queue-4.14/powerpc-tm-limit-tm-code-inside-ppc_transactional_me.patch

   1 From 4e9a1c280eee932a7a100607a471f56466311eb7 Mon Sep 17 00:00:00 2001
   2 From: Breno Leitao <leitao@debian.org>
   3 Date: Mon, 8 Apr 2019 16:32:38 +1000
   4 Subject: powerpc/tm: Limit TM code inside PPC_TRANSACTIONAL_MEM
   5
   6 [ Upstream commit 897bc3df8c5aebb54c32d831f917592e873d0559 ]
   7
   8 Commit e1c3743e1a20 ("powerpc/tm: Set MSR[TS] just prior to recheckpoint")
   9 moved a code block around and this block uses a 'msr' variable outside of
  10 the CONFIG_PPC_TRANSACTIONAL_MEM, however the 'msr' variable is declared
  11 inside a CONFIG_PPC_TRANSACTIONAL_MEM block, causing a possible error when
  12 CONFIG_PPC_TRANSACTION_MEM is not defined.
  13
  14         error: 'msr' undeclared (first use in this function)
  15
  16 This is not causing a compilation error in the mainline kernel, because
  17 'msr' is being used as an argument of MSR_TM_ACTIVE(), which is defined as
  18 the following when CONFIG_PPC_TRANSACTIONAL_MEM is *not* set:
  19
  20         #define MSR_TM_ACTIVE(x) 0
  21
  22 This patch just fixes this issue avoiding the 'msr' variable usage outside
  23 the CONFIG_PPC_TRANSACTIONAL_MEM block, avoiding trusting in the
  24 MSR_TM_ACTIVE() definition.
  25
  26 Cc: stable@vger.kernel.org
  27 Reported-by: Christoph Biedl <linux-kernel.bfrz@manchmal.in-ulm.de>
  28 Fixes: e1c3743e1a20 ("powerpc/tm: Set MSR[TS] just prior to recheckpoint")
  29 Signed-off-by: Breno Leitao <leitao@debian.org>
  30 Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
  31 Signed-off-by: Sasha Levin <sashal@kernel.org>
  32 ---
  33  arch/powerpc/kernel/signal_64.c | 23 ++++++++++++++++++-----
  34  1 file changed, 18 insertions(+), 5 deletions(-)
  35
  36 diff --git a/arch/powerpc/kernel/signal_64.c b/arch/powerpc/kernel/signal_64.c
  37 index 979b9463e17b..927384d85faf 100644
  38 --- a/arch/powerpc/kernel/signal_64.c
  39 +++ b/arch/powerpc/kernel/signal_64.c
  40 @@ -746,12 +746,25 @@ int sys_rt_sigreturn(unsigned long r3, unsigned long r4, unsigned long r5,
  41                 if (restore_tm_sigcontexts(current, &uc->uc_mcontext,
  42                                            &uc_transact->uc_mcontext))
  43                         goto badframe;
  44 -       }
  45 -       else
  46 -       /* Fall through, for non-TM restore */
  47 +       } else
  48  #endif
  49 -       if (restore_sigcontext(current, NULL, 1, &uc->uc_mcontext))
  50 -               goto badframe;
  51 +       {
  52 +               /*
  53 +                * Fall through, for non-TM restore
  54 +                *
  55 +                * Unset MSR[TS] on the thread regs since MSR from user
  56 +                * context does not have MSR active, and recheckpoint was
  57 +                * not called since restore_tm_sigcontexts() was not called
  58 +                * also.
  59 +                *
  60 +                * If not unsetting it, the code can RFID to userspace with
  61 +                * MSR[TS] set, but without CPU in the proper state,
  62 +                * causing a TM bad thing.
  63 +                */
  64 +               current->thread.regs->msr &= ~MSR_TS_MASK;
  65 +               if (restore_sigcontext(current, NULL, 1, &uc->uc_mcontext))
  66 +                       goto badframe;
  67 +       }
  68
  69         if (restore_altstack(&uc->uc_stack))
  70                 goto badframe;
  71 --
  72 2.19.1
  73