RFR: 8308975: Fix signed integer overflow in compiler code, part 2

[Coleen Phillimore]

On Fri, 26 May 2023 22:20:17 GMT, Dean Long <dlong at openjdk.org> wrote:

> This is a followup from part 1.  I had to use java_add/subtract/multiply/negate in a few more places.
> cpu_loads_process had an overflow on BSD when the first snapshot is relative to the epoch in 1970.
> Use unsigned for C2 Type hash functions.

I wandered into this wondering if -Wconversion would have found these signed integer overflow but it looks like maybe -Wsign-conversion might have found this?  We've been enabling these flags and disabling -Werror and searching the result of the build.

src/hotspot/os/bsd/os_perf_bsd.cpp line 173:

> 171:     result = OS_ERR;
> 172:   } else {
> 173:     long delta_nanos = active_processor_count * (total_cpu_nanos - _total_cpu_nanos);

We shouldn't use 'long' in the source code even though long is always 64 bits on bsd and this is specific to that, we often scan for this as a potential error since it's 32 bits on Windows.  int64_t is preferable.  Can you use that?

src/hotspot/share/opto/type.hpp line 525:

> 523: public:
> 524:   virtual bool eq( const Type *t ) const;
> 525:   virtual uint  hash() const;            // Type specific hashing

nit: align hash to singleton (a few places)

-------------

PR Review: https://git.openjdk.org/jdk/pull/14188#pullrequestreview-1450763238
PR Review Comment: https://git.openjdk.org/jdk/pull/14188#discussion_r1210162360
PR Review Comment: https://git.openjdk.org/jdk/pull/14188#discussion_r1210164058