Split telemetry into global and per-core to avoid race printing core telemetry

[palacios.git] / palacios / src / palacios / vmm_direct_paging_64.h

diff --git a/palacios/src/palacios/vmm_direct_paging_64.h b/palacios/src/palacios/vmm_direct_paging_64.h
index d45ae17..217373a 100644 (file)
--- a/palacios/src/palacios/vmm_direct_paging_64.h
+++ b/palacios/src/palacios/vmm_direct_paging_64.h
@@ -29,75 +29,6 @@
 
 // Reference: AMD Software Developer Manual Vol.2 Ch.5 "Page Translation and Protection"
 
-static uint32_t get_page_size(struct guest_info * core, addr_t fault_addr) {
-    addr_t pg_start = 0UL, pg_end = 0UL; // 2MiB page containing the faulting address
-    struct v3_mem_region * pg_next_reg = NULL; // next immediate mem reg after page start addr
-    uint32_t page_size = PAGE_SIZE_4KB;
-
-   /* If the guest has been configured for 2MiB pages, then we must check for hooked regions of
-     * memory which may overlap with the 2MiB page containing the faulting address (due to
-     * potentially differing access policies in place for e.g. i/o devices and APIC). A 2MiB page
-     * can be used if a) no region overlaps the page [or b) a region does overlap but fully contains
-     * the page]. The [bracketed] text pertains to the #if 0'd code below, state D. TODO modify this
-     * note if someone decides to enable this optimization. It can be tested with the SeaStar
-     * mapping.
-     *
-     * Examples: (CAPS regions are returned by v3_get_next_mem_region; state A returns the base reg)
-     *
-     *    |region| |region|                               2MiB mapped (state A)
-     *                   |reg|          |REG|             2MiB mapped (state B)
-     *   |region|     |reg|   |REG| |region|   |reg|      4KiB mapped (state C)
-     *        |reg|  |reg|   |--REGION---|                [2MiB mapped (state D)]
-     * |--------------------------------------------|     RAM
-     *                             ^                      fault addr
-     * |----|----|----|----|----|page|----|----|----|     2MB pages
-     *                           >>>>>>>>>>>>>>>>>>>>     search space
-     */
-
-
-    // guest page maps to a host page + offset (so when we shift, it aligns with a host page)
-    pg_start = PAGE_ADDR_2MB(fault_addr);
-    pg_end = (pg_start + PAGE_SIZE_2MB);
-
-    PrintDebug("%s: page   [%p,%p) contains address\n", __FUNCTION__, (void *)pg_start, (void *)pg_end);
-
-    pg_next_reg = v3_get_next_mem_region(core->vm_info, core->cpu_id, pg_start);
-
-    if (pg_next_reg == NULL) {
-       PrintError("%s: Error: address not in base region, %p\n", __FUNCTION__, (void *)fault_addr);
-       return PAGE_SIZE_4KB;
-    }
-
-    if (pg_next_reg->flags.base == 1) {
-       page_size = PAGE_SIZE_2MB; // State A
-    } else {
-#if 0       // State B/C and D optimization
-       if ((pg_next_reg->guest_end >= pg_end) &&
-           ((pg_next_reg->guest_start >= pg_end) || (pg_next_reg->guest_start <= pg_start))) {     
-           page_size = PAGE_SIZE_2MB;
-       }
-
-       PrintDebug("%s: region [%p,%p) %s partially overlap with page\n", __FUNCTION__,
-                  (void *)pg_next_reg->guest_start, (void *)pg_next_reg->guest_end, 
-                  (page_size == PAGE_SIZE_2MB) ? "does not" : "does");
-
-#else       // State B/C
-       if (pg_next_reg->guest_start >= pg_end) {
-           
-           page_size = PAGE_SIZE_2MB;
-       }
-
-       PrintDebug("%s: region [%p,%p) %s overlap with page\n", __FUNCTION__,
-                  (void *)pg_next_reg->guest_start, (void *)pg_next_reg->guest_end,
-                  (page_size == PAGE_SIZE_2MB) ? "does not" : "does");
-
-#endif
-    }
-
-    return page_size;
-}
-
-
 static inline int handle_passthrough_pagefault_64(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) {
     pml4e64_t * pml      = NULL;
     pdpe64_t * pdpe      = NULL;
@@ -111,7 +42,7 @@ static inline int handle_passthrough_pagefault_64(struct guest_info * core, addr
     int pde_index  = PDE64_INDEX(fault_addr);
     int pte_index  = PTE64_INDEX(fault_addr);
 
-    struct v3_mem_region * region =  v3_get_mem_region(core->vm_info, core->cpu_id, fault_addr);
+    struct v3_mem_region * region =  v3_get_mem_region(core->vm_info, core->vcpu_id, fault_addr);
     int page_size = PAGE_SIZE_4KB;
 
     if (region == NULL) {
@@ -123,8 +54,8 @@ static inline int handle_passthrough_pagefault_64(struct guest_info * core, addr
      *  1. the guest is configured to use large pages and 
      *         2. the memory regions can be referenced by a large page
      */
-    if ((core->use_large_pages == 1) ) {
-       page_size = get_page_size(core, fault_addr);
+    if ((core->use_large_pages == 1) || (core->use_giant_pages == 1)) {
+       page_size = v3_get_max_page_size(core, fault_addr, LONG);
     }
 
     PrintDebug("Using page size of %dKB\n", page_size / 1024);
@@ -287,6 +218,8 @@ static inline int invalidate_addr_64(struct guest_info * core, addr_t inv_addr)
        return 0;
     } else if (pdpe[pdpe_index].large_page == 1) { // 1GiB
        pdpe[pdpe_index].present = 0;
+       pdpe[pdpe_index].writable = 0;
+       pdpe[pdpe_index].user_page = 0;
        return 0;
     }
 
@@ -296,12 +229,16 @@ static inline int invalidate_addr_64(struct guest_info * core, addr_t inv_addr)
        return 0;
     } else if (pde[pde_index].large_page == 1) { // 2MiB
        pde[pde_index].present = 0;
+       pde[pde_index].writable = 0;
+       pde[pde_index].user_page = 0;
        return 0;
     }
 
     pte = V3_VAddr((void*)BASE_TO_PAGE_ADDR(pde[pde_index].pt_base_addr));
 
     pte[pte_index].present = 0; // 4KiB
+    pte[pte_index].writable = 0;
+    pte[pte_index].user_page = 0;
 
     return 0;
 }