On 29.11.21 15:21, Sven Schnelle wrote:

Hi,

Yafang Shao [off-list ref] writes:

quoted

On Mon, Nov 29, 2021 at 6:13 PM Sven Schnelle [off-list ref] wrote:

quoted

quoted

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 78c351e35fec..cecd4806edc6 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h

@@ -274,8 +274,13 @@ struct task_group;

 #define get_current_state()  READ_ONCE(current->__state)

-/* Task command name length: */
-#define TASK_COMM_LEN                        16
+/*
+ * Define the task command name length as enum, then it can be visible to
+ * BPF programs.
+ */
+enum {
+     TASK_COMM_LEN = 16,
+};

This breaks the trigger-field-variable-support.tc from the ftrace test
suite at least on s390:

echo
'hist:keys=next_comm:wakeup_lat=common_timestamp.usecs-$ts0:onmatch(sched.sched_waking).wakeup_latency($wakeup_lat,next_pid,sched.sched_waking.prio,next_comm)
if next_comm=="ping"'
linux/tools/testing/selftests/ftrace/test.d/trigger/inter-event/trigger-field-variable-support.tc: line 15: echo: write error: Invalid argument

I added a debugging line into check_synth_field():

[   44.091037] field->size 16, hist_field->size 16, field->is_signed 1, hist_field->is_signed 0

Note the difference in the signed field.

Hi Sven,

Thanks for the report and debugging!
Seems we should explicitly define it as signed ?
Could you pls. help verify it?

diff --git a/include/linux/sched.h b/include/linux/sched.h
index cecd4806edc6..44d36c6af3e1 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h

@@ -278,7 +278,7 @@ struct task_group;
  * Define the task command name length as enum, then it can be visible to
  * BPF programs.
  */
-enum {
+enum SignedEnum {
        TASK_COMM_LEN = 16,
 };

Umm no. What you're doing here is to define the name of the enum as
'SignedEnum'. This doesn't change the type. I think before C++0x you
couldn't force an enum type.

I think there are only some "hacks" to modify the type with GCC. For
example, with "__attribute__((packed))" we can instruct GCC to use the
smallest type possible for the defined enum values.

I think with some fake entries one can eventually instruct GCC to use an
unsigned type in some cases:

https://stackoverflow.com/questions/14635833/is-there-a-way-to-make-an-enum-unsigned-in-the-c90-standard-misra-c-2004-compl

enum {
	TASK_COMM_LEN = 16,
	TASK_FORCE_UNSIGNED = 0x80000000,
};

Haven't tested it, though, and I'm not sure if we should really do that
... :)

-- 
Thanks,

David / dhildenb