In restore_tm_sigcontexts() we take the trap value directly from the
user sigcontext with no checking:
err |= __get_user(regs->trap, &sc->gp_regs[PT_TRAP]);
This means we can be in the kernel with an arbitrary regs->trap value.
Although that's not immediately problematic, there is a risk we could
trigger one of the uses of CHECK_FULL_REGS():
#define CHECK_FULL_REGS(regs) BUG_ON(regs->trap & 1)
It can also cause us to unnecessarily save non-volatile GPRs again in
save_nvgprs(), which shouldn't be problematic but is still wrong.
It's also possible it could trick the syscall restart machinery, which
relies on regs->trap not being == 0xc00 (see 9a81c16b52 ("powerpc:
fix double syscall restarts")), though I haven't been able to make
that happen.
Finally it doesn't match the behaviour of the non-TM case, in
restore_sigcontext() which zeroes regs->trap.
So change restore_tm_sigcontexts() to zero regs->trap.
This was discovered while testing Nick's upcoming rewrite of the
syscall entry path. In that series the call to save_nvgprs() prior to
signal handling (do_notify_resume()) is removed, which leaves the
low-bit of regs->trap uncleared which can then trigger the FULL_REGS()
WARNs in setup_tm_sigcontexts().
Fixes: 2b0a576d15 ("powerpc: Add new transactional memory state to the signal context")
Cc: stable@vger.kernel.org # v3.9+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200401023836.3286664-1-mpe@ellerman.id.au
c7def7fbde