aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--accel/tcg/translate-all.c63
-rw-r--r--bsd-user/main.c1
-rw-r--r--cpus.c12
-rw-r--r--linux-user/main.c1
-rw-r--r--tcg/tcg.c222
-rw-r--r--tcg/tcg.h6
6 files changed, 260 insertions, 45 deletions
diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
index 9061c0508c..f99bfd9309 100644
--- a/accel/tcg/translate-all.c
+++ b/accel/tcg/translate-all.c
@@ -606,15 +606,13 @@ static inline void *alloc_code_gen_buffer(void)
{
void *buf = static_code_gen_buffer;
void *end = static_code_gen_buffer + sizeof(static_code_gen_buffer);
- size_t full_size, size;
+ size_t size;
/* page-align the beginning and end of the buffer */
buf = QEMU_ALIGN_PTR_UP(buf, qemu_real_host_page_size);
end = QEMU_ALIGN_PTR_DOWN(end, qemu_real_host_page_size);
- /* Reserve a guard page. */
- full_size = end - buf;
- size = full_size - qemu_real_host_page_size;
+ size = end - buf;
/* Honor a command-line option limiting the size of the buffer. */
if (size > tcg_ctx->code_gen_buffer_size) {
@@ -633,9 +631,6 @@ static inline void *alloc_code_gen_buffer(void)
if (qemu_mprotect_rwx(buf, size)) {
abort();
}
- if (qemu_mprotect_none(buf + size, qemu_real_host_page_size)) {
- abort();
- }
qemu_madvise(buf, size, QEMU_MADV_HUGEPAGE);
return buf;
@@ -644,22 +639,16 @@ static inline void *alloc_code_gen_buffer(void)
static inline void *alloc_code_gen_buffer(void)
{
size_t size = tcg_ctx->code_gen_buffer_size;
- void *buf1, *buf2;
-
- /* Perform the allocation in two steps, so that the guard page
- is reserved but uncommitted. */
- buf1 = VirtualAlloc(NULL, size + qemu_real_host_page_size,
- MEM_RESERVE, PAGE_NOACCESS);
- if (buf1 != NULL) {
- buf2 = VirtualAlloc(buf1, size, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
- assert(buf1 == buf2);
- }
+ void *buf;
- return buf1;
+ buf = VirtualAlloc(NULL, size, MEM_RESERVE | MEM_COMMIT,
+ PAGE_EXECUTE_READWRITE);
+ return buf;
}
#else
static inline void *alloc_code_gen_buffer(void)
{
+ int prot = PROT_WRITE | PROT_READ | PROT_EXEC;
int flags = MAP_PRIVATE | MAP_ANONYMOUS;
uintptr_t start = 0;
size_t size = tcg_ctx->code_gen_buffer_size;
@@ -693,8 +682,7 @@ static inline void *alloc_code_gen_buffer(void)
# endif
# endif
- buf = mmap((void *)start, size + qemu_real_host_page_size,
- PROT_NONE, flags, -1, 0);
+ buf = mmap((void *)start, size, prot, flags, -1, 0);
if (buf == MAP_FAILED) {
return NULL;
}
@@ -704,24 +692,23 @@ static inline void *alloc_code_gen_buffer(void)
/* Try again, with the original still mapped, to avoid re-acquiring
that 256mb crossing. This time don't specify an address. */
size_t size2;
- void *buf2 = mmap(NULL, size + qemu_real_host_page_size,
- PROT_NONE, flags, -1, 0);
+ void *buf2 = mmap(NULL, size, prot, flags, -1, 0);
switch ((int)(buf2 != MAP_FAILED)) {
case 1:
if (!cross_256mb(buf2, size)) {
/* Success! Use the new buffer. */
- munmap(buf, size + qemu_real_host_page_size);
+ munmap(buf, size);
break;
}
/* Failure. Work with what we had. */
- munmap(buf2, size + qemu_real_host_page_size);
+ munmap(buf2, size);
/* fallthru */
default:
/* Split the original buffer. Free the smaller half. */
buf2 = split_cross_256mb(buf, size);
size2 = tcg_ctx->code_gen_buffer_size;
if (buf == buf2) {
- munmap(buf + size2 + qemu_real_host_page_size, size - size2);
+ munmap(buf + size2, size - size2);
} else {
munmap(buf, size - size2);
}
@@ -732,10 +719,6 @@ static inline void *alloc_code_gen_buffer(void)
}
#endif
- /* Make the final buffer accessible. The guard page at the end
- will remain inaccessible with PROT_NONE. */
- mprotect(buf, size, PROT_WRITE | PROT_READ | PROT_EXEC);
-
/* Request large pages for the buffer. */
qemu_madvise(buf, size, QEMU_MADV_HUGEPAGE);
@@ -916,13 +899,8 @@ static void do_tb_flush(CPUState *cpu, run_on_cpu_data tb_flush_count)
size_t host_size = 0;
g_tree_foreach(tb_ctx.tb_tree, tb_host_size_iter, &host_size);
- printf("qemu: flush code_size=%td nb_tbs=%zu avg_tb_size=%zu\n",
- tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer, nb_tbs,
- nb_tbs > 0 ? host_size / nb_tbs : 0);
- }
- if ((unsigned long)(tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer)
- > tcg_ctx->code_gen_buffer_size) {
- cpu_abort(cpu, "Internal error: code buffer overflow\n");
+ printf("qemu: flush code_size=%zu nb_tbs=%zu avg_tb_size=%zu\n",
+ tcg_code_size(), nb_tbs, nb_tbs > 0 ? host_size / nb_tbs : 0);
}
CPU_FOREACH(cpu) {
@@ -936,7 +914,7 @@ static void do_tb_flush(CPUState *cpu, run_on_cpu_data tb_flush_count)
qht_reset_size(&tb_ctx.htable, CODE_GEN_HTABLE_SIZE);
page_flush_tb();
- tcg_ctx->code_gen_ptr = tcg_ctx->code_gen_buffer;
+ tcg_region_reset_all();
/* XXX: flush processor icache at this point if cache flush is
expensive */
atomic_mb_set(&tb_ctx.tb_flush_count, tb_ctx.tb_flush_count + 1);
@@ -1274,9 +1252,9 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
phys_pc = get_page_addr_code(env, pc);
+ buffer_overflow:
tb = tb_alloc(pc);
if (unlikely(!tb)) {
- buffer_overflow:
/* flush must be done */
tb_flush(cpu);
mmap_unlock();
@@ -1380,9 +1358,9 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
}
#endif
- tcg_ctx->code_gen_ptr = (void *)
+ atomic_set(&tcg_ctx->code_gen_ptr, (void *)
ROUND_UP((uintptr_t)gen_code_buf + gen_code_size + search_size,
- CODE_GEN_ALIGN);
+ CODE_GEN_ALIGN));
/* init jump list */
assert(((uintptr_t)tb & 3) == 0);
@@ -1908,9 +1886,8 @@ void dump_exec_info(FILE *f, fprintf_function cpu_fprintf)
* otherwise users might think "-tb-size" is not honoured.
* For avg host size we use the precise numbers from tb_tree_stats though.
*/
- cpu_fprintf(f, "gen code size %td/%zd\n",
- tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer,
- tcg_ctx->code_gen_highwater - tcg_ctx->code_gen_buffer);
+ cpu_fprintf(f, "gen code size %zu/%zu\n",
+ tcg_code_size(), tcg_code_capacity());
cpu_fprintf(f, "TB count %zu\n", nb_tbs);
cpu_fprintf(f, "TB avg target size %zu max=%zu bytes\n",
nb_tbs ? tst.target_size / nb_tbs : 0,
diff --git a/bsd-user/main.c b/bsd-user/main.c
index 392c0ed5fb..f1b244b59b 100644
--- a/bsd-user/main.c
+++ b/bsd-user/main.c
@@ -978,6 +978,7 @@ int main(int argc, char **argv)
generating the prologue until now so that the prologue can take
the real value of GUEST_BASE into account. */
tcg_prologue_init(tcg_ctx);
+ tcg_region_init();
/* build Task State */
memset(ts, 0, sizeof(TaskState));
diff --git a/cpus.c b/cpus.c
index c9a624003a..8e06257a74 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1664,6 +1664,18 @@ static void qemu_tcg_init_vcpu(CPUState *cpu)
char thread_name[VCPU_THREAD_NAME_SIZE];
static QemuCond *single_tcg_halt_cond;
static QemuThread *single_tcg_cpu_thread;
+ static int tcg_region_inited;
+
+ /*
+ * Initialize TCG regions--once. Now is a good time, because:
+ * (1) TCG's init context, prologue and target globals have been set up.
+ * (2) qemu_tcg_mttcg_enabled() works now (TCG init code runs before the
+ * -accel flag is processed, so the check doesn't work then).
+ */
+ if (!tcg_region_inited) {
+ tcg_region_inited = 1;
+ tcg_region_init();
+ }
if (qemu_tcg_mttcg_enabled() || !single_tcg_cpu_thread) {
cpu->thread = g_malloc0(sizeof(QemuThread));
diff --git a/linux-user/main.c b/linux-user/main.c
index 8814906409..28353f1a75 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -4477,6 +4477,7 @@ int main(int argc, char **argv, char **envp)
generating the prologue until now so that the prologue can take
the real value of GUEST_BASE into account. */
tcg_prologue_init(tcg_ctx);
+ tcg_region_init();
#if defined(TARGET_I386)
env->cr[0] = CR0_PG_MASK | CR0_WP_MASK | CR0_PE_MASK;
diff --git a/tcg/tcg.c b/tcg/tcg.c
index f1bbfe37ff..3de5f7cf97 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -121,6 +121,30 @@ static bool tcg_out_ldst_finalize(TCGContext *s);
static TCGContext **tcg_ctxs;
static unsigned int n_tcg_ctxs;
+/*
+ * We divide code_gen_buffer into equally-sized "regions" that TCG threads
+ * dynamically allocate from as demand dictates. Given appropriate region
+ * sizing, this minimizes flushes even when some TCG threads generate a lot
+ * more code than others.
+ */
+struct tcg_region_state {
+ QemuMutex lock;
+
+ /* fields set at init time */
+ void *start;
+ void *start_aligned;
+ void *end;
+ size_t n;
+ size_t size; /* size of one region */
+ size_t stride; /* .size + guard size */
+
+ /* fields protected by the lock */
+ size_t current; /* current region index */
+ size_t agg_size_full; /* aggregate size of full regions */
+};
+
+static struct tcg_region_state region;
+
static TCGRegSet tcg_target_available_regs[2];
static TCGRegSet tcg_target_call_clobber_regs;
@@ -258,6 +282,196 @@ TCGLabel *gen_new_label(void)
#include "tcg-target.inc.c"
+static void tcg_region_bounds(size_t curr_region, void **pstart, void **pend)
+{
+ void *start, *end;
+
+ start = region.start_aligned + curr_region * region.stride;
+ end = start + region.size;
+
+ if (curr_region == 0) {
+ start = region.start;
+ }
+ if (curr_region == region.n - 1) {
+ end = region.end;
+ }
+
+ *pstart = start;
+ *pend = end;
+}
+
+static void tcg_region_assign(TCGContext *s, size_t curr_region)
+{
+ void *start, *end;
+
+ tcg_region_bounds(curr_region, &start, &end);
+
+ s->code_gen_buffer = start;
+ s->code_gen_ptr = start;
+ s->code_gen_buffer_size = end - start;
+ s->code_gen_highwater = end - TCG_HIGHWATER;
+}
+
+static bool tcg_region_alloc__locked(TCGContext *s)
+{
+ if (region.current == region.n) {
+ return true;
+ }
+ tcg_region_assign(s, region.current);
+ region.current++;
+ return false;
+}
+
+/*
+ * Request a new region once the one in use has filled up.
+ * Returns true on error.
+ */
+static bool tcg_region_alloc(TCGContext *s)
+{
+ bool err;
+ /* read the region size now; alloc__locked will overwrite it on success */
+ size_t size_full = s->code_gen_buffer_size;
+
+ qemu_mutex_lock(&region.lock);
+ err = tcg_region_alloc__locked(s);
+ if (!err) {
+ region.agg_size_full += size_full - TCG_HIGHWATER;
+ }
+ qemu_mutex_unlock(&region.lock);
+ return err;
+}
+
+/*
+ * Perform a context's first region allocation.
+ * This function does _not_ increment region.agg_size_full.
+ */
+static inline bool tcg_region_initial_alloc__locked(TCGContext *s)
+{
+ return tcg_region_alloc__locked(s);
+}
+
+/* Call from a safe-work context */
+void tcg_region_reset_all(void)
+{
+ unsigned int i;
+
+ qemu_mutex_lock(&region.lock);
+ region.current = 0;
+ region.agg_size_full = 0;
+
+ for (i = 0; i < n_tcg_ctxs; i++) {
+ bool err = tcg_region_initial_alloc__locked(tcg_ctxs[i]);
+
+ g_assert(!err);
+ }
+ qemu_mutex_unlock(&region.lock);
+}
+
+/*
+ * Initializes region partitioning.
+ *
+ * Called at init time from the parent thread (i.e. the one calling
+ * tcg_context_init), after the target's TCG globals have been set.
+ */
+void tcg_region_init(void)
+{
+ void *buf = tcg_init_ctx.code_gen_buffer;
+ void *aligned;
+ size_t size = tcg_init_ctx.code_gen_buffer_size;
+ size_t page_size = qemu_real_host_page_size;
+ size_t region_size;
+ size_t n_regions;
+ size_t i;
+
+ /* We do not yet support multiple TCG contexts, so use one region for now */
+ n_regions = 1;
+
+ /* The first region will be 'aligned - buf' bytes larger than the others */
+ aligned = QEMU_ALIGN_PTR_UP(buf, page_size);
+ g_assert(aligned < tcg_init_ctx.code_gen_buffer + size);
+ /*
+ * Make region_size a multiple of page_size, using aligned as the start.
+ * As a result of this we might end up with a few extra pages at the end of
+ * the buffer; we will assign those to the last region.
+ */
+ region_size = (size - (aligned - buf)) / n_regions;
+ region_size = QEMU_ALIGN_DOWN(region_size, page_size);
+
+ /* A region must have at least 2 pages; one code, one guard */
+ g_assert(region_size >= 2 * page_size);
+
+ /* init the region struct */
+ qemu_mutex_init(&region.lock);
+ region.n = n_regions;
+ region.size = region_size - page_size;
+ region.stride = region_size;
+ region.start = buf;
+ region.start_aligned = aligned;
+ /* page-align the end, since its last page will be a guard page */
+ region.end = QEMU_ALIGN_PTR_DOWN(buf + size, page_size);
+ /* account for that last guard page */
+ region.end -= page_size;
+
+ /* set guard pages */
+ for (i = 0; i < region.n; i++) {
+ void *start, *end;
+ int rc;
+
+ tcg_region_bounds(i, &start, &end);
+ rc = qemu_mprotect_none(end, page_size);
+ g_assert(!rc);
+ }
+
+ /* We do not yet support multiple TCG contexts so allocate the region now */
+ {
+ bool err = tcg_region_initial_alloc__locked(tcg_ctx);
+
+ g_assert(!err);
+ }
+}
+
+/*
+ * Returns the size (in bytes) of all translated code (i.e. from all regions)
+ * currently in the cache.
+ * See also: tcg_code_capacity()
+ * Do not confuse with tcg_current_code_size(); that one applies to a single
+ * TCG context.
+ */
+size_t tcg_code_size(void)
+{
+ unsigned int i;
+ size_t total;
+
+ qemu_mutex_lock(&region.lock);
+ total = region.agg_size_full;
+ for (i = 0; i < n_tcg_ctxs; i++) {
+ const TCGContext *s = tcg_ctxs[i];
+ size_t size;
+
+ size = atomic_read(&s->code_gen_ptr) - s->code_gen_buffer;
+ g_assert(size <= s->code_gen_buffer_size);
+ total += size;
+ }
+ qemu_mutex_unlock(&region.lock);
+ return total;
+}
+
+/*
+ * Returns the code capacity (in bytes) of the entire cache, i.e. including all
+ * regions.
+ * See also: tcg_code_size()
+ */
+size_t tcg_code_capacity(void)
+{
+ size_t guard_size, capacity;
+
+ /* no need for synchronization; these variables are set at init time */
+ guard_size = region.stride - region.size;
+ capacity = region.end + guard_size - region.start;
+ capacity -= region.n * (guard_size + TCG_HIGHWATER);
+ return capacity;
+}
+
/* pool based memory allocation */
void *tcg_malloc_internal(TCGContext *s, int size)
{
@@ -401,13 +615,17 @@ TranslationBlock *tcg_tb_alloc(TCGContext *s)
TranslationBlock *tb;
void *next;
+ retry:
tb = (void *)ROUND_UP((uintptr_t)s->code_gen_ptr, align);
next = (void *)ROUND_UP((uintptr_t)(tb + 1), align);
if (unlikely(next > s->code_gen_highwater)) {
- return NULL;
+ if (tcg_region_alloc(s)) {
+ return NULL;
+ }
+ goto retry;
}
- s->code_gen_ptr = next;
+ atomic_set(&s->code_gen_ptr, next);
s->data_gen_ptr = NULL;
return tb;
}
diff --git a/tcg/tcg.h b/tcg/tcg.h
index 4b9958a179..9f95648282 100644
--- a/tcg/tcg.h
+++ b/tcg/tcg.h
@@ -802,6 +802,12 @@ void *tcg_malloc_internal(TCGContext *s, int size);
void tcg_pool_reset(TCGContext *s);
TranslationBlock *tcg_tb_alloc(TCGContext *s);
+void tcg_region_init(void);
+void tcg_region_reset_all(void);
+
+size_t tcg_code_size(void);
+size_t tcg_code_capacity(void);
+
/* Called with tb_lock held. */
static inline void *tcg_malloc(int size)
{