target/i386/hax-mem.c

   1 /*
   2  * HAX memory mapping operations
   3  *
   4  * Copyright (c) 2015-16 Intel Corporation
   5  * Copyright 2016 Google, Inc.
   6  *
   7  * This work is licensed under the terms of the GNU GPL, version 2.  See
   8  * the COPYING file in the top-level directory.
   9  */
  10
  11 #include "qemu/osdep.h"
  12 #include "cpu.h"
  13 #include "exec/address-spaces.h"
  14 #include "exec/exec-all.h"
  15
  16 #include "target/i386/hax-i386.h"
  17 #include "qemu/queue.h"
  18
  19 #define DEBUG_HAX_MEM 0
  20
  21 #define DPRINTF(fmt, ...) \
  22     do { \
  23         if (DEBUG_HAX_MEM) { \
  24             fprintf(stdout, fmt, ## __VA_ARGS__); \
  25         } \
  26     } while (0)
  27
  28 /**
  29  * HAXMapping: describes a pending guest physical memory mapping
  30  *
  31  * @start_pa: a guest physical address marking the start of the region; must be
  32  *            page-aligned
  33  * @size: a guest physical address marking the end of the region; must be
  34  *          page-aligned
  35  * @host_va: the host virtual address of the start of the mapping
  36  * @flags: mapping parameters e.g. HAX_RAM_INFO_ROM or HAX_RAM_INFO_INVALID
  37  * @entry: additional fields for linking #HAXMapping instances together
  38  */
  39 typedef struct HAXMapping {
  40     uint64_t start_pa;
  41     uint32_t size;
  42     uint64_t host_va;
  43     int flags;
  44     QTAILQ_ENTRY(HAXMapping) entry;
  45 } HAXMapping;
  46
  47 /*
  48  * A doubly-linked list (actually a tail queue) of the pending page mappings
  49  * for the ongoing memory transaction.
  50  *
  51  * It is used to optimize the number of page mapping updates done through the
  52  * kernel module. For example, it's effective when a driver is digging an MMIO
  53  * hole inside an existing memory mapping. It will get a deletion of the whole
  54  * region, then the addition of the 2 remaining RAM areas around the hole and
  55  * finally the memory transaction commit. During the commit, it will effectively
  56  * send to the kernel only the removal of the pages from the MMIO hole after
  57  * having computed locally the result of the deletion and additions.
  58  */
  59 static QTAILQ_HEAD(HAXMappingListHead, HAXMapping) mappings =
  60     QTAILQ_HEAD_INITIALIZER(mappings);
  61
  62 /**
  63  * hax_mapping_dump_list: dumps @mappings to stdout (for debugging)
  64  */
  65 static void hax_mapping_dump_list(void)
  66 {
  67     HAXMapping *entry;
  68
  69     DPRINTF("%s updates:\n", __func__);
  70     QTAILQ_FOREACH(entry, &mappings, entry) {
  71         DPRINTF("\t%c 0x%016" PRIx64 "->0x%016" PRIx64 " VA 0x%016" PRIx64
  72                 "%s\n", entry->flags & HAX_RAM_INFO_INVALID ? '-' : '+',
  73                 entry->start_pa, entry->start_pa + entry->size, entry->host_va,
  74                 entry->flags & HAX_RAM_INFO_ROM ? " ROM" : "");
  75     }
  76 }
  77
  78 static void hax_insert_mapping_before(HAXMapping *next, uint64_t start_pa,
  79                                       uint32_t size, uint64_t host_va,
  80                                       uint8_t flags)
  81 {
  82     HAXMapping *entry;
  83
  84     entry = g_malloc0(sizeof(*entry));
  85     entry->start_pa = start_pa;
  86     entry->size = size;
  87     entry->host_va = host_va;
  88     entry->flags = flags;
  89     if (!next) {
  90         QTAILQ_INSERT_TAIL(&mappings, entry, entry);
  91     } else {
  92         QTAILQ_INSERT_BEFORE(next, entry, entry);
  93     }
  94 }
  95
  96 static bool hax_mapping_is_opposite(HAXMapping *entry, uint64_t host_va,
  97                                     uint8_t flags)
  98 {
  99     /* removed then added without change for the read-only flag */
 100     bool nop_flags = (entry->flags ^ flags) == HAX_RAM_INFO_INVALID;
 101
 102     return (entry->host_va == host_va) && nop_flags;
 103 }
 104
 105 static void hax_update_mapping(uint64_t start_pa, uint32_t size,
 106                                uint64_t host_va, uint8_t flags)
 107 {
 108     uint64_t end_pa = start_pa + size;
 109     uint32_t chunk_sz;
 110     HAXMapping *entry, *next;
 111
 112     QTAILQ_FOREACH_SAFE(entry, &mappings, entry, next) {
 113         if (start_pa >= entry->start_pa + entry->size) {
 114             continue;
 115         }
 116         if (start_pa < entry->start_pa) {
 117             chunk_sz = end_pa <= entry->start_pa ? size
 118                                                  : entry->start_pa - start_pa;
 119             hax_insert_mapping_before(entry, start_pa, chunk_sz,
 120                                       host_va, flags);
 121             start_pa += chunk_sz;
 122             host_va += chunk_sz;
 123             size -= chunk_sz;
 124         }
 125         chunk_sz = MIN(size, entry->size);
 126         if (chunk_sz) {
 127             bool nop = hax_mapping_is_opposite(entry, host_va, flags);
 128             bool partial = chunk_sz < entry->size;
 129             if (partial) {
 130                 /* remove the beginning of the existing chunk */
 131                 entry->start_pa += chunk_sz;
 132                 entry->host_va += chunk_sz;
 133                 entry->size -= chunk_sz;
 134                 if (!nop) {
 135                     hax_insert_mapping_before(entry, start_pa, chunk_sz,
 136                                               host_va, flags);
 137                 }
 138             } else { /* affects the full mapping entry */
 139                 if (nop) { /* no change to this mapping, remove it */
 140                     QTAILQ_REMOVE(&mappings, entry, entry);
 141                     g_free(entry);
 142                 } else { /* update mapping properties */
 143                     entry->host_va = host_va;
 144                     entry->flags = flags;
 145                 }
 146             }
 147             start_pa += chunk_sz;
 148             host_va += chunk_sz;
 149             size -= chunk_sz;
 150         }
 151         if (!size) { /* we are done */
 152             break;
 153         }
 154     }
 155     if (size) { /* add the leftover */
 156         hax_insert_mapping_before(NULL, start_pa, size, host_va, flags);
 157     }
 158 }
 159
 160 static void hax_process_section(MemoryRegionSection *section, uint8_t flags)
 161 {
 162     MemoryRegion *mr = section->mr;
 163     hwaddr start_pa = section->offset_within_address_space;
 164     ram_addr_t size = int128_get64(section->size);
 165     unsigned int delta;
 166     uint64_t host_va;
 167
 168     /* We only care about RAM pages */
 169     if (!memory_region_is_ram(mr)) {
 170         return;
 171     }
 172
 173     /* Adjust start_pa and size so that they are page-aligned. (Cf
 174      * kvm_set_phys_mem() in kvm-all.c).
 175      */
 176     delta = qemu_real_host_page_size - (start_pa & ~qemu_real_host_page_mask);
 177     delta &= ~qemu_real_host_page_mask;
 178     if (delta > size) {
 179         return;
 180     }
 181     start_pa += delta;
 182     size -= delta;
 183     size &= qemu_real_host_page_mask;
 184     if (!size || (start_pa & ~qemu_real_host_page_mask)) {
 185         return;
 186     }
 187
 188     host_va = (uintptr_t)memory_region_get_ram_ptr(mr)
 189             + section->offset_within_region + delta;
 190     if (memory_region_is_rom(section->mr)) {
 191         flags |= HAX_RAM_INFO_ROM;
 192     }
 193
 194     /* the kernel module interface uses 32-bit sizes (but we could split...) */
 195     g_assert(size <= UINT32_MAX);
 196
 197     hax_update_mapping(start_pa, size, host_va, flags);
 198 }
 199
 200 static void hax_region_add(MemoryListener *listener,
 201                            MemoryRegionSection *section)
 202 {
 203     memory_region_ref(section->mr);
 204     hax_process_section(section, 0);
 205 }
 206
 207 static void hax_region_del(MemoryListener *listener,
 208                            MemoryRegionSection *section)
 209 {
 210     hax_process_section(section, HAX_RAM_INFO_INVALID);
 211     memory_region_unref(section->mr);
 212 }
 213
 214 static void hax_transaction_begin(MemoryListener *listener)
 215 {
 216     g_assert(QTAILQ_EMPTY(&mappings));
 217 }
 218
 219 static void hax_transaction_commit(MemoryListener *listener)
 220 {
 221     if (!QTAILQ_EMPTY(&mappings)) {
 222         HAXMapping *entry, *next;
 223
 224         if (DEBUG_HAX_MEM) {
 225             hax_mapping_dump_list();
 226         }
 227         QTAILQ_FOREACH_SAFE(entry, &mappings, entry, next) {
 228             if (entry->flags & HAX_RAM_INFO_INVALID) {
 229                 /* for unmapping, put the values expected by the kernel */
 230                 entry->flags = HAX_RAM_INFO_INVALID;
 231                 entry->host_va = 0;
 232             }
 233             if (hax_set_ram(entry->start_pa, entry->size,
 234                             entry->host_va, entry->flags)) {
 235                 fprintf(stderr, "%s: Failed mapping @0x%016" PRIx64 "+0x%"
 236                         PRIx32 " flags %02x\n", __func__, entry->start_pa,
 237                         entry->size, entry->flags);
 238             }
 239             QTAILQ_REMOVE(&mappings, entry, entry);
 240             g_free(entry);
 241         }
 242     }
 243 }
 244
 245 /* currently we fake the dirty bitmap sync, always dirty */
 246 static void hax_log_sync(MemoryListener *listener,
 247                          MemoryRegionSection *section)
 248 {
 249     MemoryRegion *mr = section->mr;
 250
 251     if (!memory_region_is_ram(mr)) {
 252         /* Skip MMIO regions */
 253         return;
 254     }
 255
 256     memory_region_set_dirty(mr, 0, int128_get64(section->size));
 257 }
 258
 259 static MemoryListener hax_memory_listener = {
 260     .begin = hax_transaction_begin,
 261     .commit = hax_transaction_commit,
 262     .region_add = hax_region_add,
 263     .region_del = hax_region_del,
 264     .log_sync = hax_log_sync,
 265     .priority = 10,
 266 };
 267
 268 static void hax_ram_block_added(RAMBlockNotifier *n, void *host, size_t size)
 269 {
 270     /*
 271      * In HAX, QEMU allocates the virtual address, and HAX kernel
 272      * populates the memory with physical memory. Currently we have no
 273      * paging, so user should make sure enough free memory in advance.
 274      */
 275     if (hax_populate_ram((uint64_t)(uintptr_t)host, size) < 0) {
 276         fprintf(stderr, "HAX failed to populate RAM");
 277         abort();
 278     }
 279 }
 280
 281 static struct RAMBlockNotifier hax_ram_notifier = {
 282     .ram_block_added = hax_ram_block_added,
 283 };
 284
 285 void hax_memory_init(void)
 286 {
 287     ram_block_notifier_add(&hax_ram_notifier);
 288     memory_listener_register(&hax_memory_listener, &address_space_memory);
 289 }