Commit Message
On Friday, November 14 2014, Pedro Alves wrote:
> On 11/13/2014 12:18 AM, Sergio Durigan Junior wrote:
>> This patch implements the testcase for this fix. The test is very
>> simple: we just have to verify if the syscall number for each
>> architecture has different meanings. I chose to test i386 and x86_64
>> here, but it could be any other architecture supported by the "catch
>> syscall" command.
>
> This only works if the built GDB has these architectures configured
> in.
>
> E.g., an --enable-targets=all build on x86:
>
> (gdb) set architecture aarch64
> The target architecture is assumed to be aarch64
>
> while on a default x86 build:
>
> (gdb) set architecture aarch64
> Undefined item: "aarch64".
>
> From that, you can see that:
>
> (gdb) set architecture i386
>
> would fail on non-x86 builds that don't include x86 in --enable-targets=foo.
True, thanks for catching this. When I tested it in my machine, I made
some confusion and used the same GDB to test both scenarios. Sorry
about that.
>>
>> +proc test_catch_syscall_multi_target {} {
>
> Please make this "multi_arch". Let's leave "multi-target" for
> https://sourceware.org/gdb/wiki/MultiTarget.
Ouch, sorry. Fixed.
>
>> + global decimal binfile
>> +
>> + with_test_prefix "multiple targets" {
>> + clean_restart $binfile
>> +
>> + gdb_test "set architecture i386" \
>> + "The target architecture is assumed to be i386" \
>> + "set arch to i386"
>> +
>> + gdb_test "catch syscall 1" \
>> + "Catchpoint $decimal \\(syscall .exit. \\\[1\\\]\\)" \
>> + "insert catch syscall on syscall 1 -- exit on i386"
>> +
>> + gdb_test "set architecture i386:x86-64" \
>> + "The target architecture is assumed to be i386:x86-64" \
>> + "set arch to x86_64"
>> +
>> + gdb_test "catch syscall 1" \
>> + "Catchpoint $decimal \\(syscall .write. \\\[1\\\]\\)" \
>> + "insert catch syscall on syscall 1 -- exit on i386"
>
> The "exit on i386" part seems stale here.
Fixed.
> I think we should do something like this:
>
> if { [istarget "i*86-*-*"] || [istarget "x86_64-*-*"] } {
> set arch1 "i386"
> set syscall1 "exit"
> set arch2 "i386:x86-64"
> set syscall2 "write"
> } elseif { [istarget "powerpc-*-linux*"] || [istarget "powerpc64-*-linux*"] } {
> ...
> } elseif { [istarget "sparc-*-linux*"] && ![istarget "sparc64-*-linux*"] } {
> ...
> } elseif { [istarget "mips*-linux*"] } {
> ...
> } elseif { [istarget "arm*-linux*"] } {
> ...
> } elseif { [istarget "s390*-linux*"] } {
> ...
> } else {
> error "please port me"
> }
That was my first idea, but I was trying to avoid having yet another set
of if...elseif... on catch-syscall.exp. And due to my mistake when
testing this patch, I thought I wouldn't need to worry about that after
all. Silly me.
Anyway, here is the updated patch. WDYT?
Comments
On 11/18/2014 03:30 AM, Sergio Durigan Junior wrote:
> + with_test_prefix "multiple targets" {
s/multiple targets/multiple architectures/
Otherwise looks good. One question:
> + # We are not interested in loading any binary here,
This is fine, though ...
> and in
> + # some systems (PowerPC, for example), if we load a binary
> + # there is no way to set other architecture.
... eh. I wasn't aware of that. What does GDB say?
> + gdb_exit
> + gdb_start
Thanks,
Pedro Alves
On Tuesday, November 18 2014, Pedro Alves wrote:
> On 11/18/2014 03:30 AM, Sergio Durigan Junior wrote:
>> + with_test_prefix "multiple targets" {
>
> s/multiple targets/multiple architectures/
Sorry, fixed.
> Otherwise looks good. One question:
>
>> + # We are not interested in loading any binary here,
>
> This is fine, though ...
>
>> and in
>> + # some systems (PowerPC, for example), if we load a binary
>> + # there is no way to set other architecture.
>
> ... eh. I wasn't aware of that. What does GDB say?
Here is what I see on a Fedora 20 PPC64 box:
(gdb) set architecture
Requires an argument. Valid arguments are rs6000:6000, rs6000:rs1,
rs6000:rsc, rs6000:rs2, powerpc:common64, powerpc:common, powerpc:603,
powerpc:EC603e, powerpc:604, powerpc:403, powerpc:601, powerpc:620,
powerpc:630, powerpc:a35, powerpc:rs64ii, powerpc:rs64iii, powerpc:7400,
powerpc:e500, powerpc:e500mc, powerpc:e500mc64, powerpc:MPC8XX,
powerpc:750, powerpc:titan, powerpc:vle, powerpc:e5500, powerpc:e6500,
auto.
(gdb) set architecture powerpc:common
Architecture `powerpc:common' not recognized.
The target architecture is set automatically (currently
powerpc:common64)
On 11/18/2014 08:20 PM, Sergio Durigan Junior wrote:
>>> and in
>>> + # some systems (PowerPC, for example), if we load a binary
>>> + # there is no way to set other architecture.
>>
>> ... eh. I wasn't aware of that. What does GDB say?
>
> Here is what I see on a Fedora 20 PPC64 box:
>
> (gdb) set architecture
> Requires an argument. Valid arguments are rs6000:6000, rs6000:rs1,
> rs6000:rsc, rs6000:rs2, powerpc:common64, powerpc:common, powerpc:603,
> powerpc:EC603e, powerpc:604, powerpc:403, powerpc:601, powerpc:620,
> powerpc:630, powerpc:a35, powerpc:rs64ii, powerpc:rs64iii, powerpc:7400,
> powerpc:e500, powerpc:e500mc, powerpc:e500mc64, powerpc:MPC8XX,
> powerpc:750, powerpc:titan, powerpc:vle, powerpc:e5500, powerpc:e6500,
> auto.
> (gdb) set architecture powerpc:common
> Architecture `powerpc:common' not recognized.
Eh, this has the looks of a bug. At the very least, if on purpose for
some reason, the error string is misleading.
> The target architecture is set automatically (currently
> powerpc:common64)
Thanks,
Pedro Alves
On Wednesday, November 19 2014, Pedro Alves wrote:
> On 11/18/2014 08:20 PM, Sergio Durigan Junior wrote:
>
>>>> and in
>>>> + # some systems (PowerPC, for example), if we load a binary
>>>> + # there is no way to set other architecture.
>>>
>>> ... eh. I wasn't aware of that. What does GDB say?
>>
>> Here is what I see on a Fedora 20 PPC64 box:
>>
>> (gdb) set architecture
>> Requires an argument. Valid arguments are rs6000:6000, rs6000:rs1,
>> rs6000:rsc, rs6000:rs2, powerpc:common64, powerpc:common, powerpc:603,
>> powerpc:EC603e, powerpc:604, powerpc:403, powerpc:601, powerpc:620,
>> powerpc:630, powerpc:a35, powerpc:rs64ii, powerpc:rs64iii, powerpc:7400,
>> powerpc:e500, powerpc:e500mc, powerpc:e500mc64, powerpc:MPC8XX,
>> powerpc:750, powerpc:titan, powerpc:vle, powerpc:e5500, powerpc:e6500,
>> auto.
>> (gdb) set architecture powerpc:common
>> Architecture `powerpc:common' not recognized.
>
> Eh, this has the looks of a bug. At the very least, if on purpose for
> some reason, the error string is misleading.
Yeah, that's what I thought, though I still did not investigate it. I
plan to get to it soon, or file a bug. Meanwhile, I believe I can check
this series in, right?
On 11/19/2014 06:45 PM, Sergio Durigan Junior wrote:
> Yeah, that's what I thought, though I still did not investigate it. I
> plan to get to it soon, or file a bug. Meanwhile, I believe I can check
> this series in, right?
Yes.
(Please squash it to a single commit, to avoid breaking bisects.)
Thanks,
Pedro Alves
On Thursday, November 20 2014, Pedro Alves wrote:
> On 11/19/2014 06:45 PM, Sergio Durigan Junior wrote:
>> Yeah, that's what I thought, though I still did not investigate it. I
>> plan to get to it soon, or file a bug. Meanwhile, I believe I can check
>> this series in, right?
>
> Yes.
Thanks.
> (Please squash it to a single commit, to avoid breaking bisects.)
Sure; I was already going to do that, but forgot to mention.
Cheers,
@@ -311,6 +311,10 @@ proc do_syscall_tests {} {
# Testing the 'catch' syscall command during a restart of
# the inferior.
if [runto_main] then { test_catch_syscall_restarting_inferior }
+
+ # Testing if the 'catch syscall' command works when switching to
+ # different architectures on-the-fly (PR gdb/10737).
+ if [runto_main] then { test_catch_syscall_multi_arch }
}
proc test_catch_syscall_without_args_noxml {} {
@@ -372,6 +376,73 @@ proc test_catch_syscall_with_wrong_args_noxml {} {
}
}
+proc test_catch_syscall_multi_arch {} {
+ global decimal binfile
+
+ if { [istarget "i*86-*-*"] || [istarget "x86_64-*-*"] } {
+ set arch1 "i386"
+ set arch2 "i386:x86-64"
+ set syscall1_name "exit"
+ set syscall2_name "write"
+ set syscall_number 1
+ } elseif { [istarget "powerpc-*-linux*"] \
+ || [istarget "powerpc64-*-linux*"] } {
+ set arch1 "powerpc:common"
+ set arch2 "powerpc:common64"
+ set syscall1_name "openat"
+ set syscall2_name "unlinkat"
+ set syscall_number 286
+ } elseif { [istarget "sparc-*-linux*"] \
+ || [istarget "sparc64-*-linux*"] } {
+ set arch1 "sparc"
+ set arch2 "sparc:v9"
+ set syscall1_name "setresuid32"
+ set syscall2_name "setresuid"
+ set syscall_number 108
+ } elseif { [istarget "mips*-linux*"] } {
+ # MIPS does not use the same numbers for syscalls on 32 and 64
+ # bits.
+ verbose "Not testing MIPS for multi-arch syscall support"
+ return
+ } elseif { [istarget "arm*-linux*"] } {
+ # catch syscall supports only 32-bit ARM for now.
+ verbose "Not testing ARM for multi-arch syscall support"
+ return
+ } elseif { [istarget "s390*-linux*"] } {
+ set arch1 ""
+ set arch2 "s390:64-bit"
+ set syscall1_name "_newselect"
+ set syscall2_name "select"
+ set syscall_number 142
+ }
+
+ with_test_prefix "multiple targets" {
+ # We are not interested in loading any binary here, and in
+ # some systems (PowerPC, for example), if we load a binary
+ # there is no way to set other architecture.
+ gdb_exit
+ gdb_start
+
+ gdb_test "set architecture $arch1" \
+ "The target architecture is assumed to be $arch1" \
+ "set arch to $arch1"
+
+ gdb_test "catch syscall $syscall_number" \
+ "Catchpoint $decimal \\(syscall .${syscall1_name}. \\\[${syscall_number}\\\]\\)" \
+ "insert catch syscall on syscall $syscall_number -- $syscall1_name on $arch1"
+
+ gdb_test "set architecture $arch2" \
+ "The target architecture is assumed to be $arch2" \
+ "set arch to $arch2"
+
+ gdb_test "catch syscall $syscall_number" \
+ "Catchpoint $decimal \\(syscall .${syscall2_name}. \\\[${syscall_number}\\\]\\)" \
+ "insert catch syscall on syscall $syscall_number -- $syscall2_name on $arch2"
+
+ clean_restart $binfile
+ }
+}
+
proc do_syscall_tests_without_xml {} {
# Make sure GDB doesn't load the syscalls xml from the system data
# directory.