free = km_vacache_free;
pa->pa_pagesz = (unsigned int)size;
+ pa->pa_backingmap = map;
+ pa->pa_backingmapptr = NULL;
pool_init(pp, PAGE_SIZE, 0, 0, PR_NOTOUCH | PR_RECURSIVE, name, pa);
-
- /* XXX for now.. */
- pool_sethiwat(pp, 0);
}
void
@@ -252,6 +252,30 @@ uvm_km_vacache_init(struct vm_map *map,
}
#endif /* !defined(PMAP_MAP_POOLPAGE) */
+
+void
+uvm_km_va_drain(struct vm_map *map, uvm_flag_t flags)
+{
+ struct vm_map_kernel *vmk = vm_map_to_kernel(map);
+ const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
+#if 0
+ const int rflags =
+ (flags & (UVM_FLAG_NOWAIT|UVM_FLAG_WAITVA)) == UVM_FLAG_WAITVA ?
+ 0 : RECLAIM_FLAG_NOWAIT;
+ struct reclaim_args args = {
+ .ra_flags = rflags,
+ };
+#endif
+ int s = 0xdeadbeaf; /* XXX: gcc */
+
+ if (intrsafe) {
+ s = splvm();
+ }
+ callback_run_roundrobin(&vmk->vmk_reclaim_callback, NULL);
+ if (intrsafe) {
+ splx(s);
+ }
+}
/*
* uvm_km_init: init kernel maps and objects to reflect reality (i.e.
Index: uvm/uvm_km.h
===================================================================
--- uvm/uvm_km.h (revision 1464)
+++ uvm/uvm_km.h (working copy)
@@ -55,6 +55,7 @@ void uvm_km_check_empty(vaddr_t, vaddr_t
#else
#define uvm_km_check_empty(a, b, c) /* nothing */
#endif /* defined(DEBUG) */
+void uvm_km_va_drain(struct vm_map *, uvm_flag_t);
#endif /* _KERNEL */
Index: uvm/uvm_map.c
===================================================================
--- uvm/uvm_map.c (revision 1587)
+++ uvm/uvm_map.c (working copy)
@@ -742,7 +742,17 @@ uvm_map_clip_end(struct vm_map *map, str
uvm_tree_sanity(map, "clip_end leave");
}
+static void
+vm_map_drain(struct vm_map *map, uvm_flag_t flags)
+{
+ if (!VM_MAP_IS_KERNEL(map)) {
+ return;
+ }
+
+ uvm_km_va_drain(map, flags);
+}
+
/*
* M A P - m a i n e n t r y p o i n t
*/
@@ -875,16 +885,11 @@ retry:
}
vm_map_lock(map); /* could sleep here */
}
- if ((prev_entry = uvm_map_findspace(map, start, size, &start,
- uobj, uoffset, align, flags)) == NULL) {
+ prev_entry = uvm_map_findspace(map, start, size, &start,
+ uobj, uoffset, align, flags);
+ if (prev_entry == NULL) {
unsigned int timestamp;
- if ((flags & UVM_FLAG_WAITVA) == 0) {
- UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",
- 0,0,0,0);
- vm_map_unlock(map);
- return ENOMEM;
- }
timestamp = map->timestamp;
UVMHIST_LOG(maphist,"waiting va timestamp=0x%x",
timestamp,0,0,0);
@@ -894,15 +899,24 @@ retry:
vm_map_unlock(map);
/*
- * wait until someone does unmap.
+ * try to reclaim kva and wait until someone does unmap.
* XXX fragile locking
*/
+ vm_map_drain(map, flags);
+
simple_lock(&map->flags_lock);
while ((map->flags & VM_MAP_WANTVA) != 0 &&
map->timestamp == timestamp) {
- ltsleep(&map->header, PVM, "vmmapva", 0,
- &map->flags_lock);
+ if ((flags & UVM_FLAG_WAITVA) == 0) {
+ simple_unlock(&map->flags_lock);
+ UVMHIST_LOG(maphist,
+ "<- uvm_map_findspace failed!", 0,0,0,0);
+ return ENOMEM;
+ } else {
+ ltsleep(&map->header, PVM, "vmmapva", 0,
+ &map->flags_lock);
+ }
}
simple_unlock(&map->flags_lock);
goto retry;
@@ -2655,6 +2669,7 @@ uvm_map_setup_kernel(struct vm_map_kerne
uvm_map_setup(&map->vmk_map, vmin, vmax, flags);
+ callback_head_init(&map->vmk_reclaim_callback);
LIST_INIT(&map->vmk_kentry_free);
map->vmk_merged_entries = NULL;
}
@@ -4789,4 +4804,18 @@ vm_map_to_kernel(struct vm_map *map)
KASSERT(VM_MAP_IS_KERNEL(map));
return (struct vm_map_kernel *)map;
+}
+
+boolean_t
+vm_map_starved_p(struct vm_map *map)
+{
+
+ if ((map->flags & VM_MAP_WANTVA) != 0) {
+ return TRUE;
+ }
+ /* XXX */
+ if ((vm_map_max(map) - vm_map_min(map)) / 16 * 15 < map->size) {
+ return TRUE;
+ }
+ return FALSE;
}
Index: uvm/uvm_map.h
===================================================================
--- uvm/uvm_map.h (revision 1571)
+++ uvm/uvm_map.h (working copy)
@@ -234,6 +234,9 @@ struct vm_map {
};
#if defined(_KERNEL)
+
+#include <sys/callback.h>
+
struct vm_map_kernel {
struct vm_map vmk_map;
LIST_HEAD(, uvm_kmapent_hdr) vmk_kentry_free;
@@ -241,6 +244,7 @@ struct vm_map_kernel {
struct vm_map_entry *vmk_merged_entries;
/* Merged entries, kept for later splitting */
+ struct callback_head vmk_reclaim_callback;
#if !defined(PMAP_MAP_POOLPAGE)
struct pool vmk_vacache; /* kva cache */
struct pool_allocator vmk_vacache_allocator; /* ... and its allocator */
@@ -506,6 +510,8 @@ do { \
if (oflags & VM_MAP_WANTLOCK) \
wakeup(&(map)->flags); \
} while (/*CONSTCOND*/ 0)
+
+boolean_t vm_map_starved_p(struct vm_map *);
#endif /* _KERNEL */
--NextPart-20060406202103-1948600--
From: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
To: gnats-bugs@netbsd.org
Cc: tech-kern@netbsd.org
Subject: Re: kern/33185: kva shortage problems
Date: Sat, 22 Apr 2006 16:00:34 +0900
> > 1. lack of kva reclamation mechanism.
> > http://mail-index.NetBSD.org/tech-kern/2005/12/17/0028.html
>
> the attached diff is an attempt to fix this part.
> it basically moves wait points for kva to vm_map.
>
> comments?
no one interested? :-)
a concern is kva fragmentation.
YAMAMOTO Takashi
From: Izumi Tsutsui <tsutsui@ceres.dti.ne.jp>
To: gnats-bugs@NetBSD.org
Cc: tech-kern@NetBSD.org, tsutsui@ceres.dti.ne.jp
Subject: Re: kern/33185: kva shortage problems
Date: Sun, 23 Apr 2006 06:02:49 +0900
In article <1145689234.922880.10618.nullmailer@yamt.dyndns.org>
yamt@mwd.biglobe.ne.jp wrote:
> a concern is kva fragmentation.
I'm not a VM expert, but "kva fragmentation" reminds me
STE shortage on 68040 and sun3x pmaps...
---
Izumi Tsutsui
From: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
To: tsutsui@ceres.dti.ne.jp
Cc: gnats-bugs@NetBSD.org, tech-kern@NetBSD.org
Subject: Re: kern/33185: kva shortage problems
Date: Thu, 11 May 2006 07:50:02 +0900
> > a concern is kva fragmentation.
>
> I'm not a VM expert, but "kva fragmentation" reminds me
> STE shortage on 68040 and sun3x pmaps...
> ---
> Izumi Tsutsui
i'm not sure what's "STE shortage". can you explain with MI terms? :-)
i don't think my patch makes kva fragmentation much worse.
my concern is, it can be sufferred from fragmentation.
ie. it assumes, if we have enough reclaimable kva regions,
freeing them yields some free kva at vm_map layer. it isn't true if
kva is fragmented much, given "slab" layers between.
now, as i don't think it's more problematic than the current situation,
i'll just go ahead unless anyone objects.
YAMAMOTO Takashi
From: Izumi Tsutsui <tsutsui@ceres.dti.ne.jp>
To: yamt@mwd.biglobe.ne.jp
Cc: gnats-bugs@NetBSD.org, tech-kern@NetBSD.org,
tsutsui@ceres.dti.ne.jp
Subject: Re: kern/33185: kva shortage problems
Date: Thu, 11 May 2006 20:08:44 +0900
yamt@mwd.biglobe.ne.jp wrote:
> > > a concern is kva fragmentation.
> >
> > I'm not a VM expert, but "kva fragmentation" reminds me
> > STE shortage on 68040 and sun3x pmaps...
> i'm not sure what's "STE shortage". can you explain with MI terms? :-)
Well, sun3x uses 7-6-6-13 bits 3-level MMU, and memory
for each Segment Table Entry is allocated statically.
Recently (for a year?) I often got "get_x_tabe: out of X tables."
panics on my 3/80 and I just thought it was caused by kva
fragmentation.
(yes, just sun3x pmap should be fixed, but it's not easy...)
> i don't think my patch makes kva fragmentation much worse.
Ok, then there is no problem on your patch.
Thank you for your explanation.
---
Izumi Tsutsui
From: YAMAMOTO Takashi <yamt@netbsd.org>
To: gnats-bugs@NetBSD.org
Cc:
Subject: PR/33185 CVS commit: src/sys
Date: Thu, 25 May 2006 14:27:28 +0000 (UTC)
Module Name: src
Committed By: yamt
Date: Thu May 25 14:27:28 UTC 2006
Modified Files:
src/sys/conf: files
src/sys/kern: subr_pool.c uipc_mbuf.c uipc_socket.c vfs_bio.c
src/sys/sys: pool.h
src/sys/uvm: uvm_init.c uvm_km.c uvm_km.h uvm_map.c uvm_map.h
Added Files:
src/sys/kern: subr_callback.c
src/sys/sys: callback.h
Log Message:
move wait points for kva from upper layers to vm_map. PR/33185 #1.
XXX there is a concern about interaction with kva fragmentation.
see: http://mail-index.NetBSD.org/tech-kern/2006/05/11/0000.html
To generate a diff of this commit:
cvs rdiff -r1.775 -r1.776 src/sys/conf/files
cvs rdiff -r0 -r1.1 src/sys/kern/subr_callback.c
cvs rdiff -r1.116 -r1.117 src/sys/kern/subr_pool.c
cvs rdiff -r1.110 -r1.111 src/sys/kern/uipc_mbuf.c
cvs rdiff -r1.118 -r1.119 src/sys/kern/uipc_socket.c
cvs rdiff -r1.160 -r1.161 src/sys/kern/vfs_bio.c
cvs rdiff -r0 -r1.1 src/sys/sys/callback.h
cvs rdiff -r1.48 -r1.49 src/sys/sys/pool.h
cvs rdiff -r1.24 -r1.25 src/sys/uvm/uvm_init.c
cvs rdiff -r1.87 -r1.88 src/sys/uvm/uvm_km.c
cvs rdiff -r1.15 -r1.16 src/sys/uvm/uvm_km.h
cvs rdiff -r1.225 -r1.226 src/sys/uvm/uvm_map.c
cvs rdiff -r1.53 -r1.54 src/sys/uvm/uvm_map.h
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
From: YAMAMOTO Takashi <yamt@netbsd.org>
To: gnats-bugs@gnats.NetBSD.org
Cc:
Subject: PR/33185 CVS commit: src/sys
Date: Sun, 23 Mar 2008 10:39:52 +0000 (UTC)
Module Name: src
Committed By: yamt
Date: Sun Mar 23 10:39:52 UTC 2008
Modified Files:
src/sys/kern: init_main.c vfs_bio.c
src/sys/sys: systm.h
Log Message:
when calculating some cache sizes, consider the amount of available kva.
PR/33185.
To generate a diff of this commit:
cvs rdiff -r1.344 -r1.345 src/sys/kern/init_main.c
cvs rdiff -r1.191 -r1.192 src/sys/kern/vfs_bio.c
cvs rdiff -r1.216 -r1.217 src/sys/sys/systm.h
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
From: YAMAMOTO Takashi <yamt@netbsd.org>
To: gnats-bugs@gnats.NetBSD.org
Cc:
Subject: PR/33185 CVS commit: src/sys/rump/librump/rumpkern
Date: Sun, 23 Mar 2008 14:12:39 +0000 (UTC)
Module Name: src
Committed By: yamt
Date: Sun Mar 23 14:12:39 UTC 2008
Modified Files:
src/sys/rump/librump/rumpkern: emul.c
Log Message:
when calculating some cache sizes, consider the amount of available kva.
PR/33185.
To generate a diff of this commit:
cvs rdiff -r1.33 -r1.34 src/sys/rump/librump/rumpkern/emul.c
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
From: Markus W Kilbinger <mk@kilbi.de>
To: gnats-bugs@NetBSD.org
Cc: kern-bug-people@netbsd.org,
gnats-admin@netbsd.org,
netbsd-bugs@netbsd.org,
yamt@mwd.biglobe.ne.jp
Subject: Re: PR/33185 CVS commit: src/sys
Date: Mon, 24 Mar 2008 09:13:15 +0100
Hi!
Your
http://mail-index.netbsd.org/source-changes/2008/03/23/msg003851.html
seems to break kernels on/for my cobalt qube2. After loading the
kernel it takes several seconds to 'display' a lot of random chars on
its serial console and then the qube reboots.
Just reverting this changes makes this problem disappear.
So it's difficult for me to provide more detailed info about the
problem...
Markus.
From: yamt@mwd.biglobe.ne.jp (YAMAMOTO Takashi)
To: mk@kilbi.de
Cc: gnats-bugs@NetBSD.org, kern-bug-people@netbsd.org, gnats-admin@netbsd.org,
netbsd-bugs@netbsd.org
Subject: Re: PR/33185 CVS commit: src/sys
Date: Mon, 24 Mar 2008 17:30:18 +0900 (JST)
> Hi!
>
> Your
>
> http://mail-index.netbsd.org/source-changes/2008/03/23/msg003851.html
>
> seems to break kernels on/for my cobalt qube2. After loading the
> kernel it takes several seconds to 'display' a lot of random chars on
> its serial console and then the qube reboots.
>
> Just reverting this changes makes this problem disappear.
>
> So it's difficult for me to provide more detailed info about the
> problem...
>
> Markus.
i have no idea right now.
have you tried DEBUG/DIAGNOSTIC/LOCKDEBUG/etc?
YAMAMOTO Takashi
From: Markus W Kilbinger <mk@kilbi.de>
To: yamt@mwd.biglobe.ne.jp (YAMAMOTO Takashi)
Cc: mk@kilbi.de,
gnats-bugs@NetBSD.org,
kern-bug-people@netbsd.org,
gnats-admin@netbsd.org,
netbsd-bugs@netbsd.org
Subject: Re: PR/33185 CVS commit: src/sys
Date: Mon, 24 Mar 2008 11:43:35 +0100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From: Markus W Kilbinger <mk@kilbi.de>
To: gnats-bugs@NetBSD.org
Cc: kern-bug-people@netbsd.org,
gnats-admin@netbsd.org,
netbsd-bugs@netbsd.org,
yamt@mwd.biglobe.ne.jp
Subject: Re: PR/33185 CVS commit: src/sys
Date: Mon, 24 Mar 2008 15:50:24 +0100
Hi again!
> have you tried DEBUG/DIAGNOSTIC/LOCKDEBUG/etc?
A kernel with all 3 options activated is not really more informative:
>> NetBSD/cobalt 4.99.55 Bootloader, Revision 0.8 [@0x80f00000]
>> (root@qie, Fri Mar 21 12:06:58 MET 2008)
>> Model: Cobalt Qube 2
>> Memory: 262144 k
>> PROM boot string: root=/dev/hda1 ro
Boot [wd0a:netbsd]:
Loading: wd0a:netbsd
3491760+412384[193728+185026]=0x415ef8
Starting at 0x80001000
Loaded initial symtab at 0x803b9290, strtab at 0x803e8c34, # entries 11460
[lots of nonsens chars]
Simply reverting your changes in
http://mail-index.netbsd.org/source-changes/2008/03/23/msg003851.html
makes the kernel behave normally again.
Hm, what else I can test?
Markus.
From: Izumi Tsutsui <tsutsui@ceres.dti.ne.jp>
To: yamt@mwd.biglobe.ne.jp
Cc: mk@kilbi.de, gnats-bugs@NetBSD.org, kern-bug-people@NetBSD.org,
gnats-admin@NetBSD.org, netbsd-bugs@NetBSD.org,
tsutsui@ceres.dti.ne.jp
Subject: Re: PR/33185 CVS commit: src/sys
Date: Tue, 25 Mar 2008 00:47:43 +0900
My cobalt kernel with sprinkled printf seems dying at:
mach_init() -> pmap_bootstrap() -> buf_memcalc() -> calc_cache_size()
and buf_map passed to calc_cache_size() seems NULL at that point.
buf_memcalc() should be deferred on mips?
---
Izumi Tsutsui
From: yamt@mwd.biglobe.ne.jp (YAMAMOTO Takashi)
To: tsutsui@ceres.dti.ne.jp
Cc: mk@kilbi.de, gnats-bugs@NetBSD.org, kern-bug-people@NetBSD.org,
gnats-admin@NetBSD.org, netbsd-bugs@NetBSD.org
Subject: Re: PR/33185 CVS commit: src/sys
Date: Tue, 25 Mar 2008 08:15:03 +0900 (JST)
> My cobalt kernel with sprinkled printf seems dying at:
> mach_init() -> pmap_bootstrap() -> buf_memcalc() -> calc_cache_size()
> and buf_map passed to calc_cache_size() seems NULL at that point.
> buf_memcalc() should be deferred on mips?
> ---
> Izumi Tsutsui
oops, i see. i'll fix it this evening.
YAMAMOTO Takashi
State-Changed-From-To: open->feedback
State-Changed-By: yamt@NetBSD.org
State-Changed-When: Tue, 08 Apr 2008 23:01:58 +0000
State-Changed-Why:
can anyone test -current?
State-Changed-From-To: feedback->closed
State-Changed-By: ad@NetBSD.org
State-Changed-When: Wed, 24 Sep 2008 10:11:31 +0000
State-Changed-Why:
1. original problem is fixed.
2. desiredvnodes has been retarded further.
3. feedback timeout on cobalt issue.
>Unformatted:
(Contact us)
$NetBSD: query-full-pr,v 1.39 2013/11/01 18:47:49 spz Exp $
$NetBSD: gnats_config.sh,v 1.8 2006/05/07 09:23:38 tsutsui Exp $
Copyright © 1994-2007
The NetBSD Foundation, Inc. ALL RIGHTS RESERVED.
Home