d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix OOM handling of some corner cases.

Browse Source 
In addition to pagedaemon initiating OOM, also do it from the
vm_fault() internals.  Namely, if the thread waits for a free page to
satisfy page fault some preconfigured amount of time, trigger OOM.
These triggers are rate-limited, due to a usual case of several
threads of the same multi-threaded process to enter fault handler
simultaneously.  The faults from pagedaemon threads participate in the
calculation of OOM rate, but are not under the limit.

Reviewed by:	markj (previous version)
Tested by:	pho
Discussed with:	alc
Sponsored by:	The FreeBSD Foundation
MFC after:	2 weeks
Differential revision:	https://reviews.freebsd.org/D13671
This commit is contained in:
Konstantin Belousov 2019-08-16 09:43:49 +00:00
parent 3a79e8e772
commit 245139c69d
Notes: svn2git 2020-12-20 02:59:44 +00:00 
svn path=/head/; revision=351114
4 changed files with 63 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
sys/vm/vm_fault.c
View File

@ -135,6 +135,18 @@ static void vm_fault_dontneed(const struct faultstate *fs, vm_offset_t vaddr,
static void vm_fault_prefault(const struct faultstate *fs, vm_offset_t addra,
int backward, int forward, bool obj_locked);
static int vm_pfault_oom_attempts = 3;
SYSCTL_INT(_vm, OID_AUTO, pfault_oom_attempts, CTLFLAG_RWTUN,
&vm_pfault_oom_attempts, 0,
"Number of page allocation attempts in page fault handler before it "
"triggers OOM handling");
static int vm_pfault_oom_wait = 10;
SYSCTL_INT(_vm, OID_AUTO, pfault_oom_wait, CTLFLAG_RWTUN,
&vm_pfault_oom_wait, 0,
"Number of seconds to wait for free pages before retrying "
"the page fault handler");
static inline void
release_page(struct faultstate *fs)
{
@ -570,7 +582,7 @@ vm_fault_hold(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
vm_pindex_t retry_pindex;
vm_prot_t prot, retry_prot;
int ahead, alloc_req, behind, cluster_offset, error, era, faultcount;
int locked, nera, result, rv;
int locked, nera, oom, result, rv;
u_char behavior;
boolean_t wired; /* Passed by reference. */
bool dead, hardfault, is_first_object_locked;
@ -581,7 +593,9 @@ vm_fault_hold(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
nera = -1;
hardfault = false;
RetryFault:;
RetryFault:
oom = 0;
RetryFault_oom:
/*
* Find the backing store object and offset into it to begin the
@ -827,7 +841,18 @@ RetryFault:;
}
if (fs.m == NULL) {
unlock_and_deallocate(&fs);
vm_waitpfault(dset);
if (vm_pfault_oom_attempts < 0 ||
oom < vm_pfault_oom_attempts) {
oom++;
vm_waitpfault(dset,
vm_pfault_oom_wait * hz);
goto RetryFault_oom;
}
if (bootverbose)
printf(
"proc %d (%s) failed to alloc page on fault, starting OOM\n",
curproc->p_pid, curproc->p_comm);
vm_pageout_oom(VM_OOM_MEM_PF);
goto RetryFault;
}
}

4
sys/vm/vm_page.c
View File

@ -3032,7 +3032,7 @@ vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object, int req)
* this balance without careful testing first.
*/
void
vm_waitpfault(struct domainset *dset)
vm_waitpfault(struct domainset *dset, int timo)
{
/*
@ -3044,7 +3044,7 @@ vm_waitpfault(struct domainset *dset)
if (vm_page_count_min_set(&dset->ds_mask)) {
vm_min_waiters++;
msleep(&vm_min_domains, &vm_domainset_lock, PUSER | PDROP,
"pfault", 0);
"pfault", timo);
} else
mtx_unlock(&vm_domainset_lock);
}

31
sys/vm/vm_pageout.c
View File

@ -1720,6 +1720,12 @@ vm_pageout_oom_pagecount(struct vmspace *vmspace)
return (res);
}
static int vm_oom_ratelim_last;
static int vm_oom_pf_secs = 10;
SYSCTL_INT(_vm, OID_AUTO, oom_pf_secs, CTLFLAG_RWTUN, &vm_oom_pf_secs, 0,
"");
static struct mtx vm_oom_ratelim_mtx;
void
vm_pageout_oom(int shortage)
{
@ -1727,8 +1733,30 @@ vm_pageout_oom(int shortage)
vm_offset_t size, bigsize;
struct thread *td;
struct vmspace *vm;
int now;
bool breakout;
/*
* For OOM requests originating from vm_fault(), there is a high
* chance that a single large process faults simultaneously in
* several threads. Also, on an active system running many
* processes of middle-size, like buildworld, all of them
* could fault almost simultaneously as well.
*
* To avoid killing too many processes, rate-limit OOMs
* initiated by vm_fault() time-outs on the waits for free
* pages.
*/
mtx_lock(&vm_oom_ratelim_mtx);
now = ticks;
if (shortage == VM_OOM_MEM_PF &&
(u_int)(now - vm_oom_ratelim_last) < hz * vm_oom_pf_secs) {
mtx_unlock(&vm_oom_ratelim_mtx);
return;
}
vm_oom_ratelim_last = now;
mtx_unlock(&vm_oom_ratelim_mtx);
/*
* We keep the process bigproc locked once we find it to keep anyone
* from messing with it; however, there is a possibility of
@ -1793,7 +1821,7 @@ vm_pageout_oom(int shortage)
continue;
}
size = vmspace_swap_count(vm);
if (shortage == VM_OOM_MEM)
if (shortage == VM_OOM_MEM || shortage == VM_OOM_MEM_PF)
size += vm_pageout_oom_pagecount(vm);
vm_map_unlock_read(&vm->vm_map);
vmspace_free(vm);
@ -2048,6 +2076,7 @@ vm_pageout(void)
p = curproc;
td = curthread;
mtx_init(&vm_oom_ratelim_mtx, "vmoomr", NULL, MTX_DEF);
swap_pager_swap_init();
for (first = -1, i = 0; i < vm_ndomains; i++) {
if (VM_DOMAIN_EMPTY(i)) {

5
sys/vm/vm_pageout.h
View File

@ -79,7 +79,8 @@ extern u_long vm_page_max_user_wired;
extern int vm_pageout_page_count;
#define VM_OOM_MEM 1
#define VM_OOM_SWAPZ 2
#define VM_OOM_MEM_PF 2
#define VM_OOM_SWAPZ 3
/*
* vm_lowmem flags.
@ -96,7 +97,7 @@ extern int vm_pageout_page_count;
*/
void vm_wait(vm_object_t obj);
void vm_waitpfault(struct domainset *);
void vm_waitpfault(struct domainset *, int timo);
void vm_wait_domain(int domain);
void vm_wait_min(void);
void vm_wait_severe(void);