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.


Updated userspace guest memory access tools for regions, plus added new fast function...
[palacios.git] / linux_usr / v3_guest_mem.c
index 0bdc6e4..ed1cd44 100644 (file)
@@ -7,66 +7,88 @@
 #include "v3_guest_mem.h"
 
 
-# Warning FIX THE PARSER TO CONFORM TO NEW /proc/v3 VM output format
 
-
-#define GUEST_FILE "/proc/v3vee/v3-guests"
-//#define GUEST_FILE "/441/pdinda/test.proc"
+#define GUEST_FILE "/proc/v3vee/v3-guests-details"
+//#define GUEST_FILE "/v-test/numa/palacios-devel/test.proc"
 #define MAXLINE 65536
 
 struct v3_guest_mem_map * v3_guest_mem_get_map(char *vmdev)
 {
   FILE *f;
+  int rc;
+  int i;
   char buf[MAXLINE];
-  char name[MAXLINE];
   char dev[MAXLINE];
-  char state[MAXLINE];
-  uint64_t start, end;
+  uint64_t start, end, num;
+  uint64_t guest_cur;
+  uint64_t num_regions;
   
 
   if (!(f=fopen(GUEST_FILE,"r"))) { 
-    fprintf(stderr,"Cannot open %s - is Palacios active?\n",GUEST_FILE);
-    return 0;
-  }
-  
-  // This is using the current "single memory region" model
-  // and will change when /proc convention changes to conform to
-  // multiple region model
-  while (fgets(buf,MAXLINE,f)) {
-    if (sscanf(buf,
-              "%s %s %s [0x%llx-0x%llx]",
-              name,
-              dev,
-              state,
-              &start,
-              &end)!=5) { 
-      fprintf(stderr, "Cannot parse following line\n%s\n",buf);
-      fclose(f);
+      fprintf(stderr,"Cannot open %s - is Palacios active?\n",GUEST_FILE);
       return 0;
-    }
-    if (!strcmp(dev,vmdev)) { 
-      struct v3_guest_mem_map *m = 
-       (struct v3_guest_mem_map *) malloc(sizeof(struct v3_guest_mem_map)+1*sizeof(struct v3_guest_mem_block));
-      if (!m) { 
-       fprintf(stderr, "Cannot allocate space\n");
-       fclose(f);
-       return 0;
-      }
+  }
 
-      memset(m,0,sizeof(struct v3_guest_mem_map)+1*sizeof(struct v3_guest_mem_block));
-      
-      m->numblocks=1;
-      m->block[0].gpa=0;
-      m->block[0].hpa=(void*)start;
-      m->block[0].numpages = (end-start+1) / 4096;
+  while (1) {
+      if (!fgets(buf,MAXLINE,f)) {
+         fprintf(stderr,"Could not find info for %s\n",vmdev);
+         return 0;
+      }
+      if (sscanf(buf,"Device: %s",dev)==1) { 
+         if (!strcmp(dev,vmdev)) {
+             // found our VM
+             break;
+         } 
+      }
+  }
+       
+  // Now we need the number of regions
+  while (1) {
+      if (!fgets(buf,MAXLINE,f)) {
+         fprintf(stderr,"Could not find number of regions for %s\n",vmdev);
+         return 0;
+      }
+      if (sscanf(buf,"Regions: %llu",&num_regions)==1) {
+         break;
+      }
+  }
+  struct v3_guest_mem_map *m = 
+      (struct v3_guest_mem_map *) malloc(sizeof(struct v3_guest_mem_map)+num_regions*sizeof(struct v3_guest_mem_block));
+  if (!m) { 
+      fprintf(stderr, "Cannot allocate space\n");
       fclose(f);
-      return m;
-    }
+      return 0;
   }
   
-  fprintf(stderr,"%s not found\n",vmdev);
+  memset(m,0,sizeof(struct v3_guest_mem_map)+num_regions*sizeof(struct v3_guest_mem_block));
+
+  m->numblocks=num_regions;
+
+  // Now collect the region info
+  guest_cur=0;
+  while (i<num_regions) { 
+      if (!fgets(buf,MAXLINE,f)) {
+         fprintf(stderr,"Did not find all regions...\n");
+         free(m);
+         return 0;
+      }
+      if (sscanf(buf," region %d has HPAs %llx-%llx",&num,&start,&end)==3) { 
+         m->block[i].gpa = (void*)guest_cur;
+         m->block[i].hpa = (void*)start;
+         m->block[i].numpages = (end-start) / 4096 + !!((end-start) % 4096);
+         if ((end-start)%4096) { 
+             fprintf(stderr,"Odd, region %d is a non-integral number of pages");
+         }
+         guest_cur+=end-start;
+         m->block[i].cumgpa=(void*)(guest_cur-1);
+         i++;
+      }
+  }
+
   fclose(f);
-  return 0;
+
+  return m;
 
 }
 
@@ -88,8 +110,8 @@ int v3_map_guest_mem(struct v3_guest_mem_map *map)
   }
 
   for (i=0; i<map->numblocks; i++) { 
-    fprintf(stderr,"Mapping %llu bytes of /dev/mem offset 0x%llx\n",
-           map->block[i].numpages*4096, (off_t)(map->block[i].hpa));
+      //fprintf(stderr,"Mapping %llu bytes of /dev/mem offset 0x%llx\n",
+      //    map->block[i].numpages*4096, (off_t)(map->block[i].hpa));
     map->block[i].uva = mmap(NULL, 
                             map->block[i].numpages*4096,
                             PROT_READ | PROT_WRITE, 
@@ -125,3 +147,99 @@ int v3_unmap_guest_mem(struct v3_guest_mem_map *map)
   }
   return 0;
 }
+
+
+void *v3_gpa_start(struct v3_guest_mem_map *map)
+{
+    return 0; // all guests start at zero for now
+}
+
+void *v3_gpa_end(struct v3_guest_mem_map *map)
+{
+    struct v3_guest_mem_block *l = &(map->block[map->numblocks-1]); 
+
+    // currently, the regions are consecutive, so we just need the last block
+    return l->gpa+l->numpages*4096-1; 
+}
+
+
+int v3_guest_mem_apply(void (*func)(void *data, uint64_t num_bytes, void *priv),
+                      struct v3_guest_mem_map *map, void *gpa, uint64_t num_bytes, void *priv)
+{
+    void *cur_gpa;
+    void *cur_uva;
+    uint64_t left_bytes;
+    uint64_t block_bytes;
+    
+    if (!(map->fd)) {
+       return -1;
+    }
+    
+    if (gpa < v3_gpa_start(map) || gpa+num_bytes-1 > v3_gpa_end(map)) { 
+       return -1;
+    }
+
+    cur_gpa = gpa;
+    left_bytes = num_bytes;
+
+    while (left_bytes) { 
+       cur_uva = v3_gpa_to_uva(map, cur_gpa, &block_bytes);
+       if (!cur_uva) { 
+           return -1;
+       }
+       if (block_bytes>left_bytes) { 
+           block_bytes = left_bytes;
+       }
+       func(cur_uva,block_bytes,priv);
+       left_bytes-=block_bytes;
+       cur_gpa+=block_bytes;
+    }
+    
+    return 0;
+}
+
+
+
+static void copy_out(void *uva, uint64_t num_bytes, void *curoff)
+{
+    memcpy(*((void**)(curoff)), uva, num_bytes);
+    *(void**)curoff += num_bytes;
+}
+
+static void copy_in(void *uva, uint64_t num_bytes, void *curoff)
+{
+    memcpy(uva, *((void**)(curoff)), num_bytes);
+    *(void**)curoff += num_bytes;
+}
+
+static void do_hash(void *uva, uint64_t num_bytes, void *priv)
+{
+    uint64_t i;
+    uint64_t *curhash = (uint64_t *)priv;
+
+    for (i=0;i<num_bytes;i++) { 
+       *curhash += ((uint8_t*)uva)[i];
+    }
+}
+
+int v3_guest_mem_read(struct v3_guest_mem_map *map, void *gpa, uint64_t num_bytes, char *data)
+{
+    void *cpy_ptr=data;
+
+    return v3_guest_mem_apply(copy_out,map,gpa,num_bytes,&cpy_ptr);
+}
+
+int v3_guest_mem_write(struct v3_guest_mem_map *map, void *gpa, uint64_t num_bytes, char *data)
+{
+    void *cpy_ptr=data;
+
+    return v3_guest_mem_apply(copy_in,map,gpa,num_bytes,&cpy_ptr);
+}
+
+int v3_guest_mem_hash(struct v3_guest_mem_map *map, void *gpa, uint64_t num_bytes, uint64_t *hash)
+{
+    *hash = 0;
+
+    return v3_guest_mem_apply(do_hash,map,gpa,num_bytes,hash);
+}
+