From: Robert Love <robert.w.love@intel.com>
To: James.Bottomley@HansenPartnership.com, linux-scsi@vger.kernel.org
Cc: Vasu Dev <vasu.dev@intel.com>, Robert Love <robert.w.love@intel.com>
Subject: [PATCH 04/64] fcoe, libfc: adds per cpu exch pool within exchange manager(EM)
Date: Tue, 25 Aug 2009 13:58:47 -0700 [thread overview]
Message-ID: <20090825205847.1553.96759.stgit@localhost.localdomain> (raw)
In-Reply-To: <20090825205826.1553.94414.stgit@localhost.localdomain>
From: Vasu Dev <vasu.dev@intel.com>
Adds per cpu exch pool for these reasons:-
1. Currently an EM instance is shared across all cpus to manage
all exches for all cpus. This required em_lock across all
cpus for an exch alloc, free, lookup and reset each frame
and that made em_lock expensive, so instead having per cpu
exch pool with their own per cpu pool lock will likely reduce
locking contention in fast path for an exch alloc, free and
lookup.
2. Per cpu exch pool will likely improve cache hit ratio since
all frames of an exch will be processed on the same cpu on
which exch originated.
This patch is only prep work to help in keeping complexity of next
patch low, so this patch only sets up per cpu exch pool and related
helper funcs to be used by next patch. The next patch fully makes
use of per cpu exch pool in all code paths ie. tx, rx and reset.
Divides per EM exch id range equally across all cpus to setup per
cpu exch pool. This division is such that lower bits of exch id
carries cpu number info on which exch originated, later a simple
bitwise AND operation on exch id of incoming frame with fc_cpu_mask
retrieves cpu number info to direct all frames to same cpu on which
exch originated. This required a global fc_cpu_mask and fc_cpu_order
initialized to max possible cpus number nr_cpu_ids rounded up to 2's
power, this will be used in mapping exch id and exch ptr array
index in pool during exch allocation, find or reset code paths.
Adds a check in fc_exch_mgr_alloc() to ensure specified min_xid
lower bits are zero since these bits are used to carry cpu info.
Adds and initializes struct fc_exch_pool with all required fields
to manage exches in pool.
Allocates per cpu struct fc_exch_pool with memory for exches array
for range of exches per pool. The exches array memory is followed
by struct fc_exch_pool.
Adds fc_exch_ptr_get/set() helper functions to get/set exch ptr in
pool exches array at specified array index.
Increases default FCOE_MAX_XID to 0x0FFF from 0x07EF, so that more
exches are available per cpu after above described exch id range
division across all cpus to each pool.
Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
---
drivers/scsi/fcoe/fcoe.h | 2 -
drivers/scsi/libfc/fc_exch.c | 89 +++++++++++++++++++++++++++++++++++++++++-
include/scsi/libfc.h | 1
3 files changed, 88 insertions(+), 4 deletions(-)
diff --git a/drivers/scsi/fcoe/fcoe.h b/drivers/scsi/fcoe/fcoe.h
index 65120e2..550d1e4 100644
--- a/drivers/scsi/fcoe/fcoe.h
+++ b/drivers/scsi/fcoe/fcoe.h
@@ -38,7 +38,7 @@
#define FCOE_MAX_OUTSTANDING_COMMANDS 1024
#define FCOE_MIN_XID 0x0000 /* the min xid supported by fcoe_sw */
-#define FCOE_MAX_XID 0x07ef /* the max xid supported by fcoe_sw */
+#define FCOE_MAX_XID 0x0FFF /* the max xid supported by fcoe_sw */
unsigned int fcoe_debug_logging;
module_param_named(debug_logging, fcoe_debug_logging, int, S_IRUGO|S_IWUSR);
diff --git a/drivers/scsi/libfc/fc_exch.c b/drivers/scsi/libfc/fc_exch.c
index 40c3427..9cbe8d6 100644
--- a/drivers/scsi/libfc/fc_exch.c
+++ b/drivers/scsi/libfc/fc_exch.c
@@ -32,6 +32,9 @@
#include <scsi/libfc.h>
#include <scsi/fc_encode.h>
+u16 fc_cpu_mask; /* cpu mask for possible cpus */
+EXPORT_SYMBOL(fc_cpu_mask);
+static u16 fc_cpu_order; /* 2's power to represent total possible cpus */
static struct kmem_cache *fc_em_cachep; /* cache for exchanges */
/*
@@ -48,6 +51,20 @@ static struct kmem_cache *fc_em_cachep; /* cache for exchanges */
*/
/*
+ * Per cpu exchange pool
+ *
+ * This structure manages per cpu exchanges in array of exchange pointers.
+ * This array is allocated followed by struct fc_exch_pool memory for
+ * assigned range of exchanges to per cpu pool.
+ */
+struct fc_exch_pool {
+ u16 next_index; /* next possible free exchange index */
+ u16 total_exches; /* total allocated exchanges */
+ spinlock_t lock; /* exch pool lock */
+ struct list_head ex_list; /* allocated exchanges list */
+};
+
+/*
* Exchange manager.
*
* This structure is the center for creating exchanges and sequences.
@@ -66,6 +83,8 @@ struct fc_exch_mgr {
u32 total_exches; /* total allocated exchanges */
struct list_head ex_list; /* allocated exchanges list */
mempool_t *ep_pool; /* reserve ep's */
+ u16 pool_max_index; /* max exch array index in exch pool */
+ struct fc_exch_pool *pool; /* per cpu exch pool */
/*
* currently exchange mgr stats are updated but not used.
@@ -303,6 +322,19 @@ static int fc_exch_done_locked(struct fc_exch *ep)
return rc;
}
+static inline struct fc_exch *fc_exch_ptr_get(struct fc_exch_pool *pool,
+ u16 index)
+{
+ struct fc_exch **exches = (struct fc_exch **)(pool + 1);
+ return exches[index];
+}
+
+static inline void fc_exch_ptr_set(struct fc_exch_pool *pool, u16 index,
+ struct fc_exch *ep)
+{
+ ((struct fc_exch **)(pool + 1))[index] = ep;
+}
+
static void fc_exch_mgr_delete_ep(struct fc_exch *ep)
{
struct fc_exch_mgr *mp;
@@ -1751,6 +1783,7 @@ static void fc_exch_mgr_destroy(struct kref *kref)
*/
WARN_ON(mp->total_exches != 0);
mempool_destroy(mp->ep_pool);
+ free_percpu(mp->pool);
kfree(mp);
}
@@ -1770,8 +1803,13 @@ struct fc_exch_mgr *fc_exch_mgr_alloc(struct fc_lport *lp,
{
struct fc_exch_mgr *mp;
size_t len;
+ u16 pool_exch_range;
+ size_t pool_size;
+ unsigned int cpu;
+ struct fc_exch_pool *pool;
- if (max_xid <= min_xid || max_xid == FC_XID_UNKNOWN) {
+ if (max_xid <= min_xid || max_xid == FC_XID_UNKNOWN ||
+ (min_xid & fc_cpu_mask) != 0) {
FC_LPORT_DBG(lp, "Invalid min_xid 0x:%x and max_xid 0x:%x\n",
min_xid, max_xid);
return NULL;
@@ -1802,10 +1840,31 @@ struct fc_exch_mgr *fc_exch_mgr_alloc(struct fc_lport *lp,
if (!mp->ep_pool)
goto free_mp;
+ /*
+ * Setup per cpu exch pool with entire exchange id range equally
+ * divided across all cpus. The exch pointers array memory is
+ * allocated for exch range per pool.
+ */
+ pool_exch_range = (mp->max_xid - mp->min_xid + 1) / (fc_cpu_mask + 1);
+ mp->pool_max_index = pool_exch_range - 1;
+
+ /*
+ * Allocate and initialize per cpu exch pool
+ */
+ pool_size = sizeof(*pool) + pool_exch_range * sizeof(struct fc_exch *);
+ mp->pool = __alloc_percpu(pool_size, __alignof__(struct fc_exch_pool));
+ if (!mp->pool)
+ goto free_mempool;
+ for_each_possible_cpu(cpu) {
+ pool = per_cpu_ptr(mp->pool, cpu);
+ spin_lock_init(&pool->lock);
+ INIT_LIST_HEAD(&pool->ex_list);
+ }
+
kref_init(&mp->kref);
if (!fc_exch_mgr_add(lp, mp, match)) {
- mempool_destroy(mp->ep_pool);
- goto free_mp;
+ free_percpu(mp->pool);
+ goto free_mempool;
}
/*
@@ -1816,6 +1875,8 @@ struct fc_exch_mgr *fc_exch_mgr_alloc(struct fc_lport *lp,
kref_put(&mp->kref, fc_exch_mgr_destroy);
return mp;
+free_mempool:
+ mempool_destroy(mp->ep_pool);
free_mp:
kfree(mp);
return NULL;
@@ -1975,6 +2036,28 @@ int fc_exch_init(struct fc_lport *lp)
if (!lp->tt.seq_exch_abort)
lp->tt.seq_exch_abort = fc_seq_exch_abort;
+ /*
+ * Initialize fc_cpu_mask and fc_cpu_order. The
+ * fc_cpu_mask is set for nr_cpu_ids rounded up
+ * to order of 2's * power and order is stored
+ * in fc_cpu_order as this is later required in
+ * mapping between an exch id and exch array index
+ * in per cpu exch pool.
+ *
+ * This round up is required to align fc_cpu_mask
+ * to exchange id's lower bits such that all incoming
+ * frames of an exchange gets delivered to the same
+ * cpu on which exchange originated by simple bitwise
+ * AND operation between fc_cpu_mask and exchange id.
+ */
+ fc_cpu_mask = 1;
+ fc_cpu_order = 0;
+ while (fc_cpu_mask < nr_cpu_ids) {
+ fc_cpu_mask <<= 1;
+ fc_cpu_order++;
+ }
+ fc_cpu_mask--;
+
return 0;
}
EXPORT_SYMBOL(fc_exch_init);
diff --git a/include/scsi/libfc.h b/include/scsi/libfc.h
index c2b928c..3206338 100644
--- a/include/scsi/libfc.h
+++ b/include/scsi/libfc.h
@@ -343,6 +343,7 @@ static inline bool fc_fcp_is_read(const struct fc_fcp_pkt *fsp)
struct fc_exch_mgr;
struct fc_exch_mgr_anchor;
+extern u16 fc_cpu_mask; /* cpu mask for possible cpus */
/*
* Sequence.
next prev parent reply other threads:[~2009-08-25 20:59 UTC|newest]
Thread overview: 69+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-08-25 20:58 [PATCH 00/64] libfc, libfcoe and fcoe updates for scsi-misc Robert Love
2009-08-25 20:58 ` [PATCH 01/64] fcoe: Add format spacing to FCOE_NETDEV_DBG debug macro Robert Love
2009-08-25 20:58 ` [PATCH 02/64] libfc: Fix misleading debug statement Robert Love
2009-08-25 20:58 ` [PATCH 03/64] fcoe: libfcoe: extra semicolon in CHECK_LOGGING macros causes compile error Robert Love
2009-08-25 20:58 ` Robert Love [this message]
2009-08-25 20:58 ` [PATCH 05/64] fcoe, libfc: fully makes use of per cpu exch pool and then removes em_lock Robert Love
2009-08-25 20:58 ` [PATCH 06/64] libfc: Export FC headers Robert Love
2009-08-27 10:27 ` Christof Schmitt
2009-08-27 17:42 ` Robert Love
2009-08-28 8:32 ` Swen Schillig
2009-08-28 16:58 ` Robert Love
2009-08-25 20:59 ` [PATCH 07/64] fcoe: Add sysfs parameter to fcoe for minimum DDP read I/O size Robert Love
2009-08-25 20:59 ` [PATCH 08/64] libfcoe: fcoe_ctlr_destroy use cancel_work_sync instead of flush_work Robert Love
2009-08-25 20:59 ` [PATCH 09/64] fcoe: fix missing error check in call to fcoe_if_init Robert Love
2009-08-25 20:59 ` [PATCH 10/64] fcoe: remove unnecessary list and lock initializations Robert Love
2009-08-25 20:59 ` [PATCH 11/64] fcoe: interface changes to fcoe_if_create and fcoe_if_destroy Robert Love
2009-08-25 20:59 ` [PATCH 12/64] fcoe: Introduce and allocate fcoe_interface structure, 1:1 with net_device Robert Love
2009-08-25 20:59 ` [PATCH 13/64] fcoe: move netdev to fcoe_interface Robert Love
2009-08-25 20:59 ` [PATCH 14/64] fcoe: move packet handlers from fcoe_port " Robert Love
2009-08-25 20:59 ` [PATCH 15/64] fcoe: move FIP controller " Robert Love
2009-08-25 20:59 ` [PATCH 16/64] fcoe: move offload exchange manager pointer " Robert Love
2009-08-25 20:59 ` [PATCH 17/64] fcoe: remove fcoe_interface->priv pointer Robert Love
2009-08-25 21:00 ` [PATCH 18/64] fcoe: fcoe_interface create, destroy and refcounting Robert Love
2009-08-25 21:00 ` [PATCH 19/64] fcoe: split out per interface setup Robert Love
2009-08-25 21:00 ` [PATCH 20/64] fcoe: add mutex to protect create and destroy Robert Love
2009-08-25 21:00 ` [PATCH 21/64] fcoe: move the host-list add/remove to keep out VN_Ports Robert Love
2009-08-25 21:00 ` [PATCH 22/64] fcoe: Fix module ref count bug by adding NETDEV UNREGISTER handling Robert Love
2009-08-25 21:00 ` [PATCH 23/64] fcoe: use rtnl mutex in place of hostlist lock Robert Love
2009-08-25 21:00 ` [PATCH 24/64] libfc: prepare to split off struct fc_rport_priv from fc_rport_libfc_priv Robert Love
2009-08-25 21:00 ` [PATCH 25/64] libfc: change interface for rport_create Robert Love
2009-08-25 21:00 ` [PATCH 26/64] libfc: fix RPORT_TO_PRIV and PRIV_TO_RPORT() macros Robert Love
2009-08-25 21:00 ` [PATCH 27/64] libfc: make fc_rport_priv the primary rport interface Robert Love
2009-08-25 21:00 ` [PATCH 28/64] libfc: change elsct to use FC_ID instead of rdata Robert Love
2009-08-25 21:01 ` [PATCH 29/64] libfc: make rport structure optional Robert Love
2009-08-25 21:01 ` [PATCH 30/64] libfc: rearrange code in fc_rport_work Robert Love
2009-08-25 21:01 ` [PATCH 31/64] libfc: rename rport event CREATED to READY Robert Love
2009-08-25 21:01 ` [PATCH 32/64] libfc: don't create dummy (rogue) remote ports Robert Love
2009-08-25 21:01 ` [PATCH 33/64] libfc: fix rport event race between READY and LOGO Robert Love
2009-08-25 21:01 ` [PATCH 34/64] libfc: eliminate disc->event Robert Love
2009-08-25 21:01 ` [PATCH 35/64] libfc: remove unused disc->delay element Robert Love
2009-08-25 21:01 ` [PATCH 36/64] libfc: rport debug messages were printing pointer values Robert Love
2009-08-25 21:01 ` [PATCH 37/64] libfc: simplify fc_lport_rport_callback Robert Love
2009-08-25 21:01 ` [PATCH 38/64] libfc: make rport module maintain the rport list Robert Love
2009-08-25 21:01 ` [PATCH 39/64] libfc: have rport_create do a lookup for pre-existing rports first Robert Love
2009-08-25 21:02 ` [PATCH 40/64] libfc: change to make remote port callback optional Robert Love
2009-08-25 21:02 ` [PATCH 41/64] libfc: move rport_lookup into fc_rport.c Robert Love
2009-08-25 21:02 ` [PATCH 42/64] libfc: do not log off rports before or after discovery Robert Love
2009-08-25 21:02 ` [PATCH 43/64] libfc: discovery restart sequence error fix Robert Love
2009-08-25 21:02 ` [PATCH 44/64] libfc: rearrange code in fc_disc_gpn_ft_resp() Robert Love
2009-08-25 21:02 ` [PATCH 45/64] libfc: handle discovery failure more correctly Robert Love
2009-08-25 21:02 ` [PATCH 46/64] libfc: fix: empty zone causes endless discovery retries Robert Love
2009-08-25 21:02 ` [PATCH 47/64] libfc: discovery retry should clear pending first Robert Love
2009-08-25 21:02 ` [PATCH 48/64] libfc: discovery gpn_ft parse bug Robert Love
2009-08-25 21:02 ` [PATCH 49/64] libfc: clean up point-to-point discovery code Robert Love
2009-08-25 21:02 ` [PATCH 50/64] libfc: don't do discovery before callback is set Robert Love
2009-08-25 21:02 ` [PATCH 51/64] libfc: Initialize fc_rport_identifiers inside fc_rport_create Robert Love
2009-08-25 21:03 ` [PATCH 52/64] libfc: Always reset remote port roles when receiving PRLI Robert Love
2009-08-25 21:03 ` [PATCH 53/64] libfc: move remote port lookup for ELS requests into fc_rport.c Robert Love
2009-08-25 21:03 ` [PATCH 54/64] libfc: fix: rport_recv_req needs disc_mutex when calling rport_lookup Robert Love
2009-08-25 21:03 ` [PATCH 55/64] libfc: improve debug messages for ELS response handlers Robert Love
2009-08-25 21:03 ` [PATCH 56/64] libfc: correctly handle incoming PLOGI request Robert Love
2009-08-25 21:03 ` [PATCH 57/64] libfc: fix rport error handling for login-required and invalid ops Robert Love
2009-08-25 21:03 ` [PATCH 58/64] libfc: re-login to remote ports that send us LOGO Robert Love
2009-08-25 21:03 ` [PATCH 59/64] libfc: LOGO response code had extraeous enter_rtv Robert Love
2009-08-25 21:03 ` [PATCH 60/64] libfc: use ADISC to verify rport login state Robert Love
2009-08-25 21:03 ` [PATCH 61/64] libfc: fix handling of incoming Discover Address (ADISC) requests Robert Love
2009-08-25 21:03 ` [PATCH 62/64] libfc: send GPN_ID in reaction to single-port RSCNs Robert Love
2009-08-25 21:04 ` [PATCH 63/64] libfc: don't swap OX_ID and RX_ID when sending BA_RJT Robert Love
2009-08-25 21:04 ` [PATCH 64/64] fcoe: flush per-cpu thread work when destroying interface Robert Love
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=20090825205847.1553.96759.stgit@localhost.localdomain \
--to=robert.w.love@intel.com \
--cc=James.Bottomley@HansenPartnership.com \
--cc=linux-scsi@vger.kernel.org \
--cc=vasu.dev@intel.com \
/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