From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: with ECARTIS (v1.0.0; list xfs); Mon, 18 Feb 2008 15:01:01 -0800 (PST) Received: from larry.melbourne.sgi.com (larry.melbourne.sgi.com [134.14.52.130]) by oss.sgi.com (8.12.11.20060308/8.12.11/SuSE Linux 0.7) with SMTP id m1IN0t1j001860 for ; Mon, 18 Feb 2008 15:00:57 -0800 Date: Tue, 19 Feb 2008 10:01:12 +1100 From: David Chinner Subject: [patch, debug, 2/2] Use power-of-2 size ktrace buffers Message-ID: <20080218230112.GU155407@sgi.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Sender: xfs-bounce@oss.sgi.com Errors-to: xfs-bounce@oss.sgi.com List-Id: xfs To: xfs-dev Cc: xfs-oss Now that the ktrace_enter() code is using atomics, the non-power-of-2 buffer sizes - which require modulus operations to get the index - are showing up as using substantial CPU in the profiles. Force the buffer sizes to be rounded up to the nearest power of two and use masking rather than modulus operations to convert the index counter to the buffer index. This reduces ktrace_enter overhead to 8% of a CPU time, and again almost halves the trace intensive test runtime. Signed-off-by: Dave Chinner --- fs/xfs/support/ktrace.c | 22 ++++++++++++++-------- fs/xfs/support/ktrace.h | 1 + 2 files changed, 15 insertions(+), 8 deletions(-) Index: 2.6.x-xfs-new/fs/xfs/support/ktrace.c =================================================================== --- 2.6.x-xfs-new.orig/fs/xfs/support/ktrace.c 2008-02-18 15:33:13.849314747 +1100 +++ 2.6.x-xfs-new/fs/xfs/support/ktrace.c 2008-02-18 15:33:16.920921133 +1100 @@ -24,7 +24,7 @@ static int ktrace_zentries; void __init ktrace_init(int zentries) { - ktrace_zentries = zentries; + ktrace_zentries = roundup_pow_of_two(zentries); ktrace_hdr_zone = kmem_zone_init(sizeof(ktrace_t), "ktrace_hdr"); @@ -47,13 +47,16 @@ ktrace_uninit(void) * ktrace_alloc() * * Allocate a ktrace header and enough buffering for the given - * number of entries. + * number of entries. Round the number of entries up to a + * power of 2 so we can do fast masking to get the index from + * the atomic index counter. */ ktrace_t * ktrace_alloc(int nentries, unsigned int __nocast sleep) { ktrace_t *ktp; ktrace_entry_t *ktep; + int entries; ktp = (ktrace_t*)kmem_zone_alloc(ktrace_hdr_zone, sleep); @@ -70,11 +73,12 @@ ktrace_alloc(int nentries, unsigned int /* * Special treatment for buffers with the ktrace_zentries entries */ - if (nentries == ktrace_zentries) { + entries = roundup_pow_of_two(nentries); + if (entries == ktrace_zentries) { ktep = (ktrace_entry_t*)kmem_zone_zalloc(ktrace_ent_zone, sleep); } else { - ktep = (ktrace_entry_t*)kmem_zalloc((nentries * sizeof(*ktep)), + ktep = (ktrace_entry_t*)kmem_zalloc((entries * sizeof(*ktep)), sleep | KM_LARGE); } @@ -91,7 +95,9 @@ ktrace_alloc(int nentries, unsigned int } ktp->kt_entries = ktep; - ktp->kt_nentries = nentries; + ktp->kt_nentries = entries; + ASSERT(is_power_of_2(entries)); + ktp->kt_index_mask = entries - 1; atomic_set(&ktp->kt_index, 0); ktp->kt_rollover = 0; return ktp; @@ -160,7 +166,7 @@ ktrace_enter( * Grab an entry by pushing the index up to the next one. */ index = atomic_add_return(1, &ktp->kt_index); - index = (index - 1) % ktp->kt_nentries; + index = (index - 1) & ktp->kt_index_mask; if (!ktp->kt_rollover && index == ktp->kt_nentries - 1) ktp->kt_rollover = 1; @@ -197,7 +203,7 @@ ktrace_nentries( if (ktp == NULL) return 0; - index = atomic_read(&ktp->kt_index) % ktp->kt_nentries; + index = atomic_read(&ktp->kt_index) & ktp->kt_index_mask; return (ktp->kt_rollover ? ktp->kt_nentries : index); } @@ -223,7 +229,7 @@ ktrace_first(ktrace_t *ktp, ktrace_sna int nentries; if (ktp->kt_rollover) - index = atomic_read(&ktp->kt_index) % ktp->kt_nentries; + index = atomic_read(&ktp->kt_index) & ktp->kt_index_mask; else index = 0; Index: 2.6.x-xfs-new/fs/xfs/support/ktrace.h =================================================================== --- 2.6.x-xfs-new.orig/fs/xfs/support/ktrace.h 2008-02-18 15:32:28.911073469 +1100 +++ 2.6.x-xfs-new/fs/xfs/support/ktrace.h 2008-02-18 15:33:16.920921133 +1100 @@ -31,6 +31,7 @@ typedef struct ktrace_entry { typedef struct ktrace { int kt_nentries; /* number of entries in trace buf */ atomic_t kt_index; /* current index in entries */ + unsigned int kt_index_mask; int kt_rollover; ktrace_entry_t *kt_entries; /* buffer of entries */ } ktrace_t;