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.


reorganized swapbypass to isolate it to a special shadow paging implementation, and...
[palacios.git] / palacios / src / palacios / vmm_shdw_pg_swapbypass.c
index 2330f6d..af23fed 100644 (file)
  */
 
 #include <palacios/vmm_shadow_paging.h>
+#include <palacios/vmm_swapbypass.h>
 #include <palacios/vmm_ctrl_regs.h>
 
 #include <palacios/vm_guest.h>
 #include <palacios/vm_guest_mem.h>
+#include <palacios/vmm_paging.h>
+#include <palacios/vmm_hashtable.h>
+
+
+#ifdef CONFIG_SWAPBYPASS_TELEMETRY
+#include <palacios/vmm_telemetry.h>
+#endif
+
+
+
+
+
+
+// This is a hack and 32 bit linux specific.... need to fix...
+struct swap_pte {
+    uint32_t present    : 1;
+    uint32_t dev_index  : 8;
+    uint32_t pg_index   : 23;
+};
+
+
+struct shadow_pointer {
+    uint32_t pg_index;
+    uint32_t dev_index;
+
+    pte32_t * shadow_pte;
+    
+    addr_t guest_pte;
+    
+    struct list_head node;
+};
+
+
+
+
+
 
 struct shadow_page_data {
     v3_reg_t cr3;
@@ -37,10 +74,170 @@ struct swapbypass_local_state {
 
 };
 
+struct v3_swap_dev {
+    uint8_t present;
+
+    struct v3_swap_ops * ops;
+
+    void * private_data;
+};
+
+
+struct swapbypass_vm_state {
+    struct v3_swap_dev devs[256];
+
+#ifdef CONFIG_SWAPBYPASS_TELEMETRY
+    uint32_t read_faults;
+    uint32_t write_faults;
+    uint32_t flushes;
+    uint32_t mapped_pages;
+    uint32_t list_size;
+#endif
+
+    // shadow pointers
+    struct hashtable * shdw_ptr_ht;
+};
+
+
+
+
+static uint_t swap_hash_fn(addr_t key) {
+    return v3_hash_long(key, 32);
+}
+
+
+static int swap_eq_fn(addr_t key1, addr_t key2) {
+    return (key1 == key2);
+}
+
+
+
+static inline uint32_t get_pg_index(pte32_t * pte) {
+    return ((struct swap_pte *)pte)->pg_index;
+}
+
+
+static inline uint32_t get_dev_index(pte32_t * pte) {
+    return ((struct swap_pte *)pte)->dev_index;
+}
+
+
+// Present = 0 and Dirty = 0
+// fixme
+static inline int is_swapped_pte32(pte32_t * pte) {
+    return ((pte->present == 0) && (*(uint32_t *)pte != 0));
+}
+
+
+
 
 static struct shadow_page_data * create_new_shadow_pt(struct guest_info * core);
 
 
+
+#ifdef CONFIG_SWAPBYPASS_TELEMETRY
+static void telemetry_cb(struct v3_vm_info * vm, void * private_data, char * hdr) {
+    struct swapbypass_vm_state * swap_state = (struct swapbypass_vm_state *)(vm->shdw_impl.impl_data);
+
+    V3_Print("%sSymbiotic Swap:\n", hdr);
+    V3_Print("%s\tRead faults=%d\n", hdr, swap_state->read_faults);
+    V3_Print("%s\tWrite faults=%d\n", hdr, swap_state->write_faults);
+    V3_Print("%s\tMapped Pages=%d\n", hdr, swap_state->mapped_pages);
+    V3_Print("%s\tFlushes=%d\n", hdr, swap_state->flushes);
+    V3_Print("%s\tlist size=%d\n", hdr, swap_state->list_size);
+}
+#endif
+
+
+
+
+
+
+
+
+
+static int get_vaddr_perms(struct guest_info * info, addr_t vaddr, pte32_t * guest_pte, pf_error_t * page_perms) {
+    uint64_t pte_val = (uint64_t)*(uint32_t *)guest_pte;
+
+    // symcall to check if page is in cache or on swap disk
+    if (v3_sym_call3(info, SYMCALL_MEM_LOOKUP, (uint64_t *)&vaddr, (uint64_t *)&pte_val, (uint64_t *)page_perms) == -1) {
+       PrintError("Sym call error?? that's weird... \n");
+       return -1;
+    }
+
+    //    V3_Print("page perms = %x\n", *(uint32_t *)page_perms);
+
+    if (vaddr == 0) {
+       return 1;
+    }
+
+    return 0;
+}
+
+
+static addr_t get_swapped_pg_addr(struct v3_vm_info * vm, pte32_t * guest_pte) {
+   struct swapbypass_vm_state * swap_state = (struct swapbypass_vm_state *)(vm->shdw_impl.impl_data);
+    int dev_index = get_dev_index(guest_pte);
+    struct v3_swap_dev * swp_dev = &(swap_state->devs[dev_index]);
+
+
+    if (! swp_dev->present ) {
+       return 0;
+    }
+
+    return (addr_t)swp_dev->ops->get_swap_entry(get_pg_index(guest_pte), swp_dev->private_data);
+}
+
+
+
+static addr_t map_swp_page(struct v3_vm_info * vm, pte32_t * shadow_pte, pte32_t * guest_pte, void * swp_page_ptr) {
+   struct swapbypass_vm_state * swap_state = (struct swapbypass_vm_state *)(vm->shdw_impl.impl_data);
+    struct list_head * shdw_ptr_list = NULL;
+    struct shadow_pointer * shdw_ptr = NULL;
+
+
+
+    if (swp_page_ptr == NULL) {
+       //      PrintError("Swapped out page not found on swap device\n");
+       return 0;
+    }
+
+    shdw_ptr_list = (struct list_head *)v3_htable_search(swap_state->shdw_ptr_ht, (addr_t)*(uint32_t *)guest_pte);
+
+    if (shdw_ptr_list == NULL) {
+       shdw_ptr_list = (struct list_head *)V3_Malloc(sizeof(struct list_head *));
+#ifdef CONFIG_SWAPBYPASS_TELEMETRY
+       swap_state->list_size++;
+#endif
+       INIT_LIST_HEAD(shdw_ptr_list);
+       v3_htable_insert(swap_state->shdw_ptr_ht, (addr_t)*(uint32_t *)guest_pte, (addr_t)shdw_ptr_list);
+    }
+
+    shdw_ptr = (struct shadow_pointer *)V3_Malloc(sizeof(struct shadow_pointer));
+
+    if (shdw_ptr == NULL) {
+       PrintError("MEMORY LEAK\n");
+#ifdef CONFIG_SWAPBYPASS_TELEMETRY
+       telemetry_cb(vm, NULL, "");
+#endif
+       return 0;
+    }
+
+    shdw_ptr->shadow_pte = shadow_pte;
+    shdw_ptr->guest_pte = *(uint32_t *)guest_pte;
+    shdw_ptr->pg_index = get_pg_index(guest_pte);
+    shdw_ptr->dev_index = get_dev_index(guest_pte);
+
+    // We don't check for conflicts, because it should not happen...
+    list_add(&(shdw_ptr->node), shdw_ptr_list);
+
+    return PAGE_BASE_ADDR((addr_t)V3_PAddr(swp_page_ptr));
+}
+
+
+
+
+
 #include "vmm_shdw_pg_swapbypass_32.h"
 #include "vmm_shdw_pg_swapbypass_32pae.h"
 #include "vmm_shdw_pg_swapbypass_64.h"
@@ -92,9 +289,121 @@ static struct shadow_page_data * create_new_shadow_pt(struct guest_info * core)
 
 
 
+
+
+
+int v3_register_swap_disk(struct v3_vm_info * vm, int dev_index, 
+                         struct v3_swap_ops * ops, void * private_data) {
+    struct swapbypass_vm_state * swap_state = (struct swapbypass_vm_state *)(vm->shdw_impl.impl_data);
+
+    swap_state->devs[dev_index].present = 1;
+    swap_state->devs[dev_index].private_data = private_data;
+    swap_state->devs[dev_index].ops = ops;
+
+    return 0;
+}
+
+
+
+
+int v3_swap_in_notify(struct v3_vm_info * vm, int pg_index, int dev_index) {
+    struct list_head * shdw_ptr_list = NULL;
+    struct swapbypass_vm_state * swap_state = (struct swapbypass_vm_state *)(vm->shdw_impl.impl_data);
+    struct shadow_pointer * tmp_shdw_ptr = NULL;
+    struct shadow_pointer * shdw_ptr = NULL;
+    struct swap_pte guest_pte = {0, dev_index, pg_index};
+
+    shdw_ptr_list = (struct list_head * )v3_htable_search(swap_state->shdw_ptr_ht, *(addr_t *)&(guest_pte));
+
+    if (shdw_ptr_list == NULL) {
+       return 0;
+    }
+
+    list_for_each_entry_safe(shdw_ptr, tmp_shdw_ptr, shdw_ptr_list, node) {
+       if ((shdw_ptr->pg_index == pg_index) &&
+           (shdw_ptr->dev_index == dev_index)) {
+
+           // Trigger faults for next shadow access
+           shdw_ptr->shadow_pte->present = 0;
+
+           // Delete entry from list
+           list_del(&(shdw_ptr->node));
+           V3_Free(shdw_ptr);
+       }
+    }
+
+    return 0;
+}
+
+
+
+int v3_swap_flush(struct v3_vm_info * vm) {
+    struct swapbypass_vm_state * swap_state = (struct swapbypass_vm_state *)(vm->shdw_impl.impl_data);
+    struct hashtable_iter * ht_iter = v3_create_htable_iter(swap_state->shdw_ptr_ht);
+
+    //    PrintDebug("Flushing Symbiotic Swap table\n");
+
+#ifdef CONFIG_SWAPBYPASS_TELEMETRY
+    swap_state->flushes++;
+#endif
+
+    if (!ht_iter) {
+       PrintError("NULL iterator in swap flush!! Probably will crash soon...\n");
+    }
+
+    while (ht_iter->entry) {
+       struct shadow_pointer * tmp_shdw_ptr = NULL;
+       struct shadow_pointer * shdw_ptr = NULL;
+       struct list_head * shdw_ptr_list = (struct list_head *)v3_htable_get_iter_value(ht_iter);
+
+       // delete all swapped entries
+       // we can leave the list_head structures and reuse them for the next round
+       
+       list_for_each_entry_safe(shdw_ptr, tmp_shdw_ptr, shdw_ptr_list, node) {
+           if (shdw_ptr == NULL) {
+               PrintError("Null shadow pointer in swap flush!! Probably crashing soon...\n");
+           }
+
+           // Trigger faults for next shadow access
+           shdw_ptr->shadow_pte->present = 0;
+           
+           // Delete entry from list
+           list_del(&(shdw_ptr->node));
+           V3_Free(shdw_ptr);
+       }
+
+       v3_htable_iter_advance(ht_iter);
+    }
+
+    V3_Free(ht_iter);
+
+    return 0;
+}
+
+
+
+
+
+
+
 static int sb_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
+    struct v3_shdw_impl_state * impl_state = &(vm->shdw_impl);
+    struct swapbypass_vm_state * sb_state = NULL;
+
+    memset(sb_state, 0, sizeof(struct swapbypass_vm_state));
+    sb_state->shdw_ptr_ht = v3_create_htable(0, swap_hash_fn, swap_eq_fn);
+
+#ifdef CONFIG_SWAPBYPASS_TELEMETRY
+    if (vm->enable_telemetry) {
+       v3_add_telemetry_cb(vm, telemetry_cb, NULL);
+    }
+#endif
+
+    impl_state->impl_data = sb_state;
+
+    PrintDebug("Initialized SwapBypass\n");
+
 
-    V3_Print("SWAPBYPASS initialization\n");
     return 0;
 }
 
@@ -108,7 +417,6 @@ static int sb_local_init(struct guest_info * core) {
 
     V3_Print("SWAPBYPASS local initialization\n");
 
-
     swapbypass_state = (struct swapbypass_local_state *)V3_Malloc(sizeof(struct swapbypass_local_state));
 
     INIT_LIST_HEAD(&(swapbypass_state->page_list));