From: Claudio Imbrenda <imbrenda@linux.ibm.com>
To: Janis Schoetterl-Glausch <scgl@linux.vnet.ibm.com>
Cc: Janis Schoetterl-Glausch <scgl@linux.ibm.com>,
Thomas Huth <thuth@redhat.com>,
Janosch Frank <frankja@linux.ibm.com>,
David Hildenbrand <david@redhat.com>,
kvm@vger.kernel.org, linux-s390@vger.kernel.org
Subject: Re: [kvm-unit-tests PATCH v3 2/2] s390x: Test specification exceptions during transaction
Date: Tue, 26 Oct 2021 16:55:49 +0200 [thread overview]
Message-ID: <20211026165549.18137134@p-imbrenda> (raw)
In-Reply-To: <32dfb400-4191-44f8-354e-809fac890b63@linux.vnet.ibm.com>
On Tue, 26 Oct 2021 16:22:40 +0200
Janis Schoetterl-Glausch <scgl@linux.vnet.ibm.com> wrote:
> On 10/25/21 19:30, Claudio Imbrenda wrote:
> > On Fri, 22 Oct 2021 14:01:56 +0200
> > Janis Schoetterl-Glausch <scgl@linux.ibm.com> wrote:
> >
> >> Program interruptions during transactional execution cause other
> >> interruption codes.
> >> Check that we see the expected code for (some) specification exceptions.
> >>
> >> Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
> >> ---
>
> [...]
>
> >> +#define TRANSACTION_MAX_RETRIES 5
> >> +
> >> +/* NULL must be passed to __builtin_tbegin via constant, forbid diagnose from
> >> + * being NULL to keep things simple
> >> + */
> >> +static int __attribute__((nonnull))
> >> +with_transaction(void (*trigger)(void), struct __htm_tdb *diagnose)
> >> +{
> >> + int cc;
> >> +
> >
> > if you want to be extra sure, put an assert here (although I'm not sure
> > how nonnull works, I have never seen it before)
>
> Ok, with nonnull, the compiler might warn you if you pass NULL.
fair enough
> >
> >> + cc = __builtin_tbegin(diagnose);
> >> + if (cc == _HTM_TBEGIN_STARTED) {
> >> + trigger();
> >> + __builtin_tend();
> >> + return -TRANSACTION_COMPLETED;
> >> + } else {
> >> + return -cc;
> >> + }
> >> +}
> >> +
> >> +static int retry_transaction(const struct spec_ex_trigger *trigger, unsigned int max_retries,
> >> + struct __htm_tdb *tdb, uint16_t expected_pgm)
> >> +{
> >> + int trans_result, i;
> >> + uint16_t pgm;
> >> +
> >> + for (i = 0; i < max_retries; i++) {
> >> + expect_pgm_int();
> >> + trans_result = with_transaction(trigger->func, tdb);
> >> + if (trans_result == -_HTM_TBEGIN_TRANSIENT) {
> >> + mb();
> >> + pgm = lc->pgm_int_code;
> >> + if (pgm == 0)
> >> + continue;
> >> + else if (pgm == expected_pgm)
> >> + return 0;
> >> + }
> >> + return trans_result;
> >> + }
> >> + return -TRANSACTION_MAX_RETRIES;
> >
> > so this means that a test will be considered failed if the transaction
> > failed too many times?
>
> Yes.
> >
> > this means that could fail if the test is run on busy system, even if
> > the host running the unit test is correct
>
> I suppose so, don't know how likely that is.
I don't like the idea of failing a test when the implementation is
correct, just because the system might be a little more busy than
expected.
if you can't find a way to refactor the test so that it doesn't fail if
there are too many retries, then at least make it a skip?
but I'd really like to see something that does not fail on a correctly
implemented system just because the test machine was too busy.
> >
> > also, do you really need to use negative values? it's probably easier
> > to read if you stick to positive values, and less prone to mistakes if
> > you accidentally forget a - somewhere.
>
> Ok.
> >
> >> +}
> >> +
> >> +static void test_spec_ex_trans(struct args *args, const struct spec_ex_trigger *trigger)
> >> +{
> >> + const uint16_t expected_pgm = PGM_INT_CODE_SPECIFICATION
> >> + | PGM_INT_CODE_TX_ABORTED_EVENT;
> >> + union {
> >> + struct __htm_tdb tdb;
> >> + uint64_t dwords[sizeof(struct __htm_tdb) / sizeof(uint64_t)];
> >> + } diag;
> >> + unsigned int i, failures = 0;
> >> + int trans_result;
> >> +
> >> + if (!test_facility(73)) {
> >> + report_skip("transactional-execution facility not installed");
> >> + return;
> >> + }
> >> + ctl_set_bit(0, CTL0_TRANSACT_EX_CTL); /* enable transactional-exec */
> >> +
> >> + for (i = 0; i < args->iterations && failures <= args->max_failures; i++) {
> >> + register_pgm_cleanup_func(trigger->fixup);
> >> + trans_result = retry_transaction(trigger, args->max_retries, &diag.tdb, expected_pgm);
> >
> > so you retry each iteration up to args->max_retries times, and if a
> > transaction aborts too many times (maybe because the host system is
> > very busy), then you consider it a fail
> >
>
> [...]
next prev parent reply other threads:[~2021-10-26 14:56 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-10-22 12:01 [kvm-unit-tests PATCH v3 0/2] Add specification exception tests Janis Schoetterl-Glausch
2021-10-22 12:01 ` [kvm-unit-tests PATCH v3 1/2] s390x: Add specification exception test Janis Schoetterl-Glausch
2021-10-25 17:17 ` Claudio Imbrenda
2021-10-26 12:00 ` Janis Schoetterl-Glausch
2021-10-26 13:41 ` Claudio Imbrenda
2021-10-27 10:00 ` Janis Schoetterl-Glausch
2021-10-27 12:08 ` Thomas Huth
2021-10-22 12:01 ` [kvm-unit-tests PATCH v3 2/2] s390x: Test specification exceptions during transaction Janis Schoetterl-Glausch
2021-10-25 17:30 ` Claudio Imbrenda
2021-10-25 18:28 ` Christian Borntraeger
2021-10-26 14:22 ` Janis Schoetterl-Glausch
2021-10-26 14:55 ` Claudio Imbrenda [this message]
2021-10-27 10:05 ` Janis Schoetterl-Glausch
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=20211026165549.18137134@p-imbrenda \
--to=imbrenda@linux.ibm.com \
--cc=david@redhat.com \
--cc=frankja@linux.ibm.com \
--cc=kvm@vger.kernel.org \
--cc=linux-s390@vger.kernel.org \
--cc=scgl@linux.ibm.com \
--cc=scgl@linux.vnet.ibm.com \
--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;
as well as URLs for NNTP newsgroup(s).