[PATCH/RFA] Testsuite improvements for non-decr_pc_after_break i386

[PATCH/RFA] Testsuite improvements for non-decr_pc_after_break i386

[Mark Kettenis]

Toying with PR_BPTADJ on Solaris uncovered a problem with the
i386-prologue.exp test.  Here's a patch that makes sure we skip the
hardcoded breakpoints before continueing.

Since this isn't completely trivial, I'd like another pair of eyes to
look at this before I check this in.

Tested on i386-unknown-freebsd4.7 and i386-unknown-solaris2.9 (with
PR_BPTADJ patch).

Mark


Index: testsuite/ChangeLog
from  Mark Kettenis  <kettenis@gnu.org>

	* gdb.arch/i386-prologue.exp (skip_breakpoint): New function.  Use
	it to skip the breakpoints encoded in the inline assembly.

Index: testsuite/gdb.arch/i386-prologue.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.arch/i386-prologue.exp,v
retrieving revision 1.9
diff -u -p -r1.9 i386-prologue.exp
--- testsuite/gdb.arch/i386-prologue.exp 8 Aug 2004 22:15:08 -0000 1.9
+++ testsuite/gdb.arch/i386-prologue.exp 14 Aug 2004 20:03:14 -0000
@@ -50,6 +50,22 @@ if { [gdb_compile "${srcdir}/${subdir}/$
     gdb_suppress_entire_file "Testcase compile failed, so all tests in this file will automatically fail."
 }
 
+
+# The code used by the tests here encodes some breakpoints by using
+# inline assembler.  This will generate a SIGTRAP which will be caught
+# by GDB.  At that point the instruction pointer will point at the
+# next instruction, and execution can continue without any problems.
+# Some systems however (QNX Neutrino, Solaris) will adjust the
+# instruction pointer to point at the breakpoint instruction instead.
+# On these systems we cannot continue unless we skip it.  This
+# procedure takes care of that.
+
+proc skip_breakpoint { msg } {
+    gdb_test "if (*(unsigned char *)\$pc == 0xcc)\nset \$pc = \$pc + 1\nend" \
+	    "" "skip breakpoint in ${msg}"
+}
+
+
 gdb_exit
 gdb_start
 gdb_reinitialize_dir $srcdir/$subdir
@@ -67,6 +83,8 @@ if ![runto_main] then {
 
 gdb_test "continue" "Program received signal SIGTRAP.*" "continue to standard"
 
+skip_breakpoint standard
+
 gdb_test "backtrace 10" \
 	"#0\[ \t\]*$hex in standard.*\r\n#1\[ \t\]*$hex in main.*" \
 	"backtrace in standard"
@@ -79,6 +97,8 @@ gdb_test "info frame" \
 
 gdb_test "continue" "Program received signal SIGTRAP.*" "continue to gdb1253"
 
+skip_breakpoint gdb1253
+
 gdb_test "backtrace 10" \
 	"#0\[ \t\]*$hex in gdb1253.*\r\n#1\[ \t\]*$hex in main.*" \
 	"backtrace in gdb1253"
@@ -91,6 +111,8 @@ gdb_test "info frame" \
 
 gdb_test "continue" "Program received signal SIGTRAP.*" "continue to gdb1718"
 
+skip_breakpoint gdb1718
+
 gdb_test "backtrace 10" \
 	"#0\[ \t\]*$hex in gdb1718.*\r\n#1\[ \t\]*$hex in main.*" \
 	"backtrace in gdb1718"
@@ -104,6 +126,8 @@ gdb_test "info frame" \
 
 gdb_test "continue" "Program received signal SIGTRAP.*" "continue to gdb1338"
 
+skip_breakpoint gdb1338
+
 gdb_test "backtrace 10" \
 	"#0\[ \t\]*$hex in gdb1338.*\r\n#1\[ \t\]*$hex in main.*" \
 	"backtrace in gdb1338"

Re: [PATCH/RFA] Testsuite improvements for non-decr_pc_after_break i386

[Michael Chastain]

This patch is approved.

I proofread it.  I also tested it on native i686-pc-linux-gnu,
which is a no-adjust platform ($eip points after the breakpoint).
I turned it into an adjust platform with this:

    gdb_test "set \$pc = \$pc - 1" "" "decr pc"

... and the logic works.

Michael C

===

Index: testsuite/ChangeLog
from  Mark Kettenis  <kettenis@gnu.org>

	* gdb.arch/i386-prologue.exp (skip_breakpoint): New function.  Use
	it to skip the breakpoints encoded in the inline assembly.

Re: [PATCH/RFA] Testsuite improvements for non-decr_pc_after_break i386

[Daniel Jacobowitz]

On Sun, Aug 15, 2004 at 03:04:10AM -0400, Michael Chastain wrote:
> This patch is approved.
> 
> I proofread it.  I also tested it on native i686-pc-linux-gnu,
> which is a no-adjust platform ($eip points after the breakpoint).
> I turned it into an adjust platform with this:
> 
>     gdb_test "set \$pc = \$pc - 1" "" "decr pc"
> 
> ... and the logic works.

FYI, you've got the nomenclature backwards - adjust-pc-after-break
means "GDB must adjust the PC after break to figure out where the
breakpoint was".

-- 
Daniel Jacobowitz

Re: [PATCH/RFA] Testsuite improvements for non-decr_pc_after_break i386

[Michael Chastain]

mec> I proofread it.  I also tested it on native i686-pc-linux-gnu,
mec> which is a no-adjust platform ($eip points after the breakpoint).

drow> FYI, you've got the nomenclature backwards - adjust-pc-after-break
drow> means "GDB must adjust the PC after break to figure out where the
drow> breakpoint was".

Gah.  I knew that!  Let me retry:

native i686-pc-linux-gnu has $eip pointing after the breakpoint,
so I threw in "set $pc [expr $pc - 1]" to make it look like the
program counter was pointing at the breakpoint instruction.