Re: [PATCH v5 08/12] perf annotate: Add 'T' hot key to toggle data type display

[Namhyung Kim]

On Wed, Aug 20, 2025 at 06:13:15PM -0300, Arnaldo Carvalho de Melo wrote:
> On Fri, Aug 15, 2025 at 08:16:31PM -0700, Namhyung Kim wrote:
> > Support data type display with a key press so that users can toggle the
> > output dynamically on TUI.  Also display "[Type]" in the title line if
> > it's enabled.
> 
> Testing here I see:
> 
>    0.81 │       mov     0x9c(%r8),%r11d         # data-type: struct cfs_rq +0x9c
> 
> If I ask for source code by pressing 's':
> 
>         │     delta += sa->period_contrib;                                                                                                                              ▒
>    0.97 │       mov     0x9c(%r8),%r11d         # data-type: struct cfs_rq +0x9c
> 
> So it is the 'period_contrib' field of 'struct cfs_rq', humm, not:
> 
> root@number:~# pahole -E --hex cfs_rq | grep 0x9c -B 10 -A10
> 	struct sched_entity *      next;                                                 /*  0x60   0x8 */
> 
> 	/* XXX 24 bytes hole, try to pack */
> 
> 	/* --- cacheline 2 boundary (128 bytes) --- */
> 	struct sched_avg {
> 		/* typedef u64 -> __u64 */ long long unsigned int last_update_time;      /*  0x80   0x8 */
> 		/* typedef u64 -> __u64 */ long long unsigned int load_sum;              /*  0x88   0x8 */
> 		/* typedef u64 -> __u64 */ long long unsigned int runnable_sum;          /*  0x90   0x8 */
> 		/* typedef u32 -> __u32 */ unsigned int       util_sum;                  /*  0x98   0x4 */
> 		/* typedef u32 -> __u32 */ unsigned int       period_contrib;            /*  0x9c   0x4 */
> 		long unsigned int  load_avg;                                             /*  0xa0   0x8 */
> 		long unsigned int  runnable_avg;                                         /*  0xa8   0x8 */
> 		long unsigned int  util_avg;                                             /*  0xb0   0x8 */
> 		unsigned int       util_est;                                             /*  0xb8   0x4 */
> 	} avg __attribute__((__aligned__(64))); /*  0x80  0x40 */
> 
> 	/* XXX last struct has 4 bytes of padding */
> 
> 	/* --- cacheline 3 boundary (192 bytes) --- */
> 	struct {
> root@number:~#
> 
> So it is in a subtype and probably this is an improvement to be made,
> right?

It seems %r8 contains a pointer to cfs_rq and there's a variable for sa
(sched_avg) but it actually uses the same register as it's embeded.  I
think DWARF has correct location expression to find the containing data
type and the output looks ok.  In general, I'd prefer seeing outer types.

Thanks,
Namhyung