On Wed, Jul 23, 2014 at 02:40:10PM +0100, Alex Smith wrote:
> Copy FCSR in the FP regset to match the original pre-regset core dumper.
> The code paths for where sizeof(union fpureg) == sizeof(elf_fpreg_t)
> already do so, but they actually copy 4 bytes more than they should do
> as FCSR is only 32 bits. The not equal code paths do not copy it at all.
> Therefore change the copy to be done explicitly (with the correct size)
> for both paths.
Ah, I hadn't realised that ELF_NFPREG == 33, sneaky! That together with
the "XXX fcr31" comment led me to believe the FP regset didn't include
FCSR which is why I hadn't fixed the oops there or taken it into account
for the case where FPR size != sizeof(elf_fpreg_t) (ie. when MSA support
is enabled).
> Additionally, clear the cause bits from FCSR when setting the FP regset
> to avoid the possibility of causing an FP exception (and an oops) in the
> kernel.
>
> Signed-off-by: Alex Smith <alex@xxxxxxxxxxxxxxxx>
> Cc: Paul Burton <paul.burton@xxxxxxxxxx>
> Cc: <stable@xxxxxxxxxxxxxxx> # v3.13+
> ---
> This patch incorporates a fix for another instance of the bug fixed by
> Paul Burton's patch "MIPS: prevent user from setting FCSR cause bits" -
> the code path in fpr_set for sizeof(fpureg) == sizeof(elf_fpreg_t)
> copied fcr31 without clearing cause bits. I've incorporated a fix for
> it into this patch to so that it's easier to apply both patches without
> conflicts.
> ---
> arch/mips/kernel/ptrace.c | 61 +++++++++++++++++++++++++++++------------------
> 1 file changed, 38 insertions(+), 23 deletions(-)
>
> diff --git a/arch/mips/kernel/ptrace.c b/arch/mips/kernel/ptrace.c
> index 8bd13ed..ffc2e37 100644
> --- a/arch/mips/kernel/ptrace.c
> +++ b/arch/mips/kernel/ptrace.c
> @@ -409,23 +409,28 @@ static int fpr_get(struct task_struct *target,
> int err;
> u64 fpr_val;
>
> - /* XXX fcr31 */
> -
> - if (sizeof(target->thread.fpu.fpr[i]) == sizeof(elf_fpreg_t))
> - return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
> - &target->thread.fpu,
> - 0, sizeof(elf_fpregset_t));
> -
> - for (i = 0; i < NUM_FPU_REGS; i++) {
> - fpr_val = get_fpr64(&target->thread.fpu.fpr[i], 0);
> + if (sizeof(target->thread.fpu.fpr[i]) == sizeof(elf_fpreg_t)) {
> err = user_regset_copyout(&pos, &count, &kbuf, &ubuf,
> - &fpr_val, i * sizeof(elf_fpreg_t),
> - (i + 1) * sizeof(elf_fpreg_t));
> + &target->thread.fpu.fpr,
> + 0, NUM_FPU_REGS * sizeof(elf_fpreg_t));
> if (err)
> return err;
> + } else {
> + for (i = 0; i < NUM_FPU_REGS; i++) {
> + fpr_val = get_fpr64(&target->thread.fpu.fpr[i], 0);
> + err = user_regset_copyout(&pos, &count, &kbuf, &ubuf,
> + &fpr_val,
> + i * sizeof(elf_fpreg_t),
> + (i + 1) * sizeof(elf_fpreg_t));
> + if (err)
> + return err;
> + }
> }
>
> - return 0;
> + return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
> + &target->thread.fpu.fcr31,
> + NUM_FPU_REGS * sizeof(elf_fpreg_t),
> + (NUM_FPU_REGS * sizeof(elf_fpreg_t)) + sizeof(u32));
The only problem I can think of is that the final register in the regset
will still be treated as 64b (regset->size) as far as ptrace is
concerned, so I'm not sure how best to handle this. I presume the pre
regset core dump format placed the 32b FCSR value immediately after
the 64b $f31, as you have here? In which case we should probably at
least zero out the other 4 bytes of this final "register", assuming
the extra 4 bytes compared to the pre-regset version isn't a problem?
Thanks,
Paul
> }
>
> static int fpr_set(struct task_struct *target,
> @@ -436,23 +441,33 @@ static int fpr_set(struct task_struct *target,
> unsigned i;
> int err;
> u64 fpr_val;
> + u32 fcr31;
>
> - /* XXX fcr31 */
> -
> - if (sizeof(target->thread.fpu.fpr[i]) == sizeof(elf_fpreg_t))
> - return user_regset_copyin(&pos, &count, &kbuf, &ubuf,
> - &target->thread.fpu,
> - 0, sizeof(elf_fpregset_t));
> -
> - for (i = 0; i < NUM_FPU_REGS; i++) {
> + if (sizeof(target->thread.fpu.fpr[i]) == sizeof(elf_fpreg_t)) {
> err = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
> - &fpr_val, i * sizeof(elf_fpreg_t),
> - (i + 1) * sizeof(elf_fpreg_t));
> + &target->thread.fpu.fpr,
> + 0, NUM_FPU_REGS * sizeof(elf_fpreg_t));
> if (err)
> return err;
> - set_fpr64(&target->thread.fpu.fpr[i], 0, fpr_val);
> + } else {
> + for (i = 0; i < NUM_FPU_REGS; i++) {
> + err = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
> + &fpr_val,
> + i * sizeof(elf_fpreg_t),
> + (i + 1) * sizeof(elf_fpreg_t));
> + if (err)
> + return err;
> + set_fpr64(&target->thread.fpu.fpr[i], 0, fpr_val);
> + }
> }
>
> + err = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &fcr31,
> + NUM_FPU_REGS * sizeof(elf_fpreg_t),
> + (NUM_FPU_REGS * sizeof(elf_fpreg_t)) + sizeof(u32));
> + if (err)
> + return err;
> +
> + target->thread.fpu.fcr31 = fcr31 & ~FPU_CSR_ALL_X;
> return 0;
> }
>
> --
> 1.9.1
>
--
To unsubscribe from this list: send the line "unsubscribe stable" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html