# define EXPORT_SYMTAB
#endif
-#ifdef PROPRIETARY_ELAN
-# include <qsw/kernel.h>
-#else
-# include <qsnet/kernel.h>
-#endif
-
+#include <qsnet/kernel.h>
#undef printf /* nasty QSW #define */
#include <linux/config.h>
#include <linux/module.h>
-#include <elan3/elanregs.h>
-#include <elan3/elandev.h>
-#include <elan3/elanvp.h>
-#include <elan3/elan3mmu.h>
-#include <elan3/elanctxt.h>
-#include <elan3/elandebug.h>
-#include <elan3/urom_addrs.h>
-#include <elan3/busops.h>
-#include <elan3/kcomm.h>
+#if MULTIRAIL_EKC
+# include <elan/epcomms.h>
+#else
+# include <elan3/elanregs.h>
+# include <elan3/elandev.h>
+# include <elan3/elanvp.h>
+# include <elan3/elan3mmu.h>
+# include <elan3/elanctxt.h>
+# include <elan3/elandebug.h>
+# include <elan3/urom_addrs.h>
+# include <elan3/busops.h>
+# include <elan3/kcomm.h>
+#endif
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/stat.h>
#include <linux/errno.h>
-#include <linux/locks.h>
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+#include <linux/locks.h> /* wait_on_buffer */
+#else
+#include <linux/buffer_head.h> /* wait_on_buffer */
+#endif
#include <linux/unistd.h>
#include <net/sock.h>
#include <linux/uio.h>
#include <linux/file.h>
#include <linux/stat.h>
#include <linux/list.h>
-#include <asm/uaccess.h>
+#include <linux/sysctl.h>
#include <asm/segment.h>
#define DEBUG_SUBSYSTEM S_QSWNAL
#include <linux/kp30.h>
+#include <linux/kpr.h>
#include <portals/p30.h>
#include <portals/lib-p30.h>
-
-#define KQSW_OPTIMIZE_GETS 1
+#include <portals/nal.h>
+#include <linux/lustre_idl.h>
#define KQSW_CHECKSUM 0
#if KQSW_CHECKSUM
#define KQSW_HDR_SIZE (sizeof (ptl_hdr_t) + KQSW_CSUM_SIZE)
/*
- * Elan NAL
- */
-#define EP_SVC_LARGE_PORTALS_SMALL (0x10) /* Portals over elan port number (large payloads) */
-#define EP_SVC_LARGE_PORTALS_LARGE (0x11) /* Portals over elan port number (small payloads) */
-/* NB small/large message sizes are GLOBAL constants */
-
-/*
* Performance Tuning defines
* NB no mention of PAGE_SIZE for interoperability
*/
#define KQSW_RESCHED 100 /* # busy loops that forces scheduler to yield */
+#define KQSW_OPTIMIZED_GETS 1 /* optimize gets >= this size */
+#define KQSW_OPTIMIZED_PUTS (32<<10) /* optimize puts >= this size */
+#define KQSW_COPY_SMALL_FWD 0 /* copy small fwd messages to pre-mapped buffer? */
+
/*
* derived constants
*/
/* Remote memory descriptor */
typedef struct
{
- __u32 kqrmd_neiov; /* # frags */
- EP_IOVEC kqrmd_eiov[0]; /* actual frags */
+ __u32 kqrmd_nfrag; /* # frags */
+#if MULTIRAIL_EKC
+ EP_NMD kqrmd_frag[0]; /* actual frags */
+#else
+ EP_IOVEC kqrmd_frag[0]; /* actual frags */
+#endif
} kqswnal_remotemd_t;
typedef struct
struct list_head krx_list; /* enqueue -> thread */
EP_RCVR *krx_eprx; /* port to post receives to */
EP_RXD *krx_rxd; /* receive descriptor (for repost) */
- E3_Addr krx_elanaddr; /* Elan address of buffer (contiguous in elan vm) */
+#if MULTIRAIL_EKC
+ EP_NMD krx_elanbuffer; /* contiguous Elan buffer */
+#else
+ E3_Addr krx_elanbuffer; /* contiguous Elan buffer */
+#endif
int krx_npages; /* # pages in receive buffer */
int krx_nob; /* Number Of Bytes received into buffer */
- atomic_t krx_refcount; /* who's using me? */
- int krx_rpc_completed; /* I completed peer's RPC */
+ int krx_rpc_reply_needed; /* peer waiting for EKC RPC reply */
+ int krx_rpc_reply_status; /* what status to send */
+ int krx_state; /* what this RX is doing */
+ atomic_t krx_refcount; /* how to tell when rpc is done */
kpr_fwd_desc_t krx_fwd; /* embedded forwarding descriptor */
- struct page *krx_pages[KQSW_NRXMSGPAGES_LARGE]; /* pages allocated */
- struct iovec krx_iov[KQSW_NRXMSGPAGES_LARGE]; /* iovec for forwarding */
+ ptl_kiov_t krx_kiov[KQSW_NRXMSGPAGES_LARGE]; /* buffer frags */
} kqswnal_rx_t;
+#define KRX_POSTED 1 /* receiving */
+#define KRX_PARSE 2 /* ready to be parsed */
+#define KRX_COMPLETING 3 /* waiting to be completed */
+
+
typedef struct
{
struct list_head ktx_list; /* enqueue idle/active */
struct list_head ktx_delayed_list; /* enqueue delayedtxds */
- int ktx_isnblk:1; /* reserved descriptor? */
- int ktx_state:7; /* What I'm doing */
+ unsigned int ktx_isnblk:1; /* reserved descriptor? */
+ unsigned int ktx_state:7; /* What I'm doing */
+ unsigned int ktx_firsttmpfrag:1; /* ktx_frags[0] is in my ebuffer ? 0 : 1 */
uint32_t ktx_basepage; /* page offset in reserved elan tx vaddrs for mapping pages */
int ktx_npages; /* pages reserved for mapping messages */
int ktx_nmappedpages; /* # pages mapped for current message */
int ktx_port; /* destination ep port */
ptl_nid_t ktx_nid; /* destination node */
- void *ktx_args[2]; /* completion passthru */
- E3_Addr ktx_ebuffer; /* elan address of ktx_buffer */
+ void *ktx_args[3]; /* completion passthru */
char *ktx_buffer; /* pre-allocated contiguous buffer for hdr + small payloads */
unsigned long ktx_launchtime; /* when (in jiffies) the transmit was launched */
/* debug/info fields */
pid_t ktx_launcher; /* pid of launching process */
- ptl_hdr_t *ktx_wire_hdr; /* portals header (wire endian) */
int ktx_nfrag; /* # message frags */
- union {
- EP_IOVEC iov[EP_MAXFRAG]; /* msg frags (elan vaddrs) */
- EP_DATAVEC datav[EP_MAXFRAG]; /* DMA frags (eolan vaddrs) */
- } ktx_frags;
+#if MULTIRAIL_EKC
+ int ktx_rail; /* preferred rail */
+ EP_NMD ktx_ebuffer; /* elan mapping of ktx_buffer */
+ EP_NMD ktx_frags[EP_MAXFRAG];/* elan mapping of msg frags */
+#else
+ E3_Addr ktx_ebuffer; /* elan address of ktx_buffer */
+ EP_IOVEC ktx_frags[EP_MAXFRAG];/* msg frags (elan vaddrs) */
+#endif
} kqswnal_tx_t;
#define KTX_IDLE 0 /* on kqn_(nblk_)idletxds */
-#define KTX_SENDING 1 /* local send */
-#define KTX_FORWARDING 2 /* routing a packet */
-#define KTX_GETTING 3 /* local optimised get */
+#define KTX_FORWARDING 1 /* sending a forwarded packet */
+#define KTX_SENDING 2 /* normal send */
+#define KTX_GETTING 3 /* sending optimised get */
+#define KTX_PUTTING 4 /* sending optimised put */
+#define KTX_RDMAING 5 /* handling optimised put/get */
+
+typedef struct
+{
+ /* dynamic tunables... */
+ int kqn_optimized_puts; /* optimized PUTs? */
+ int kqn_optimized_gets; /* optimized GETs? */
+#if CONFIG_SYSCTL
+ struct ctl_table_header *kqn_sysctl; /* sysctl interface */
+#endif
+} kqswnal_tunables_t;
typedef struct
{
char kqn_init; /* what's been initialised */
char kqn_shuttingdown; /* I'm trying to shut down */
- atomic_t kqn_nthreads; /* # threads still running */
+ atomic_t kqn_nthreads; /* # threads running */
kqswnal_rx_t *kqn_rxds; /* all the receive descriptors */
kqswnal_tx_t *kqn_txds; /* all the transmit descriptors */
spinlock_t kqn_idletxd_lock; /* serialise idle txd access */
wait_queue_head_t kqn_idletxd_waitq; /* sender blocks here waiting for idle txd */
struct list_head kqn_idletxd_fwdq; /* forwarded packets block here waiting for idle txd */
+ atomic_t kqn_pending_txs; /* # transmits being prepped */
spinlock_t kqn_sched_lock; /* serialise packet schedulers */
wait_queue_head_t kqn_sched_waitq; /* scheduler blocks here */
struct list_head kqn_delayedfwds; /* delayed forwards */
struct list_head kqn_delayedtxds; /* delayed transmits */
- spinlock_t kqn_statelock; /* cb_cli/cb_sti */
- nal_cb_t *kqn_cb; /* -> kqswnal_lib */
- EP_DEV *kqn_epdev; /* elan device */
+#if MULTIRAIL_EKC
+ EP_SYS *kqn_ep; /* elan system */
+ EP_NMH *kqn_ep_tx_nmh; /* elan reserved tx vaddrs */
+ EP_NMH *kqn_ep_rx_nmh; /* elan reserved rx vaddrs */
+#else
+ EP_DEV *kqn_ep; /* elan device */
+ ELAN3_DMA_HANDLE *kqn_eptxdmahandle; /* elan reserved tx vaddrs */
+ ELAN3_DMA_HANDLE *kqn_eprxdmahandle; /* elan reserved rx vaddrs */
+#endif
EP_XMTR *kqn_eptx; /* elan transmitter */
EP_RCVR *kqn_eprx_small; /* elan receiver (small messages) */
EP_RCVR *kqn_eprx_large; /* elan receiver (large messages) */
- ELAN3_DMA_HANDLE *kqn_eptxdmahandle; /* elan reserved tx vaddrs */
- ELAN3_DMA_HANDLE *kqn_eprxdmahandle; /* elan reserved rx vaddrs */
kpr_router_t kqn_router; /* connection to Kernel Portals Router module */
ptl_nid_t kqn_nid_offset; /* this cluster's NID offset */
int kqn_nnodes; /* this cluster's size */
int kqn_elanid; /* this nodes's elan ID */
+
+ EP_STATUSBLK kqn_rpc_success; /* preset RPC reply status blocks */
+ EP_STATUSBLK kqn_rpc_failed;
} kqswnal_data_t;
/* kqn_init state */
#define KQN_INIT_NOTHING 0 /* MUST BE ZERO so zeroed state is initialised OK */
#define KQN_INIT_DATA 1
-#define KQN_INIT_PTL 2
+#define KQN_INIT_LIB 2
#define KQN_INIT_ALL 3
-extern nal_cb_t kqswnal_lib;
-extern nal_t kqswnal_api;
-extern kqswnal_data_t kqswnal_data;
+extern lib_nal_t kqswnal_lib;
+extern nal_t kqswnal_api;
+extern kqswnal_tunables_t kqswnal_tunables;
+extern kqswnal_data_t kqswnal_data;
extern int kqswnal_thread_start (int (*fn)(void *arg), void *arg);
extern void kqswnal_rxhandler(EP_RXD *rxd);
extern int kqswnal_scheduler (void *);
extern void kqswnal_fwd_packet (void *arg, kpr_fwd_desc_t *fwd);
-extern void kqswnal_reply_complete (EP_RXD *rxd);
-extern void kqswnal_requeue_rx (kqswnal_rx_t *krx);
+extern void kqswnal_rx_done (kqswnal_rx_t *krx);
static inline ptl_nid_t
kqswnal_elanid2nid (int elanid)
return (nid - kqswnal_data.kqn_nid_offset);
}
+static inline ptl_nid_t
+kqswnal_rx_nid(kqswnal_rx_t *krx)
+{
+ return (kqswnal_elanid2nid(ep_rxd_node(krx->krx_rxd)));
+}
+
static inline int
kqswnal_pages_spanned (void *base, int nob)
{
}
#endif
+static inline void kqswnal_rx_decref (kqswnal_rx_t *krx)
+{
+ LASSERT (atomic_read (&krx->krx_refcount) > 0);
+ if (atomic_dec_and_test (&krx->krx_refcount))
+ kqswnal_rx_done(krx);
+}
+
+#if MULTIRAIL_EKC
+# ifndef EP_RAILMASK_ALL
+# error "old (unsupported) version of EKC headers"
+# endif
+#else
+/* multirail defines these in <elan/epcomms.h> */
+#define EP_MSG_SVC_PORTALS_SMALL (0x10) /* Portals over elan port number (large payloads) */
+#define EP_MSG_SVC_PORTALS_LARGE (0x11) /* Portals over elan port number (small payloads) */
+/* NB small/large message sizes are GLOBAL constants */
+
+/* A minimal attempt to minimise inline #ifdeffing */
+
+#define EP_SUCCESS ESUCCESS
+#define EP_ENOMEM ENOMEM
+
+static inline EP_XMTR *
+ep_alloc_xmtr(EP_DEV *e)
+{
+ return (ep_alloc_large_xmtr(e));
+}
+
+static inline EP_RCVR *
+ep_alloc_rcvr(EP_DEV *e, int svc, int nenv)
+{
+ return (ep_install_large_rcvr(e, svc, nenv));
+}
+
+static inline void
+ep_free_xmtr(EP_XMTR *x)
+{
+ ep_free_large_xmtr(x);
+}
+
+static inline void
+ep_free_rcvr(EP_RCVR *r)
+{
+ ep_remove_large_rcvr(r);
+}
+#endif
+
#endif /* _QSWNAL_H */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff