Skip to contentSkip to conversation

Linus 2.6.29-rc4

PATCH In review Linus Torvalds 15 participants 40 messages Feb 8, 2009 – Apr 26, 2009

Subsystems (7)
lore
Share & tools
lore
#0 lore raw

Another week (and a half), another -rc.

Arch updates (sparc, blackfin), driver updates (dvb, mmc, ide), ACPI updates, FS updates (ubifs, btrfs), you name it. It's all there.

But more importantly, people really have been working on regressions, and hopefully this closes a nice set of the top one, and hopefully without introducing too many new ones.

Linus

---
Aaro Koskinen (1):
ARM: OMAP: gptimer min_delta_ns corrected

Adrian Hunter (5):
UBIFS: ensure orphan area head is initialized
UBIFS: fix LPT out-of-space bug (again)
UBIFS: sync wbufs after syncing inodes and pages
UBIFS: spelling fix 'date' -> 'data'
UBIFS: remount ro fixes

Al Viro (1):
braino in sg_ioctl_trans()

Alberto Bertogli (3):
Fix small typo in bio.h's documentation
Fix misleading comment in bio.h
bio.h: If they MUST be inlined, then use __always_inline

Alex Buell (1):
fbdev: incorrect URL given in drivers/video/Kconfig

Alex Chiang (1):
PCI: properly clean up ASPM link state on device remove

Alexander Beregalov (1):
sata_sil: Fix build breakage

Alexander Duyck (3):
igb: Fix DCA errors and do not use context index for 82576
igb: prevent skb_over panic w/ mtu smaller than 1K
igb: fix link reporting when using sgmii

Alexey Dobriyan (3):
sky2: fix hard hang with netconsoling and iface going up
alpha: fixup BUG macro
seq_file: fix big-enough lseek() + read()

Alexey Klimov (1):

      V4L/DVB (10317): radio-mr800: fix radio->muted and radio->stereo

Andre Noll (1):
md: Fix a bug in linear.c causing which_dev() to return the wrong device.

Andrea Righi (1):
fbmem: don't call copy_from/to_user() with mutex held

Andreas Schwab (1):
x86 setup: fix asm constraints in vesa_store_edid

Andrew Morton (4):
drivers/gpu/drm/i915/intel_lvds.c: fix locking snafu kernel/async.c: fix printk warnings
revert "rlimit: permit setting RLIMIT_NOFILE to RLIM_INFINITY"
hp-wmi: fix error path in hp_wmi_bios_setup()

Andy Fleming (2):
gianfar: Fix BD_LENGTH_MASK definition
gianfar: Fix potential soft reset race

Andy Walls (3):
V4L/DVB (10218): cx23885: Fix Oops for mixed install of analog and digital only cards V4L/DVB (10219): saa7134: Prevent Oops due to stale IRQ status when enabling interrupts V4L/DVB (10229): cx88-dvb: Fix order of frontend allocations

Anton Vorontsov (1):
gianfar: Fix Wake-on-LAN support

Antti Palosaari (2):
V4L/DVB (10287): af9015: fix second FE
V4L/DVB (10288): af9015: bug fix: stick does not work always when plugged

Arjan van de Ven (1):
ahci: add a module parameter to ignore the SSS flags for async scanning

Artem Bityutskiy (20):
UBI: improve ioctl commentaries
UBI: remove unnecessry header inclusion
UBI: allow all ioctls
UBIFS: do not commit twice
UBIFS: constify operations
UBI: add ioctl compatibility
UBI: use nicer 64-bit math
UBIFS: do not treat all data as short term
UBIFS: document dark_wm and dead_wm better
UBI: remove unused variable
UBI: fix resource de-allocation
UBIFS: simplify locking
UBIFS: fix LEB list freeing
UBIFS: add re-mount debugging checks
UBIFS: always clean up GC LEB space
UBIFS: fix assertions
UBIFS: fix no_chk_data_crc
UBIFS: return sensible error codes
UBIFS: remove fast unmounting
write-back: fix nr_to_write counter

Arve Hjønnevåg (3):
Staging: android: binder: fix arm build errors
Staging: android: timed_gpio: Fix build to build on kernels after 2.6.25.
Staging: android: Add lowmemorykiller documentation.

Atsushi Nemoto (2):
MIPS: TXx9: Add support for TX4939 internal RTC
tx4939ide: typo fix and minor cleanup

Atsushi SAKAI (1):
lguest: typos fix

Balaji Rao (1):
mfd: Remove non exported references from pcf50633

Bartlomiej Zolnierkiewicz (4):
block: export SSD/non-rotational queue flag through sysfs
ide: fix ide_register_port() failure handling
ide: add CS5536 host driver (v3)
icside: fix PCB version 6 support (v2)

Ben Hutchings (5):
sfc: SFN4111T: Fix GPIO sharing between I2C and FLASH_CFG_1
sfc: Update board info for hardware monitor on SFN4111T-R5 and later
sfc: SFT9001: Always enable XNP exchange on SFT9001 rev B
sfc: SFX7101/SFT9001: Fix AN advertisements
sfc: Replace stats_enabled flag with a disable count

Benjamin Herrenschmidt (4):
powerpc: Fix oops on some machines due to incorrect pr_debug()
atyfb: Properly save PCI state before changing PCI PM level
aty128fb: Properly save PCI state before changing PCI PM level
radeonfb: Fix resume from D3Cold on some platforms

Benjamin Zores (1):
ipv4: fix infinite retry loop in IP-Config

Bernd Schmidt (1):

      Blackfin arch: fix bug - traps test case 19 for exception 0x2d fails

Bharath Ramesh (1):
hwmon: applesmc: add support for MacPro 3 temperature sensors

Bill Nottingham (1):
Documentation/Changes: add required versions for new filesystems

Boaz Harrosh (1):
include/linux: Add bsg.h to the Kernel exported headers

Bob Copeland (1):
ath5k: fix locking in ath5k_config

Borislav Petkov (2):
ide-cd: fix DMA for non bio-backed requests
x86: APIC: enable workaround on AMD Fam10h CPUs

Bryan Wu (1):
Blackfin arch: Add ANOMALY_05000380 to BF54x to kill the compile warning

Carmelo AMOROSO (1):
sh: fcnvds fix with denormalized numbers on SH-4 FPU.

Carsten Otte (1):
do_wp_page: fix regression with execute in place

Chris Ball (1):
Btrfs: Handle SGID bit when creating inodes Before this patch, new files/dirs would ignore the SGID bit on their parent directory and always be owned by the creating user's uid/gid.

Chris Mason (18):
Btrfs: stop providing a bmap operation to avoid swapfile corruptions
Btrfs: do less aggressive btree readahead
Btrfs: fix readdir on 32 bit machines
Btrfs: Catch missed bios in the async bio submission thread
Btrfs: async threads should try harder to find work
Btrfs: sort references by byte number during btrfs_inc_ref
Btrfs: disable leak debugging checks in extent_io.c
Btrfs: hash_lock is no longer needed
Btrfs: Change btree locking to use explicit blocking points
Btrfs: Make btrfs_drop_snapshot work in larger and more efficient chunks
Btrfs: Don't try to compress pages past i_size
Btrfs: Change btrfs_truncate_inode_items to stop when it hits the inode
Btrfs: change btrfs_del_leaf to drop locks earlier
Btrfs: fix btrfs_unlock_up_safe to walk the entire path
Btrfs: Only prep for btree deletion balances when nodes are mostly empty
Btrfs: don't return congestion in write_cache_pages as often
Btrfs: Fix memory leak in cache_drop_leaf_ref
Btrfs: Make sure dir is non-null before doing S_ISGID checks

Christian Hesse (1):
Btrfs: make btrfs acls selectable This patch adds a menu entry to kconfig to enable acls for btrfs. This allows you to enable FS_POSIX_ACL at kernel compile time.

Christopher SMITH (1):

      sh: Only reserve memory under CONFIG_ZERO_PAGE_OFFSET when it != 0.

Clemens Ladisch (1):
sound: usb-audio: handle wMaxPacketSize for FIXED_ENDPOINT devices

Cliff Brake (2):
pxamci: replace #ifdef CONFIG_PXA27x with if (cpu_is_pxa27x())
pxamci: enable DMA for write ops after CMD/RESP

Cord Walter (1):
pcnet_cs: Fix misuse of the equality operator.

Corentin Chary (10):
UBI: add ioctl for map operation
UBI: add ioctl for unmap operation
UBI: add ioctl for is_mapped operation
eeepc-laptop: split eeepc_backlight_exit()
asus_acpi: Add R1F support
asus-laptop: use generic netlink interface
asus-laptop: hotkeys via the generic input interface
asus-laptop: update Kconfig for input layer
asus-laptop: fix label indentation
eeepc-laptop: use netlink interface

Cornelia Huck (4):
async: Fix running list handling.
async: Handle kthread_run() return codes.
async: Add some documentation.
async: Rename _special -> _domain for clarity.

Daisuke Nishimura (1):
memcg: get/put parents at create/free

Dan Carpenter (2):
fix emacs indenting howto filename expansion
sx.c: fix missed unlock_kernel() on error path in sx_fw_ioctl()

Daniel Marjamäki (1):
netxen: fix memory leak in drivers/net/netxen_nic_init.c

Darren Salt (1):
eeepc-laptop: fix oops when changing backlight brightness during eeepc-laptop init

Darrick J. Wong (1):
PCI hotplug: fakephp: Allocate PCI resources before adding the device

Dave Chinner (1):
xfs: Check buffer lengths in log recovery

Dave Jones (2):
ipv6: compile fix for ip6mr.c
x86: add cache descriptors for Intel Core i7

Dave Kleikamp (1):
vfs: Don't call attach_nobh_buffers() with an empty list

David Altobelli (2):
hpilo: increment version
hpilo: open/close fix

David Brownell (1):
rtc: rtc-dm355evm driver

David Daney (3):
MIPS: Fix a typo in watchpoint register structure.
MIPS: Read watch registers with interrupts disabled.
MIPS: Use hardware watchpoints on all R1 and R2 CPUs.

David Howells (1):
CRED: Fix SUID exec regression

David Rientjes (1):
slub: fix per cpu kmem_cache_cpu array memory leak

David S. Miller (10):
ipv6: Make mc_forwarding sysctl read-only.
sparc64: Move generic PCR support code to seperate file.
sparc: Probe PMU type and record in sparc_pmu_type.
tun: Add some missing TUN compat ioctl translations.
sparc64: Implement NMI watchdog on capable cpus.
net: Fix userland breakage wrt. linux/if_tunnel.h
sparc64: On non-Niagara, need to touch NMI watchdog in NOHZ mode.
sparc64: Validate kernel generated fault addresses on sparc64.
sparc64: Kill bogus TPC/address truncation during 32-bit faults. Revert "tcp: Always set urgent pointer if it's beyond snd_nxt"

David Vrabel (4):
uwb: safely remove all reservations
wusb: return -ENOTCONN when resetting a port with no connected device
wusb: timeout when waiting for ASL/PZL updates in whci-hcd
uwb: lock rc->rsvs_lock with spin_lock_bh()

Davide Libenzi (1):
epoll: drop max_user_instances and rely only on max_user_watches

Dean Nelson (2):
sgi-xp: link XPNET's net_device_ops to its net_device structure
sgi-xp: fix writing past the end of kzalloc()'d space

Devin Heitmueller (2):
V4L/DVB (10261): em28xx: fix kernel panic on audio shutdown V4L/DVB (10411): s5h1409: Perform s5h1409 soft reset after tuning

Dhananjay Phadke (1):
netxen: revert jumbo ringsize

Dimitris Michailidis (1):
tcp: Fix length tcp_splice_data_recv passes to skb_splice_bits.

Divy Le Ray (1):
cxgb3: Fix lro switch

Divyesh Shah (1):
cfq-iosched: Allow RT requests to pre-empt ongoing BE timeslice

Ed Swierk (1):
signals, debug: fix BUG: using smp_processor_id() in preemptible code in print_fatal_signal()

Eero Nurkkala (1):
OMAP: ASoC: Fix spinlock misuse in omap-pcm.c

Eric Biederman (1):
seq_file: move traverse so it can be used from seq_read

Eric Dumazet (4):
udp: optimize bind(0) if many ports are in use
net: wrong test in inet_ehash_locks_alloc()
udp: increments sk_drops in __udp_queue_rcv_skb()
modules: Use a better scheme for refcounting

Eric Piel (1):
lis3lv02d: add axes knowledge for HP 6530

Eric Sandeen (1):
don't reallocate sxp variable passed into xfs_swapext

Evgeniy Polyakov (1):
mm: OOM documentation update

Felix Blyakher (1):
[XFS] Warn on transaction in flight on read-only remount

Filip Aben (1):
hso: add new device id's

Frank Seidel (1):
ACPI: add missing KERN_* constants to printks

Frans Pop (2):
hp-wmi: fix regressions caused by missing if statement
hp-wmi: set initial docking state

Frederic Weisbecker (2):
tracing/function-graph-tracer: fix a regression while suspend to disk
hrtimers: increase clock min delta threshold while interrupt hanging

Geert Uytterhoeven (1):
staging: agnx: drivers/staging/agnx/agnx.h needs <linux/io.h>

Gerhard Pircher (1):
ide: Force VIA IDE legacy interrupts for AmigaOne boards

Grace Pan (1):
Blackfin arch: Add in cflag to support mlong-calls for kgdb_test

Graf Yang (3):
Blackfin arch: Fix bug - BF518 port F, G, and H have different mux offset compare to BF527 Blackfin arch: Fix bug - 561 SMP kernel can't boot from jffs2
Blackfin arch: add platform device bfin_mii-bus and KSZ8893M switch driver platform resources to board files

Grant Likely (3):
powerpc/5200: update defconfigs
powerpc/5200: Bugfix for PCI mapping of memory and IMMR powerpc/5200: update device tree binding documentation

Greg Kroah-Hartman (3):
Staging: comedi: fix Kbuild
Staging: meilhaus: fix Kbuild
Staging: android: fix build error on 64bit boxes

H. Peter Anvin (2):
x86: correct the CPUID pattern for MSR_IA32_MISC_ENABLE availability
x86: add [email protected] to MAINTAINERS

Haiying Wang (1):
ucc_geth: Change uec phy id to the same format as gianfar's

Hans J. Koch (1):
UIO: Add missing documentation of features added recently

Hans Verkuil (8):

      V4L/DVB (10214): Fix 'stb0899_get_srate' defined but not used warning
      V4L/DVB (10230): v4l2-device: fix buggy macro
      V4L/DVB (10243): em28xx: fix compile warning
      V4L/DVB (10248): v4l-dvb: fix a bunch of compile warnings.
      V4L/DVB (10250): cx25840: fix regression: fw not loaded on first use
      V4L/DVB (10270): saa7146: fix unbalanced mutex_lock/unlock
      V4L/DVB (10314): cx25840: ignore TUNER_SET_CONFIG in the command callback.
      V4L/DVB (10229): ivtv: fix memory leak

Hans-Christian Egtvedt (1):
atmel-ssc: fix misuse of dev_dbg when requested ssc instance is not found

Herbert Xu (3):
net: Fix frag_list handling in skb_seq_read
packet: Avoid lock_sock in mmap handler
ipv6: Copy cork options in ip6_append_data

Holger Macht (1):
ACPI: dock: Don't eval _STA on every show_docked sysfs read

Huang Weiyi (5):
uwb: remove unused #include <version.h>'s
Btrfs: removed unused #include <version.h>'s
Btrfs: remove duplicated #include
V4L/DVB (10193): removed unused #include <version.h>'s
MIPS: Octeon: Remove duplicated #includes

Ian Campbell (1):
IDE: Unregister and disable devices if initialization fails.

Inaky Perez-Gonzalez (1):
wimax: fix build issue when debugfs is disabled

Ingo Molnar (2):
irq: export __set_irq_handler() and handle_level_irq()
x86: tone down mtrr_trim_uncached_memory() warning

Ira W. Snyder (1):
virtio_net: use correct accessors for scatterlists

Ivan Kokshaysky (3):
alpha: use syscall wrappers
alpha: compile fixes
alpha: fix the BUG() macro

James Bottomley (1):
x86/Voyager: make it build and boot

Jan Engelhardt (2):
UBI: constify file operations
Btrfs: change/remove typedef

Jan Kara (2):
ocfs2: Push out dropping of dentry lock to ocfs2_wq
ocfs2: Fix possible deadlock in ocfs2_write_dquot()

Jarkko Nikula (1):
ARM: OMAP: DMA: Fix uninitialized channel flags

Jarod Wilson (1):
HID: add antec-branded soundgraph imon devices to blacklist

Jaswinder Singh Rajput (90):
headers_check fix: can/bcm.h
headers_check fix: dvb/audio.h
headers_check fix: dvb/dmx.h
headers_check fix: dvb/frontend.h
headers_check fix: dvb/net.h
headers_check fix: dvb/video.h
headers_check fix: netfilter/xt_conntrack.h
headers_check fix: nfsd/export.h
headers_check fix: nfsd/nfsfh.h
headers_check fix: nfsd/syscall.h
headers_check fix: raid/md_p.h
headers_check fix: spi/spidev.h
headers_check fix: tc_act/tc_gact.h
headers_check fix: tc_act/tc_mirred.h
headers_check fix: tc_act/tc_pedit.h
headers_check fix: tc_ematch/tc_em_cmp.h
headers_check fix: tc_ematch/tc_em_meta.h
headers_check fix: tc_ematch/tc_em_nbyte.h
headers_check fix: tc_ematch/tc_em_text.h
headers_check fix: usb/cdc.h
headers_check fix: usb/gadgetfs.h
headers_check fix: linux/aio_abi.h
headers_check fix: linux/atalk.h
headers_check fix: linux/atmbr2684.h
headers_check fix: linux/auto_fs4.h
headers_check fix: linux/bfs_fs.h
headers_check fix: linux/blktrace_api.h
headers_check fix: linux/capability.h
headers_check fix: linux/cdrom.h
headers_check fix: linux/cgroupstats.h
headers_check fix: linux/dlm_plock.h
headers_check fix: linux/dn.h
headers_check fix: linux/edd.h
headers_check fix: linux/efs_fs_sb.h
headers_check fix: linux/elf-fdpic.h
headers_check fix: linux/elf.h
headers_check fix: linux/errqueue.h
headers_check fix: linux/genetlink.h
headers_check fix: linux/gfs2_ondisk.h
headers_check fix: linux/hid.h
headers_check fix: linux/hiddev.h
headers_check fix: linux/icmpv6.h
headers_check fix: linux/if_addr.h
headers_check fix: linux/if_addrlabel.h
headers_check fix: linux/if_fc.h
headers_check fix: linux/if_hippi.h
headers_check fix: linux/if_link.h
headers_check fix: linux/if_ppp.h
headers_check fix: linux/if_strip.h
headers_check fix: linux/if_tr.h
headers_check fix: linux/igmp.h
headers_check fix: linux/inet_diag.h
headers_check fix: linux/ip6_tunnel.h
headers_check fix: linux/ipv6.h
headers_check fix: linux/ipv6_route.h
headers_check fix: linux/ipx.h
headers_check fix: linux/irda.h
headers_check fix: linux/minix_fs.h
headers_check fix: linux/msdos_fs.h
headers_check fix: linux/neighbour.h
headers_check fix: linux/nfs_idmap.h
headers_check fix: linux/phonet.h
headers_check fix: linux/pkt_cls.h
headers_check fix: linux/pkt_sched.h
headers_check fix: linux/ppp_defs.h
headers_check fix: linux/random.h
headers_check fix: linux/signalfd.h
headers_check fix: linux/sound.h
headers_check fix: linux/synclink.h
headers_check fix: linux/taskstats.h
headers_check fix: linux/video_decoder.h
headers_check fix: linux/video_encoder.h
headers_check fix: linux/videodev.h
headers_check fix: linux/virtio_blk.h
headers_check fix: linux/virtio_console.h
headers_check fix: linux/virtio_net.h
headers_check fix: mtd/inftl-user.h
headers_check fix: sound/hdsp.h
headers_check fix: video/sisfb.h
headers_check fix: video/uvesafb.h
headers_check fix: linux/nubus.h
headers_check fix: linux/rtnetlink.h
headers_check fix: x86, e820.h
headers_check fix: x86, kvm.h
headers_check fix: x86, mce.h
headers_check fix: x86, mtrr.h
headers_check fix: x86, ptrace-abi.h
headers_check fix: x86, sigcontext.h
headers_check fix: x86, sigcontext32.h
headers_check fix: x86, swab.h

Jean-Francois Moine (1):
V4L/DVB (10385): gspca - main: Fix memory leak when USB disconnection while streaming.

Jeff Layton (2):
cifs: turn smb_send into a wrapper around smb_sendv
cifs: make sure we allocate enough storage for socket address

Jeff Mahoney (1):
kmalloc: return NULL instead of link failure

Jens Axboe (8):
block: get rid of the manual directory counting in blktrace
block: seperate bio/request unplug and sync bits
block: add bio_rw_flagged() for testing bio->bi_rw
block: silently error an unsupported barrier bio
block: add sysfs file for controlling io stats accounting
block: fix oops in blk_queue_io_stat()
block: fix inconsistent parenthesisation of QUEUE_FLAG_DEFAULT
block: add text file detailing queue/ sysfs files

Jeremy Fitzhardinge (2):
xen: make sysfs files behave as their names suggest
xen: disable interrupts before saving in percpu

Jesper Dangaard Brouer (1):
udp: Fix UDP short packet false positive

Jesse Brandeburg (1):
e1000: fix bug with shared interrupt during reset

Jie Zhang (2):
Blackfin arch: Add one more check on `fp' to prevent double fault Blackfin arch: fix 2 bugs related to debug

Jim Owens (1):
Btrfs: selinux support Add call to LSM security initialization and save resulting security xattr for new inodes.

Jiri Kosina (2):
HID: adjust report descriptor fixup for MS 1028 receiver
HID: document difference between hid_blacklist and hid_ignore_list

Jiri Pirko (1):
net: fix xfrm reverse flow lookup for icmp6

Jiri Tersel (1):
lis3lv02d: add axes knowledge for HP 6510b

Joe Perches (1):
Btrfs: MAINTAINERS entry

Joel Becker (1):
configfs: Silence lockdep on mkdir(), rmdir() and configfs_depend_item()

Joerg Schirottke (1):
ALSA: hda - Add quirk for HP DV6700 laptop

Johannes Berg (1):
mac80211: remove Michael Wu as maintainer

Johannes Weiner (1):
wait: prevent exclusive waiter starvation

Josef Bacik (2):
Btrfs: cleanup xattr code
Btrfs: join the transaction in __btrfs_setxattr

Juha Yrjola (1):
ARM: OMAP: Fix race in OMAP2/3 DMA IRQ handling

KAMEZAWA Hiroyuki (4):
cgroups: use hierarchy mutex in creation failure path
memcg: fix refcnt handling at swapoff
memcg: update document to mention that swapoff should be tested
memcg: NULL pointer dereference at rmdir on some NUMA systems

Karsten Keil (1):
e1000: Fix PCI enable to honor the need_ioport flag

Kay Sievers (2):
ide: struct device - replace bus_id with dev_name(), dev_set_name()
ACPI: struct device - replace bus_id with dev_name(), dev_set_name()

Kevin Hilman (1):
ARM: OMAP: fix fault in enter_full_retention()

Kuninori Morimoto (4):
sh: mach-migor: Enable ov772x and tw9910 in defconfig.
sh: ap325rxa: control camera power toggling.
sh: ap325rxa: Add ov772x support.
sh: ap325rxa: Enable ov772x in defconfig.

Lai Jiangshan (1):
ring_buffer: reset write when reserve buffer fail

Larry Finger (1):
rtl8187: Fix error in setting OFDM power settings for RTL8187L

Laurent Pinchart (3):
V4L/DVB (10197): uvcvideo: Whitespace and comments cleanup, copyright updates. V4L/DVB (10198): uvcvideo: Print the UVC version number in binary-coded decimal. V4L/DVB (10199): uvcvideo: Fix GET_DEF failure detection.

Len Brown (6):
ACPI: remove locking from PM1x_STS register reads
ACPI: remove BM_RLD access from idle entry path
ACPICA: add debug dump of BIOS _OSI strings
ACPI: make some IO ports off-limits to AML
ACPI: delete CPU_IDLE=n code
ACPI: disable ACPI cleanly when bad RSDP found

Li Zefan (2):
cgroups: fix lock inconsistency in cgroup_clone() fork.c: fix NULL pointer dereference when nr_threads == threads-max

Liam Girdwood (1):
ALSA: ASoC: email - update email addresses.

Linus Torvalds (5):
Fix OOPS in mmap_region() when merging adjacent VM_LOCKED file segments Allow opportunistic merging of VM_CAN_NONLINEAR areas
Stop playing silly games with the VM_ACCOUNT flag
Manually revert "mlock: downgrade mmap sem while populating mlocked regions" Linux 2.6.29-rc4

Linus Walleij (1):
mmci: Add support for ST Micro derivate

Luis R. Rodriguez (2):
cfg80211: Fix sanity check on 5 GHz when processing country IE
cfg80211: print correct intersected regulatory domain

Luke Yelavich (1):
ALSA: hda - add another MacBook Pro 4, 1 subsystem ID

Maciej Sosnowski (1):
dca: redesign locks to fix deadlocks

Magnus Damm (1):
gpiolib: fix request related issue

Manish Katiyar (2):
r6040: Remove unused variable pdev from drivers/net/r6040.c
rtc-ds1390: fix compilation warnings in drivers/rtc/rtc-ds1390.c

Manuel Lauss (1):
MIPS: Alchemy: time.c build fix

Mark Brown (1):
ASoC: Fix null string usage with WM8753 DAIs

Mark Eggleston (1):
ALSA: hda - Add support of iMac 24 Aluminium

Mark Fasheh (2):
ocfs2: add quota call to ocfs2_remove_btree_range()
Revert "configfs: Silence lockdep on mkdir(), rmdir() and configfs_depend_item()"

Mark Lord (1):
sata_mv: Fix chip type for Hightpoint RocketRaid 1740/1742

Mark McLoughlin (1):
virtio-pci: do not oops on config change if driver not loaded

Mark Wallis (1):
lguest: Fix a memory leak with the lg object during launcher close

Markus Metzger (1):
x86, ds, bts: cleanup/fix DS configuration

Martin Dauskardt (1):

      V4L/DVB (10216): saa7127: fix broken S-Video with saa7129

Martin Hicks (1):
x86: push old stack address on irqstack for unwinder

Martin K. Petersen (3):
block: Don't verify integrity metadata on read error
block: Remove obsolete BUG_ON
block: Allow empty integrity profile

Martin Kebert (1):
lis3lv02d: add axes knowledge for HP 6710

Masami Hiramatsu (2):
kprobes: fix module compilation error with CONFIG_KPROBES=n prevent kprobes from catching spurious page faults

Matthew Garrett (5):
eeepc-laptop: Add support for extended hotkeys
eeepc-laptop: Check return values from rfkill_register
eeepc-laptop: Implement rfkill hotplugging in eeepc-laptop PCI hotplug: Change link order of pciehp & acpiphp
misc: dell-laptop should depend on POWER_SUPPLY

Matthew Wilcox (1):
PCI MSI: Fix undefined shift by 32

Matthias Dahl (1):
V4L/DVB (9054): implement proper locking in the dvb ca en50221 driver

Mauro Carvalho Chehab (6):

      V4L/DVB (10192): em28xx: fix input selection
      V4L/DVB (10201): Fixes cx88 compilation bug
      V4L/DVB (10208): zoran: Re-adds udev entry removed by changeset 60b4bde4
      V4L/DVB (10209): tveeprom: Properly initialize tuner type (BZ#11367)
      V4L/DVB (10228): em28xx: fix audio output PCM IN selection
      V4L/DVB (10403): saa7134-alsa: saa7130 doesn't support digital audio

Miao Xie (1):
cpuset: fix possible deadlock in async_rebuild_sched_domains

Michael Hennerich (9):
Blackfin arch: Fix Bug - Kernel does not boot if re-program clocks
Blackfin arch: Fix Bug - request lines with peripheral_request_list, but don't get freed with peripheral_free_list Blackfin arch: reset POLAR setting when acquiring a gpio for the first time
Blackfin arch: Faster Implementation of csum_tcpudp_nofold()
Blackfin arch: Update CM-BF527 kernel config
Blackfin arch: Add BF561 PPI POLS, POLC Masks
Blackfin arch: Update Copyright information
Blackfin arch: Fix udelay implementation
Blackfin arch: Remove outdated code

Mike Frysinger (15):
Blackfin arch: scrub comments/whitespace/cvs keywords
Blackfin arch: add support for mobile ddr bf54x parts
Blackfin arch: line up machine-/cpu- vars after BF54xM addition Blackfin arch: fix bug - Dmacopy failed in BF537-STAMP
Blackfin arch: cleanup bf54x ifdef mess in gpio code
MAINTIANERS: Blackfin: remove subscribers-only marking
Blackfin arch: sync reboot handler with version in u-boot Blackfin arch: bfin_reset->_bfin_reset redirection no longer needed Blackfin arch: drop CONFIG_I2C_BOARDINFO ifdefs
Blackfin arch: fixup board init function name
Blackfin arch: read SYSCR on newer parts that mirror the bits of SWRST in it Blackfin arch: base SIC_IWR# programming on whether the MMR exists Blackfin arch: build jtag tty driver as a module by default Blackfin arch: cplb mananger: use a do...while loop rather than a for loop Blackfin arch: define bfin_memmap as static since it is only used here

Mike Galbraith (1):
sched: clear buddies more aggressively

Mike Isely (1):
V4L/DVB (10240): Fix obvious swapped names in v4l2_subdev logic

Mike Rapoport (1):
drivers/video/backlight: rename da903x to da903x_bl

Mikulas Patocka (1):
Fix memory corruption in console selection

Misael Lopez Cruz (1):
ASoC: OMAP: Initialize XCCR and RCCR registers in McBSP DAI driver

Myron Stowe (1):
ACPICA: Fix table entry truncation calculation

NeilBrown (2):
md: Allow read error in a single drive raid1 to be passed up.
md: Ensure an md array never has too many devices.

Nicolas Fournier (1):
V4L/DVB (10233): [PATCH] Terratec Cinergy DT XS Diversity new USB ID (0ccd:0081)

Nicolas Pitre (1):
[ARM] 5366/1: fix shared memory coherency with VIVT L1 + L2 caches

Nikanth Karthikesan (1):
Mark mandatory elevator functions in the biodoc.txt

Oleg Nesterov (2):
Staging: android: task_get_unused_fd_flags: fix the wrong usage of tsk->signal
ftrace: do_each_pid_task() needs rcu lock

Oliver Neukum (1):
HID: fix reversed logic in disconnect testing of hiddev

Pallipadi, Venkatesh (1):
ACPI: Enable bit 11 in _PDC to advertise hw coord

Pasi Kärkkäinen (1):
[libata] sata_sil: Fix compilation error with libata debugging enabled

Paul Larson (1):
Add enable_ms to jsm driver

Paul Menage (2):
cgroups: add cpu_relax() calls in css_tryget() and cgroup_clear_css_refs()
cgroup: fix root_count when mount fails due to busy subsystem

Paul Mundt (1):
sh: Fix up spurious syscall restarting.

Pavel Herrmann (1):
lis3lv02d: add axes knowledge for HP 6730

Peter Zijlstra (4):
sched: fix sync wakeups
sched: symmetric sync vs avg_overlap
sched: fix buddie group latency
generic swap(): don't return a value from swap()

Petr Vandrovec (1):
firewire: core: Remove card from list of cards when enable fails

Phil Sutter (14):
MIPS: RB532: Fix bit swapping in rb532_set_bit()
MIPS: RC32434: Define io_map_base for PCI controller
MIPS: RB532: Add set_type() function to IRQ struct.
MIPS: RB532: Auto disable GPIO alternate function
MIPS: RB532: remove useless CF GPIO initialisation
MIPS: RB532: Detect uart type, add platform device
MIPS: RB532: Use driver_data instead of platform_data
MIPS: RB532: Fix init of rb532_dev3_ctl_res
MIPS: RB532: Fix set_latch_u5()
MIPS: RB532: Move dev3 init code to devices.c
MIPS: RB532: Remove {get,set}_434_reg()
MIPS: RB532: Simplify dev3 init
MIPS: RB532: Update headers
MIPS: RB532: Export rb532_gpio_set_func()

Philipp Zabel (1):
regulator: move bq24022 init back to module_init instead of subsys_initcall

Philippe De Muyter (1):
tulip: fix 21142 with 10Mbps without negotiation

Qinghuang Feng (3):
Btrfs: check return value for kthread_run() correctly
Btrfs: simplify iteration codes
Btrfs: open_ctree() error handling can oops on fs_info

Rafael J. Wysocki (13):
PM: Fix compilation warning in kernel/power/main.c
ACPI suspend: Fix compilation warnings in drivers/acpi/sleep.c PCI PM: Fix suspend error paths and testing facility breakage PCI PM: Fix hibernation breakage on EeePC 701
PCI PM: Power up devices before restoring their state
PCI PM: Do not wait for buses in B2 or B3 during resume
PCI PM: Fix handling of devices without drivers
PCI PM: Check if the state has been saved before trying to restore it PCI PM: Fix saving of device state in pci_legacy_suspend
PCI: PCIe portdrv: Simplify suspend and resume
PCI PM: Do not disable and enable bridges during suspend-resume PCI PM: Read power state from device after trying to change it on resume PCI PM: make the PM core more careful with drivers using the new PM framework

Ralf Baechle (7):
MIPS: SMTC: Fix build after recent creditial changes.
MIPS: Avoid destructive invalidation on partial cachelines.
MIPS: Port "mm: invoke oom-killer from page fault" from UML / x86
MIPS: atomic_*(): Change type of intermediate variables.
MIPS: Add return value checks to user_termio_to_kernel_termios()
MIPS: IP27: Switch from DMA_IP27 to DMA_COHERENT
MIPS: R2: Fix broken installation of cache error handler.

Randy Dunlap (7):
driver-core: fix kernel-doc parameter name
Documentation: move DMA-mapping.txt to Doc/PCI/
libata: fix kernel-doc warnings
sched: add missing kernel-doc in sched.h
sunrpc: fix rdma dependencies
maintainers: [email protected] is moderated
atyfb: fix CONFIG_ namespace violations

Reinette Chatre (2):
iwlwifi: clean key table in iwl_clear_stations_table
iwlwifi: save PCI state before suspend, restore after resume

Risto Suominen (1):
fbdev/atyfb: Fix DSP config on some PowerMacs & PowerBooks

Robert Krakora (4):
V4L/DVB (10254): em28xx: Fix audio URB transfer buffer race condition V4L/DVB (10256): em28xx: Fix for KWorld 330U AC97
V4L/DVB (10257): em28xx: Fix for KWorld 330U Board
V4L/DVB (10325): em28xx: Fix for fail to submit URB with IRQs and Pre-emption Disabled

Robin Getz (5):
Blackfin arch: Print out where the bootmode is coming from (for easier debugging). Blackfin arch: don't accidently re-enable interrupts
Blackfin arch: Fix URL
Blackfin arch: Add a few more instructions that can cause the trace buffer to be discontiguous Blackfin arch: Add ability to count and display number of NMI interrupts

Robin Holt (3):
sgi-xpc: ensure flags are updated before bte_copy
sgi-xpc: Remove NULL pointer dereference.
sgi-xpc: fix up stale DBUG_ON statements

Roel Kluin (10):
panasonic-laptop: fix X[ ARRAY_SIZE(X) ]
Staging: usbip: usbip_start_threads(): handle kernel_thread failure
Staging: poch: fix verification of memory area
smsc911x: timeout reaches -1
drivers/net/skfp: if !capable(CAP_NET_ADMIN): inverted logic
leds: Fix bounds checking of wm8350->pmic.led
ALSA: alsa: time reaches -1, tested 0
net: variables reach -1, but 0 tested
cassini/sungem: limit reaches -1, but 0 tested
ALSA: pcm_oss: AFMT_S24_LE is set twice in return value

Roland Dreier (1):
Btrfs: Remove extra KERN_INFO in the middle of a line

Roland McGrath (2):
elf core dump: fix get_user use
x86-64: fix int $0x80 -ENOSYS return

Ron Mercer (3):
qlge: bugfix: Fix endian issue when reading flash.
qlge: bugfix: Add flash offset for second port.
qlge: bugfix: Add missing netif_napi_del call.

Russell King (3):
[ARM] etherh: continue fixing build failure
[ARM] msm: fix build errors
[ARM] call undefined instruction exception handler with irqs enabled

Rusty Russell (3):
cpumask: convert lib/smp_processor_id to new cpumask ops
sched_rt: don't use first_cpu on cpumask created with cpumask_and
module: remove over-zealous check in __module_get()

Samuel Thibault (1):
Fix my email address in qd65xx.[ch]/pata_qdi.c

Sascha Hauer (1):
mmc: Add a MX2/MX3 specific SDHC driver

Sebastian Andrzej Siewior (1):

      V4L/DVB (10202): [PATCH] v4l/tvp514x: Don't write after line end

Sebastiano Di Paola (1):
net: packet socket packet_lookup_frame fix

Sebastien Dugue (1):
hrtimers: allow the hot-unplugging of all cpus

Sergei Shtylyov (1):
ide/libata: fix ata_id_is_cfa() (take 4)

Seth Heasley (1):
PCI: irq and pci_ids patch for Intel Tigerpoint DeviceIDs

Shyam Iyer (1):
net: Fix OOPS in skb_seq_read().

Sidney Amani (1):
UBI: allow direct user-space I/O

Simon Harrison (1):

      V4L/DVB (10210): Fix a bug on v4lgrab.c

Simon Holm Thøgersen (1):
ext4: fix wrong use of do_div

Sonic Zhang (2):
Blackfin arch: enable bfin_eth in bf51x by default Blackfin arch: Fix bug - Run "reboot" hangs bf518-ezbrd

Stanley.Miao (1):
ARM: OMAP: Fix McBSP spin_lock deadlock

Stefan Richter (18):
firewire: insist on successive self ID complete events
firewire: unnecessary BM delay after generation rollover
firewire: keep highlevel drivers attached during brief connection loss
firewire: ohci: change "context_stop: still active" log message
firewire: ohci: increase AT req. retries, fix ack_busy_X from Panasonic camcorders and others
ieee1394: ohci1394: increase AT req. retries, fix ack_busy_X from Panasonic camcorders and others
firewire: core: optimize card shutdown
ieee1394: support for speeds greater than S800
ieee1394: sbp2: update a help string
ieee1394: sbp2: fix payload limit at S1600 and S3200
ieee1394: sbp2: don't assume zero model_id or firmware_revision if there is none
firewire: sbp2: fix payload limit at S1600 and S3200
firewire: sbp2: define some magic numbers as macros
firewire: sbp2: fix DMA mapping leak on the failure path
firewire: sbp2: add workarounds for 2nd and 3rd generation iPods
ieee1394: sbp2: add workarounds for 2nd and 3rd generation iPods
ieee1394: dv1394: move deprecation message from module init to file open
async: use list_move_tail

Stephen Hemminger (1):
net: update documentation ip aliases

Stephen Rothwell (2):
powerpc/ps3: Printing fixups for l64 to ll64 convserion drivers/net
sparc: fixup for sparseirq changes

Steve French (4):
[CIFS] Rename md5 functions to avoid collision with new rt modules [CIFS] revalidate parent inode when rmdir done within that directory [CIFS] some cleanup to dir.c prior to addition of posix_open
[CIFS] Make socket retry timeouts consistent between blocking and nonblocking cases

Steve Glendinning (1):
smsc9420: fix interrupt signalling test failures

Steve Hodgson (6):
sfc: SFT9001: Enable robust link training
sfc: SFX7101: Remove workaround for bad link training
sfc: SFT9001: Fix speed reporting in 1G PHY loopback
sfc: Fix post-reset MAC selection
sfc: Reinitialise the PHY completely in case of a PHY or NIC reset
sfc: Test for PHYXS faults whenever we cannot test link state bits

Steven Rostedt (6):
ring-buffer: fix alignment problem
trace: print ftrace_dump at KERN_EMERG log level
trace: stop all recording to ring buffer on ftrace_dump
trace: set max latency variable to zero on default
ring-buffer: reset timestamps when ring buffer is reset
generic-ipi: use per cpu data for single cpu ipi calls

Stuart Menefy (1):
sh: Handle calling csum_partial with misaligned data

Sunil Mushran (1):
ocfs2: Wakeup the downconvert thread after a successful cancel convert

Takashi Iwai (8):
ALSA: hda - Fix PCM reference NID for STAC/IDT analog outputs
ALSA: hda - Fix compile warning with CONFIG_SND_JACK=n
ALSA: hda - Add quirk for another HP dv5 model
ALSA: hda - No widget selection for volume knob widgets in proc output
ALSA: hda - Add quirk for FSC Amilo Xi2550
ALSA: hda - Fix misc workqueue issues
ALSA: hda - Add missing initialization for ALC272
ALSA: hda - Add missing COEF initialization for ALC887

Takashi Yoshii (1):
sh: Fix up T-bit error handling in SH-4A mutex fastpath.

Tao Ma (1):
ocfs2: Access the xattr bucket only before modifying it.

Teemu Likonen (1):
Documentation: update CodingStyle tips for Emacs users

Tejun Heo (9):
sata_nv: ck804 has borked hardreset too
libata: fix EH device failure handling
libata: move ata_dev_disable() to libata-eh.c
libata: check onlineness before using SPD in sata_down_spd_limit()
libata: clear dev->ering in smarter way
libata: add @spd_limit to sata_down_spd_limit()
libata: improve probe failure handling
libata: add no penalty retry request for EH device handling routines
libata: implement HORKAGE_1_5_GBPS and apply it to WD My Book

Tero Roponen (1):
ACPI: Fix crash on ASUS laptops

Thadeu Lima de Souza Cascardo (1):
ext4: Fix building with EXT4FS_DEBUG

Theodore Ts'o (8):
ext4: only use i_size_high for regular files
ext4: Add sanity check to make_indexed_dir
ext3: Add sanity check to make_indexed_dir
jbd2: On a __journal_expect() assertion failure printk "JBD2", not "EXT3-fs"
ext4: Fix ext4_free_blocks() w/o a journal when files have indirect blocks
ext4: Initialize the new group descriptor when resizing the filesystem
ext4: Remove bogus BUG() check in ext4_bmap()
block: Fix documentation for blkdev_issue_flush()

Thierry Vignaud (1):
ACPI: Kconfig text - Fix the ACPI_CONTAINER module name according to the real module name.

Thomas Gleixner (1):
hrtimer: prevent negative expiry value after clock_was_set()

Thomas Goff (1):
IPv6: Fix multicast routing bugs.

Thomas Renninger (1):
ACPI: cpufreq: Remove deprecated /proc/acpi/processor/../performance proc entries

Tim 'mithro' Ansell (1):
lguest: disable the FORTIFY for lguest.

Timothy S. Nelson (1):
PCI: return error on failure to read PCI ROMs

Tony Battersby (1):
shm: fix shmctl(SHM_INFO) lockup with !CONFIG_SHMEM

Tony Broad (1):
V4L/DVB (10265): budget.c driver: Kernel oops: "BUG: unable to handle kernel paging request at ffffffff

Tony Lindgren (2):
ARM: OMAP: Fix omap34xx revision detection for ES3.1
ARM: OMAP: Fix hsmmc init, v2

Trent Piepho (6):
V4L/DVB (10212): Convert to be a pci driver
V4L/DVB (10222): zoran: Better syntax for initializing array module params V4L/DVB (10223): zoran: Remove global device array
V4L/DVB (10224): zoran: Use pci device table to get card type V4L/DVB (10225): zoran: Remove zr36057_adr field
V4L/DVB (10226): zoran: Get rid of extra module ref count

Tsugikazu Shibata (1):
Sync patch for jp_JP/stable_kernel_rules.txt

Tyler Hicks (1):
eCryptfs: Regression in unencrypted filename symlinks

Uwe Kleine-Koenig (1):
Blackfin arch: explicit add a might sleep to gpio_free

Uwe Kleine-König (6):
Use __SPIN_LOCK_UNLOCKED to initialize bad_irq_desc.lock annotate that [fp, #-4] is the saved lr
rename platform_driver name "flash" to "sa1100-mtd" NVRAM depends on RTC_DRV_CMOS
add another mailmap entry for Uwe Kleine-König Add Sascha Hauer to .mailmap

Wang Cong (1):
Btrfs: cleanup fs/btrfs/super.c::btrfs_control_ioctl()

Yan Zheng (4):
Btrfs: fix locking issue in btrfs_remove_block_group
Btrfs: Fix infinite loop in btrfs_extent_post_op
Btrfs: fix stop searching test in replace_one_extent
Btrfs: fix tree logs parallel sync

Yehuda Sadeh (1):
Btrfs: fiemap support

Yinghai Lu (1):
irq, x86: fix lock status with numa_migrate_irq_desc

Yusuf Altin (1):
V4L/DVB (10195): [PATCH] add Terratec Cinergy T Express to dibcom driver

Zhao Yakui (2):
ACPI: Skip the first two elements in the _BCL package
ACPI: proc_dir_entry 'video/VGA' already registered

Zhu, Yi (1):
iwlwifi: fix kernel oops when ucode DMA memory allocation failure

[email protected] (1):
ricoh_mmc: Use suspend_late/resume_early

김규원 (1):
ARM: OMAP: Mask interrupts when disabling interrupts, v2

#1 lore raw
Quoted 11 lines · On Sun, 8 Feb 2009 13:01:04 -0800 (PST) Linus Torvalds…

Another week (and a half), another -rc.

Arch updates (sparc, blackfin), driver updates (dvb, mmc, ide), ACPI updates, FS updates (ubifs, btrfs), you name it. It's all there.

But more importantly, people really have been working on regressions, and hopefully this closes a nice set of the top one, and hopefully without introducing too many new ones.

Linus

I just looked at the top -rc3 oopses/warnings/etc

quick summary of the top ones:
(manual rather than the normal automated scripts, so the format is different)
"regulars" that happened a lot prior to .29 (eg clearly non-regressions) I marked with a *

Oopses: 78 oopses reported
--------------------------
12 times i915_driver_irq_handler http://www.kerneloops.org/searchweek.php?search=i915_driver_irq_handler
8 times apic_mmio_read http://www.kerneloops.org/searchweek.php?search=apic_mmio_read

BUG/BUG_ON: 16 reports
--------------------------
6 times i915_gem_object_unpin http://www.kerneloops.org/searchweek.php?i915_gem_object_unpin
2 times __schedule http://www.kerneloops.org/searchweek.php?__schedule

WARN/WARN_ON/etc: 999 reports
-----------------------------------
213 times check_unmap http://www.kerneloops.org/searchweek.php?search=check_unmap
164 times last_sector_hacks http://www.kerneloops.org/searchweek.php?search=last_sector_hack
94 times *mpt_config http://www.kerneloops.org/searchweek.php?search=mpt_config
92 times reserve_pfn_range http://www.kerneloops.org/searchweek.php?search=reserve_pfn_range
79 times *init_ck804xrom http://www.kerneloops.org/searchweek.php?search=init_ck804xrom
56 times *iwl_set_dynamic_key http://www.kerneloops.org/searchweek.php?search=iwl_set_dynamic_key
39 times check_sync http://www.kerneloops.org/searchweek.php?search=check_sync
35 times track_pfn_vma_copy http://www.kerneloops.org/searchweek.php?search=track_pfn_vma_cop

Signature · 3 lines
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
#2 lore raw
Quoted 23 lines · Arjan van de Ven
Quoted 11 lines · On Sun, 8 Feb 2009 13:01:04 -0800 (PST) Linus Torvalds…

Another week (and a half), another -rc.

Arch updates (sparc, blackfin), driver updates (dvb, mmc, ide), ACPI updates, FS updates (ubifs, btrfs), you name it. It's all there.

But more importantly, people really have been working on regressions, and hopefully this closes a nice set of the top one, and hopefully without introducing too many new ones.

Linus

I just looked at the top -rc3 oopses/warnings/etc

quick summary of the top ones:
(manual rather than the normal automated scripts, so the format is different)
"regulars" that happened a lot prior to .29 (eg clearly non-regressions) I marked with a *

Quoted 4 lines

Oopses: 78 oopses reported
--------------------------
12 times i915_driver_irq_handler http://www.kerneloops.org/searchweek.php?search=i915_driver_irq_handler
8 times apic_mmio_read http://www.kerneloops.org/searchweek.php?search=apic_mmio_read

The last one seems KVM related.

Ingo

#5 lore raw
Quoted 7 lines · Gerd Hoffmann

Hmm, Linu*s* 2.6.29-rc4?
Tyops or hidden message?

/me remembers the "Linus 2.0" announcement ...

cheers,
Gerd

Funny part is I didn't notice it at all until you said something. I wonder how many other people didn't notice...

- Steven

#6 lore raw

Oopses and ACPI problems (Linus 2.6.29-rc4)

In reply to Linus Torvalds

I'm seeing various oopses with -rc4, where -rc3 is fine; config, dmesg (captured via netconsole) and decoded oops data attached ¹. The last few oopses were followed by a scrambled display and a hard lockup.

All oopses follow an otherwise successful suspend-to-RAM.

Also, I see that Fn-Fx don't seem to be reported via /proc/acpi/events any more, or at least I'm seeing no evidence of them via acpi_listen (and, not coincidentally, I can't adjust the volume etc. via eeepc-acpi-scripts).

The only other (significant) change is that I've installed X from Debian experimental, though -rc3 and 2.6.28.* seem fine with this. (Oh, and I can get UXA to cause OOM quite trivially... several Iceweasel tabs is enough.)

¹ Yes, I decided not to include symbol names in the running kernel...

Signature · 5 lines
| Darren Salt    | linux or ds at              | nr. Ashington, | Toon
| RISC OS, Linux | youmustbejoking,demon,co,uk | Northumberland | Army
| + Use more efficient products. Use less.          BE MORE ENERGY EFFICIENT.

I'll have what the bloke on the floor is having...

ksymoops 2.4.11 on i686 2.6.29-rc4-eee901. Options used
-V (default)
-k /proc/ksyms (default)
-l /proc/modules (default)
-o /lib/modules/2.6.29-rc4-eee901/ (default) -m /boot/System.map-2.6.29-rc4-eee901 (default)

Warning: You did not tell me where to find symbol information. I will assume that the log matches the kernel and modules that are running right now and I'll use the default options above for symbol resolution. If the current kernel and/or modules do not match the log, you can get more accurate output by telling me the kernel version and where to find map, modules, ksyms etc. ksymoops -h explains the options.

Error (regular_file): read_ksyms stat /proc/ksyms failed No modules in ksyms, skipping objects
No ksyms, skipping lsmod
CPU 1 is now offline

WARNING: at fs/sysfs/group.c:138 ()
Pid: 330, comm: pciehpd Not tainted 2.6.29-rc4-eee901 #2
Call Trace:
 [<c011fab7>] 0xc011fab7
 [<c032ca9d>] 0xc032ca9d
 [<f81760d5>] 0xf81760d5
 [<c01f112c>] 0xc01f112c
 [<c0118940>] 0xc0118940
 [<c011fc82>] 0xc011fc82
 [<c016b8aa>] 0xc016b8aa
 [<c0190b6b>] 0xc0190b6b
 [<c016b9a8>] 0xc016b9a8
 [<c016b738>] 0xc016b738
 [<c019102a>] 0xc019102a
 [<c0191faa>] 0xc0191faa
 [<c025a8d4>] 0xc025a8d4
 [<c025a9c9>] 0xc025a9c9
 [<c01fae3c>] 0xc01fae3c
 [<c01faf00>] 0xc01faf00
 [<c0205ca4>] 0xc0205ca4
 [<c0204a68>] 0xc0204a68
 [<c0204eb3>] 0xc0204eb3
 [<c0204e2d>] 0xc0204e2d
 [<c012bd8f>] 0xc012bd8f
 [<c012c51d>] 0xc012c51d
 [<c012c5d1>] 0xc012c5d1
 [<c012e9ac>] 0xc012e9ac
 [<c012c51d>] 0xc012c51d
 [<c012e8de>] 0xc012e8de
 [<c012e8a6>] 0xc012e8a6
 [<c010349b>] 0xc010349b
BUG: unable to handle kernel NULL pointer dereference at 00000010
*pde = 00000000 
Oops: 0000 [#1] PREEMPT SMP 

Pid: 330, comm: pciehpd Tainted: G W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c03ccef7>] EFLAGS: 00010246 CPU: 0
Using defaults from ksymoops -t elf32-i386 -a i386
EAX: 00000000 EBX: 00000001 ECX: 00000046 EDX: 00000001
ESI: f17dd470 EDI: f70f7c5c EBP: 00000000 ESP: f7339f00
DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
f17dd45c 00000000 f70f7c5c c025a9c9 f7060b40 f7311a80ff 83 c0 89 83 cf db 74 19 f6 06 01 74 11 6a 00 ---[ end trace 449537779ba90603 ]---
Warning (Oops_read): Code line not seen, dumping what data is available

Trace; c011fab7 <warn_slowpath+7d/94> Trace; c032ca9d <__alloc_skb+4b/103>
Trace; f81760d5 <END_OF_CODE+37c160d5/????> Trace; c01f112c <cpumask_next_and+23/33> Trace; c0118940 <find_busiest_group+242/725> Trace; c011fc82 <__call_console_drivers+51/5d> Trace; c016b8aa <find_inode+1b/56>
Trace; c0190b6b <sysfs_ilookup_test+0/d> Trace; c016b9a8 <ifind+3a/74>
Trace; c016b738 <iput+21/4a>
Trace; c019102a <sysfs_addrm_finish+4a/1ae> Trace; c0191faa <sysfs_remove_group+34/8d> Trace; c025a8d4 <device_del+32/11f>
Trace; c025a9c9 <device_unregister+8/10> Trace; c01fae3c <pci_stop_dev+20/3a>
Trace; c01faf00 <pci_remove_bus_device+27/88> Trace; c0205ca4 <pciehp_unconfigure_device+108/180> Trace; c0204a68 <pciehp_disable_slot+12d/1d6> Trace; c0204eb3 <pciehp_power_thread+86/cd> Trace; c0204e2d <pciehp_power_thread+0/cd> Trace; c012bd8f <run_workqueue+71/f1> Trace; c012c51d <worker_thread+0/bf>
Trace; c012c5d1 <worker_thread+b4/bf> Trace; c012e9ac <autoremove_wake_function+0/2b> Trace; c012c51d <worker_thread+0/bf>
Trace; c012e8de <kthread+38/5e>
Trace; c012e8a6 <kthread+0/5e>
Trace; c010349b <kernel_thread_helper+7/10>

Quoted 1 line
Quoted 1 line

EIP; c03ccef7 <klist_put+f/59> <=====

BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#2] PREEMPT SMP 

Pid: 4094, comm: bash Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015bb0a>] EFLAGS: 00010006 CPU: 1
EAX: 00000000 EBX: c04c7d90 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00000282 EDI: c04c7d90 EBP: 00000040 ESP: ee9f1e38
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000080d0 f173f7c0 c194c488 7ac532a8 fe280a2e c194c488 c0194004 f173f7c0 c194c488 f6e64498 0000003e f6e02458 c019aaa0 c194c488 00000000 ee949600 c194cff8 00000000 ee949600 ee949600 f6e02458 00000000 c019ab55

Call Trace:
 [<c0194004>] 0xc0194004
 [<c019aaa0>] 0xc019aaa0
 [<c019ab55>] 0xc019ab55
 [<c01405ce>] 0xc01405ce
 [<c0102cf1>] 0xc0102cf1
56 BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#3] PREEMPT SMP 

Pid: 3147, comm: xfce4-systemloa Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015b4e5>] EFLAGS: 00010006 CPU: 1
EAX: 00000000 EBX: 00000292 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00000040 EDI: c018d53c EBP: c04c7d90 ESP: f16ebe34
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000000d0 f702c00a ee949300 f73e35e0 c018d53c 00000000 c017059b 00000000 f73e35e0 fffffff4 c018d53c ee949300 c0170678 00000000 f7029000 fffffff4 00001000 ee949300 c018d51e f173f2a0 f70e9680 c018d4d0 f70e96cc

Call Trace:
 [<c018d53c>] 0xc018d53c
 [<c017059b>] 0xc017059b
 [<c018d53c>] 0xc018d53c
 [<c0170678>] 0xc0170678
 [<c018d51e>] 0xc018d51e
 [<c018d4d0>] 0xc018d4d0
 [<c01887b8>] 0xc01887b8
 [<c017006d>] 0xc017006d
 [<c012e9ac>] 0xc012e9ac
 [<c0102cf1>] 0xc0102cf1
BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#4] PREEMPT SMP 

Pid: 3137, comm: xfce4-panel Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015b4e5>] EFLAGS: 00010006 CPU: 1
EAX: 00000000 EBX: 00000296 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00000040 EDI: ee939f7c EBP: c04c7d90 ESP: f1653ecc
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000000d0 00000000 ee814440 00000000 ee939f7c 01200011 c012c900 ee938380 eea4fff8 c048d53c ee814440 c0101a87 01200011 ee814440 00000000 ee939f7c ee814440 00000000 ee939f7c 01200011 c011e9cb f1653fb8 bf9e8d30

Call Trace:
 [<c012c900>] 0xc012c900
 [<c0101a87>] 0xc0101a87
 [<c011e9cb>] 0xc011e9cb
 [<c011f085>] 0xc011f085
eb BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#5] PREEMPT SMP 

Pid: 2942, comm: Xorg Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015bb0a>] EFLAGS: 00213006 CPU: 1
EAX: 00000000 EBX: c04c7d90 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00203282 EDI: c04c7d90 EBP: 00000040 ESP: f68adef4
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000000d0 00203286 00000000 fffffff4 40286454 f6a86800 f81ea6f4 f6b2d420 f8348c59 f83583e4 f6950080 bfbc1eb4 40286454 c0167680 bfbc1eb4 f6950080 f6f1e7f8 08a84b40 f68ac000 c0167bb4 0000000d 00203082 f68adf9c

Call Trace:
 [<f81ea6f4>] 0xf81ea6f4
 [<f8348c59>] 0xf8348c59
 [<c0167680>] 0xc0167680
 [<c0167bb4>] 0xc0167bb4
 [<c0122ac3>] 0xc0122ac3
 [<c0122c59>] 0xc0122c59
f4 BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#6] PREEMPT SMP 

Pid: 1219, comm: udevd Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015b4e5>] EFLAGS: 00010006 CPU: 1
EAX: 00000000 EBX: 00000292 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00000040 EDI: c018d53c EBP: c04c7d90 ESP: f68fbe34
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000000d0 f708c00a ee949c80 f73e3720 c018d53c 00000000 c017059b f6c12964 f73e3720 fffffff4 c018d53c ee949c80 c0170678 00000000 f708e000 fffffff4 00001000 ee949c80 c018d51e f173fa20 f70e9680 c018d4d0 f70e96cc

Call Trace:
 [<c018d53c>] 0xc018d53c
 [<c017059b>] 0xc017059b
 [<c018d53c>] 0xc018d53c
 [<c0170678>] 0xc0170678

Warning (Oops_read): Code line not seen, dumping what data is available

Quoted 5 lines
Quoted 5 lines

EIP; c015bb0a <__kmalloc+83/c0> <=====

EBX; c04c7d90 <kmalloc_caches+210/480>
ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
EDI; c04c7d90 <kmalloc_caches+210/480>

Trace; c0194004 <ext3_htree_store_dirent+26/de>
Trace; c019aaa0 <htree_dirblock_to_tree+d6/11b>
Trace; c019ab55 <ext3_htree_fill_tree+70/1cb>
Trace; c01405ce <find_lock_page+10/43>
Trace; c0102cf1 <sysenter_do_call+12/25>

Quoted 5 lines
Quoted 5 lines

EIP; c015b4e5 <kmem_cache_alloc+48/83> <=====

ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
EDI; c018d53c <show_stat+0/680>
EBP; c04c7d90 <kmalloc_caches+210/480>

Trace; c018d53c <show_stat+0/680>
Trace; c017059b <seq_open+23/6d>
Trace; c018d53c <show_stat+0/680>
Trace; c0170678 <single_open+48/68>
Trace; c018d51e <stat_open+4e/6c>
Trace; c018d4d0 <stat_open+0/6c>
Trace; c01887b8 <proc_reg_open+89/ed>
Trace; c017006d <single_release+0/1b>
Trace; c012e9ac <autoremove_wake_function+0/2b>
Trace; c0102cf1 <sysenter_do_call+12/25>

Quoted 4 lines
Quoted 4 lines

EIP; c015b4e5 <kmem_cache_alloc+48/83> <=====

ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
EBP; c04c7d90 <kmalloc_caches+210/480>

Trace; c012c900 <alloc_pid+18/2b6>
Trace; c0101a87 <copy_thread+2f/f3>
Trace; c011e9cb <copy_process+917/e42>
Trace; c011f085 <do_fork+11a/268>

Quoted 6 lines
Quoted 6 lines

EIP; c015bb0a <__kmalloc+83/c0> <=====

EBX; c04c7d90 <kmalloc_caches+210/480>
ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
ESI; 00203282 <phys_startup_32+103282/c0000000>
EDI; c04c7d90 <kmalloc_caches+210/480>

Trace; f81ea6f4 <END_OF_CODE+37c8a6f4/????>
Trace; f8348c59 <END_OF_CODE+37de8c59/????>
Trace; c0167680 <vfs_ioctl+47/5d>
Trace; c0167bb4 <do_vfs_ioctl+446/47c>
Trace; c0122ac3 <do_setitimer+157/2a9>
Trace; c0122c59 <sys_setitimer+44/71>

Quoted 5 lines
Quoted 5 lines

EIP; c015b4e5 <kmem_cache_alloc+48/83> <=====

ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
EDI; c018d53c <show_stat+0/680>
EBP; c04c7d90 <kmalloc_caches+210/480>

Trace; c018d53c <show_stat+0/680>
Trace; c017059b <seq_open+23/6d>
Trace; c018d53c <show_stat+0/680>
Trace; c0170678 <single_open+48/68>

3 warnings and 1 error issued. Results may not be reliable.

#7 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

Quoted 5 lines · Darren Salt

I'm seeing various oopses with -rc4, where -rc3 is fine; config, dmesg (captured via netconsole) and decoded oops data attached ¹. The last few oopses were followed by a scrambled display and a hard lockup.

All oopses follow an otherwise successful suspend-to-RAM.

Various folks Cc:-ed. There's been a late, rather large ACPI merge, so one post-rc3 suspect would be one of these commits:

f12b12a: Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6
2d29c6a: Merge branches 'release', 'asus', 'bugzilla-12450', 'cpuidle', 'debug', 'ec', 'misc', 'printk' and 'processor' into release 370154b: ACPI: Kconfig text - Fix the ACPI_CONTAINER module name according to the real module name.
7695fb0: eeepc-laptop: fix oops when changing backlight brightness during eeepc-laptop init
386e4a8: ACPICA: Fix table entry truncation calculation
d96f94c: ACPI: Enable bit 11 in _PDC to advertise hw coord
db1461a: ACPI: struct device - replace bus_id with dev_name(), dev_set_name()
4d93915: ACPI: add missing KERN_* constants to printks
fc5a9f8: ACPI: dock: Don't eval _STA on every show_docked sysfs read
9e3a9d1: ACPI: disable ACPI cleanly when bad RSDP found
9fdd54f: ACPI: delete CPU_IDLE=n code
62663ea: ACPI: cpufreq: Remove deprecated /proc/acpi/processor/../performance proc entries
5ec5d38: ACPI: make some IO ports off-limits to AML
3e0676a: ACPICA: add debug dump of BIOS _OSI strings
f3b39f1: ACPI: proc_dir_entry 'video/VGA' already registered
0a3db1c: ACPI: Skip the first two elements in the _BCL package
31878dd: ACPI: remove BM_RLD access from idle entry path
a2b7b01: ACPI: remove locking from PM1x_STS register reads
2b25c9f: eeepc-laptop: use netlink interface
5740294: eeepc-laptop: Implement rfkill hotplugging in eeepc-laptop
c9ddf8f: eeepc-laptop: Check return values from rfkill_register
b5f6f26: eeepc-laptop: Add support for extended hotkeys
ed6f442: asus-laptop: fix label indentation
12d6f35: asus-laptop: update Kconfig for input layer
034ce90: asus-laptop: hotkeys via the generic input interface
2a7dc0d: asus-laptop: use generic netlink interface
1021e21: asus_acpi: Add R1F support
a9df80c: eeepc-laptop: split eeepc_backlight_exit()
2b190e7: panasonic-laptop: fix X[ ARRAY_SIZE(X) ]
4312495: ACPI: Fix crash on ASUS laptops
5d8b532: ACPI suspend: Fix compilation warnings in drivers/acpi/sleep.c
091d71e: PM: Fix compilation warning in kernel/power/main.c

Quoted 4 lines

Also, I see that Fn-Fx don't seem to be reported via /proc/acpi/events any more, or at least I'm seeing no evidence of them via acpi_listen (and, not coincidentally, I can't adjust the volume etc. via eeepc-acpi-scripts).

Quoted 267 lines

The only other (significant) change is that I've installed X from Debian experimental, though -rc3 and 2.6.28.* seem fine with this. (Oh, and I can get UXA to cause OOM quite trivially... several Iceweasel tabs is enough.)

¹ Yes, I decided not to include symbol names in the running kernel...

Signature · 98 lines
| Darren Salt    | linux or ds at              | nr. Ashington, | Toon
| RISC OS, Linux | youmustbejoking,demon,co,uk | Northumberland | Army
| + Use more efficient products. Use less.          BE MORE ENERGY EFFICIENT.

I'll have what the bloke on the floor is having...



ksymoops 2.4.11 on i686 2.6.29-rc4-eee901.  Options used
     -V (default)
     -k /proc/ksyms (default)
     -l /proc/modules (default)
     -o /lib/modules/2.6.29-rc4-eee901/ (default)
     -m /boot/System.map-2.6.29-rc4-eee901 (default)

Warning: You did not tell me where to find symbol information.  I will
assume that the log matches the kernel and modules that are running
right now and I'll use the default options above for symbol resolution.
If the current kernel and/or modules do not match the log, you can get
more accurate output by telling me the kernel version and where to find
map, modules, ksyms etc.  ksymoops -h explains the options.

Error (regular_file): read_ksyms stat /proc/ksyms failed
No modules in ksyms, skipping objects
No ksyms, skipping lsmod
CPU 1 is now offline
WARNING: at fs/sysfs/group.c:138 ()
Pid: 330, comm: pciehpd Not tainted 2.6.29-rc4-eee901 #2
Call Trace:
 [<c011fab7>] 0xc011fab7
 [<c032ca9d>] 0xc032ca9d
 [<f81760d5>] 0xf81760d5
 [<c01f112c>] 0xc01f112c
 [<c0118940>] 0xc0118940
 [<c011fc82>] 0xc011fc82
 [<c016b8aa>] 0xc016b8aa
 [<c0190b6b>] 0xc0190b6b
 [<c016b9a8>] 0xc016b9a8
 [<c016b738>] 0xc016b738
 [<c019102a>] 0xc019102a
 [<c0191faa>] 0xc0191faa
 [<c025a8d4>] 0xc025a8d4
 [<c025a9c9>] 0xc025a9c9
 [<c01fae3c>] 0xc01fae3c
 [<c01faf00>] 0xc01faf00
 [<c0205ca4>] 0xc0205ca4
 [<c0204a68>] 0xc0204a68
 [<c0204eb3>] 0xc0204eb3
 [<c0204e2d>] 0xc0204e2d
 [<c012bd8f>] 0xc012bd8f
 [<c012c51d>] 0xc012c51d
 [<c012c5d1>] 0xc012c5d1
 [<c012e9ac>] 0xc012e9ac
 [<c012c51d>] 0xc012c51d
 [<c012e8de>] 0xc012e8de
 [<c012e8a6>] 0xc012e8a6
 [<c010349b>] 0xc010349b
BUG: unable to handle kernel NULL pointer dereference at 00000010
*pde = 00000000 
Oops: 0000 [#1] PREEMPT SMP 
Pid: 330, comm: pciehpd Tainted: G        W  (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c03ccef7>] EFLAGS: 00010246 CPU: 0
Using defaults from ksymoops -t elf32-i386 -a i386
EAX: 00000000 EBX: 00000001 ECX: 00000046 EDX: 00000001
ESI: f17dd470 EDI: f70f7c5c EBP: 00000000 ESP: f7339f00
 DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
 f17dd45c 00000000 f70f7c5c c025a9c9 f7060b40 f7311a80ff 83 c0 89 83 cf db 74 19 f6 06 01 74 11 6a 00 ---[ end trace 449537779ba90603 ]---
Warning (Oops_read): Code line not seen, dumping what data is available


Trace; c011fab7 <warn_slowpath+7d/94>
Trace; c032ca9d <__alloc_skb+4b/103>
Trace; f81760d5 <END_OF_CODE+37c160d5/????>
Trace; c01f112c <cpumask_next_and+23/33>
Trace; c0118940 <find_busiest_group+242/725>
Trace; c011fc82 <__call_console_drivers+51/5d>
Trace; c016b8aa <find_inode+1b/56>
Trace; c0190b6b <sysfs_ilookup_test+0/d>
Trace; c016b9a8 <ifind+3a/74>
Trace; c016b738 <iput+21/4a>
Trace; c019102a <sysfs_addrm_finish+4a/1ae>
Trace; c0191faa <sysfs_remove_group+34/8d>
Trace; c025a8d4 <device_del+32/11f>
Trace; c025a9c9 <device_unregister+8/10>
Trace; c01fae3c <pci_stop_dev+20/3a>
Trace; c01faf00 <pci_remove_bus_device+27/88>
Trace; c0205ca4 <pciehp_unconfigure_device+108/180>
Trace; c0204a68 <pciehp_disable_slot+12d/1d6>
Trace; c0204eb3 <pciehp_power_thread+86/cd>
Trace; c0204e2d <pciehp_power_thread+0/cd>
Trace; c012bd8f <run_workqueue+71/f1>
Trace; c012c51d <worker_thread+0/bf>
Trace; c012c5d1 <worker_thread+b4/bf>
Trace; c012e9ac <autoremove_wake_function+0/2b>
Trace; c012c51d <worker_thread+0/bf>
Trace; c012e8de <kthread+38/5e>
Trace; c012e8a6 <kthread+0/5e>
Trace; c010349b <kernel_thread_helper+7/10>
Quoted 1 line
Quoted 1 line

EIP; c03ccef7 <klist_put+f/59> <=====

BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#2] PREEMPT SMP 

Pid: 4094, comm: bash Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015bb0a>] EFLAGS: 00010006 CPU: 1
EAX: 00000000 EBX: c04c7d90 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00000282 EDI: c04c7d90 EBP: 00000040 ESP: ee9f1e38
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000080d0 f173f7c0 c194c488 7ac532a8 fe280a2e c194c488 c0194004 f173f7c0 c194c488 f6e64498 0000003e f6e02458 c019aaa0 c194c488 00000000 ee949600 c194cff8 00000000 ee949600 ee949600 f6e02458 00000000 c019ab55

Call Trace:
 [<c0194004>] 0xc0194004
 [<c019aaa0>] 0xc019aaa0
 [<c019ab55>] 0xc019ab55
 [<c01405ce>] 0xc01405ce
 [<c0102cf1>] 0xc0102cf1
56 BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#3] PREEMPT SMP 

Pid: 3147, comm: xfce4-systemloa Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015b4e5>] EFLAGS: 00010006 CPU: 1
EAX: 00000000 EBX: 00000292 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00000040 EDI: c018d53c EBP: c04c7d90 ESP: f16ebe34
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000000d0 f702c00a ee949300 f73e35e0 c018d53c 00000000 c017059b 00000000 f73e35e0 fffffff4 c018d53c ee949300 c0170678 00000000 f7029000 fffffff4 00001000 ee949300 c018d51e f173f2a0 f70e9680 c018d4d0 f70e96cc

Call Trace:
 [<c018d53c>] 0xc018d53c
 [<c017059b>] 0xc017059b
 [<c018d53c>] 0xc018d53c
 [<c0170678>] 0xc0170678
 [<c018d51e>] 0xc018d51e
 [<c018d4d0>] 0xc018d4d0
 [<c01887b8>] 0xc01887b8
 [<c017006d>] 0xc017006d
 [<c012e9ac>] 0xc012e9ac
 [<c0102cf1>] 0xc0102cf1
BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#4] PREEMPT SMP 

Pid: 3137, comm: xfce4-panel Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015b4e5>] EFLAGS: 00010006 CPU: 1
EAX: 00000000 EBX: 00000296 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00000040 EDI: ee939f7c EBP: c04c7d90 ESP: f1653ecc
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000000d0 00000000 ee814440 00000000 ee939f7c 01200011 c012c900 ee938380 eea4fff8 c048d53c ee814440 c0101a87 01200011 ee814440 00000000 ee939f7c ee814440 00000000 ee939f7c 01200011 c011e9cb f1653fb8 bf9e8d30

Call Trace:
 [<c012c900>] 0xc012c900
 [<c0101a87>] 0xc0101a87
 [<c011e9cb>] 0xc011e9cb
 [<c011f085>] 0xc011f085
eb BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#5] PREEMPT SMP 

Pid: 2942, comm: Xorg Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015bb0a>] EFLAGS: 00213006 CPU: 1
EAX: 00000000 EBX: c04c7d90 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00203282 EDI: c04c7d90 EBP: 00000040 ESP: f68adef4
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000000d0 00203286 00000000 fffffff4 40286454 f6a86800 f81ea6f4 f6b2d420 f8348c59 f83583e4 f6950080 bfbc1eb4 40286454 c0167680 bfbc1eb4 f6950080 f6f1e7f8 08a84b40 f68ac000 c0167bb4 0000000d 00203082 f68adf9c

Call Trace:
 [<f81ea6f4>] 0xf81ea6f4
 [<f8348c59>] 0xf8348c59
 [<c0167680>] 0xc0167680
 [<c0167bb4>] 0xc0167bb4
 [<c0122ac3>] 0xc0122ac3
 [<c0122c59>] 0xc0122c59
f4 BUG: unable to handle kernel paging request at 7a9d5f5c
*pde = 00000000 
Oops: 0000 [#6] PREEMPT SMP 

Pid: 1219, comm: udevd Tainted: G D W (2.6.29-rc4-eee901 #2) 901
EIP: 0060:[<c015b4e5>] EFLAGS: 00010006 CPU: 1
EAX: 00000000 EBX: 00000292 ECX: 7a9d5f5c EDX: c180ff6c
ESI: 00000040 EDI: c018d53c EBP: c04c7d90 ESP: f68fbe34
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
7a9d5f5c 000000d0 f708c00a ee949c80 f73e3720 c018d53c 00000000 c017059b f6c12964 f73e3720 fffffff4 c018d53c ee949c80 c0170678 00000000 f708e000 fffffff4 00001000 ee949c80 c018d51e f173fa20 f70e9680 c018d4d0 f70e96cc

Call Trace:
 [<c018d53c>] 0xc018d53c
 [<c017059b>] 0xc017059b
 [<c018d53c>] 0xc018d53c
 [<c0170678>] 0xc0170678

Warning (Oops_read): Code line not seen, dumping what data is available

Quoted 5 lines
Quoted 5 lines

EIP; c015bb0a <__kmalloc+83/c0> <=====

EBX; c04c7d90 <kmalloc_caches+210/480>
ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
EDI; c04c7d90 <kmalloc_caches+210/480>

Trace; c0194004 <ext3_htree_store_dirent+26/de>
Trace; c019aaa0 <htree_dirblock_to_tree+d6/11b>
Trace; c019ab55 <ext3_htree_fill_tree+70/1cb>
Trace; c01405ce <find_lock_page+10/43>
Trace; c0102cf1 <sysenter_do_call+12/25>

Quoted 5 lines
Quoted 5 lines

EIP; c015b4e5 <kmem_cache_alloc+48/83> <=====

ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
EDI; c018d53c <show_stat+0/680>
EBP; c04c7d90 <kmalloc_caches+210/480>

Trace; c018d53c <show_stat+0/680>
Trace; c017059b <seq_open+23/6d>
Trace; c018d53c <show_stat+0/680>
Trace; c0170678 <single_open+48/68>
Trace; c018d51e <stat_open+4e/6c>
Trace; c018d4d0 <stat_open+0/6c>
Trace; c01887b8 <proc_reg_open+89/ed>
Trace; c017006d <single_release+0/1b>
Trace; c012e9ac <autoremove_wake_function+0/2b>
Trace; c0102cf1 <sysenter_do_call+12/25>

Quoted 4 lines
Quoted 4 lines

EIP; c015b4e5 <kmem_cache_alloc+48/83> <=====

ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
EBP; c04c7d90 <kmalloc_caches+210/480>

Trace; c012c900 <alloc_pid+18/2b6>
Trace; c0101a87 <copy_thread+2f/f3>
Trace; c011e9cb <copy_process+917/e42>
Trace; c011f085 <do_fork+11a/268>

Quoted 6 lines
Quoted 6 lines

EIP; c015bb0a <__kmalloc+83/c0> <=====

EBX; c04c7d90 <kmalloc_caches+210/480>
ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
ESI; 00203282 <phys_startup_32+103282/c0000000>
EDI; c04c7d90 <kmalloc_caches+210/480>

Trace; f81ea6f4 <END_OF_CODE+37c8a6f4/????>
Trace; f8348c59 <END_OF_CODE+37de8c59/????>
Trace; c0167680 <vfs_ioctl+47/5d>
Trace; c0167bb4 <do_vfs_ioctl+446/47c>
Trace; c0122ac3 <do_setitimer+157/2a9>
Trace; c0122c59 <sys_setitimer+44/71>

Quoted 5 lines
Quoted 5 lines

EIP; c015b4e5 <kmem_cache_alloc+48/83> <=====

ECX; 7a9d5f5c <phys_startup_32+7a8d5f5c/c0000000>
EDI; c018d53c <show_stat+0/680>
EBP; c04c7d90 <kmalloc_caches+210/480>

Trace; c018d53c <show_stat+0/680>
Trace; c017059b <seq_open+23/6d>
Trace; c018d53c <show_stat+0/680>
Trace; c0170678 <single_open+48/68>

3 warnings and 1 error issued. Results may not be reliable.

--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#8 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

[Adding Matthew Garrett & Corentin Chary to Cc]

I demand that Ingo Molnar may or may not have written...

Quoted 9 lines · Darren Salt
Quoted 5 lines

I'm seeing various oopses with -rc4, where -rc3 is fine; config, dmesg (captured via netconsole) and decoded oops data attached ¹. The last few oopses were followed by a scrambled display and a hard lockup.

All oopses follow an otherwise successful suspend-to-RAM.

Various folks Cc:-ed. There's been a late, rather large ACPI merge, so one post-rc3 suspect would be one of these commits:

[snip]

Quoted 1 line

7695fb0: eeepc-laptop: fix oops when changing backlight brightness during

eeepc-laptop init

No, that one's harmless :-)

[snip]

Quoted 1 line

2b25c9f: eeepc-laptop: use netlink interface

That one makes acpid (apparently) useless. The rest of the commit description is:
To be prepared for /proc/acpi/event removal we export events also through generic netlink interface.

Note that "also". The actual effect of the patch is "instead".

Adding the two removed lines (but removing the post-increment of the event count) fixes this.

Quoted 1 line

5740294: eeepc-laptop: Implement rfkill hotplugging in eeepc-laptop

Reverting this fixes the rfkill oops; things work correctly again. I don't know why this appears to be needed, since things work fine without this patch ‒ both Fn-F2 and echoing to /sys/class/rfkill/rfkill0/state ‒ even when I start up with init=/bin/sh and test directly from that shell.

(901; BIOS rev. 1808.)

[snip]

Quoted 1 line

2a7dc0d: asus-laptop: use generic netlink interface

That one also says "also" where it should say "instead" (not that it affects me).

... oh yes, there's a long delay between output of these two lines during boot:
input: ETPS/2 Elantech Touchpad as /class/input/input5
eeepc: Hotkey init flags 0x41
(not seen in .28 or .28.1; not checked later .28.*.)

Anyway, a patch for the ACPI reporting problems will follow this message.

[snip]

Signature · 9 lines
| Darren Salt    | linux or ds at              | nr. Ashington, | Toon
| RISC OS, Linux | youmustbejoking,demon,co,uk | Northumberland | Army
| + Output less CO2 => avoid massive flooding.    TIME IS RUNNING OUT *FAST*.

Life is that brief interlude between nothingness and eternity.
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#9 lore raw

PATCH 1/1 Restore ACPI reporting via /proc/acpi/events for EeePC & other Asus laptops.

The descriptions for the following two commits
2a7dc0d8c60325e9bf820900bf919430e5a419ab (asus-laptop) 2b25c9f01aa58d48129b2f93748dfb5d1f7ab0a2 (eeepc-laptop)
say that events should ALSO be exported via netlink, but the effect is that they are INSTEAD exported via netlink. The latter commit breaks use of acpid on my EeePC 901.

(quoted diff) +0 -0

b/drivers/platform/x86/asus-laptop.c
index 56af6cf..42ecf6a 100644

drivers/platform/x86/asus-laptop.c +0 -0
@@ -832,6 +832,8 @@ static void asus_hotk_notify(acpi_handle handle, u32

event, void *data)
lcd_blank(FB_BLANK_POWERDOWN);
}

+ acpi_bus_generate_proc_event(ehotk->device, event,
+ ehotk->event_count[event % 128]);
acpi_bus_generate_netlink_event(hotk->device->pnp.device_class,
dev_name(&hotk->device->dev), event, hotk->event_count[event % 128]++);

(quoted diff) +0 -0

b/drivers/platform/x86/eeepc-laptop.c
index 786ed86..8511891 100644

drivers/platform/x86/eeepc-laptop.c +0 -0
@@ -561,6 +561,8 @@ static void eeepc_hotk_notify(acpi_handle handle, u32

event, void *data)
return;
if (event >= NOTIFY_BRN_MIN && event <= NOTIFY_BRN_MAX)
notify_brn();
+ acpi_bus_generate_proc_event(ehotk->device, event,
+ ehotk->event_count[event % 128]);
acpi_bus_generate_netlink_event(ehotk->device->pnp.device_class,
dev_name(&ehotk->device->dev), event, ehotk->event_count[event % 128]++);

Signature · 5 lines
| Darren Salt    | d @ youmustbejoking,demon,co,uk | nr. Ashington, | Toon
| RISC OS, Linux | s   zap,tartarus,org            | Northumberland | Army
| + Use more efficient products. Use less.          BE MORE ENERGY EFFICIENT.

Never drink from your finger bowl; it contains only water.
#10 lore raw

Re: [PATCH 2.6.29-rc4] Restore ACPI reporting via /proc/acpi/events for EeePC & other Asus laptops.

Quoted 6 lines · On Tuesday 10 February 2009 15:54:56 Darren Salt wrote:

The descriptions for the following two commits
2a7dc0d8c60325e9bf820900bf919430e5a419ab (asus-laptop) 2b25c9f01aa58d48129b2f93748dfb5d1f7ab0a2 (eeepc-laptop)
say that events should ALSO be exported via netlink, but the effect is that they are INSTEAD exported via netlink. The latter commit breaks use of acpid on my EeePC 901.

Hi,
There was already two patch for that on the ml

http://patchwork.kernel.org/patch/7356/
http://patchwork.kernel.org/patch/7357/

Thanks

Signature · 2 lines
Corentin Chary
http://xf.iksaif.net
#11 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

In reply to Darren Salt
Quoted 4 lines · Darren Salt

Reverting this fixes the rfkill oops; things work correctly again. I don't know why this appears to be needed, since things work fine without this patch ‒ both Fn-F2 and echoing to /sys/class/rfkill/rfkill0/state ‒ even when I start up with init=/bin/sh and test directly from that shell.

If this is BIOS version specific I'm going to be upset, but the general case behaviour is that rfkill only works on the 901 if you either (a) pass force=1 to pciehp (which you shouldn't) or (b) have native support in a driver. My /guess/ is that you've got pciehp loaded, and there's some kind of awkward race between it and eeepc-laptop.

Quoted 3 lines

input: ETPS/2 Elantech Touchpad as /class/input/input5
eeepc: Hotkey init flags 0x41
(not seen in .28 or .28.1; not checked later .28.*.)

BIOS bug. There's an explicit delay in the eee bios for some reason, and I haven't found any straightforward way to avoid it.

Signature · 5 lines
Matthew Garrett | [email protected]
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#12 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

I demand that Matthew Garrett may or may not have written...

Quoted 11 lines · Darren Salt
Quoted 4 lines

Reverting this fixes the rfkill oops; things work correctly again. I don't know why this appears to be needed, since things work fine without this patch ‒ both Fn-F2 and echoing to /sys/class/rfkill/rfkill0/state ‒ even when I start up with init=/bin/sh and test directly from that shell.

If this is BIOS version specific I'm going to be upset, but the general case behaviour is that rfkill only works on the 901 if you either (a) pass force=1 to pciehp (which you shouldn't) or (b) have native support in a driver. My /guess/ is that you've got pciehp loaded, and there's some kind of awkward race between it and eeepc-laptop.

I have; and yes, pciehp_force=1.

As things stand, this is likely to be a problem for anybody using lenny on Eee hardware and upgrading (on their own) to 2.6.29, given that that workaround is present in eeepc-acpi-scripts in lenny.

Anyway, removing that option fixes the problem.

Quoted 7 lines
Quoted 4 lines

[long delay between these two lines]
input: ETPS/2 Elantech Touchpad as /class/input/input5
eeepc: Hotkey init flags 0x41
(not seen in .28 or .28.1; not checked later .28.*.)

BIOS bug. There's an explicit delay in the eee bios for some reason, and I haven't found any straightforward way to avoid it.

BIOS bug or no, the fact remains that this is (AFAICS) a regression.

Signature · 9 lines
| Darren Salt    | linux or ds at              | nr. Ashington, | Toon
| RISC OS, Linux | youmustbejoking,demon,co,uk | Northumberland | Army
| + Use more efficient products. Use less.          BE MORE ENERGY EFFICIENT.

  10 PRINT "I used to do this in Dixons": POKE 23692,255: GO TO 10
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#13 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

Quoted 5 lines · Darren Salt

As things stand, this is likely to be a problem for anybody using lenny on Eee hardware and upgrading (on their own) to 2.6.29, given that that workaround is present in eeepc-acpi-scripts in lenny.

Anyway, removing that option fixes the problem.

Ok. I'll look into ensuring that these don't collide, but fundamentally the problem here is distributions using broken options rather than fixing the problem properly in the first place.

Quoted 4 lines
Quoted 2 lines

BIOS bug. There's an explicit delay in the eee bios for some reason, and I haven't found any straightforward way to avoid it.

BIOS bug or no, the fact remains that this is (AFAICS) a regression.

You don't get properly working hotkeys otherwise, to the best of my recollection. There's an entry on the kernel bugzilla about this somewhere.

Signature · 1 line
Matthew Garrett | [email protected]
maintainer
#14 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

Quoted 15 lines · Matthew Garrett
Quoted 5 lines · Darren Salt

As things stand, this is likely to be a problem for anybody using lenny on Eee hardware and upgrading (on their own) to 2.6.29, given that that workaround is present in eeepc-acpi-scripts in lenny.

Anyway, removing that option fixes the problem.

Ok. I'll look into ensuring that these don't collide, but fundamentally the problem here is distributions using broken options rather than fixing the problem properly in the first place.

Quoted 3 lines
Quoted 2 lines

BIOS bug. There's an explicit delay in the eee bios for some reason, and I haven't found any straightforward way to avoid it.

Quoted 7 lines
Quoted 1 line

BIOS bug or no, the fact remains that this is (AFAICS) a regression.

You don't get properly working hotkeys otherwise, to the best of my recollection. There's an entry on the kernel bugzilla about this somewhere.
--
Matthew Garrett | [email protected]

http://bugzilla.kernel.org/show_bug.cgi?id=12243
If I understand things correctly, this was already the case before, but the kernel boot was slower, so we could'nt see that. If it's not the case, someone with a 901 should try bisecting (I only have a 701).

Signature · 2 lines
Corentin Chary
http://xf.iksaif.net
#15 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

I demand that Corentin Chary may or may not have written...

Quoted 2 lines

On Tue, Feb 10, 2009 at 4:45 PM, Matthew Garrett <[email protected]> wrote:

[snip; long delay when initialising eeepc-laptop]

Quoted 11 lines
Quoted 6 lines
Quoted 3 lines
Quoted 2 lines

BIOS bug. There's an explicit delay in the eee bios for some reason, and I haven't found any straightforward way to avoid it.

BIOS bug or no, the fact remains that this is (AFAICS) a regression.

You don't get properly working hotkeys otherwise, to the best of my recollection. There's an entry on the kernel bugzilla about this somewhere.

http://bugzilla.kernel.org/show_bug.cgi?id=12243
If I understand things correctly, this was already the case before, but the kernel boot was slower, so we couldn't see that. If it's not the case, someone with a 901 should try bisecting (I only have a 701).

I have some timings...

Quoted 1 line

From 2.6.28.4:

[    0.898903] eeepc: Eee PC Hotkey Driver
[    0.902414] eeepc: Hotkey init (getting ACPI bus status)
[    0.905913] eeepc: Hotkey init (init flags)
[    2.620193] eeepc: Hotkey init flags 0x41
[    2.626667] eeepc: Get control methods supported: 0x101713
[    2.630215] input: Asus EeePC extra buttons as /class/input/input4
Quoted 1 line

From 2.6.29-rc4:

[    2.141971] eeepc: Eee PC Hotkey Driver
[    2.145325] eeepc: Hotkey init (getting ACPI bus status)
[    2.148555] eeepc: Hotkey init (init flags)
[    2.286445] usb 5-1: new full speed USB device using uhci_hcd and address 2
[... USB Bluetooth device; Elantech test ...]
[    2.780187] input: ETPS/2 Elantech Touchpad as /class/input/input5
[   28.088399] eeepc: Hotkey init flags 0x41
[   28.094880] eeepc: Get control methods supported: 0x101713
[   28.098425] input: Asus EeePC extra buttons as /class/input/input6

Config is as in my first posting in this thread; kernel logs are attached. The following patch is applied:

drivers/platform/x86/eeepc-laptop.c renamed from drivers/platform/x86/eeepc-laptop.c~ +5 -0
@@ -452,10 +452,15 @@
 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 	int result;
 
+	printk(EEEPC_NOTICE "Hotkey init (getting ACPI bus status)\n");
 	result = acpi_bus_get_status(ehotk->device);
 	if (result)
+	{
+		printk(EEEPC_ERR "Hotkey init aborted (%d)\n", result);
 		return result;
+	}
 	if (ehotk->device->status.present) {
+		printk(EEEPC_NOTICE "Hotkey init (init flags)\n");
 		if (write_acpi_int(ehotk->handle, "INIT", ehotk->init_flag,
 				    &buffer)) {
 			printk(EEEPC_ERR "Hotkey initialization failed\n");
Signature · 5 lines
| Darren Salt    | linux or ds at              | nr. Ashington, | Toon
| RISC OS, Linux | youmustbejoking,demon,co,uk | Northumberland | Army
| + Buy less and make it last longer.         INDUSTRY CAUSES GLOBAL WARMING.

If at first you don't succeed, give up. No use being a damn fool.
#16 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

Quoted 7 lines · Darren Salt

From 2.6.28.4:

[    0.898903] eeepc: Eee PC Hotkey Driver
[    0.902414] eeepc: Hotkey init (getting ACPI bus status)
[    0.905913] eeepc: Hotkey init (init flags)
[    2.620193] eeepc: Hotkey init flags 0x41
[    2.626667] eeepc: Get control methods supported: 0x101713
[    2.630215] input: Asus EeePC extra buttons as /class/input/input4

So we see the expected BIOS bizarro pause there...

Quoted 11 lines

From 2.6.29-rc4:

[    2.141971] eeepc: Eee PC Hotkey Driver
[    2.145325] eeepc: Hotkey init (getting ACPI bus status)
[    2.148555] eeepc: Hotkey init (init flags)
[    2.286445] usb 5-1: new full speed USB device using uhci_hcd and address 2
[... USB Bluetooth device; Elantech test ...]
[    2.780187] input: ETPS/2 Elantech Touchpad as /class/input/input5
[   28.088399] eeepc: Hotkey init flags 0x41
[   28.094880] eeepc: Get control methods supported: 0x101713
[   28.098425] input: Asus EeePC extra buttons as /class/input/input6

And a rather more upsetting 26 second pause here. Yeah, ok, that's clearly a regression. I don't see any obvious way that it's an eeepc-laptop bug, though...

Signature · 1 line
Matthew Garrett | [email protected]
#17 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

In reply to Corentin Chary
Quoted 24 lines · Corentin Chary
Quoted 15 lines · Matthew Garrett
Quoted 5 lines · Darren Salt

As things stand, this is likely to be a problem for anybody using lenny on Eee hardware and upgrading (on their own) to 2.6.29, given that that workaround is present in eeepc-acpi-scripts in lenny.

Anyway, removing that option fixes the problem.

Ok. I'll look into ensuring that these don't collide, but fundamentally the problem here is distributions using broken options rather than fixing the problem properly in the first place.

Quoted 3 lines
Quoted 2 lines

BIOS bug. There's an explicit delay in the eee bios for some reason, and I haven't found any straightforward way to avoid it.

Quoted 7 lines
Quoted 1 line

BIOS bug or no, the fact remains that this is (AFAICS) a regression.

You don't get properly working hotkeys otherwise, to the best of my recollection. There's an entry on the kernel bugzilla about this somewhere.
--
Matthew Garrett | [email protected]

Quoted 5 lines

http://bugzilla.kernel.org/show_bug.cgi?id=12243
If I understand things correctly, this was already the case before, but the kernel boot was slower, so we could'nt see that. If it's not the case, someone with a 901 should try bisecting (I only have a 701).

I have an Asus EEEPC901. And it will take about 1 second to execute the ACPI INIT object, which is defined in BIOS.
From the ACPI AML code it seems that there exist some explicit
delays. But we have nothing to avoid it.

Quoted 2 lines
#18 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

In reply to Matthew Garrett

I demand that Matthew Garrett may or may not have written...

Quoted 9 lines · Darren Salt
Quoted 4 lines

As things stand, this is likely to be a problem for anybody using lenny on Eee hardware and upgrading (on their own) to 2.6.29, given that that workaround is present in eeepc-acpi-scripts in lenny.
Anyway, removing that option fixes the problem.

Ok. I'll look into ensuring that these don't collide, but fundamentally the problem here is distributions using broken options rather than fixing the problem properly in the first place.

Not so much "broken options" as "using the ony available workaround at the time", I think. I have no idea what action other than using this option was taken.

Quoted 6 lines
Quoted 3 lines
Quoted 2 lines

BIOS bug. There's an explicit delay in the eee bios for some reason, and I haven't found any straightforward way to avoid it.

BIOS bug or no, the fact remains that this is (AFAICS) a regression.

You don't get properly working hotkeys otherwise, to the best of my recollection.

I can't say that I've noticed any problems.

Quoted 1 line

There's an entry on the kernel bugzilla about this somewhere.

This one, I think: http://bugzilla.kernel.org/show_bug.cgi?id=12243

Signature · 5 lines
| Darren Salt    | linux or ds at              | nr. Ashington, | Toon
| RISC OS, Linux | youmustbejoking,demon,co,uk | Northumberland | Army
|   Kill all extremists!

"The name's Borg, James Borg. Prepare to be assimilated, Miss Moneypenny."
#19 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

Can you try this (entirely untested - I don't have access to the hardware right now) patch? It should register the eee code with the PCI hotplug core. With luck that'll avoid the oops.

drivers/platform/x86/Kconfig +1 -0
index b3866ad..75a560b 100644
@@ -307,6 +307,7 @@ config EEEPC_LAPTOP
 	select BACKLIGHT_CLASS_DEVICE
 	select HWMON
 	select RFKILL
+	select HOTPLUG_PCI
 	---help---
 	  This driver supports the Fn-Fx keys on Eee PC laptops.
 	  It also adds the ability to switch camera/wlan on/off.
drivers/platform/x86/eeepc-laptop.c +80 -1

127 lines not on this page · raw message

Signature · 1 line
Matthew Garrett | [email protected]
#20 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

I demand that Matthew Garrett may or may not have written...

drivers/platform/x86/Kconfig config EEEPC_LAPTOP · line 310

Quoted 13 lines

Can you try this (entirely untested - I don't have access to the hardware right now) patch? It should register the eee code with the PCI hotplug core. With luck that'll avoid the oops.

drivers/platform/x86/Kconfig +1 -0
index b3866ad..75a560b 100644
@@ -307,6 +307,7 @@ config EEEPC_LAPTOP
 	select BACKLIGHT_CLASS_DEVICE
 	select HWMON
 	select RFKILL
+	select HOTPLUG_PCI

Not HOTPLUG_PCI_PCIE?

With that selected (I see no point in testing this otherwise) *and* pciehp_force=1, I get -EBUSY when the hotplug slot registration is attempted and no wireless, bluetooth or hot key reporting; wireless & bluetooth are switched off at that point (I made sure that they were enabled via BIOS setup). Without that option, all is fine.

(As usual with my kernel builds, these are built in.)

[snip]

Signature · 5 lines
| Darren Salt    | linux or ds at              | nr. Ashington, | Toon
| RISC OS, Linux | youmustbejoking,demon,co,uk | Northumberland | Army
| + Travel less. Share transport more.           PRODUCE LESS CARBON DIOXIDE.

The man who raises a fist has run out of ideas.
#21 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

Quoted 1 line · Darren Salt

Not HOTPLUG_PCI_PCIE?

No - it just depends on the hotplug PCI core.

Quoted 5 lines

With that selected (I see no point in testing this otherwise) *and* pciehp_force=1, I get -EBUSY when the hotplug slot registration is attempted and no wireless, bluetooth or hot key reporting; wireless & bluetooth are switched off at that point (I made sure that they were enabled via BIOS setup). Without that option, all is fine.

Right, so that sounds like everything works as expected. I guess to avoid breaking existing setups we should probably be a bit nicer there. Can you test whether this one works?

drivers/platform/x86/Kconfig +1 -0
index b3866ad..75a560b 100644
@@ -307,6 +307,7 @@ config EEEPC_LAPTOP
 	select BACKLIGHT_CLASS_DEVICE
 	select HWMON
 	select RFKILL
+	select HOTPLUG_PCI
 	---help---
 	  This driver supports the Fn-Fx keys on Eee PC laptops.
 	  It also adds the ability to switch camera/wlan on/off.
drivers/platform/x86/eeepc-laptop.c +92 -3

141 lines not on this page · raw message

Signature · 1 line
Matthew Garrett | [email protected]
Tested-by
#22 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

I demand that Matthew Garrett may or may not have written...

Quoted 11 lines · Darren Salt
Quoted 1 line

Not HOTPLUG_PCI_PCIE?

No - it just depends on the hotplug PCI core.

Quoted 5 lines

With that selected (I see no point in testing this otherwise) *and* pciehp_force=1, I get -EBUSY when the hotplug slot registration is attempted and no wireless, bluetooth or hot key reporting; wireless & bluetooth are switched off at that point (I made sure that they were enabled via BIOS setup). Without that option, all is fine.

Quoted 3 lines

Right, so that sounds like everything works as expected. I guess to avoid breaking existing setups we should probably be a bit nicer there. Can you test whether this one works?

Works fine, AFAICS. If that's what's to be committed, then:

[snip]

Signature · 5 lines
| Darren Salt    | linux or ds at              | nr. Ashington, | Toon
| RISC OS, Linux | youmustbejoking,demon,co,uk | Northumberland | Army
| + Burn less waste. Use less packaging. Waste less.     USE FEWER RESOURCES.

Beware of quantum ducks. Quark! Quark!
#23 lore raw

PATCH 1/1 eee-laptop: Register as a pci-hotplug device

In reply to Darren Salt

eee-laptop: Register as a pci-hotplug device

The eee contains a logically (but not physically) hotpluggable PCIe slot. Currently this is handled by adding or removing the PCI device in response to rfkill events, but if a user has forced pciehp to bind to it (with the force=1 argument) then both drivers will try to handle the event and hilarity (in the form of oopses) will ensue. This can be avoided by having eee-laptop register the slot as a hotplug slot. Only one of pciehp and eee-laptop will successfully register this, avoiding the problem.

drivers/platform/x86/Kconfig +1 -0
index 284ebac..afbe441 100644
@@ -341,6 +341,7 @@ config EEEPC_LAPTOP
 	select BACKLIGHT_CLASS_DEVICE
 	select HWMON
 	select RFKILL
+	select HOTPLUG_PCI
 	---help---
 	  This driver supports the Fn-Fx keys on Eee PC laptops.
 	  It also adds the ability to switch camera/wlan on/off.
drivers/platform/x86/eeepc-laptop.c +92 -3

141 lines not on this page · raw message

Signature · 1 line
Matthew Garrett | [email protected]
maintainer
#24 lore raw

Re: [PATCH] eee-laptop: Register as a pci-hotplug device

Quoted 13 lines
drivers/platform/x86/eeepc-laptop.c +12 -0
+       result = eeepc_setup_pci_hotplug();
+
+       /*
+        * If we get -EBUSY then something else is handling the PCI hotplug -
+        * don't fail in this case
+        */
+
+       if (result == -EBUSY)
+               return 0;
+       else if (result)
+               goto pci_fail;
+
       eeepc_register_rfkill_notifier("\\_SB.PCI0.P0P6");

drivers/platform/x86/eeepc-laptop.c static int eeepc_hotk_add(struct acpi_device *device)

Quoted 10 lines
drivers/platform/x86/eeepc-laptop.c +5 -2
       eeepc_register_rfkill_notifier("\\_SB.PCI0.P0P7");
 
 
-       eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P6");
-       eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P7");
+       if (ehotk->hotplug_slot) {
+               eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P6");
+               eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P7");
+               pci_hp_deregister(ehotk->hotplug_slot);
+       }

Hi,
This patch conflict with
http://git.iksaif.net/?p=acpi4asus.git;a=commitdiff;h=8082f39bb3d4a27c7bb03a76e72c70c86d55117b;hp=7ec0a7290797f57b780f792d12f4bcc19c83aa4f
A quick fix is to do :

eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P6"); eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P7"); if (ehotk->hotplug_slot)

      pci_hp_deregister(ehotk->hotplug_slot);

but I'm not sure it's correct.
I should have sent a series sooner to avoid that :/ Thanks

#25 lore raw

Re: [PATCH] eee-laptop: Register as a pci-hotplug device

Quoted 5 lines · Corentin Chary

Hi,
This patch conflict with
http://git.iksaif.net/?p=acpi4asus.git;a=commitdiff;h=8082f39bb3d4a27c7bb03a76e72c70c86d55117b;hp=7ec0a7290797f57b780f792d12f4bcc19c83aa4f
A quick fix is to do :

Quoted 7 lines

eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P6"); eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P7"); if (ehotk->hotplug_slot)

      pci_hp_deregister(ehotk->hotplug_slot);

but I'm not sure it's correct.
I should have sent a series sooner to avoid that :/

Ok - I'll grab that and rediff.

Signature · 1 line
Matthew Garrett | [email protected]
maintainer Tested-by Darren Salt
#26 lore raw

Re: [PATCH] eee-laptop: Register as a pci-hotplug device

Hi,
I take your patch and make it works against acpi4asus tree. If you think it's ok, then I'll send a serie including this patch to Len. (the tree is at http://git.iksaif.net/ )

From: Matthew Garrett <[email protected]>

The eee contains a logically (but not physically) hotpluggable PCIe slot. Currently this is handled by adding or removing the PCI device in response to rfkill events, but if a user has forced pciehp to bind to it (with the force=1 argument) then both drivers will try to handle the event and hilarity (in the form of oopses) will ensue. This can be avoided by having eee-laptop register the slot as a hotplug slot. Only one of pciehp and eee-laptop will successfully register this, avoiding the problem.

---
 drivers/platform/x86/Kconfig        |    1 +
 drivers/platform/x86/eeepc-laptop.c |   87 +++++++++++++++++++++++++++++++++++
 2 files changed, 88 insertions(+), 0 deletions(-)
drivers/platform/x86/Kconfig +1 -0
index 284ebac..afbe441 100644
@@ -341,6 +341,7 @@ config EEEPC_LAPTOP
 	select BACKLIGHT_CLASS_DEVICE
 	select HWMON
 	select RFKILL
+	select HOTPLUG_PCI
 	---help---
 	  This driver supports the Fn-Fx keys on Eee PC laptops.
 	  It also adds the ability to switch camera/wlan on/off.
(quoted diff) +0 -0

b/drivers/platform/x86/eeepc-laptop.c
index 7aaf587..4440f6c 100644

drivers/platform/x86/eeepc-laptop.c +85 -0

123 lines not on this page · raw message

*device, int type)

eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P6"); eeepc_unregister_rfkill_notifier("\\_SB.PCI0.P0P7");
+ if (ehotk->hotplug_slot)
+ pci_hp_deregister(ehotk->hotplug_slot);

kfree(ehotk);
return 0;

Signature · 1 line
1.6.2.3
#27 lore raw

Re: [PATCH] eee-laptop: Register as a pci-hotplug device

Hm. I /think/ that version leaks notification handlers on the failure path. Does this look ok? I don't have an eee to hand right now, so can't test this at the moment.

drivers/platform/x86/Kconfig +1 -0
index 284ebac..afbe441 100644
@@ -341,6 +341,7 @@ config EEEPC_LAPTOP
 	select BACKLIGHT_CLASS_DEVICE
 	select HWMON
 	select RFKILL
+	select HOTPLUG_PCI
 	---help---
 	  This driver supports the Fn-Fx keys on Eee PC laptops.
 	  It also adds the ability to switch camera/wlan on/off.
drivers/platform/x86/eeepc-laptop.c +91 -0

141 lines not on this page · raw message

Signature · 1 line
Matthew Garrett | [email protected]
maintainer
#28 lore raw

Re: [PATCH] eee-laptop: Register as a pci-hotplug device

Quoted 3 lines · Matthew Garrett

Hm. I /think/ that version leaks notification handlers on the failure path. Does this look ok? I don't have an eee to hand right now, so can't test this at the moment.

I'll test it tomorow, but it seems ok.
Thanks

Signature · 2 lines
Corentin Chary
http://xf.iksaif.net
#29 lore raw

Re: Oopses and ACPI problems (Linus 2.6.29-rc4)

In reply to Darren Salt
Quoted 10 lines · Darren Salt

I'm seeing various oopses with -rc4, where -rc3 is fine; config, dmesg (captured via netconsole) and decoded oops data attached ¹. The last few oopses were followed by a scrambled display and a hard lockup.

All oopses follow an otherwise successful suspend-to-RAM.

Also, I see that Fn-Fx don't seem to be reported via /proc/acpi/events any more, or at least I'm seeing no evidence of them via acpi_listen (and, not coincidentally, I can't adjust the volume etc. via eeepc-acpi-scripts).

Quoted 5 lines

The only other (significant) change is that I've installed X from Debian experimental, though -rc3 and 2.6.28.* seem fine with this. (Oh, and I can get UXA to cause OOM quite trivially... several Iceweasel tabs is enough.)

From the description it seems that this is a regression.

Will you please use git-bisect to identify which commit causes the regression?
Thanks.

Quoted 3 lines

¹ Yes, I decided not to include symbol names in the running kernel...

--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#30 lore raw

PATCH 1/1 Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

In reply to Linus Torvalds
Quoted 10 lines · Linus Torvalds

Another week (and a half), another -rc.

Arch updates (sparc, blackfin), driver updates (dvb, mmc, ide), ACPI updates, FS updates (ubifs, btrfs), you name it. It's all there.

But more importantly, people really have been working on regressions, and hopefully this closes a nice set of the top one, and hopefully without introducing too many new ones.

Hugepages are currently busted due to commit fc8744adc870a8d4366908221508bb113d8b72ee and the problem is in 2.6.29-rc4. There was a bit of a discussion but it didn't get very far and then I went offline for the weekend. Here is a revised patch that tries to bring hugetlbfs more in line with what the core VM is doing by dealing with VM_ACCOUNT and VM_NORESERVE.

=============
Do not account for the address space used by hugetlbfs using VM_ACCOUNT

When overcommit is disabled, the core VM accounts for pages used by anonymous shared, private mappings and special mappings. It keeps track of VMAs that should be accounted for with VM_ACCOUNT and VMAs that never had a reserve with VM_NORESERVE.

Overcommit for hugetlbfs is much riskier than overcommit for base pages due to contiguity requirements. It avoids overcommiting on both shared and private mappings using reservation counters that are checked and updated during mmap(). This ensures (within limits) that hugepages exist in the future when faults occurs or it is too easy to applications to be SIGKILLed.

As hugetlbfs makes its own reservations of a different unit to the base page size, VM_ACCOUNT should never be set. Even if the units were correct, we would double account for the usage in the core VM and hugetlbfs. VM_NORESERVE may be set because an application can request no reserves be made for hugetlbfs at the risk of getting killed later.

With commit fc8744adc870a8d4366908221508bb113d8b72ee, VM_NORESERVE and VM_ACCOUNT are getting unconditionally set for hugetlbfs-backed mappings. This breaks the accounting for both the core VM and hugetlbfs, can trigger an OOM storm when hugepage pools are too small lockups and corrupted counters otherwise are used. This patch brings hugetlbfs more in line with how the core VM treats VM_NORESERVE but prevents VM_ACCOUNT being set.

---

 fs/hugetlbfs/inode.c    |    8 +++++---
 include/linux/hugetlb.h |    5 +++--
 include/linux/mm.h      |    3 +--
 ipc/shm.c               |    8 +++++---
 mm/fremap.c             |    2 +-
 mm/hugetlb.c            |   39 +++++++++++++++++++++++++--------------
 mm/mmap.c               |   38 ++++++++++++++++++++++----------------
 mm/mprotect.c           |    5 +++--
 8 files changed, 65 insertions(+), 43 deletions(-)
fs/hugetlbfs/inode.c +5 -3
index 6903d37..9b800d9 100644
@@ -108,7 +108,8 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 
 	if (hugetlb_reserve_pages(inode,
 				vma->vm_pgoff >> huge_page_order(h),
-				len >> huge_page_shift(h), vma))
+				len >> huge_page_shift(h), vma,
+				vma->vm_flags))
 		goto out;
 
 	ret = 0;
@@ -947,7 +948,7 @@ static int can_do_hugetlb_shm(void)
 			can_do_mlock());
 }
 
-struct file *hugetlb_file_setup(const char *name, size_t size)
+struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag)
 {
 	int error = -ENOMEM;
 	struct file *file;
@@ -981,7 +982,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size)
 
 	error = -ENOMEM;
 	if (hugetlb_reserve_pages(inode, 0,
-			size >> huge_page_shift(hstate_inode(inode)), NULL))
+			size >> huge_page_shift(hstate_inode(inode)), NULL,
+			acctflag))
 		goto out_inode;
 
 	d_instantiate(dentry, inode);
include/linux/hugetlb.h +3 -2
index f1d2fba..af09660 100644
@@ -33,7 +33,8 @@ unsigned long hugetlb_total_pages(void);
 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, int write_access);
 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
-						struct vm_area_struct *vma);
+						struct vm_area_struct *vma,
+						int acctflags);
 void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed);
 
 extern unsigned long hugepages_treat_as_movable;
@@ -138,7 +139,7 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 
 extern const struct file_operations hugetlbfs_file_operations;
 extern struct vm_operations_struct hugetlb_vm_ops;
-struct file *hugetlb_file_setup(const char *name, size_t);
+struct file *hugetlb_file_setup(const char *name, size_t, int);
 int hugetlb_get_quota(struct address_space *mapping, long delta);
 void hugetlb_put_quota(struct address_space *mapping, long delta);
 
include/linux/mm.h +1 -2
index e8ddc98..3235615 100644
@@ -1129,8 +1129,7 @@ extern unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	unsigned long flag, unsigned long pgoff);
 extern unsigned long mmap_region(struct file *file, unsigned long addr,
 	unsigned long len, unsigned long flags,
-	unsigned int vm_flags, unsigned long pgoff,
-	int accountable);
+	unsigned int vm_flags, unsigned long pgoff);
 
 static inline unsigned long do_mmap(struct file *file, unsigned long addr,
 	unsigned long len, unsigned long prot,
ipc/shm.c +5 -3
index f8f69fa..05d51d2 100644
@@ -340,6 +340,7 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 	struct file * file;
 	char name[13];
 	int id;
+	int acctflag = 0;
 
 	if (size < SHMMIN || size > ns->shm_ctlmax)
 		return -EINVAL;
@@ -364,11 +365,12 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 
 	sprintf (name, "SYSV%08x", key);
 	if (shmflg & SHM_HUGETLB) {
-		/* hugetlb_file_setup takes care of mlock user accounting */
-		file = hugetlb_file_setup(name, size);
+		/* hugetlb_file_setup applies strict accounting */
+		if (shmflg & SHM_NORESERVE)
+			acctflag = VM_NORESERVE;
+		file = hugetlb_file_setup(name, size, acctflag);
 		shp->mlock_user = current_user();
 	} else {
-		int acctflag = 0;
 		/*
 		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
 	 	 * if it's asked for.
mm/fremap.c +1 -1
index 736ba7f..b6ec85a 100644
@@ -198,7 +198,7 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
 			flags &= MAP_NONBLOCK;
 			get_file(file);
 			addr = mmap_region(file, start, size,
-					flags, vma->vm_flags, pgoff, 1);
+					flags, vma->vm_flags, pgoff);
 			fput(file);
 			if (IS_ERR_VALUE(addr)) {
 				err = addr;
mm/hugetlb.c +25 -14
index 618e983..2074642 100644
@@ -2269,14 +2269,12 @@ void hugetlb_change_protection(struct vm_area_struct *vma,
 
 int hugetlb_reserve_pages(struct inode *inode,
 					long from, long to,
-					struct vm_area_struct *vma)
+					struct vm_area_struct *vma,
+					int acctflag)
 {
-	long ret, chg;
+	long ret = 0, chg;
 	struct hstate *h = hstate_inode(inode);
 
-	if (vma && vma->vm_flags & VM_NORESERVE)
-		return 0;
-
 	/*
 	 * Shared mappings base their reservation on the number of pages that
 	 * are already allocated on behalf of the file. Private mappings need
@@ -2285,22 +2283,25 @@ int hugetlb_reserve_pages(struct inode *inode,
 	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		chg = region_chg(&inode->i_mapping->private_list, from, to);
-	else {
-		struct resv_map *resv_map = resv_map_alloc();
-		if (!resv_map)
-			return -ENOMEM;
-
+	else
 		chg = to - from;
 
-		set_vma_resv_map(vma, resv_map);
-		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
-	}
-
 	if (chg < 0)
 		return chg;
 
 	if (hugetlb_get_quota(inode->i_mapping, chg))
 		return -ENOSPC;
+
+	/*
+	 * Only apply hugepage reservation if asked. We still have to
+	 * take the filesystem quota because it is an upper limit
+	 * defined for the mount and not necessarily memory as a whole
+	 */
+	if (acctflag & VM_NORESERVE) {
+		reset_vma_resv_huge_pages(vma);
+		return 0;
+	}
+
 	ret = hugetlb_acct_memory(h, chg);
 	if (ret < 0) {
 		hugetlb_put_quota(inode->i_mapping, chg);
@@ -2308,6 +2309,16 @@ int hugetlb_reserve_pages(struct inode *inode,
 	}
 	if (!vma || vma->vm_flags & VM_SHARED)
 		region_add(&inode->i_mapping->private_list, from, to);
+	else {
+		struct resv_map *resv_map = resv_map_alloc();
+
+		if (!resv_map)
+			return -ENOMEM;
+
+		set_vma_resv_map(vma, resv_map);
+		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
+	}
+
 	return 0;
 }
 
mm/mmap.c +22 -16
index 214b6a2..6f70db8 100644
@@ -918,7 +918,6 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	struct inode *inode;
 	unsigned int vm_flags;
 	int error;
-	int accountable = 1;
 	unsigned long reqprot = prot;
 
 	/*
@@ -1019,8 +1018,6 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 					return -EPERM;
 				vm_flags &= ~VM_MAYEXEC;
 			}
-			if (is_file_hugepages(file))
-				accountable = 0;
 
 			if (!file->f_op || !file->f_op->mmap)
 				return -ENODEV;
@@ -1053,8 +1050,7 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	if (error)
 		return error;
 
-	return mmap_region(file, addr, len, flags, vm_flags, pgoff,
-			   accountable);
+	return mmap_region(file, addr, len, flags, vm_flags, pgoff);
 }
 EXPORT_SYMBOL(do_mmap_pgoff);
 
@@ -1092,17 +1088,23 @@ int vma_wants_writenotify(struct vm_area_struct *vma)
 
 /*
  * We account for memory if it's a private writeable mapping,
- * and VM_NORESERVE wasn't set.
+ * not hugepages and VM_NORESERVE wasn't set.
  */
-static inline int accountable_mapping(unsigned int vm_flags)
+static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
 {
+	/*
+	 * hugetlb has its own accounting separate from the core VM
+	 * VM_HUGETLB may not be set yet so we cannot check for that flag.
+	 */
+	if (file && is_file_hugepages(file))
+		return 0;
+
 	return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
 }
 
 unsigned long mmap_region(struct file *file, unsigned long addr,
 			  unsigned long len, unsigned long flags,
-			  unsigned int vm_flags, unsigned long pgoff,
-			  int accountable)
+			  unsigned int vm_flags, unsigned long pgoff)
 {
 	struct mm_struct *mm = current->mm;
 	struct vm_area_struct *vma, *prev;
@@ -1128,18 +1130,22 @@ munmap_back:
 
 	/*
 	 * Set 'VM_NORESERVE' if we should not account for the
-	 * memory use of this mapping. We only honor MAP_NORESERVE
-	 * if we're allowed to overcommit memory.
+	 * memory use of this mapping.
 	 */
-	if ((flags & MAP_NORESERVE) && sysctl_overcommit_memory != OVERCOMMIT_NEVER)
-		vm_flags |= VM_NORESERVE;
-	if (!accountable)
-		vm_flags |= VM_NORESERVE;
+	if ((flags & MAP_NORESERVE)) {
+	 	/* * We honor MAP_NORESERVE if allowed to overcommit */
+		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
+			vm_flags |= VM_NORESERVE;
+
+		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
+		if (file && is_file_hugepages(file))
+			vm_flags |= VM_NORESERVE;
+	}
 
 	/*
 	 * Private writable mapping: check memory availability
 	 */
-	if (accountable_mapping(vm_flags)) {
+	if (accountable_mapping(file, vm_flags)) {
 		charged = len >> PAGE_SHIFT;
 		if (security_vm_enough_memory(charged))
 			return -ENOMEM;
mm/mprotect.c +3 -2
index abe2694..258197b 100644
@@ -151,10 +151,11 @@ mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 	/*
 	 * If we make a private mapping writable we increase our commit;
 	 * but (without finer accounting) cannot reduce our commit if we
-	 * make it unwritable again.
+	 * make it unwritable again. hugetlb mapping were accounted for
+	 * even if read-only so there is no need to account for them here
 	 */
 	if (newflags & VM_WRITE) {
-		if (!(oldflags & (VM_ACCOUNT|VM_WRITE|
+		if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 						VM_SHARED|VM_NORESERVE))) {
 			charged = nrpages;
 			if (security_vm_enough_memory(charged))
maintainer
#31 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

Quoted 61 lines · On Tue, 10 Feb 2009 14:02:27 +0000 Mel Gorman <[email protected]> wrote:
Quoted 10 lines · Linus Torvalds

Another week (and a half), another -rc.

Arch updates (sparc, blackfin), driver updates (dvb, mmc, ide), ACPI updates, FS updates (ubifs, btrfs), you name it. It's all there.

But more importantly, people really have been working on regressions, and hopefully this closes a nice set of the top one, and hopefully without introducing too many new ones.

Hugepages are currently busted due to commit fc8744adc870a8d4366908221508bb113d8b72ee and the problem is in 2.6.29-rc4. There was a bit of a discussion but it didn't get very far and then I went offline for the weekend. Here is a revised patch that tries to bring hugetlbfs more in line with what the core VM is doing by dealing with VM_ACCOUNT and VM_NORESERVE.

=============
Do not account for the address space used by hugetlbfs using VM_ACCOUNT

When overcommit is disabled, the core VM accounts for pages used by anonymous shared, private mappings and special mappings. It keeps track of VMAs that should be accounted for with VM_ACCOUNT and VMAs that never had a reserve with VM_NORESERVE.

Overcommit for hugetlbfs is much riskier than overcommit for base pages due to contiguity requirements. It avoids overcommiting on both shared and private mappings using reservation counters that are checked and updated during mmap(). This ensures (within limits) that hugepages exist in the future when faults occurs or it is too easy to applications to be SIGKILLed.

As hugetlbfs makes its own reservations of a different unit to the base page size, VM_ACCOUNT should never be set. Even if the units were correct, we would double account for the usage in the core VM and hugetlbfs. VM_NORESERVE may be set because an application can request no reserves be made for hugetlbfs at the risk of getting killed later.

With commit fc8744adc870a8d4366908221508bb113d8b72ee, VM_NORESERVE and VM_ACCOUNT are getting unconditionally set for hugetlbfs-backed mappings. This breaks the accounting for both the core VM and hugetlbfs, can trigger an OOM storm when hugepage pools are too small lockups and corrupted counters otherwise are used. This patch brings hugetlbfs more in line with how the core VM treats VM_NORESERVE but prevents VM_ACCOUNT being set.

...

fs/hugetlbfs/inode.c +2 -1
@@ -108,7 +108,8 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 
 	if (hugetlb_reserve_pages(inode,
 				vma->vm_pgoff >> huge_page_order(h),
-				len >> huge_page_shift(h), vma))
+				len >> huge_page_shift(h), vma,
+				vma->vm_flags))
 		goto out;
 
 	ret = 0;

fs/hugetlbfs/inode.c static int can_do_hugetlb_shm(void) · line 951

Quoted 5 lines
fs/hugetlbfs/inode.c +1 -1
@@ -947,7 +948,7 @@ static int can_do_hugetlb_shm(void)
 			can_do_mlock());
 }
 
-struct file *hugetlb_file_setup(const char *name, size_t size)
+struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag)

This superundocumented acctflag looks like a poorly-named boolean. But it is fact a composition of VM_foo bits.

Would it not be clearer to name it `vm_flags'?

Quoted 18 lines
fs/hugetlbfs/inode.c +2 -1
 {
 	int error = -ENOMEM;
 	struct file *file;
@@ -981,7 +982,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size)
 
 	error = -ENOMEM;
 	if (hugetlb_reserve_pages(inode, 0,
-			size >> huge_page_shift(hstate_inode(inode)), NULL))
+			size >> huge_page_shift(hstate_inode(inode)), NULL,
+			acctflag))
 		goto out_inode;
 
 	d_instantiate(dentry, inode);
include/linux/hugetlb.h +0 -0
index f1d2fba..af09660 100644

include/linux/hugetlb.h unsigned long hugetlb_total_pages(void); · line 37

Quoted 6 lines
include/linux/hugetlb.h +2 -1
@@ -33,7 +33,8 @@ unsigned long hugetlb_total_pages(void);
 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, int write_access);
 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
-						struct vm_area_struct *vma);
+						struct vm_area_struct *vma,
+						int acctflags);

here it went plural.

include/linux/hugetlb.h static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb) · line 142

Quoted 9 lines
include/linux/hugetlb.h +1 -1
 void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed);
 
 extern unsigned long hugepages_treat_as_movable;
@@ -138,7 +139,7 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 
 extern const struct file_operations hugetlbfs_file_operations;
 extern struct vm_operations_struct hugetlb_vm_ops;
-struct file *hugetlb_file_setup(const char *name, size_t);
+struct file *hugetlb_file_setup(const char *name, size_t, int);

Here it is omitted altogether. We now have one named parameter and two secret ones.

Also, the patch forgot to update this:

#define hugetlb_file_setup(name,size) ERR_PTR(-ENOSYS)

Which I assume breaks CONFIG_HUGETLBFS=n.

include/linux/mm.h extern unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, · line 1132

Quoted 11 lines
include/linux/mm.h +1 -2
index e8ddc98..3235615 100644
@@ -1129,8 +1129,7 @@ extern unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	unsigned long flag, unsigned long pgoff);
 extern unsigned long mmap_region(struct file *file, unsigned long addr,
 	unsigned long len, unsigned long flags,
-	unsigned int vm_flags, unsigned long pgoff,
-	int accountable);
+	unsigned int vm_flags, unsigned long pgoff);

And there is a vm_flags which has type `unsigned int'.

Your newly-added should-have-been-called-vm_flags has type `int'.

The vm_flagses in struct vm_region' and struct vm_area_struct' have type `unsigned long'.

It'd be nice to get these consistent. We only have two bits left in the vm_flags namespace so arguably we could add a vm_flags_t while fixing this up, with a view to makeing it u64 later. Maybe.

#32 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

Quoted 67 lines · Andrew Morton
Quoted 64 lines · On Tue, 10 Feb 2009 14:02:27 +0000 Mel Gorman <[email protected]> wrote:
Quoted 10 lines · Linus Torvalds

Another week (and a half), another -rc.

Arch updates (sparc, blackfin), driver updates (dvb, mmc, ide), ACPI updates, FS updates (ubifs, btrfs), you name it. It's all there.

But more importantly, people really have been working on regressions, and hopefully this closes a nice set of the top one, and hopefully without introducing too many new ones.

Hugepages are currently busted due to commit fc8744adc870a8d4366908221508bb113d8b72ee and the problem is in 2.6.29-rc4. There was a bit of a discussion but it didn't get very far and then I went offline for the weekend. Here is a revised patch that tries to bring hugetlbfs more in line with what the core VM is doing by dealing with VM_ACCOUNT and VM_NORESERVE.

=============
Do not account for the address space used by hugetlbfs using VM_ACCOUNT

When overcommit is disabled, the core VM accounts for pages used by anonymous shared, private mappings and special mappings. It keeps track of VMAs that should be accounted for with VM_ACCOUNT and VMAs that never had a reserve with VM_NORESERVE.

Overcommit for hugetlbfs is much riskier than overcommit for base pages due to contiguity requirements. It avoids overcommiting on both shared and private mappings using reservation counters that are checked and updated during mmap(). This ensures (within limits) that hugepages exist in the future when faults occurs or it is too easy to applications to be SIGKILLed.

As hugetlbfs makes its own reservations of a different unit to the base page size, VM_ACCOUNT should never be set. Even if the units were correct, we would double account for the usage in the core VM and hugetlbfs. VM_NORESERVE may be set because an application can request no reserves be made for hugetlbfs at the risk of getting killed later.

With commit fc8744adc870a8d4366908221508bb113d8b72ee, VM_NORESERVE and VM_ACCOUNT are getting unconditionally set for hugetlbfs-backed mappings. This breaks the accounting for both the core VM and hugetlbfs, can trigger an OOM storm when hugepage pools are too small lockups and corrupted counters otherwise are used. This patch brings hugetlbfs more in line with how the core VM treats VM_NORESERVE but prevents VM_ACCOUNT being set.

...

fs/hugetlbfs/inode.c +2 -1
@@ -108,7 +108,8 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 
 	if (hugetlb_reserve_pages(inode,
 				vma->vm_pgoff >> huge_page_order(h),
-				len >> huge_page_shift(h), vma))
+				len >> huge_page_shift(h), vma,
+				vma->vm_flags))
 		goto out;
 
 	ret = 0;
@@ -947,7 +948,7 @@ static int can_do_hugetlb_shm(void)
 			can_do_mlock());
 }
 
Quoted 6 lines

fs/hugetlbfs/inode.c static int can_do_hugetlb_shm(void) · line 951

Quoted 2 lines
fs/hugetlbfs/inode.c +1 -1
-struct file *hugetlb_file_setup(const char *name, size_t size)
+struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag)

This superundocumented acctflag looks like a poorly-named boolean. But it is fact a composition of VM_foo bits.

It's matching naming at the call site. The call to shmem_file_setup() looks like

	file = shmem_file_setup(name, size, acctflag);

so, the call to hugetlb_file_setup looks like

	hugetlb_file_setup(name, size, acctflag);

and the name in the signature matches. shmem doesn't match like this, it just called its parameter "flags" and uses a different unit for size. The different in unit is bad in itself and I need to check that out.

Quoted 1 line

Would it not be clearer to name it `vm_flags'?

It's not all the flags either so vm_flags is confusing for other reasons. At least, at the time of writing the flags were meant to be VM_ACCOUNT and VM_NORESERVE but in reality, only VM_NORESERVE appears to be sent in but maybe that will change.

That said..... newseg() passes VM_NORESERVE as a flag to shmem_file_setup(). It in turn checks for VM_ACCOUNT in shmem_acct_size(), not VM_NORESERVE. Right now we may have an issue there too. Later it uses shmem_acct_block() but at that point VM_ACCOUNT has been set properly. shmem also has a VM_ACCT macro that has very little to do with VM_ACCOUNT. Naming rocks.

I'll look at a patch that makes the signatures of *_file_setup() match. I'll then check if it should be vm_acct_flags (both VM_ACCOUNT and VM_NORESERVE possible) or vm_resv_flag (VM_NORESERVE only).

Quoted 13 lines
Quoted 13 lines
fs/hugetlbfs/inode.c +2 -1
 {
 	int error = -ENOMEM;
 	struct file *file;
@@ -981,7 +982,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size)
 
 	error = -ENOMEM;
 	if (hugetlb_reserve_pages(inode, 0,
-			size >> huge_page_shift(hstate_inode(inode)), NULL))
+			size >> huge_page_shift(hstate_inode(inode)), NULL,
+			acctflag))
 		goto out_inode;
 
Quoted 15 lines

include/linux/hugetlb.h unsigned long hugetlb_total_pages(void); · line 37

Quoted 12 lines
fs/hugetlbfs/inode.c +0 -0
 	d_instantiate(dentry, inode);
include/linux/hugetlb.h +2 -1
index f1d2fba..af09660 100644
@@ -33,7 +33,8 @@ unsigned long hugetlb_total_pages(void);
 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, int write_access);
 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
-						struct vm_area_struct *vma);
+						struct vm_area_struct *vma,
+						int acctflags);

here it went plural.

Because I was thinking of VM_ACCOUNT and VM_NORESERVE, hence plural.

Quoted 5 lines
Quoted 5 lines
include/linux/hugetlb.h +0 -0
 void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed);
 
 extern unsigned long hugepages_treat_as_movable;
@@ -138,7 +139,7 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 
Quoted 8 lines

include/linux/hugetlb.h static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb) · line 142

Quoted 4 lines
include/linux/hugetlb.h +1 -1
 extern const struct file_operations hugetlbfs_file_operations;
 extern struct vm_operations_struct hugetlb_vm_ops;
-struct file *hugetlb_file_setup(const char *name, size_t);
+struct file *hugetlb_file_setup(const char *name, size_t, int);

Here it is omitted altogether. We now have one named parameter and two secret ones.

Dirt, that was matching what was happening for size_t but it's fugly. I should have fixed size_t while I was there rather than making it look worse.

Quoted 6 lines

Also, the patch forgot to update this:

#define hugetlb_file_setup(name,size) ERR_PTR(-ENOSYS)

Which I assume breaks CONFIG_HUGETLBFS=n.

Yeah, it broke and has been fixed since.

Quoted 14 lines

include/linux/mm.h extern unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, · line 1132

Quoted 11 lines
include/linux/mm.h +1 -2
index e8ddc98..3235615 100644
@@ -1129,8 +1129,7 @@ extern unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	unsigned long flag, unsigned long pgoff);
 extern unsigned long mmap_region(struct file *file, unsigned long addr,
 	unsigned long len, unsigned long flags,
-	unsigned int vm_flags, unsigned long pgoff,
-	int accountable);
+	unsigned int vm_flags, unsigned long pgoff);

And there is a vm_flags which has type `unsigned int'.

Quoted 5 lines

Your newly-added should-have-been-called-vm_flags has type `int'.

The vm_flagses in struct vm_region' and struct vm_area_struct' have type `unsigned long'.

Yes, they all should have been unsigned long.

Quoted 4 lines

It'd be nice to get these consistent. We only have two bits left in the vm_flags namespace so arguably we could add a vm_flags_t while fixing this up, with a view to makeing it u64 later. Maybe.

I'll look at a vm_flags_t that works similar in principal to gfp_t and making the signatures match up.

Signature · 3 lines
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
#33 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

In reply to Mel Gorman
Quoted 141 lines · Mel Gorman
Quoted 10 lines · Linus Torvalds

Another week (and a half), another -rc.

Arch updates (sparc, blackfin), driver updates (dvb, mmc, ide), ACPI updates, FS updates (ubifs, btrfs), you name it. It's all there.

But more importantly, people really have been working on regressions, and hopefully this closes a nice set of the top one, and hopefully without introducing too many new ones.

Hugepages are currently busted due to commit fc8744adc870a8d4366908221508bb113d8b72ee and the problem is in 2.6.29-rc4. There was a bit of a discussion but it didn't get very far and then I went offline for the weekend. Here is a revised patch that tries to bring hugetlbfs more in line with what the core VM is doing by dealing with VM_ACCOUNT and VM_NORESERVE.

=============
Do not account for the address space used by hugetlbfs using VM_ACCOUNT

When overcommit is disabled, the core VM accounts for pages used by anonymous shared, private mappings and special mappings. It keeps track of VMAs that should be accounted for with VM_ACCOUNT and VMAs that never had a reserve with VM_NORESERVE.

Overcommit for hugetlbfs is much riskier than overcommit for base pages due to contiguity requirements. It avoids overcommiting on both shared and private mappings using reservation counters that are checked and updated during mmap(). This ensures (within limits) that hugepages exist in the future when faults occurs or it is too easy to applications to be SIGKILLed.

As hugetlbfs makes its own reservations of a different unit to the base page size, VM_ACCOUNT should never be set. Even if the units were correct, we would double account for the usage in the core VM and hugetlbfs. VM_NORESERVE may be set because an application can request no reserves be made for hugetlbfs at the risk of getting killed later.

With commit fc8744adc870a8d4366908221508bb113d8b72ee, VM_NORESERVE and VM_ACCOUNT are getting unconditionally set for hugetlbfs-backed mappings. This breaks the accounting for both the core VM and hugetlbfs, can trigger an OOM storm when hugepage pools are too small lockups and corrupted counters otherwise are used. This patch brings hugetlbfs more in line with how the core VM treats VM_NORESERVE but prevents VM_ACCOUNT being set.

---

 fs/hugetlbfs/inode.c    |    8 +++++---
 include/linux/hugetlb.h |    5 +++--
 include/linux/mm.h      |    3 +--
 ipc/shm.c               |    8 +++++---
 mm/fremap.c             |    2 +-
 mm/hugetlb.c            |   39 +++++++++++++++++++++++++--------------
 mm/mmap.c               |   38 ++++++++++++++++++++++----------------
 mm/mprotect.c           |    5 +++--
 8 files changed, 65 insertions(+), 43 deletions(-)
fs/hugetlbfs/inode.c +5 -3
index 6903d37..9b800d9 100644
@@ -108,7 +108,8 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 
 	if (hugetlb_reserve_pages(inode,
 				vma->vm_pgoff >> huge_page_order(h),
-				len >> huge_page_shift(h), vma))
+				len >> huge_page_shift(h), vma,
+				vma->vm_flags))
 		goto out;
 
 	ret = 0;
@@ -947,7 +948,7 @@ static int can_do_hugetlb_shm(void)
 			can_do_mlock());
 }
 
-struct file *hugetlb_file_setup(const char *name, size_t size)
+struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag)
 {
 	int error = -ENOMEM;
 	struct file *file;
@@ -981,7 +982,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size)
 
 	error = -ENOMEM;
 	if (hugetlb_reserve_pages(inode, 0,
-			size >> huge_page_shift(hstate_inode(inode)), NULL))
+			size >> huge_page_shift(hstate_inode(inode)), NULL,
+			acctflag))
 		goto out_inode;
 
 	d_instantiate(dentry, inode);
include/linux/hugetlb.h +3 -2
index f1d2fba..af09660 100644
@@ -33,7 +33,8 @@ unsigned long hugetlb_total_pages(void);
 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, int write_access);
 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
-						struct vm_area_struct *vma);
+						struct vm_area_struct *vma,
+						int acctflags);
 void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed);
 
 extern unsigned long hugepages_treat_as_movable;
@@ -138,7 +139,7 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 
 extern const struct file_operations hugetlbfs_file_operations;
 extern struct vm_operations_struct hugetlb_vm_ops;
-struct file *hugetlb_file_setup(const char *name, size_t);
+struct file *hugetlb_file_setup(const char *name, size_t, int);
 int hugetlb_get_quota(struct address_space *mapping, long delta);
 void hugetlb_put_quota(struct address_space *mapping, long delta);
 
include/linux/mm.h +1 -2
index e8ddc98..3235615 100644
@@ -1129,8 +1129,7 @@ extern unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	unsigned long flag, unsigned long pgoff);
 extern unsigned long mmap_region(struct file *file, unsigned long addr,
 	unsigned long len, unsigned long flags,
-	unsigned int vm_flags, unsigned long pgoff,
-	int accountable);
+	unsigned int vm_flags, unsigned long pgoff);
 
 static inline unsigned long do_mmap(struct file *file, unsigned long addr,
 	unsigned long len, unsigned long prot,
ipc/shm.c +1 -0
index f8f69fa..05d51d2 100644
@@ -340,6 +340,7 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 	struct file * file;
 	char name[13];
 	int id;
+	int acctflag = 0;
 
 	if (size < SHMMIN || size > ns->shm_ctlmax)
 		return -EINVAL;

ipc/shm.c static int newseg(struct ipc_namespace *ns, struct ipc_params *params) · line 371

Quoted 9 lines
ipc/shm.c +4 -2
@@ -364,11 +365,12 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 
 	sprintf (name, "SYSV%08x", key);
 	if (shmflg & SHM_HUGETLB) {
-		/* hugetlb_file_setup takes care of mlock user accounting */
-		file = hugetlb_file_setup(name, size);
+		/* hugetlb_file_setup applies strict accounting */
+		if (shmflg & SHM_NORESERVE)
+			acctflag = VM_NORESERVE;
+		file = hugetlb_file_setup(name, size, acctflag);

Does this change semantics? At first look it appears that you are now implementing SHM_NORESERVE for SHM_HUGETLB where we did not previously do so? Not that this is necesarily a bad thing, just feels like its changing.

Quoted 42 lines
ipc/shm.c +0 -1
 		shp->mlock_user = current_user();
 	} else {
-		int acctflag = 0;
 		/*
 		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
 	 	 * if it's asked for.
mm/fremap.c +1 -1
index 736ba7f..b6ec85a 100644
@@ -198,7 +198,7 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
 			flags &= MAP_NONBLOCK;
 			get_file(file);
 			addr = mmap_region(file, start, size,
-					flags, vma->vm_flags, pgoff, 1);
+					flags, vma->vm_flags, pgoff);
 			fput(file);
 			if (IS_ERR_VALUE(addr)) {
 				err = addr;
mm/hugetlb.c +3 -5
index 618e983..2074642 100644
@@ -2269,14 +2269,12 @@ void hugetlb_change_protection(struct vm_area_struct *vma,
 
 int hugetlb_reserve_pages(struct inode *inode,
 					long from, long to,
-					struct vm_area_struct *vma)
+					struct vm_area_struct *vma,
+					int acctflag)
 {
-	long ret, chg;
+	long ret = 0, chg;
 	struct hstate *h = hstate_inode(inode);
 
-	if (vma && vma->vm_flags & VM_NORESERVE)
-		return 0;
-
 	/*
 	 * Shared mappings base their reservation on the number of pages that
 	 * are already allocated on behalf of the file. Private mappings need

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2285

Quoted 3 lines
mm/hugetlb.c +0 -0
@@ -2285,22 +2283,25 @@ int hugetlb_reserve_pages(struct inode *inode,
 	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		chg = region_chg(&inode->i_mapping->private_list, from, to);

I thought the region map for a VM_SHARED mapping is meant to contain those pages for which we already have a reservation allocated. So that if we have overlapping VM_RESERVE and VM_NORESERVE mappings we know that we did have a page reserved at fault time and know whether we can take it from the reserve portion of the pool. By letting this get executed for VM_NORESERVE mappings that would seem to be getting out of sync, which doesn't sound right. I don't see any updates to the documentation on the meanings of reservation map if you are changing it?

Here is the commentry from mm/hugetlb.c:

* The private mapping reservation is represented in a subtly different
* manner to a shared mapping. A shared mapping has a region map associated
* with the underlying file, this region map represents the backing file
* pages which have ever had a reservation assigned which this persists even
* after the page is instantiated. A private mapping has a region map
* associated with the original mmap which is attached to all VMAs which
* reference it, this region map represents those offsets which have consumed
* reservation ie. where pages have been instantiated.

Quoted 15 lines
mm/hugetlb.c +1 -9
-	else {
-		struct resv_map *resv_map = resv_map_alloc();
-		if (!resv_map)
-			return -ENOMEM;
-
+	else
 		chg = to - from;
 
-		set_vma_resv_map(vma, resv_map);
-		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
-	}
-
 	if (chg < 0)
 		return chg;
 

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2301

Quoted 10 lines
mm/hugetlb.c +8 -0
 	if (hugetlb_get_quota(inode->i_mapping, chg))
 		return -ENOSPC;
+
+	/*
+	 * Only apply hugepage reservation if asked. We still have to
+	 * take the filesystem quota because it is an upper limit
+	 * defined for the mount and not necessarily memory as a whole
+	 */
+	if (acctflag & VM_NORESERVE) {
+		reset_vma_resv_huge_pages(vma);

Why do we now need to do this in the non-reserve case? We didn't need to do it before.

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2304

Quoted 3 lines
mm/hugetlb.c +3 -0
+		return 0;
+	}
+

This also seems like a semantic change. Previously NO_RESERVE did not take quota, now it does. NO_RESERVE used to mean that we took our chances on there being pages available at fault time both for quota and in the pools. Now it means that we only risk there being no pages. Does that not significantly change semantics. The main use case for NO_RESERVE was mapping a massive area to expand into, this might now be refused for quota when previously it was not.

Quoted 118 lines
mm/hugetlb.c +10 -0
 	ret = hugetlb_acct_memory(h, chg);
 	if (ret < 0) {
 		hugetlb_put_quota(inode->i_mapping, chg);
@@ -2308,6 +2309,16 @@ int hugetlb_reserve_pages(struct inode *inode,
 	}
 	if (!vma || vma->vm_flags & VM_SHARED)
 		region_add(&inode->i_mapping->private_list, from, to);
+	else {
+		struct resv_map *resv_map = resv_map_alloc();
+
+		if (!resv_map)
+			return -ENOMEM;
+
+		set_vma_resv_map(vma, resv_map);
+		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
+	}
+
 	return 0;
 }
 
mm/mmap.c +22 -16
index 214b6a2..6f70db8 100644
@@ -918,7 +918,6 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	struct inode *inode;
 	unsigned int vm_flags;
 	int error;
-	int accountable = 1;
 	unsigned long reqprot = prot;
 
 	/*
@@ -1019,8 +1018,6 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 					return -EPERM;
 				vm_flags &= ~VM_MAYEXEC;
 			}
-			if (is_file_hugepages(file))
-				accountable = 0;
 
 			if (!file->f_op || !file->f_op->mmap)
 				return -ENODEV;
@@ -1053,8 +1050,7 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	if (error)
 		return error;
 
-	return mmap_region(file, addr, len, flags, vm_flags, pgoff,
-			   accountable);
+	return mmap_region(file, addr, len, flags, vm_flags, pgoff);
 }
 EXPORT_SYMBOL(do_mmap_pgoff);
 
@@ -1092,17 +1088,23 @@ int vma_wants_writenotify(struct vm_area_struct *vma)
 
 /*
  * We account for memory if it's a private writeable mapping,
- * and VM_NORESERVE wasn't set.
+ * not hugepages and VM_NORESERVE wasn't set.
  */
-static inline int accountable_mapping(unsigned int vm_flags)
+static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
 {
+	/*
+	 * hugetlb has its own accounting separate from the core VM
+	 * VM_HUGETLB may not be set yet so we cannot check for that flag.
+	 */
+	if (file && is_file_hugepages(file))
+		return 0;
+
 	return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
 }
 
 unsigned long mmap_region(struct file *file, unsigned long addr,
 			  unsigned long len, unsigned long flags,
-			  unsigned int vm_flags, unsigned long pgoff,
-			  int accountable)
+			  unsigned int vm_flags, unsigned long pgoff)
 {
 	struct mm_struct *mm = current->mm;
 	struct vm_area_struct *vma, *prev;
@@ -1128,18 +1130,22 @@ munmap_back:
 
 	/*
 	 * Set 'VM_NORESERVE' if we should not account for the
-	 * memory use of this mapping. We only honor MAP_NORESERVE
-	 * if we're allowed to overcommit memory.
+	 * memory use of this mapping.
 	 */
-	if ((flags & MAP_NORESERVE) && sysctl_overcommit_memory != OVERCOMMIT_NEVER)
-		vm_flags |= VM_NORESERVE;
-	if (!accountable)
-		vm_flags |= VM_NORESERVE;
+	if ((flags & MAP_NORESERVE)) {
+	 	/* * We honor MAP_NORESERVE if allowed to overcommit */
+		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
+			vm_flags |= VM_NORESERVE;
+
+		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
+		if (file && is_file_hugepages(file))
+			vm_flags |= VM_NORESERVE;
+	}
 
 	/*
 	 * Private writable mapping: check memory availability
 	 */
-	if (accountable_mapping(vm_flags)) {
+	if (accountable_mapping(file, vm_flags)) {
 		charged = len >> PAGE_SHIFT;
 		if (security_vm_enough_memory(charged))
 			return -ENOMEM;
mm/mprotect.c +0 -0
index abe2694..258197b 100644
@@ -151,10 +151,11 @@ mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 	/*
 	 * If we make a private mapping writable we increase our commit;
 	 * but (without finer accounting) cannot reduce our commit if we

mm/mprotect.c mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, · line 161

Quoted 10 lines
mm/mprotect.c +3 -2
-	 * make it unwritable again.
+	 * make it unwritable again. hugetlb mapping were accounted for
+	 * even if read-only so there is no need to account for them here
 	 */
 	if (newflags & VM_WRITE) {
-		if (!(oldflags & (VM_ACCOUNT|VM_WRITE|
+		if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 						VM_SHARED|VM_NORESERVE))) {
 			charged = nrpages;
 			if (security_vm_enough_memory(charged))

-apw

#34 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

Quoted 143 lines · Andy Whitcroft
Quoted 142 lines · Mel Gorman
Quoted 10 lines · Linus Torvalds

Another week (and a half), another -rc.

Arch updates (sparc, blackfin), driver updates (dvb, mmc, ide), ACPI updates, FS updates (ubifs, btrfs), you name it. It's all there.

But more importantly, people really have been working on regressions, and hopefully this closes a nice set of the top one, and hopefully without introducing too many new ones.

Hugepages are currently busted due to commit fc8744adc870a8d4366908221508bb113d8b72ee and the problem is in 2.6.29-rc4. There was a bit of a discussion but it didn't get very far and then I went offline for the weekend. Here is a revised patch that tries to bring hugetlbfs more in line with what the core VM is doing by dealing with VM_ACCOUNT and VM_NORESERVE.

=============
Do not account for the address space used by hugetlbfs using VM_ACCOUNT

When overcommit is disabled, the core VM accounts for pages used by anonymous shared, private mappings and special mappings. It keeps track of VMAs that should be accounted for with VM_ACCOUNT and VMAs that never had a reserve with VM_NORESERVE.

Overcommit for hugetlbfs is much riskier than overcommit for base pages due to contiguity requirements. It avoids overcommiting on both shared and private mappings using reservation counters that are checked and updated during mmap(). This ensures (within limits) that hugepages exist in the future when faults occurs or it is too easy to applications to be SIGKILLed.

As hugetlbfs makes its own reservations of a different unit to the base page size, VM_ACCOUNT should never be set. Even if the units were correct, we would double account for the usage in the core VM and hugetlbfs. VM_NORESERVE may be set because an application can request no reserves be made for hugetlbfs at the risk of getting killed later.

With commit fc8744adc870a8d4366908221508bb113d8b72ee, VM_NORESERVE and VM_ACCOUNT are getting unconditionally set for hugetlbfs-backed mappings. This breaks the accounting for both the core VM and hugetlbfs, can trigger an OOM storm when hugepage pools are too small lockups and corrupted counters otherwise are used. This patch brings hugetlbfs more in line with how the core VM treats VM_NORESERVE but prevents VM_ACCOUNT being set.

---

 fs/hugetlbfs/inode.c    |    8 +++++---
 include/linux/hugetlb.h |    5 +++--
 include/linux/mm.h      |    3 +--
 ipc/shm.c               |    8 +++++---
 mm/fremap.c             |    2 +-
 mm/hugetlb.c            |   39 +++++++++++++++++++++++++--------------
 mm/mmap.c               |   38 ++++++++++++++++++++++----------------
 mm/mprotect.c           |    5 +++--
 8 files changed, 65 insertions(+), 43 deletions(-)
fs/hugetlbfs/inode.c +5 -3
index 6903d37..9b800d9 100644
@@ -108,7 +108,8 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 
 	if (hugetlb_reserve_pages(inode,
 				vma->vm_pgoff >> huge_page_order(h),
-				len >> huge_page_shift(h), vma))
+				len >> huge_page_shift(h), vma,
+				vma->vm_flags))
 		goto out;
 
 	ret = 0;
@@ -947,7 +948,7 @@ static int can_do_hugetlb_shm(void)
 			can_do_mlock());
 }
 
-struct file *hugetlb_file_setup(const char *name, size_t size)
+struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag)
 {
 	int error = -ENOMEM;
 	struct file *file;
@@ -981,7 +982,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size)
 
 	error = -ENOMEM;
 	if (hugetlb_reserve_pages(inode, 0,
-			size >> huge_page_shift(hstate_inode(inode)), NULL))
+			size >> huge_page_shift(hstate_inode(inode)), NULL,
+			acctflag))
 		goto out_inode;
 
 	d_instantiate(dentry, inode);
include/linux/hugetlb.h +3 -2
index f1d2fba..af09660 100644
@@ -33,7 +33,8 @@ unsigned long hugetlb_total_pages(void);
 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, int write_access);
 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
-						struct vm_area_struct *vma);
+						struct vm_area_struct *vma,
+						int acctflags);
 void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed);
 
 extern unsigned long hugepages_treat_as_movable;
@@ -138,7 +139,7 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 
 extern const struct file_operations hugetlbfs_file_operations;
 extern struct vm_operations_struct hugetlb_vm_ops;
-struct file *hugetlb_file_setup(const char *name, size_t);
+struct file *hugetlb_file_setup(const char *name, size_t, int);
 int hugetlb_get_quota(struct address_space *mapping, long delta);
 void hugetlb_put_quota(struct address_space *mapping, long delta);
 
include/linux/mm.h +1 -2
index e8ddc98..3235615 100644
@@ -1129,8 +1129,7 @@ extern unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	unsigned long flag, unsigned long pgoff);
 extern unsigned long mmap_region(struct file *file, unsigned long addr,
 	unsigned long len, unsigned long flags,
-	unsigned int vm_flags, unsigned long pgoff,
-	int accountable);
+	unsigned int vm_flags, unsigned long pgoff);
 
 static inline unsigned long do_mmap(struct file *file, unsigned long addr,
 	unsigned long len, unsigned long prot,
ipc/shm.c +1 -0
index f8f69fa..05d51d2 100644
@@ -340,6 +340,7 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 	struct file * file;
 	char name[13];
 	int id;
+	int acctflag = 0;
 
 	if (size < SHMMIN || size > ns->shm_ctlmax)
 		return -EINVAL;
@@ -364,11 +365,12 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 
Quoted 14 lines

ipc/shm.c static int newseg(struct ipc_namespace *ns, struct ipc_params *params) · line 371

Quoted 8 lines
ipc/shm.c +4 -2
 	sprintf (name, "SYSV%08x", key);
 	if (shmflg & SHM_HUGETLB) {
-		/* hugetlb_file_setup takes care of mlock user accounting */
-		file = hugetlb_file_setup(name, size);
+		/* hugetlb_file_setup applies strict accounting */
+		if (shmflg & SHM_NORESERVE)
+			acctflag = VM_NORESERVE;
+		file = hugetlb_file_setup(name, size, acctflag);

Does this change semantics? At first look it appears that you are now implementing SHM_NORESERVE for SHM_HUGETLB where we did not previously do so? Not that this is necesarily a bad thing, just feels like its changing.

Yes, semantics for shared mappings are changed so that they obey SHM_NORESERVE. This was part of bringing hugetlbfs closer to the behaviour of the core VM but I should have called it out better in the changelog.

Quoted 46 lines
Quoted 42 lines
ipc/shm.c +0 -1
 		shp->mlock_user = current_user();
 	} else {
-		int acctflag = 0;
 		/*
 		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
 	 	 * if it's asked for.
mm/fremap.c +1 -1
index 736ba7f..b6ec85a 100644
@@ -198,7 +198,7 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
 			flags &= MAP_NONBLOCK;
 			get_file(file);
 			addr = mmap_region(file, start, size,
-					flags, vma->vm_flags, pgoff, 1);
+					flags, vma->vm_flags, pgoff);
 			fput(file);
 			if (IS_ERR_VALUE(addr)) {
 				err = addr;
mm/hugetlb.c +3 -5
index 618e983..2074642 100644
@@ -2269,14 +2269,12 @@ void hugetlb_change_protection(struct vm_area_struct *vma,
 
 int hugetlb_reserve_pages(struct inode *inode,
 					long from, long to,
-					struct vm_area_struct *vma)
+					struct vm_area_struct *vma,
+					int acctflag)
 {
-	long ret, chg;
+	long ret = 0, chg;
 	struct hstate *h = hstate_inode(inode);
 
-	if (vma && vma->vm_flags & VM_NORESERVE)
-		return 0;
-
 	/*
 	 * Shared mappings base their reservation on the number of pages that
 	 * are already allocated on behalf of the file. Private mappings need

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2285

Quoted 3 lines
mm/hugetlb.c +0 -0
@@ -2285,22 +2283,25 @@ int hugetlb_reserve_pages(struct inode *inode,
 	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		chg = region_chg(&inode->i_mapping->private_list, from, to);
Quoted 7 lines

I thought the region map for a VM_SHARED mapping is meant to contain those pages for which we already have a reservation allocated. So that if we have overlapping VM_RESERVE and VM_NORESERVE mappings we know that we did have a page reserved at fault time and know whether we can take it from the reserve portion of the pool. By letting this get executed for VM_NORESERVE mappings that would seem to be getting out of sync, which doesn't sound right.

This part doesn't get executed for NORESERVE so while region_chg() took place to calculate a reservation, region_add() did not get called to commit it.

Quoted 5 lines

I don't see any updates to the documentation
on the meanings of reservation map if you are changing it?

Here is the commentry from mm/hugetlb.c:

Quoted 9 lines

* The private mapping reservation is represented in a subtly different
* manner to a shared mapping. A shared mapping has a region map associated
* with the underlying file, this region map represents the backing file
* pages which have ever had a reservation assigned which this persists even
* after the page is instantiated. A private mapping has a region map
* associated with the original mmap which is attached to all VMAs which
* reference it, this region map represents those offsets which have consumed
* reservation ie. where pages have been instantiated.

This is still accurate.

Quoted 15 lines
Quoted 15 lines
mm/hugetlb.c +1 -9
-	else {
-		struct resv_map *resv_map = resv_map_alloc();
-		if (!resv_map)
-			return -ENOMEM;
-
+	else
 		chg = to - from;
 
-		set_vma_resv_map(vma, resv_map);
-		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
-	}
-
 	if (chg < 0)
 		return chg;
 
Quoted 14 lines

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2301

Quoted 10 lines
mm/hugetlb.c +8 -0
 	if (hugetlb_get_quota(inode->i_mapping, chg))
 		return -ENOSPC;
+
+	/*
+	 * Only apply hugepage reservation if asked. We still have to
+	 * take the filesystem quota because it is an upper limit
+	 * defined for the mount and not necessarily memory as a whole
+	 */
+	if (acctflag & VM_NORESERVE) {
+		reset_vma_resv_huge_pages(vma);

Why do we now need to do this in the non-reserve case? We didn't need to do it before.

True, it was largely defensive against anything being in page_private that would make it think it had reserves.

Quoted 9 lines

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2304

Quoted 3 lines
mm/hugetlb.c +3 -0
+		return 0;
+	}
+

This also seems like a semantic change. Previously NO_RESERVE did not take quota, now it does. NO_RESERVE used to mean that we took our chances on there being pages available at fault time both for quota and in the pools. Now it means that we only risk there being no pages. Does that not significantly change semantics.

Yes, and this was a mistake. For noreserve mappings, we may now be taking twice the amount of quota and probably leaking it. This is wrong and I need to move the check for quota below the check for VM_NORESERVE. Good spot.

Quoted 124 lines

The main use case for
NO_RESERVE was mapping a massive area to expand into, this might now be refused for quota when previously it was not.

Quoted 120 lines
mm/hugetlb.c +10 -0
 	ret = hugetlb_acct_memory(h, chg);
 	if (ret < 0) {
 		hugetlb_put_quota(inode->i_mapping, chg);
@@ -2308,6 +2309,16 @@ int hugetlb_reserve_pages(struct inode *inode,
 	}
 	if (!vma || vma->vm_flags & VM_SHARED)
 		region_add(&inode->i_mapping->private_list, from, to);
+	else {
+		struct resv_map *resv_map = resv_map_alloc();
+
+		if (!resv_map)
+			return -ENOMEM;
+
+		set_vma_resv_map(vma, resv_map);
+		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
+	}
+
 	return 0;
 }
 
mm/mmap.c +22 -16
index 214b6a2..6f70db8 100644
@@ -918,7 +918,6 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	struct inode *inode;
 	unsigned int vm_flags;
 	int error;
-	int accountable = 1;
 	unsigned long reqprot = prot;
 
 	/*
@@ -1019,8 +1018,6 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 					return -EPERM;
 				vm_flags &= ~VM_MAYEXEC;
 			}
-			if (is_file_hugepages(file))
-				accountable = 0;
 
 			if (!file->f_op || !file->f_op->mmap)
 				return -ENODEV;
@@ -1053,8 +1050,7 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 	if (error)
 		return error;
 
-	return mmap_region(file, addr, len, flags, vm_flags, pgoff,
-			   accountable);
+	return mmap_region(file, addr, len, flags, vm_flags, pgoff);
 }
 EXPORT_SYMBOL(do_mmap_pgoff);
 
@@ -1092,17 +1088,23 @@ int vma_wants_writenotify(struct vm_area_struct *vma)
 
 /*
  * We account for memory if it's a private writeable mapping,
- * and VM_NORESERVE wasn't set.
+ * not hugepages and VM_NORESERVE wasn't set.
  */
-static inline int accountable_mapping(unsigned int vm_flags)
+static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
 {
+	/*
+	 * hugetlb has its own accounting separate from the core VM
+	 * VM_HUGETLB may not be set yet so we cannot check for that flag.
+	 */
+	if (file && is_file_hugepages(file))
+		return 0;
+
 	return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
 }
 
 unsigned long mmap_region(struct file *file, unsigned long addr,
 			  unsigned long len, unsigned long flags,
-			  unsigned int vm_flags, unsigned long pgoff,
-			  int accountable)
+			  unsigned int vm_flags, unsigned long pgoff)
 {
 	struct mm_struct *mm = current->mm;
 	struct vm_area_struct *vma, *prev;
@@ -1128,18 +1130,22 @@ munmap_back:
 
 	/*
 	 * Set 'VM_NORESERVE' if we should not account for the
-	 * memory use of this mapping. We only honor MAP_NORESERVE
-	 * if we're allowed to overcommit memory.
+	 * memory use of this mapping.
 	 */
-	if ((flags & MAP_NORESERVE) && sysctl_overcommit_memory != OVERCOMMIT_NEVER)
-		vm_flags |= VM_NORESERVE;
-	if (!accountable)
-		vm_flags |= VM_NORESERVE;
+	if ((flags & MAP_NORESERVE)) {
+	 	/* * We honor MAP_NORESERVE if allowed to overcommit */
+		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
+			vm_flags |= VM_NORESERVE;
+
+		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
+		if (file && is_file_hugepages(file))
+			vm_flags |= VM_NORESERVE;
+	}
 
 	/*
 	 * Private writable mapping: check memory availability
 	 */
-	if (accountable_mapping(vm_flags)) {
+	if (accountable_mapping(file, vm_flags)) {
 		charged = len >> PAGE_SHIFT;
 		if (security_vm_enough_memory(charged))
 			return -ENOMEM;
mm/mprotect.c +2 -1
index abe2694..258197b 100644
@@ -151,10 +151,11 @@ mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 	/*
 	 * If we make a private mapping writable we increase our commit;
 	 * but (without finer accounting) cannot reduce our commit if we
-	 * make it unwritable again.
+	 * make it unwritable again. hugetlb mapping were accounted for
+	 * even if read-only so there is no need to account for them here
Quoted 9 lines

mm/mprotect.c mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, · line 161

Quoted 8 lines
mm/mprotect.c +1 -1
 	 */
 	if (newflags & VM_WRITE) {
-		if (!(oldflags & (VM_ACCOUNT|VM_WRITE|
+		if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 						VM_SHARED|VM_NORESERVE))) {
 			charged = nrpages;
 			if (security_vm_enough_memory(charged))

Thanks, I'll get the quota thing fixed up.

Signature · 3 lines
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
#35 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

Quoted 16 lines · Mel Gorman
Quoted 8 lines

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2285

Quoted 7 lines
mm/hugetlb.c +0 -0
 	/*
 	 * Shared mappings base their reservation on the number of pages that
 	 * are already allocated on behalf of the file. Private mappings need
@@ -2285,22 +2283,25 @@ int hugetlb_reserve_pages(struct inode *inode,
 	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		chg = region_chg(&inode->i_mapping->private_list, from, to);
Quoted 7 lines

I thought the region map for a VM_SHARED mapping is meant to contain those pages for which we already have a reservation allocated. So that if we have overlapping VM_RESERVE and VM_NORESERVE mappings we know that we did have a page reserved at fault time and know whether we can take it from the reserve portion of the pool. By letting this get executed for VM_NORESERVE mappings that would seem to be getting out of sync, which doesn't sound right.

Quoted 3 lines

This part doesn't get executed for NORESERVE so while region_chg() took place to calculate a reservation, region_add() did not get called to commit it.

Bah I wrote those damn routines and even I get confused. As you say they are a prepare/commit pair and this is only the prepare phase you are executing, no persistant change is made; so its safe.

Quoted 9 lines
Quoted 6 lines

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2301

Quoted 2 lines
mm/hugetlb.c +2 -0
+	if (acctflag & VM_NORESERVE) {
+		reset_vma_resv_huge_pages(vma);

Why do we now need to do this in the non-reserve case? We didn't need to do it before.

True, it was largely defensive against anything being in page_private that would make it think it had reserves.

Defensive is good.

Quoted 10 lines
Quoted 9 lines

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2304

Quoted 3 lines
mm/hugetlb.c +3 -0
+		return 0;
+	}
+

This also seems like a semantic change. Previously NO_RESERVE did not take quota, now it does. NO_RESERVE used to mean that we took our chances on there being pages available at fault time both for quota and in the pools. Now it means that we only risk there being no pages. Does that not significantly change semantics.

Quoted 3 lines

Yes, and this was a mistake. For noreserve mappings, we may now be taking twice the amount of quota and probably leaking it. This is wrong and I need to move the check for quota below the check for VM_NORESERVE. Good spot.

Thanks.

-apw

#36 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

Quoted 8 lines · Andy Whitcroft
Quoted 5 lines

<SNIP>

Yes, and this was a mistake. For noreserve mappings, we may now be taking twice the amount of quota and probably leaking it. This is wrong and I need to move the check for quota below the check for VM_NORESERVE. Good spot.

Thanks.

How about this?

=====
[PATCH] Do not account for hugetlbfs quota at mmap() time if mapping *_NORESERVE

Commit 5a6fe125950676015f5108fb71b2a67441755003 brought hugetlbfs more in line with the core VM by obeying VM_NORESERVE and not reserving hugepages for both shared and private mappings when [SHM|MAP]_NORESERVE are specified. However, it is still taking filesystem quota unconditionally and this leads to double-accounting.

At fault time, if there are no reserves and attempt is made to allocate the page and account for filesystem quota. If either fail, the fault fails. This patch prevents quota being taken when [SHM|MAP]_NORESERVE is specified. To help prevent this mistake happening again, it improves the documentation of hugetlb_reserve_pages().

---

 hugetlb.c |   29 +++++++++++++++++++++++------
 1 file changed, 23 insertions(+), 6 deletions(-)
mm/hugetlb.c +23 -6
index 2074642..b0b63cd 100644
@@ -2289,24 +2289,41 @@ int hugetlb_reserve_pages(struct inode *inode,
 	if (chg < 0)
 		return chg;
 
-	if (hugetlb_get_quota(inode->i_mapping, chg))
-		return -ENOSPC;
-
 	/*
-	 * Only apply hugepage reservation if asked. We still have to
-	 * take the filesystem quota because it is an upper limit
-	 * defined for the mount and not necessarily memory as a whole
+	 * Only apply hugepage reservation if asked. At fault time, an
+	 * attempt will be made for VM_NORESERVE to allocate a page
+	 * and filesystem quota without using reserves
 	 */
 	if (acctflag & VM_NORESERVE) {
 		reset_vma_resv_huge_pages(vma);
 		return 0;
 	}
 
+	/* There must be enough filesystem quota for the mapping */
+	if (hugetlb_get_quota(inode->i_mapping, chg))
+		return -ENOSPC;
+
+	/*
+	 * Check enough hugepages are available for the reservation.
+	 * Hand back the quota if there are not
+	 */
 	ret = hugetlb_acct_memory(h, chg);
 	if (ret < 0) {
 		hugetlb_put_quota(inode->i_mapping, chg);
 		return ret;
 	}
+
+	/*
+	 * Account for the reservations made. Shared mappings record regions
+	 * that have reservations as they are shared by multiple VMAs.
+	 * When the last VMA disappears, the region map says how much
+	 * the reservation was and the page cache tells how much of
+	 * the reservation was consumed. Private mappings are per-VMA and
+	 * only the consumed reservations are tracked. When the VMA
+	 * disappears, the original reservation is the VMA size and the
+	 * consumed reservations are stored in the map. Here, we just need
+	 * to allocate the region map.
+	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		region_add(&inode->i_mapping->private_list, from, to);
 	else {
#37 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

Quoted 66 lines · Mel Gorman
Quoted 8 lines · Andy Whitcroft
Quoted 5 lines

<SNIP>

Yes, and this was a mistake. For noreserve mappings, we may now be taking twice the amount of quota and probably leaking it. This is wrong and I need to move the check for quota below the check for VM_NORESERVE. Good spot.

Thanks.

How about this?

=====
[PATCH] Do not account for hugetlbfs quota at mmap() time if mapping *_NORESERVE

Commit 5a6fe125950676015f5108fb71b2a67441755003 brought hugetlbfs more in line with the core VM by obeying VM_NORESERVE and not reserving hugepages for both shared and private mappings when [SHM|MAP]_NORESERVE are specified. However, it is still taking filesystem quota unconditionally and this leads to double-accounting.

At fault time, if there are no reserves and attempt is made to allocate the page and account for filesystem quota. If either fail, the fault fails. This patch prevents quota being taken when [SHM|MAP]_NORESERVE is specified. To help prevent this mistake happening again, it improves the documentation of hugetlb_reserve_pages().

---

 hugetlb.c |   29 +++++++++++++++++++++++------
 1 file changed, 23 insertions(+), 6 deletions(-)
mm/hugetlb.c +11 -6
index 2074642..b0b63cd 100644
@@ -2289,24 +2289,41 @@ int hugetlb_reserve_pages(struct inode *inode,
 	if (chg < 0)
 		return chg;
 
-	if (hugetlb_get_quota(inode->i_mapping, chg))
-		return -ENOSPC;
-
 	/*
-	 * Only apply hugepage reservation if asked. We still have to
-	 * take the filesystem quota because it is an upper limit
-	 * defined for the mount and not necessarily memory as a whole
+	 * Only apply hugepage reservation if asked. At fault time, an
+	 * attempt will be made for VM_NORESERVE to allocate a page
+	 * and filesystem quota without using reserves
 	 */
 	if (acctflag & VM_NORESERVE) {
 		reset_vma_resv_huge_pages(vma);
 		return 0;
 	}
 
+	/* There must be enough filesystem quota for the mapping */
+	if (hugetlb_get_quota(inode->i_mapping, chg))
+		return -ENOSPC;
+
+	/*
+	 * Check enough hugepages are available for the reservation.
+	 * Hand back the quota if there are not
+	 */
 	ret = hugetlb_acct_memory(h, chg);

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2320

Quoted 10 lines
mm/hugetlb.c +6 -0
 	if (ret < 0) {
 		hugetlb_put_quota(inode->i_mapping, chg);
 		return ret;
 	}
+
+	/*
+	 * Account for the reservations made. Shared mappings record regions
+	 * that have reservations as they are shared by multiple VMAs.
+	 * When the last VMA disappears, the region map says how much
+	 * the reservation was and the page cache tells how much of

how much of the area had a reservation and the page cache ...

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2329

Quoted 9 lines
mm/hugetlb.c +6 -0
+	 * the reservation was consumed. Private mappings are per-VMA and
+	 * only the consumed reservations are tracked. When the VMA
+	 * disappears, the original reservation is the VMA size and the
+	 * consumed reservations are stored in the map. Here, we just need
+	 * to allocate the region map.
+	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		region_add(&inode->i_mapping->private_list, from, to);
 	else {

I am worried about the failure semantics here. For shared mappings we alloc any memory we need with region_chg(), then take quota, then apply it and done. For private mappings we take the quota and then allocate any memory we need. If that fails we do not return the quota? Sounds like a quota leak to me?

I think we do need to allocate the private map up when we do the region_chg() for the shared case. And I think neither of those is needed for NORESERVE mappings?

-apw

#38 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

Quoted 80 lines · Andy Whitcroft
Quoted 66 lines · Mel Gorman
Quoted 8 lines · Andy Whitcroft
Quoted 5 lines

<SNIP>

Yes, and this was a mistake. For noreserve mappings, we may now be taking twice the amount of quota and probably leaking it. This is wrong and I need to move the check for quota below the check for VM_NORESERVE. Good spot.

Thanks.

How about this?

=====
[PATCH] Do not account for hugetlbfs quota at mmap() time if mapping *_NORESERVE

Commit 5a6fe125950676015f5108fb71b2a67441755003 brought hugetlbfs more in line with the core VM by obeying VM_NORESERVE and not reserving hugepages for both shared and private mappings when [SHM|MAP]_NORESERVE are specified. However, it is still taking filesystem quota unconditionally and this leads to double-accounting.

At fault time, if there are no reserves and attempt is made to allocate the page and account for filesystem quota. If either fail, the fault fails. This patch prevents quota being taken when [SHM|MAP]_NORESERVE is specified. To help prevent this mistake happening again, it improves the documentation of hugetlb_reserve_pages().

---

 hugetlb.c |   29 +++++++++++++++++++++++------
 1 file changed, 23 insertions(+), 6 deletions(-)
mm/hugetlb.c +11 -6
index 2074642..b0b63cd 100644
@@ -2289,24 +2289,41 @@ int hugetlb_reserve_pages(struct inode *inode,
 	if (chg < 0)
 		return chg;
 
-	if (hugetlb_get_quota(inode->i_mapping, chg))
-		return -ENOSPC;
-
 	/*
-	 * Only apply hugepage reservation if asked. We still have to
-	 * take the filesystem quota because it is an upper limit
-	 * defined for the mount and not necessarily memory as a whole
+	 * Only apply hugepage reservation if asked. At fault time, an
+	 * attempt will be made for VM_NORESERVE to allocate a page
+	 * and filesystem quota without using reserves
 	 */
 	if (acctflag & VM_NORESERVE) {
 		reset_vma_resv_huge_pages(vma);
 		return 0;
 	}
 
+	/* There must be enough filesystem quota for the mapping */
+	if (hugetlb_get_quota(inode->i_mapping, chg))
+		return -ENOSPC;
+
+	/*
+	 * Check enough hugepages are available for the reservation.
+	 * Hand back the quota if there are not
+	 */
 	ret = hugetlb_acct_memory(h, chg);

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2320

Quoted 10 lines
mm/hugetlb.c +6 -0
 	if (ret < 0) {
 		hugetlb_put_quota(inode->i_mapping, chg);
 		return ret;
 	}
+
+	/*
+	 * Account for the reservations made. Shared mappings record regions
+	 * that have reservations as they are shared by multiple VMAs.
+	 * When the last VMA disappears, the region map says how much
+	 * the reservation was and the page cache tells how much of

how much of the area had a reservation and the page cache ...

Quoted 16 lines

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2329

Quoted 9 lines
mm/hugetlb.c +6 -0
+	 * the reservation was consumed. Private mappings are per-VMA and
+	 * only the consumed reservations are tracked. When the VMA
+	 * disappears, the original reservation is the VMA size and the
+	 * consumed reservations are stored in the map. Here, we just need
+	 * to allocate the region map.
+	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		region_add(&inode->i_mapping->private_list, from, to);
 	else {

I am worried about the failure semantics here. For shared mappings we alloc any memory we need with region_chg(), then take quota, then apply it and done. For private mappings we take the quota and then allocate any memory we need. If that fails we do not return the quota? Sounds like a quota leak to me?

It is a quota leak.

Quoted 4 lines

I think we do need to allocate the private map up when we do the region_chg() for the shared case. And I think neither of those is needed for NORESERVE mappings?

Right. This shuffle was based on the assumption quota mattered and had to be accounted for. As you have pointed out, this was wrong and so all the shuffling was pointless. The correct thing to do is a partial revert of the hunk that alters hugetlb_reserve_pages() and then add back acctflag to check VM_NORESERVE so that SHM_NORESERVE still has meaning.

After the following patch is applied, the net change to hugetlb_reserve_pages() is small - mainly additional comments documenting what's going on.

======
Subject: [PATCH] Do not account for hugetlbfs quota at mmap() time if mapping [SHM|MAP]_NORESERVE

Commit 5a6fe125950676015f5108fb71b2a67441755003 brought hugetlbfs more in line with the core VM by obeying VM_NORESERVE and not reserving hugepages for both shared and private mappings when [SHM|MAP]_NORESERVE are specified. However, it is still taking filesystem quota unconditionally.

At fault time, if there are no reserves and attempt is made to allocate the page and account for filesystem quota. If either fail, the fault fails. The impact is that quota is getting accounted for twice. This patch partially reverts 5a6fe125950676015f5108fb71b2a67441755003. To help prevent this mistake happening again, it improves the documentation of hugetlb_reserve_pages()

mm/hugetlb.c +33 -20
index 2074642..107da3d 100644
@@ -2272,10 +2272,18 @@ int hugetlb_reserve_pages(struct inode *inode,
 					struct vm_area_struct *vma,
 					int acctflag)
 {
-	long ret = 0, chg;
+	long ret, chg;
 	struct hstate *h = hstate_inode(inode);
 
 	/*
+	 * Only apply hugepage reservation if asked. At fault time, an
+	 * attempt will be made for VM_NORESERVE to allocate a page
+	 * and filesystem quota without using reserves
+	 */
+	if (acctflag & VM_NORESERVE)
+		return 0;
+
+	/*
 	 * Shared mappings base their reservation on the number of pages that
 	 * are already allocated on behalf of the file. Private mappings need
 	 * to reserve the full area even if read-only as mprotect() may be
@@ -2283,42 +2291,47 @@ int hugetlb_reserve_pages(struct inode *inode,
 	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		chg = region_chg(&inode->i_mapping->private_list, from, to);
-	else
+	else {
+		struct resv_map *resv_map = resv_map_alloc();
+		if (!resv_map)
+			return -ENOMEM;
+
 		chg = to - from;
 
+		set_vma_resv_map(vma, resv_map);
+		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
+	}
+
 	if (chg < 0)
 		return chg;
 
+	/* There must be enough filesystem quota for the mapping */
 	if (hugetlb_get_quota(inode->i_mapping, chg))
 		return -ENOSPC;
 
 	/*
-	 * Only apply hugepage reservation if asked. We still have to
-	 * take the filesystem quota because it is an upper limit
-	 * defined for the mount and not necessarily memory as a whole
+	 * Check enough hugepages are available for the reservation.
+	 * Hand back the quota if there are not
 	 */
-	if (acctflag & VM_NORESERVE) {
-		reset_vma_resv_huge_pages(vma);
-		return 0;
-	}
-
 	ret = hugetlb_acct_memory(h, chg);
 	if (ret < 0) {
 		hugetlb_put_quota(inode->i_mapping, chg);
 		return ret;
 	}
+
+	/*
+	 * Account for the reservations made. Shared mappings record regions
+	 * that have reservations as they are shared by multiple VMAs.
+	 * When the last VMA disappears, the region map says how much
+	 * the reservation was and the page cache tells how much of
+	 * the reservation was consumed. Private mappings are per-VMA and
+	 * only the consumed reservations are tracked. When the VMA
+	 * disappears, the original reservation is the VMA size and the
+	 * consumed reservations are stored in the map. Hence, nothing
+	 * else has to be done for private mappings here
+	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		region_add(&inode->i_mapping->private_list, from, to);
-	else {
-		struct resv_map *resv_map = resv_map_alloc();
-
-		if (!resv_map)
-			return -ENOMEM;
-
-		set_vma_resv_map(vma, resv_map);
-		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
-	}
-
 	return 0;
 }
 
#39 lore raw

Re: [PATCH] Do not account for the address space used by hugetlbfs using VM_ACCOUNT V2 (Was Linus 2.6.29-rc4)

Quoted 97 lines

Subject: [PATCH] Do not account for hugetlbfs quota at mmap() time if mapping [SHM|MAP]_NORESERVE

Commit 5a6fe125950676015f5108fb71b2a67441755003 brought hugetlbfs more in line with the core VM by obeying VM_NORESERVE and not reserving hugepages for both shared and private mappings when [SHM|MAP]_NORESERVE are specified. However, it is still taking filesystem quota unconditionally.

At fault time, if there are no reserves and attempt is made to allocate the page and account for filesystem quota. If either fail, the fault fails. The impact is that quota is getting accounted for twice. This patch partially reverts 5a6fe125950676015f5108fb71b2a67441755003. To help prevent this mistake happening again, it improves the documentation of hugetlb_reserve_pages()

mm/hugetlb.c +33 -12
index 2074642..107da3d 100644
@@ -2272,10 +2272,18 @@ int hugetlb_reserve_pages(struct inode *inode,
 					struct vm_area_struct *vma,
 					int acctflag)
 {
-	long ret = 0, chg;
+	long ret, chg;
 	struct hstate *h = hstate_inode(inode);
 
 	/*
+	 * Only apply hugepage reservation if asked. At fault time, an
+	 * attempt will be made for VM_NORESERVE to allocate a page
+	 * and filesystem quota without using reserves
+	 */
+	if (acctflag & VM_NORESERVE)
+		return 0;
+
+	/*
 	 * Shared mappings base their reservation on the number of pages that
 	 * are already allocated on behalf of the file. Private mappings need
 	 * to reserve the full area even if read-only as mprotect() may be
@@ -2283,42 +2291,47 @@ int hugetlb_reserve_pages(struct inode *inode,
 	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		chg = region_chg(&inode->i_mapping->private_list, from, to);
-	else
+	else {
+		struct resv_map *resv_map = resv_map_alloc();
+		if (!resv_map)
+			return -ENOMEM;
+
 		chg = to - from;
 
+		set_vma_resv_map(vma, resv_map);
+		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
+	}
+
 	if (chg < 0)
 		return chg;
 
+	/* There must be enough filesystem quota for the mapping */
 	if (hugetlb_get_quota(inode->i_mapping, chg))
 		return -ENOSPC;
 
 	/*
-	 * Only apply hugepage reservation if asked. We still have to
-	 * take the filesystem quota because it is an upper limit
-	 * defined for the mount and not necessarily memory as a whole
+	 * Check enough hugepages are available for the reservation.
+	 * Hand back the quota if there are not
 	 */
-	if (acctflag & VM_NORESERVE) {
-		reset_vma_resv_huge_pages(vma);
-		return 0;
-	}
-
 	ret = hugetlb_acct_memory(h, chg);
 	if (ret < 0) {
 		hugetlb_put_quota(inode->i_mapping, chg);
 		return ret;
 	}
+
+	/*
+	 * Account for the reservations made. Shared mappings record regions
+	 * that have reservations as they are shared by multiple VMAs.
+	 * When the last VMA disappears, the region map says how much
+	 * the reservation was and the page cache tells how much of
+	 * the reservation was consumed. Private mappings are per-VMA and
+	 * only the consumed reservations are tracked. When the VMA
+	 * disappears, the original reservation is the VMA size and the
+	 * consumed reservations are stored in the map. Hence, nothing
+	 * else has to be done for private mappings here
+	 */
 	if (!vma || vma->vm_flags & VM_SHARED)
 		region_add(&inode->i_mapping->private_list, from, to);
-	else {
-		struct resv_map *resv_map = resv_map_alloc();

mm/hugetlb.c int hugetlb_reserve_pages(struct inode *inode, · line 2336

Quoted 10 lines
mm/hugetlb.c +0 -8
-
-		if (!resv_map)
-			return -ENOMEM;
-
-		set_vma_resv_map(vma, resv_map);
-		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
-	}
-
 	return 0;
 }

-apw