lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Wed, 13 May 2015 16:09:18 +0200
From:	Denys Vlasenko <dvlasenk@...hat.com>
To:	Ingo Molnar <mingo@...nel.org>
CC:	Linus Torvalds <torvalds@...ux-foundation.org>,
	Thomas Graf <tgraf@...g.ch>,
	"David S. Miller" <davem@...emloft.net>,
	Bart Van Assche <bvanassche@....org>,
	Peter Zijlstra <peterz@...radead.org>,
	David Rientjes <rientjes@...gle.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Oleg Nesterov <oleg@...hat.com>,
	"Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] force inlining of spinlock ops

On 05/13/2015 12:43 PM, Ingo Molnar wrote:
> 
> * Denys Vlasenko <dvlasenk@...hat.com> wrote:
> 
>> On 05/13/2015 12:17 PM, Ingo Molnar wrote:
>>>>> In any case, the interesting measurement would not be -Os comparisons 
>>>>> (which causes GCC to be too crazy), but to see the size effect of your 
>>>>> _patch_ that always-inlines spinlock ops, on plain defconfig and on 
>>>>> defconfig-Os.
>>>>
>>>> Here it is:
>>>>
>>>>     text    data     bss      dec    hex filename
>>>> 12335864 1746152 1081344 15163360 e75fe0 vmlinuxO2.before
>>>> 12335930 1746152 1081344 15163426 e76022 vmlinux
>>>
>>> Hm, that's a (small) size increase on O2.
>>>
>>> That might be a net positive though: because now we've eliminated 
>>> quite a few function calls. Do we know which individual functions 
>>> bloat and which debloat?
>>
>>>>     text    data     bss      dec    hex filename
>>>> 10373764 1684200 1077248 13135212 c86d6c vmlinuxOs.before
>>>> 10363621 1684200 1077248 13125069 c845cd vmlinux
>>>
>>> A decrease - which gets exploded on allyesconfig.
>>>
>>> So as long as the -O2 case does not get hurt we can do -Os fixes.
>>>
>>> I think this needs a bit more work to ensure that the O2 case is a 
>>> net win.
>>
>> I think O2 difference is just noise: with -O2 gcc is far less prone 
>> to bogus deinlining, my patch should have negligible effect. And 
>> effect is indeed negligible: +70 bytes on 12 megabytes.
> 
> So the patch force-inlines about a dozen locking APIs:
> 
>  - Some of those decrease the defconfig kernel size.
>    Which ones and by how much?
> 
>  - Some of those increase the defconfig kernel size.
>    Which ones and by how much?
> 
> We only know that the net effect is +70 bytes. Does that come out of:
> 
>  - large fluctuations such as -1000-1000+1000+1070, which happens to 
>    net out into a small net number?
> 
>  - or does it come from much smaller fluctuations?
> 
> So to make an informed decision we need to know those details.

Fair enough. Let's investigate.

I produced a list of functions with their sizes from each vmlinux,
and diffed them:

$ nm --size-sort vmlinux | sed 's/\.[0-9]*.*/.NNN/' >vmlinux.nm
$ nm --size-sort vmlinuxO2.before | sed 's/\.[0-9]*.*/.NNN/' >vmlinuxO2.before.nm
$ diff -u vmlinuxO2.before.nm vmlinux.nm | grep -v '^[ @]' >vmlinux.nm.dif

I see the following:

- spin_[un]lock_foo's are gone as expected.
- some other functions got spuriously deinlined
  (such as __raw_spin_unlock).
- yet other functions which were spuriously deinlined before,
  now aren't deinlined (such as nf_conntrack_put).

--- vmlinuxO2.before.nm	2015-05-13 15:46:37.147058665 +0200
+++ vmlinux.nm	2015-05-13 15:46:26.079086233 +0200
+0000000000000009 t ipc_unlock_object
+0000000000000009 t __raw_spin_unlock
+0000000000000009 t __raw_spin_unlock
-0000000000000009 t spin_unlock
-000000000000000b t spin_lock
-000000000000000b t spin_lock
-000000000000000b t spin_unlock_irqrestore
+000000000000000d t task_unlock
+0000000000000011 t arch_spin_is_locked
+0000000000000013 t double_unlock_hb
-000000000000001c t nf_conntrack_put
-000000000000001d t ctnetlink_done_list
+000000000000001e t unix_state_double_unlock
+0000000000000025 t check_and_drop
-0000000000000025 t hugetlbfs_inc_free_inodes.NNN
-0000000000000025 t sem_lock.NNN
-0000000000000027 t check_and_drop

- many other functions now have slightly different sizes.
  For example:
  ext4_mb_pa_callback grew from 0x28 to 0x2a bytes (+2 bytes)
  hid_alloc_report_buf shrank from 0x2a to 0x28 bytes (-2 bytes)

-0000000000000028 t ext4_mb_pa_callback
+0000000000000028 T hid_alloc_report_buf
+0000000000000028 t nf_conntrack_double_unlock
+0000000000000029 t do_shm_rmid
+000000000000002a t ext4_mb_pa_callback
-000000000000002a T hid_alloc_report_buf

Let's take a look at ext4_mb_pa_callback.
The difference stems from gcc choosing a different scratch register:

<ext4_mb_pa_callback>:                                          <ext4_mb_pa_callback>:
8b 47 14                mov    0x14(%rdi),%eax                  8b 47 14                mov    0x14(%rdi),%eax
48 8d 77 e0             lea    -0x20(%rdi),%rsi                 48 8d 77 e0             lea    -0x20(%rdi),%rsi
85 c0                   test   %eax,%eax                        85 c0                   test   %eax,%eax
75 1b                   jne    <ext4_mb_pa_callback+0x26>       75 19                   jne    <ext4_mb_pa_callback+0x24>
44 8b 57 18             mov    0x18(%rdi),%r10d                 8b 7f 18                mov    0x18(%rdi),%edi
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
45 85 d2                test   %r10d,%r10d                      85 ff                   test   %edi,%edi
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
74 14                   je     <ext4_mb_pa_callback+0x28>       74 14                   je     <ext4_mb_pa_callback+0x26>
55                      push   %rbp                             55                      push   %rbp
48 8b 3d e4 b8 eb 00    mov    0xebb8e4(%rip),%rdi              48 8b 3d 96 c0 eb 00    mov    0xebc096(%rip),%rdi
48 89 e5                mov    %rsp,%rbp                        48 89 e5                mov    %rsp,%rbp
e8 8c 3f f4 ff          callq  <kmem_cache_free>                e8 fe 46 f4 ff          callq  <kmem_cache_free>
5d                      pop    %rbp                             5d                      pop    %rbp
c3                      retq                                    c3                      retq
0f 0b                   ud2                                     0f 0b                   ud2
0f 0b                   ud2                                     0f 0b                   ud2

Working with r10 requires REX prefix in two underlined instructions,
which caused function to grow by 2 bytes.

This is not in any way related to my patch.

I think this confirms my hypothesis that the difference of +70 bytes is noise.

The remainder of the diff is below:

-000000000000002b t do_shm_rmid
+000000000000002b T jbd2_journal_ack_err
-000000000000002c t ctnetlink_done
+000000000000002d t ctnetlink_done_list
-000000000000002d T jbd2_journal_ack_err
-000000000000002e t amd_iommu_stats_add
-000000000000002e t hid_parser_reserved
+0000000000000030 t amd_iommu_stats_add
+0000000000000030 t hid_parser_reserved
-0000000000000034 T acpi_ec_unblock_transactions
+0000000000000034 T usb_hub_to_struct_hub
-0000000000000036 t acpi_ec_stopped
+0000000000000036 T acpi_ec_unblock_transactions
-0000000000000037 T completion_done
-000000000000003c T usb_hub_to_struct_hub
+000000000000003e T completion_done
-000000000000003e t ehci_clear_tt_buffer.NNN
+000000000000003f t acpi_ec_gpe_handler
+000000000000003f t ehci_clear_tt_buffer.NNN
-0000000000000041 t acpi_ec_gpe_handler
+0000000000000042 T usb_wakeup_notification
-0000000000000044 T acpi_boot_ec_enable
-0000000000000044 T usb_wakeup_notification
+0000000000000045 T acpi_boot_ec_enable
-0000000000000046 t hugetlbfs_destroy_inode
+0000000000000047 t ctnetlink_done
+0000000000000049 T efivar_entry_iter
+000000000000004a t turn_on_io_watchdog
-000000000000004b T efivar_entry_iter
-000000000000004c t init_fat_fs
-000000000000004c t turn_on_io_watchdog
+000000000000004e t init_fat_fs
+0000000000000054 T shm_destroy_orphaned
-0000000000000055 t sem_wait_array
+0000000000000056 t dm_dirty_log_init
-0000000000000056 T shm_destroy_orphaned
-0000000000000057 t dm_dirty_log_init
+0000000000000058 t fat_evict_inode
+0000000000000058 t mirror_available
-0000000000000059 t mirror_available
-000000000000005a t fat_evict_inode
+000000000000005b t sem_wait_array
-000000000000005c T blk_pre_runtime_suspend
+000000000000005c t free_dev_data
-000000000000005d t free_dev_data
+000000000000005f t hugetlbfs_destroy_inode
+0000000000000064 T blk_pre_runtime_suspend
+0000000000000071 t metadata_show
+0000000000000071 t update_changeattr
-0000000000000072 t update_changeattr
-0000000000000073 t hugetlbfs_alloc_inode
-0000000000000076 t fat_calc_dir_size
+0000000000000079 t fat_calc_dir_size
-0000000000000079 t metadata_show
+000000000000007c t rtc_dev_open
+000000000000007d t hugetlbfs_alloc_inode
-000000000000007e t ptrace_unfreeze_traced.NNN
-000000000000007f t rtc_dev_open
+0000000000000080 t ptrace_unfreeze_traced.NNN
-0000000000000083 T vfs_whiteout
+0000000000000086 T vfs_whiteout
-0000000000000087 t read_disk_sb.NNN
+0000000000000089 T assert_forcewakes_inactive
+000000000000008a t read_disk_sb.NNN
-000000000000008b T assert_forcewakes_inactive
-000000000000008e t cache_ioctl.NNN
+0000000000000093 T usbhid_close
-0000000000000094 t nf_nat_init
+0000000000000095 t nf_nat_init
+0000000000000096 t cache_ioctl.NNN
+0000000000000096 t __lock_sock
-0000000000000096 T usbhid_close
-0000000000000097 t __lock_sock
+0000000000000098 T jbd2_trans_will_send_data_barrier
-000000000000009a T jbd2_trans_will_send_data_barrier
+000000000000009b T drm_crtc_vblank_reset
-000000000000009d T drm_crtc_vblank_reset
+00000000000000a6 T md_setup_cluster
+00000000000000aa t hugetlbfs_statfs
+00000000000000ab t uhci_hcd_init
-00000000000000ac t hugetlbfs_statfs
-00000000000000ac t uhci_hcd_init
+00000000000000b1 t gro_cell_poll
-00000000000000b2 t gro_cell_poll
+00000000000000b2 t nf_conntrack_double_lock
-00000000000000b4 T md_setup_cluster
-00000000000000b4 t pps_init
+00000000000000b5 t pps_init
-00000000000000bb T svc_authenticate
+00000000000000bd T svc_authenticate
-00000000000000be t ehci_poll_ASS
+00000000000000c1 t ehci_poll_PSS
+00000000000000c1 t loop_queue_write_work
+00000000000000c5 T usb_disable_lpm
+00000000000000c6 t ehci_poll_ASS
-00000000000000c7 t nf_conntrack_double_lock
-00000000000000c7 T usb_disable_lpm
-00000000000000c9 t ehci_poll_PSS
+00000000000000cb t xfrm_del_sa
+00000000000000cd t snd_timer_user_ccallback
-00000000000000d0 t xfrm_del_sa
+00000000000000d2 T d_prune_aliases
-00000000000000d4 T usb_sg_cancel
-00000000000000d5 t pci_pm_freeze_noirq
-00000000000000d5 t snd_timer_user_ccallback
+00000000000000d5 T usb_sg_cancel
+00000000000000d7 t pci_pm_freeze_noirq
-00000000000000d9 T d_prune_aliases
-00000000000000db t loop_queue_write_work
-00000000000000e9 t dentry_lru_isolate
+00000000000000ed t dentry_lru_isolate
+00000000000000f6 t max_sync_store
+00000000000000fa t slab_sysfs_init
+00000000000000fc T fat_attach
-00000000000000fc t slab_sysfs_init
-00000000000000fe T fat_attach
-00000000000000fe t max_sync_store
-00000000000000ff t slab_out_of_memory
+0000000000000100 t slab_out_of_memory
+0000000000000100 t tg3_nway_reset
-0000000000000102 t ctnetlink_dump_exp_ct
+0000000000000103 t calgary_fixup_tce_spaces
+0000000000000103 t ext4_inode_csum_set
-0000000000000103 t tg3_nway_reset
-0000000000000104 t calgary_fixup_tce_spaces
-0000000000000105 t __unmap_single.NNN
-0000000000000106 t acpi_ec_stop
-0000000000000107 t pktsched_init
+0000000000000107 t __unmap_single.NNN
+0000000000000108 t pktsched_init
-000000000000010b t ext4_inode_csum_set
+000000000000010f T drm_vblank_on
+0000000000000113 t ctnetlink_dump_exp_ct
-0000000000000115 t mddev_put
+0000000000000117 t nl80211_set_mac_acl
-0000000000000119 T drm_vblank_on
+000000000000011a t mddev_put
+000000000000011b T wait_for_completion_interruptible_timeout
-000000000000011c T wait_for_completion_interruptible_timeout
+0000000000000123 t acpi_ec_stop
+0000000000000125 t ext4_mb_simple_scan_group
-0000000000000125 t nfs_do_filldir
+0000000000000125 T remove_proc_subtree
-0000000000000127 t nl80211_set_mac_acl
+000000000000012b t azx_irq_pending_work
+000000000000012d t nfs_do_filldir
-000000000000012e T remove_proc_subtree
+0000000000000131 t store_uframe_periodic_max
-0000000000000133 t store_uframe_periodic_max
-0000000000000137 t ext4_mb_simple_scan_group
-0000000000000139 T jbd2_journal_put_journal_head
-000000000000013b t azx_irq_pending_work
+000000000000013b T jbd2_journal_put_journal_head
+000000000000013d t xfrm_add_policy
+0000000000000142 t serial8250_backup_timeout
-0000000000000145 t xfrm_add_policy
+000000000000014a T hid_dump_device
-000000000000014c t alloc_buddy_huge_page
+0000000000000151 t finish_urb
+0000000000000151 t submit_flushes
-0000000000000152 T hid_dump_device
-0000000000000152 t serial8250_backup_timeout
+0000000000000154 t alloc_buddy_huge_page
-0000000000000159 t submit_flushes
+000000000000015a T vfs_mknod
+000000000000015c T remove_proc_entry
-000000000000015e T remove_proc_entry
-0000000000000161 t finish_urb
+0000000000000162 t nv_remove
-0000000000000162 T vfs_mknod
+0000000000000165 T nf_ct_delete
-0000000000000166 t nv_nic_irq_rx
+0000000000000167 t nv_nic_irq_rx
-000000000000016a t nv_remove
-0000000000000174 T iommu_tbl_pool_init
-0000000000000175 t queue_process
+000000000000017c T iommu_tbl_pool_init
+0000000000000180 t queue_process
-0000000000000183 t ctnetlink_del_conntrack
-0000000000000185 T nf_ct_delete
-000000000000019f t ctnetlink_new_conntrack
-00000000000001aa t azx_attach_pcm_stream
-00000000000001aa t dump_header.NNN
+00000000000001ab t azx_attach_pcm_stream
+00000000000001ab t dump_header.NNN
+00000000000001ae t acpi_ec_add
-00000000000001b0 t acpi_ec_add
+00000000000001b2 t ctnetlink_del_conntrack
-00000000000001b8 t nv_close
+00000000000001c9 t ctnetlink_new_conntrack
-00000000000001d1 t ctnetlink_create_expect.NNN
+00000000000001dc t nv_close
+00000000000001e1 t ctnetlink_create_expect.NNN
-00000000000001e3 t ctnetlink_get_conntrack
-00000000000001f0 t ctnetlink_dump_table
+00000000000001f0 t hid_debug_rdesc_show
-00000000000001f8 t hid_debug_rdesc_show
-00000000000001fa T ata_task_ioctl
-00000000000001fa T dm_kcopyd_copy
+00000000000001fb t ctnetlink_dump_table
+00000000000001fc T dm_kcopyd_copy
+0000000000000202 T ata_task_ioctl
-0000000000000207 T usb_reset_device
-000000000000020c T inet_bind
+000000000000020f T usb_reset_device
+0000000000000212 t ctnetlink_get_conntrack
-000000000000021a t nl80211_send_mlme_event.NNN
+000000000000021c T inet_bind
+000000000000021c t nl80211_send_mlme_event.NNN
+0000000000000226 t nv_nic_irq_other
-000000000000022e T destroy_workqueue
-000000000000022e t nv_nic_irq_other
+0000000000000230 T destroy_workqueue
-0000000000000233 T netpoll_send_skb_on_dev
+0000000000000242 t snd_timer_user_read
-0000000000000245 t snd_timer_user_read
-000000000000024a t univ8250_setup_irq
+000000000000024b T netpoll_send_skb_on_dev
-000000000000024c t ctnetlink_dump_list
+000000000000024f t ctnetlink_dump_list
+0000000000000252 t univ8250_setup_irq
+0000000000000253 t unix_dgram_connect
-000000000000026f t ieee80211_tx_h_select_key
-0000000000000275 t remove_and_add_spares
+0000000000000277 t ieee80211_tx_h_select_key
+0000000000000277 t remove_and_add_spares
+0000000000000278 t slot_store
-0000000000000280 t slot_store
-0000000000000286 t d_walk
-0000000000000288 T sys_shmctl
-0000000000000288 T SyS_shmctl
-0000000000000291 t unix_dgram_connect
+00000000000002a3 t prepend_path
+00000000000002a4 t d_walk
+00000000000002a4 T sys_shmctl
+00000000000002a4 T SyS_shmctl
+00000000000002b6 t ext4_mb_find_by_goal
+00000000000002b6 T nf_conntrack_hash_check_insert
-00000000000002b7 t prepend_path
+00000000000002bf t nl80211_set_reg
-00000000000002c6 t ext4_mb_find_by_goal
+00000000000002c7 T __nf_conntrack_confirm
-00000000000002cf t nl80211_set_reg
-00000000000002f6 T nf_conntrack_hash_check_insert
-00000000000002ff T __nf_conntrack_confirm
-0000000000000308 T detect_calgary
+0000000000000309 T detect_calgary
-0000000000000326 t hiddev_read
-0000000000000328 t ext4_mb_generate_buddy
+0000000000000328 t hiddev_read
+0000000000000330 t ext4_mb_generate_buddy
-0000000000000345 t alps_process_touchpad_packet_v3_v5
-0000000000000346 t bsg_map_hdr.NNN
+0000000000000347 t bsg_map_hdr.NNN
+000000000000034c t nfs_end_delegation_return
+000000000000034d t alps_process_touchpad_packet_v3_v5
-000000000000035d t nfs_end_delegation_return
-0000000000000372 T oom_kill_process
+0000000000000373 T oom_kill_process
+0000000000000379 t acpi_ec_transaction
-0000000000000394 t acpi_ec_transaction
+00000000000003a3 t chv_read16
-00000000000003ab t chv_read16
-00000000000003d9 t hiddev_ioctl_usage.NNN
+00000000000003e9 t do_timerfd_settime
+00000000000003e9 t hiddev_ioctl_usage.NNN
-00000000000003eb t do_timerfd_settime
-0000000000000405 T ohci_hub_status_data
-0000000000000406 t chv_write16
-0000000000000406 t super_90_load
+0000000000000407 t super_90_load
-000000000000040b t loop_clr_fd
+000000000000040e t chv_write16
+0000000000000415 T ohci_hub_status_data
+000000000000041b t loop_clr_fd
+0000000000000425 t worker_thread
-0000000000000428 t worker_thread
+0000000000000444 T azx_init_chip
-000000000000044c T azx_init_chip
-0000000000000453 T ext4_discard_preallocations
-0000000000000475 T do_shmat
+000000000000047a T do_shmat
+000000000000047b T ext4_discard_preallocations
-0000000000000488 t ext4_direct_IO
+0000000000000498 t ext4_direct_IO
+00000000000004f9 t ext4_mb_normalize_request
-0000000000000509 t ext4_mb_normalize_request
-000000000000050b t __dev_queue_xmit
+0000000000000511 t __dev_queue_xmit
-00000000000005d6 T serial8250_do_startup
+00000000000005e6 T serial8250_do_startup
-000000000000061c t nv_update_linkspeed
+000000000000061b t nv_napi_poll
-0000000000000625 t nv_napi_poll
+000000000000062c t nv_update_linkspeed
-00000000000007e8 t nv_start_xmit_optimized
+00000000000007ea t nv_start_xmit_optimized
-0000000000000893 t futex_requeue
+0000000000000891 t futex_requeue
+0000000000000898 t nl80211_parse_sched_scan.NNN
-00000000000008a0 t nl80211_parse_sched_scan.NNN
-000000000000094c T sock_setsockopt
+0000000000000950 T sock_setsockopt
-00000000000009ee T do_futex
+00000000000009fe T do_futex
-0000000000000a66 t nv_self_test
+0000000000000aaa t nv_self_test
-0000000000000af3 T sata_pmp_error_handler
+0000000000000afb T sata_pmp_error_handler
-0000000000000b5d t ohci_urb_enqueue
+0000000000000b65 t ohci_urb_enqueue
-0000000000000b72 t display_crc_ctl_write
+0000000000000b79 t display_crc_ctl_write
-0000000000000ca7 t SYSC_semtimedop
+0000000000000c9b t SYSC_semtimedop
+0000000000000e31 T md_do_sync
-0000000000000e3f T md_do_sync
+0000000000000fd3 t ehci_urb_enqueue
-0000000000000ffb t ehci_urb_enqueue
-0000000000001027 t packet_sendmsg
+0000000000001029 t packet_sendmsg
-0000000000001e17 t do_blockdev_direct_IO
+0000000000001e07 t do_blockdev_direct_IO
-000000000000261e t nl80211_send_wiphy
+000000000000262e t nl80211_send_wiphy
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ