md: replace STRIPE_OP_COMPUTE_BLK with STRIPE_COMPUTE_RUN
authorDan Williams <dan.j.williams@intel.com>
Fri, 27 Jun 2008 22:32:03 +0000 (08:32 +1000)
committerNeil Brown <neilb@notabene.brown>
Fri, 27 Jun 2008 22:32:03 +0000 (08:32 +1000)
From: Dan Williams <dan.j.williams@intel.com>

Track the state of compute operations (recalculating a block from all the other
blocks in a stripe) with a state flag.  Reduces the scope of the
STRIPE_OP_COMPUTE_BLK flag to only tracking whether a compute operation has
been requested via the ops_request field of struct stripe_head_state.

Note, the compute operation that is performed in the course of doing a 'repair'
operation (check the parity block, recalculate it and write it back if the
check result is not zero) is tracked separately with the 'check_state'
variable.  Compute operations are held off while a 'check' is in progress, and
moving this check out to handle_issuing_new_read_requests5 the helper routine
__handle_issuing_new_read_requests5 can be simplified.

This is another step towards the removal of ops.{pending,ack,complete,count},
i.e. STRIPE_OP_COMPUTE_BLK only requests an operation and does not track the
state of the operation.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Neil Brown <neilb@suse.de>
drivers/md/raid5.c

index b9c0a32a4f9549d8e768ffdff6aeb85225bf179b..835046bf384e035913db9dc88a167d6cf174d6b8 100644 (file)
@@ -604,8 +604,6 @@ static void ops_complete_compute5(void *stripe_head_ref)
        clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
        if (sh->check_state == check_state_compute_run)
                sh->check_state = check_state_compute_result;
-       else
-               set_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.complete);
        set_bit(STRIPE_HANDLE, &sh->state);
        release_stripe(sh);
 }
@@ -881,8 +879,7 @@ static void raid5_run_ops(struct stripe_head *sh, unsigned long pending,
                overlap_clear++;
        }
 
-       if (test_bit(STRIPE_OP_COMPUTE_BLK, &pending) ||
-           test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request))
+       if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request))
                tx = ops_run_compute5(sh, pending);
 
        if (test_bit(STRIPE_OP_PREXOR, &pending))
@@ -1960,12 +1957,6 @@ static int __handle_issuing_new_read_requests5(struct stripe_head *sh,
        struct r5dev *dev = &sh->dev[disk_idx];
        struct r5dev *failed_dev = &sh->dev[s->failed_num];
 
-       /* don't schedule compute operations or reads on the parity block while
-        * a check is in flight
-        */
-       if (disk_idx == sh->pd_idx && sh->check_state)
-               return ~0;
-
        /* is the data in this block needed, and can we get it? */
        if (!test_bit(R5_LOCKED, &dev->flags) &&
            !test_bit(R5_UPTODATE, &dev->flags) && (dev->toread ||
@@ -1974,23 +1965,16 @@ static int __handle_issuing_new_read_requests5(struct stripe_head *sh,
             (failed_dev->toread || (failed_dev->towrite &&
             !test_bit(R5_OVERWRITE, &failed_dev->flags)
             ))))) {
-               /* 1/ We would like to get this block, possibly by computing it,
-                * but we might not be able to.
-                *
-                * 2/ Since parity check operations potentially make the parity
-                * block !uptodate it will need to be refreshed before any
-                * compute operations on data disks are scheduled.
-                *
-                * 3/ We hold off parity block re-reads until check operations
-                * have quiesced.
+               /* We would like to get this block, possibly by computing it,
+                * otherwise read it if the backing disk is insync
                 */
-               if ((s->uptodate == disks - 1) && !sh->check_state &&
+               if ((s->uptodate == disks - 1) &&
                    (s->failed && disk_idx == s->failed_num)) {
-                       set_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending);
+                       set_bit(STRIPE_COMPUTE_RUN, &sh->state);
+                       set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
                        set_bit(R5_Wantcompute, &dev->flags);
                        sh->ops.target = disk_idx;
                        s->req_compute = 1;
-                       sh->ops.count++;
                        /* Careful: from this point on 'uptodate' is in the eye
                         * of raid5_run_ops which services 'compute' operations
                         * before writes. R5_Wantcompute flags a block that will
@@ -1999,12 +1983,7 @@ static int __handle_issuing_new_read_requests5(struct stripe_head *sh,
                         */
                        s->uptodate++;
                        return 0; /* uptodate + compute == disks */
-               } else if ((s->uptodate < disks - 1) &&
-                       test_bit(R5_Insync, &dev->flags)) {
-                       /* Note: we hold off compute operations while checks are
-                        * in flight, but we still prefer 'compute' over 'read'
-                        * hence we only read if (uptodate < * disks-1)
-                        */
+               } else if (test_bit(R5_Insync, &dev->flags)) {
                        set_bit(R5_LOCKED, &dev->flags);
                        set_bit(R5_Wantread, &dev->flags);
                        s->locked++;
@@ -2021,20 +2000,13 @@ static void handle_issuing_new_read_requests5(struct stripe_head *sh,
 {
        int i;
 
-       /* Clear completed compute operations */
-       if (test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.complete)) {
-               clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.complete);
-               clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.ack);
-               clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending);
-       }
-
        /* look for blocks to read/compute, skip this if a compute
         * is already in flight, or if the stripe contents are in the
         * midst of changing due to a write
         */
-       if (!test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending) &&
-               !test_bit(STRIPE_OP_PREXOR, &sh->ops.pending) &&
-               !test_bit(STRIPE_OP_POSTXOR, &sh->ops.pending)) {
+       if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
+           !test_bit(STRIPE_OP_PREXOR, &sh->ops.pending) &&
+           !test_bit(STRIPE_OP_POSTXOR, &sh->ops.pending)) {
                for (i = disks; i--; )
                        if (__handle_issuing_new_read_requests5(
                                sh, s, i, disks) == 0)
@@ -2236,10 +2208,9 @@ static void handle_issuing_new_write_requests5(raid5_conf_t *conf,
         * simultaneously.  If this is not the case then new writes need to be
         * held off until the compute completes.
         */
-       if ((s->req_compute ||
-           !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending)) &&
-               (s->locked == 0 && (rcw == 0 || rmw == 0) &&
-               !test_bit(STRIPE_BIT_DELAY, &sh->state)))
+       if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
+           (s->locked == 0 && (rcw == 0 || rmw == 0) &&
+           !test_bit(STRIPE_BIT_DELAY, &sh->state)))
                s->locked += handle_write_operations5(sh, rcw == 0, 0);
 }
 
@@ -2410,6 +2381,7 @@ static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
                                set_bit(STRIPE_INSYNC, &sh->state);
                        else {
                                sh->check_state = check_state_compute_run;
+                               set_bit(STRIPE_COMPUTE_RUN, &sh->state);
                                set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
                                set_bit(R5_Wantcompute,
                                        &sh->dev[sh->pd_idx].flags);
@@ -2725,8 +2697,7 @@ static void handle_stripe5(struct stripe_head *sh)
         * or to load a block that is being partially written.
         */
        if (s.to_read || s.non_overwrite ||
-           (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding ||
-           test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending))
+           (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
                handle_issuing_new_read_requests5(sh, &s, disks);
 
        /* Now we check to see if any write operations have recently
@@ -2803,7 +2774,7 @@ static void handle_stripe5(struct stripe_head *sh)
         */
        if (sh->check_state ||
            (s.syncing && s.locked == 0 &&
-            !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending) &&
+            !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
             !test_bit(STRIPE_INSYNC, &sh->state)))
                handle_parity_checks5(conf, sh, &s, disks);
 
@@ -2869,7 +2840,7 @@ static void handle_stripe5(struct stripe_head *sh)
        }
 
        if (s.expanding && s.locked == 0 &&
-           !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending))
+           !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
                handle_stripe_expansion(conf, sh, NULL);
 
        if (sh->ops.count)
@@ -3089,7 +3060,7 @@ static void handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
        }
 
        if (s.expanding && s.locked == 0 &&
-           !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending))
+           !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
                handle_stripe_expansion(conf, sh, &r6s);
 
  unlock: