From: David Hildenbrand <david@redhat.com>
To: Janosch Frank <frankja@linux.ibm.com>, kvm@vger.kernel.org
Cc: thuth@redhat.com, borntraeger@de.ibm.com,
linux-s390@vger.kernel.org, cohuck@redhat.com
Subject: Re: [kvm-unit-tests PATCH v4 2/9] s390x: smp: Only use smp_cpu_setup once
Date: Fri, 24 Jan 2020 09:38:06 +0100 [thread overview]
Message-ID: <8deec98c-d384-bc2e-2790-7b47fcfd2d62@redhat.com> (raw)
In-Reply-To: <b16cbc73-20b2-4955-52a7-1b133aa33cc4@redhat.com>
On 24.01.20 09:28, David Hildenbrand wrote:
> On 23.01.20 14:56, David Hildenbrand wrote:
>> On 23.01.20 14:54, Janosch Frank wrote:
>>> On 1/23/20 2:45 PM, David Hildenbrand wrote:
>>>> On 21.01.20 14:42, Janosch Frank wrote:
>>>>> Let's stop and start instead of using setup to run a function on a
>>>>> cpu.
>>>>>
>>>>> Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
>>>>> Reviewed-by: Thomas Huth <thuth@redhat.com>
>>>>> Reviewed-by: Cornelia Huck <cohuck@redhat.com>
>>>>> Acked-by: David Hildenbrand <david@redhat.com>
>>>>> ---
>>>>> s390x/smp.c | 21 ++++++++++++++-------
>>>>> 1 file changed, 14 insertions(+), 7 deletions(-)
>>>>>
>>>>> diff --git a/s390x/smp.c b/s390x/smp.c
>>>>> index e37eb56..3e8cf3e 100644
>>>>> --- a/s390x/smp.c
>>>>> +++ b/s390x/smp.c
>>>>> @@ -53,7 +53,7 @@ static void test_start(void)
>>>>> psw.addr = (unsigned long)test_func;
>>>>>
>>>>> set_flag(0);
>>>>> - smp_cpu_setup(1, psw);
>>>>> + smp_cpu_start(1, psw);
>>>>> wait_for_flag();
>>>>> report(1, "start");
>>>>> }
>>>>> @@ -109,6 +109,7 @@ static void test_store_status(void)
>>>>> report(1, "status written");
>>>>> free_pages(status, PAGE_SIZE * 2);
>>>>> report_prefix_pop();
>>>>> + smp_cpu_stop(1);
>>>>>
>>>>> report_prefix_pop();
>>>>> }
>>>>> @@ -137,9 +138,8 @@ static void test_ecall(void)
>>>>>
>>>>> report_prefix_push("ecall");
>>>>> set_flag(0);
>>>>> - smp_cpu_destroy(1);
>>>>>
>>>>> - smp_cpu_setup(1, psw);
>>>>> + smp_cpu_start(1, psw);
>>>>> wait_for_flag();
>>>>> set_flag(0);
>>>>> sigp(1, SIGP_EXTERNAL_CALL, 0, NULL);
>>>>> @@ -172,9 +172,8 @@ static void test_emcall(void)
>>>>>
>>>>> report_prefix_push("emcall");
>>>>> set_flag(0);
>>>>> - smp_cpu_destroy(1);
>>>>>
>>>>> - smp_cpu_setup(1, psw);
>>>>> + smp_cpu_start(1, psw);
>>>>> wait_for_flag();
>>>>> set_flag(0);
>>>>> sigp(1, SIGP_EMERGENCY_SIGNAL, 0, NULL);
>>>>> @@ -192,7 +191,7 @@ static void test_reset_initial(void)
>>>>> psw.addr = (unsigned long)test_func;
>>>>>
>>>>> report_prefix_push("reset initial");
>>>>> - smp_cpu_setup(1, psw);
>>>>> + smp_cpu_start(1, psw);
>>>>>
>>>>> sigp_retry(1, SIGP_INITIAL_CPU_RESET, 0, NULL);
>>>>> sigp(1, SIGP_STORE_STATUS_AT_ADDRESS, (uintptr_t)status, NULL);
>>>>> @@ -223,7 +222,7 @@ static void test_reset(void)
>>>>> psw.addr = (unsigned long)test_func;
>>>>>
>>>>> report_prefix_push("cpu reset");
>>>>> - smp_cpu_setup(1, psw);
>>>>> + smp_cpu_start(1, psw);
>>>>>
>>>>> sigp_retry(1, SIGP_CPU_RESET, 0, NULL);
>>>>> report(smp_cpu_stopped(1), "cpu stopped");
>>>>> @@ -232,6 +231,7 @@ static void test_reset(void)
>>>>>
>>>>> int main(void)
>>>>> {
>>>>> + struct psw psw;
>>>>> report_prefix_push("smp");
>>>>>
>>>>> if (smp_query_num_cpus() == 1) {
>>>>> @@ -239,6 +239,12 @@ int main(void)
>>>>> goto done;
>>>>> }
>>>>>
>>>>> + /* Setting up the cpu to give it a stack and lowcore */
>>>>> + psw.mask = extract_psw_mask();
>>>>> + psw.addr = (unsigned long)cpu_loop;
>>>>> + smp_cpu_setup(1, psw);
>>>>> + smp_cpu_stop(1);
>>>>> +
>>>>> test_start();
>>>>> test_stop();
>>>>> test_stop_store_status();
>>>>> @@ -247,6 +253,7 @@ int main(void)
>>>>> test_emcall();
>>>>> test_reset();
>>>>> test_reset_initial();
>>>>> + smp_cpu_destroy(1);
>>>>>
>>>>> done:
>>>>> report_prefix_pop();
>>>>>
>>>>
>>>> With this patch, I get timeouts under TCG. Seems to loop forever.
>>>>
>>> The branch works on lpar and kvm without a problem.
>>
>> Which could mean that either TCG is broken or your test is broken (e.g.,
>> a race condition that does not trigger under LPAR because it's faster,
>> or some undocumented/not guaranteed behavior).
>>
>
> So, the test works every now and then under TCG.
>
> It seems to work very reliably with "-smp 2"
>
> With smp 8, it sometimes works, sometimes not.
>
> t480s: ~/git/kvm-unit-tests (kein Branch, Rebase von s390x-prep) $
> ./s390x-run s390x/smp.elf -smp 8
> /home/dhildenb/git/qemu/s390x-softmmu/qemu-system-s390x -nodefaults
> -nographic -machine s390-ccw-virtio,accel=tcg -chardev stdio,id=con0
> -device sclpconsole,chardev=con0 -kernel s390x/smp.elf -smp 8 # -initrd
> /tmp/tmp.UL9ZaqoKBW
> SMP: Initializing, found 8 cpus
> PASS: smp: start
> PASS: smp: stop
> PASS: smp: stop store status: prefix
> PASS: smp: stop store status: stack
> PASS: smp: store status at address: running: incorrect state
> PASS: smp: store status at address: running: status not written
> PASS: smp: store status at address: stopped: status written
> PASS: smp: ecall: ecall
> PASS: smp: emcall: ecall
> PASS: smp: cpu reset: cpu stopped
> PASS: smp: reset initial: clear: psw
> PASS: smp: reset initial: clear: prefix
> PASS: smp: reset initial: clear: fpc
> PASS: smp: reset initial: clear: cpu timer
> PASS: smp: reset initial: clear: todpr
> PASS: smp: reset initial: initialized: cr0 == 0xE0
> PASS: smp: reset initial: initialized: cr14 == 0xC2000000
> PASS: smp: reset initial: cpu stopped
> SUMMARY: 18 tests
>
> EXIT: STATUS=1
>
> t480s: ~/git/kvm-unit-tests (kein Branch, Rebase von s390x-prep) $
> ./s390x-run s390x/smp.elf -smp 8
> /home/dhildenb/git/qemu/s390x-softmmu/qemu-system-s390x -nodefaults
> -nographic -machine s390-ccw-virtio,accel=tcg -chardev stdio,id=con0
> -device sclpconsole,chardev=con0 -kernel s390x/smp.elf -smp 8 # -initrd
> /tmp/tmp.csifrEDgnC
> SMP: Initializing, found 8 cpus
> PASS: smp: start
> PASS: smp: stop
> PASS: smp: stop store status: prefix
> FAIL: smp: stop store status: stack
> PASS: smp: store status at address: running: incorrect state
> PASS: smp: store status at address: running: status not written
> PASS: smp: store status at address: stopped: status written
> [... hang]
>
>
> Note that there is a previous failure for "smp: stop store status:
> stack" whenever the test will hang later.
>
>
In smp_cpu_stop_nolock(), when you do a SIGP_STOP_AND_STORE_STATUS, you
wait until the target CPU is stopped.
This does, however, not guarantee that the state has already been stored
- especially not if the target CPU is already stopped.
Stop and Store Status
The addressed CPU performs the stop function, fol-
lowed by the store-status operation (see “Store Sta-
tus” on page 4-82). The CPU does not necessarily
complete the operation, or even enter the stopped
state, during the execution of SIGNAL PROCES-
SOR ..."
So there is a race in your code. I do wonder how you can test if the
order on the target CPU has been executed, without issuing another SIGP
operation (e.g., restart) or by testing if the content to be stored was
all changed ...
--
Thanks,
David / dhildenb
next prev parent reply other threads:[~2020-01-24 8:38 UTC|newest]
Thread overview: 26+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-01-21 13:42 [kvm-unit-tests PATCH v4 0/9] s390x: smp: Improve smp code and reset checks Janosch Frank
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 1/9] s390x: smp: Cleanup smp.c Janosch Frank
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 2/9] s390x: smp: Only use smp_cpu_setup once Janosch Frank
2020-01-23 13:45 ` David Hildenbrand
2020-01-23 13:54 ` Janosch Frank
2020-01-23 13:56 ` David Hildenbrand
2020-01-24 8:28 ` David Hildenbrand
2020-01-24 8:38 ` David Hildenbrand [this message]
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 3/9] s390x: Add cpu id to interrupt error prints Janosch Frank
2020-01-23 10:02 ` Janosch Frank
2020-01-23 10:03 ` [kvm-unit-tests PATCH] " Janosch Frank
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 4/9] s390x: smp: Rework cpu start and active tracking Janosch Frank
2020-01-21 17:40 ` David Hildenbrand
2020-01-22 8:25 ` Janosch Frank
2020-01-23 9:14 ` [kvm-unit-tests PATCH] " Janosch Frank
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 5/9] s390x: smp: Wait for cpu setup to finish Janosch Frank
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 6/9] s390x: smp: Loop if secondary cpu returns into cpu setup again Janosch Frank
2020-01-21 14:28 ` Cornelia Huck
2020-01-21 17:36 ` David Hildenbrand
2020-01-23 13:32 ` David Hildenbrand
2020-01-23 13:47 ` Janosch Frank
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 7/9] s390x: smp: Remove unneeded cpu loops Janosch Frank
2020-01-21 14:30 ` Cornelia Huck
2020-01-21 17:36 ` David Hildenbrand
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 8/9] s390x: smp: Test all CRs on initial reset Janosch Frank
2020-01-21 13:42 ` [kvm-unit-tests PATCH v4 9/9] s390x: smp: Dirty fpc before initial reset test Janosch Frank
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=8deec98c-d384-bc2e-2790-7b47fcfd2d62@redhat.com \
--to=david@redhat.com \
--cc=borntraeger@de.ibm.com \
--cc=cohuck@redhat.com \
--cc=frankja@linux.ibm.com \
--cc=kvm@vger.kernel.org \
--cc=linux-s390@vger.kernel.org \
--cc=thuth@redhat.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