[8u] RFR(S): 8244548: JDK 8u: sun.misc.Version.jdkUpdateVersion() returns wrong result

[Severin Gehwolf]

Hi Andrew,

Thanks for looking at this.

On Tue, 2020-05-12 at 14:07 +0100, Andrew Haley wrote:
> On 5/7/20 4:09 PM, Severin Gehwolf wrote:
> > jvm_version_info.jvm_version currently holds this quadruplet:
> > 
> > Most significant 8 bits => major version, followed by 8 bits => minor
> > version, followed by 8 bits => micro version, followed by 8 bits =>
> > build version. Note that JVM minor version represents the update
> > version as passed in via configure and the micro version is currently
> > not used (always 0). See vm_version.cpp lines 100-102 where only major,
> > minor and build number are ever been set. Knowing this, we can still
> > preserve the same behavior after patch by defining JVM_VERSION_MICRO to
> > 0 for any version.
> 
> This is tricky. JVM_GetVersionInfo is a function exported by
> libjvm.so, and the version is simply encoded as
> 
> unsigned int Abstract_VM_Version::jvm_version() {
>   return ((Abstract_VM_Version::vm_major_version() & 0xFF) << 24) |
>          ((Abstract_VM_Version::vm_minor_version() & 0xFF) << 16) |
>          (Abstract_VM_Version::vm_build_number() & 0xFF);
> }
> 
> I guess we could argue that this is for JVM-JDK internal use only, and
> no-one else cares.
> 
> Or we could encode it in a different way such that at least we have a
> jvm_version that is monotonically increasing, perhaps by (ab)using the
> lower 8 bits of the version, the vm_build_number. But I guess I'm
> being paranoid, and no tools are going to care about minor versions
> anyway,even if they do call JVM_GetVersionInfo.

Yes, this is indeed tricky. The trouble is that
Abstract_VM_Version::vm_build_number() actually holds the configured
JDK build number, we can't really use it. It's already being used.

(gdb)
99	
100	  _vm_major_version = atoi(vm_major_ver);
101	  _vm_minor_version = atoi(vm_minor_ver);
102	  _vm_build_number  = atoi(vm_build_num);
103	
104	  os::free(vm_version);
105	  _initialized = true;
106	}
107	
108	#if defined(_LP64)
(gdb) p _vm_build_number 
$1 = 2

The only bits which seem unused are bits 8 through 16 of this unsigned
int. So even if JVM_GetVersionInfo *is* being used somewhere in the
wild, it *should* continue to work. Hence, the proposal to now also use
those unused bits for the minor version.

Thoughts?

Thanks,
Severin