Overhaul of resource LP to support 'blocking'

codes/resource-lp.h

@@ -35,6 +35,7 @@ void resource_lp_configure();
 void resource_lp_get(
         msg_header *header,
         uint64_t req, 
+        int block_on_unavail,
         int msg_size, 
         int msg_header_offset,
         int msg_callback_offset,
@@ -44,6 +45,7 @@ void resource_lp_free(uint64_t req, tw_lp *sender);
 void resource_lp_reserve(
         msg_header *header, 
         uint64_t req,
+        int block_on_unavail,
         int msg_size,
         int msg_header_offset,
         int msg_callback_offset,
@@ -52,6 +54,7 @@ void resource_lp_get_reserved(
         msg_header *header,
         uint64_t req,
         resource_token_t tok,
+        int block_on_unavail,
         int msg_size, 
         int msg_header_offset,
         int msg_callback_offset,

codes/resource.h

@@ -16,18 +16,21 @@
 
 #include <stdint.h>
 
+
 typedef struct resource_s resource;
 typedef unsigned int resource_token_t;
 
 /* initialize with avail capacity, all unreserved */
 void resource_init(uint64_t avail, resource *r);
 
-/* Acquire req units of the resource from the general pool. 
- * Returns 0 on success, 1 on failure (not enough available). */
-int resource_get(uint64_t req, resource *r);
+/* Acquire req units of the resource. 
+ * Returns 0 on success, 1 on failure (not enough available), 2 on invalid
+ * token. */
+int resource_get(uint64_t req, resource_token_t tok, resource *r);
 
-/* Release req units of the resource from the general pool. */
-void resource_free(uint64_t req, resource *r);
+/* Release req units of the resource.
+ * Returns 0 on success, 2 on invalid token */
+int resource_free(uint64_t req, resource_token_t tok, resource *r);
 
 /* Reservation functions, same return value as get. 
  * These functions expect exactly one caller per LP group as 
@@ -35,16 +38,10 @@ void resource_free(uint64_t req, resource *r);
  * TODO: "un-reserving" not yet supported */
 int resource_reserve(uint64_t req, resource_token_t *tok, resource *r);
 
-/* Acquire req units of the resource from a reserved pool */ 
-int reserved_get(uint64_t req, resource_token_t tok, resource *r);
-
-/* Release req units of the resource from the general pool. */
-void reserved_free(uint64_t req, resource_token_t tok, resource *r);
-
 #define MAX_RESERVE 8
 struct resource_s {
-    uint64_t avail;
-    uint64_t reserved_avail[MAX_RESERVE];
+    /* index 0 is the general pool, 1... are the reserved pools */
+    uint64_t avail[MAX_RESERVE+1];
     unsigned int num_tokens;
 };
 

src/util/resource-lp.c

@@ -8,6 +8,7 @@
 #include "codes/resource.h"
 #include "codes/codes_mapping.h"
 #include "codes/jenkins-hash.h"
+#include "codes/quicklist.h"
 #include "ross.h"
 #include <assert.h>
 #include <stdio.h>
@@ -23,35 +24,55 @@ static uint64_t avail_global;
 
 typedef struct resource_state resource_state;
 typedef struct resource_msg resource_msg;
+typedef struct pending_op pending_op;
 
 #define TOKEN_DUMMY ((resource_token_t)-1)
 
 /* event types */
 enum resource_event
 {
-    RESOURCE_INIT = 100,
-    RESOURCE_GET,
+    RESOURCE_GET = 100,
     RESOURCE_FREE,
+    RESOURCE_DEQ,
     RESOURCE_RESERVE,
-    RESOURCE_GET_RESERVED,
-    RESOURCE_FREE_RESERVED
 };
 
 struct resource_state {
     resource r;
-    int is_init;
+    /* pending operations - if OOM and we are using the 'blocking' method, 
+     * then need to stash parameters.
+     * Index 0 is the general pool, index 1.. are the reservation-specific
+     * pools. We take advantage of resource_token_t's status as a simple 
+     * array index to do the proper indexing */
+    struct qlist_head pending[MAX_RESERVE+1];
 };
 
-struct resource_msg {
+/* following struct exists because we want to basically cache a message within
+ * a message for rc (ewww) */
+struct resource_msg_internal{
     msg_header h;
     /* request data */
     uint64_t req;
     resource_token_t tok; /* only for reserved calls */
+    /* behavior when sending response to caller
+     * 0 - send the callback immediately if resource unavailable. 
+     * 1 - send the callback when memory is available (danger - deadlock
+     * possible) */
+    int block_on_unavail; 
     /* callback data */
     msg_header h_callback;
     int msg_size;
     int msg_header_offset;
     int msg_callback_offset;
+}; 
+
+struct resource_msg {
+    struct resource_msg_internal i, i_rc;
+};
+
+struct pending_op {
+    struct resource_msg_internal m;
+    struct qlist_head ql;
 };
 
 /**** END SIMULATION DATA STRUCTURES ****/
@@ -60,21 +81,21 @@ struct resource_msg {
 
 /* ROSS LP processing functions */  
 static void resource_lp_ind_init(
-    resource_state * ns,
-    tw_lp * lp);
+        resource_state * ns,
+        tw_lp * lp);
 static void resource_event_handler(
-    resource_state * ns,
-    tw_bf * b,
-    resource_msg * m,
-    tw_lp * lp);
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp);
 static void resource_rev_handler(
-    resource_state * ns,
-    tw_bf * b,
-    resource_msg * m,
-    tw_lp * lp);
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp);
 static void resource_finalize(
-    resource_state * ns,
-    tw_lp * lp);
+        resource_state * ns,
+        tw_lp * lp);
 
 /* ROSS function pointer table for this LP */
 static tw_lptype resource_lp = {
@@ -94,8 +115,202 @@ void resource_lp_ind_init(
         resource_state * ns,
         tw_lp * lp){
     /* currently use global to initialize, may need to have other LPs init */
-    ns->is_init = 1;
     resource_init(avail_global, &ns->r);
+    int i;
+    for (i = 0; i < MAX_RESERVE+1; i++){
+        INIT_QLIST_HEAD(&ns->pending[i]);
+    }
+}
+
+static void resource_response(
+        resource_msg *m,
+        tw_lp *lp,
+        int ret,
+        resource_token_t tok){
+    /* send return message */
+    msg_header h;
+    msg_set_header(m->i.h_callback.magic, m->i.h_callback.event_type, 
+            lp->gid, &h);
+
+    resource_callback c;
+    c.ret = ret;
+    c.tok = tok;
+
+    /* before we send the message, sanity check the sizes */
+    if (m->i.msg_size >= m->i.msg_header_offset+sizeof(h) &&
+            m->i.msg_size >= m->i.msg_callback_offset+sizeof(c)){
+        tw_event *e = codes_event_new(m->i.h_callback.src, 
+                codes_local_latency(lp), lp);
+        void *msg = tw_event_data(e);
+        memcpy(((char*)msg)+m->i.msg_header_offset, &h, sizeof(h));
+        memcpy(((char*)msg)+m->i.msg_callback_offset, &c, sizeof(c));
+        tw_event_send(e);
+    }
+    else{
+        tw_error(TW_LOC, 
+                "message size not large enough to hold header/callback "
+                "structures");
+    }
+}
+static void resource_response_rc(tw_lp *lp){
+    codes_local_latency_reverse(lp);
+}
+
+/* bitfield usage:
+ * c0 - enqueued a message 
+ * c1 - sent an ack 
+ * c2 - successfully got the resource */
+static void handle_resource_get(
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp){
+    int ret = 1;
+    int send_ack = 1;
+    if (!qlist_empty(&ns->pending[m->i.tok]) || 
+            (ret = resource_get(m->i.req, m->i.tok, &ns->r))){
+        /* failed to receive data */
+        assert(ret != 2);
+        if (m->i.block_on_unavail){
+            /* queue up operation, save til later */
+            b->c0 = 1;
+            pending_op *op = malloc(sizeof(pending_op));
+            op->m = m->i; /* no need to set rc msg here */
+            qlist_add_tail(&op->ql, &ns->pending[m->i.tok]);
+        }
+    }
+    if (send_ack){
+        b->c1 = 1;
+        resource_response(m, lp, ret, TOKEN_DUMMY);
+    }
+
+    b->c2 = !ret;
+}
+
+/* bitfield usage:
+ * c0 - enqueued a message 
+ * c1 - sent an ack 
+ * c2 - successfully got the resource */
+static void handle_resource_get_rc(
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp){
+    if (b->c0){
+        assert(!qlist_empty(&ns->pending[m->i.tok]));
+        struct qlist_head *ql = qlist_pop_back(&ns->pending[m->i.tok]);
+        free(qlist_entry(ql, pending_op, ql));
+    }
+    else if (b->c1){
+        resource_response_rc(lp);
+    }
+
+    if (b->c2){
+        int ret = resource_free(m->i.req, m->i.tok, &ns->r);
+        assert(ret != 2);
+    }
+}
+
+static void handle_resource_free(
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp){
+    assert(!resource_free(m->i.req, m->i.tok, &ns->r));
+    /* create an event to pop the next queue item */
+    tw_event *e = codes_event_new(lp->gid, codes_local_latency(lp), lp);
+    resource_msg *m_deq = tw_event_data(e);
+    msg_set_header(resource_magic, RESOURCE_DEQ, lp->gid, &m_deq->i.h);
+    m_deq->i.tok = m->i.tok; /* only tok is needed, all others grabbed from q */
+    tw_event_send(e);
+}
+static void handle_resource_free_rc(
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp){
+    assert(!resource_get(m->i.req, m->i.tok, &ns->r));
+    codes_local_latency_reverse(lp);
+}
+
+/* bitfield usage:
+ * c0 - dequeue+alloc success */ 
+static void handle_resource_deq(
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp){
+    if (qlist_empty(&ns->pending[m->i.tok])){
+        /* nothing to do */
+        return;
+    }
+
+    struct qlist_head *front = ns->pending[m->i.tok].next;
+    pending_op *p = qlist_entry(front, pending_op, ql);
+    int ret = resource_get(p->m.req, p->m.tok, &ns->r);
+    assert(ret != 2);
+    b->c0 = !ret;
+    if (!ret){
+        /* success, dequeue (saving as rc) and send to client */
+        qlist_del(front);
+        m->i_rc = p->m;
+        resource_response(m, lp, ret, TOKEN_DUMMY);
+        free(p);
+        /* additionally attempt to dequeue next one down */
+        tw_event *e = codes_event_new(lp->gid, codes_local_latency(lp), lp);
+        resource_msg *m_deq = tw_event_data(e);
+        msg_set_header(resource_magic, RESOURCE_DEQ, lp->gid, &m_deq->i.h);
+        /* only tok is needed, all others grabbed from q */
+        m_deq->i.tok = m->i.tok; 
+        tw_event_send(e);
+    }
+    /* else do nothing */
+}
+
+/* bitfield usage:
+ * c0 - dequeue+alloc success */ 
+static void handle_resource_deq_rc(
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp){
+    if (qlist_empty(&ns->pending[m->i.tok])){
+        /* nothing to do */
+        return;
+    }
+
+    if (b->c0){
+        /* add operation back to the front of the queue */
+        pending_op *op = malloc(sizeof(pending_op));
+        op->m = m->i_rc;
+        qlist_add(&op->ql, &ns->pending[m->i.tok]);
+        resource_response_rc(lp);
+        /* reverse "deq next" op */
+        codes_local_latency_reverse(lp);
+    }
+}
+
+static void handle_resource_reserve(
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp){
+    resource_token_t tok;
+    int ret = resource_reserve(m->i.req, &tok, &ns->r);
+    assert(!ret);
+    resource_response(m, lp, ret, tok);
+}
+static void handle_resource_reserve_rc(
+        resource_state * ns,
+        tw_bf * b,
+        resource_msg * m,
+        tw_lp * lp){
+    /* this reversal method is essentially a hack that relies on each
+     * sequential reserve appending to the end of the list 
+     * - we expect reserves to happen strictly at the beginning of the
+     *   simulation */
+    ns->r.num_tokens--;
+    resource_response_rc(lp);
 }
 
 void resource_event_handler(
@@ -103,122 +318,46 @@ void resource_event_handler(
         tw_bf * b,
         resource_msg * m,
         tw_lp * lp){
-    assert(m->h.magic == resource_magic);
-    
-    int send_ack = 0;
-    int ret;
-    resource_token_t tok = TOKEN_DUMMY;
-    switch (m->h.event_type){
-        case RESOURCE_INIT:
-            assert(0);/* this should not be called */
-            assert(ns->is_init == 0);
-            resource_init(m->req, &ns->r);
-            ns->is_init = 1;
-            break;
+    assert(m->i.h.magic == resource_magic);
+    switch(m->i.h.event_type){
         case RESOURCE_GET:
-            assert(ns->is_init);
-            ret = resource_get(m->req, &ns->r);
-            send_ack = 1;
+            handle_resource_get(ns,b,m,lp);
             break;
         case RESOURCE_FREE:
-            assert(ns->is_init);
-            resource_free(m->req, &ns->r);
-            break;
-        case RESOURCE_RESERVE:
-            assert(ns->is_init);
-            ret = resource_reserve(m->req, &tok, &ns->r);
-            send_ack = 1;
-            /* even though we don't expect reserve to RC, set the tok in the
-             * message anyways */
-            m->tok = tok;
+            handle_resource_free(ns,b,m,lp);
             break;
-        case RESOURCE_GET_RESERVED:
-            assert(ns->is_init);
-            ret = reserved_get(m->req, m->tok, &ns->r);
-            send_ack = 1;
+        case RESOURCE_DEQ:
+            handle_resource_deq(ns,b,m,lp);
             break;
-        case RESOURCE_FREE_RESERVED:
-            assert(ns->is_init);
-            reserved_free(m->req, m->tok, &ns->r);
+        case RESOURCE_RESERVE:
+            handle_resource_reserve(ns,b,m,lp);
             break;
         default:
-            tw_error(TW_LOC, "resource event type not known");
-            break;
-    }
-    if (send_ack){
-        /* we use bitfield to determine whether previous op was a success,
-         * allowing us to do rollback a bit easier */
-        b->c0 = !ret;
-        /* send return message */
-
-        msg_header h;
-        msg_set_header(m->h_callback.magic, m->h_callback.event_type, 
-                lp->gid, &h);
-
-        resource_callback c;
-        c.ret = ret;
-        c.tok = tok;
-
-        /* before we send the message, sanity check the sizes */
-        if (m->msg_size >= m->msg_header_offset+sizeof(h) &&
-                m->msg_size >= m->msg_callback_offset+sizeof(c)){
-            tw_event *e = codes_event_new(m->h_callback.src, 
-                    codes_local_latency(lp), lp);
-            void *msg = tw_event_data(e);
-            memcpy(((char*)msg)+m->msg_header_offset, &h, sizeof(h));
-            memcpy(((char*)msg)+m->msg_callback_offset, &c, sizeof(c));
-            tw_event_send(e);
-        }
-        else{
-            tw_error(TW_LOC, 
-                    "message size not large enough to hold header/callback "
-                    "structures");
-        }
+            assert(0);
     }
 }
-
 void resource_rev_handler(
         resource_state * ns,
         tw_bf * b,
         resource_msg * m,
         tw_lp * lp){
-    assert(m->h.magic == resource_magic);
-    
-    int send_ack = 0;
-    switch (m->h.event_type){
-        case RESOURCE_INIT:
-            assert(0); /* not currently used */
-            /* this really shouldn't happen, but who knows... */
-            ns->is_init = 0;
-            break;
+    assert(m->i.h.magic == resource_magic);
+    switch(m->i.h.event_type){
         case RESOURCE_GET:
-            send_ack = 1;
-            if (b->c0){ resource_free(m->req, &ns->r); }
+            handle_resource_get_rc(ns,b,m,lp);
             break;
         case RESOURCE_FREE:
-            /* "re-allocate" the resource (this MUST work under the current
-             * implementation) */
-            assert(0==resource_get(m->req, &ns->r));
-            break;
-        case RESOURCE_RESERVE:
-            /* this reversal method is essentially a hack that relies on each
-             * sequential reserve appending to the end of the list */
-            send_ack = 1;
-            if (b->c0){ ns->r.num_tokens--; }
+            handle_resource_free_rc(ns,b,m,lp);
             break;
-        case RESOURCE_GET_RESERVED:
-            send_ack = 1;
-            if (b->c0){ reserved_free(m->req, m->tok, &ns->r); }
+        case RESOURCE_DEQ:
+            handle_resource_deq_rc(ns,b,m,lp);
             break;
-        case RESOURCE_FREE_RESERVED:
-            /* "re-allocate" the resource (this MUST work under the current
-             * implementation) */
-            assert(0==reserved_get(m->req, m->tok, &ns->r));
+        case RESOURCE_RESERVE:
+            handle_resource_reserve_rc(ns,b,m,lp);
             break;
         default:
-            tw_error(TW_LOC, "resource event type not known");
+            assert(0);
     }
-    if (send_ack){ codes_local_latency_reverse(lp); }
 }
 
 void resource_finalize(
@@ -256,6 +395,7 @@ static void resource_lp_issue_event(
         msg_header *header,
         uint64_t req,
         resource_token_t tok, /* only used in reserve_get/free */
+        int block_on_unavail,
         int msg_size,
         int msg_header_offset,
         int msg_callback_offset,
@@ -278,17 +418,18 @@ static void resource_lp_issue_event(
 
     /* set message info */
     resource_msg *m = tw_event_data(e);
-    msg_set_header(resource_magic, type, sender->gid, &m->h);
-    m->req = req;
-    m->tok = tok;
+    msg_set_header(resource_magic, type, sender->gid, &m->i.h);
+    m->i.req = req;
+    m->i.tok = tok;
+    m->i.block_on_unavail = block_on_unavail;
 
     /* set callback info */
     if (header != NULL){
-        m->h_callback = *header;
+        m->i.h_callback = *header;
     }
-    m->msg_size = msg_size;
-    m->msg_header_offset = msg_header_offset;
-    m->msg_callback_offset = msg_callback_offset;
+    m->i.msg_size = msg_size;
+    m->i.msg_header_offset = msg_header_offset;
+    m->i.msg_callback_offset = msg_callback_offset;
 
     tw_event_send(e);
 }
@@ -296,46 +437,51 @@ static void resource_lp_issue_event(
 void resource_lp_get(
         msg_header *header,
         uint64_t req, 
+        int block_on_unavail,
         int msg_size, 
         int msg_header_offset,
         int msg_callback_offset,
         tw_lp *sender){
-    resource_lp_issue_event(header, req, TOKEN_DUMMY, msg_size, 
-            msg_header_offset, msg_callback_offset, RESOURCE_GET, sender);
+    resource_lp_issue_event(header, req, 0, block_on_unavail,
+            msg_size, msg_header_offset, msg_callback_offset, RESOURCE_GET,
+            sender);
 }
 
 /* no callback for frees thus far */
 void resource_lp_free(uint64_t req, tw_lp *sender){
-    resource_lp_issue_event(NULL, req, TOKEN_DUMMY, -1,-1,-1,
+    resource_lp_issue_event(NULL, req, 0, -1, -1,-1,-1,
             RESOURCE_FREE, sender);
 }
 void resource_lp_reserve(
         msg_header *header, 
         uint64_t req,
+        int block_on_unavail,
         int msg_size,
         int msg_header_offset,
         int msg_callback_offset,
         tw_lp *sender){
-    resource_lp_issue_event(header, req, TOKEN_DUMMY, msg_size, 
-            msg_header_offset, msg_callback_offset, RESOURCE_RESERVE, sender);
+    resource_lp_issue_event(header, req, 0, msg_size,
+            block_on_unavail, msg_header_offset, msg_callback_offset,
+            RESOURCE_RESERVE, sender);
 }
 void resource_lp_get_reserved(
         msg_header *header,
         uint64_t req,
         resource_token_t tok,
+        int block_on_unavail,
         int msg_size, 
         int msg_header_offset,
         int msg_callback_offset,
         tw_lp *sender){
-    resource_lp_issue_event(header, req, tok, msg_size, 
-            msg_header_offset, msg_callback_offset, RESOURCE_GET_RESERVED,
+    resource_lp_issue_event(header, req, tok, msg_size, block_on_unavail,
+            msg_header_offset, msg_callback_offset, RESOURCE_GET,
             sender);
 }
 void resource_lp_free_reserved(
         uint64_t req, 
         resource_token_t tok,
         tw_lp *sender){
-    resource_lp_issue_event(NULL, req, tok, -1,-1,-1, RESOURCE_FREE_RESERVED,
+    resource_lp_issue_event(NULL, req, tok, -1,-1,-1,-1, RESOURCE_FREE,
             sender);
 }
 

src/util/resource.c

@@ -9,25 +9,36 @@
 
 /* initialize with avail capacity, all unreserved */
 void resource_init(uint64_t avail, resource *r){
-    r->avail = avail;
+    r->avail[0] = avail;
     r->num_tokens = 0;
 }
 
-/* Acquire req units of the resource from the general pool. 
- * Returns 0 on success, 1 on failure (not enough available). */
-int resource_get(uint64_t req, resource *r){