From: Luc Michel <luc.michel@amd.com>
To: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Cc: qemu-devel@nongnu.org,
"Philippe Mathieu-Daudé" <philmd@linaro.org>,
"Richard Henderson" <richard.henderson@linaro.org>,
"Marcel Apfelbaum" <marcel.apfelbaum@gmail.com>,
"Yanan Wang" <wangyanan55@huawei.com>,
"Paolo Bonzini" <pbonzini@redhat.com>,
"Mahmoud Mandour" <ma.mandourr@gmail.com>,
"Alexandre Iooss" <erdnaxe@crans.org>,
"Alex Bennée" <alex.bennee@linaro.org>,
"Eduardo Habkost" <eduardo@habkost.net>
Subject: Re: [PATCH v5 09/12] contrib/plugins/hotblocks: migrate to new per_vcpu API
Date: Tue, 27 Feb 2024 15:26:34 +0100 [thread overview]
Message-ID: <Zd3xGrZetKoAbIn1@luc-work-vm> (raw)
In-Reply-To: <e0688c8b-fac2-4b4f-8cda-a0c4f5ee2bf6@linaro.org>
On 15:09 Tue 27 Feb , Pierrick Bouvier wrote:
> On 2/27/24 2:54 PM, Luc Michel wrote:
> > Hi Pierrick,
> >
> > On 13:14 Mon 26 Feb , Pierrick Bouvier wrote:
> > > Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
> > > ---
> > > contrib/plugins/hotblocks.c | 50 ++++++++++++++++++++++---------------
> > > 1 file changed, 30 insertions(+), 20 deletions(-)
> > >
> > > diff --git a/contrib/plugins/hotblocks.c b/contrib/plugins/hotblocks.c
> > > index 4de1b134944..02bc5078bdd 100644
> > > --- a/contrib/plugins/hotblocks.c
> > > +++ b/contrib/plugins/hotblocks.c
> > > @@ -34,8 +34,8 @@ static guint64 limit = 20;
> > > */
> > > typedef struct {
> > > uint64_t start_addr;
> > > - uint64_t exec_count;
> > > - int trans_count;
> > > + struct qemu_plugin_scoreboard *exec_count;
> > > + int trans_count;
> > > unsigned long insns;
> > > } ExecCount;
> > >
> > > @@ -43,7 +43,17 @@ static gint cmp_exec_count(gconstpointer a, gconstpointer b)
> > > {
> > > ExecCount *ea = (ExecCount *) a;
> > > ExecCount *eb = (ExecCount *) b;
> > > - return ea->exec_count > eb->exec_count ? -1 : 1;
> > > + uint64_t count_a =
> > > + qemu_plugin_u64_sum(qemu_plugin_scoreboard_u64(ea->exec_count));
> > > + uint64_t count_b =
> > > + qemu_plugin_u64_sum(qemu_plugin_scoreboard_u64(eb->exec_count));
> > > + return count_a > count_b ? -1 : 1;
> > > +}
> > > +
> > > +static void exec_count_free(gpointer key, gpointer value, gpointer user_data)
> > > +{
> > > + ExecCount *cnt = value;
> > > + qemu_plugin_scoreboard_free(cnt->exec_count);
> > > }
> > >
> > > static void plugin_exit(qemu_plugin_id_t id, void *p)
> > > @@ -52,7 +62,6 @@ static void plugin_exit(qemu_plugin_id_t id, void *p)
> > > GList *counts, *it;
> > > int i;
> > >
> > > - g_mutex_lock(&lock);
> >
> > I encountered cases before where the vCPUs continue executing while
> > plugin_exit is called. This can happen e.g., when QEMU calls exit(3)
> > from one CPU thread. Others will continue to run at the same time the
> > atexit callbacks are called.
> >
> > This also means that you can't really free the resources as you do at
> > the end of plugin_exit.
> >
>
> Interesting...
>
> The current documentation [1] mentions it's the right place to free
> resources, and from what I saw, existing plugins assume this too (see
> contrib/plugins/cache.c for instance).
>
> There is probably a bug related to the case you mention, and I'll try to
> reproduce this, and see if we can have a proper fix.
>
> I'm not sure that removing cleanup code from existing plugins is the
> right thing to do at the moment, even though there is an existing issue
> with some programs.
Yes absolutely. The problem is on the QEMU side. FWIW I tried to fix one
of those exit cases (semihosted exit syscall) some time ago:
https://lore.kernel.org/qemu-devel/20220621125916.25257-1-lmichel@kalray.eu/
IIRC Peter was not fundamentally against it. But then I was off for
several months due to medical reasons.
I can take some time to rebase it and address Peter's comments if there
is interest to it.
Luc
>
> [1]
> https://www.qemu.org/docs/master/devel/tcg-plugins.html#c.qemu_plugin_register_atexit_cb
>
> > > g_string_append_printf(report, "%d entries in the hash table\n",
> > > g_hash_table_size(hotblocks));
> > > counts = g_hash_table_get_values(hotblocks);
> > > @@ -63,16 +72,21 @@ static void plugin_exit(qemu_plugin_id_t id, void *p)
> > >
> > > for (i = 0; i < limit && it->next; i++, it = it->next) {
> > > ExecCount *rec = (ExecCount *) it->data;
> > > - g_string_append_printf(report, "0x%016"PRIx64", %d, %ld, %"PRId64"\n",
> > > - rec->start_addr, rec->trans_count,
> > > - rec->insns, rec->exec_count);
> > > + g_string_append_printf(
> > > + report, "0x%016"PRIx64", %d, %ld, %"PRId64"\n",
> > > + rec->start_addr, rec->trans_count,
> > > + rec->insns,
> > > + qemu_plugin_u64_sum(
> > > + qemu_plugin_scoreboard_u64(rec->exec_count)));
> > > }
> > >
> > > g_list_free(it);
> > > }
> > > - g_mutex_unlock(&lock);
> > >
> > > qemu_plugin_outs(report->str);
> > > +
> > > + g_hash_table_foreach(hotblocks, exec_count_free, NULL);
> > > + g_hash_table_destroy(hotblocks);
> > > }
> > >
> > > static void plugin_init(void)
> > > @@ -82,15 +96,9 @@ static void plugin_init(void)
> > >
> > > static void vcpu_tb_exec(unsigned int cpu_index, void *udata)
> > > {
> > > - ExecCount *cnt;
> > > - uint64_t hash = (uint64_t) udata;
> > > -
> > > - g_mutex_lock(&lock);
> > > - cnt = (ExecCount *) g_hash_table_lookup(hotblocks, (gconstpointer) hash);
> > > - /* should always succeed */
> > > - g_assert(cnt);
> > > - cnt->exec_count++;
> > > - g_mutex_unlock(&lock);
> > > + ExecCount *cnt = (ExecCount *)udata;
> > > + qemu_plugin_u64_add(qemu_plugin_scoreboard_u64(cnt->exec_count),
> > > + cpu_index, 1);
> > > }
> > >
> > > /*
> > > @@ -114,18 +122,20 @@ static void vcpu_tb_trans(qemu_plugin_id_t id, struct qemu_plugin_tb *tb)
> > > cnt->start_addr = pc;
> > > cnt->trans_count = 1;
> > > cnt->insns = insns;
> > > + cnt->exec_count = qemu_plugin_scoreboard_new(sizeof(uint64_t));
> > > g_hash_table_insert(hotblocks, (gpointer) hash, (gpointer) cnt);
> > > }
> > >
> > > g_mutex_unlock(&lock);
> > >
> > > if (do_inline) {
> > > - qemu_plugin_register_vcpu_tb_exec_inline(tb, QEMU_PLUGIN_INLINE_ADD_U64,
> > > - &cnt->exec_count, 1);
> > > + qemu_plugin_register_vcpu_tb_exec_inline_per_vcpu(
> > > + tb, QEMU_PLUGIN_INLINE_ADD_U64,
> > > + qemu_plugin_scoreboard_u64(cnt->exec_count), 1);
> > > } else {
> > > qemu_plugin_register_vcpu_tb_exec_cb(tb, vcpu_tb_exec,
> > > QEMU_PLUGIN_CB_NO_REGS,
> > > - (void *)hash);
> > > + (void *)cnt);
> > > }
> > > }
> > >
> > > --
> > > 2.43.0
> > >
> > >
> >
--
next prev parent reply other threads:[~2024-02-27 14:27 UTC|newest]
Thread overview: 38+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-02-26 9:14 [PATCH v5 00/12] TCG Plugin inline operation enhancement Pierrick Bouvier
2024-02-26 9:14 ` [PATCH v5 01/12] plugins: scoreboard API Pierrick Bouvier
2024-02-26 9:14 ` [PATCH v5 02/12] plugins: define qemu_plugin_u64 Pierrick Bouvier
2024-02-26 9:14 ` [PATCH v5 03/12] plugins: implement inline operation relative to cpu_index Pierrick Bouvier
2024-02-26 9:14 ` [PATCH v5 04/12] plugins: add inline operation per vcpu Pierrick Bouvier
2024-02-26 9:14 ` [PATCH v5 05/12] tests/plugin: add test plugin for inline operations Pierrick Bouvier
2024-02-26 9:14 ` [PATCH v5 06/12] tests/plugin/mem: migrate to new per_vcpu API Pierrick Bouvier
2024-02-27 9:35 ` Luc Michel
2024-02-27 10:56 ` Pierrick Bouvier
2024-02-27 14:27 ` Luc Michel
2024-02-28 22:08 ` Alex Bennée
2024-02-29 5:19 ` Pierrick Bouvier
2024-02-29 7:08 ` Alex Bennée
2024-02-29 7:12 ` Pierrick Bouvier
2024-02-29 13:46 ` Alex Bennée
2024-03-01 9:41 ` Pierrick Bouvier
2024-02-26 9:14 ` [PATCH v5 07/12] tests/plugin/insn: " Pierrick Bouvier
2024-02-27 9:50 ` Luc Michel
2024-02-26 9:14 ` [PATCH v5 08/12] tests/plugin/bb: " Pierrick Bouvier
2024-02-27 10:02 ` Luc Michel
2024-02-29 14:21 ` Alex Bennée
2024-03-01 9:58 ` Pierrick Bouvier
2024-03-01 10:26 ` Alex Bennée
2024-03-01 10:47 ` Pierrick Bouvier
2024-02-26 9:14 ` [PATCH v5 09/12] contrib/plugins/hotblocks: " Pierrick Bouvier
2024-02-27 10:54 ` Luc Michel
2024-02-27 11:09 ` Pierrick Bouvier
2024-02-27 14:26 ` Luc Michel [this message]
2024-02-28 8:04 ` Pierrick Bouvier
2024-02-29 14:33 ` Alex Bennée
2024-03-01 13:21 ` Luc Michel
2024-02-26 9:14 ` [PATCH v5 10/12] contrib/plugins/howvec: " Pierrick Bouvier
2024-02-29 14:35 ` Alex Bennée
2024-02-29 14:40 ` Alex Bennée
2024-02-26 9:14 ` [PATCH v5 11/12] plugins: remove non per_vcpu inline operation from API Pierrick Bouvier
2024-02-29 14:41 ` Alex Bennée
2024-02-26 9:14 ` [PATCH v5 12/12] plugins: cleanup codepath for previous inline operation Pierrick Bouvier
2024-02-29 14:42 ` 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=Zd3xGrZetKoAbIn1@luc-work-vm \
--to=luc.michel@amd.com \
--cc=alex.bennee@linaro.org \
--cc=eduardo@habkost.net \
--cc=erdnaxe@crans.org \
--cc=ma.mandourr@gmail.com \
--cc=marcel.apfelbaum@gmail.com \
--cc=pbonzini@redhat.com \
--cc=philmd@linaro.org \
--cc=pierrick.bouvier@linaro.org \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
--cc=wangyanan55@huawei.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).