selftests/x86: Support Atom for syscall_arg_fault test
authorTong Bo <bo.tong@intel.com>
Fri, 19 Apr 2019 07:10:55 +0000 (15:10 +0800)
committerShuah Khan <skhan@linuxfoundation.org>
Tue, 21 May 2019 15:24:30 +0000 (09:24 -0600)
commita20d452a2db7e092a9b214976306a36857b88ade
treea606e24366f9598ee93da9a9a75c88acd7ab3121
parenta188339ca5a396acc588e5851ed7e19f66b0ebd9
selftests/x86: Support Atom for syscall_arg_fault test

Atom-based CPUs trigger stack fault when invoke 32-bit SYSENTER instruction
with invalid register values. So we also need SIGBUS handling in this case.

Following is assembly when the fault exception happens.

(gdb) disassemble $eip
Dump of assembler code for function __kernel_vsyscall:
   0xf7fd8fe0 <+0>:     push   %ecx
   0xf7fd8fe1 <+1>:     push   %edx
   0xf7fd8fe2 <+2>:     push   %ebp
   0xf7fd8fe3 <+3>:     mov    %esp,%ebp
   0xf7fd8fe5 <+5>:     sysenter
   0xf7fd8fe7 <+7>:     int    $0x80
=> 0xf7fd8fe9 <+9>:     pop    %ebp
   0xf7fd8fea <+10>:    pop    %edx
   0xf7fd8feb <+11>:    pop    %ecx
   0xf7fd8fec <+12>:    ret
End of assembler dump.

According to Intel SDM, this could also be a Stack Segment Fault(#SS, 12),
except a normal Page Fault(#PF, 14). Especially, in section 6.9 of Vol.3A,
both stack and page faults are within the 10th(lowest priority) class, and
as it said, "exceptions within each class are implementation-dependent and
may vary from processor to processor". It's expected for processors like
Intel Atom to trigger stack fault(SIGBUS), while we get page fault(SIGSEGV)
from common Core processors.

Signed-off-by: Tong Bo <bo.tong@intel.com>
Acked-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
tools/testing/selftests/x86/syscall_arg_fault.c