ARM linux-3.11.1-300.fc19.armv7hl incorrect handling uint16_t in probe_kernel_read() probe_kernel_write() functions

[William Cohen]

Hello Everyone,

I was reviewing the systemtap testsuite results for ARM on Fedora 19 and I found that one
of the tests (systemtap.base/deref.stp) failing on ARM for uint16_t reads and writes.  The
same tests work on x86_64. There appears to be ARM specific code in
linux/arch/arm/include/asm/uaccess.h to read and write memory.  I suspect that the 16-bit
case for copies isn't being handled correctly.

The issue can be observed by   

# yum install kernel-devel systemtap-testsuite -y
# debuginfo-install kernel-`uname -r` -y

$ uname -a
Linux dhcp129-3.rdu.redhat.com 3.11.1-300.fc19.armv7hl #1 SMP Thu Sep 19 16:31:05 EDT 2013
armv7l armv7l armv7l GNU/Linux
$ rpm -q systemtap
systemtap-2.3-1.fc19.armv7hl

$cd /usr/share/systemtap/testsuite
$sudo make installcheck RUNTESTFLAGS="--debug systemtap.base/deref.exp"

On ARM Fedora 19 The tests reports some failures:

		=== systemtap Summary ===

# of expected passes		2
# of unexpected failures	1

Looking at the systemtap.log shows the tests starting up, but the 16-bit read/write
operations failing:

PASS: ./systemtap.base/deref.stp startup
PASS: ./systemtap.base/deref.stp load generation
systemtap ending probe
systemtap test failure - kread u16
systemtap test failure - kread const u16
systemtap test failure - kwrite u16


On x86_64 the tests all passed:

		=== systemtap Summary ===

# of expected passes		6


I modified the deref.stp script to see precisely what the arm machine was reasing and
writing.  It looked like it only got the lower 8-bits of the 16-bit quanity correct.  The
uper 8-bits are non-zero but it isn't clear where those values are coming from.

-Will