exec: reintroduce MemoryRegion caching

MemoryRegionCache was reverted to "normal" address_space_* operations
for 2.9, due to lack of support for IOMMUs.  Reinstate the
optimizations, caching only the IOMMU translation at address_cache_init
but not the IOMMU lookup and target AddressSpace translation are not
cached; now that MemoryRegionCache supports IOMMUs, it becomes more widely
applicable too.

The inlined fast path is defined in memory_ldst_cached.inc.h, while the
slow path uses memory_ldst.inc.c as before.  The smaller fast path causes
a little code size reduction in MemoryRegionCache users:

    hw/virtio/virtio.o text size before: 32373
    hw/virtio/virtio.o text size after: 31941

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This commit is contained in:
Paolo Bonzini 2018-03-18 18:26:36 +01:00
parent a411c84b56
commit 48564041a7
6 changed files with 280 additions and 20 deletions

View file

@ -1688,12 +1688,16 @@ MemTxResult address_space_write(AddressSpace *as, hwaddr addr,
#include "exec/memory_ldst_phys.inc.h"
struct MemoryRegionCache {
void *ptr;
hwaddr xlat;
hwaddr len;
AddressSpace *as;
FlatView *fv;
MemoryRegionSection mrs;
bool is_write;
};
#define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .as = NULL })
#define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .mrs.mr = NULL })
/* address_space_ld*_cached: load from a cached #MemoryRegion
* address_space_st*_cached: store into a cached #MemoryRegion
@ -1719,11 +1723,40 @@ struct MemoryRegionCache {
* if NULL, this information is discarded
*/
#define SUFFIX _cached
#define SUFFIX _cached_slow
#define ARG1 cache
#define ARG1_DECL MemoryRegionCache *cache
#include "exec/memory_ldst.inc.h"
/* Inline fast path for direct RAM access. */
static inline uint8_t address_space_ldub_cached(MemoryRegionCache *cache,
hwaddr addr, MemTxAttrs attrs, MemTxResult *result)
{
assert(addr < cache->len);
if (likely(cache->ptr)) {
return ldub_p(cache->ptr + addr);
} else {
return address_space_ldub_cached_slow(cache, addr, attrs, result);
}
}
static inline void address_space_stb_cached(MemoryRegionCache *cache,
hwaddr addr, uint32_t val, MemTxAttrs attrs, MemTxResult *result)
{
assert(addr < cache->len);
if (likely(cache->ptr)) {
stb_p(cache->ptr + addr, val);
} else {
address_space_stb_cached_slow(cache, addr, val, attrs, result);
}
}
#define ENDIANNESS _le
#include "exec/memory_ldst_cached.inc.h"
#define ENDIANNESS _be
#include "exec/memory_ldst_cached.inc.h"
#define SUFFIX _cached
#define ARG1 cache
#define ARG1_DECL MemoryRegionCache *cache
@ -1860,6 +1893,13 @@ MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr,
MemoryRegion *mr);
void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr);
/* Internal functions, part of the implementation of address_space_read_cached
* and address_space_write_cached. */
void address_space_read_cached_slow(MemoryRegionCache *cache,
hwaddr addr, void *buf, int len);
void address_space_write_cached_slow(MemoryRegionCache *cache,
hwaddr addr, const void *buf, int len);
static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
{
if (is_write) {
@ -1928,7 +1968,11 @@ address_space_read_cached(MemoryRegionCache *cache, hwaddr addr,
void *buf, int len)
{
assert(addr < cache->len && len <= cache->len - addr);
address_space_read(cache->as, cache->xlat + addr, MEMTXATTRS_UNSPECIFIED, buf, len);
if (likely(cache->ptr)) {
memcpy(buf, cache->ptr + addr, len);
} else {
address_space_read_cached_slow(cache, addr, buf, len);
}
}
/**
@ -1944,7 +1988,11 @@ address_space_write_cached(MemoryRegionCache *cache, hwaddr addr,
void *buf, int len)
{
assert(addr < cache->len && len <= cache->len - addr);
address_space_write(cache->as, cache->xlat + addr, MEMTXATTRS_UNSPECIFIED, buf, len);
if (likely(cache->ptr)) {
memcpy(cache->ptr + addr, buf, len);
} else {
address_space_write_cached_slow(cache, addr, buf, len);
}
}
#endif