From: "Alex Bennée" <alex.bennee@linaro.org>
To: mttcg@listserver.greensocs.com, qemu-devel@nongnu.org,
fred.konrad@greensocs.com, a.rigo@virtualopensystems.com,
serge.fdrv@gmail.com, cota@braap.org, bobby.prani@gmail.com,
rth@twiddle.net
Cc: mark.burton@greensocs.com, pbonzini@redhat.com,
jan.kiszka@siemens.com, peter.maydell@linaro.org,
claudio.fontana@huawei.com,
"Alex Bennée" <alex.bennee@linaro.org>,
"Peter Crosthwaite" <crosthwaite.peter@gmail.com>
Subject: [Qemu-devel] [PATCH v2 6/6] tcg: cpu-exec: roll-up tb_find_fast/slow
Date: Tue, 5 Jul 2016 17:18:16 +0100 [thread overview]
Message-ID: <1467735496-16256-7-git-send-email-alex.bennee@linaro.org> (raw)
In-Reply-To: <1467735496-16256-1-git-send-email-alex.bennee@linaro.org>
In an effort not to re-take locks I've re-factored the code so
everything is done inside the new tb_find function. The fast path will
only take a lock if the previous TB needs patching. The generation path
will do the patching inside the generation critical section.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
---
cpu-exec.c | 117 ++++++++++++++++++++++++++++++++++---------------------------
1 file changed, 66 insertions(+), 51 deletions(-)
diff --git a/cpu-exec.c b/cpu-exec.c
index 59c2ec5..529eae1 100644
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -281,7 +281,8 @@ static TranslationBlock *tb_find_physical(CPUState *cpu,
/*
* Patch the last TB with a jump to the current TB.
*
- * Modification of the TB has to be protected with tb_lock.
+ * Modification of the TB has to be protected with tb_lock which can
+ * either be already held or taken here.
*/
static inline void maybe_patch_last_tb(CPUState *cpu,
TranslationBlock *tb,
@@ -306,72 +307,86 @@ static inline void maybe_patch_last_tb(CPUState *cpu,
#endif
/* See if we can patch the calling TB. */
if (*last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
- tb_lock();
+ bool locked = tb_lock_recursive();
tb_add_jump(*last_tb, tb_exit, tb);
- tb_unlock();
- }
-}
-
-static TranslationBlock *tb_find_slow(CPUState *cpu,
- target_ulong pc,
- target_ulong cs_base,
- uint32_t flags)
-{
- TranslationBlock *tb;
-
- /* Ensure that we won't find a TB in the shared hash table
- * if it is being invalidated by some other thread.
- * Otherwise we'd put it back to CPU's local cache.
- * Pairs with smp_wmb() in tb_phys_invalidate(). */
- smp_rmb();
- tb = tb_find_physical(cpu, pc, cs_base, flags);
- if (!tb) {
-
- /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
- * taken outside tb_lock. As system emulation is currently
- * single threaded the locks are NOPs.
- */
- mmap_lock();
- tb_lock();
-
- /* There's a chance that our desired tb has been translated while
- * taking the locks so we check again inside the lock.
- */
- tb = tb_find_physical(cpu, pc, cs_base, flags);
- if (!tb) {
- /* if no translated code available, then translate it now */
- tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
+ if (locked) {
+ tb_unlock();
}
-
- tb_unlock();
- mmap_unlock();
}
-
- /* We add the TB in the virtual pc hash table for the fast lookup */
- atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb);
- return tb;
}
-static inline TranslationBlock *tb_find_fast(CPUState *cpu,
- TranslationBlock **last_tb,
- int tb_exit)
+/*
+ * tb_find - find next TB, possibly generating it
+ *
+ * There is a multi-level lookup for finding the next TB which avoids
+ * locks unless generation is required.
+ *
+ * 1. Lookup via the per-vcpu tb_jmp_cache
+ * 2. Lookup via tb_find_physical (using QHT)
+ *
+ * If both of those fail then we need to grab the mmap_lock and
+ * tb_lock and do code generation.
+ *
+ * As the jump patching of code also needs to be protected by locks we
+ * have multiple paths into maybe_patch_last_tb taking advantage of
+ * the fact we may already have locks held for code generation.
+ */
+static TranslationBlock *tb_find(CPUState *cpu,
+ TranslationBlock **last_tb,
+ int tb_exit)
{
CPUArchState *env = (CPUArchState *)cpu->env_ptr;
TranslationBlock *tb;
target_ulong cs_base, pc;
+ unsigned int h;
uint32_t flags;
/* we record a subset of the CPU state. It will
always be the same before a given translated block
is executed. */
cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
- tb = atomic_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)]);
+ h = tb_jmp_cache_hash_func(pc);
+ tb = atomic_read(&cpu->tb_jmp_cache[h]);
+
if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
tb->flags != flags)) {
- tb = tb_find_slow(cpu, pc, cs_base, flags);
- }
- maybe_patch_last_tb(cpu, tb, last_tb, tb_exit);
+ /* Ensure that we won't find a TB in the shared hash table
+ * if it is being invalidated by some other thread.
+ * Otherwise we'd put it back to CPU's local cache.
+ * Pairs with smp_wmb() in tb_phys_invalidate(). */
+ smp_rmb();
+ tb = tb_find_physical(cpu, pc, cs_base, flags);
+
+ if (!tb) {
+ /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
+ * taken outside tb_lock. As system emulation is currently
+ * single threaded the locks are NOPs.
+ */
+ mmap_lock();
+ tb_lock();
+
+ /* There's a chance that our desired tb has been translated while
+ * taking the locks so we check again inside the lock.
+ */
+ tb = tb_find_physical(cpu, pc, cs_base, flags);
+ if (!tb) {
+ /* if no translated code available, then translate it now */
+ tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
+ }
+ maybe_patch_last_tb(cpu, tb, last_tb, tb_exit);
+
+ tb_unlock();
+ mmap_unlock();
+ } else {
+ maybe_patch_last_tb(cpu, tb, last_tb, tb_exit);
+ }
+
+ /* We update the TB in the virtual pc hash table for the fast lookup */
+ atomic_set(&cpu->tb_jmp_cache[h], tb);
+ } else {
+ maybe_patch_last_tb(cpu, tb, last_tb, tb_exit);
+ }
return tb;
}
@@ -452,7 +467,7 @@ static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
&& cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
/* try to cause an exception pending in the log */
TranslationBlock *last_tb = NULL; /* Avoid chaining TBs */
- cpu_exec_nocache(cpu, 1, tb_find_fast(cpu, &last_tb, 0), true);
+ cpu_exec_nocache(cpu, 1, tb_find(cpu, &last_tb, 0), true);
*ret = -1;
return true;
#endif
@@ -636,7 +651,7 @@ int cpu_exec(CPUState *cpu)
cpu->tb_flushed = false; /* reset before first TB lookup */
for(;;) {
cpu_handle_interrupt(cpu, &last_tb);
- tb = tb_find_fast(cpu, &last_tb, tb_exit);
+ tb = tb_find(cpu, &last_tb, tb_exit);
cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc);
/* Try to align the host and virtual clocks
if the guest is in advance */
--
2.7.4
next prev parent reply other threads:[~2016-07-05 16:18 UTC|newest]
Thread overview: 41+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-07-05 16:18 [Qemu-devel] [PATCH v2 0/6] Reduce lock contention on TCG hot-path Alex Bennée
2016-07-05 16:18 ` [Qemu-devel] [PATCH v2 1/6] tcg: Ensure safe tb_jmp_cache lookup out of 'tb_lock' Alex Bennée
2016-07-07 13:52 ` Sergey Fedorov
2016-07-08 14:51 ` Sergey Fedorov
2016-07-05 16:18 ` [Qemu-devel] [PATCH v2 2/6] tcg: set up tb->page_addr before insertion Alex Bennée
2016-07-07 14:08 ` Sergey Fedorov
2016-07-08 9:40 ` Sergey Fedorov
2016-07-05 16:18 ` [Qemu-devel] [PATCH v2 3/6] tcg: cpu-exec: remove tb_lock from the hot-path Alex Bennée
2016-07-07 14:18 ` Sergey Fedorov
2016-07-08 15:50 ` Sergey Fedorov
2016-07-08 17:34 ` Sergey Fedorov
2016-07-08 18:03 ` Alex Bennée
2016-07-08 18:20 ` Sergey Fedorov
2016-07-08 20:09 ` Sergey Fedorov
2016-07-05 16:18 ` [Qemu-devel] [PATCH v2 4/6] tcg: cpu-exec: factor out TB patching code Alex Bennée
2016-07-05 16:18 ` [Qemu-devel] [PATCH v2 5/6] tcg: introduce tb_lock_recursive() Alex Bennée
2016-07-05 16:18 ` Alex Bennée [this message]
2016-07-07 16:44 ` [Qemu-devel] [PATCH v2 6/6] tcg: cpu-exec: roll-up tb_find_fast/slow Sergey Fedorov
2016-07-07 16:44 ` [Qemu-devel] [PATCH 1/3] tcg: Introduce mmap_lock_reset() Sergey Fedorov
2016-07-07 16:44 ` [Qemu-devel] [PATCH 2/3] tcg: Introduce tb_lock_locked() Sergey Fedorov
2016-07-07 16:44 ` [Qemu-devel] [PATCH 3/3] tcg: Avoid bouncing tb_lock between tb_gen_code() and tb_add_jump() Sergey Fedorov
2016-07-07 19:36 ` Alex Bennée
2016-07-07 19:46 ` Sergey Fedorov
2016-07-07 20:36 ` Sergey Fedorov
2016-07-07 21:40 ` Alex Bennée
2016-07-08 8:40 ` Paolo Bonzini
2016-07-08 10:25 ` Sergey Fedorov
2016-07-08 11:02 ` Paolo Bonzini
2016-07-08 12:32 ` Sergey Fedorov
2016-07-08 14:07 ` Paolo Bonzini
2016-07-08 19:55 ` Sergey Fedorov
2016-07-08 20:18 ` Paolo Bonzini
2016-07-08 20:24 ` Sergey Fedorov
2016-07-08 20:52 ` Paolo Bonzini
2016-07-11 13:06 ` Sergey Fedorov
2016-07-11 14:03 ` Paolo Bonzini
2016-07-11 14:27 ` Sergey Fedorov
2016-07-07 16:04 ` [Qemu-devel] [PATCH v2 0/6] Reduce lock contention on TCG hot-path Emilio G. Cota
2016-07-07 16:13 ` Paolo Bonzini
2016-07-07 19:33 ` Alex Bennée
2016-07-07 19:38 ` Alex Bennée
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1467735496-16256-7-git-send-email-alex.bennee@linaro.org \
--to=alex.bennee@linaro.org \
--cc=a.rigo@virtualopensystems.com \
--cc=bobby.prani@gmail.com \
--cc=claudio.fontana@huawei.com \
--cc=cota@braap.org \
--cc=crosthwaite.peter@gmail.com \
--cc=fred.konrad@greensocs.com \
--cc=jan.kiszka@siemens.com \
--cc=mark.burton@greensocs.com \
--cc=mttcg@listserver.greensocs.com \
--cc=pbonzini@redhat.com \
--cc=peter.maydell@linaro.org \
--cc=qemu-devel@nongnu.org \
--cc=rth@twiddle.net \
--cc=serge.fdrv@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).