RFR(XS): 8204135: jtreg: Fix failing RTM test TestUseRTMXendForLockBusy

[Igor Ignatyev]

Hi Gustavo,

Unfortunately, I don't recall any detail on this test. Filipp (added) might remember something. but I wouldn't expect him to answer though as he hasn't been actively contributing in open jdk lately.  

I have checked the history, the test hasn't been changed since its integration -- http://hg.openjdk.java.net/jdk9/jdk9/hotspot/rev/3c9c3ba62dfd <http://hg.openjdk.java.net/jdk9/jdk9/hotspot/rev/3c9c3ba62dfd>. although your explanation sounds reasonable to me, I don't understand how the test passed before.

Thanks,
-- Igor

> On May 30, 2018, at 7:04 PM, Gustavo Romero <gromero at linux.vnet.ibm.com> wrote:
> 
> Hi,
> 
> Could the following change be reviewed please?
> 
> webrev: http://cr.openjdk.java.net/~gromero/8204135/v1
> bug   : https://bugs.openjdk.java.net/browse/JDK-8204135
> 
> It fixes currently failing RTM test "TestUseRTMXendForLockBusy" by changing
> in which cases UseRTMForStackLocks flags must be enabled/disabled.
> 
> I could not track down on the timeline when exactly that test stopped to
> pass, however my understanding is that if inflateMonitor is 'false' that
> should /not/ imply that UseRTMForStackLocks is also 'false'. Thus I
> understand that for this particular test if inflated monitors are used,
> then UseRTMForStackLocks has to be disabled, i.e. 'false'. On the other
> hand, if inflated monitors are /not/ used, then UseRTMForStackLocks must be
> enabled.
> 
> Currently the test is failing for the following cases where inflateMonitor
> is 'false', because it implies that UseRTMForStackLocks is disabled and so
> no abort statistics is generated (which is correct from the JVM's
> perspective):
> 
>        // stack lock, xabort on lock busy
>        verifyXendForLockBusy(false, false);
>        // stack lock, xend on lock busy
>        verifyXendForLockBusy(false, true);
> 
> In other words, "stack lock" case should imply UseRTMForStackLocks = 'true',
> not 'false'.
> 
> Cases:
>        // inflated lock, xabort on lock busy
>        verifyXendForLockBusy(true, false);
>        // inflated lock, xend on lock busy
>        verifyXendForLockBusy(true, true);
> 
> are not failing because UseRTMForStackLocks = 'true' and they will generate
> proper abort statistics when  +UseRTMXendForLockBusy ('xend' is used to
> finish the transaction) or when -UseRTMXendForLockBusy ('xabort' is used
> to finish the transaction).
> 
> So, in summing up: (a) inflated lock cases are not being tested correctly
> (since UseRTMForStackLocks is always enabled) and (b) stack lock cases are
> failing because UseRTMForStackLocks is always disabled.
> 
> @Igor, it's a long time ago, but maybe you recall some details on that
> test... :-)
> 
> 
> Thank you.
> 
> Best regards,
> Gustavo
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20180530/cdc5e456/attachment.html>