/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (c) 2003 Los Alamos National Laboratory (LANL)
+ * Copyright (c) 2003 Los Alamos National Laboratory (LANL)
+ * Copyright (C) 2005 Cluster File Systems, Inc. All rights reserved.
*
* This file is part of Lustre, http://www.lustre.org/
*
#ifndef EXPORT_SYMTAB
# define EXPORT_SYMTAB
#endif
-
-#include "linux/config.h"
+#ifndef AUTOCONF_INCLUDED
+#include <linux/config.h>
+#endif
#include "linux/module.h"
#include "linux/tty.h"
#include "linux/kernel.h"
#include "linux/vmalloc.h"
#include "linux/sysctl.h"
-#define DEBUG_SUBSYSTEM S_NAL
+#define DEBUG_SUBSYSTEM S_LND
-#include "portals/nal.h"
-#include "portals/api.h"
-#include "portals/errno.h"
#include "libcfs/kp30.h"
-#include "portals/p30.h"
-
-#include "portals/nal.h"
-#include "portals/lib-p30.h"
+#include "lnet/lnet.h"
+#include "lnet/lib-lnet.h"
/* undefine these before including the GM headers which clash */
#undef PACKAGE_BUGREPORT
#include "gm.h"
#include "gm_internal.h"
-/*
- * Defines for the API NAL
- */
+/* Fixed tunables */
+#define GMNAL_RESCHED 100 /* # busy loops to force scheduler to yield */
+#define GMNAL_NETADDR_BASE 0x10000000 /* where we start in network VM */
+#define GMNAL_LARGE_PRIORITY GM_LOW_PRIORITY /* large message GM priority */
+#define GMNAL_SMALL_PRIORITY GM_LOW_PRIORITY /* small message GM priority */
/* Wire protocol */
-
typedef struct {
- ptl_hdr_t gmim_hdr; /* portals header */
+ lnet_hdr_t gmim_hdr; /* portals header */
char gmim_payload[0]; /* payload */
} gmnal_immediate_msg_t;
} gmm_u;
} WIRE_ATTR gmnal_msg_t;
-#define GMNAL_MSG_MAGIC 0x6d797269 /* 'myri'! */
+#define GMNAL_MSG_MAGIC LNET_PROTO_GM_MAGIC
#define GMNAL_MSG_VERSION 1
#define GMNAL_MSG_IMMEDIATE 1
+typedef struct netbuf {
+ __u64 nb_netaddr; /* network VM address */
+ lnet_kiov_t nb_kiov[1]; /* the pages (at least 1) */
+} gmnal_netbuf_t;
+
+#define GMNAL_NETBUF_MSG(nb) ((gmnal_msg_t *)page_address((nb)->nb_kiov[0].kiov_page))
+#define GMNAL_NETBUF_LOCAL_NETADDR(nb) ((void *)((unsigned long)(nb)->nb_netaddr))
+
+typedef struct gmnal_txbuf {
+ struct list_head txb_list; /* queue on gmni_idle_ltxbs */
+ struct gmnal_txbuf *txb_next; /* stash on gmni_ltxs */
+ gmnal_netbuf_t txb_buf; /* space */
+} gmnal_txbuf_t;
+
typedef struct gmnal_tx {
- struct gmnal_tx *tx_next;
- gmnal_msg_t *tx_msg;
- int tx_buffer_size;
- gm_size_t tx_gm_size;
- int tx_msg_size;
- int tx_gmlid;
- int tx_gm_priority;
- ptl_nid_t tx_nid;
- struct gmnal_ni *tx_gmni;
- lib_msg_t *tx_libmsg;
- int tx_rxt;
+ struct list_head tx_list; /* queue */
+ int tx_credit:1; /* consumed a credit? */
+ int tx_large_iskiov:1; /* large is in kiovs? */
+ struct gmnal_ni *tx_gmni; /* owning NI */
+ lnet_nid_t tx_nid; /* destination NID */
+ int tx_gmlid; /* destination GM local ID */
+ lnet_msg_t *tx_lntmsg; /* lntmsg to finalize on completion */
+
+ gmnal_netbuf_t tx_buf; /* small tx buffer */
+ gmnal_txbuf_t *tx_ltxb; /* large buffer (to free on completion) */
+ int tx_msgnob; /* message size (so far) */
+
+ int tx_large_nob; /* # bytes large buffer payload */
+ int tx_large_offset; /* offset within frags */
+ int tx_large_niov; /* # VM frags */
+ union {
+ struct iovec *iov; /* mapped frags */
+ lnet_kiov_t *kiov; /* page frags */
+ } tx_large_frags;
+ unsigned long tx_launchtime; /* when (in jiffies) the transmit was launched */
+ struct gmnal_tx *tx_next; /* stash on gmni_txs */
} gmnal_tx_t;
-/*
- * as for gmnal_tx_t
- * a hash table in nal_data find rxs from
- * the rx buffer address. hash table populated at init time
- */
typedef struct gmnal_rx {
- struct list_head rx_list;
- gmnal_msg_t *rx_msg;
- int rx_size;
- gm_size_t rx_gmsize;
- unsigned int rx_recv_nob;
- __u16 rx_recv_gmid;
- __u8 rx_recv_port;
- __u8 rx_recv_type;
- struct gmnal_rx *rx_next;
+ struct list_head rx_list; /* enqueue on gmni_rxq for handling */
+ int rx_islarge:1; /* large receive buffer? */
+ unsigned int rx_recv_nob; /* bytes received */
+ __u16 rx_recv_gmid; /* sender */
+ __u8 rx_recv_port; /* sender's port */
+ __u8 rx_recv_type; /* ?? */
+ struct gmnal_rx *rx_next; /* stash on gmni_rxs */
+ gmnal_netbuf_t rx_buf; /* the buffer */
} gmnal_rx_t;
-
-/*
- * 1 receive thread started on each CPU
- */
-#define NRXTHREADS 10 /* max number of receiver threads */
-
typedef struct gmnal_ni {
- spinlock_t gmni_tx_lock;
- struct semaphore gmni_tx_token;
- gmnal_tx_t *gmni_tx;
- spinlock_t gmni_rxt_tx_lock;
- struct semaphore gmni_rxt_tx_token;
- gmnal_tx_t *gmni_rxt_tx;
- gmnal_rx_t *gmni_rx;
- struct gm_hash *gmni_rx_hash;
- lib_nal_t *gmni_libnal;
- struct gm_port *gmni_port;
- spinlock_t gmni_gm_lock; /* serialise GM calls */
- long gmni_rxthread_pid[NRXTHREADS];
- int gmni_rxthread_stop_flag;
- spinlock_t gmni_rxthread_flag_lock;
- long gmni_rxthread_flag;
- long gmni_ctthread_pid;
- int gmni_ctthread_flag;
- gm_alarm_t gmni_ctthread_alarm;
- int gmni_msg_size;
- struct list_head gmni_rxq;
- spinlock_t gmni_rxq_lock;
- struct semaphore gmni_rxq_wait;
+ lnet_ni_t *gmni_ni; /* generic NI */
+ struct gm_port *gmni_port; /* GM port */
+ spinlock_t gmni_gm_lock; /* serialise GM calls */
+ int gmni_large_pages; /* # pages in a large message buffer */
+ int gmni_large_msgsize; /* nob in large message buffers */
+ int gmni_large_gmsize; /* large message GM bucket */
+ int gmni_small_msgsize; /* nob in small message buffers */
+ int gmni_small_gmsize; /* small message GM bucket */
+ __u64 gmni_netaddr_base; /* base of mapped network VM */
+ int gmni_netaddr_size; /* # bytes of mapped network VM */
+
+ gmnal_tx_t *gmni_txs; /* all txs */
+ gmnal_rx_t *gmni_rxs; /* all rx descs */
+ gmnal_txbuf_t *gmni_ltxbs; /* all large tx bufs */
+
+ atomic_t gmni_nthreads; /* total # threads */
+ gm_alarm_t gmni_alarm; /* alarm to wake caretaker */
+ int gmni_shutdown; /* tell all threads to exit */
+
+ struct list_head gmni_idle_txs; /* idle tx's */
+ int gmni_tx_credits; /* # transmits still possible */
+ struct list_head gmni_idle_ltxbs; /* idle large tx buffers */
+ struct list_head gmni_buf_txq; /* tx's waiting for buffers */
+ struct list_head gmni_cred_txq; /* tx's waiting for credits */
+ spinlock_t gmni_tx_lock; /* serialise */
+
+ struct gm_hash *gmni_rx_hash; /* buffer->rx lookup */
+ struct semaphore gmni_rx_mutex; /* serialise blocking on GM */
} gmnal_ni_t;
-/*
- * Flags to start/stop and check status of threads
- * each rxthread sets 1 bit (any bit) of the flag on startup
- * and clears 1 bit when exiting
- */
-#define GMNAL_THREAD_RESET 0
-#define GMNAL_THREAD_STOP 666
-#define GMNAL_CTTHREAD_STARTED 333
-#define GMNAL_RXTHREADS_STARTED ( (1<<num_rx_threads)-1)
-
-
-/*
- * for ioctl get pid
- */
-#define GMNAL_IOC_GET_GNID 1
+typedef struct {
+ int *gm_port;
+ int *gm_ntx;
+ int *gm_credits;
+ int *gm_peer_credits;
+ int *gm_nlarge_tx_bufs;
+ int *gm_nrx_small;
+ int *gm_nrx_large;
+
+#if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
+ cfs_sysctl_table_header_t *gm_sysctl; /* sysctl interface */
+#endif
+} gmnal_tunables_t;
/* gmnal_api.c */
int gmnal_init(void);
-void gmnal_fini(void);
+void gmnal_fini(void);
+int gmnal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
+int gmnal_startup(lnet_ni_t *ni);
+void gmnal_shutdown(lnet_ni_t *ni);
/* gmnal_cb.c */
-ptl_err_t gmnal_cb_recv(lib_nal_t *libnal, void *private,
- lib_msg_t *libmsg,
- unsigned int niov, struct iovec *iov,
- size_t offset, size_t mlen, size_t rlen);
-ptl_err_t gmnal_cb_recv_pages(lib_nal_t *libnal, void *private,
- lib_msg_t *libmsg,
- unsigned int nkiov, ptl_kiov_t *kiov,
- size_t offset, size_t mlen, size_t rlen);
-ptl_err_t gmnal_cb_send(lib_nal_t *libnal, void *private,
- lib_msg_t *libmsg, ptl_hdr_t *hdr, int type,
- ptl_nid_t nid, ptl_pid_t pid,
- unsigned int niov, struct iovec *iov,
- size_t offset, size_t len);
-ptl_err_t gmnal_cb_send_pages(lib_nal_t *libnal, void *private,
- lib_msg_t *libmsg, ptl_hdr_t *hdr, int type,
- ptl_nid_t nid, ptl_pid_t pid,
- unsigned int nkiov, ptl_kiov_t *kiov,
- size_t offset, size_t len);
-int gmnal_cb_dist(lib_nal_t *libnal, ptl_nid_t nid, unsigned long *dist);
+int gmnal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
+ int delayed, unsigned int niov,
+ struct iovec *iov, lnet_kiov_t *kiov,
+ unsigned int offset, unsigned int mlen, unsigned int rlen);
+int gmnal_send(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
/* gmnal_util.c */
-int gmnal_is_rxthread(gmnal_ni_t *gmnalni);
-int gmnal_alloc_txs(gmnal_ni_t *gmnalni);
-void gmnal_free_txs(gmnal_ni_t *gmnalni);
-gmnal_tx_t *gmnal_get_tx(gmnal_ni_t *gmnalni, int block);
-void gmnal_return_tx(gmnal_ni_t *gmnalni, gmnal_tx_t *tx);
-int gmnal_alloc_rxs(gmnal_ni_t *gmnalni);
-void gmnal_free_rxs(gmnal_ni_t *gmnalni);
-void gmnal_stop_rxthread(gmnal_ni_t *gmnalni);
-void gmnal_stop_ctthread(gmnal_ni_t *gmnalni);
+void gmnal_free_ltxbufs(gmnal_ni_t *gmni);
+int gmnal_alloc_ltxbufs(gmnal_ni_t *gmni);
+void gmnal_free_txs(gmnal_ni_t *gmni);
+int gmnal_alloc_txs(gmnal_ni_t *gmni);
+void gmnal_free_rxs(gmnal_ni_t *gmni);
+int gmnal_alloc_rxs(gmnal_ni_t *gmni);
char *gmnal_gmstatus2str(gm_status_t status);
char *gmnal_rxevent2str(gm_recv_event_t *ev);
void gmnal_yield(int delay);
-int gmnal_enqueue_rx(gmnal_ni_t *gmnalni, gm_recv_t *recv);
-gmnal_rx_t *gmnal_dequeue_rx(gmnal_ni_t *gmnalni);
-int gmnal_start_kernel_threads(gmnal_ni_t *gmnalni);
/* gmnal_comm.c */
-void gmnal_pack_msg(gmnal_ni_t *gmnalni, gmnal_tx_t *tx,
- ptl_nid_t dstnid, int type);
-int gmnal_ct_thread(void *arg);
-int gmnal_rx_thread(void *arg);
-void gmnal_post_rx(gmnal_ni_t *gmnalni, gmnal_rx_t *rx);
-ptl_err_t gmnal_post_tx(gmnal_ni_t *gmnalni, gmnal_tx_t *tx,
- lib_msg_t *libmsg, ptl_nid_t nid, int nob);
+void gmnal_post_rx(gmnal_ni_t *gmni, gmnal_rx_t *rx);
+gmnal_tx_t *gmnal_get_tx(gmnal_ni_t *gmni);
+void gmnal_tx_done(gmnal_tx_t *tx, int rc);
+void gmnal_pack_msg(gmnal_ni_t *gmni, gmnal_msg_t *msg,
+ lnet_nid_t dstnid, int type);
+void gmnal_stop_threads(gmnal_ni_t *gmni);
+int gmnal_start_threads(gmnal_ni_t *gmni);
+void gmnal_check_txqueues_locked (gmnal_ni_t *gmni);
/* Module Parameters */
-extern int num_rx_threads;
-extern int num_txds;
-extern int gm_port_id;
+extern gmnal_tunables_t gmnal_tunables;
#endif /*__INCLUDE_GMNAL_H__*/
git://git.whamcloud.com / fs / lustre-release.git / blobdiff