Reuse the top bits of _JB_SIGFLAG field previously used to store a
boolean to store a cookie that's validated by [sig]longjmp to make it
harder to use as a ROP gadget. Additionally, encrypt saved registers
with the cookie so that an attacker can't modify a register's value to
a specific value without knowing the cookie.
Bug: http://b/23942752
Change-Id: Id0eb8d06916e89d5d776bfcaa9458f8826717ba3
Although the LP32 mips sigset_t is large enough to represent all signals,
their jmp_buf is too small. This test succeeded on arm and x86 because the
RT signals were never in the 'expected' sigset_t, so the equality comparison
with the 'actual' sigset_t worked fine --- everyone was blind to the RT
signal. On mips the tests fail because the 'expected' sigset_t does contain
the RT signal but the 'actual' doesn't because the jmp_buf only saves and
restores the first 32 signals.
There are 32 free bits (currently used as padding) in the LP32 mips jmp_buf,
and they might choose to use those to provide better support than the other
two platforms, but I'll leave that to them. It will be easy to just remove
the #if defined(__LP64__) from this change in that case.
For mips64 it's not to late to increase the size of the jmp_buf and fix
the setjmp family, but since there are decisions to be made here for LP32,
I'll leave it all to Imagination folks...
Bug: 16918359
Change-Id: I6b723712fce0e9210dafa165d8599d950b2d3500
