From: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
To: Bob Liu <lliubbo@gmail.com>
Cc: xen-devel@lists.xenproject.org, keir@xen.org,
ian.campbell@citrix.com, JBeulich@suse.com
Subject: Re: [PATCH 15/16] tmem: refator function tmem_ensure_avail_pages()
Date: Fri, 22 Nov 2013 13:22:10 -0500 [thread overview]
Message-ID: <20131122182210.GI8120@phenom.dumpdata.com> (raw)
In-Reply-To: <1384937185-24749-15-git-send-email-bob.liu@oracle.com>
On Wed, Nov 20, 2013 at 04:46:24PM +0800, Bob Liu wrote:
> tmem_ensure_avail_pages() doesn't return a value which is incorrect because
> the caller need to confirm there is enough memory.
Which it was not doing until now. Ouch.
>
> Signed-off-by: Bob Liu <bob.liu@oracle.com>
> ---
> xen/common/tmem.c | 43 +++++++++++++++++++------------------------
> xen/include/xen/tmem_xen.h | 6 ------
> 2 files changed, 19 insertions(+), 30 deletions(-)
>
> diff --git a/xen/common/tmem.c b/xen/common/tmem.c
> index 2e807d4..0bc33ee 100644
> --- a/xen/common/tmem.c
> +++ b/xen/common/tmem.c
> @@ -1296,22 +1296,28 @@ static unsigned long tmem_relinquish_npages(unsigned long n)
> return avail_pages;
> }
>
> -/* Under certain conditions (e.g. if each client is putting pages for exactly
> +/*
> + * Under certain conditions (e.g. if each client is putting pages for exactly
> * one object), once locks are held, freeing up memory may
> * result in livelocks and very long "put" times, so we try to ensure there
> * is a minimum amount of memory (1MB) available BEFORE any data structure
> - * locks are held */
> -static inline void tmem_ensure_avail_pages(void)
> + * locks are held.
> + */
> +static inline bool_t tmem_ensure_avail_pages(void)
> {
> int failed_evict = 10;
> + unsigned long free_mem;
>
> - while ( !tmem_free_mb() )
> - {
> - if ( tmem_evict() )
> - continue;
> - else if ( failed_evict-- <= 0 )
> - break;
> - }
> + do {
> + free_mem = (tmem_page_list_pages + total_free_pages())
> + >> (20 - PAGE_SHIFT);
> + if ( free_mem )
> + return 1;
> + if ( !tmem_evict() )
> + failed_evict--;
> + } while ( failed_evict > 0 );
> +
> + return 0;
> }
>
> /************ TMEM CORE OPERATIONS ************************************/
> @@ -2282,9 +2288,6 @@ long do_tmem_op(tmem_cli_op_t uops)
> struct tmem_pool *pool = NULL;
> struct oid *oidp;
> int rc = 0;
> - bool_t succ_get = 0, succ_put = 0;
> - bool_t non_succ_get = 0, non_succ_put = 0;
> - bool_t flush = 0, flush_obj = 0;
Um, OK, they seem to belong to a different patch?
> bool_t write_lock_set = 0, read_lock_set = 0;
>
> if ( !tmem_initialized )
> @@ -2299,8 +2302,7 @@ long do_tmem_op(tmem_cli_op_t uops)
> if ( unlikely(tmem_get_tmemop_from_client(&op, uops) != 0) )
> {
> tmem_client_err("tmem: can't get tmem struct from %s\n", tmem_client_str);
> - rc = -EFAULT;
> - return rc;
> + return -EFAULT;
This too.
> }
>
> if ( op.cmd == TMEM_CONTROL )
> @@ -2369,28 +2371,21 @@ long do_tmem_op(tmem_cli_op_t uops)
> op.u.creat.uuid[0], op.u.creat.uuid[1]);
> break;
> case TMEM_PUT_PAGE:
> - tmem_ensure_avail_pages();
> - rc = do_tmem_put(pool, oidp, op.u.gen.index, op.u.gen.cmfn,
> + if ( tmem_ensure_avail_pages() )
> + rc = do_tmem_put(pool, oidp, op.u.gen.index, op.u.gen.cmfn,
> tmem_cli_buf_null);
> - if (rc == 1) succ_put = 1;
> - else non_succ_put = 1;
which we ignore..
> break;
> case TMEM_GET_PAGE:
> rc = do_tmem_get(pool, oidp, op.u.gen.index, op.u.gen.cmfn,
> tmem_cli_buf_null);
> - if (rc == 1) succ_get = 1;
> - else non_succ_get = 1;
ugh.
> break;
> case TMEM_FLUSH_PAGE:
> - flush = 1;
> rc = do_tmem_flush_page(pool, oidp, op.u.gen.index);
> break;
> case TMEM_FLUSH_OBJECT:
> rc = do_tmem_flush_object(pool, oidp);
> - flush_obj = 1;
> break;
> case TMEM_DESTROY_POOL:
> - flush = 1;
Those really belong to a seperate cleanup patch.
> rc = do_tmem_destroy_pool(op.pool_id);
> break;
> default:
> diff --git a/xen/include/xen/tmem_xen.h b/xen/include/xen/tmem_xen.h
> index bf3be62..258e8e7 100644
> --- a/xen/include/xen/tmem_xen.h
> +++ b/xen/include/xen/tmem_xen.h
> @@ -164,13 +164,7 @@ static inline void __tmem_free_page(struct page_info *pi)
> atomic_dec(&freeable_page_count);
> }
>
> -static inline unsigned long tmem_free_mb(void)
> -{
> - return (tmem_page_list_pages + total_free_pages()) >> (20 - PAGE_SHIFT);
> -}
> -
> /* "Client" (==domain) abstraction */
> -
> struct client;
> static inline struct client *tmem_client_from_cli_id(domid_t cli_id)
> {
> --
> 1.7.10.4
>
next prev parent reply other threads:[~2013-11-22 18:22 UTC|newest]
Thread overview: 49+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-11-20 8:46 [PATCH 01/16] tmem: cleanup: drop some debug code Bob Liu
2013-11-20 8:46 ` [PATCH 02/16] tmem: cleanup: drop useless function 'tmem_copy_page' Bob Liu
2013-11-20 8:46 ` [PATCH 03/16] tmem: cleanup: rm unused tmem_op Bob Liu
2013-11-22 17:38 ` Konrad Rzeszutek Wilk
2013-11-25 9:43 ` Jan Beulich
2013-11-25 9:52 ` Ian Campbell
2013-11-25 9:58 ` Jan Beulich
2013-11-25 16:37 ` Konrad Rzeszutek Wilk
2013-11-25 16:40 ` Ian Campbell
2013-11-25 17:09 ` Konrad Rzeszutek Wilk
2013-11-25 17:12 ` Ian Campbell
2013-11-25 19:56 ` Konrad Rzeszutek Wilk
2013-11-26 8:56 ` Bob Liu
2013-11-20 8:46 ` [PATCH 04/16] tmem: cleanup: rm unneeded parameters from put path Bob Liu
2013-11-22 17:54 ` Konrad Rzeszutek Wilk
2013-11-26 8:22 ` Bob Liu
2013-11-20 8:46 ` [PATCH 05/16] tmem: cleanup: rm unneeded parameters from get path Bob Liu
2013-11-22 17:55 ` Konrad Rzeszutek Wilk
2013-11-20 8:46 ` [PATCH 06/16] tmem: cleanup: reorg do_tmem_put() Bob Liu
2013-11-22 18:04 ` Konrad Rzeszutek Wilk
2013-11-20 8:46 ` [PATCH 07/16] tmem: drop unneeded is_ephemeral() and is_private() Bob Liu
2013-11-20 8:46 ` [PATCH 08/16] tmem: cleanup: rm useless EXPORT/FORWARD define Bob Liu
2013-11-22 18:05 ` Konrad Rzeszutek Wilk
2013-11-20 8:46 ` [PATCH 09/16] tmem: cleanup: drop tmemc_list() temporary Bob Liu
2013-11-22 18:07 ` Konrad Rzeszutek Wilk
2013-11-26 8:28 ` Bob Liu
2013-11-22 21:00 ` Konrad Rzeszutek Wilk
2013-11-20 8:46 ` [PATCH 10/16] tmem: cleanup: drop runtime statistics Bob Liu
2013-11-22 18:08 ` Konrad Rzeszutek Wilk
2013-11-20 8:46 ` [PATCH 11/16] tmem: cleanup: drop tmem_lock_all Bob Liu
2013-11-20 8:46 ` [PATCH 12/16] tmem: cleanup: refactor the alloc/free path Bob Liu
2013-11-20 8:46 ` [PATCH 13/16] tmem: cleanup: __tmem_alloc_page: drop unneed parameters Bob Liu
2013-11-22 18:17 ` Konrad Rzeszutek Wilk
2013-11-26 8:41 ` Bob Liu
2013-11-26 17:38 ` Konrad Rzeszutek Wilk
2013-11-20 8:46 ` [PATCH 14/16] tmem: cleanup: drop useless functions from head file Bob Liu
2013-11-27 14:38 ` Andrew Cooper
2013-11-27 14:52 ` Konrad Rzeszutek Wilk
2013-11-27 14:59 ` Andrew Cooper
2013-11-27 15:55 ` Jan Beulich
2013-11-20 8:46 ` [PATCH 15/16] tmem: refator function tmem_ensure_avail_pages() Bob Liu
2013-11-22 18:22 ` Konrad Rzeszutek Wilk [this message]
2013-11-20 8:46 ` [PATCH 16/16] tmem: cleanup: rename tmem_relinquish_npages() Bob Liu
2013-11-20 9:08 ` [PATCH 01/16] tmem: cleanup: drop some debug code Jan Beulich
2013-11-20 9:19 ` Bob Liu
2013-11-20 9:25 ` Jan Beulich
2013-11-20 13:51 ` Konrad Rzeszutek Wilk
2013-11-20 14:21 ` Jan Beulich
2013-11-20 18:46 ` Konrad Rzeszutek Wilk
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20131122182210.GI8120@phenom.dumpdata.com \
--to=konrad.wilk@oracle.com \
--cc=JBeulich@suse.com \
--cc=ian.campbell@citrix.com \
--cc=keir@xen.org \
--cc=lliubbo@gmail.com \
--cc=xen-devel@lists.xenproject.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).