Signed-off-by: Steffen Maier <maier@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Message-Id: <1562149189-1417-4-git-send-email-maier@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

For non-static-inlines, debug.c already had non-compliant function
header docs. So move the pure prototype kdocs of
("s390: include/asm/debug.h add kerneldoc markups")
from debug.h to debug.c and merge them with the old function docs.
Also, I had the impression that kdoc typically is at the implementation
in the compile unit rather than at the prototype in the header file.

While at it, update the short kdoc description to distinguish the
different functions. And a few more consistency cleanups.

Added a new kdoc for debug_set_critical() since debug.h comments it
as part of the API.

Signed-off-by: Steffen Maier <maier@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Message-Id: <1562149189-1417-3-git-send-email-maier@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Complements previous ("s390: include/asm/debug.h add kerneldoc markups")
which seemed to have dropped important non-kdoc parts such as
user space interface (level, size, flush)
as well as views and caution regarding strings in the sprintf view.

Signed-off-by: Steffen Maier <maier@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Message-Id: <1562149189-1417-2-git-send-email-maier@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Merge tag 'vfio-ccw-20190705' of https://git.kernel.org/pub/scm/linux/kernel/git/kvms390/vfio-ccw into features

Fix a bug introduced in the refactoring.

* tag 'vfio-ccw-20190705' of https://git.kernel.org/pub/scm/linux/kernel/git/kvms390/vfio-ccw

:
  vfio-ccw: Fix the conversion of Format-0 CCWs to Format-1

Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

When processing Format-0 CCWs, we use the "len" variable as the
number of CCWs to convert to Format-1.  But that variable
contains zero here, and is not a meaningful CCW count until
ccwchain_calc_length() returns.  Since that routine requires and
expects Format-1 CCWs to identify the chaining behavior, the
format conversion must be done first.

Convert the 2KB we copied even if it's more than we need.

Fixes: 7f8e89a8

 ("vfio-ccw: Factor out the ccw0-to-ccw1 transition")
Reported-by: Farhan Ali <alifm@linux.ibm.com>
Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190702180928.18113-1-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

Do not issue CLP_SET_ENABLE_MIO after opting out of MIO instruction
usage. This should not fix a bug but reduce overhead within firmware.

Signed-off-by: Sebastian Ott <sebott@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Unfortunately we have to handle a class of devices that don't support the
new MIO instructions. Adjust resource assignment and mapping accordingly.

Signed-off-by: Sebastian Ott <sebott@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

AP Queue Interruption Control (AQIC) facility gives
the guest the possibility to control interruption for
the Cryptographic Adjunct Processor queues.

Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
Reviewed-by: Tony Krowiak <akrowiak@linux.ibm.com>
Acked-by: Harald Freudenberger <freude@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
[ Modified while picking: we may not expose STFLE facility 65
unconditionally because AIV is a pre-requirement.]
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

We register a AP PQAP instruction hook during the open
of the mediated device. And unregister it on release.

During the probe of the AP device, we allocate a vfio_ap_queue
structure to keep track of the information we need for the
PQAP/AQIC instruction interception.

In the AP PQAP instruction hook, if we receive a demand to
enable IRQs,
- we retrieve the vfio_ap_queue based on the APQN we receive
  in REG1,
- we retrieve the page of the guest address, (NIB), from
  register REG2
- we retrieve the mediated device to use the VFIO pinning
  infrastructure to pin the page of the guest address,
- we retrieve the pointer to KVM to register the guest ISC
  and retrieve the host ISC
- finaly we activate GISA

If we receive a demand to disable IRQs,
- we deactivate GISA
- unregister from the GIB
- unpin the NIB

When removing the AP device from the driver the device is
reseted and this process unregisters the GISA from the GIB,
and unpins the NIB address then we free the vfio_ap_queue
structure.

Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
Acked-by: Tony Krowiak <akrowiak@linux.ibm.com>
Acked-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

To be able to use the VFIO interface to facilitate the
mediated device memory pinning/unpinning we need to register
a notifier for IOMMU.

While we will start to pin one guest page for the interrupt indicator
byte, this is still ok with ballooning as this page will never be
used by the guest virtio-balloon driver.
So the pinned page will never be freed. And even a broken guest does
so, that would not impact the host as the original page is still
in control by vfio.

Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Tony Krowiak <akrowiak@linux.ibm.com>
Acked-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

We prepare the interception of the PQAP/AQIC instruction for
the case the AQIC facility is enabled in the guest.

First of all we do not want to change existing behavior when
intercepting AP instructions without the SIE allowing the guest
to use AP instructions.

In this patch we only handle the AQIC interception allowed by
facility 65 which will be enabled when the complete interception
infrastructure will be present.

We add a callback inside the KVM arch structure for s390 for
a VFIO driver to handle a specific response to the PQAP
instruction with the AQIC command and only this command.

But we want to be able to return a correct answer to the guest
even there is no VFIO AP driver in the kernel.
Therefor, we inject the correct exceptions from inside KVM for the
case the callback is not initialized, which happens when the vfio_ap
driver is not loaded.

We do consider the responsibility of the driver to always initialize
the PQAP callback if it defines queues by initializing the CRYCB for
a guest.
If the callback has been setup we call it.
If not we setup an answer considering that no queue is available
for the guest when no callback has been setup.

Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
Reviewed-by: Tony Krowiak <akrowiak@linux.ibm.com>
Acked-by: Harald Freudenberger <freude@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Kasan instrumentation of backchain unwinder stack reads is disabled
completely and simply uses READ_ONCE_NOCHECK now.
READ_ONCE_TASK_STACK macro is unused and could be removed.

Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Avoid kasan false positive when current task is interrupted in-between
stack frame allocation and backchain write instructions leaving new stack
frame backchain invalid. In particular if backchain is 0 the unwinder
tries to read pt_regs from the stack and might hit kasan poisoned bytes,
leading to kasan "stack-out-of-bounds" report.

Disable kasan instrumentation of unwinder stack reads, since this
limitation couldn't be handled otherwise with current backchain unwinder
implementation.

Fixes: 78c98f90

 ("s390/unwind: introduce stack unwind API")
Reported-by: Julian Wiedmann <jwi@linux.ibm.com>
Tested-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Current code sets the dsci to 0x00000080. Which doesn't make any sense,
as the indicator area is located in the _left-most_ byte.

Worse: if the dsci is the _shared_ indicator, this potentially clears
the indication of activity for a _different_ device.
tiqdio_thinint_handler() will then have no reason to call that device's
IRQ handler, and the device ends up stalling.

Fixes: d0c9d4a8

 ("[S390] qdio: set correct bit in dsci")
Cc: <stable@vger.kernel.org>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

When tiqdio_remove_input_queues() removes a queue from the tiq_list as
part of qdio_shutdown(), it doesn't re-initialize the queue's list entry
and the prev/next pointers go stale.

If a subsequent qdio_establish() fails while sending the ESTABLISH cmd,
it calls qdio_shutdown() again in QDIO_IRQ_STATE_ERR state and
tiqdio_remove_input_queues() will attempt to remove the queue entry a
second time. This dereferences the stale pointers, and bad things ensue.
Fix this by re-initializing the list entry after removing it from the
list.

For good practice also initialize the list entry when the queue is first
allocated, and remove the quirky checks that papered over this omission.
Note that prior to
commit e5218134 ("s390/qdio: fix access to uninitialized qdio_q fields"),
these checks were bogus anyway.

setup_queues_misc() clears the whole queue struct, and thus needs to
re-init the prev/next pointers as well.

Fixes: 779e6e1c

 ("[S390] qdio: new qdio driver.")
Cc: <stable@vger.kernel.org>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

The "len" variable is the length of the option up to the next option or
to the end of the string which ever first.  We want to print the invalid
option so we want precision "%.*s" but the format is width "%*s" so it
prints up to the end of the string.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Stefan Haberland <sth@linux.ibm.com>
Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Sometimes, we want to control which of the matching drivers
binds to a subchannel device (e.g. for subchannels we want to
handle via vfio-ccw).

For pci devices, a mechanism to do so has been introduced in
782a985d

 ("PCI: Introduce new device binding path using
pci_dev.driver_override"). It makes sense to introduce the
driver_override attribute for subchannel devices as well, so
that we can easily extend the 'driverctl' tool (which makes
use of the driver_override attribute for pci).

Note that unlike pci we still require a driver override to
match the subchannel type; matching more than one subchannel
type is probably not useful anyway.

Signed-off-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Halil Pasic <pasic@linux.ibm.com>
Reviewed-by: Sebastian Ott <sebott@linux.ibm.com>
Signed-off-by: Sebastian Ott <sebott@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Reported by sparse.

Fixes: 7f8e89a8

 ("vfio-ccw: Factor out the ccw0-to-ccw1 transition")
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190624090721.16241-1-cohuck@redhat.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Merge tag 'vfio-ccw-20190621' of https://git.kernel.org/pub/scm/linux/kernel/git/kvms390/vfio-ccw into features

Refactoring of the vfio-ccw cp handling, simplifying the
code and avoiding unneeded allocating/copying.

* tag 'vfio-ccw-20190621' of https://git.kernel.org/pub/scm/linux/kernel/git/kvms390/vfio-ccw

:
  vfio-ccw: Remove copy_ccw_from_iova()
  vfio-ccw: Factor out the ccw0-to-ccw1 transition
  vfio-ccw: Copy CCW data outside length calculation
  vfio-ccw: Skip second copy of guest cp to host
  vfio-ccw: Move guest_cp storage into common struct
  s390/cio: Combine direct and indirect CCW paths
  vfio-ccw: Rearrange IDAL allocation in direct CCW
  vfio-ccw: Remove pfn_array_table
  vfio-ccw: Adjust the first IDAW outside of the nested loops
  vfio-ccw: Rearrange pfn_array and pfn_array_table arrays
  s390/cio: Use generalized CCW handler in cp_init()
  s390/cio: Generalize the TIC handler
  s390/cio: Refactor the routine that handles TIC CCWs
  s390/cio: Squash cp_free() and cp_unpin_free()

Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Just to keep things tidy.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Message-Id: <20190618202352.39702-6-farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Farhan Ali <alifm@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

This is a really useful function, but it's buried in the
copy_ccw_from_iova() routine so that ccwchain_calc_length()
can just work with Format-1 CCWs while doing its counting.
But it means we're translating a full 2K of "CCWs" to Format-1,
when in reality there's probably far fewer in that space.

Let's factor it out, so maybe we can do something with it later.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Message-Id: <20190618202352.39702-5-farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Farhan Ali <alifm@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

It doesn't make much sense to "hide" the copy to the channel_program
struct inside a routine that calculates the length of the chain.

Let's move it to the calling routine, which will later copy from
channel_program to the memory it allocated itself.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Message-Id: <20190618202352.39702-4-farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Farhan Ali <alifm@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

We already pinned/copied/unpinned 2K (256 CCWs) of guest memory
to the host space anchored off vfio_ccw_private.  There's no need
to do that again once we have the length calculated, when we could
just copy the section we need to the "permanent" space for the I/O.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Message-Id: <20190618202352.39702-3-farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Farhan Ali <alifm@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

Rather than allocating/freeing a piece of memory every time
we try to figure out how long a CCW chain is, let's use a piece
of memory allocated for each device.

The io_mutex added with commit 4f766173

 ("vfio-ccw: protect
the I/O region") is held for the duration of the VFIO_CCW_EVENT_IO_REQ
event that accesses/uses this space, so there should be no race
concerns with another CPU attempting an (unexpected) SSCH for the
same device.

Suggested-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Eric Farman <farman@linux.ibm.com>
Message-Id: <20190618202352.39702-2-farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Farhan Ali <alifm@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

This allows device drivers (eg. qeth) to use the struct when processing
information retrieved via RCD.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Acked-by: Sebastian Ott <sebott@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

The stfle inline assembly returns the number of double words written
(condition code 0) or the double words it would have written
(condition code 3), if the memory array it got as parameter would have
been large enough.

The current stfle implementation assumes that the array is always
large enough and clears those parts of the array that have not been
written to with a subsequent memset call.

If however the array is not large enough memset will get a negative
length parameter, which means that memset clears memory until it gets
an exception and the kernel crashes.

To fix this simply limit the maximum length. Move also the inline
assembly to an extra function to avoid clobbering of register 0, which
might happen because of the added min_t invocation together with code
instrumentation.

The bug was introduced with commit 14375bc4 ("[S390] cleanup
facility list handling") but was rather harmless, since it would only
write to a rather large array. It became a potential problem with
commit 3ab121ab ("[S390] kernel: Add z/VM LGR detection"). Since
then it writes to an array with only four double words, while some
machines already deliver three double words. As soon as machines have
a facility bit within the fifth double a crash on IPL would happen.

Fixes: 14375bc4

 ("[S390] cleanup facility list handling")
Cc: <stable@vger.kernel.org> # v2.6.37+
Reviewed-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

This feature has never been used, so remove it.

Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Acked-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Replace defconfig with performance_defconfig. defconfig had some more
or less random debug options enabled, where nobody knows why anymore.

Just remove the old defconfig and replace it with performance_defconfig,
which reduces the number of configs to maintain. A config with debugging
options enabled is debug_defconfig which is supposed to be rather close
to performance_defconfig except that is has debug options enabled.

Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

With both the direct-addressed and indirect-addressed CCW paths
simplified to this point, the amount of shared code between them is
(hopefully) more easily visible.  Move the processing of IDA-specific
bits into the direct-addressed path, and add some useful commentary of
what the individual pieces are doing.  This allows us to remove the
entire ccwchain_fetch_idal() routine and maintain a single function
for any non-TIC CCW.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-10-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

This is purely deck furniture, to help understand the merge of the
direct and indirect handlers.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-9-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

Now that both CCW codepaths build this nested array:

  ccwchain->pfn_array_table[1]->pfn_array[#idaws/#pages]

We can collapse this into simply:

  ccwchain->pfn_array[#idaws/#pages]

Let's do that, so that we don't have to continually navigate two
nested arrays when the first array always has a count of one.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-8-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

Now that pfn_array_table[] is always an array of 1, it seems silly to
check for the very first entry in an array in the middle of two nested
loops, since we know it'll only ever happen once.

Let's move this outside the loops to simplify things, even though
the "k" variable is still necessary.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-7-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

While processing a channel program, we currently have two nested
arrays that carry a slightly different structure.  The direct CCW
path creates this:

  ccwchain->pfn_array_table[1]->pfn_array[#pages]

while an IDA CCW creates:

  ccwchain->pfn_array_table[#idaws]->pfn_array[1]

The distinction appears to state that each pfn_array_table entry
points to an array of contiguous pages, represented by a pfn_array,
um, array.  Since the direct-addressed scenario can ONLY represent
contiguous pages, it makes the intermediate array necessary but
difficult to recognize.  Meanwhile, since an IDAL can contain
non-contiguous pages and there is no logic in vfio-ccw to detect
adjacent IDAWs, it is the second array that is necessary but appearing
to be superfluous.

I am not aware of any documentation that states the pfn_array[] needs
to be of contiguous pages; it is just what the code does today.
I don't see any reason for this either, let's just flip the IDA
codepath around so that it generates:

  ch_pat->pfn_array_table[1]->pfn_array[#idaws]

This will bring it in line with the direct-addressed codepath,
so that we can understand the behavior of this memory regardless
of what type of CCW is being processed.  And it means the casual
observer does not need to know/care whether the pfn_array[]
represents contiguous pages or not.

NB: The existing vfio-ccw code only supports 4K-block Format-2 IDAs,
so that "#pages" == "#idaws" in this area.  This means that we will
have difficulty with this overlap in terminology if support for
Format-1 or 2K-block Format-2 IDAs is ever added.  I don't think that
this patch changes our ability to make that distinction.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-6-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

It is now pretty apparent that ccwchain_handle_ccw()
(nee ccwchain_handle_tic()) does everything that cp_init()
wants to do.

Let's remove that duplicated code from cp_init() and let
ccwchain_handle_ccw() handle it itself.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-5-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

Refactor ccwchain_handle_tic() into a routine that handles a channel
program address (which itself is a CCW pointer), rather than a CCW pointer
that is only a TIC CCW.  This will make it easier to reuse this code for
other CCW commands.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-4-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

Extract the "does the target of this TIC already exist?" check from
ccwchain_handle_tic(), so that it's easier to refactor that function
into one that cp_init() is able to use.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-3-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

The routine cp_free() does nothing but call cp_unpin_free(), and while
most places call cp_free() there is one caller of cp_unpin_free() used
when the cp is guaranteed to have not been marked initialized.

This seems like a dubious way to make a distinction, so let's combine
these routines and make cp_free() do all the work.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20190606202831.44135-2-farman@linux.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>

stop_machine is the only user left of cpu_relax_yield. Given that it
now has special semantics which are tied to stop_machine introduce a
weak stop_machine_yield function which architectures can override, and
get rid of the generic cpu_relax_yield implementation.

Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>

The stop_machine loop to advance the state machine and to wait for all
affected CPUs to check-in calls cpu_relax_yield in a tight loop until
the last missing CPUs acknowledged the state transition.

On a virtual system where not all logical CPUs are backed by real CPUs
all the time it can take a while for all CPUs to check-in. With the
current definition of cpu_relax_yield a diagnose 0x44 is done which
tells the hypervisor to schedule *some* other CPU. That can be any
CPU and not necessarily one of the CPUs that need to run in order to
advance the state machine. This can lead to a pretty bad diagnose 0x44
storm until the last missing CPU finally checked-in.

Replace the undirected cpu_relax_yield based on diagnose 0x44 with a
directed yield. Each CPU in the wait loop will pick up the next CPU
in the cpumask of stop_machine. The diagnose 0x9c is used to tell the
hypervisor to run this next CPU instead of the current one. If there
is only a limited number of real CPUs backing the virtual CPUs we
end up with the real CPUs passed around in a round-robin fashion.

[heiko.carstens@de.ibm.com]:
    Use cpumask_next_wrap as suggested by Peter Zijlstra.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>