From: "Alex Bennée" <alex.bennee@linaro.org>
To: Peter Maydell <peter.maydell@linaro.org>
Cc: Riku Voipio <riku.voipio@iki.fi>,
1836558@bugs.launchpad.net,
QEMU Developers <qemu-devel@nongnu.org>,
Laurent Vivier <laurent@vivier.eu>
Subject: Re: [Qemu-devel] [RFC PATCH for 4.1] linux-user: unparent CPU object before unref
Date: Tue, 16 Jul 2019 16:02:52 +0100 [thread overview]
Message-ID: <87blxuuin7.fsf@zen.linaroharston> (raw)
In-Reply-To: <CAFEAcA8V2h3pqC_g5aaD2gsv08c18YZKRcxLM+hDNuMv_C55=Q@mail.gmail.com>
Peter Maydell <peter.maydell@linaro.org> writes:
> Ccing the QOM maintainers to make sure we have the
> QOM lifecycle operations right here...
>
> On Tue, 16 Jul 2019 at 15:02, Alex Bennée <alex.bennee@linaro.org> wrote:
>>
>> When a CPU object is created it is parented during it's realize stage.
>> If we don't unparent before the "final" unref we will never finzalize
>> the object leading to a memory leak. For most setups you probably
>> won't notice but with anything that creates and destroys a lot of
>> threads this will add up. This goes especially for architectures which
>> allocate a lot of memory in their CPU structures.
>>
>> Fixes: https://bugs.launchpad.net/qemu/+bug/1836558
>> Cc: 1836558@bugs.launchpad.net
>> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
>> ---
>> linux-user/syscall.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
>> index 39a37496fed..4c9313fd9d0 100644
>> --- a/linux-user/syscall.c
>> +++ b/linux-user/syscall.c
>> @@ -7183,6 +7183,7 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
>> NULL, NULL, 0);
>> }
>> thread_cpu = NULL;
>> + object_unparent(OBJECT(cpu));
>> object_unref(OBJECT(cpu));
>> g_free(ts);
>> rcu_unregister_thread();
>
> I think (as I mentioned on IRC) that we also need to unrealize
> the CPU object, because target/ppc at least does some freeing
> of memory in its unrealize method. I think we do that by
> setting the QOM "realize" property back to "false" -- but that
> might barf if we try it on a CPU that isn't hotpluggable...
I have tried:
thread_cpu = NULL;
+ object_unparent(OBJECT(cpu));
+ object_property_set_bool(OBJECT(cpu), false, "realized", NULL);
object_unref(OBJECT(cpu));
but it didn't manifestly change anything (i.e. both with and without
setting realized the thread allocated stuff is freed). Valgrind still
complains about:
==22483== 6,656 bytes in 26 blocks are possibly lost in loss record 1,639 of 1,654
==22483== at 0x483577F: malloc (vg_replace_malloc.c:299)
==22483== by 0x4D7F8D0: g_malloc (in /usr/lib/x86_64-linux-gnu/libglib-2.0.so.0.5800.3)
==22483== by 0x27D692: create_new_table (translate_init.inc.c:9252)
==22483== by 0x27D7CD: register_ind_in_table (translate_init.inc.c:9291)
==22483== by 0x27D975: register_dblind_insn (translate_init.inc.c:9337)
==22483== by 0x27DBBB: register_insn (translate_init.inc.c:9384)
==22483== by 0x27DE4E: create_ppc_opcodes (translate_init.inc.c:9449)
==22483== by 0x27E79C: ppc_cpu_realize (translate_init.inc.c:9818)
==22483== by 0x2C6FE8: device_set_realized (qdev.c:834)
==22483== by 0x2D1E3D: property_set_bool (object.c:2076)
==22483== by 0x2D00B3: object_property_set (object.c:1268)
==22483== by 0x2D3185: object_property_set_qobject (qom-qobject.c:26)
But I don't know if that is just because the final exit_group of the
main CPU just exits without attempting to clean-up. However it is better
than it was.
--
Alex Bennée
WARNING: multiple messages have this Message-ID (diff)
From: "Alex Bennée" <alex.bennee@linaro.org>
To: qemu-devel@nongnu.org
Subject: [Qemu-devel] [Bug 1836558] Re: [RFC PATCH for 4.1] linux-user: unparent CPU object before unref
Date: Tue, 16 Jul 2019 15:02:52 -0000 [thread overview]
Message-ID: <87blxuuin7.fsf@zen.linaroharston> (raw)
Message-ID: <20190716150252.z54rMT9OgBQABHYAWdfLiyn3QM0V0gD4cZf9byllEkU@z> (raw)
In-Reply-To: CAFEAcA8V2h3pqC_g5aaD2gsv08c18YZKRcxLM+hDNuMv_C55=Q@mail.gmail.com
Peter Maydell <peter.maydell@linaro.org> writes:
> Ccing the QOM maintainers to make sure we have the
> QOM lifecycle operations right here...
>
> On Tue, 16 Jul 2019 at 15:02, Alex Bennée <alex.bennee@linaro.org> wrote:
>>
>> When a CPU object is created it is parented during it's realize stage.
>> If we don't unparent before the "final" unref we will never finzalize
>> the object leading to a memory leak. For most setups you probably
>> won't notice but with anything that creates and destroys a lot of
>> threads this will add up. This goes especially for architectures which
>> allocate a lot of memory in their CPU structures.
>>
>> Fixes: https://bugs.launchpad.net/qemu/+bug/1836558
>> Cc: 1836558@bugs.launchpad.net
>> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
>> ---
>> linux-user/syscall.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
>> index 39a37496fed..4c9313fd9d0 100644
>> --- a/linux-user/syscall.c
>> +++ b/linux-user/syscall.c
>> @@ -7183,6 +7183,7 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
>> NULL, NULL, 0);
>> }
>> thread_cpu = NULL;
>> + object_unparent(OBJECT(cpu));
>> object_unref(OBJECT(cpu));
>> g_free(ts);
>> rcu_unregister_thread();
>
> I think (as I mentioned on IRC) that we also need to unrealize
> the CPU object, because target/ppc at least does some freeing
> of memory in its unrealize method. I think we do that by
> setting the QOM "realize" property back to "false" -- but that
> might barf if we try it on a CPU that isn't hotpluggable...
I have tried:
thread_cpu = NULL;
+ object_unparent(OBJECT(cpu));
+ object_property_set_bool(OBJECT(cpu), false, "realized", NULL);
object_unref(OBJECT(cpu));
but it didn't manifestly change anything (i.e. both with and without
setting realized the thread allocated stuff is freed). Valgrind still
complains about:
==22483== 6,656 bytes in 26 blocks are possibly lost in loss record 1,639 of 1,654
==22483== at 0x483577F: malloc (vg_replace_malloc.c:299)
==22483== by 0x4D7F8D0: g_malloc (in /usr/lib/x86_64-linux-gnu/libglib-2.0.so.0.5800.3)
==22483== by 0x27D692: create_new_table (translate_init.inc.c:9252)
==22483== by 0x27D7CD: register_ind_in_table (translate_init.inc.c:9291)
==22483== by 0x27D975: register_dblind_insn (translate_init.inc.c:9337)
==22483== by 0x27DBBB: register_insn (translate_init.inc.c:9384)
==22483== by 0x27DE4E: create_ppc_opcodes (translate_init.inc.c:9449)
==22483== by 0x27E79C: ppc_cpu_realize (translate_init.inc.c:9818)
==22483== by 0x2C6FE8: device_set_realized (qdev.c:834)
==22483== by 0x2D1E3D: property_set_bool (object.c:2076)
==22483== by 0x2D00B3: object_property_set (object.c:1268)
==22483== by 0x2D3185: object_property_set_qobject (qom-qobject.c:26)
But I don't know if that is just because the final exit_group of the
main CPU just exits without attempting to clean-up. However it is better
than it was.
--
Alex Bennée
--
You received this bug notification because you are a member of qemu-
devel-ml, which is subscribed to QEMU.
https://bugs.launchpad.net/bugs/1836558
Title:
Qemu-ppc Memory leak creating threads
Status in QEMU:
Confirmed
Bug description:
When creating c++ threads (with c++ std::thread), the resulting binary
has memory leaks when running with qemu-ppc.
Eg the following c++ program, when compiled with gcc, consumes more
and more memory while running at qemu-ppc. (does not have memory leaks
when compiling for Intel, when running same binary on real powerpc CPU
hardware also no memory leaks).
(Note I used function getCurrentRSS to show available memory, see
https://stackoverflow.com/questions/669438/how-to-get-memory-usage-at-
runtime-using-c; calls commented out here)
Compiler: powerpc-linux-gnu-g++ (Debian 8.3.0-2) 8.3.0 (but same problem with older g++ compilers even 4.9)
Os: Debian 10.0 ( Buster) (but same problem seen on Debian 9/stetch)
qemu: qemu-ppc version 3.1.50
---
#include <iostream>
#include <thread>
#include <chrono>
using namespace std::chrono_literals;
// Create/run and join a 100 threads.
void Fun100()
{
// auto b4 = getCurrentRSS();
// std::cout << getCurrentRSS() << std::endl;
for(int n = 0; n < 100; n++)
{
std::thread t([]
{
std::this_thread::sleep_for( 10ms );
});
// std::cout << n << ' ' << getCurrentRSS() << std::endl;
t.join();
}
std::this_thread::sleep_for( 500ms ); // to give OS some time to wipe memory...
// auto after = getCurrentRSS();
std::cout << b4 << ' ' << after << std::endl;
}
int main(int, char **)
{
Fun100();
Fun100(); // memory used keeps increasing
}
To manage notifications about this bug go to:
https://bugs.launchpad.net/qemu/+bug/1836558/+subscriptions
next prev parent reply other threads:[~2019-07-16 15:03 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-07-15 10:18 [Qemu-devel] [Bug 1836558] [NEW] Qemu-ppc Memory leak creating threads Daan Scherft
2019-07-15 13:34 ` [Qemu-devel] [Bug 1836558] " Daan Scherft
2019-07-15 14:15 ` Alex Bennée
2019-07-15 15:10 ` Daan Scherft
2019-07-15 15:52 ` Alex Bennée
2019-07-15 16:21 ` Alex Bennée
2019-07-15 16:50 ` Alex Bennée
2019-07-16 12:13 ` [Qemu-devel] [RFC PATCH for 4.1?] target/ppc: move opcode decode tables to PowerPCCPU Alex Bennée
2019-07-16 12:13 ` [Qemu-devel] [Bug 1836558] " Alex Bennée
2019-07-16 14:50 ` [Qemu-devel] " Richard Henderson
2019-07-17 1:33 ` David Gibson
2019-07-17 9:41 ` Alex Bennée
2019-07-17 9:41 ` [Qemu-devel] [Bug 1836558] " Alex Bennée
2019-07-17 12:13 ` [Qemu-devel] " no-reply
2019-07-16 14:01 ` [Qemu-devel] [RFC PATCH for 4.1] linux-user: unparent CPU object before unref Alex Bennée
2019-07-16 14:01 ` [Qemu-devel] [Bug 1836558] " Alex Bennée
2019-07-16 14:43 ` [Qemu-devel] " Peter Maydell
2019-07-16 14:43 ` [Qemu-devel] [Bug 1836558] " Peter Maydell
2019-07-16 15:02 ` Alex Bennée [this message]
2019-07-16 15:02 ` Alex Bennée
2019-07-16 15:17 ` [Qemu-devel] " Philippe Mathieu-Daudé
2020-01-09 13:22 ` [Bug 1836558] Re: Qemu-ppc Memory leak creating threads Thomas Huth
2020-03-10 8:48 ` Laurent Vivier
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=87blxuuin7.fsf@zen.linaroharston \
--to=alex.bennee@linaro.org \
--cc=1836558@bugs.launchpad.net \
--cc=laurent@vivier.eu \
--cc=peter.maydell@linaro.org \
--cc=qemu-devel@nongnu.org \
--cc=riku.voipio@iki.fi \
/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).