FreeBSD Bugzilla – Attachment 146423 Details for
Bug 187594
[zfs] [patch] ZFS ARC behavior problem and fix
Home
|
New
|
Browse
|
Search
|
[?]
|
Reports
|
Help
|
New Account
|
Log In
Remember
[x]
|
Forgot Password
Login:
[x]
[patch]
arc reclaim refactor (against head)
arc-reclaim-head.patch (text/plain), 7.33 KB, created by
Steven Hartland
on 2014-08-27 23:15:07 UTC
(
hide
)
Description:
arc reclaim refactor (against head)
Filename:
MIME Type:
Creator:
Steven Hartland
Created:
2014-08-27 23:15:07 UTC
Size:
7.33 KB
patch
obsolete
>Index: sys/cddl/compat/opensolaris/kern/opensolaris_kmem.c >=================================================================== >--- sys/cddl/compat/opensolaris/kern/opensolaris_kmem.c (revision 270422) >+++ sys/cddl/compat/opensolaris/kern/opensolaris_kmem.c (working copy) >@@ -126,18 +126,47 @@ kmem_size_init(void *unused __unused) > } > SYSINIT(kmem_size_init, SI_SUB_KMEM, SI_ORDER_ANY, kmem_size_init, NULL); > >-uint64_t >-kmem_size(void) >+/* >+ * The return values from kmem_free_* are only valid once the pagedaemon >+ * has been initialised, before then they return 0. >+ * >+ * To ensure the returns are valid the caller can use a SYSINIT with >+ * subsystem set to SI_SUB_KTHREAD_PAGE and an order of at least >+ * SI_ORDER_SECOND. >+ */ >+u_int >+kmem_free_target(void) > { > >- return (kmem_size_val); >+ return (vm_cnt.v_free_target); > } > >+u_int >+kmem_free_min(void) >+{ >+ >+ return (vm_cnt.v_free_min); >+} >+ >+u_int >+kmem_free_count(void) >+{ >+ >+ return (vm_cnt.v_free_count); >+} >+ >+u_int >+kmem_page_count(void) >+{ >+ >+ return (vm_cnt.v_page_count); >+} >+ > uint64_t >-kmem_used(void) >+kmem_size(void) > { > >- return (vmem_size(kmem_arena, VMEM_ALLOC)); >+ return (kmem_size_val); > } > > static int >Index: sys/cddl/compat/opensolaris/sys/kmem.h >=================================================================== >--- sys/cddl/compat/opensolaris/sys/kmem.h (revision 270422) >+++ sys/cddl/compat/opensolaris/sys/kmem.h (working copy) >@@ -66,7 +66,16 @@ typedef struct kmem_cache { > void *zfs_kmem_alloc(size_t size, int kmflags); > void zfs_kmem_free(void *buf, size_t size); > uint64_t kmem_size(void); >-uint64_t kmem_used(void); >+u_int kmem_page_count(void); >+ >+/* >+ * The return values from kmem_free_* are only valid once the pagedaemon >+ * has been initialised, before then they return 0. >+ */ >+u_int kmem_free_count(void); >+u_int kmem_free_target(void); >+u_int kmem_free_min(void); >+ > kmem_cache_t *kmem_cache_create(char *name, size_t bufsize, size_t align, > int (*constructor)(void *, void *, int), void (*destructor)(void *, void *), > void (*reclaim)(void *) __unused, void *private, vmem_t *vmp, int cflags); >Index: sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c >=================================================================== >--- sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c (revision 270422) >+++ sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c (working copy) >@@ -193,9 +193,6 @@ extern int zfs_prefetch_disable; > */ > static boolean_t arc_warm; > >-/* >- * These tunables are for performance analysis. >- */ > uint64_t zfs_arc_max; > uint64_t zfs_arc_min; > uint64_t zfs_arc_meta_limit = 0; >@@ -204,7 +201,21 @@ int zfs_arc_shrink_shift = 0; > int zfs_arc_p_min_shift = 0; > int zfs_disable_dup_eviction = 0; > uint64_t zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ >+u_int zfs_arc_free_target = (1 << 19); /* default before pagedaemon init only */ > >+static int sysctl_vfs_zfs_arc_free_target(SYSCTL_HANDLER_ARGS); >+ >+#ifdef _KERNEL >+static void >+arc_free_target_init(void *unused __unused) >+{ >+ >+ zfs_arc_free_target = kmem_free_target(); >+} >+SYSINIT(arc_free_target_init, SI_SUB_KTHREAD_PAGE, SI_ORDER_ANY, >+ arc_free_target_init, NULL); >+#endif >+ > TUNABLE_QUAD("vfs.zfs.arc_meta_limit", &zfs_arc_meta_limit); > SYSCTL_DECL(_vfs_zfs); > SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, arc_max, CTLFLAG_RDTUN, &zfs_arc_max, 0, >@@ -214,7 +225,36 @@ SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, arc_min, CTLFLAG_ > SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, arc_average_blocksize, CTLFLAG_RDTUN, > &zfs_arc_average_blocksize, 0, > "ARC average blocksize"); >+/* >+ * We don't have a tunable for arc_free_target due to the dependency on >+ * pagedaemon initialisation. >+ */ >+SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_free_target, >+ CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(u_int), >+ sysctl_vfs_zfs_arc_free_target, "IU", >+ "Desired number of free pages below which ARC triggers reclaim"); > >+static int >+sysctl_vfs_zfs_arc_free_target(SYSCTL_HANDLER_ARGS) >+{ >+ u_int val; >+ int err; >+ >+ val = zfs_arc_free_target; >+ err = sysctl_handle_int(oidp, &val, 0, req); >+ if (err != 0 || req->newptr == NULL) >+ return (err); >+ >+ if (val < kmem_free_min()) >+ return (EINVAL); >+ if (val > kmem_page_count()) >+ return (EINVAL); >+ >+ zfs_arc_free_target = val; >+ >+ return (0); >+} >+ > /* > * Note that buffers can be in one of 6 states: > * ARC_anon - anonymous (discussed below) >@@ -2418,9 +2458,12 @@ arc_flush(spa_t *spa) > void > arc_shrink(void) > { >+ > if (arc_c > arc_c_min) { > uint64_t to_free; > >+ DTRACE_PROBE2(arc__shrink, uint64_t, arc_c, uint64_t, >+ arc_c_min); > #ifdef _KERNEL > to_free = arc_c >> arc_shrink_shift; > #else >@@ -2440,8 +2483,11 @@ arc_shrink(void) > ASSERT((int64_t)arc_p >= 0); > } > >- if (arc_size > arc_c) >+ if (arc_size > arc_c) { >+ DTRACE_PROBE2(arc__shrink_adjust, uint64_t, arc_size, >+ uint64_t, arc_c); > arc_adjust(); >+ } > } > > static int needfree = 0; >@@ -2452,15 +2498,25 @@ arc_reclaim_needed(void) > > #ifdef _KERNEL > >- if (needfree) >+ if (needfree) { >+ DTRACE_PROBE(arc__reclaim_needfree); > return (1); >+ } > >+ if (kmem_free_count() < zfs_arc_free_target) { >+ DTRACE_PROBE2(arc__reclaim_freetarget, uint64_t, >+ kmem_free_count(), uint64_t, zfs_arc_free_target); >+ return (1); >+ } >+ > /* > * Cooperate with pagedaemon when it's time for it to scan > * and reclaim some pages. > */ >- if (vm_paging_needed()) >+ if (vm_paging_needed()) { >+ DTRACE_PROBE(arc__reclaim_paging); > return (1); >+ } > > #ifdef sun > /* >@@ -2504,9 +2560,6 @@ arc_reclaim_needed(void) > (btop(vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC)) >> 2)) > return (1); > #endif >-#else /* !sun */ >- if (kmem_used() > (kmem_size() * 3) / 4) >- return (1); > #endif /* sun */ > > #else >@@ -2513,6 +2566,8 @@ arc_reclaim_needed(void) > if (spa_get_random(100) == 0) > return (1); > #endif >+ DTRACE_PROBE(arc__reclaim_no); >+ > return (0); > } > >Index: sys/vm/vm_pageout.c >=================================================================== >--- sys/vm/vm_pageout.c (revision 270422) >+++ sys/vm/vm_pageout.c (working copy) >@@ -115,10 +115,14 @@ __FBSDID("$FreeBSD$"); > > /* the kernel process "vm_pageout"*/ > static void vm_pageout(void); >+static void vm_pageout_init(void); > static int vm_pageout_clean(vm_page_t); > static void vm_pageout_scan(struct vm_domain *vmd, int pass); > static void vm_pageout_mightbe_oom(struct vm_domain *vmd, int pass); > >+SYSINIT(pagedaemon_init, SI_SUB_KTHREAD_PAGE, SI_ORDER_FIRST, vm_pageout_init, >+ NULL); >+ > struct proc *pageproc; > > static struct kproc_desc page_kp = { >@@ -126,7 +130,7 @@ static struct kproc_desc page_kp = { > vm_pageout, > &pageproc > }; >-SYSINIT(pagedaemon, SI_SUB_KTHREAD_PAGE, SI_ORDER_FIRST, kproc_start, >+SYSINIT(pagedaemon, SI_SUB_KTHREAD_PAGE, SI_ORDER_SECOND, kproc_start, > &page_kp); > > #if !defined(NO_SWAPPING) >@@ -1637,15 +1641,11 @@ vm_pageout_worker(void *arg) > } > > /* >- * vm_pageout is the high level pageout daemon. >+ * vm_pageout_init initialises basic pageout daemon settings. > */ > static void >-vm_pageout(void) >+vm_pageout_init(void) > { >-#if MAXMEMDOM > 1 >- int error, i; >-#endif >- > /* > * Initialize some paging parameters. > */ >@@ -1691,7 +1691,18 @@ static void > /* XXX does not really belong here */ > if (vm_page_max_wired == 0) > vm_page_max_wired = vm_cnt.v_free_count / 3; >+} > >+/* >+ * vm_pageout is the high level pageout daemon. >+ */ >+static void >+vm_pageout(void) >+{ >+#if MAXMEMDOM > 1 >+ int error, i; >+#endif >+ > swap_pager_swap_init(); > #if MAXMEMDOM > 1 > for (i = 1; i < vm_ndomains; i++) {
You cannot view the attachment while viewing its details because your browser does not support IFRAMEs.
View the attachment on a separate page
.
View Attachment As Diff
View Attachment As Raw
Actions:
View
|
Diff
Attachments on
bug 187594
:
140882
|
140883
|
140884
|
140885
|
140886
|
140887
|
140888
|
140889
|
140890
|
140891
|
140892
|
146178
|
146203
|
146249
|
146251
|
146287
|
146300
|
146373
|
146423
|
146424
|
146456
|
146816
|
146817
|
146851
|
146852
|
146854
|
146859
|
146861
|
146946
|
146947
|
146948
|
146949
|
147014
|
147068
|
147069
|
147070
|
147265
|
147274
|
147275
|
147276
|
147286
|
147459
|
147607
|
147609
|
147733
|
147738
|
147754
|
147815
|
152852
|
158809
|
159207
|
159688
|
159859
|
159905
|
161691
|
161692
|
161943
|
164051
|
174197
|
174198
|
174231
|
174232
|
174254
|
186818