On Fri, Oct 16, 2020, 6:46 AM Jann Horn <jannh(a)google.com>
wrote:
> +openrisc folks
>
> On Thu, Oct 15, 2020 at 11:28 PM kernel test robot <lkp(a)intel.com> wrote:
> > tree:
>
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable-rc.git
> linux-5.4.y
> > head: 85b0841aab15c12948af951d477183ab3df7de14
> > commit: c5665cafbedd2e2a523fe933e452391a02d3adb3 [665/2391] binder:
> Prevent context manager from incrementing ref 0
> > config: openrisc-randconfig-r002-20201014 (attached as .config)
> > compiler: or1k-linux-gcc (GCC) 9.3.0
> > reproduce (this is a W=1 build):
> > wget
>
https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross
> -O ~/bin/make.cross
> > chmod +x ~/bin/make.cross
> > #
>
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable-rc.gi...
> > git remote add linux-stable-rc
>
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable-rc.git
> > git fetch --no-tags linux-stable-rc linux-5.4.y
> > git checkout c5665cafbedd2e2a523fe933e452391a02d3adb3
> > # save the attached .config to linux build tree
> > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-9.3.0 make.cross
> ARCH=openrisc
> >
> > If you fix the issue, kindly add following tag as appropriate
> > Reported-by: kernel test robot <lkp(a)intel.com>
> >
> > All errors (new ones prefixed by >>):
> >
> > drivers/android/binder.c: Assembler messages:
> > >> drivers/android/binder.c:3776: Error: unrecognized keyword/register
> name `l.lwz ?ap,4(r25)'
> > drivers/android/binder.c:3781: Error: unrecognized keyword/register
> name `l.addi ?ap,r0,0'
>
> binder is basically doing this:
>
> u64 data_ptr;
> if (get_user(data_ptr, (u64 __user *)ptr))
> return -EFAULT;
>
> and GCC complains that that doesn't turn into valid assembly on
> openrisc, where get_user() of size 8 expands into this:
>
> #define __get_user_asm2(x, addr, err) \
> { \
> unsigned long long __gu_tmp; \
> __asm__ __volatile__( \
> "1: l.lwz %1,0(%2)\n" \
> "2: l.lwz %H1,4(%2)\n" \
> "3:\n" \
> ".section .fixup,\"ax\"\n" \
> "4: l.addi %0,r0,%3\n" \
> " l.addi %1,r0,0\n" \
> " l.addi %H1,r0,0\n" \
> " l.j 3b\n" \
> " l.nop\n" \
> ".previous\n" \
> ".section __ex_table,\"a\"\n" \
> " .align 2\n" \
> " .long 1b,4b\n" \
> " .long 2b,4b\n" \
> ".previous" \
> : "=r"(err), "=&r"(__gu_tmp) \
> : "r"(addr), "i"(-EFAULT), "0"(err));
\
> (x) = (__typeof__(*(addr)))( \
> (__typeof__((x)-(x)))__gu_tmp); \
> }
>
> and apparently the "l.lwz %H1,4(%2)" and "l.addi %H1,r0,0"
don't turn
> into valid assembly when %H1 expands to "?ap"?
>
> I don't know anything about OpenRISC, but this seems like it's
> probably an issue in the get_user() implementation.
>
This is fixed in 5.9. I think the patch can be cherry picked by itself.
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit...
Does not apply cleanly to 5.8.y or 5.4.y, can someone please properly
backport it and send it to stable(a)vger.kernel.org?
thanks,
greg k-h