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.


Added support for using large shadow pages with 32-bit 4MB page table entries
Patrick Bridges [Thu, 12 Aug 2010 17:23:22 +0000 (11:23 -0600)]
palacios/src/palacios/mmu/vmm_shdw_pg_tlb_32.h

index 10c3f9d..83bfb10 100644 (file)
@@ -46,8 +46,11 @@ static inline int activate_shadow_pt_32(struct guest_info * core) {
  * *
  * *
  */
-static int handle_4MB_shadow_pagefault_32(struct guest_info * info,  addr_t fault_addr, pf_error_t error_code, 
-                                         pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde);
+static int handle_4MB_shadow_pagefault_pde_32(struct guest_info * info,  addr_t fault_addr, pf_error_t error_code, 
+                                             pt_access_status_t shadow_pde_access, pde32_4MB_t * large_shadow_pde, 
+                                             pde32_4MB_t * large_guest_pde);
+static int handle_4MB_shadow_pagefault_pte_32(struct guest_info * info,  addr_t fault_addr, pf_error_t error_code, 
+                                             pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde);
 
 static int handle_pte_shadow_pagefault_32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code,
                                          pte32_t * shadow_pt,  pte32_t * guest_pt);
@@ -129,6 +132,28 @@ static inline int handle_shadow_pagefault_32(struct guest_info * info, addr_t fa
     // Get the next shadow page level, allocate if not present
 
     if (shadow_pde_access == PT_ACCESS_NOT_PRESENT) {
+
+        if (info->use_large_pages && guest_pde->large_page) {
+            // Check underlying physical memory map to see if a large page is viable
+            addr_t guest_pa = BASE_TO_PAGE_ADDR_4MB(((pde32_4MB_t *)guest_pde)->page_base_addr);
+            addr_t host_pa;
+            if (v3_get_max_page_size(info, guest_pa, PAGE_SIZE_4MB) < PAGE_SIZE_4MB) {
+                PrintDebug("Underlying physical memory map doesn't allow use of a large page.\n");
+                // Fallthrough to small pages
+            } else if ((v3_gpa_to_hpa(info, guest_pa, &host_pa) != 0)
+                       || (v3_compute_page_alignment(host_pa) < PAGE_SIZE_4MB)) {
+                PrintDebug("Host memory alignment doesn't allow use of a large page.\n");
+                // Fallthrough to small pages
+            } else if (handle_4MB_shadow_pagefault_pde_32(info, fault_addr, error_code, shadow_pde_access,
+                                                          (pde32_4MB_t *)shadow_pde, (pde32_4MB_t *)guest_pde) == 0) {
+                return 0;
+            } else {
+                PrintError("Error handling large pagefault with large page\n");
+                return -1;
+            }
+            // Fallthrough to handle the region with small pages
+        }
+
        struct shadow_page_data * shdw_page =  create_new_shadow_pt(info);
        shadow_pt = (pte32_t *)V3_VAddr((void *)shdw_page->page_pa);
 
@@ -181,7 +206,7 @@ static inline int handle_shadow_pagefault_32(struct guest_info * info, addr_t fa
            return -1;
        }
     } else {
-       if (handle_4MB_shadow_pagefault_32(info, fault_addr, error_code, shadow_pt, (pde32_4MB_t *)guest_pde) == -1) {
+       if (handle_4MB_shadow_pagefault_pte_32(info, fault_addr, error_code, shadow_pt, (pde32_4MB_t *)guest_pde) == -1) {
            PrintError("Error handling large pagefault\n");
            return -1;
        }       
@@ -322,9 +347,8 @@ static int handle_pte_shadow_pagefault_32(struct guest_info * info, addr_t fault
     return 0;
 }
 
-
-
-static int handle_4MB_shadow_pagefault_32(struct guest_info * info, 
+// Handle a 4MB page fault with small pages in the PTE
+static int handle_4MB_shadow_pagefault_pte_32(struct guest_info * info, 
                                     addr_t fault_addr, pf_error_t error_code, 
                                     pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde) 
 {
@@ -414,20 +438,98 @@ static int handle_4MB_shadow_pagefault_32(struct guest_info * info,
        return -1;
     }
 
-    PrintDebug("Returning from large page fault handler\n");
+    PrintDebug("Returning from large page->small page fault handler\n");
     return 0;
 }
 
 
+// Handle a 4MB page fault with a 4MB page in the PDE
+static int handle_4MB_shadow_pagefault_pde_32(struct guest_info * info, 
+                                    addr_t fault_addr, pf_error_t error_code, 
+                                    pt_access_status_t shadow_pde_access,
+                                    pde32_4MB_t * large_shadow_pde, pde32_4MB_t * large_guest_pde) 
+{
+    addr_t guest_fault_pa = BASE_TO_PAGE_ADDR_4MB(large_guest_pde->page_base_addr) + PAGE_OFFSET_4MB(fault_addr);  
+
+
+    PrintDebug("Handling 4MB fault with large page (guest_fault_pa=%p) (error_code=%x)\n", (void *)guest_fault_pa, *(uint_t*)&error_code);
+    PrintDebug("LargeShadowPDE=%p, LargeGuestPDE=%p\n", large_shadow_pde, large_guest_pde);
+
+    struct v3_mem_region * shdw_reg = v3_get_mem_region(info->vm_info, info->cpu_id, guest_fault_pa);
+
+    if (shdw_reg == NULL) {
+       // Inject a machine check in the guest
+       PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_fault_pa);
+       v3_raise_exception(info, MC_EXCEPTION);
+       return -1;
+    }
+
+    if (shadow_pde_access == PT_ACCESS_OK) {
+       // Inconsistent state...
+       // Guest Re-Entry will flush tables and everything should now workd
+       PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n");
+       return 0;
+    }
+
+  
+    if (shadow_pde_access == PT_ACCESS_NOT_PRESENT) {
+       // Get the guest physical address of the fault
+
+       if ((shdw_reg->flags.alloced == 1) && 
+           (shdw_reg->flags.read  == 1)) {
+           addr_t shadow_pa = 0;
+
 
+           if (v3_gpa_to_hpa(info, guest_fault_pa, &shadow_pa) == -1) {
+               PrintError("could not translate page fault address (%p)\n", (void *)guest_fault_pa);
+               return -1;
+           }
 
+           PrintDebug("\tMapping shadow page (%p)\n", (void *)BASE_TO_PAGE_ADDR(shadow_pte->page_base_addr));
 
+            large_guest_pde->vmm_info = V3_LARGE_PG; /* For invalidations */
+            large_shadow_pde->page_base_addr = PAGE_BASE_ADDR_4MB(shadow_pa);
+            large_shadow_pde->large_page = 1;
+            large_shadow_pde->present = 1;
+            large_shadow_pde->user_page = 1;
 
+            if (shdw_reg->flags.write == 0) {
+                large_shadow_pde->writable = 0;
+            } else {
+                large_shadow_pde->writable = 1;
+            }
 
+           //set according to VMM policy
+           large_shadow_pde->write_through = large_guest_pde->write_through;
+           large_shadow_pde->cache_disable = large_guest_pde->cache_disable;
+           large_shadow_pde->global_page = large_guest_pde->global_page;
+           //
 
+       } else {
+           if (shdw_reg->unhandled(info, fault_addr, guest_fault_pa, shdw_reg, error_code) == -1) {
+               PrintError("Special Page Fault handler returned error for address: %p\n", (void *)fault_addr);
+               return -1;
+           }
+       }
+    } else if (shadow_pde_access == PT_ACCESS_WRITE_ERROR) {
 
+       if (shdw_reg->flags.write == 0) {
+           if (shdw_reg->unhandled(info, fault_addr, guest_fault_pa, shdw_reg, error_code) == -1) {
+               PrintError("Special Page Fault handler returned error for address: %p\n", (void *)fault_addr);
+               return -1;
+           }
+       }
 
+    } else {
+       PrintError("Error in large page fault handler...\n");
+       PrintError("This case should have been handled at the top level handler\n");
+       return -1;
+    }
 
+    PrintDebug("Returning from large page->large page fault handler\n");
+    return 0;
+}
 
 /* If we start to optimize we should look up the guest pages in the cache... */
 static inline int handle_shadow_invlpg_32(struct guest_info * info, addr_t vaddr) {