mm: factor commit limit calculation
Subsystems (6)
PATCH 1/2 mm: factor commit limit calculation
The same calculation is currently done in three different places. Factor that code so future changes has to be made at only one place.
--- fs/proc/meminfo.c | 5 +---- include/linux/mman.h | 12 ++++++++++++ mm/mmap.c | 4 +--- mm/nommu.c | 3 +-- 4 files changed, 15 insertions(+), 9 deletions(-)
fs/proc/meminfo.c +1 -4
@@ -24,7 +24,6 @@ static int meminfo_proc_show(struct seq_file *m, void *v) {struct sysinfo i;unsigned long committed; -unsigned long allowed;struct vmalloc_info vmi;long cached;unsigned long pages[NR_LRU_LISTS ];
@@ -37,8 +36,6 @@ static int meminfo_proc_show(struct seq_file *m, void *v)si_meminfo (&i);si_swapinfo (&i); committed =percpu_counter_read_positive (&vm_committed_as); - allowed = ((totalram_pages -hugetlb_total_pages ()) - * sysctl_overcommit_ratio /100 ) + total_swap_pages; cached =global_page_state (NR_FILE_PAGES ) -total_swapcache_pages () - i.bufferram;
@@ -153,7 +150,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v)K (global_page_state (NR_UNSTABLE_NFS )),K (global_page_state (NR_BOUNCE )),K (global_page_state (NR_WRITEBACK_TEMP )), -K (allowed), +K (vm_commit_limit ()),K (committed), (unsigned long )VMALLOC_TOTAL >>10 , vmi.used >> 10,
include/linux/mman.h +12 -0
@@ -7,6 +7,9 @@#include <linux/atomic.h> #include <uapi/linux/mman.h> +#include <linux/hugetlb.h> +#include <linux/swap.h> +extern int sysctl_overcommit_memory;extern int sysctl_overcommit_ratio;extern struct percpu_counter vm_committed_as;
@@ -87,4 +90,13 @@ calc_vm_flag_bits(unsigned long flags)_calc_vm_trans (flags,MAP_DENYWRITE ,VM_DENYWRITE ) | _calc_vm_trans(flags, MAP_LOCKED, VM_LOCKED ); } + +/* +* Commited memory limit enforced when OVERCOMMIT_NEVER policy is used +*/ +static inline unsigned long vm_commit_limit () +{ +return ((totalram_pages -hugetlb_total_pages ()) + * sysctl_overcommit_ratio /100 ) + total_swap_pages; +}#endif /* _LINUX_MMAN_H */
mm/mmap.c +1 -3
@@ -179,14 +179,12 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) goto error; } - allowed = (totalram_pages -hugetlb_total_pages ()) - * sysctl_overcommit_ratio /100 ; + allowed =vm_commit_limit ();/* * Reserve some for root */ if (!cap_sys_admin) allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT -10 ); - allowed += total_swap_pages;/* * Don't let a single process grow so big a user can't recover
mm/nommu.c +1 -2
@@ -1948,13 +1948,12 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) goto error; } - allowed = totalram_pages * sysctl_overcommit_ratio /100 ; + allowed =vm_commit_limit ();/* * Reserve some 3% for root */ if (!cap_sys_admin) allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT -10 ); - allowed += total_swap_pages;/* * Don't let a single process grow so big a user can't recover
Signature · 7 lines
1.7.7.6 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
PATCH 2/2 mm: add overcommit_kbytes sysctl variable
Some applications that run on HPC clusters are designed around the availability of RAM and the overcommit ratio is fine tuned to get the maximum usage of memory without swapping. With growing memory, the 1%-of-all-RAM grain provided by overcommit_ratio has become too coarse for these workload (on a 2TB machine it represents no less than 20GB).
This patch adds the new overcommit_kbytes sysctl variable that allow a much finer grain.
--- Documentation/sysctl/vm.txt | 12 ++++++++++++ Documentation/vm/overcommit-accounting | 7 ++++--- include/linux/mm.h | 5 +++++ include/linux/mman.h | 13 +++++++++++-- kernel/sysctl.c | 10 +++++++++- mm/mmap.c | 25 +++++++++++++++++++++++++ mm/nommu.c | 1 + 7 files changed, 67 insertions(+), 6 deletions(-)
Documentation/sysctl/vm.txt +12 -0
@@ -47,6 +47,7 @@ Currently, these files are in /proc/sys/vm: - numa_zonelist_order - oom_dump_tasks - oom_kill_allocating_task +- overcommit_kbytes - overcommit_memory - overcommit_ratio - page-cluster
@@ -561,6 +562,17 @@ The default value is 0. ============================================================== +overcommit_kbytes: + +When overcommit_memory is set to 2, the committed address space is not +permitted to exceed swap plus this amount of physical RAM. See below. + +Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one +of them may be specified at a time. Setting one disable the other (which +then appears as 0 when read). + +============================================================== + overcommit_memory: This value contains a flag that enables memory overcommitment.
Documentation/vm/overcommit-accounting +4 -3
@@ -14,8 +14,8 @@ The Linux kernel supports the following overcommit handling modes 2 - Don't overcommit. The total address space commit for the system is not permitted to exceed swap + a - configurable percentage (default is 50) of physical RAM. - Depending on the percentage you use, in most situations + configurable amount (default is 50%) of physical RAM. + Depending on the amount you use, in most situations this means a process will not be killed while accessing pages but will receive errors on memory allocation as appropriate.
@@ -26,7 +26,8 @@ The Linux kernel supports the following overcommit handling modes The overcommit policy is set via the sysctl `vm.overcommit_memory'. -The overcommit percentage is set via `vm.overcommit_ratio'. +The overcommit amount can be set via `vm.overcommit_ratio' (percentage) +or `vm.overcommit_kbytes' (absolute value). The current overcommit limit and amount committed are viewable in /proc/meminfo as CommitLimit and Committed_AS respectively.
include/linux/mm.h +5 -0
@@ -53,6 +53,11 @@ extern int sysctl_legacy_va_layout;extern unsigned long sysctl_user_reserve_kbytes;extern unsigned long sysctl_admin_reserve_kbytes; +extern int overcommit_ratio_handler (struct ctl_table *,int ,void __user *, +size_t *,loff_t *); +extern int overcommit_kbytes_handler (struct ctl_table *,int ,void __user *, +size_t *,loff_t *); +#define nth_page (page,n) pfn_to_page(page_to_pfn((page)) + (n))/* to align the pointer to the (next) page boundary */
include/linux/mman.h +11 -2
@@ -12,6 +12,7 @@extern int sysctl_overcommit_memory;extern int sysctl_overcommit_ratio; +extern unsigned long sysctl_overcommit_kbytes;extern struct percpu_counter vm_committed_as;#ifdef CONFIG_SMP
@@ -96,7 +97,15 @@ calc_vm_flag_bits(unsigned long flags)*/ static inline unsigned long vm_commit_limit () { -return ((totalram_pages -hugetlb_total_pages ()) - * sysctl_overcommit_ratio /100 ) + total_swap_pages; +unsigned long allowed; + +if (sysctl_overcommit_kbytes) + allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT -10 ); +else + allowed = ((totalram_pages -hugetlb_total_pages ()) + * sysctl_overcommit_ratio /100 ); + allowed += total_swap_pages; + +return allowed; }#endif /* _LINUX_MMAN_H */
kernel/sysctl.c +9 -1
@@ -97,6 +97,7 @@/* External variables not in a header file. */ extern int sysctl_overcommit_memory;extern int sysctl_overcommit_ratio; +extern unsigned long sysctl_overcommit_kbytes;extern int max_threads;extern int suid_dumpable;#ifdef CONFIG_COREDUMP
@@ -1119,7 +1120,14 @@ static struct ctl_table vm_table[] = { .data = &sysctl_overcommit_ratio, .maxlen = sizeof(sysctl_overcommit_ratio), .mode = 0644, - .proc_handler = proc_dointvec, + .proc_handler = overcommit_ratio_handler, + }, + { + .procname = "overcommit_kbytes", + .data = &sysctl_overcommit_kbytes, + .maxlen = sizeof(sysctl_overcommit_kbytes), + .mode = 0644, + .proc_handler = overcommit_kbytes_handler, }, { .procname = "page-cluster",
mm/mmap.c +25 -0
@@ -86,6 +86,7 @@ EXPORT_SYMBOL(vm_get_page_prot);int sysctl_overcommit_memory__read_mostly =OVERCOMMIT_GUESS ;/* heuristic overcommit */ int sysctl_overcommit_ratio__read_mostly =50 ;/* default is 50% */ +unsigned long sysctl_overcommit_kbytes__read_mostly =0 ;int sysctl_max_map_count__read_mostly =DEFAULT_MAX_MAP_COUNT ;unsigned long sysctl_user_reserve_kbytes__read_mostly =1UL <<17 ;/* 128MB */ unsigned long sysctl_admin_reserve_kbytes__read_mostly =1UL <<13 ;/* 8MB */
@@ -95,6 +96,30 @@ unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */*/ struct percpu_counter vm_committed_as____cacheline_aligned_in_smp ; +int overcommit_ratio_handler (struct ctl_table *table,int write, +void __user *buffer,size_t *lenp, +loff_t *ppos) +{ +int ret; + + ret =proc_dointvec (table, write, buffer, lenp, ppos); +if (ret ==0 && write) + sysctl_overcommit_kbytes =0 ; +return ret; +} + +int overcommit_kbytes_handler (struct ctl_table *table,int write, +void __user *buffer,size_t *lenp, +loff_t *ppos) +{ +int ret; + + ret =proc_doulongvec_minmax (table, write, buffer, lenp, ppos); +if (ret ==0 && write) + sysctl_overcommit_ratio =0 ; +return ret; +} +/* * The global memory commitment made in the system can be a metric * that can be used to drive ballooning decisions when Linux is hosted
mm/nommu.c +1 -0
@@ -60,6 +60,7 @@ unsigned long highest_memmap_pfn;struct percpu_counter vm_committed_as;int sysctl_overcommit_memory =OVERCOMMIT_GUESS ;/* heuristic overcommit */ int sysctl_overcommit_ratio =50 ;/* default is 50% */ +unsigned long sysctl_overcommit_kbytes__read_mostly =0 ;int sysctl_max_map_count =DEFAULT_MAX_MAP_COUNT ;int sysctl_nr_trim_pages =CONFIG_NOMMU_INITIAL_TRIM_EXCESS ;unsigned long sysctl_user_reserve_kbytes__read_mostly =1UL <<17 ;/* 128MB */
Signature · 7 lines
1.7.7.6 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2] mm: add overcommit_kbytes sysctl variable
Quoted 9 lines · Jerome Marchand
Some applications that run on HPC clusters are designed around the availability of RAM and the overcommit ratio is fine tuned to get the maximum usage of memory without swapping. With growing memory, the 1%-of-all-RAM grain provided by
overcommit_ratiohas become too coarse for these workload (on a 2TB machine it represents no less than 20GB).This patch adds the new
overcommit_kbytessysctl variable that allow a much finer grain.
Instead of introducing yet another tunable, why don't we just make the ratio that comes in from the user more fine-grained?
sysctl overcommit_ratio=0.2
We change the internal 'sysctl_overcommit_ratio' to store tenths or hundreths of a percent (or whatever), then parse the input as two integers. I don't think we need fully correct floating point parsing and rounding here, so it shouldn't be too much of a chore. It'd probably end up being less code than you have as it stands.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2] mm: add overcommit_kbytes sysctl variable
Quoted 14 lines · Dave Hansen
Quoted 9 lines · Jerome Marchand
Some applications that run on HPC clusters are designed around the availability of RAM and the overcommit ratio is fine tuned to get the maximum usage of memory without swapping. With growing memory, the 1%-of-all-RAM grain provided by
overcommit_ratiohas become too coarse for these workload (on a 2TB machine it represents no less than 20GB).This patch adds the new
overcommit_kbytessysctl variable that allow a much finer grain.Instead of introducing yet another tunable, why don't we just make the ratio that comes in from the user more fine-grained?
Quoted 8 lines
sysctl overcommit_ratio=0.2
We change the internal '
sysctl_overcommit_ratio' to store tenths or hundreths of a percent (or whatever), then parse the input as two integers. I don't think we need fully correct floating point parsing and rounding here, so it shouldn't be too much of a chore. It'd probably end up being less code than you have as it stands.
Whatever works for me. I did it in that way to be more consistent with what has already been done for dirty*_ratio/bytes.
Thanks,
Jerome
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2] mm: add overcommit_kbytes sysctl variable
In reply to Dave HansenQuoted 14 lines · Dave Hansen
Quoted 9 lines · Jerome Marchand
Some applications that run on HPC clusters are designed around the availability of RAM and the overcommit ratio is fine tuned to get the maximum usage of memory without swapping. With growing memory, the 1%-of-all-RAM grain provided by
overcommit_ratiohas become too coarse for these workload (on a 2TB machine it represents no less than 20GB).This patch adds the new
overcommit_kbytessysctl variable that allow a much finer grain.Instead of introducing yet another tunable, why don't we just make the ratio that comes in from the user more fine-grained?
Quoted 8 lines
sysctl overcommit_ratio=0.2
We change the internal '
sysctl_overcommit_ratio' to store tenths or hundreths of a percent (or whatever), then parse the input as two integers. I don't think we need fully correct floating point parsing and rounding here, so it shouldn't be too much of a chore. It'd probably end up being less code than you have as it stands.
Now that I think about it, that could break user space. Sure write access wouldn't be a problem (one can still write a plain integer), but a script that reads a fractional value when it expects an integer might not be able to cope with it.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2] mm: add overcommit_kbytes sysctl variable
Quoted 11 lines · Jerome Marchand
Quoted 11 lines
Quoted 11 lines
Instead of introducing yet another tunable, why don't we just make the ratio that comes in from the user more fine-grained?
sysctl overcommit_ratio=0.2
We change the internal '
sysctl_overcommit_ratio' to store tenths or hundreths of a percent (or whatever), then parse the input as two integers. I don't think we need fully correct floating point parsing and rounding here, so it shouldn't be too much of a chore. It'd probably end up being less code than you have as it stands.
Quoted 4 lines
Now that I think about it, that could break user space. Sure write access wouldn't be a problem (one can still write a plain integer), but a script that reads a fractional value when it expects an integer might not be able to cope with it.
You're right. Something doing FOO=$(cat overcommit_ratio) and then trying do do arithmetic would just fail loudly. But, it would probably fail silently if we create another tunable that all of a sudden returns 0 (when the kernel is not _behaving_ like it is set to 0).
I'm not sure there's a good way out of this without breakage (or at least confusing) of _some_ old scripts/programs. Either way has ups and downs.
The existing dirty_ratio/bytes stuff just annoys me because I end up having to check two places whenever I go looking for it.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2] mm: add overcommit_kbytes sysctl variable
Quoted 22 lines · Dave Hansen
Quoted 11 lines · Jerome Marchand
Quoted 11 lines
Quoted 11 lines
Instead of introducing yet another tunable, why don't we just make the ratio that comes in from the user more fine-grained?
sysctl overcommit_ratio=0.2
We change the internal '
sysctl_overcommit_ratio' to store tenths or hundreths of a percent (or whatever), then parse the input as two integers. I don't think we need fully correct floating point parsing and rounding here, so it shouldn't be too much of a chore. It'd probably end up being less code than you have as it stands.Quoted 4 lines
Now that I think about it, that could break user space. Sure write access wouldn't be a problem (one can still write a plain integer), but a script that reads a fractional value when it expects an integer might not be able to cope with it.
You're right. Something doing FOO=$(cat
overcommit_ratio) and then trying do do arithmetic would just fail loudly. But, it would probably fail silently if we create another tunable that all of a sudden returns 0 (when the kernel is not _behaving_ like it is set to 0).
Quoted 7 lines
I'm not sure there's a good way out of this without breakage (or at least confusing) of _some_ old
scripts/programs. Either way has ups and downs.The existing dirty_ratio/bytes stuff just annoys me because I end up having to check two places whenever I go looking for it.
Right. Then we could just use some overcommit_fine_ratio internally and overcommit_ratio would show and set a rounded value. I doubt that a script that reads 80% would notice the difference if it is actually 79.5%.
We could also use overcommit_kbytes internally, but then overcommit_ratio would fluctuate if RAM ram is added/removed (e.g. memory hotplug or baloon driver). That might be a problem.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
PATCH v2 2/2 mm: allow to set overcommit ratio more precisely
In reply to Jerome MarchandChanges since v1:
- use overcommit_ratio_ppm instead of overcommit_kbytes
- keep both variables in sync
Some applications that run on HPC clusters are designed around the availability of RAM and the overcommit ratio is fine tuned to get the maximum usage of memory without swapping. With growing memory, the 1% of all RAM grain provided by overcommit_ratio has become too coarse for these workload (on a 2TB machine it represents no less than 20GB).
This patch adds the new overcommit_ratio_ppm sysctl variable that allow to set overcommit ratio with a part per million precision. The old overcommit_ratio variable can still be used to set and read the ratio with a 1% precision. That way, overcommit_ratio interface isn't broken in any way that I can imagine.
--- include/linux/mman.h | 6 ++-- include/linux/sysctl.h | 2 + kernel/sysctl.c | 63 +++++++++++++++++++++++++++++++++++++++++++++-- mm/mmap.c | 2 +- mm/nommu.c | 2 +- 5 files changed, 67 insertions(+), 8 deletions(-)
include/linux/mman.h +3 -3
@@ -11,7 +11,7 @@#include <linux/swap.h> extern int sysctl_overcommit_memory; -extern int sysctl_overcommit_ratio; +extern int sysctl_overcommit_ratio_ppm;extern struct percpu_counter vm_committed_as;#ifdef CONFIG_SMP
@@ -96,7 +96,7 @@ calc_vm_flag_bits(unsigned long flags)*/ static inline unsigned long vm_commit_limit () { -return ((totalram_pages -hugetlb_total_pages ()) - * sysctl_overcommit_ratio /100 ) + total_swap_pages; +return ((u64) (totalram_pages -hugetlb_total_pages ()) + * sysctl_overcommit_ratio_ppm /100000 ) + total_swap_pages; }#endif /* _LINUX_MMAN_H */
include/linux/sysctl.h +2 -0
@@ -51,6 +51,8 @@ extern int proc_dointvec_userhz_jiffies(struct ctl_table *, int, void __user *, size_t *, loff_t *);extern int proc_dointvec_ms_jiffies (struct ctl_table *,int ,void __user *,size_t *,loff_t *); +extern int proc_dointvec_percent_ppm (struct ctl_table *,int , +void __user *,size_t *,loff_t *);extern int proc_doulongvec_minmax (struct ctl_table *,int ,void __user *,size_t *,loff_t *); extern int proc_doulongvec_ms_jiffies_minmax(struct ctl_table *table, int,
kernel/sysctl.c +60 -3
81 lines not on this page · raw message
mm/mmap.c +1 -1
@@ -85,7 +85,7 @@ pgprot_t vm_get_page_prot(unsigned long vm_flags)EXPORT_SYMBOL (vm_get_page_prot);int sysctl_overcommit_memory__read_mostly =OVERCOMMIT_GUESS ;/* heuristic overcommit */ -int sysctl_overcommit_ratio__read_mostly =50 ;/* default is 50% */ +int sysctl_overcommit_ratio_ppm__read_mostly =500000 ;/* default is 50% */ int sysctl_max_map_count__read_mostly =DEFAULT_MAX_MAP_COUNT ;unsigned long sysctl_user_reserve_kbytes__read_mostly =1UL <<17 ;/* 128MB */ unsigned long sysctl_admin_reserve_kbytes__read_mostly =1UL <<13 ;/* 8MB */
mm/nommu.c +1 -1
@@ -59,7 +59,7 @@ unsigned long max_mapnr;unsigned long highest_memmap_pfn;struct percpu_counter vm_committed_as;int sysctl_overcommit_memory =OVERCOMMIT_GUESS ;/* heuristic overcommit */ -int sysctl_overcommit_ratio =50 ;/* default is 50% */ +int sysctl_overcommit_ratio_ppm =500000 ;/* default is 50% */ int sysctl_max_map_count =DEFAULT_MAX_MAP_COUNT ;int sysctl_nr_trim_pages =CONFIG_NOMMU_INITIAL_TRIM_EXCESS ;unsigned long sysctl_user_reserve_kbytes__read_mostly =1UL <<17 ;/* 128MB */
Signature · 7 lines
1.7.7.6 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2 v2] mm: allow to set overcommit ratio more precisely
Quoted 5 lines · Jerome Marchand
This patch adds the new
overcommit_ratio_ppmsysctl variable that allow to set overcommit ratio with a part per million precision. The oldovercommit_ratiovariable can still be used to set and read the ratio with a 1% precision. That way,overcommit_ratiointerface isn't broken in any way that I can imagine.
Looks like a pretty sane solution. Could you also make a Documentation/ update, please?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2 v2] mm: allow to set overcommit ratio more precisely
Quoted 9 lines · Dave Hansen
Quoted 5 lines · Jerome Marchand
This patch adds the new
overcommit_ratio_ppmsysctl variable that allow to set overcommit ratio with a part per million precision. The oldovercommit_ratiovariable can still be used to set and read the ratio with a 1% precision. That way,overcommit_ratiointerface isn't broken in any way that I can imagine.Looks like a pretty sane solution. Could you also make a
Documentation/update, please?
Damn! I forgot. Will do.
Thanks
Quoted 7 lines
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2 v2] mm: allow to set overcommit ratio more precisely
hi!
Quoted 10 lines
Quoted 8 lines
Quoted 5 lines
This patch adds the new
overcommit_ratio_ppmsysctl variable that allow to set overcommit ratio with a part per million precision. The oldovercommit_ratiovariable can still be used to set and read the ratio with a 1% precision. That way,overcommit_ratiointerface isn't broken in any way that I can imagine.Looks like a pretty sane solution. Could you also make a
Documentation/update, please?Damn! I forgot. Will do.
Actually... would something like overcommit_bytes be better interface? overcommit_pages?
If system would normally allow allocating "n" pages, with overcommit it would allow allocating "n + overcommit_pages" pages. That seems like right granularity...
Signature · 8 lines
(english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [PATCH 2/2 v2] mm: allow to set overcommit ratio more precisely
Quoted 11 lines · Pavel Machek
hi!
Quoted 9 lines
Quoted 8 lines
Quoted 5 lines
This patch adds the new
overcommit_ratio_ppmsysctl variable that allow to set overcommit ratio with a part per million precision. The oldovercommit_ratiovariable can still be used to set and read the ratio with a 1% precision. That way,overcommit_ratiointerface isn't broken in any way that I can imagine.Looks like a pretty sane solution. Could you also make a
Documentation/update, please?
Quoted 8 lines
Quoted 1 line
Damn! I forgot. Will do.
Actually... would something like
overcommit_bytesbe better interface?overcommit_pages?If system would normally allow allocating "n" pages, with overcommit it would allow allocating "n +
overcommit_pages" pages. That seems like right granularity...
I don't know what do you mean by "normally".
Anyway, I've considered that option: my concern about mixing absolute and proportional values is that they would diverge if the amount of ram varies (e.g. memory hotplug or virt baloon driver).
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
PATCH v3 2/2 mm: allow to set overcommit ratio more precisely
In reply to Dave HansenChanges since v2:
- update documentation
Changes since v1:
- use overcommit_ratio_ppm instead of overcommit_kbytes
- keep both variables in sync
Some applications that run on HPC clusters are designed around the availability of RAM and the overcommit ratio is fine tuned to get the maximum usage of memory without swapping. With growing memory, the 1% of all RAM grain provided by overcommit_ratio has become too coarse for these workload (on a 2TB machine it represents no less than 20GB).
This patch adds the new overcommit_ratio_ppm sysctl variable that allow to set overcommit ratio with a part per million precision. The old overcommit_ratio variable can still be used to set and read the ratio with a 1% precision. That way, overcommit_ratio interface isn't broken in any way that I can imagine.
--- Documentation/sysctl/vm.txt | 10 +++++ Documentation/vm/overcommit-accounting | 7 ++-- include/linux/mman.h | 6 ++-- include/linux/sysctl.h | 2 + kernel/sysctl.c | 63 ++++++++++++++++++++++++++++++-- mm/mmap.c | 2 +- mm/nommu.c | 2 +- 7 files changed, 81 insertions(+), 11 deletions(-)
Documentation/sysctl/vm.txt +10 -0
@@ -49,6 +49,7 @@ Currently, these files are in /proc/sys/vm: - oom_kill_allocating_task - overcommit_memory - overcommit_ratio +- overcommit_ratio_ppm - page-cluster - panic_on_oom - percpu_pagelist_fraction
@@ -591,6 +592,15 @@ overcommit_ratio: When overcommit_memory is set to 2, the committed address space is not permitted to exceed swap plus this percentage of physical RAM. See above. +If overcommit_ratio_ppm has been set, overcommit_ratio shows a +rounded value. + +============================================================== + +overcommit_ratio_ppm: + +Same as overcommit_ratio, but allows to set the ratio with a finer +grain (part per million). ==============================================================
Documentation/vm/overcommit-accounting +4 -3
@@ -14,8 +14,8 @@ The Linux kernel supports the following overcommit handling modes 2 - Don't overcommit. The total address space commit for the system is not permitted to exceed swap + a - configurable percentage (default is 50) of physical RAM. - Depending on the percentage you use, in most situations + configurable ratio (default is 50%) of physical RAM. + Depending on the ratio you use, in most situations this means a process will not be killed while accessing pages but will receive errors on memory allocation as appropriate.
@@ -26,7 +26,8 @@ The Linux kernel supports the following overcommit handling modes The overcommit policy is set via the sysctl `vm.overcommit_memory'. -The overcommit percentage is set via `vm.overcommit_ratio'. +The overcommit percentage is set via `vm.overcommit_ratio' or +`vm.overcommit_ratio_ppm'. The current overcommit limit and amount committed are viewable in /proc/meminfo as CommitLimit and Committed_AS respectively.
include/linux/mman.h +3 -3
@@ -11,7 +11,7 @@#include <linux/swap.h> extern int sysctl_overcommit_memory; -extern int sysctl_overcommit_ratio; +extern int sysctl_overcommit_ratio_ppm;extern struct percpu_counter vm_committed_as;#ifdef CONFIG_SMP
@@ -96,7 +96,7 @@ calc_vm_flag_bits(unsigned long flags)*/ static inline unsigned long vm_commit_limit () { -return ((totalram_pages -hugetlb_total_pages ()) - * sysctl_overcommit_ratio /100 ) + total_swap_pages; +return ((u64) (totalram_pages -hugetlb_total_pages ()) + * sysctl_overcommit_ratio_ppm /100000 ) + total_swap_pages; }#endif /* _LINUX_MMAN_H */
include/linux/sysctl.h +2 -0
@@ -51,6 +51,8 @@ extern int proc_dointvec_userhz_jiffies(struct ctl_table *, int, void __user *, size_t *, loff_t *);extern int proc_dointvec_ms_jiffies (struct ctl_table *,int ,void __user *,size_t *,loff_t *); +extern int proc_dointvec_percent_ppm (struct ctl_table *,int , +void __user *,size_t *,loff_t *);extern int proc_doulongvec_minmax (struct ctl_table *,int ,void __user *,size_t *,loff_t *); extern int proc_doulongvec_ms_jiffies_minmax(struct ctl_table *table, int,
kernel/sysctl.c +60 -3
81 lines not on this page · raw message
mm/mmap.c +1 -1
@@ -85,7 +85,7 @@ pgprot_t vm_get_page_prot(unsigned long vm_flags)EXPORT_SYMBOL (vm_get_page_prot);int sysctl_overcommit_memory__read_mostly =OVERCOMMIT_GUESS ;/* heuristic overcommit */ -int sysctl_overcommit_ratio__read_mostly =50 ;/* default is 50% */ +int sysctl_overcommit_ratio_ppm__read_mostly =500000 ;/* default is 50% */ int sysctl_max_map_count__read_mostly =DEFAULT_MAX_MAP_COUNT ;unsigned long sysctl_user_reserve_kbytes__read_mostly =1UL <<17 ;/* 128MB */ unsigned long sysctl_admin_reserve_kbytes__read_mostly =1UL <<13 ;/* 8MB */
mm/nommu.c +1 -1
@@ -59,7 +59,7 @@ unsigned long max_mapnr;unsigned long highest_memmap_pfn;struct percpu_counter vm_committed_as;int sysctl_overcommit_memory =OVERCOMMIT_GUESS ;/* heuristic overcommit */ -int sysctl_overcommit_ratio =50 ;/* default is 50% */ +int sysctl_overcommit_ratio_ppm =500000 ;/* default is 50% */ int sysctl_max_map_count =DEFAULT_MAX_MAP_COUNT ;int sysctl_nr_trim_pages =CONFIG_NOMMU_INITIAL_TRIM_EXCESS ;unsigned long sysctl_user_reserve_kbytes__read_mostly =1UL <<17 ;/* 128MB */
Signature · 7 lines
1.7.7.6 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>