diff -r -u /tmp/rds-tools-2.0.4/rds-stress.c rds-tools-2.0.7/rds-stress.c

--- /tmp/rds-tools-2.0.4/rds-stress.c	Wed Aug  4 15:25:10 2010

+++ rds-tools-2.0.7/rds-stress.c	Thu Feb 24 13:27:52 2011

@@ -15,7 +15,13 @@

 #include <sys/time.h>

 #include <time.h>

 #include <inttypes.h>

+#if defined(__SVR4) && defined(__sun)

+#include <sys/syscall.h>

+#include <signal.h>

+#include <sys/lgrp_user.h>

+#else

 #include <syscall.h>

+#endif

 #include <sys/stat.h>

 #include <sys/poll.h>

 #include <ctype.h>

@@ -22,8 +28,13 @@

 #include <fcntl.h>

 #include <sched.h>

 #include <getopt.h>

+#if !(defined(__SVR4) && defined(__sun))

 #include <byteswap.h>

 #include "rds.h"

+#else

+#include <infiniband/ofa_solaris.h>

+#include <sys/rds.h>

+#endif

 #include "pfhack.h"

@@ -110,6 +121,7 @@

 struct child_control {

 	pid_t pid;

 	int ready;

+	int stopping;

 	struct timeval start;

 	struct counter cur[NR_STATS];

 	struct counter last[NR_STATS];

@@ -254,7 +266,20 @@

 	die("invalid host name or dotted quad '%s'\n", ptr);

 }

+#if defined(__SVR4) && defined(__sun)

+static lgrp_id_t lgrp_id = -1;

+static void

+set_my_lgrp(void)

+{

+	if (lgrp_id != -1) {

+		lgrp_affinity_set(P_LWPID, P_MYID, lgrp_id,

+			LGRP_AFF_STRONG);

+		yield(); /* force a context switch */

+	}

+}

+#endif

+

 static void usage(void)

 {

         fprintf(stderr, "rds-stress version %s\n", RDS_VERSION);

@@ -281,6 +306,9 @@

 	" -c                measure cpu use with per-cpu soak processes\n"

 	" -V                trace execution\n"

 	" -z                print a summary at end of test only\n"

+#if defined(__SVR4) && defined(__sun)

+	" -g [lgrpid]       bind the process to the specified lgrp\n"

+#endif

 	"\n"

 	"Example:\n"

 	"  recv$ rds-stress\n"

@@ -310,7 +338,7 @@

 static void check_parent(pid_t pid)

 {

 	if (pid != getppid())

-		die("parent %u exited\n", pid);

+		die("parent %u exited\n", (int)pid);

 }

 /*

@@ -334,6 +362,7 @@

 		msg_pattern[i] = k;

 }

+#if !(defined(__SVR4) && defined(__sun))

 #if __BYTE_ORDER == __LITTLE_ENDIAN

 #define htonll(x)	bswap_64(x)

 #define ntohll(x)	bswap_64(x)

@@ -341,6 +370,7 @@

 #define htonll(x)	(x)

 #define ntohll(x)	(x)

 #endif

+#endif /* Not sun */

 static void encode_hdr(struct header *dst, const struct header *hdr)

 {

@@ -584,7 +614,11 @@

 	if (opts->receive_addr == 0)

 		return 1;

+#if defined(__SVR4) && defined(__sun)

+	sin.sin_family = AF_INET_OFFLOAD;

+#else

 	sin.sin_family = AF_INET;

+#endif

 	sin.sin_port = htons(opts->starting_port);

 	sin.sin_addr.s_addr = htonl(opts->receive_addr);

@@ -677,7 +711,11 @@

 	size = sizeof(struct rdma_key_o_meter)

 			+ 2 * nr_tasks * sizeof(*kt)

 			+ 2 * RDMA_MAX_TRACKED_KEYS * sizeof(*ks);

+#if defined(__SVR4) && defined(__sun)

+	base = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_SHARED, -1, 0);

+#else

 	base = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_SHARED, 0, 0);

+#endif

 	if (base == MAP_FAILED)

 		die_errno("alloc_rdma_buffers: mmap failed");

@@ -828,7 +866,7 @@

 	}

 	if (!failed)

-		trace("compare pass pattern %Lx addr %p\n",

+		trace("compare pass pattern 0x%Lx addr %p\n",

 			(unsigned long long) pattern, addr);

 }

@@ -865,7 +903,11 @@

 	/* We use mmap here rather than malloc, because it is always

 	 * page aligned. */

 	len = 2 * opts->nr_tasks * opts->req_depth * (opts->rdma_vector * opts->rdma_size) + sys_page_size;

+#if defined(__SVR4) && defined(__sun)

+	base = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);

+#else	

 	base = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, 0, 0);

+#endif

 	if (base == MAP_FAILED)

 		die_errno("alloc_rdma_buffers: mmap failed");

 	memset(base, 0x2f, len);

@@ -915,17 +957,16 @@

 	if (RDMA_OP_READ == hdr->rdma_op) {

 		if (opt.verify)

 			rds_fill_buffer(rdma_addr, rdma_size, hdr->rdma_pattern);

-		trace("Requesting RDMA read for pattern %Lx "

-				"local addr to rdma read %p\n",

-				(unsigned long long) hdr->rdma_pattern,

+		trace("Requesting RDMA read for pattern 0x%Lx"

+				"local addr to rdma read 0x%p\n",

+				hdr->rdma_pattern,

 				rdma_addr);

 	} else {

 		if (opt.verify)

 			rds_fill_buffer(rdma_addr, rdma_size, 0);

-		trace("Requesting RDMA write for pattern %Lx "

-				"local addr to rdma write %p\n",

-				(unsigned long long) hdr->rdma_pattern,

-				rdma_addr);

+

+		trace("Requesting RDMA write for pattern 0x%Lx",

+				hdr->rdma_pattern);

 	}

 }

@@ -947,7 +988,7 @@

 		die("Unexpected RDMA op %u in request\n", in_hdr->rdma_op);

-	trace("RDS received request to issue rdma %s len %lu rva %Lx key %Lx pattern %Lx\n",

+	trace("RDS received request to issue rdma %s len %lu rva 0x%Lx key 0x%Lx pattern 0x%Lx\n",

 		in_hdr->rdma_op == RDMA_OP_WRITE? "write to" : "read from",

 		rdma_size,

 		(unsigned long long) in_hdr->rdma_addr,

@@ -1007,6 +1048,9 @@

 	t->drain_rdmas = 0;

 }

+#if defined(__SVR4) && defined(__sun)

+#undef MSG_MAXIOVLEN

+#endif

 #define MSG_MAXIOVLEN 2

 /*

@@ -1560,7 +1604,12 @@

 	struct timeval start;

         int do_work = opts->simplex ? active : 1;

+#if defined(__SVR4) && defined(__sun)

+	set_my_lgrp();

+	sin.sin_family = AF_INET_OFFLOAD;

+#else

 	sin.sin_family = AF_INET;

+#endif

 	sin.sin_port = htons(opts->starting_port + 1 + id);

 	sin.sin_addr.s_addr = htonl(opts->receive_addr);

@@ -1572,7 +1621,11 @@

 	for (i = 0; i < opts->nr_tasks; i++) {

 		tasks[i].nr = i;

 		tasks[i].src_addr = sin;

+#if defined(__SVR4) && defined(__sun)

+		tasks[i].dst_addr.sin_family = AF_INET_OFFLOAD;

+#else

 		tasks[i].dst_addr.sin_family = AF_INET;

+#endif

 		tasks[i].dst_addr.sin_addr.s_addr = htonl(opts->send_addr);

 		tasks[i].dst_addr.sin_port = htons(opts->starting_port + 1 + i);

 		tasks[i].send_time = alloca(opts->req_depth * sizeof(struct timeval));

@@ -1625,6 +1678,10 @@

 				;

 		}

+		/* stop sending if in shutdown phase */

+		if (ctl->stopping)

+			continue;

+

 		/* keep the pipeline full */

 		can_send = !!(pfd.revents & POLLOUT);

 		for (i = 0, t = tasks; i < opts->nr_tasks; i++, t++) {

@@ -1665,8 +1722,12 @@

 	uint32_t i;

 	len = opts->nr_tasks * sizeof(*ctl);

+#if defined(__SVR4) && defined(__sun)

+	ctl = (struct child_control *)mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_SHARED, -1, 0);

+#else

 	ctl = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_SHARED,

 		   0, 0);

+#endif

 	if (ctl == MAP_FAILED)

 		die("mmap of %u child control structs failed", opts->nr_tasks);

@@ -1699,7 +1760,7 @@

 			continue;

 		pid = waitpid(-1, NULL, WNOHANG);

 		if (pid)

-			die("child %u (pid %u) exited\n", i, pid);

+			die("child %u (pid %u) exited\n", i, (int)pid);

 		sleep(1);

 		i--; /* try this child again */

 	}

@@ -1967,7 +2028,7 @@

 	pid = waitpid(-1, &status, wflags);

 	if (pid < 0)

-		die("waitpid returned %u", pid);

+		die("waitpid returned %u", (int)pid);

 	if (pid == 0)

 		return 0;

@@ -1975,15 +2036,15 @@

 		if (WEXITSTATUS(status) == 0)

 			return 1;

 		die("child pid %u exited with status %d\n",

-				pid, WEXITSTATUS(status));

+				(int)pid, WEXITSTATUS(status));

 	}

 	if (WIFSIGNALED(status)) {

 		if (WTERMSIG(status) == SIGTERM)

 			return 1;

 		die("child pid %u exited with signal %d\n",

-				pid, WTERMSIG(status));

+				(int)pid, WTERMSIG(status));

 	}

-	die("child pid %u wait status %d\n", pid, status);

+	die("child pid %u wait status %d\n", (int)pid, status);

 }

 static void release_children_and_wait(struct options *opts,

@@ -2139,7 +2200,12 @@

 	control_fd = -1;

 	if (nr_running) {

+		/* let everything gracefully stop before we kill the chillins */

 		for (i = 0; i < opts->nr_tasks; i++)

+			ctl[i].stopping = 1;

+		sleep(1);

+

+		for (i = 0; i < opts->nr_tasks; i++)

 			kill(ctl[i].pid, SIGTERM);

 		stop_soakers(soak_arr);

 	}

@@ -2517,7 +2583,11 @@

 	/* an extra terminating entry which will be all 0s */

 	len = (nr_soak + 1) * sizeof(struct soak_control);

 	soak_arr = mmap(NULL, len, PROT_READ|PROT_WRITE,

+#if defined(__SVR4) && defined(__sun)

+			MAP_ANONYMOUS|MAP_SHARED, -1, 0);

+#else

 			MAP_ANONYMOUS|MAP_SHARED, 0, 0);

+#endif

 	if (soak_arr == MAP_FAILED)

 		die("mmap of %ld soak control structs failed", nr_soak);

@@ -2589,6 +2659,7 @@

 { "rtprio",		no_argument,		NULL,	'R'	},

 { "verify",		no_argument,		NULL,	'v'	},

 { "trace",		no_argument,		NULL,	'V'	},

+{ "lgrpid",		required_argument,	NULL,	'g'	},

 { "rdma-use-once",	required_argument,	NULL,	OPT_RDMA_USE_ONCE },

 { "rdma-use-get-mr",	required_argument,	NULL,	OPT_RDMA_USE_GET_MR },

@@ -2652,7 +2723,7 @@

 	while(1) {

 		int c, index;

-		c = getopt_long(argc, argv, "+a:cD:d:hI:M:op:q:Rr:s:t:T:vVz",

+		c = getopt_long(argc, argv, "+a:cD:d:hI:M:op:q:Rr:s:t:T:vVg:z",

 				long_options, &index);

 		if (c == -1)

 			break;

@@ -2711,6 +2782,10 @@

 			case 'V':

 				opts.tracing = 1;

 				break;

+			case 'g':

+				lgrp_id = (lgrp_id_t)parse_ull(optarg,

+				    (uint32_t)~0);

+				break;

 			case OPT_USE_CONG_MONITOR:

 				opts.use_cong_monitor = parse_ull(optarg, 1);

 				break;

@@ -2786,6 +2861,7 @@

 	if (opts.rdma_size && 0)

 		opts.rdma_size = (opts.rdma_size + 4095) & ~4095;

+	set_my_lgrp();

 	opt = opts;

 	return active_parent(&opts, soak_arr);

 }

diff -r -u /tmp/rds-tools-2.0.4/pfhack.h rds-tools-2.0.7/pfhack.h

--- /tmp/rds-tools-2.0.4/pfhack.h	Wed Aug  4 15:25:11 2010

+++ rds-tools-2.0.7/pfhack.h	Thu Feb 24 13:27:51 2011

@@ -44,9 +44,11 @@

 #ifndef __PF_HACK_H

 #define __PF_HACK_H

+#if !((defined(__SVR4) && defined(__sun)))

 #define PF_RDS		21

 #define AF_RDS		21

 #define SOL_RDS		276

+#endif

 extern int discover_pf_rds();

 extern int discover_sol_rds();

diff -r -u /tmp/rds-tools-2.0.4/rds-info.c rds-tools-2.0.7/rds-info.c

--- /tmp/rds-tools-2.0.4/rds-info.c	Wed Aug  4 15:25:10 2010

+++ rds-tools-2.0.7/rds-info.c	Thu Feb 24 13:27:51 2011

@@ -42,16 +42,27 @@

 #include <sys/types.h>

 #include <sys/socket.h>

 #include <errno.h>

+#if defined(__SVR4) && defined(__sun)

+#include <strings.h>

+#else

 #include <string.h>

+#endif

 #include <inttypes.h>

 #include <netinet/in.h>

 #include <arpa/inet.h>

+#if defined(__SVR4) && defined(__sun)

+#include <sys/rds.h>

+#else

 #include "rds.h"

+#endif

 #include "pfhack.h"

+/* WHUPS changed the struct rds_info_connection definition b/w rds in 1.4 & 1.5. gotta support both

+   for now. TODO remove check of transport[15] once ofed pre-1.5 is extinct. */

 #define rds_conn_flag(conn, flag, letter) \

-	(conn.flags & RDS_INFO_CONNECTION_FLAG_##flag ? letter : '-')

+	(conn.flags & RDS_INFO_CONNECTION_FLAG_##flag \

+	|| conn.transport[15] & RDS_INFO_CONNECTION_FLAG_##flag ? letter : '-')

 #define min(a, b) (a < b ? a : b)

 #define array_size(foo) (sizeof(foo) / sizeof(foo[0]))

@@ -234,8 +245,10 @@

 		print_msgs, "Send", 0 },

 	['t'] = { RDS_INFO_RETRANS_MESSAGES, "retransmit queue messages",

 		  print_msgs, "Retransmit", 0 },

+#if !(defined(__SVR4) && defined(__sun))

 	['T'] = { RDS_INFO_TCP_SOCKETS, "TCP transport sockets",

 		  print_tcp_socks, NULL, 0 },

+#endif

 	['I'] = { RDS_INFO_IB_CONNECTIONS, "IB transport connections",

 		  print_ib_conns, NULL, 0 },

 };

@@ -266,6 +279,10 @@

 	char optstring[258] = "v+";

 	int given_options = 0;

 	socklen_t len = 0;

+#if defined(__SVR4) && defined(__sun)

+	socklen_t ulen;

+	struct rds_info_arg arg;

+#endif

 	void *data = NULL;

 	int fd;

 	int each;

@@ -322,6 +339,7 @@

 		    (given_options && !infos[i].option_given))

 			continue;

+#if !(defined(__SVR4) && defined(__sun))

 		/* read in the info until we get a full snapshot */

 		while ((each = getsockopt(fd, sol, infos[i].opt_val, data,

 				   &len)) < 0) {

@@ -345,15 +363,60 @@

 				return 1;

 			}

 		}

+#else

+		/* 1st call gets the length of the data available */

+		ulen = 0;

+		bzero(&arg, sizeof (struct rds_info_arg));

+		arg.lenp = (uint64_t)(uintptr_t)&ulen;

+		arg.datap = NULL;

+		each = ioctl(fd, infos[i].opt_val, &arg);

+		if ((each < 0) && (errno != ENOSPC)) {

+			verbosef(0, stderr, "%s: Unable get statistics: %s\n",

+			    progname, strerror(errno));

+			return 1;

+		}

+		/* No data at the driver */

+		if (ulen == 0)

+			invalid_opt = 1;;

+#endif

+

 		if (invalid_opt)

 			continue;

+#if !(defined(__SVR4) && defined(__sun))

 		infos[i].print(data, each, len, infos[i].extra);

+#else

+		do {

+			arg.datap = (uint64_t)(uintptr_t)realloc(

+			    (char *)(uintptr_t)arg.datap, ulen);

+			if (arg.datap == NULL) {

+			    verbosef(0, stderr, "%s: Unable to allocate memory "

+				"for %u bytes of info: %s\n",

+				progname, ulen, strerror(errno));

+				return 1;

+			}

+			/* 2nd call gets the data */

+			len = ulen;

+			each = ioctl(fd, infos[i].opt_val, &arg);

+			if ((each < 0) && (errno != ENOSPC)) {

+				verbosef(0, stderr,

+				    "%s: Unable get statistics: %s\n",

+				    progname, strerror(errno));

+				return 1;

+			}

+		} while (ulen > len);

+

+		infos[i].print((void *)(uintptr_t)arg.datap, each, ulen,

+		    infos[i].extra);

+#endif

 		if (given_options && --given_options == 0)

 			break;

 	}

+#if defined(__SVR4) && defined(__sun)

+	free((void *)(uintptr_t)arg.datap);

+#endif

 	return 0;

 }

diff -r -u /tmp/rds-tools-2.0.4/rds.7 rds-tools-2.0.7/rds.7

--- /tmp/rds-tools-2.0.4/rds.7	Wed Aug  4 15:25:11 2010

+++ rds-tools-2.0.7/rds.7	Thu Feb 24 13:27:52 2011

@@ -6,6 +6,7 @@

 .nf

 .B #include <sys/socket.h>

 .B #include <netinet/in.h>

+.B #include <sys/rds.h>

 .fi

 .SH DESCRIPTION

 This is an implementation of the RDS socket API. It provides reliable,

@@ -14,18 +15,13 @@

 Currently, RDS can be transported over Infiniband, and loopback.

 RDS over TCP is disabled, but will be re-enabled in the near future.

 .PP

-RDS uses standard

-.B AF_INET

-addresses as described in

-.BR ip (7)

+RDS uses 

+.B AF_INET_OFFLOAD address family 

 to identify end points.

 .\"------------------------------------------------------------------

 .SS Socket Creation

 RDS is still in development and as such does not have a reserved protocol

-family constant. Applications must read the string representation of the

-protocol family value from the

-.B pf_rds

-sysctl parameter file described below.

+family constant.  Applications should use AF_INET_OFFLOAD.

 .PP

 .nf

 .B rds_socket = socket(pf_rds, SOCK_SEQPACKET, 0);

@@ -58,9 +54,6 @@

 .BR SOL_RDS ).

 Just as with the RDS protocol family, an official value has not been

 assigned yet, so the kernel will assign a value dynamically.

-The assigned value can be retrieved from the

-.B sol_rds

-sysctl parameter file.

 .PP

 RDS specific socket options will be described in a separate section

 below.

@@ -77,7 +70,7 @@

 .PP

 For instance, when binding to the address of an Infiniband interface

 such as

-.BR ib0 ,

+.BR ibd0 ,

 the socket will use the Infiniband transport. If RDS is not able

 to associate a transport with the given address, it will return

 .BR EADDRNOTAVAIL .

@@ -394,47 +387,6 @@

 be delivered in the order they're sent. Messages sent from different

 sockets, or to different destinations, may be delivered in any order.

 .\"------------------------------------------------------------------

-.SH SYSCTL VALUES

-These parameteres may only be accessed through their files in

-.BR /proc/sys/net/rds .

-Access through

-.BR sysctl (2)