Palacios Public Git Repository

To checkout Palacios execute

  git clone http://v3vee.org/palacios/palacios.web/palacios.git
This will give you the master branch. You probably want the devel branch or one of the release branches. To switch to the devel branch, simply execute
  cd palacios
  git checkout --track -b devel origin/devel
The other branches are similar.


Dynamic migration of memory from node to node
[palacios.git] / palacios / src / palacios / vmm.c
index e9634d6..cfef4f9 100644 (file)
@@ -29,6 +29,8 @@
 #include <palacios/vmm_timeout.h>
 #include <palacios/vmm_options.h>
 #include <palacios/vmm_cpu_mapper.h>
+#include <palacios/vmm_direct_paging.h>
+#include <interfaces/vmm_numa.h>
 
 #ifdef V3_CONFIG_SVM
 #include <palacios/svm.h>
@@ -489,7 +491,121 @@ int v3_move_vm_core(struct v3_vm_info * vm, int vcore_id, int target_cpu) {
     return 0;
 }
 
+/* move a memory region to memory with affinity for a specific physical core */
+int v3_move_vm_mem(struct v3_vm_info * vm, void *gpa, int target_cpu) {
+    int old_node;
+    int new_node;
+    struct v3_mem_region *reg;
+    void *new_hpa;
+    int num_pages;
+    void *old_hpa;
+    int i;
+
+    old_node = v3_numa_gpa_to_node(vm,(addr_t)gpa);
+
+    if (old_node<0) { 
+       PrintError(vm, VCORE_NONE, "Cannot determine current node of gpa %p\n",gpa);
+       return -1;
+    }
+
+    new_node = v3_numa_cpu_to_node(target_cpu);
+
+    if (new_node<0) { 
+       PrintError(vm, VCORE_NONE, "Cannot determine current node of cpu %d\n",target_cpu);
+       return -1;
+    }
+
+    if (new_node==old_node) { 
+       PrintDebug(vm, VCORE_NONE, "Affinity is already established - ignoring request\n");
+       return 0;
+    }
+
+    // We are now going to change the universe, so 
+    // we'll barrier everyone first
+
+    while (v3_raise_barrier(vm, NULL) == -1);
+
+    // get region
+    
+    reg = v3_get_mem_region(vm, V3_MEM_CORE_ANY, (addr_t) gpa);
+
+    if (!reg) { 
+       PrintError(vm, VCORE_NONE, "Attempt to migrate non-existent memory\n");
+       goto out_fail;
+    }
+    
+    if (!(reg->flags.base) || !(reg->flags.alloced)) { 
+       PrintError(vm, VCORE_NONE, "Attempt to migrate invalid region: base=%d alloced=%d\n", reg->flags.base, reg->flags.alloced);
+       goto out_fail;
+    }
+
+    // we now have the allocated base region corresponding to  - and not a copy
+    // we will rewrite this region after moving its contents
+    
+    // first, let's double check that we are in fact changing the numa_id...
 
+    if (reg->numa_id==new_node) { 
+       PrintDebug(vm, VCORE_NONE, "Affinity for this base region is already established - ignoring...\n");
+       goto out_success;
+    }
+
+    // region uses exclusive addressing [guest_start,guest_end)
+    num_pages = (reg->guest_end-reg->guest_start)/PAGE_SIZE;
+
+    // Now we allocate space for the new region with the same constraints as
+    // it originally had
+    new_hpa = V3_AllocPagesExtended(num_pages,
+                                   PAGE_SIZE_4KB,
+                                   new_node,
+                                   reg->flags.limit32 ? V3_ALLOC_PAGES_CONSTRAINT_4GB : 0);
+
+    if (!new_hpa) { 
+       PrintError(vm, VCORE_NONE, "Cannot allocate memory for new base region...\n");
+       goto out_fail;
+    }
+
+    // Note, assumes virtual contiguity in the host OS... 
+    memcpy(V3_VAddr((void*)new_hpa), V3_VAddr((void*)(reg->host_addr)), num_pages*PAGE_SIZE);
+
+    old_hpa = (void*)(reg->host_addr);
+    old_node = (int)(reg->numa_id);
+
+    reg->host_addr = (addr_t)new_hpa;
+    reg->numa_id = v3_numa_hpa_to_node((addr_t)new_hpa);
+
+    // flush all page tables / kill all humans 
+
+    for (i=0;i<vm->num_cores;i++) { 
+       if (vm->cores[i].shdw_pg_mode==SHADOW_PAGING) { 
+           v3_invalidate_shadow_pts(&(vm->cores[i]));
+       } else if (vm->cores[i].shdw_pg_mode==NESTED_PAGING) { 
+           // nested invalidator uses inclusive addressing [start,end], not [start,end)
+           v3_invalidate_nested_addr_range(&(vm->cores[i]),reg->guest_start,reg->guest_end-1);
+       } else {
+           PrintError(vm,VCORE_NONE, "Cannot determine how to invalidate paging structures! Reverting to previous region.\n");
+           // We'll restore things...
+           reg->host_addr = (addr_t) old_hpa;
+           reg->numa_id = old_node;
+           V3_FreePages(new_hpa,num_pages);
+           goto out_fail;
+       }
+    }
+    
+    // Now the old region can go away...
+    V3_FreePages(old_hpa,num_pages);
+    
+    PrintDebug(vm,VCORE_NONE,"Migration of memory complete - new region is %p to %p\n",
+              (void*)(reg->host_addr),(void*)(reg->host_addr+num_pages*PAGE_SIZE-1));
+    
+ out_success:
+    v3_lower_barrier(vm);
+    return 0;
+    
+    
+ out_fail:
+    v3_lower_barrier(vm);
+    return -1;
+}
 
 int v3_stop_vm(struct v3_vm_info * vm) {