#ifndef _GNILND_GNILND_H_
#define _GNILND_GNILND_H_
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <libcfs/libcfs.h>
#include <lnet/lnet.h>
#include <lnet/lib-lnet.h>
-#include <lnet/lnet-sysctl.h>
#include <gni_pub.h>
#include "gnilnd_version.h"
/* tune down some COMPUTE options as they won't see the same number of connections and
* don't need the throughput of multiple threads by default */
#if defined(CONFIG_CRAY_COMPUTE)
+#ifdef CONFIG_MK1OM
+#define GNILND_SCHED_THREADS 2 /* default # of kgnilnd_scheduler threads */
+#else
#define GNILND_SCHED_THREADS 1 /* default # of kgnilnd_scheduler threads */
+#endif
#define GNILND_FMABLK 64 /* default number of mboxes per fmablk */
#define GNILND_SCHED_NICE 0 /* default nice value for scheduler threads */
#define GNILND_COMPUTE 1 /* compute image */
int *kgn_dgram_timeout; /* max time for dgram mover to run before scheduling */
int *kgn_sched_nice; /* nice value for kgnilnd scheduler threads */
int *kgn_reverse_rdma; /* Reverse RDMA setting */
+ int *kgn_eager_credits; /* allocated eager buffers */
+ int *kgn_efault_lbug; /* Should we LBUG on receiving an EFAULT */
#if CONFIG_SYSCTL && !CFS_SYSFS_MODULE_PARM
cfs_sysctl_table_header_t *kgn_sysctl; /* sysctl interface */
#endif
int gnd_id; /* device id, also index in kgn_devices */
__u32 gnd_nid; /* ph host ID translated to NID */
struct list_head gnd_fma_buffs; /* list of FMA memory blocks */
- struct semaphore gnd_fmablk_sem; /* semaphore for FMA block memory alloc/free */
+ struct mutex gnd_fmablk_mutex; /* mutex for FMA block memory alloc/free */
spinlock_t gnd_fmablk_lock; /* lock for mbox alloc/release */
atomic_t gnd_nfmablk; /* # of fmablk live */
atomic_t gnd_fmablk_vers; /* gnd_fma_bufs stamp */
unsigned long gnp_last_alive; /* last time I had valid comms */
int gnp_last_dgram_errno; /* last error dgrams saw */
unsigned long gnp_last_dgram_time; /* last time I tried to connect */
- unsigned long gnp_reconnect_time; /* CURRENT_SECONDS when reconnect OK */
+ unsigned long gnp_reconnect_time; /* get_seconds() when reconnect OK */
unsigned long gnp_reconnect_interval; /* exponential backoff */
atomic_t gnp_dirty_eps; /* # of old but yet to be destroyed EPs from conns */
int gnp_down; /* rca says peer down */
wait_queue_head_t kgn_ruhroh_waitq; /* ruhroh thread wakeup */
int kgn_quiesce_trigger; /* should we quiesce ? */
atomic_t kgn_nquiesce; /* how many quiesced ? */
- struct semaphore kgn_quiesce_sem; /* serialize ruhroh task, startup and shutdown */
+ struct mutex kgn_quiesce_mutex; /* serialize ruhroh task, startup and shutdown */
int kgn_needs_reset; /* we need stack reset */
/* These next three members implement communication from gnilnd into
struct list_head *kgn_conns; /* conns hashed by cqid */
atomic_t kgn_nconns; /* # connections extant */
+ atomic_t kgn_neager_allocs; /* # of eager allocations */
__u64 kgn_peerstamp; /* when I started up */
__u64 kgn_connstamp; /* conn stamp generator */
int kgn_conn_version; /* version flag for conn tables */
wait_queue_head_t kgn_reaper_waitq; /* reaper sleeps here */
spinlock_t kgn_reaper_lock; /* serialise */
- cfs_mem_cache_t *kgn_rx_cache; /* rx descriptor space */
- cfs_mem_cache_t *kgn_tx_cache; /* tx descriptor memory */
- cfs_mem_cache_t *kgn_tx_phys_cache; /* tx phys descriptor memory */
+ struct kmem_cache *kgn_rx_cache; /* rx descriptor space */
+ struct kmem_cache *kgn_tx_cache; /* tx descriptor memory */
+ struct kmem_cache *kgn_tx_phys_cache; /* tx phys descriptor memory */
atomic_t kgn_ntx; /* # tx in use */
- cfs_mem_cache_t *kgn_dgram_cache; /* outgoing datagrams */
+ struct kmem_cache *kgn_dgram_cache; /* outgoing datagrams */
struct page ***kgn_cksum_map_pages; /* page arrays for mapping pages on checksum */
__u64 kgn_cksum_npages; /* Number of pages allocated for checksumming */
atomic_t kgn_rev_offset; /* number of time REV rdma have been misaligned offsets */
atomic_t kgn_rev_length; /* Number of times REV rdma have been misaligned lengths */
atomic_t kgn_rev_copy_buff; /* Number of times REV rdma have had to make a copy buffer */
+ struct socket *kgn_sock; /* for Apollo */
} kgn_data_t;
extern kgn_data_t kgnilnd_data;
(atomic_read(&kgnilnd_data.kgn_nquiesce) == \
atomic_read(&kgnilnd_data.kgn_nthreads))
-#define KGNILND_SPIN_QUIESCE \
-do { \
- /* E.T phone home */ \
- atomic_inc(&kgnilnd_data.kgn_nquiesce); \
- CDEBUG(D_NET, "Waiting for thread pause to be over...\n"); \
- while (kgnilnd_data.kgn_quiesce_trigger) { \
- set_current_state(TASK_INTERRUPTIBLE); \
- cfs_schedule_timeout_and_set_state(TASK_INTERRUPTIBLE, \
- cfs_time_seconds(1)); \
- } \
- /* Mom, my homework is done */ \
- CDEBUG(D_NET, "Waking up from thread pause\n"); \
- atomic_dec(&kgnilnd_data.kgn_nquiesce); \
+#define KGNILND_SPIN_QUIESCE \
+do { \
+ /* E.T phone home */ \
+ atomic_inc(&kgnilnd_data.kgn_nquiesce); \
+ CDEBUG(D_NET, "Waiting for thread pause to be over...\n"); \
+ while (kgnilnd_data.kgn_quiesce_trigger) { \
+ set_current_state(TASK_INTERRUPTIBLE); \
+ schedule_timeout(HZ); \
+ } \
+ /* Mom, my homework is done */ \
+ CDEBUG(D_NET, "Waking up from thread pause\n"); \
+ atomic_dec(&kgnilnd_data.kgn_nquiesce); \
} while(0)
/* use macros for addref/decref to get the calling function name in the CDEBUG */
int kgnilnd_allocate_phys_fmablk(kgn_device_t *device);
int kgnilnd_map_phys_fmablk(kgn_device_t *device);
-void kgnilnd_unmap_phys_fmablk(kgn_device_t *device);
+void kgnilnd_unmap_fma_blocks(kgn_device_t *device);
void kgnilnd_free_phys_fmablk(kgn_device_t *device);
int kgnilnd_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
int kgnilnd_schedule_process_conn(kgn_conn_t *conn, int sched_intent);
void kgnilnd_schedule_dgram(kgn_device_t *dev);
-int kgnilnd_create_peer_safe(kgn_peer_t **peerp, lnet_nid_t nid, kgn_net_t *net);
+int kgnilnd_create_peer_safe(kgn_peer_t **peerp, lnet_nid_t nid, kgn_net_t *net, int node_state);
void kgnilnd_add_peer_locked(lnet_nid_t nid, kgn_peer_t *new_stub_peer, kgn_peer_t **peerp);
int kgnilnd_add_peer(kgn_net_t *net, lnet_nid_t nid, kgn_peer_t **peerp);
int kgnilnd_report_node_state(lnet_nid_t nid, int down);
void kgnilnd_wakeup_rca_thread(void);
int kgnilnd_start_rca_thread(void);
+int kgnilnd_get_node_state(__u32 nid);
int kgnilnd_tunables_init(void);
void kgnilnd_tunables_fini(void);
int kgnilnd_find_and_cancel_dgram(kgn_device_t *dev, lnet_nid_t dst_nid);
void kgnilnd_cancel_dgram_locked(kgn_dgram_t *dgram);
-void kgnilnd_release_dgram(kgn_device_t *dev, kgn_dgram_t *dgram);
+void kgnilnd_release_dgram(kgn_device_t *dev, kgn_dgram_t *dgram, int shutdown);
int kgnilnd_setup_wildcard_dgram(kgn_device_t *dev);
int kgnilnd_cancel_net_dgrams(kgn_net_t *net);
int kgnilnd_cancel_wc_dgrams(kgn_device_t *dev);
+int kgnilnd_cancel_dgrams(kgn_device_t *dev);
void kgnilnd_wait_for_canceled_dgrams(kgn_device_t *dev);
int kgnilnd_dgram_waitq(void *arg);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff