[openssl-users] 1.1.1 pre1 tests failing on Solaris SPARC

[Andy Polyakov]

>> And "the default for all v9 architectures is -xmemalign=8s".
> I'm getting confused.  Since I did not specify -xmemalign at all,

And not specifying -xmemalign is equivalent of specifying 8s in 64-bit
build such as one in question.

> why
> did the test fail with SIGBUS in the first place?  Seems like there
> should have been no alignment problem if the compiler is doing the right
> thing by default as you say.

Once again, objects on stack are *customarily* aligned at pointer size,
i.e. at 8 bytes in -xarch=v9 case, even if their declaration doesn't
imply corresponding guarantee. Or in other words, specifically in
context of this question, even though 'buf' is *not required* to be
aligned at 8 bytes by language standard, so far it was effectively
observed to be aligned anyway, at least on other RISC platforms. Now,
I'm *not* saying that we should *rely* on this custom, rather contrary,
we definitely should *not*. Question is what does the fact that it went
unnoticed till now mean. Or in more practical terms are there more such
dependencies that shouldn't be there? Because if possible, it would only
be appropriate to locate and eliminate them. In yet other words, mystery
is not why this specific test crashed on you, but rather what else can
crash (but doesn't for some reason).