Patrick.Le-Dot wrote:
> Add accounting at the page level.
>
> Signed-off-by: Patrick Le Dot <[EMAIL PROTECTED]>
> ---
>
> include/linux/mm_types.h | 12 +++++
> kernel/res_group/memctlr.c | 98
> +++++++++++++++++++++++++++++++++++----------
> 2 files changed, 90 insertions(+), 20 deletions(-)
>
> diff -puN a/include/linux/mm_types.h b/include/linux/mm_types.h
> --- a/include/linux/mm_types.h 2006-12-12 13:39:19.000000000 +0100
> +++ b/include/linux/mm_types.h 2006-12-11 10:47:22.000000000 +0100
> @@ -6,6 +6,14 @@
> #include <linux/list.h>
> #include <linux/spinlock.h>
>
> +#ifdef CONFIG_RES_GROUPS_MEMORY
> +struct rg_shared_pg_cnt {
> + unsigned long rg_bit_id; /* rgroup id */
> + atomic_t count; /* 1 counter per shared page per rgroup
> */
> + struct rg_shared_pg_cnt *next; /* another rgroup use the page */
> +};
> +#endif /* CONFIG_RES_GROUPS_MEMORY */
> +
> struct address_space;
>
> /*
> @@ -62,6 +70,10 @@ struct page {
> void *virtual; /* Kernel virtual address (NULL if
> not kmapped, ie. highmem) */
> #endif /* WANT_PAGE_VIRTUAL */
> +#ifdef CONFIG_RES_GROUPS_MEMORY
> + unsigned long rg_bitmap; /* 1 bit per rgroup using this
> page */
> + struct rg_shared_pg_cnt *shared_count; /* and 1 counter per rgroup */
> +#endif /* CONFIG_RES_GROUPS_MEMORY */
> };
>
This would limit the numbers to groups to the word size on the machine.
It would be interesting if we can support shared pages without any
changes to struct page. I am working on an approach of using the rmap
information for tracking shared pages (I have one dirty patch which
needs a lot of cleaning up).
> #endif /* _LINUX_MM_TYPES_H */
> diff -puN a/kernel/res_group/memctlr.c b/kernel/res_group/memctlr.c
> --- a/kernel/res_group/memctlr.c 2006-12-12 13:20:24.000000000 +0100
> +++ b/kernel/res_group/memctlr.c 2006-12-12 12:57:30.000000000 +0100
> @@ -42,6 +42,12 @@ static const char res_ctlr_name[] = "mem
> static struct resource_group *root_rgroup;
> static const char version[] = "0.01";
> static struct memctlr *memctlr_root;
> +static unsigned long rg_bitmap_shift_index = 0;
> +
> +/*
> + * first implementation : use a global lock
> + */
> +static spinlock_t memctlr_lock;
>
> /*
> * this struct is used in mm_struct
> @@ -56,6 +62,7 @@ struct mem_counter {
> struct memctlr {
> struct res_shares shares; /* My shares */
> struct mem_counter counter; /* Accounting information */
> + unsigned long bit_id; /* rgroup_id bitmap : only 1 bit on */
> spinlock_t lock;
> };
>
> @@ -120,10 +127,17 @@ static inline struct memctlr *get_task_m
> return res;
> }
>
> +void memctlr_inc_rss(struct page *page)
> +{
> + struct mm_struct *mm = current->mm;
> +
> + memctlr_inc_rss_mm(page, mm);
> +}
>
> void memctlr_inc_rss_mm(struct page *page, struct mm_struct *mm)
> {
> struct memctlr *res;
> + struct rg_shared_pg_cnt *shared_pg_cnt;
>
> res = get_task_memctlr(current);
> if (!res) {
> @@ -131,32 +145,44 @@ void memctlr_inc_rss_mm(struct page *pag
> return;
> }
>
> - spin_lock(&res->lock);
> - atomic_long_inc(¤t->mm->counter->rss);
> - atomic_long_inc(&res->counter.rss);
> - spin_unlock(&res->lock);
> -}
> -
> -void memctlr_inc_rss(struct page *page)
> -{
> - struct memctlr *res;
> - struct mm_struct *mm = current->mm;
> + spin_lock(&memctlr_lock);
>
> - res = get_task_memctlr(current);
> - if (!res) {
> - printk(KERN_INFO "inc_rss no res set *---*\n");
> - return;
> + if ((page->rg_bitmap & res->bit_id) != 0) {
> + /* search the counter for this rgroup */
> + shared_pg_cnt = page->shared_count;
> + while (shared_pg_cnt != NULL) {
> + if (shared_pg_cnt->rg_bit_id == res->bit_id) {
> + atomic_inc(&shared_pg_cnt->count);
> + spin_unlock(&memctlr_lock);
> + return;
> + }
> + shared_pg_cnt = shared_pg_cnt->next;
> + }
> + /* should never get here */
> + BUG();
> }
>
> - spin_lock(&res->lock);
> + /* first mapping for this rgroup : add a new counter */
> + shared_pg_cnt = kzalloc(sizeof(struct rg_shared_pg_cnt), GFP_ATOMIC);
> + shared_pg_cnt->rg_bit_id = res->bit_id;
> + atomic_set(&shared_pg_cnt->count, 1);
> +
> + shared_pg_cnt->next = page->shared_count;
> + page->shared_count = shared_pg_cnt;
> +
> + page->rg_bitmap = (page->rg_bitmap ^ res->bit_id);
> +
> atomic_long_inc(&mm->counter->rss);
> atomic_long_inc(&res->counter.rss);
> - spin_unlock(&res->lock);
> +
> + spin_unlock(&memctlr_lock);
> }
>
> void memctlr_dec_rss_mm(struct page *page, struct mm_struct *mm)
> {
> struct memctlr *res;
> + struct rg_shared_pg_cnt *shared_pg_cnt;
> + struct rg_shared_pg_cnt *previous = NULL;
>
> res = get_task_memctlr(current);
> if (!res) {
> @@ -164,10 +190,36 @@ void memctlr_dec_rss_mm(struct page *pag
> return;
> }
>
> - spin_lock(&res->lock);
> - atomic_long_dec(&res->counter.rss);
> - atomic_long_dec(&mm->counter->rss);
> - spin_unlock(&res->lock);
> + spin_lock(&memctlr_lock);
> + shared_pg_cnt = page->shared_count;
> +
> + while (shared_pg_cnt != NULL) {
> + if (shared_pg_cnt->rg_bit_id == res->bit_id) {
> + atomic_dec(&shared_pg_cnt->count);
> +
> + if (atomic_read(&shared_pg_cnt->count) == 0) {
> + /* this group don't use this page anymore */
> + /* remove the rg_shared_pg_cnt struct from the
> list */
> + if (previous != NULL)
> + previous->next = shared_pg_cnt->next;
> + else
> + page->shared_count = NULL;
> + /* and free it */
> + kfree(shared_pg_cnt);
> + /* remove the rgroup from the page bitmap */
> + page->rg_bitmap = (page->rg_bitmap &
> ~res->bit_id);
> +
> + atomic_long_dec(&res->counter.rss);
> + atomic_long_dec(&mm->counter->rss);
> + }
> + spin_unlock(&memctlr_lock);
> + return;
> + }
> + previous = shared_pg_cnt;
> + shared_pg_cnt = shared_pg_cnt->next;
> + }
> + /* should never get here */
> + BUG();
> }
>
> static void memctlr_init_new(struct memctlr *res)
> @@ -176,6 +228,7 @@ static void memctlr_init_new(struct memc
> res->shares.max_shares = SHARE_DONT_CARE;
> res->shares.child_shares_divisor = SHARE_DEFAULT_DIVISOR;
> res->shares.unused_min_shares = SHARE_DEFAULT_DIVISOR;
> + res->bit_id = (1 << rg_bitmap_shift_index);
>
> memctlr_init_mem_counter(&res->counter);
> spin_lock_init(&res->lock);
> @@ -189,6 +242,7 @@ static struct res_shares *memctlr_alloc_
> if (!res)
> return NULL;
> memctlr_init_new(res);
> + rg_bitmap_shift_index++; /* overflow TBC */
> if (is_res_group_root(rgroup)) {
> root_rgroup = rgroup;
> memctlr_root = res;
> @@ -386,6 +440,9 @@ static void memctlr_move_task(struct tas
> if (p->pid != p->tgid)
> return;
>
> + printk("memctlr_move_task not yet available \n");
> + return;
> +
Any particular reason for not implementing migration in this patch.
> oldres = get_memctlr_from_shares(old);
> newres = get_memctlr_from_shares(new);
>
> @@ -414,6 +471,7 @@ int __init memctlr_init(void)
> {
> if (memctlr_rg.ctlr_id != NO_RES_ID)
> return -EBUSY; /* already registered */
> + spin_lock_init(&memctlr_lock);
> return register_controller(&memctlr_rg);
> }
>
>
Do you have any test results with this patch? Showing the effect of
tracking shared pages
--
Balbir Singh,
Linux Technology Center,
IBM Software Labs
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
ckrm-tech mailing list
https://lists.sourceforge.net/lists/listinfo/ckrm-tech
