diff mbox

[v2] Pass semihosting exit code back to system.

Message ID 1416340889-4991-1-git-send-email-ilg@livius.net
State New
Headers show

Commit Message

Liviu Ionescu Nov. 18, 2014, 8:01 p.m. UTC
In order to run unit tests under semihosting, it is necessary to pass the
application exit code back to the system.

ARM defines only the code to be used for non-error application exit
(ADP_Stopped_ApplicationExit), all other codes should return non-zero
exit codes.

This patch checks if the application code passed via TARGET_SYS_EXIT is
ADP_Stopped_ApplicationExit, and return 0, otherwise return 1.

Signed-off-by: Liviu Ionescu <ilg@livius.net>
---
 target-arm/arm-semi.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

Comments

Peter Maydell Nov. 18, 2014, 8:08 p.m. UTC | #1
On 18 November 2014 20:01, Liviu Ionescu <ilg@livius.net> wrote:
> In order to run unit tests under semihosting, it is necessary to pass the
> application exit code back to the system.
>
> ARM defines only the code to be used for non-error application exit
> (ADP_Stopped_ApplicationExit), all other codes should return non-zero
> exit codes.
>
> This patch checks if the application code passed via TARGET_SYS_EXIT is
> ADP_Stopped_ApplicationExit, and return 0, otherwise return 1.
>
> Signed-off-by: Liviu Ionescu <ilg@livius.net>

Applied to target-arm.next, thanks.
(That's
https://git.linaro.org/people/peter.maydell/qemu-arm.git target-arm.next
which is a *rebasing* branch.)

-- PMM
Paolo Bonzini Nov. 18, 2014, 9:19 p.m. UTC | #2
On 18/11/2014 21:08, Peter Maydell wrote:
> Applied to target-arm.next, thanks.
> (That's
> https://git.linaro.org/people/peter.maydell/qemu-arm.git target-arm.next
> which is a *rebasing* branch.)

The isa-debugexit and testdev character device do not exit with exitcode
0, in order to distinguish an exit from a system power down; the
low-order bit is always 1.  The return values then should be 1 and 3
instead of 0 and 1.

Paolo
Liviu Ionescu Nov. 18, 2014, 9:43 p.m. UTC | #3
On 18 Nov 2014, at 23:19, Paolo Bonzini <pbonzini@redhat.com> wrote:

> The isa-debugexit and testdev character device do not exit with exitcode
> 0, in order to distinguish an exit from a system power down; the
> low-order bit is always 1.  The return values then should be 1 and 3
> instead of 0 and 1.

I have no experience with the isa-debugexit and testdev character device, and if there is any relationship between them and the semihosting code, but the original code always returned 0, regardless of the value passed by TARGET_SYS_EXIT, so I doubt there are any existing use cases that expect the semihosting code to return 1 or 3. 

in addition to passing trace messages, the main use of the semihosting calls is to run unit tests, and all automated testing infrastructures assumes the test executables to return 0/non-zero, as any well behaved POSIX processes.

I personally would not change this.

regards,

Liviu
Peter Maydell Nov. 18, 2014, 10:50 p.m. UTC | #4
On 18 November 2014 21:19, Paolo Bonzini <pbonzini@redhat.com> wrote:
> The isa-debugexit and testdev character device do not exit with exitcode
> 0, in order to distinguish an exit from a system power down; the
> low-order bit is always 1.  The return values then should be 1 and 3
> instead of 0 and 1.

Semihosting isn't a test device -- it's an implementation of an
ABI (mostly intended for almost-but-not-quite-bare-metal programs).
I'm pretty sure that exiting anything except 0 on a successful
exit request by the guest will break usage of QEMU in scenarios
like the gcc test suite.

thanks
-- PMM
Paolo Bonzini Nov. 19, 2014, 9:34 a.m. UTC | #5
On 18/11/2014 23:50, Peter Maydell wrote:
> On 18 November 2014 21:19, Paolo Bonzini <pbonzini@redhat.com> wrote:
>> The isa-debugexit and testdev character device do not exit with exitcode
>> 0, in order to distinguish an exit from a system power down; the
>> low-order bit is always 1.  The return values then should be 1 and 3
>> instead of 0 and 1.
> 
> Semihosting isn't a test device -- it's an implementation of an
> ABI (mostly intended for almost-but-not-quite-bare-metal programs).
> I'm pretty sure that exiting anything except 0 on a successful
> exit request by the guest will break usage of QEMU in scenarios
> like the gcc test suite.

Thanks!  (FWIW, I don't find the behavior of those devices very useful
either... just mentioning them for the sake of consistency).

Paolo
diff mbox

Patch

diff --git a/target-arm/arm-semi.c b/target-arm/arm-semi.c
index ebb5235..a8b83e6 100644
--- a/target-arm/arm-semi.c
+++ b/target-arm/arm-semi.c
@@ -58,6 +58,10 @@ 
 #define TARGET_SYS_HEAPINFO    0x16
 #define TARGET_SYS_EXIT        0x18
 
+/* ADP_Stopped_ApplicationExit is used for exit(0),
+ * anything else is implemented as exit(1) */
+#define ADP_Stopped_ApplicationExit     (0x20026)
+
 #ifndef O_BINARY
 #define O_BINARY 0
 #endif
@@ -551,8 +555,11 @@  uint32_t do_arm_semihosting(CPUARMState *env)
             return 0;
         }
     case TARGET_SYS_EXIT:
-        gdb_exit(env, 0);
-        exit(0);
+        /* ARM specifies only Stopped_ApplicationExit as normal
+         * exit, everything else is considered an error */
+        ret = (args == ADP_Stopped_ApplicationExit) ? 0 : 1;
+        gdb_exit(env, ret);
+        exit(ret);
     default:
         fprintf(stderr, "qemu: Unsupported SemiHosting SWI 0x%02x\n", nr);
         cpu_dump_state(cs, stderr, fprintf, 0);