On Wed, Apr 3, 2019 at 5:12 PM Arnaldo Carvalho de Melo
<arnaldo.melo@xxxxxxxxx> wrote:
>
> Em Wed, Apr 03, 2019 at 09:07:47PM -0300, Arnaldo Carvalho de Melo escreveu:
> > Em Wed, Apr 03, 2019 at 02:50:14PM -0700, Andrii Nakryiko escreveu:
> > > On Wed, Apr 3, 2019 at 2:36 PM Arnaldo Carvalho de Melo <arnaldo.melo@xxxxxxxxx> wrote:
> > > > Em Wed, Apr 03, 2019 at 06:06:18PM -0300, Arnaldo Carvalho de Melo escreveu:
> > > > [acme@quaco pahole]$ pahole -F btf --packable vmlinux | cut -f 1 | while read struct ; do pahole -F btf -C $struct --reorganize vmlinux | grep BRAIN ; done
> > > > /* BRAIN FART ALERT! 200 bytes != 176 (member bytes) + 0 (member bits) + 80 (byte holes) + 0 (bit holes), diff = -448 bits */
> > > > /* BRAIN FART ALERT! 760 bytes != 756 (member bytes) + 0 (member bits) + 4 (byte holes) + 0 (bit holes), diff = -32 bits */
> > > > /* BRAIN FART ALERT! 72 bytes != 28 (member bytes) + 0 (member bits) + 100 (byte holes) + 0 (bit holes), diff = -448 bits */
> > > > /* BRAIN FART ALERT! 40 bytes != 36 (member bytes) + 0 (member bits) + 44 (byte holes) + 0 (bit holes), diff = -352 bits */
> > > > /* BRAIN FART ALERT! 184 bytes != 182 (member bytes) + 0 (member bits) + 6 (byte holes) + 0 (bit holes), diff = -32 bits */
> > > > [acme@quaco pahole]$ pahole -F btf --packable vmlinux | wc -l
> > > > 503
> > > > [acme@quaco pahole]$
> > > >
> > > > 1% getting closer...
> > > >
> > > > [acme@quaco pahole]$ pahole -F btf --reorganize -C netns_frags vmlinux
> > > > struct netns_frags {
> > > > long int high_thresh; /* 0 8 */
> > > > long int low_thresh; /* 8 8 */
> > > > int timeout; /* 16 4 */
> > > > int max_dist; /* 20 4 */
> > > > struct inet_frags * f; /* 24 8 */
> > > > atomic_long_t mem; /* 32 8 */
> > > >
> > > > /* XXX 24 bytes hole, try to pack */
> > > >
> > > > /* --- cacheline 1 boundary (64 bytes) --- */
> > > > struct rhashtable rhashtable; /* 64 136 */
> > > >
> > > > /* XXX 56 bytes hole, try to pack */
> > >
> > > This one is strange, there should be no hole. Will take a look.
> >
> > So I've pushed everything to my master branch, removed the last patch of
> > your latest series, and have some patches I've worked on to take
> > advantage of DW_AT_alignment, and the ones at the end are the 1% of the
> > ones that currently are considered --packable.
> >
> > The problem seems to be related to the last member hole, which should
> > probably be just trowed away at the end, together with fixing up the
> > class size.
>
> Mid-air colision:
heh :)
>
> I'm assuming the first three patches are the same as before, with the
correct. only patch 4 is different (and seems to be fixing the rest of
BFAs, yay!).
Thanks for working on this actively!
> 4th applied (I had already dropped the 4th of the v1 series), all seems
> clear:
>
> [acme@quaco pahole]$ cat /tmp/BFA | while read struct ; do pahole --reorganize -C $struct -F btf vmlinux ; done
> struct netns_frags {
> long int high_thresh; /* 0 8 */
> long int low_thresh; /* 8 8 */
> int timeout; /* 16 4 */
> int max_dist; /* 20 4 */
> struct inet_frags * f; /* 24 8 */
> atomic_long_t mem; /* 32 8 */
>
> /* XXX 24 bytes hole, try to pack */
>
> /* --- cacheline 1 boundary (64 bytes) --- */
> struct rhashtable rhashtable; /* 64 136 */
>
> /* size: 200, cachelines: 4, members: 7 */
> /* sum members: 176, holes: 1, sum holes: 24 */
> /* last cacheline: 8 bytes */
> }; /* saved 120 bytes and 1 cacheline! */
> struct worker_pool {
> spinlock_t lock; /* 0 4 */
> int cpu; /* 4 4 */
> int node; /* 8 4 */
> int id; /* 12 4 */
> unsigned int flags; /* 16 4 */
> int refcnt; /* 20 4 */
> long unsigned int watchdog_ts; /* 24 8 */
> struct list_head worklist; /* 32 16 */
> int nr_workers; /* 48 4 */
> int nr_idle; /* 52 4 */
> struct list_head idle_list; /* 56 16 */
> /* --- cacheline 1 boundary (64 bytes) was 8 bytes ago --- */
> struct timer_list idle_timer; /* 72 40 */
>
> /* XXX last struct has 4 bytes of padding */
>
> struct timer_list mayday_timer; /* 112 40 */
>
> /* XXX last struct has 4 bytes of padding */
>
> /* --- cacheline 2 boundary (128 bytes) was 24 bytes ago --- */
> struct hlist_head busy_hash[64]; /* 152 512 */
> /* --- cacheline 10 boundary (640 bytes) was 24 bytes ago --- */
> struct worker * manager; /* 664 8 */
> struct list_head workers; /* 672 16 */
> struct completion * detach_completion; /* 688 8 */
> struct ida worker_ida; /* 696 16 */
> /* --- cacheline 11 boundary (704 bytes) was 8 bytes ago --- */
> struct workqueue_attrs * attrs; /* 712 8 */
> struct hlist_node hash_node; /* 720 16 */
> struct callback_head rcu; /* 736 16 */
> atomic_t nr_running; /* 752 4 */
>
> /* size: 760, cachelines: 12, members: 22 */
> /* padding: 4 */
> /* paddings: 2, sum paddings: 8 */
> /* last cacheline: 56 bytes */
> }; /* saved 72 bytes and 1 cacheline! */
> struct sbitmap_word {
> long unsigned int depth; /* 0 8 */
> spinlock_t swap_lock; /* 8 4 */
>
> /* XXX 4 bytes hole, try to pack */
>
> long unsigned int cleared; /* 16 8 */
>
> /* XXX 40 bytes hole, try to pack */
>
> /* --- cacheline 1 boundary (64 bytes) --- */
> long unsigned int word; /* 64 8 */
>
> /* size: 72, cachelines: 2, members: 4 */
> /* sum members: 28, holes: 2, sum holes: 44 */
> /* last cacheline: 8 bytes */
> }; /* saved 120 bytes and 1 cacheline! */
> struct ptr_ring {
> int producer; /* 0 4 */
> spinlock_t producer_lock; /* 4 4 */
> spinlock_t consumer_lock; /* 8 4 */
> int consumer_tail; /* 12 4 */
> void * * queue; /* 16 8 */
> int batch; /* 24 4 */
> int size; /* 28 4 */
> int consumer_head; /* 32 4 */
>
> /* size: 40, cachelines: 1, members: 8 */
> /* padding: 4 */
> /* last cacheline: 40 bytes */
> }; /* saved 152 bytes and 2 cachelines! */
> struct acpi_ec {
> acpi_handle handle; /* 0 8 */
> u32 gpe; /* 8 4 */
> bool global_lock; /* 12 1 */
> bool busy_polling; /* 13 1 */
>
> /* XXX 2 bytes hole, try to pack */
>
> long unsigned int command_addr; /* 16 8 */
> long unsigned int data_addr; /* 24 8 */
> long unsigned int flags; /* 32 8 */
> long unsigned int reference_count; /* 40 8 */
> struct mutex mutex; /* 48 32 */
> /* --- cacheline 1 boundary (64 bytes) was 16 bytes ago --- */
> wait_queue_head_t wait; /* 80 24 */
> struct list_head list; /* 104 16 */
> struct transaction * curr; /* 120 8 */
> /* --- cacheline 2 boundary (128 bytes) --- */
> spinlock_t lock; /* 128 4 */
> unsigned int polling_guard; /* 132 4 */
> struct work_struct work; /* 136 32 */
> long unsigned int timestamp; /* 168 8 */
> long unsigned int nr_pending_queries; /* 176 8 */
>
> /* size: 184, cachelines: 3, members: 17 */
> /* sum members: 182, holes: 1, sum holes: 2 */
> /* last cacheline: 56 bytes */
> }; /* saved 16 bytes and 1 cacheline! */
> [acme@quaco pahole]$
