From: "Alex Bennée" <alex.bennee@linaro.org>
To: qemu-devel@nongnu.org
Cc: peter.maydell@linaro.org, "Emilio Cota" <cota@braap.org>,
"Alex Bennée" <alex.bennee@linaro.org>,
"Richard Henderson" <richard.henderson@linaro.org>,
"Paolo Bonzini" <pbonzini@redhat.com>
Subject: [PULL 31/35] plugins: fix optimization in plugin_gen_disable_mem_helpers
Date: Thu, 26 Jan 2023 11:22:46 +0000 [thread overview]
Message-ID: <20230126112250.2584701-32-alex.bennee@linaro.org> (raw)
In-Reply-To: <20230126112250.2584701-1-alex.bennee@linaro.org>
From: Emilio Cota <cota@braap.org>
We were mistakenly checking tcg_ctx->plugin_insn as a canary to know
whether the TB had emitted helpers that might have accessed memory.
The problem is that tcg_ctx->plugin_insn gets updated on every
instruction in the TB, which results in us wrongly performing the
optimization (i.e. not clearing cpu->plugin_mem_cbs) way too often,
since it's not rare that the last instruction in the TB doesn't
use helpers.
Fix it by tracking a per-TB canary.
While at it, expand documentation.
Related: #1381
Signed-off-by: Emilio Cota <cota@braap.org>
Message-Id: <20230108164731.61469-2-cota@braap.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20230124180127.1881110-32-alex.bennee@linaro.org>
diff --git a/include/qemu/plugin.h b/include/qemu/plugin.h
index a772e14193..e0ebedef84 100644
--- a/include/qemu/plugin.h
+++ b/include/qemu/plugin.h
@@ -118,7 +118,10 @@ struct qemu_plugin_insn {
void *haddr;
GArray *cbs[PLUGIN_N_CB_TYPES][PLUGIN_N_CB_SUBTYPES];
bool calls_helpers;
+
+ /* if set, the instruction calls helpers that might access guest memory */
bool mem_helper;
+
bool mem_only;
};
@@ -158,6 +161,10 @@ struct qemu_plugin_tb {
void *haddr1;
void *haddr2;
bool mem_only;
+
+ /* if set, the TB calls helpers that might access guest memory */
+ bool mem_helper;
+
GArray *cbs[PLUGIN_N_CB_SUBTYPES];
};
diff --git a/accel/tcg/plugin-gen.c b/accel/tcg/plugin-gen.c
index c7d6514840..17a686bd9e 100644
--- a/accel/tcg/plugin-gen.c
+++ b/accel/tcg/plugin-gen.c
@@ -579,7 +579,8 @@ static void inject_mem_helper(TCGOp *begin_op, GArray *arr)
* is possible that the code we generate after the instruction is
* dead, we also add checks before generating tb_exit etc.
*/
-static void inject_mem_enable_helper(struct qemu_plugin_insn *plugin_insn,
+static void inject_mem_enable_helper(struct qemu_plugin_tb *ptb,
+ struct qemu_plugin_insn *plugin_insn,
TCGOp *begin_op)
{
GArray *cbs[2];
@@ -599,6 +600,7 @@ static void inject_mem_enable_helper(struct qemu_plugin_insn *plugin_insn,
rm_ops(begin_op);
return;
}
+ ptb->mem_helper = true;
arr = g_array_sized_new(false, false,
sizeof(struct qemu_plugin_dyn_cb), n_cbs);
@@ -626,15 +628,22 @@ void plugin_gen_disable_mem_helpers(void)
{
TCGv_ptr ptr;
- if (likely(tcg_ctx->plugin_insn == NULL ||
- !tcg_ctx->plugin_insn->mem_helper)) {
+ /*
+ * We could emit the clearing unconditionally and be done. However, this can
+ * be wasteful if for instance plugins don't track memory accesses, or if
+ * most TBs don't use helpers. Instead, emit the clearing iff the TB calls
+ * helpers that might access guest memory.
+ *
+ * Note: we do not reset plugin_tb->mem_helper here; a TB might have several
+ * exit points, and we want to emit the clearing from all of them.
+ */
+ if (!tcg_ctx->plugin_tb->mem_helper) {
return;
}
ptr = tcg_const_ptr(NULL);
tcg_gen_st_ptr(ptr, cpu_env, offsetof(CPUState, plugin_mem_cbs) -
offsetof(ArchCPU, env));
tcg_temp_free_ptr(ptr);
- tcg_ctx->plugin_insn->mem_helper = false;
}
static void plugin_gen_tb_udata(const struct qemu_plugin_tb *ptb,
@@ -682,14 +691,14 @@ static void plugin_gen_mem_inline(const struct qemu_plugin_tb *ptb,
inject_inline_cb(cbs, begin_op, op_rw);
}
-static void plugin_gen_enable_mem_helper(const struct qemu_plugin_tb *ptb,
+static void plugin_gen_enable_mem_helper(struct qemu_plugin_tb *ptb,
TCGOp *begin_op, int insn_idx)
{
struct qemu_plugin_insn *insn = g_ptr_array_index(ptb->insns, insn_idx);
- inject_mem_enable_helper(insn, begin_op);
+ inject_mem_enable_helper(ptb, insn, begin_op);
}
-static void plugin_gen_disable_mem_helper(const struct qemu_plugin_tb *ptb,
+static void plugin_gen_disable_mem_helper(struct qemu_plugin_tb *ptb,
TCGOp *begin_op, int insn_idx)
{
struct qemu_plugin_insn *insn = g_ptr_array_index(ptb->insns, insn_idx);
@@ -750,7 +759,7 @@ static void pr_ops(void)
#endif
}
-static void plugin_gen_inject(const struct qemu_plugin_tb *plugin_tb)
+static void plugin_gen_inject(struct qemu_plugin_tb *plugin_tb)
{
TCGOp *op;
int insn_idx = -1;
@@ -870,6 +879,7 @@ bool plugin_gen_tb_start(CPUState *cpu, const DisasContextBase *db,
ptb->haddr1 = db->host_addr[0];
ptb->haddr2 = NULL;
ptb->mem_only = mem_only;
+ ptb->mem_helper = false;
plugin_gen_empty_callback(PLUGIN_GEN_FROM_TB);
}
--
2.34.1
next prev parent reply other threads:[~2023-01-26 11:28 UTC|newest]
Thread overview: 43+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-26 11:22 [PULL 00/35] Testing, docs, semihosting and plugin updates Alex Bennée
2023-01-26 11:22 ` [PULL 01/35] scripts/ci: update gitlab-runner playbook to use latest runner Alex Bennée
2023-01-26 11:22 ` [PULL 02/35] gitlab: add FF_SCRIPT_SECTIONS for timings Alex Bennée
2023-01-26 11:22 ` [PULL 03/35] gitlab: just use plain --cc=clang for custom runner build Alex Bennée
2023-01-26 11:22 ` [PULL 04/35] tests/unit: drop hacky race avoidance in test-io-channel-command Alex Bennée
2023-02-06 12:27 ` Philippe Mathieu-Daudé
2023-02-06 13:11 ` Alex Bennée
2023-02-06 13:22 ` Philippe Mathieu-Daudé
2023-01-26 11:22 ` [PULL 05/35] build-sys: fix crlf-ending C code Alex Bennée
2023-01-26 11:22 ` [PULL 06/35] .gitlab-ci.d/windows: do not disable opengl Alex Bennée
2023-01-26 11:22 ` [PULL 07/35] meson: replace Perl usage with Python Alex Bennée
2023-01-26 11:22 ` [PULL 08/35] docs: drop texinfo options Alex Bennée
2023-01-26 11:22 ` [PULL 09/35] gitlab: add lsan suppression file to workaround tcmalloc issues Alex Bennée
2023-01-26 11:22 ` [PULL 10/35] Update lcitool and fedora to 37 Alex Bennée
2023-01-26 11:22 ` [PULL 11/35] lcitool: drop perl from QEMU project/dependencies Alex Bennée
2023-01-26 11:22 ` [PULL 12/35] lcitool: drop texinfo " Alex Bennée
2023-01-26 11:22 ` [PULL 13/35] tests/docker: Install flex in debian-tricore-cross Alex Bennée
2023-01-26 11:22 ` [PULL 14/35] tests/docker: drop debian-tricore-cross's partial status Alex Bennée
2023-01-26 11:22 ` [PULL 15/35] tests/tcg: skip the vma-pthread test on CI Alex Bennée
2023-01-26 11:22 ` [PULL 16/35] tests/tcg: Use SIGKILL for timeout Alex Bennée
2023-01-26 11:22 ` [PULL 17/35] gitlab: wrap up test results for custom runners Alex Bennée
2023-01-26 11:22 ` [PULL 18/35] MAINTAINERS: Fix the entry for tests/tcg/nios2 Alex Bennée
2023-01-26 11:22 ` [PULL 19/35] docs: add hotlinks to about preface text Alex Bennée
2023-01-26 11:22 ` [PULL 20/35] docs: add a new section to outline emulation support Alex Bennée
2023-01-26 11:22 ` [PULL 21/35] semihosting: add semihosting section to the docs Alex Bennée
2023-01-26 11:22 ` [PULL 22/35] docs: add an introduction to the system docs Alex Bennée
2023-01-26 11:22 ` [PULL 23/35] semihosting: Write back semihosting data before completion callback Alex Bennée
2023-01-26 11:22 ` [PULL 24/35] semihosting: add O_BINARY flag in host_open for NT compatibility Alex Bennée
2023-01-26 11:22 ` [PULL 25/35] tests/tcg: add memory-sve test for aarch64 Alex Bennée
2023-01-26 11:22 ` [PULL 26/35] cpu: free cpu->tb_jmp_cache with RCU Alex Bennée
2023-01-26 11:22 ` [PULL 27/35] util/qht: add missing atomic_set(hashes[i]) Alex Bennée
2023-01-26 11:22 ` [PULL 28/35] thread: de-const qemu_spin_destroy Alex Bennée
2023-01-26 11:22 ` [PULL 29/35] util/qht: use striped locks under TSAN Alex Bennée
2023-01-26 11:22 ` [PULL 30/35] plugins: make qemu_plugin_user_exit's locking order consistent with fork_start's Alex Bennée
2023-01-26 11:22 ` Alex Bennée [this message]
2023-01-26 11:22 ` [PULL 32/35] translator: always pair plugin_gen_insn_{start, end} calls Alex Bennée
2023-01-26 11:22 ` [PULL 33/35] tcg: exclude non-memory effecting helpers from instrumentation Alex Bennée
2023-01-26 11:22 ` [PULL 34/35] cpu-exec: assert that plugin_mem_cbs is NULL after execution Alex Bennée
2023-01-26 11:22 ` [PULL 35/35] plugins: Iterate on cb_lists in qemu_plugin_user_exit Alex Bennée
2023-02-01 16:15 ` [PULL 00/35] Testing, docs, semihosting and plugin updates Peter Maydell
2023-02-01 18:06 ` Alex Bennée
2023-02-02 10:50 ` Peter Maydell
2023-02-02 17:12 ` 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=20230126112250.2584701-32-alex.bennee@linaro.org \
--to=alex.bennee@linaro.org \
--cc=cota@braap.org \
--cc=pbonzini@redhat.com \
--cc=peter.maydell@linaro.org \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
/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).