Implement sata_std_hardreset(), which simply wraps around
sata_link_hardreset().  sata_std_hardreset() becomes new standard
hardreset method for sata_port_ops and sata_sff_hardreset() moves from
ata_base_port_ops to ata_sff_port_ops, which is where it really
belongs.

ata_is_builtin_hardreset() is added so that both
ata_std_error_handler() and ata_sff_error_handler() skip both builtin
hardresets if SCR isn't accessible.

piix_sidpr_hardreset() in ata_piix.c is identical to
sata_std_hardreset() in functionality and got replaced with the
standard function.
Signed-off-by: NTejun Heo <htejun@gmail.com>

sata_sff_hardreset() contains link readiness wait logic which isn't
SFF specific.  Move that part into sata_link_hardreset(), which now
takes two more parameters - @online and @check_ready.  Both are
optional.  The former is out parameter for link onlineness after
reset.  The latter is used to wait for link readiness after hardreset.

Users of sata_link_hardreset() is updated to use new funtionality and
ahci_hardreset() is updated to use sata_link_hardreset() instead of
sata_sff_hardreset().  This doesn't really cause any behavior change.
Signed-off-by: NTejun Heo <htejun@gmail.com>

Factor out waiting logic (which is common to all ATA controllers) from
ata_sff_wait_ready() into ata_wait_ready().  ata_wait_ready() takes
@check_ready function pointer and uses it to poll for readiness.  This
allows non-SFF controllers to use ata_wait_ready() to wait for link
readiness.

This patch also implements ata_wait_after_reset() - generic version of
ata_sff_wait_after_reset() - using ata_wait_ready().

ata_sff_wait_ready() is reimplemented using ata_wait_ready() and
ata_sff_check_ready().  Functionality remains the same.
Signed-off-by: NTejun Heo <htejun@gmail.com>

Previously, post-softreset readiness is waited as follows.

1. ata_sff_wait_after_reset() waits for 150ms and then for
   ATA_TMOUT_FF_WAIT if status is 0xff and other conditions meet.

2. ata_bus_softreset() finishes with -ENODEV if status is still 0xff.
   If not, continue to #3.

3. ata_bus_post_reset() waits readiness of dev0 and/or dev1 depending
   on devmask using ata_sff_wait_ready().

And for post-hardreset readiness,

1. ata_sff_wait_after_reset() waits for 150ms and then for
   ATA_TMOUT_FF_WAIT if status is 0xff and other conditions meet.

2. sata_sff_hardreset waits for device readiness using
   ata_sff_wait_ready().

This patch merges and unifies post-reset readiness waits into
ata_sff_wait_ready() and ata_sff_wait_after_reset().

ATA_TMOUT_FF_WAIT handling is merged into ata_sff_wait_ready().  If TF
status is 0xff, link status is unknown and the port is SATA, it will
continue polling till ATA_TMOUT_FF_WAIT.

ata_sff_wait_after_reset() is updated to perform the following steps.

1. waits for 150ms.

2. waits for dev0 readiness using ata_sff_wait_ready().  Note that
   this is done regardless of devmask, as ata_sff_wait_ready() handles
   0xff status correctly, this preserves the original behavior except
   that it may wait longer after softreset if link is online but
   status is 0xff.  This behavior change is very unlikely to cause any
   actual difference and is intended.  It brings softreset behavior to
   that of hardreset.

3. waits for dev1 readiness just the same way ata_bus_post_reset() did.

Now both soft and hard resets call ata_sff_wait_after_reset() after
reset to wait for readiness after resets.  As
ata_sff_wait_after_reset() contains calls to ->sff_dev_select(),
explicit call near the end of sata_sff_hardreset() is removed.

This change makes reset implementation simpler and more consistent.

While at it, make the magical 150ms wait post-reset wait duration a
constant and ata_sff_wait_ready() and ata_sff_wait_after_reset() take
@link instead of @ap.  This is to make them consistent with other
reset helpers and ease core changes.

pata_scc is updated accordingly.
Signed-off-by: NTejun Heo <htejun@gmail.com>

Separate out generic ATA portion from ata_sff_postreset() into
ata_std_postreset() and implement ata_sff_postreset() using the std
version.

ata_base_port_ops now has ata_std_postreset() for its postreset and
ata_sff_port_ops overrides it to ata_sff_postreset().

This change affects pdc_adma, ahci, sata_fsl and sata_sil24.  pdc_adma
now specifies postreset to ata_sff_postreset() explicitly.  sata_fsl
and sata_sil24 now use ata_std_postreset() which makes no difference
to them.  ahci now calls ata_std_postreset() from its own postreset
method, which causes no behavior difference.
Signed-off-by: NTejun Heo <htejun@gmail.com>

Separate out generic ATA portion from ata_sff_prereset() into
ata_std_prereset() and implement ata_sff_prereset() using the std
version.  Waiting for device readiness is the only SFF specific part.

ata_base_port_ops now has ata_std_prereset() for its prereset and
ata_sff_port_ops overrides it to ata_sff_prereset().  This change can
affect pdc_adma, ahci, sata_fsl and sata_sil24.  pdc_adma implements
its own prereset using ata_sff_prereset() and the rest has hardreset
and thus are unaffected by this change.

This change reflects real world situation.  There is no generic way to
wait for device readiness for non-SFF controllers and some of them
don't have any mechanism for that.  Non-sff drivers which don't have
hardreset should wrap ata_std_prereset() and wait for device readiness
itself but there's no such driver now and isn't likely to be popular
in the future either.
Signed-off-by: NTejun Heo <htejun@gmail.com>

->sff_irq_clear() is called only from SFF interrupt handler, so there
is no reason to initialize it for non-SFF controllers.  Also,
ata_sff_irq_clear() can handle both BMDMA and non-BMDMA SFF
controllers.

This patch kills ata_noop_irq_clear() and removes it from base
port_ops and sets ->sff_irq_clear to ata_sff_irq_clear() in sff
port_ops instead of bmdma port_ops.
Signed-off-by: NTejun Heo <htejun@gmail.com>

Add sff_ prefix to SFF specific port ops.

This rename is in preparation of separating SFF support out of libata
core layer.  This patch strictly renames ops and doesn't introduce any
behavior difference.
Signed-off-by: NTejun Heo <htejun@gmail.com>

SFF functions have confusing names.  Some have sff prefix, some have
bmdma, some std, some pci and some none.  Unify the naming by...

* SFF functions which are common to both BMDMA and non-BMDMA are
  prefixed with ata_sff_.

* SFF functions which are specific to BMDMA are prefixed with
  ata_bmdma_.

* SFF functions which are specific to PCI but apply to both BMDMA and
  non-BMDMA are prefixed with ata_pci_sff_.

* SFF functions which are specific to PCI and BMDMA are prefixed with
  ata_pci_bmdma_.

* Drop generic prefixes from LLD specific routines.  For example,
  bfin_std_dev_select -> bfin_dev_select.

The following renames are noteworthy.

  ata_qc_issue_prot() -> ata_sff_qc_issue()
  ata_pci_default_filter() -> ata_bmdma_mode_filter()
  ata_dev_try_classify() -> ata_sff_dev_classify()

This rename is in preparation of separating SFF support out of libata
core layer.  This patch strictly renames functions and doesn't
introduce any behavior difference.
Signed-off-by: NTejun Heo <htejun@gmail.com>

It should be ATA_TAG_INTERNAL.
Signed-off-by: NYoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Currently whether a command should be retried after failure is
determined inside ata_eh_finish().  Add ATA_QCFLAG_RETRY and move the
logic into ata_eh_autopsy().  This makes things clearer and helps
extending retry determination logic.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

ata_chk_status() just calls ops->check_status and it only adds
confusion with other status functions.  Kill it.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

ata_pci_default_filter() doesn't really have anything to do with PCI.
It's generally applicable to BMDMA controllers.  Move it out of
CONFIG_PCI.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

* Move SFF related functions from libata-core.c to libata-sff.c.

  ata_[bmdma_]sff_port_ops, ata_devchk(), ata_dev_try_classify(),
  ata_std_dev_select(), ata_tf_to_host(), ata_busy_sleep(),
  ata_wait_after_reset(), ata_wait_ready(), ata_bus_post_reset(),
  ata_bus_softreset(), ata_bus_reset(), ata_std_softreset(),
  sata_std_hardreset(), ata_fill_sg(), ata_fill_sg_dumb(),
  ata_qc_prep(), ata_dump_qc_prep(), ata_data_xfer(),
  ata_data_xfer_noirq(), ata_pio_sector(), ata_pio_sectors(),
  atapi_send_cdb(), __atapi_pio_bytes(), atapi_pio_bytes(),
  ata_hsm_ok_in_wq(), ata_hsm_qc_complete(), ata_hsm_move(),
  ata_pio_task(), ata_qc_issue_prot(), ata_host_intr(),
  ata_interrupt(), ata_std_ports()

* Make ata_pio_queue_task() global as it's now called from
  libata-sff.c.

* Move SFF related stuff in include/linux/libata.h and
  drivers/ata/libata.h into one place.  While at it, move timing
  constants into the global enum definition and fortify comments a
  bit.

This patch strictly moves stuff around and as such doesn't cause any
functional difference.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Currently reset methods are not specified directly in the
ata_port_operations table.  If a LLD wants to use custom reset
methods, it should construct and use a error_handler which uses those
reset methods.  It's done this way for two reasons.

First, the ops table already contained too many methods and adding
four more of them would noticeably increase the amount of necessary
boilerplate code all over low level drivers.

Second, as ->error_handler uses those reset methods, it can get
confusing.  ie. By overriding ->error_handler, those reset ops can be
made useless making layering a bit hazy.

Now that ops table uses inheritance, the first problem doesn't exist
anymore.  The second isn't completely solved but is relieved by
providing default values - most drivers can just override what it has
implemented and don't have to concern itself about higher level
callbacks.  In fact, there currently is no driver which actually
modifies error handling behavior.  Drivers which override
->error_handler just wraps the standard error handler only to prepare
the controller for EH.  I don't think making ops layering strict has
any noticeable benefit.

This patch makes ->prereset, ->softreset, ->hardreset, ->postreset and
their PMP counterparts propoer ops.  Default ops are provided in the
base ops tables and drivers are converted to override individual reset
methods instead of creating custom error_handler.

* ata_std_error_handler() doesn't use sata_std_hardreset() if SCRs
  aren't accessible.  sata_promise doesn't need to use separate
  error_handlers for PATA and SATA anymore.

* softreset is broken for sata_inic162x and sata_sx4.  As libata now
  always prefers hardreset, this doesn't really matter but the ops are
  forced to NULL using ATA_OP_NULL for documentation purpose.

* pata_hpt374 needs to use different prereset for the first and second
  PCI functions.  This used to be done by branching from
  hpt374_error_handler().  The proper way to do this is to use
  separate ops and port_info tables for each function.  Converted.
Signed-off-by: NTejun Heo <htejun@gmail.com>

libata core layer doesn't care about sht or ->irq_handler.  Those are
only of interest to the LLD during initialization.  This is confusing
and has caused several drivers to have duplicate unused initializers
for these fields.

Currently only sata_nv uses these fields.  Make sata_nv use
->private_data, which is supposed to carry LLD-specific information,
instead and kill ->sht and ->irq_handler.  nv_pi_priv structure is
defined and struct literals are used to initialize private_data.
Notational overhead is negligible.
Signed-off-by: NTejun Heo <htejun@gmail.com>

port_info->private_data is currently used for two purposes - to record
private data about the port_info or to specify host->private_data to
use when allocating ata_host.

This overloading is confusing and counter-intuitive in that
port_info->private_data becomes host->private_data instead of
port->private_data.  In addition, port_info and host don't correspond
to each other 1-to-1.  Currently, the first non-NULL
port_info->private_data is used.

This patch makes port_info->private_data just be what it is -
private_data for the port_info where LLD can jot down extra info.
libata no longer sets host->private_data to the first non-NULL
port_info->private_data, @host_priv argument is added to
ata_pci_init_one() instead.  LLDs which use ata_pci_init_one() can use
this argument to pass in pointer to host private data.  LLDs which
don't should use init-register model anyway and can initialize
host->private_data directly.

Adding @host_priv instead of using init-register model for LLDs which
use ata_pci_init_one() is suggested by Alan Cox.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>

ata_pci_init_one() is the only function which uses ops->irq_handler
and pi->sht.  Other initialization functions take the same information
as arguments.  This causes confusion and duplicate unused entries in
structures.

Make ata_pci_init_one() take sht as an argument and use ata_interrupt
implicitly.  All current users use ata_interrupt and if different irq
handler is necessary open coding ata_pci_init_one() using
ata_prepare_sff_host() and ata_activate_sff_host can be done under ten
lines including error handling and driver which requires custom
interrupt handler is likely to require custom initialization anyway.

As ata_pci_init_one() was the last user of ops->irq_handler, this
patch also kills the field.
Signed-off-by: NTejun Heo <htejun@gmail.com>

libata lets low level drivers build ata_port_operations table and
register it with libata core layer.  This allows low level drivers
high level of flexibility but also burdens them with lots of
boilerplate entries.

This becomes worse for drivers which support related similar
controllers which differ slightly.  They share most of the operations
except for a few.  However, the driver still needs to list all
operations for each variant.  This results in large number of
duplicate entries, which is not only inefficient but also error-prone
as it becomes very difficult to tell what the actual differences are.

This duplicate boilerplates all over the low level drivers also make
updating the core layer exteremely difficult and error-prone.  When
compounded with multi-branched development model, it ends up
accumulating inconsistencies over time.  Some of those inconsistencies
cause immediate problems and fixed.  Others just remain there dormant
making maintenance increasingly difficult.

To rectify the problem, this patch implements ata_port_operations
inheritance.  To allow LLDs to easily re-use their own ops tables
overriding only specific methods, this patch implements poor man's
class inheritance.  An ops table has ->inherits field which can be set
to any ops table as long as it doesn't create a loop.  When the host
is started, the inheritance chain is followed and any operation which
isn't specified is taken from the nearest ancestor which has it
specified.  This operation is called finalization and done only once
per an ops table and the LLD doesn't have to do anything special about
it other than making the ops table non-const such that libata can
update it.

libata provides four base ops tables lower drivers can inherit from -
base, sata, pmp, sff and bmdma.  To avoid overriding these ops
accidentaly, these ops are declared const and LLDs should always
inherit these instead of using them directly.

After finalization, all the ops table are identical before and after
the patch except for setting .irq_handler to ata_interrupt in drivers
which didn't use to.  The .irq_handler doesn't have any actual effect
and the field will soon be removed by later patch.

* sata_sx4 is still using old style EH and currently doesn't take
  advantage of ops inheritance.
Signed-off-by: NTejun Heo <htejun@gmail.com>

libata lets low level drivers build scsi_host_template and register it
to the SCSI layer.  This allows low level drivers high level of
flexibility but also burdens them with lots of boilerplate entries.

This patch implements SHT initializers which can be used to initialize
all the boilerplate entries in a sht.  Three variants of them are
implemented - BASE, BMDMA and NCQ - for different types of drivers.
Note that entries can be overriden by putting individual initializers
after the helper macro.

All sht tables are identical before and after this patch.
Signed-off-by: NTejun Heo <htejun@gmail.com>

->irq_clear() is used to clear IRQ bit of a SFF controller and isn't
useful for drivers which don't use libata SFF HSM implementation.
However, it's a required callback and many drivers implement their own
noop version as placeholder.  This patch implements ata_noop_irq_clear
and use it to replace those custom placeholders.

Also, SFF drivers which don't support BMDMA don't need to use
ata_bmdma_irq_clear().  It becomes noop if BMDMA address isn't
initialized.  Convert them to use ata_noop_irq_clear().
Signed-off-by: NTejun Heo <htejun@gmail.com>

Over the time, ops in ata_port_operations has become a bit confusing.
Reorganize.  SFF/BMDMA ops are separated into separate a group as they
will be taken out of ata_port_operations later.
Signed-off-by: NTejun Heo <htejun@gmail.com>

ata_ehi_schedule_probe() was created to hide details of link-resuming
reset magic.  Now that all the softreset workarounds are gone,
scheduling probe is very simple - set probe_mask and request RESET.
Kill ata_ehi_schedule_probe() and open code it.  This also increases
consistency as ata_ehi_schedule_probe() couldn't cover individual
device probings so they were open-coded even when the helper existed.

While at it, define ATA_ALL_DEVICES as mask of all possible devices on
a link and always use it when requesting probe on link level for
simplicity and consistency.  Setting extra bits in the probe_mask
doesn't hurt anybody.
Signed-off-by: NTejun Heo <htejun@gmail.com>

Some controllers can't reliably record the initial D2H FIS after SATA
link is brought online for whatever reason.  Advanced controllers
which don't have traditional TF register based interface often have
this problem as they don't really have the TF registers to update
while the controller and link are being initialized.

SKIP_D2H_BSY works around the problem by skipping the wait for device
readiness before issuing SRST, so for such controllers libata issues
SRST blindly and hopes for the best.

Now that libata defaults to hardreset, this workaround is no longer
necessary.  For controllers which have support for hardreset, SRST is
never issued by itself.  It is only issued as follow-up SRST for
device classification and PMP initialization, so there's no need to
wait for it from prereset.

Kill ATA_LFLAG_SKIP_D2H_BSY.
Signed-off-by: NTejun Heo <htejun@gmail.com>

ATA_EHI_RESUME_LINK has two functions - promote reset to hardreset if
ATA_LFLAG_HRST_TO_RESUME is set and preventing EH from shortcutting
reset action when probing is requested.  The former is gone now and
the latter can easily be achieved by making EH to perform at least one
reset if reset is requested, which also makes more sense than
depending on RESUME_LINK flag.

As ATA_EHI_RESUME_LINK was the only EHI reset modifier, this also
kills reset modifier handling.
Signed-off-by: NTejun Heo <htejun@gmail.com>

Now that hardreset is the preferred method of resetting, there's no
need for ATA_LFLAG_HRST_TO_RESUME flag.  Kill it.
Signed-off-by: NTejun Heo <htejun@gmail.com>

When both soft and hard resets are available, libata preferred
softreset till now.  The logic behind it was to be softer to devices;
however, this doesn't really help much.  Rationales for the change:

* BIOS may freeze lock certain things during boot and softreset can't
  unlock those.  This by itself is okay but during operation PHY event
  or other error conditions can trigger hardreset and the device may
  end up with different configuration.

  For example, after a hardreset, previously unlockable HPA can be
  unlocked resulting in different device size and thus revalidation
  failure.  Similar condition can occur during or after resume.

* Certain ATAPI devices require hardreset to recover after certain
  error conditions.  On PATA, this is done by issuing the DEVICE RESET
  command.  On SATA, COMRESET has equivalent effect.  The problem is
  that DEVICE RESET needs its own execution protocol.

  For SFF controllers with bare TF access, it can be easily
  implemented but more advanced controllers (e.g. ahci and sata_sil24)
  require specialized implementations.  Simply using hardreset solves
  the problem nicely.

* COMRESET initialization sequence is the norm in SATA land and many
  SATA devices don't work properly if only SRST is used.  For example,
  some PMPs behave this way and libata works around by always issuing
  hardreset if the host supports PMP.

  Like the above example, libata has developed a number of mechanisms
  aiming to promote softreset to hardreset if softreset is not going
  to work.  This approach is time consuming and error prone.

  Also, note that, dependingon how you read the specs, it could be
  argued that PMP fan-out ports require COMRESET to start operation.
  In fact, all the PMPs on the market except one don't work properly
  if COMRESET is not issued to fan-out ports after PMP reset.

* COMRESET is an integral part of SATA connection and any working
  device should be able to handle COMRESET properly.  After all, it's
  the way to signal hardreset during reboot.  This is the most used
  and recommended (at least by the ahci spec) method of resetting
  devices.

So, this patch makes libata prefer hardreset over softreset by making
the following changes.

* Rename ATA_EH_RESET_MASK to ATA_EH_RESET and use it whereever
  ATA_EH_{SOFT|HARD}RESET used to be used.  ATA_EH_{SOFT|HARD}RESET is
  now only used to tell prereset whether soft or hard reset will be
  issued.

* Strip out now unneeded promote-to-hardreset logics from
  ata_eh_reset(), ata_std_prereset(), sata_pmp_std_prereset() and
  other places.
Signed-off-by: NTejun Heo <htejun@gmail.com>

The THERMAL_MAX_TRIPS value is set to 10.  It is too few for the Compaq AP550
machine which has 12 trip points.
Signed-off-by: NKrzysztof Helt <krzysztof.h1@wp.pl>
Cc: Len Brown <lenb@kernel.org>
Cc: Zhang Rui <rui.zhang@intel.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

mb_cache_entry_alloc() was allocating cache entries with GFP_KERNEL.  But
filesystems are calling this function while holding xattr_sem so possible
recursion into the fs violates locking ordering of xattr_sem and transaction
start / i_mutex for ext2-4.  Change mb_cache_entry_alloc() so that filesystems
can specify desired gfp mask and use GFP_NOFS from all of them.
Signed-off-by: NJan Kara <jack@suse.cz>
Reported-by: NDave Jones <davej@redhat.com>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This fixes DMA on architectures where DMA is nontrivial, like PPC64.
We must use the host-device's (PCI) struct device for any DMA
operation instead of the SSB device. For this we add a new
struct device pointer to the SSB device structure that will always
point to the right device for DMAing.

Without this patch b43 and b44 drivers won't work on complex-DMA
architectures, that for example need dev->archdata for DMA operations.
Signed-off-by: NMichael Buesch <mb@bu3sch.de>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Thanks to Nikanth Karthikesan <knikanth@suse.de> for reporting this.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

struct ipv6_opt_hdr is the common structure for IPv6 extension
headers, and it is common to increment the pointer to get
the real content.  On the other hand, since the structure
consists only of 1-byte next-header field and 1-byte length
field, size of that structure depends on architecture; 2 or 4.
Add "packed" attribute to get 2.
Signed-off-by: NYOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There's no reason for this to be in the header, and it just hurts
recompile time.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Acked-by: NMax Krasnyanskiy <maxk@qualcomm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A little more detail here wouldn't hurt.
Signed-off-by: NJ. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: NJonathan Corbet <corbet@lwn.net>

git commit 54a01510 ("asmlinkage_protect
replaces prevent_tail_call") causes this build failure on s390:

    AS      arch/s390/kernel/entry64.o
  In file included from arch/s390/kernel/entry64.S:14:
  include/linux/linkage.h:34: error: syntax error in macro parameter list
  make[1]: *** [arch/s390/kernel/entry64.o] Error 1
  make: *** [arch/s390/kernel] Error 2

and some other architectures.  The reason is that some architectures add
the "-traditional" flag to the invocation of $(AS), which disables
variadic macro argument support.

So just surround the new define with an #ifndef __ASSEMBLY__ to prevent
any side effects on asm code.

Cc: Roland McGrath <roland@redhat.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: NHeiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Increase the PNP "number of devices" limit.  We currently use an unsigned
char, which limits us to 256 devices per protocol.  This patch changes that to
an unsigned int.

Not all backends can take advantage of this: we limit ISAPNP to 10 devices in
isapnp_cfg_begin(), and PNPBIOS is limited to 256 devices because the BIOS
interfaces use a one-byte device node number.

But there is no limit on the number of PNPACPI devices we may have.  Large HP
Integrity machines have more than 256, which causes the current "unsigned char
number" to wrap around.  This causes errors like this:

    pnp: PnP ACPI init
    kobject_add failed for 00:00 with -EEXIST, don't try to register things with the same name in the same directory.

    Call Trace:
     [<a000000100010720>] show_stack+0x40/0xa0
     [<a0000001000107b0>] dump_stack+0x30/0x60
     [<a0000001001dbdf0>] kobject_add+0x290/0x2c0
     [<a0000001002bfd40>] device_add+0x160/0x860
     [<a0000001002c0470>] device_register+0x30/0x60
     [<a00000010026ba70>] __pnp_add_device+0x130/0x180
     [<a00000010026bb70>] pnp_add_device+0xb0/0xe0
     [<a0000001007f2730>] pnpacpi_add_device+0x510/0x5a0
     [<a0000001007f2810>] pnpacpi_add_device_handler+0x50/0x80

This patch increases the limit to fix this PNPACPI problem.  It should not
have any adverse effect on ISAPNP or PNPBIOS because their limits are still
enforced in the backends.
Signed-off-by: NBjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It's really a pretty ugly thing to need, and some day it will hopefully
be obviated by teaching gcc about the magic calling conventions for the
low-level system call code, but in the meantime we can at least add big
honking comments about why we need these insane and strange macros.

I took my comments from my version of the macro, but I ended up deciding
to just pick Roland's version of the actual code instead (with his
prettier syntax that uses vararg macros).  Thus the previous two commits
that actually implement it.
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The prevent_tail_call() macro works around the problem of the compiler
clobbering argument words on the stack, which for asmlinkage functions
is the caller's (user's) struct pt_regs.  The tail/sibling-call
optimization is not the only way that the compiler can decide to use
stack argument words as scratch space, which we have to prevent.
Other optimizations can do it too.

Until we have new compiler support to make "asmlinkage" binding on the
compiler's own use of the stack argument frame, we have work around all
the manifestations of this issue that crop up.

More cases seem to be prevented by also keeping the incoming argument
variables live at the end of the function.  This makes their original
stack slots attractive places to leave those variables, so the compiler
tends not clobber them for something else.  It's still no guarantee, but
it handles some observed cases that prevent_tail_call() did not.
Signed-off-by: NRoland McGrath <roland@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The 'disable_cb' callback is designed as an optimization to tell the host
we don't need callbacks now.  As it is not reliable, the debug check is
overzealous: it can happen on two CPUs at the same time.  Document this.

Even if it were reliable, the virtio_net driver doesn't disable
callbacks on transmit so the START_USE/END_USE debugging reentrance
protection can be easily tripped even on UP.

Thanks to Balaji Rao for the bug report and testing.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
CC: Balaji Rao <balajirrao@gmail.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The effects of cgroup_disable=foo are:

- foo isn't auto-mounted if you mount all cgroups in a single hierarchy
- foo isn't visible as an individually mountable subsystem

As a result there will only ever be one call to foo->create(), at init time;
all processes will stay in this group, and the group will never be mounted on
a visible hierarchy.  Any additional effects (e.g.  not allocating metadata)
are up to the foo subsystem.

This doesn't handle early_init subsystems (their "disabled" bit isn't set be,
but it could easily be extended to do so if any of the early_init systems
wanted it - I think it would just involve some nastier parameter processing
since it would occur before the command-line argument parser had been run.

Hugh said:

  Ballpark figures, I'm trying to get this question out rather than
  processing the exact numbers: CONFIG_CGROUP_MEM_RES_CTLR adds 15% overhead
  to the affected paths, booting with cgroup_disable=memory cuts that back to
  1% overhead (due to slightly bigger struct page).

  I'm no expert on distros, they may have no interest whatever in
  CONFIG_CGROUP_MEM_RES_CTLR=y; and the rest of us can easily build with or
  without it, or apply the cgroup_disable=memory patches.

Unix bench's execl test result on x86_64 was

== just after boot without mounting any cgroup fs.==
mem_cgorup=off : Execl Throughput       43.0     3150.1      732.6
mem_cgroup=on  : Execl Throughput       43.0     2932.6      682.0
==

[lizf@cn.fujitsu.com: fix boot option parsing]
Signed-off-by: NBalbir Singh <balbir@linux.vnet.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Sudhir Kumar <skumar@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>