svn commit: r347532 - in head: contrib/netbsd-tests/lib/libc/sys lib/libc/sys lib/libc/tests/sys sys/amd64/vmm sys/sys sys/vm usr.bin/vmstat
Rodney W. Grimes
freebsd at gndrsh.dnsmgr.net
Mon May 13 17:35:13 UTC 2019
> Author: markj
> Date: Mon May 13 16:38:48 2019
> New Revision: 347532
> URL: https://svnweb.freebsd.org/changeset/base/347532
>
> Log:
> Provide separate accounting for user-wired pages.
>
> Historically we have not distinguished between kernel wirings and user
> wirings for accounting purposes. User wirings (via mlock(2)) were
> subject to a global limit on the number of wired pages, so if large
> swaths of physical memory were wired by the kernel, as happens with
> the ZFS ARC among other things, the limit could be exceeded, causing
> user wirings to fail.
>
> The change adds a new counter, v_user_wire_count, which counts the
> number of virtual pages wired by user processes via mlock(2) and
> mlockall(2). Only user-wired pages are subject to the system-wide
> limit which helps provide some safety against deadlocks. In
> particular, while sources of kernel wirings typically support some
> backpressure mechanism, there is no way to reclaim user-wired pages
> shorting of killing the wiring process. The limit is exported as
> vm.max_user_wired, renamed from vm.max_wired, and changed from u_int
> to u_long.
>
> The choice to count virtual user-wired pages rather than physical
> pages was done for simplicity. There are mechanisms that can cause
> user-wired mappings to be destroyed while maintaining a wiring of
> the backing physical page; these make it difficult to accurately
> track user wirings at the physical page layer.
>
> The change also closes some holes which allowed user wirings to succeed
> even when they would cause the system limit to be exceeded. For
> instance, mmap() may now fail with ENOMEM in a process that has called
> mlockall(MCL_FUTURE) if the new mapping would cause the user wiring
> limit to be exceeded.
>
> Note that bhyve -S is subject to the user wiring limit, which defaults
> to 1/3 of physical RAM. Users that wish to exceed the limit must tune
> vm.max_user_wired.
Because of that this should probably have a:
Release Notes: Yes
> Reviewed by: kib, ngie (mlock() test changes)
> Tested by: pho (earlier version)
> MFC after: 45 days
> Sponsored by: Netflix
> Differential Revision: https://reviews.freebsd.org/D19908
>
> Modified:
> head/contrib/netbsd-tests/lib/libc/sys/t_mlock.c
> head/lib/libc/sys/mlock.2
> head/lib/libc/sys/mlockall.2
> head/lib/libc/tests/sys/mlock_helper.c
> head/sys/amd64/vmm/vmm.c
> head/sys/sys/vmmeter.h
> head/sys/vm/vm_glue.c
> head/sys/vm/vm_map.c
> head/sys/vm/vm_map.h
> head/sys/vm/vm_meter.c
> head/sys/vm/vm_mmap.c
> head/sys/vm/vm_pageout.c
> head/sys/vm/vm_pageout.h
> head/sys/vm/vm_unix.c
> head/usr.bin/vmstat/vmstat.c
>
> Modified: head/contrib/netbsd-tests/lib/libc/sys/t_mlock.c
> ==============================================================================
> --- head/contrib/netbsd-tests/lib/libc/sys/t_mlock.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/contrib/netbsd-tests/lib/libc/sys/t_mlock.c Mon May 13 16:38:48 2019 (r347532)
> @@ -51,7 +51,7 @@ __RCSID("$NetBSD: t_mlock.c,v 1.6 2016/08/09 12:02:44
> #define _KMEMUSER
> #include <machine/vmparam.h>
>
> -void set_vm_max_wired(int);
> +void set_vm_max_wired(u_long);
> void restore_vm_max_wired(void);
> #endif
>
>
> Modified: head/lib/libc/sys/mlock.2
> ==============================================================================
> --- head/lib/libc/sys/mlock.2 Mon May 13 15:39:54 2019 (r347531)
> +++ head/lib/libc/sys/mlock.2 Mon May 13 16:38:48 2019 (r347532)
> @@ -28,7 +28,7 @@
> .\" @(#)mlock.2 8.2 (Berkeley) 12/11/93
> .\" $FreeBSD$
> .\"
> -.Dd March 20, 2018
> +.Dd May 13, 2019
> .Dt MLOCK 2
> .Os
> .Sh NAME
> @@ -97,13 +97,13 @@ resource limit and the
> system-wide
> .Dq wired pages
> limit
> -.Va vm.max_wired .
> -.Va vm.max_wired
> +.Va vm.max_user_wired .
> +.Va vm.max_user_wired
> applies to the system as a whole, so the amount available to a single
> process at any given time is the difference between
> -.Va vm.max_wired
> +.Va vm.max_user_wired
> and
> -.Va vm.stats.vm.v_wire_count .
> +.Va vm.stats.vm.v_user_wire_count .
> .Pp
> If
> .Va security.bsd.unprivileged_mlock
> @@ -124,13 +124,11 @@ will fail if:
> is set to 0 and the caller is not the super-user.
> .It Bq Er EINVAL
> The address range given wraps around zero.
> -.It Bq Er EAGAIN
> -Locking the indicated range would exceed the system limit for locked memory.
> .It Bq Er ENOMEM
> Some portion of the indicated address range is not allocated.
> There was an error faulting/mapping a page.
> -Locking the indicated range would exceed the per-process limit for locked
> -memory.
> +Locking the indicated range would exceed the per-process or system-wide limits
> +for locked memory.
> .El
> The
> .Fn munlock
> @@ -171,11 +169,11 @@ system calls first appeared in
> Allocating too much wired memory can lead to a memory-allocation deadlock
> which requires a reboot to recover from.
> .Pp
> -The per-process resource limit is a limit on the amount of virtual
> -memory locked, while the system-wide limit is for the number of locked
> -physical pages.
> -Hence a process with two distinct locked mappings of the same physical page
> -counts as 2 pages against the per-process limit and as only a single page
> -in the system limit.
> +The per-process and system-wide resource limits of locked memory apply
> +to the amount of virtual memory locked, not the amount of locked physical
> +pages.
> +Hence two distinct locked mappings of the same physical page counts as
> +2 pages aginst the system limit, and also against the per-process limit
> +if both mappings belong to the same physical map.
> .Pp
> The per-process resource limit is not currently supported.
>
> Modified: head/lib/libc/sys/mlockall.2
> ==============================================================================
> --- head/lib/libc/sys/mlockall.2 Mon May 13 15:39:54 2019 (r347531)
> +++ head/lib/libc/sys/mlockall.2 Mon May 13 16:38:48 2019 (r347532)
> @@ -30,7 +30,7 @@
> .\"
> .\" $FreeBSD$
> .\"
> -.Dd December 25, 2012
> +.Dd May 13, 2019
> .Dt MLOCKALL 2
> .Os
> .Sh NAME
> @@ -69,7 +69,7 @@ limited in how much they can lock down.
> A single process can lock the minimum of a system-wide
> .Dq wired pages
> limit
> -.Va vm.max_wired
> +.Va vm.max_user_wired
> and the per-process
> .Dv RLIMIT_MEMLOCK
> resource limit.
> @@ -138,9 +138,9 @@ and
> functions first appeared in
> .Fx 5.1 .
> .Sh BUGS
> -The per-process resource limit is a limit on the amount of virtual
> -memory locked, while the system-wide limit is for the number of locked
> -physical pages.
> -Hence a process with two distinct locked mappings of the same physical page
> -counts as 2 pages against the per-process limit and as only a single page
> -in the system limit.
> +The per-process and system-wide resource limits of locked memory apply
> +to the amount of virtual memory locked, not the amount of locked physical
> +pages.
> +Hence two distinct locked mappings of the same physical page counts as
> +2 pages aginst the system limit, and also against the per-process limit
> +if both mappings belong to the same physical map.
>
> Modified: head/lib/libc/tests/sys/mlock_helper.c
> ==============================================================================
> --- head/lib/libc/tests/sys/mlock_helper.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/lib/libc/tests/sys/mlock_helper.c Mon May 13 16:38:48 2019 (r347532)
> @@ -39,16 +39,16 @@ __FBSDID("$FreeBSD$");
> #include <limits.h>
> #include <stdio.h>
>
> -#define VM_MAX_WIRED "vm.max_wired"
> +#define VM_MAX_WIRED "vm.max_user_wired"
>
> static void
> -vm_max_wired_sysctl(int *old_value, int *new_value)
> +vm_max_wired_sysctl(u_long *old_value, u_long *new_value)
> {
> size_t old_len;
> - size_t new_len = (new_value == NULL ? 0 : sizeof(int));
> + size_t new_len = (new_value == NULL ? 0 : sizeof(*new_value));
>
> if (old_value == NULL)
> - printf("Setting the new value to %d\n", *new_value);
> + printf("Setting the new value to %lu\n", *new_value);
> else {
> ATF_REQUIRE_MSG(sysctlbyname(VM_MAX_WIRED, NULL, &old_len,
> new_value, new_len) == 0,
> @@ -60,14 +60,14 @@ vm_max_wired_sysctl(int *old_value, int *new_value)
> "sysctlbyname(%s) failed: %s", VM_MAX_WIRED, strerror(errno));
>
> if (old_value != NULL)
> - printf("Saved the old value (%d)\n", *old_value);
> + printf("Saved the old value (%lu)\n", *old_value);
> }
>
> void
> -set_vm_max_wired(int new_value)
> +set_vm_max_wired(u_long new_value)
> {
> FILE *fp;
> - int old_value;
> + u_long old_value;
>
> fp = fopen(VM_MAX_WIRED, "w");
> if (fp == NULL) {
> @@ -78,7 +78,7 @@ set_vm_max_wired(int new_value)
>
> vm_max_wired_sysctl(&old_value, NULL);
>
> - ATF_REQUIRE_MSG(fprintf(fp, "%d", old_value) > 0,
> + ATF_REQUIRE_MSG(fprintf(fp, "%lu", old_value) > 0,
> "saving %s failed", VM_MAX_WIRED);
>
> fclose(fp);
> @@ -90,7 +90,7 @@ void
> restore_vm_max_wired(void)
> {
> FILE *fp;
> - int saved_max_wired;
> + u_long saved_max_wired;
>
> fp = fopen(VM_MAX_WIRED, "r");
> if (fp == NULL) {
> @@ -98,14 +98,14 @@ restore_vm_max_wired(void)
> return;
> }
>
> - if (fscanf(fp, "%d", &saved_max_wired) != 1) {
> + if (fscanf(fp, "%lu", &saved_max_wired) != 1) {
> perror("fscanf failed\n");
> fclose(fp);
> return;
> }
>
> fclose(fp);
> - printf("old value in %s: %d\n", VM_MAX_WIRED, saved_max_wired);
> + printf("old value in %s: %lu\n", VM_MAX_WIRED, saved_max_wired);
>
> if (saved_max_wired == 0) /* This will cripple the test host */
> return;
>
> Modified: head/sys/amd64/vmm/vmm.c
> ==============================================================================
> --- head/sys/amd64/vmm/vmm.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/amd64/vmm/vmm.c Mon May 13 16:38:48 2019 (r347532)
> @@ -754,7 +754,8 @@ vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segi
> VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
> if (error != KERN_SUCCESS) {
> vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
> - return (EFAULT);
> + return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM :
> + EFAULT);
> }
> }
>
>
> Modified: head/sys/sys/vmmeter.h
> ==============================================================================
> --- head/sys/sys/vmmeter.h Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/sys/vmmeter.h Mon May 13 16:38:48 2019 (r347532)
> @@ -153,6 +153,8 @@ extern domainset_t vm_severe_domains;
> #define VM_CNT_INC(var) VM_CNT_ADD(var, 1)
> #define VM_CNT_FETCH(var) counter_u64_fetch(vm_cnt.var)
>
> +extern u_long vm_user_wire_count;
> +
> static inline void
> vm_wire_add(int cnt)
> {
>
> Modified: head/sys/vm/vm_glue.c
> ==============================================================================
> --- head/sys/vm/vm_glue.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/vm/vm_glue.c Mon May 13 16:38:48 2019 (r347532)
> @@ -181,21 +181,8 @@ vslock(void *addr, size_t len)
> if (last < (vm_offset_t)addr || end < (vm_offset_t)addr)
> return (EINVAL);
> npages = atop(end - start);
> - if (npages > vm_page_max_wired)
> + if (npages > vm_page_max_user_wired)
> return (ENOMEM);
> -#if 0
> - /*
> - * XXX - not yet
> - *
> - * The limit for transient usage of wired pages should be
> - * larger than for "permanent" wired pages (mlock()).
> - *
> - * Also, the sysctl code, which is the only present user
> - * of vslock(), does a hard loop on EAGAIN.
> - */
> - if (npages + vm_wire_count() > vm_page_max_wired)
> - return (EAGAIN);
> -#endif
> error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end,
> VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
> if (error == KERN_SUCCESS) {
>
> Modified: head/sys/vm/vm_map.c
> ==============================================================================
> --- head/sys/vm/vm_map.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/vm/vm_map.c Mon May 13 16:38:48 2019 (r347532)
> @@ -90,6 +90,7 @@ __FBSDID("$FreeBSD$");
> #include <vm/pmap.h>
> #include <vm/vm_map.h>
> #include <vm/vm_page.h>
> +#include <vm/vm_pageout.h>
> #include <vm/vm_object.h>
> #include <vm/vm_pager.h>
> #include <vm/vm_kern.h>
> @@ -2917,12 +2918,12 @@ done:
>
> if (rv == KERN_SUCCESS && (!user_unwire ||
> (entry->eflags & MAP_ENTRY_USER_WIRED))) {
> - if (user_unwire)
> - entry->eflags &= ~MAP_ENTRY_USER_WIRED;
> if (entry->wired_count == 1)
> vm_map_entry_unwire(map, entry);
> else
> entry->wired_count--;
> + if (user_unwire)
> + entry->eflags &= ~MAP_ENTRY_USER_WIRED;
> }
> KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
> ("vm_map_unwire: in-transition flag missing %p", entry));
> @@ -2942,6 +2943,28 @@ done:
> return (rv);
> }
>
> +static void
> +vm_map_wire_user_count_sub(u_long npages)
> +{
> +
> + atomic_subtract_long(&vm_user_wire_count, npages);
> +}
> +
> +static bool
> +vm_map_wire_user_count_add(u_long npages)
> +{
> + u_long wired;
> +
> + wired = vm_user_wire_count;
> + do {
> + if (npages + wired > vm_page_max_user_wired)
> + return (false);
> + } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
> + npages + wired));
> +
> + return (true);
> +}
> +
> /*
> * vm_map_wire_entry_failure:
> *
> @@ -2978,37 +3001,49 @@ vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t
> entry->wired_count = -1;
> }
>
> +int
> +vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
> +{
> + int rv;
> +
> + vm_map_lock(map);
> + rv = vm_map_wire_locked(map, start, end, flags);
> + vm_map_unlock(map);
> + return (rv);
> +}
> +
> +
> /*
> - * vm_map_wire:
> + * vm_map_wire_locked:
> *
> - * Implements both kernel and user wiring.
> + * Implements both kernel and user wiring. Returns with the map locked,
> + * the map lock may be dropped.
> */
> int
> -vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
> - int flags)
> +vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
> {
> vm_map_entry_t entry, first_entry, tmp_entry;
> vm_offset_t faddr, saved_end, saved_start;
> - unsigned int last_timestamp;
> + u_long npages;
> + u_int last_timestamp;
> int rv;
> boolean_t need_wakeup, result, user_wire;
> vm_prot_t prot;
>
> + VM_MAP_ASSERT_LOCKED(map);
> +
> if (start == end)
> return (KERN_SUCCESS);
> prot = 0;
> if (flags & VM_MAP_WIRE_WRITE)
> prot |= VM_PROT_WRITE;
> user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
> - vm_map_lock(map);
> VM_MAP_RANGE_CHECK(map, start, end);
> if (!vm_map_lookup_entry(map, start, &first_entry)) {
> if (flags & VM_MAP_WIRE_HOLESOK)
> first_entry = first_entry->next;
> - else {
> - vm_map_unlock(map);
> + else
> return (KERN_INVALID_ADDRESS);
> - }
> }
> last_timestamp = map->timestamp;
> entry = first_entry;
> @@ -3042,7 +3077,6 @@ vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset
> /*
> * first_entry has been deleted.
> */
> - vm_map_unlock(map);
> return (KERN_INVALID_ADDRESS);
> }
> end = saved_start;
> @@ -3082,13 +3116,22 @@ vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset
> }
> if (entry->wired_count == 0) {
> entry->wired_count++;
> - saved_start = entry->start;
> - saved_end = entry->end;
>
> + npages = atop(entry->end - entry->start);
> + if (user_wire && !vm_map_wire_user_count_add(npages)) {
> + vm_map_wire_entry_failure(map, entry,
> + entry->start);
> + end = entry->end;
> + rv = KERN_RESOURCE_SHORTAGE;
> + goto done;
> + }
> +
> /*
> * Release the map lock, relying on the in-transition
> * mark. Mark the map busy for fork.
> */
> + saved_start = entry->start;
> + saved_end = entry->end;
> vm_map_busy(map);
> vm_map_unlock(map);
>
> @@ -3136,6 +3179,8 @@ vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset
> last_timestamp = map->timestamp;
> if (rv != KERN_SUCCESS) {
> vm_map_wire_entry_failure(map, entry, faddr);
> + if (user_wire)
> + vm_map_wire_user_count_sub(npages);
> end = entry->end;
> goto done;
> }
> @@ -3201,9 +3246,12 @@ done:
> * Undo the wiring. Wiring succeeded on this entry
> * but failed on a later entry.
> */
> - if (entry->wired_count == 1)
> + if (entry->wired_count == 1) {
> vm_map_entry_unwire(map, entry);
> - else
> + if (user_wire)
> + vm_map_wire_user_count_sub(
> + atop(entry->end - entry->start));
> + } else
> entry->wired_count--;
> }
> next_entry_done:
> @@ -3220,7 +3268,6 @@ done:
> }
> vm_map_simplify_entry(map, entry);
> }
> - vm_map_unlock(map);
> if (need_wakeup)
> vm_map_wakeup(map);
> return (rv);
> @@ -3338,13 +3385,18 @@ vm_map_sync(
> static void
> vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
> {
> + vm_size_t size;
>
> VM_MAP_ASSERT_LOCKED(map);
> KASSERT(entry->wired_count > 0,
> ("vm_map_entry_unwire: entry %p isn't wired", entry));
> +
> + size = entry->end - entry->start;
> + if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
> + vm_map_wire_user_count_sub(atop(size));
> pmap_unwire(map->pmap, entry->start, entry->end);
> - vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
> - entry->start, PQ_ACTIVE);
> + vm_object_unwire(entry->object.vm_object, entry->offset, size,
> + PQ_ACTIVE);
> entry->wired_count = 0;
> }
>
> @@ -4311,12 +4363,11 @@ retry:
> * Heed the MAP_WIREFUTURE flag if it was set for this process.
> */
> if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
> - vm_map_unlock(map);
> - vm_map_wire(map, grow_start, grow_start + grow_amount,
> + rv = vm_map_wire_locked(map, grow_start,
> + grow_start + grow_amount,
> VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
> - vm_map_lock_read(map);
> - } else
> - vm_map_lock_downgrade(map);
> + }
> + vm_map_lock_downgrade(map);
>
> out:
> #ifdef RACCT
>
> Modified: head/sys/vm/vm_map.h
> ==============================================================================
> --- head/sys/vm/vm_map.h Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/vm/vm_map.h Mon May 13 16:38:48 2019 (r347532)
> @@ -422,7 +422,8 @@ int vm_map_madvise (vm_map_t, vm_offset_t, vm_offset_t
> int vm_map_stack (vm_map_t, vm_offset_t, vm_size_t, vm_prot_t, vm_prot_t, int);
> int vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
> int flags);
> -int vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
> +int vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags);
> +int vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end,
> int flags);
> long vmspace_swap_count(struct vmspace *vmspace);
> void vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add);
>
> Modified: head/sys/vm/vm_meter.c
> ==============================================================================
> --- head/sys/vm/vm_meter.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/vm/vm_meter.c Mon May 13 16:38:48 2019 (r347532)
> @@ -97,6 +97,8 @@ struct vmmeter __read_mostly vm_cnt = {
> .v_wire_count = EARLY_COUNTER,
> };
>
> +u_long __exclusive_cache_line vm_user_wire_count;
> +
> static void
> vmcounter_startup(void)
> {
> @@ -394,6 +396,8 @@ sysctl_handle_vmstat_proc(SYSCTL_HANDLER_ARGS)
>
> #define VM_STATS_UINT(var, descr) \
> SYSCTL_UINT(_vm_stats_vm, OID_AUTO, var, CTLFLAG_RD, &vm_cnt.var, 0, descr)
> +#define VM_STATS_ULONG(var, descr) \
> + SYSCTL_ULONG(_vm_stats_vm, OID_AUTO, var, CTLFLAG_RD, &vm_cnt.var, 0, descr)
>
> VM_STATS_UINT(v_page_size, "Page size in bytes");
> VM_STATS_UINT(v_page_count, "Total number of pages in system");
> @@ -410,6 +414,9 @@ VM_STATS_PROC(v_laundry_count, "Pages eligible for lau
> VM_STATS_UINT(v_pageout_free_min, "Min pages reserved for kernel");
> VM_STATS_UINT(v_interrupt_free_min, "Reserved pages for interrupt code");
> VM_STATS_UINT(v_free_severe, "Severe page depletion point");
> +
> +SYSCTL_ULONG(_vm_stats_vm, OID_AUTO, v_user_wire_count, CTLFLAG_RD,
> + &vm_user_wire_count, 0, "User-wired virtual memory");
>
> #ifdef COMPAT_FREEBSD11
> /*
>
> Modified: head/sys/vm/vm_mmap.c
> ==============================================================================
> --- head/sys/vm/vm_mmap.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/vm/vm_mmap.c Mon May 13 16:38:48 2019 (r347532)
> @@ -1003,7 +1003,7 @@ kern_mlock(struct proc *proc, struct ucred *cred, uint
> if (last < addr || end < addr)
> return (EINVAL);
> npages = atop(end - start);
> - if (npages > vm_page_max_wired)
> + if (npages > vm_page_max_user_wired)
> return (ENOMEM);
> map = &proc->p_vmspace->vm_map;
> PROC_LOCK(proc);
> @@ -1013,8 +1013,6 @@ kern_mlock(struct proc *proc, struct ucred *cred, uint
> return (ENOMEM);
> }
> PROC_UNLOCK(proc);
> - if (npages + vm_wire_count() > vm_page_max_wired)
> - return (EAGAIN);
> #ifdef RACCT
> if (racct_enable) {
> PROC_LOCK(proc);
> @@ -1091,7 +1089,12 @@ sys_mlockall(struct thread *td, struct mlockall_args *
> */
> error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
> VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
> - error = (error == KERN_SUCCESS ? 0 : EAGAIN);
> + if (error == KERN_SUCCESS)
> + error = 0;
> + else if (error == KERN_RESOURCE_SHORTAGE)
> + error = ENOMEM;
> + else
> + error = EAGAIN;
> }
> #ifdef RACCT
> if (racct_enable && error != KERN_SUCCESS) {
> @@ -1558,10 +1561,16 @@ vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_siz
> * If the process has requested that all future mappings
> * be wired, then heed this.
> */
> - if (map->flags & MAP_WIREFUTURE) {
> - vm_map_wire(map, *addr, *addr + size,
> - VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
> - VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
> + if ((map->flags & MAP_WIREFUTURE) != 0) {
> + vm_map_lock(map);
> + if ((map->flags & MAP_WIREFUTURE) != 0)
> + rv = vm_map_wire_locked(map, *addr,
> + *addr + size, VM_MAP_WIRE_USER |
> + ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
> + VM_MAP_WIRE_NOHOLES));
> + if (rv != KERN_SUCCESS)
> + (void)vm_map_delete(map, *addr, *addr + size);
> + vm_map_unlock(map);
> }
> }
> return (vm_mmap_to_errno(rv));
>
> Modified: head/sys/vm/vm_pageout.c
> ==============================================================================
> --- head/sys/vm/vm_pageout.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/vm/vm_pageout.c Mon May 13 16:38:48 2019 (r347532)
> @@ -194,9 +194,10 @@ SYSCTL_UINT(_vm, OID_AUTO, background_launder_max, CTL
>
> int vm_pageout_page_count = 32;
>
> -int vm_page_max_wired; /* XXX max # of wired pages system-wide */
> -SYSCTL_INT(_vm, OID_AUTO, max_wired,
> - CTLFLAG_RW, &vm_page_max_wired, 0, "System-wide limit to wired page count");
> +u_long vm_page_max_user_wired;
> +SYSCTL_ULONG(_vm, OID_AUTO, max_user_wired, CTLFLAG_RW,
> + &vm_page_max_user_wired, 0,
> + "system-wide limit to user-wired page count");
>
> static u_int isqrt(u_int num);
> static int vm_pageout_launder(struct vm_domain *vmd, int launder,
> @@ -2041,8 +2042,8 @@ vm_pageout_init(void)
> if (vm_pageout_update_period == 0)
> vm_pageout_update_period = 600;
>
> - if (vm_page_max_wired == 0)
> - vm_page_max_wired = freecount / 3;
> + if (vm_page_max_user_wired == 0)
> + vm_page_max_user_wired = freecount / 3;
> }
>
> /*
>
> Modified: head/sys/vm/vm_pageout.h
> ==============================================================================
> --- head/sys/vm/vm_pageout.h Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/vm/vm_pageout.h Mon May 13 16:38:48 2019 (r347532)
> @@ -75,7 +75,7 @@
> * Exported data structures.
> */
>
> -extern int vm_page_max_wired;
> +extern u_long vm_page_max_user_wired;
> extern int vm_pageout_page_count;
>
> #define VM_OOM_MEM 1
>
> Modified: head/sys/vm/vm_unix.c
> ==============================================================================
> --- head/sys/vm/vm_unix.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/sys/vm/vm_unix.c Mon May 13 16:38:48 2019 (r347532)
> @@ -95,13 +95,11 @@ kern_break(struct thread *td, uintptr_t *addr)
> rlim_t datalim, lmemlim, vmemlim;
> int prot, rv;
> int error = 0;
> - boolean_t do_map_wirefuture;
>
> datalim = lim_cur(td, RLIMIT_DATA);
> lmemlim = lim_cur(td, RLIMIT_MEMLOCK);
> vmemlim = lim_cur(td, RLIMIT_VMEM);
>
> - do_map_wirefuture = FALSE;
> new = round_page(*addr);
> vm_map_lock(map);
>
> @@ -184,7 +182,14 @@ kern_break(struct thread *td, uintptr_t *addr)
> if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32))
> prot |= VM_PROT_EXECUTE;
> #endif
> - rv = vm_map_insert(map, NULL, 0, old, new, prot, VM_PROT_ALL, 0);
> + rv = vm_map_insert(map, NULL, 0, old, new, prot, VM_PROT_ALL,
> + 0);
> + if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
> + rv = vm_map_wire_locked(map, old, new,
> + VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
> + if (rv != KERN_SUCCESS)
> + vm_map_delete(map, old, new);
> + }
> if (rv != KERN_SUCCESS) {
> #ifdef RACCT
> if (racct_enable) {
> @@ -205,17 +210,6 @@ kern_break(struct thread *td, uintptr_t *addr)
> goto done;
> }
> vm->vm_dsize += btoc(new - old);
> - /*
> - * Handle the MAP_WIREFUTURE case for legacy applications,
> - * by marking the newly mapped range of pages as wired.
> - * We are not required to perform a corresponding
> - * vm_map_unwire() before vm_map_delete() below, as
> - * it will forcibly unwire the pages in the range.
> - *
> - * XXX If the pages cannot be wired, no error is returned.
> - */
> - if ((map->flags & MAP_WIREFUTURE) == MAP_WIREFUTURE)
> - do_map_wirefuture = TRUE;
> } else if (new < old) {
> rv = vm_map_delete(map, new, old);
> if (rv != KERN_SUCCESS) {
> @@ -238,10 +232,6 @@ kern_break(struct thread *td, uintptr_t *addr)
> }
> done:
> vm_map_unlock(map);
> -
> - if (do_map_wirefuture)
> - (void) vm_map_wire(map, old, new,
> - VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
>
> if (error == 0)
> *addr = new;
>
> Modified: head/usr.bin/vmstat/vmstat.c
> ==============================================================================
> --- head/usr.bin/vmstat/vmstat.c Mon May 13 15:39:54 2019 (r347531)
> +++ head/usr.bin/vmstat/vmstat.c Mon May 13 16:38:48 2019 (r347532)
> @@ -156,6 +156,7 @@ static struct __vmmeter {
> u_int v_free_min;
> u_int v_free_count;
> u_int v_wire_count;
> + u_long v_user_wire_count;
> u_int v_active_count;
> u_int v_inactive_target;
> u_int v_inactive_count;
> @@ -566,6 +567,7 @@ fill_vmmeter(struct __vmmeter *vmmp)
> GET_VM_STATS(vm, v_free_min);
> GET_VM_STATS(vm, v_free_count);
> GET_VM_STATS(vm, v_wire_count);
> + GET_VM_STATS(vm, v_user_wire_count);
> GET_VM_STATS(vm, v_active_count);
> GET_VM_STATS(vm, v_inactive_target);
> GET_VM_STATS(vm, v_inactive_count);
> @@ -1057,6 +1059,8 @@ dosum(void)
> sum.v_laundry_count);
> xo_emit("{:wired-pages/%9u} {N:pages wired down}\n",
> sum.v_wire_count);
> + xo_emit("{:virtual-user-wired-pages/%9lu} {N:virtual user pages wired "
> + "down}\n", sum.v_user_wire_count);
> xo_emit("{:free-pages/%9u} {N:pages free}\n",
> sum.v_free_count);
> xo_emit("{:bytes-per-page/%9u} {N:bytes per page}\n", sum.v_page_size);
>
>
--
Rod Grimes rgrimes at freebsd.org
More information about the svn-src-all
mailing list