commit 24398e39c8
Author: Johannes Berg <[email protected]>
Date: Wed Mar 28 10:58:36 2012 +0200
mac80211: set HT channel before association
removed IEEE80211_CONF_CHANGE_CHANNEL argument from ieee80211_hw_config,
which is required by iwl4965 driver, otherwise that driver does not
configure channel properly and is not able to associate.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Fix:
BUG: sleeping function called from invalid context at kernel/workqueue.c:2547
in_atomic(): 1, irqs_disabled(): 0, pid: 629, name: wpa_supplicant
2 locks held by wpa_supplicant/629:
#0: (rtnl_mutex){+.+.+.}, at: [<c08b2b84>] rtnl_lock+0x14/0x20
#1: (&trigger->leddev_list_lock){.+.?..}, at: [<c0867f41>] led_trigger_event+0x21/0x80
Pid: 629, comm: wpa_supplicant Not tainted 3.3.0-0.rc3.git5.1.fc17.i686
Call Trace:
[<c046a9f6>] __might_sleep+0x126/0x1d0
[<c0457d6c>] wait_on_work+0x2c/0x1d0
[<c045a09a>] __cancel_work_timer+0x6a/0x120
[<c045a160>] cancel_delayed_work_sync+0x10/0x20
[<f7dd3c22>] rtl8187_led_brightness_set+0x82/0xf0 [rtl8187]
[<c0867f7c>] led_trigger_event+0x5c/0x80
[<f7ff5e6d>] ieee80211_led_radio+0x1d/0x40 [mac80211]
[<f7ff3583>] ieee80211_stop_device+0x13/0x230 [mac80211]
Removing _sync is ok, because if led_on work is currently running
it will be finished before led_off work start to perform, since
they are always queued on the same mac80211 local->workqueue.
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=795176
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Larry Finger <[email protected]>
Acked-by: Hin-Tak Leung <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
commit 133d40f9a2
Author: Stanislaw Gruszka <[email protected]>
Date: Wed Mar 28 16:01:19 2012 +0200
mac80211: do not scan and monitor connection in parallel
add bug, which make possible to start a scan and never finish it, so
make every new scanning request finish with -EBUSY error. This can
happen on code paths where we finish connection monitoring and clear
IEEE80211_STA_*_POLL flags, but do not check if scan was deferred.
This patch fixes those code paths.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
There are various problems happened on 5GHz band not observed on
2.4 GHz (microcode errors, queue stuck, etc... ) . Also roaming
between 5GHz AP and 2GHz does not work very well. To workaround
the problems add option to disable 5GHz support. This will help
on environments where APs are dual-band, and devices will not try
to associate on band where issues happen.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Wey-Yi Guy <[email protected]>
ctx->vif is dereferenced in different part of iwlwifi code, so do not
nullify it.
This should address at least one of the possible reasons of WARNING at
iwlagn_mac_remove_interface, and perhaps some random crashes when
firmware reset is performed.
Cc: [email protected]
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Wey-Yi Guy <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Before we send probes in connection monitoring we check if scan is not
pending. But we do that check without locking. Fix that and also do not
start scan if connection monitoring is in progress.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This is workaround H/W or F/W bug, see in code comments. Without the fix
ping can receive duplicated ICMP frames while associated with legacy AP.
Reported-by: Walter Goldens <[email protected]>
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Helmut Schaa <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This is needed when we are concted to non 11n AP.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
While associated we should never have empty SSID, but life can be full
of surprises, and is allways better to print a warning than crash.
Before memcpy() in ieee80211_probereq_get() check ssid_len instead of
ssid pointer, sice pointer it always passed by "ssidie + 2" expression
to send probe functions, so practically never can be NULL.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Is possible that we will arm the tid_rx->reorder_timer after
del_timer_sync() in ___ieee80211_stop_rx_ba_session(). We need to stop
timer after RCU grace period finish, so move it to
ieee80211_free_tid_rx(). Timer will not be armed again, as
rcu_dereference(sta->ampdu_mlme.tid_rx[tid]) will return NULL.
Debug object detected problem with the following warning:
ODEBUG: free active (active state 0) object type: timer_list hint: sta_rx_agg_reorder_timer_expired+0x0/0xf0 [mac80211]
Bug report (with all warning messages):
https://bugzilla.redhat.com/show_bug.cgi?id=804007
Reported-by: "jan p. springer" <[email protected]>
Cc: [email protected]
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Another good catch from Jakub Kicinski. This patch fixes my
recent commit: ed61e2b020
"rt2x00: rt2800usb: rework txdone code"
We should reread status register only when nobody else start already
reading status i.e. test_and_set_bit(TX_STATUS_READING, flags) return 0.
Reported-by: Jakub Kicinski <[email protected]>
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This is fix for my current commit
ed61e2b020
"rt2x00: rt2800usb: rework txdone code"
We should schedule txdone work on timeout, otherwise if newer get
tx status from hardware, we will never report tx status to mac80211
and eventually never wakeup tx queue.
Reported-by: Jakub Kicinski <[email protected]>
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Current commit 0775f9f90c
"mac80211: remove spurious BSSID change flag" exposed bug on iwlegacy,
that we do not set BSSID address correctly and then device was not able
to receive frames after successful associate.
On the way fix scan canceling comment. Apparently ->post_associate()
do cancel scan itself, but scan cancel on BSS_CHANGED_BSSID is needed.
I'm not sure why, but when I removed it, I had frequent auth failures:
wlan4: send auth to 54:e6:fc:98:63:fe (try 1/3)
wlan4: send auth to 54:e6:fc:98:63:fe (try 2/3)
wlan4: send auth to 54:e6:fc:98:63:fe (try 3/3)
wlan4: authentication with 54:e6:fc:98:63:fe timed out
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This il->vif is dereferenced in different part of iwlegacy code, so do
not nullify it. This should fix random crashes observed in companion
with microcode errors i.e. crash in il3945_config_ap().
Additionally this should address also
WARNING: at drivers/net/wireless/iwlegacy/common.c:4656 il_mac_remove_interface
at least one of the possible reasons of that warning.
Cc: [email protected]
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
TX status fifo is limited to 16 elements. When we send more frames than
that, we can easily loose status, what is not good for rate scaling
algorithm.
On my testing the change does not degrade performance, actually make
is slightly better. Additionally with the patch I can see much less
various rt2x00 warnings in dmesg.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Acked-by: Helmut Schaa <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Tx statuses of aggregated subframes contain packetid of first subframe
in the AMPDU. We can not identify AMPDU subframes based on packedid, so
simply assume that status match first pending frame in the queue. Thats
mostly the same what 2800pci do.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Acked-by: Helmut Schaa <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Currently we read tx status register after each urb data transfer. As
callback procedure also trigger reading, that causing we have many
"threads" of reading status. To prevent that introduce TX_STATUS_READING
flags, and check if we are already in process of sequential reading
TX_STA_FIFO, before requesting new reads.
Change timer to hrtimer, that make TX_STA_FIFO overruns less possible.
Use 200 us for initial timeout, and then reschedule in 100 us period,
this values probably have to be tuned.
Make changes on txdone work. Schedule it from
rt2800usb_tx_sta_fifo_read_completed() callback when first valid status
show up. Check in callback if tx status timeout happens, and schedule
work on that condition too. That make possible to remove tx status
timeout from generic watchdog. I moved that to rt2800usb.
Loop in txdone work, that should prevent situation when we queue work,
which is already processed, and after finish work is not rescheduled
again.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Patch change txdone code to make it similar like txdone in rt2800pci,
process only one entry from queue matching tx status.
Before we processed all pending entries from queue until PACKEDID match,
that caused that we do not report tx statuses correctly.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Is possible that we stop queue and then do not wake up it again,
especially when packets are transmitted fast. That can be easily
reproduced with modified tx queue entry_num to some small value e.g. 16.
If mac80211 already hold local->queue_stop_reason_lock, then we can wait
on that lock in both rt2x00queue_pause_queue() and
rt2x00queue_unpause_queue(). After drooping ->queue_stop_reason_lock
is possible that __ieee80211_wake_queue() will be performed before
__ieee80211_stop_queue(), hence we stop queue and newer wake up it
again.
Another race condition is possible when between rt2x00queue_threshold()
check and rt2x00queue_pause_queue() we will process all pending tx
buffers on different cpu. This might happen if for example interrupt
will be triggered on cpu performing rt2x00mac_tx().
To prevent race conditions serialize pause/unpause by queue->tx_lock.
Cc: [email protected]
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
On il3945_down procedure we free tx queue data and nullify il->txq
pointer. After that we drop mutex and then cancel delayed works. There
is possibility, that after drooping mutex and before the cancel, some
delayed work will start and crash while trying to send commands to
the device. For example, here is reported crash in
il3945_bg_reg_txpower_periodic():
https://bugzilla.kernel.org/show_bug.cgi?id=42766#c10
Patch fix problem by adding il->txq check on works that send commands,
hence utilize tx queue.
Reported-by: Clemens Eisserer <[email protected]>
Cc: [email protected]
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
If device is disabled by rfkill switch, do not enable all interrupts,
but only CSR_INT_BIT_RF_KILL to receive rfkill state change. Unblocking
other interrupts might cause problems, since driver can not be prepared
for receive them.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Wey-Yi Guy <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
iwl_poll_direct_bit() return negative error value on timeout,
positive values do not indicate an error.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Wey-Yi Guy <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
That change will save us some CPU cycles at run time. Having port-based
I/O seems to be not possible for PCIe devices.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Wey-Yi Guy <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
wmb(), rmb() are not needed when writel(), readl() are used as
accessors for MMIO. We use them indirectly via iowrite32(),
ioread32().
What is needed mmiowb(), for synchronizing writes coming from
different CPUs on PCIe bridge (see in patch comments). This
fortunately is not needed on x86, where mmiowb() is just
defined as compiler barrier. As iwlwifi devices are most likely
not used on anything other than x86, this is not so important
fix.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Wey-Yi Guy <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Before we write to the device registers always check if
iwl_grap_nic_access() was successful.
On the way change return type of grab_nic_access() to bool, and add
likely()/unlikely() statement.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Wey-Yi Guy <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Print dump stack when the device is not responding. This should give
some more clue about the reason of failure. Also change the message we
print, since "MAC in deep sleep" is kinda confusing.
On the way add unlikely(), as fail to gain NIC access is hmm ...
unlikely.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Wey-Yi Guy <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
The following situation might occur:
__blkdev_get: add_disk:
register_disk()
get_gendisk()
disk_block_events()
disk->ev == NULL
disk_add_events()
__disk_unblock_events()
disk->ev != NULL
--ev->block
Then we unblock events, when they are suppose to be blocked. This can
trigger events related block/genhd.c warnings, but also can crash in
sd_check_events() or other places.
I'm able to reproduce crashes with the following scripts (with
connected usb dongle as sdb disk).
<snip>
DEV=/dev/sdb
ENABLE=/sys/bus/usb/devices/1-2/bConfigurationValue
function stop_me()
{
for i in `jobs -p` ; do kill $i 2> /dev/null ; done
exit
}
trap stop_me SIGHUP SIGINT SIGTERM
for ((i = 0; i < 10; i++)) ; do
while true; do fdisk -l $DEV 2>&1 > /dev/null ; done &
done
while true ; do
echo 1 > $ENABLE
sleep 1
echo 0 > $ENABLE
done
</snip>
I use the script to verify patch fixing oops in sd_revalidate_disk
http://marc.info/?l=linux-scsi&m=132935572512352&w=2
Without Jun'ichi Nomura patch titled "Fix NULL pointer dereference in
sd_revalidate_disk" or this one, script easily crash kernel within
a few seconds. With both patches applied I do not observe crash.
Unfortunately after some time (dozen of minutes), script will hung in:
[ 1563.906432] [<c08354f5>] schedule_timeout_uninterruptible+0x15/0x20
[ 1563.906437] [<c04532d5>] msleep+0x15/0x20
[ 1563.906443] [<c05d60b2>] blk_drain_queue+0x32/0xd0
[ 1563.906447] [<c05d6e00>] blk_cleanup_queue+0xd0/0x170
[ 1563.906454] [<c06d278f>] scsi_free_queue+0x3f/0x60
[ 1563.906459] [<c06d7e6e>] __scsi_remove_device+0x6e/0xb0
[ 1563.906463] [<c06d4aff>] scsi_forget_host+0x4f/0x60
[ 1563.906468] [<c06cd84a>] scsi_remove_host+0x5a/0xf0
[ 1563.906482] [<f7f030fb>] quiesce_and_remove_host+0x5b/0xa0 [usb_storage]
[ 1563.906490] [<f7f03203>] usb_stor_disconnect+0x13/0x20 [usb_storage]
Anyway I think this patch is some step forward.
As drawback, I do not teardown on sysfs file create error, because I do
not know how to nullify disk->ev (since it can be used). However add_disk
error handling practically does not exist too, and things will work
without this sysfs file, except events will not be exported to user
space.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Tejun Heo <[email protected]>
Cc: [email protected]
Signed-off-by: Jens Axboe <[email protected]>
If rfkill is on il_grab_nic_access() fail and we can not write to the
various registers during stop procedure. Write to those registers
unconditionally instead.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
The same data can be gathered using monitor mode.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Remove unneeded structure and cleanup temperature calibration routines
a bit.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
That change will save us some CPU cycles at run time. Having
port-based I/O seems to be not possible for PCIe devices.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
wmb(), rmb() are not needed when writel(), readl() are used as
accessors for MMIO. We use them indirectly via iowrite32(),
ioread32().
What is needed mmiowb(), for synchronizing writes coming from
different CPUs on PCIe bridge (see in patch comments). This
fortunately is not needed on x86, where mmiowb() is just
defined as compiler barrier. As iwlegacy devices are most likely
not used on anything other than x86, this is not so important
fix.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Before we write to the device register always check if
_il_grap_nic_access() was successful.
Change type return type _il_grap_nic_access() to bool, and
add likely()/unlikely() statements.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Print dump stack when the device is not responding. This should give
some more clue about the reason of failure. Also change the message we
print, since "MAC in deep sleep" is kinda confusing.
On the way add unlikely(), as fail to gain NIC access is hmm ...
unlikely.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
We create "bsg" link if q->kobj.sd is not NULL, so remove it only
when the same condition is true.
Fixes:
WARNING: at fs/sysfs/inode.c:323 sysfs_hash_and_remove+0x2b/0x77()
sysfs: can not remove 'bsg', no directory
Call Trace:
[<c0429683>] warn_slowpath_common+0x6a/0x7f
[<c0537a68>] ? sysfs_hash_and_remove+0x2b/0x77
[<c042970b>] warn_slowpath_fmt+0x2b/0x2f
[<c0537a68>] sysfs_hash_and_remove+0x2b/0x77
[<c053969a>] sysfs_remove_link+0x20/0x23
[<c05d88f1>] bsg_unregister_queue+0x40/0x6d
[<c0692263>] __scsi_remove_device+0x31/0x9d
[<c069149f>] scsi_forget_host+0x41/0x52
[<c0689fa9>] scsi_remove_host+0x71/0xe0
[<f7de5945>] quiesce_and_remove_host+0x51/0x83 [usb_storage]
[<f7de5a1e>] usb_stor_disconnect+0x18/0x22 [usb_storage]
[<c06c29de>] usb_unbind_interface+0x4e/0x109
[<c067a80f>] __device_release_driver+0x6b/0xa6
[<c067a861>] device_release_driver+0x17/0x22
[<c067a46a>] bus_remove_device+0xd6/0xe6
[<c06785e2>] device_del+0xf2/0x137
[<c06c101f>] usb_disable_device+0x94/0x1a0
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
We do not need that callback, settings parameters can be done locally.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Taken from:
2011_0719_RT3070_RT3370_RT5370_RT5372_Linux_STA_V2.5.0.3_DPO
(based on function RT33xx_ChipSwitchChannel)
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Synchronize code with Ralink driver:
2011_0719_RT3070_RT3370_RT5370_RT5372_Linux_STA_V2.5.0.3_DPO
Based on functions:
RT30xx_ChipSwitchChannel
RT33xx_ChipSwitchChannel
NICInitRT3370RFRegisters
and defines from:
include/chip/rt33xx.h
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Synchronize code with Ralink driver:
2011_0719_RT3070_RT3370_RT5370_RT5372_Linux_STA_V2.5.0.3_DPO
Based on functions:
RT33xx_ChipSwitchChannel
RT30xx_ChipSwitchChannel
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Synchronize code with Ralink driver:
2011_0719_RT3070_RT3370_RT5370_RT5372_Linux_STA_V2.5.0.3_DPO
Based on functions:
RT33xx_ChipSwitchChannel
RT30xx_ChipSwitchChannel
RT33xx_Init
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Zero all other bits than RESET_CSR and RESET_BBP when want to do the
reset, that the vendor driver behaviour.
Acked-by: Helmut Schaa <[email protected]>
Acked-by: Ivo van Doorn <[email protected]>
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
We can receive frames just after firmware load with current code, so
disable DMA just after firmware is loaded, not before.
Acked-by: Ivo van Doorn <[email protected]>
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Ralink USB driver initialize H2M_INT_SRC to 0 after firmware load, and
never touch this register later. It is not touched at all by Ralink PCI
driver.
Acked-by: Helmut Schaa <[email protected]>
Acked-by: Ivo van Doorn <[email protected]>
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This patch workaround live deadlock problem caused by infinite loop
in brcms_c_wait_for_tx_completion(). I do not consider the patch as
the proper fix, which should fix the real reason of tx queue flush
failure, but patch helps with system lockup.
Reference:
https://bugzilla.kernel.org/show_bug.cgi?id=42576
Reported-and-tested-by: Patrick <[email protected]>
Cc: [email protected] # 3.2+
Signed-off-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Same devices can generate interrupt without properly setting bit in
INT_SOURCE_CSR register (spurious interrupt), what will cause IRQ line
will be disabled by interrupts controller driver.
We discovered that clearing INT_MASK_CSR stops such behaviour. We
previously first read that register, and then clear all know interrupt
sources bits and do not touch reserved bits. After this patch, we write
to all register content (I believe writing to reserved bits on that
register will not cause any problems, I tested that on my rt2800pci
device).
This fix very bad performance problem, practically making device
unusable (since worked without interrupts), reported in:
https://bugzilla.redhat.com/show_bug.cgi?id=658451
We previously tried to workaround that issue in commit
4ba7d99978 "rt2800pci: handle spurious
interrupts", but it was reverted in commit
82e5fc2a34
as thing, that will prevent to detect real spurious interrupts.
Reported-and-tested-by: Amir Hedayaty <[email protected]>
Cc: [email protected]
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Gertjan van Wingerde <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When debugging with CONFIG_DEBUG_PAGEALLOC and debug_guardpage_minorder >
0, we have lot of free pages that are not marked so. Snapshot code
account them as savable, what cause hibernate memory preallocation
failure.
It is pretty hard to make hibernate allocation succeed with
debug_guardpage_minorder=1. This change at least make it possible when
system has relatively big amount of RAM.
Signed-off-by: Stanislaw Gruszka <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: Christoph Lameter <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Pekka Enberg <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
With CONFIG_DEBUG_PAGEALLOC configured, the CPU will generate an exception
on access (read,write) to an unallocated page, which permits us to catch
code which corrupts memory. However the kernel is trying to maximise
memory usage, hence there are usually few free pages in the system and
buggy code usually corrupts some crucial data.
This patch changes the buddy allocator to keep more free/protected pages
and to interlace free/protected and allocated pages to increase the
probability of catching corruption.
When the kernel is compiled with CONFIG_DEBUG_PAGEALLOC,
debug_guardpage_minorder defines the minimum order used by the page
allocator to grant a request. The requested size will be returned with
the remaining pages used as guard pages.
The default value of debug_guardpage_minorder is zero: no change from
current behaviour.
[[email protected]: tweak documentation, s/flg/flag/]
Signed-off-by: Stanislaw Gruszka <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: "Rafael J. Wysocki" <[email protected]>
Cc: Christoph Lameter <[email protected]>
Cc: Pekka Enberg <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Stanislaw Gruszka