Author: glebius
Date: Fri Feb 9 04:45:39 2018
New Revision: 329058
URL: https://svnweb.freebsd.org/changeset/base/329058
Log:
Fix boot_pages exhaustion on machines with many domains and cores, where
size of UMA zone allocation is greater than page size. In this case zone
of zones can not use UMA_MD_SMALL_ALLOC, and we need to postpone switch
off of this zone from startup_alloc() until full launch of VM.
o Always supply number of VM zones to uma_startup_count(). On machines
with UMA_MD_SMALL_ALLOC ignore it completely, unless zsize goes over
a page. In the latter case account VM zones for number of allocations
from the zone of zones.
o Rewrite startup_alloc() so that it will immediately switch off from
itself any zone that is already capable of running real alloc.
In worst case scenario we may leak a single page here. See comment
in uma_startup_count().
o Hardcode call to uma_startup2() into vm_mem_init(). Otherwise some
extra SYSINITs, e.g. vm_page_init() may sneak in before.
o While here, remove uma_boot_pages_mtx. With recent changes to boot
pages calculation, we are guaranteed to use all of the boot_pages
in the early single threaded stage.
Reported & tested by: mav
Modified:
head/sys/kern/kern_malloc.c
head/sys/vm/uma_core.c
head/sys/vm/vm_init.c
head/sys/vm/vm_page.c
Modified: head/sys/kern/kern_malloc.c
==============================================================================
--- head/sys/kern/kern_malloc.c Fri Feb 9 03:07:12 2018 (r329057)
+++ head/sys/kern/kern_malloc.c Fri Feb 9 04:45:39 2018 (r329058)
@@ -96,8 +96,6 @@ __FBSDID("$FreeBSD$");
dtrace_malloc_probe_func_t dtrace_malloc_probe;
#endif
-extern void uma_startup2(void);
-
#if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) || \
defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE)
#define MALLOC_DEBUG 1
@@ -928,8 +926,6 @@ mallocinit(void *dummy)
mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
kmeminit();
-
- uma_startup2();
if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
kmem_zmax = KMEM_ZMAX;
Modified: head/sys/vm/uma_core.c
==============================================================================
--- head/sys/vm/uma_core.c Fri Feb 9 03:07:12 2018 (r329057)
+++ head/sys/vm/uma_core.c Fri Feb 9 04:45:39 2018 (r329058)
@@ -134,13 +134,10 @@ static struct rwlock_padalign __exclusive_cache_line u
/*
* Pointer and counter to pool of pages, that is preallocated at
- * startup to bootstrap UMA. Early zones continue to use the pool
- * until it is depleted, so allocations may happen after boot, thus
- * we need a mutex to protect it.
+ * startup to bootstrap UMA.
*/
static char *bootmem;
static int boot_pages;
-static struct mtx uma_boot_pages_mtx;
static struct sx uma_drain_lock;
@@ -1081,37 +1078,46 @@ startup_alloc(uma_zone_t zone, vm_size_t bytes, int do
int pages;
keg = zone_first_keg(zone);
- pages = howmany(bytes, PAGE_SIZE);
- KASSERT(pages > 0, ("startup_alloc can't reserve 0 pages\n"));
/*
- * Check our small startup cache to see if it has pages remaining.
+ * If we are in BOOT_BUCKETS or higher, than switch to real
+ * allocator. Zones with page sized slabs switch at BOOT_PAGEALLOC.
*/
- mtx_lock(&uma_boot_pages_mtx);
- if (pages <= boot_pages) {
-#ifdef DIAGNOSTIC
- printf("%s from \"%s\", %d boot pages left\n", __func__,
- zone->uz_name, boot_pages);
+ switch (booted) {
+ case BOOT_COLD:
+ case BOOT_STRAPPED:
+ break;
+ case BOOT_PAGEALLOC:
+ if (keg->uk_ppera > 1)
+ break;
+ case BOOT_BUCKETS:
+ case BOOT_RUNNING:
+#ifdef UMA_MD_SMALL_ALLOC
+ keg->uk_allocf = (keg->uk_ppera > 1) ?
+ page_alloc : uma_small_alloc;
+#else
+ keg->uk_allocf = page_alloc;
#endif
- mem = bootmem;
- boot_pages -= pages;
- bootmem += pages * PAGE_SIZE;
- mtx_unlock(&uma_boot_pages_mtx);
- *pflag = UMA_SLAB_BOOT;
- return (mem);
+ return keg->uk_allocf(zone, bytes, domain, pflag, wait);
}
- mtx_unlock(&uma_boot_pages_mtx);
- if (booted < BOOT_PAGEALLOC)
- panic("UMA zone \"%s\": Increase vm.boot_pages", zone->uz_name);
+
/*
- * Now that we've booted reset these users to their real allocator.
+ * Check our small startup cache to see if it has pages remaining.
*/
-#ifdef UMA_MD_SMALL_ALLOC
- keg->uk_allocf = (keg->uk_ppera > 1) ? page_alloc : uma_small_alloc;
-#else
- keg->uk_allocf = page_alloc;
+ pages = howmany(bytes, PAGE_SIZE);
+ KASSERT(pages > 0, ("%s can't reserve 0 pages", __func__));
+ if (pages > boot_pages)
+ panic("UMA zone \"%s\": Increase vm.boot_pages", zone->uz_name);
+#ifdef DIAGNOSTIC
+ printf("%s from \"%s\", %d boot pages left\n", __func__, zone->uz_name,
+ boot_pages);
#endif
- return keg->uk_allocf(zone, bytes, domain, pflag, wait);
+ mem = bootmem;
+ boot_pages -= pages;
+ bootmem += pages * PAGE_SIZE;
+ *pflag = UMA_SLAB_BOOT;
+
+ return (mem);
}
/*
@@ -1789,9 +1795,9 @@ zone_foreach(void (*zfunc)(uma_zone_t))
#define UMA_BOOT_ALIGN 32
static int zsize, ksize;
int
-uma_startup_count(int zones)
+uma_startup_count(int vm_zones)
{
- int pages;
+ int zones, pages;
ksize = sizeof(struct uma_keg) +
(sizeof(struct uma_domain) * vm_ndomains);
@@ -1806,12 +1812,17 @@ uma_startup_count(int zones)
pages = howmany(roundup(zsize, CACHE_LINE_SIZE) * 2 +
roundup(ksize, CACHE_LINE_SIZE), PAGE_SIZE);
- zones += UMA_BOOT_ZONES;
+#ifdef UMA_MD_SMALL_ALLOC
+ zones = UMA_BOOT_ZONES;
+#else
+ zones = UMA_BOOT_ZONES + vm_zones;
+ vm_zones = 0;
+#endif
/* Memory for the rest of startup zones, UMA and VM, ... */
if (zsize > UMA_SLAB_SIZE)
- pages += zones * howmany(roundup2(zsize, UMA_BOOT_ALIGN),
- UMA_SLAB_SIZE);
+ pages += (zones + vm_zones) *
+ howmany(roundup2(zsize, UMA_BOOT_ALIGN), UMA_SLAB_SIZE);
else
pages += howmany(zones,
UMA_SLAB_SPACE / roundup2(zsize, UMA_BOOT_ALIGN));
@@ -1872,7 +1883,6 @@ uma_startup(void *mem, int npages)
args.flags = UMA_ZFLAG_INTERNAL;
zone_ctor(kegs, zsize, &args, M_WAITOK);
- mtx_init(&uma_boot_pages_mtx, "UMA boot pages", NULL, MTX_DEF);
bootmem = mem;
boot_pages = npages;
@@ -1917,6 +1927,9 @@ void
uma_startup2(void)
{
+#ifdef DIAGNOSTIC
+ printf("Entering %s with %d boot pages left\n", __func__, boot_pages);
+#endif
booted = BOOT_BUCKETS;
sx_init(&uma_drain_lock, "umadrain");
bucket_enable();
Modified: head/sys/vm/vm_init.c
==============================================================================
--- head/sys/vm/vm_init.c Fri Feb 9 03:07:12 2018 (r329057)
+++ head/sys/vm/vm_init.c Fri Feb 9 04:45:39 2018 (r329058)
@@ -95,6 +95,7 @@ __FBSDID("$FreeBSD$");
#include <vm/vm_extern.h>
extern void uma_startup1(void);
+extern void uma_startup2(void);
extern void vm_radix_reserve_kva(void);
#if VM_NRESERVLEVEL > 0
@@ -183,9 +184,9 @@ vm_mem_init(dummy)
#ifndef UMA_MD_SMALL_ALLOC
/* Set up radix zone to use noobj_alloc. */
vm_radix_reserve_kva();
- /* Announce page availability to UMA. */
- uma_startup1();
#endif
+ /* Announce full page availability to UMA. */
+ uma_startup2();
kmem_init_zero_region();
pmap_init();
vm_pager_init();
Modified: head/sys/vm/vm_page.c
==============================================================================
--- head/sys/vm/vm_page.c Fri Feb 9 03:07:12 2018 (r329057)
+++ head/sys/vm/vm_page.c Fri Feb 9 04:45:39 2018 (r329058)
@@ -506,16 +506,13 @@ vm_page_startup(vm_offset_t vaddr)
* Allocate memory for use when boot strapping the kernel memory
* allocator. Tell UMA how many zones we are going to create
* before going fully functional. UMA will add its zones.
- */
-#ifdef UMA_MD_SMALL_ALLOC
- boot_pages = uma_startup_count(0);
-#else
- /*
+ *
* VM startup zones: vmem, vmem_btag, VM OBJECT, RADIX NODE, MAP,
* KMAP ENTRY, MAP ENTRY, VMSPACE.
*/
boot_pages = uma_startup_count(8);
+#ifndef UMA_MD_SMALL_ALLOC
/* vmem_startup() calls uma_prealloc(). */
boot_pages += vmem_startup_count();
/* vm_map_startup() calls uma_prealloc(). */
_______________________________________________
svn-src-all@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/svn-src-all
To unsubscribe, send any mail to "svn-src-all-unsubscr...@freebsd.org"
