Re: [Qemu-devel] [PATCH v2 09/12] Rework ram block hash

[Dr. David Alan Gilbert]

* Michael R. Hines (address@hidden) wrote:
> On 06/11/2015 12:17 PM, Dr. David Alan Gilbert (git) wrote:
> >From: "Dr. David Alan Gilbert" <address@hidden>
> >
> >RDMA uses a hash from block offset->RAM Block; this isn't needed
> >on the destination, and it becomes harder to maintain after the next
> >patch in the series that sorts the block list.
> >
> >Split the hash so that it's only generated on the source.
> >
> >Signed-off-by: Dr. David Alan Gilbert <address@hidden>
> >---
> >  migration/rdma.c | 32 ++++++++++++++++++++------------
> >  1 file changed, 20 insertions(+), 12 deletions(-)
> >
> >diff --git a/migration/rdma.c b/migration/rdma.c
> >index 8d99378..f541586 100644
> >--- a/migration/rdma.c
> >+++ b/migration/rdma.c
> >@@ -533,23 +533,22 @@ static int rdma_add_block(RDMAContext *rdma, const 
> >char *block_name,
> >                           ram_addr_t block_offset, uint64_t length)
> >  {
> >      RDMALocalBlocks *local = &rdma->local_ram_blocks;
> >-    RDMALocalBlock *block = g_hash_table_lookup(rdma->blockmap,
> >-        (void *)(uintptr_t)block_offset);
> >+    RDMALocalBlock *block;
> >      RDMALocalBlock *old = local->block;
> >
> >-    assert(block == NULL);
> >-
> >      local->block = g_malloc0(sizeof(RDMALocalBlock) * (local->nb_blocks + 
> > 1));
> >
> >      if (local->nb_blocks) {
> >          int x;
> >
> >-        for (x = 0; x < local->nb_blocks; x++) {
> >-            g_hash_table_remove(rdma->blockmap,
> >-                                (void *)(uintptr_t)old[x].offset);
> >-            g_hash_table_insert(rdma->blockmap,
> >-                                (void *)(uintptr_t)old[x].offset,
> >-                                &local->block[x]);
> >+        if (rdma->blockmap) {
> >+            for (x = 0; x < local->nb_blocks; x++) {
> >+                g_hash_table_remove(rdma->blockmap,
> >+                                    (void *)(uintptr_t)old[x].offset);
> >+                g_hash_table_insert(rdma->blockmap,
> >+                                    (void *)(uintptr_t)old[x].offset,
> >+                                    &local->block[x]);
> >+            }
> >          }
> >          memcpy(local->block, old, sizeof(RDMALocalBlock) * 
> > local->nb_blocks);
> >          g_free(old);
> >@@ -571,7 +570,9 @@ static int rdma_add_block(RDMAContext *rdma, const char 
> >*block_name,
> >
> >      block->is_ram_block = local->init ? false : true;
> >
> >-    g_hash_table_insert(rdma->blockmap, (void *) block_offset, block);
> >+    if (rdma->blockmap) {
> >+        g_hash_table_insert(rdma->blockmap, (void *) block_offset, block);
> >+    }
> >
> >      trace_rdma_add_block(block_name, local->nb_blocks,
> >                           (uintptr_t) block->local_host_addr,
> >@@ -607,7 +608,6 @@ static int qemu_rdma_init_ram_blocks(RDMAContext *rdma)
> >      RDMALocalBlocks *local = &rdma->local_ram_blocks;
> >
> >      assert(rdma->blockmap == NULL);
> >-    rdma->blockmap = g_hash_table_new(g_direct_hash, g_direct_equal);
> >      memset(local, 0, sizeof *local);
> >      qemu_ram_foreach_block(qemu_rdma_init_one_block, rdma);
> >      trace_qemu_rdma_init_ram_blocks(local->nb_blocks);
> >@@ -2292,6 +2292,14 @@ static int qemu_rdma_source_init(RDMAContext *rdma, 
> >Error **errp, bool pin_all)
> >          goto err_rdma_source_init;
> >      }
> >
> >+    /* Build the hash that maps from offset to RAMBlock */
> >+    rdma->blockmap = g_hash_table_new(g_direct_hash, g_direct_equal);
> >+    for (idx = 0; idx < rdma->local_ram_blocks.nb_blocks; idx++) {
> >+        g_hash_table_insert(rdma->blockmap,
> >+                (void *)(uintptr_t)rdma->local_ram_blocks.block[idx].offset,
> >+                &rdma->local_ram_blocks.block[idx]);
> >+    }
> >+
> >      for (idx = 0; idx < RDMA_WRID_MAX; idx++) {
> >          ret = qemu_rdma_reg_control(rdma, idx);
> >          if (ret) {
> 
> You didn't want to use the ID string as a key? I forget....

I was trying not to; it sounded expensive to hash on strings
if we didn't need it.

> Reviewed-by: Michael R. Hines <address@hidden>

Thanks.

Dave
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK