The basic issue is to be able to do what hugetlbfs does but with
different page sizes for some other special filesystems; more
specifically, my need is:

 - Huge pages

 - SPE local store mappings using 64K pages on a 4K base page size
kernel on Cell

 - Some special 4K segments in 64K-page kernels for mapping a dodgy
type of powerpc-specific infiniband hardware that requires 4K MMU
mappings for various reasons I won't explain here.

The main issues are:

 - To maintain/keep track of the page size per "segment" (as we can
only have one page size per segment on powerpc, which are 256MB
divisions of the address space).

 - To make sure special mappings stay within their allotted
"segments" (including MAP_FIXED crap)

 - To make sure everybody else doesn't mmap/brk/grow_stack into a
"segment" that is used for a special mapping

Some of the necessary mechanisms to handle that were present in the
hugetlbfs code, but mostly in ways not suitable for anything else.

The patch relies on some changes to the generic get_unmapped_area()
that just got merged.  It still hijacks hugetlb callbacks here or
there as the generic code hasn't been entirely cleaned up yet but
that shouldn't be a problem.

So what is a slice ?  Well, I re-used the mechanism used formerly by our
hugetlbfs implementation which divides the address space in
"meta-segments" which I called "slices".  The division is done using
256MB slices below 4G, and 1T slices above.  Thus the address space is
divided currently into 16 "low" slices and 16 "high" slices.  (Special
case: high slice 0 is the area between 4G and 1T).

Doing so simplifies significantly the tracking of segments and avoids
having to keep track of all the 256MB segments in the address space.

While I used the "concepts" of hugetlbfs, I mostly re-implemented
everything in a more generic way and "ported" hugetlbfs to it.

Slices can have an associated page size, which is encoded in the mmu
context and used by the SLB miss handler to set the segment sizes.  The
hash code currently doesn't care, it has a specific check for hugepages,
though I might add a mechanism to provide per-slice hash mapping
functions in the future.

The slice code provide a pair of "generic" get_unmapped_area() (bottomup
and topdown) functions that should work with any slice size.  There is
some trickiness here so I would appreciate people to have a look at the
implementation of these and let me know if I got something wrong.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Use DEFINE_SPINLOCK instead of initializing spinlocks to
SPIN_LOCK_UNLOCKED, since DEFINE_SPINLOCK is better for lockdep.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This change fixes the case where spu_base and spufs are initialised on a
system with no SPEs - unconditionally create the spu_lists so spu_alloc
doesn't explode, and check for spu_management ops before starting spufs.
Signed-off-by: NJeremy Kerr <jk@ozlabs.org>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>

 arch/powerpc/platforms/cell/spu_base.c    |    7 ++++---
 arch/powerpc/platforms/cell/spufs/inode.c |    5 +++++
 2 files changed, 9 insertions(+), 3 deletions(-)

spu_base.c is always built into the kernel image, so there is no need
for a cleanup function.  And some of the things it does are in the
way for my following patches, so I'd rather get rid of it ASAP.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>

Until now, we have always entered the spu page fault handler
with a mutex for the spu context held. This has multiple
bad side-effects:
- it becomes impossible to suspend the context during
  page faults
- if an spu program attempts to access its own mmio
  areas through DMA, we get an immediate livelock when
  the nopage function tries to acquire the same mutex

This patch makes the page fault logic operate on a
struct spu_context instead of a struct spu, and moves it
from spu_base.c to a new file fault.c inside of spufs.

We now also need to copy the dar and dsisr contents
of the last fault into the saved context to have it
accessible in case we schedule out the context before
activating the page fault handler.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>

There is no reason to execute spu_init_channels under spu_mutex
after the spu has been taken off the freelist it's ours.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>

The SPU code doesn't properly invalidate SPUs SLBs when necessary,
for example when changing a segment size from the hugetlbfs code. In
addition, it saves and restores the SLB content on context switches
which makes it harder to properly handle those invalidations.

This patch removes the saving & restoring for now, something more
efficient might be found later on. It also adds a spu_flush_all_slbs(mm)
that can be used by the core mm code to flush the SLBs of all SPEs that
are running a given mm at the time of the flush.

In order to do that, it adds a spinlock to the list of all SPEs and move
some bits & pieces from spufs to spu_base.c
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

spu->register_lock should be held before accessing registers.
Signed-off-by: NKou Ishizaki <kou.ishizaki@toshiba.co.jp>
Acked-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Signed-off-by: NStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>

This adds a platform specific spu management abstraction and the coresponding
routines to support the IBM Cell Blade.  It also removes the hypervisor only
resources that were included in struct spu.

Three new platform specific routines are introduced, spu_enumerate_spus(),
spu_create_spu() and spu_destroy_spu().  The underlying design uses a new
type, struct spu_management_ops, to hold function pointers that the platform
setup code is expected to initialize to instances appropriate to that platform.

For the IBM Cell Blade support, I put the hypervisor only resources that were
in struct spu into a platform specific data structure struct spu_pdata.
Signed-off-by: NGeoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>

When we attempt an MFC DMA to an unmapped address, the event
returned from spu_run should be SPE_EVENT_SPE_DATA_STORAGE,
not SPE_EVENT_INVALID_DMA.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Replace the use of the platform specific variable spu.nid with the
platform independednt variable spu.node.
Signed-off-by: NGeoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This fixes a typo in the "new style" code for mapping SPE resources,
which causes it to try to map the same resource 4 times.

It also adds some pr_debug's that are useful to track down issues with
the firmware when bringinh up new machines.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This patch adds support for stopping, and restarting, spus
from xmon. We use the spu master runcntl bit to stop execution,
this is apparently the "right" way to control spu execution and
spufs will be changed in the future to use this bit.

Testing has shown that to restart execution we have to turn the
master runcntl bit on and also rewrite the spu runcntl bit, even
if it is already set to 1 (running).

Stopping spus is triggered by the xmon command 'ss' - "spus stop"
perhaps. Restarting them is triggered via 'sr'. Restart doesn't
start execution on spus unless they were running prior to being
stopped by xmon.

Walking the spu->full_list in xmon after a panic, would mean
corruption of any spu struct would make all the others
inaccessible. To avoid this, and also to make the next patch
easier, we cache pointers to all spus during boot.

We attempt to catch and recover from errors while stopping and
restarting the spus, but as with most xmon functionality there are
no guarantees that performing these operations won't crash xmon
itself.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

In order to add sysfs attributes to all spu's, there is a
need for a list of all available spu's. Adding the device_node
makes also sense, as it is needed for proper register access.
This patch also adds two functions to create and remove sysfs
attributes and attribute_groups to all spus.
That allows to group spu attributes in a subdirectory like:
/sys/devices/system/spu/spuX/group_name/what_ever
This will be used by cbe_thermal to group all attributes dealing with
thermal support in one directory.
Signed-off-by: NChristian Krafft <krafft@de.ibm.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This patch adds general support for isolated mode SPE apps.

Isolated apps are started indirectly, by a dedicated loader "kernel".
This patch starts the loader when spe_create is invoked with the
ISOLATE flag. We do this at spe_create time to allow libspe to pass the
isolated app in before calling spe_run.

The loader is read from the device tree, at the location
"/spu-isolation/loader". If the loader is not present, an attempt to
start an isolated SPE binary will fail with -ENODEV.

Update: loader needs to be correctly aligned - copy to a kmalloced buf.
Update: remove workaround for systemsim/spurom 'L-bit' bug, which has
        been fixed.
Update: don't write to runcntl on spu_run_init: SPU is already running.
Update: do spu_setup_isolated earlier

Tested on systemsim.
Signed-off-by: NJeremy Kerr <jk@ozlabs.org>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Remove the mostly unused variable isrc from struct spu and a forgotten
function declaration.
Signed-off-by: NGeoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

SPRN_SDR1 and the SPE's MFC SDR are hypervisor resources and
are not accessible from a logical partition.  This change adds an
access wrapper.

When running on bare H/W, the spufs needs to only set the SPE's MFC SDR
to the value of the PPE's SPRN_SDR1 once at SPE initialization, so this
change renames mfc_sdr_set() to mfc_sdr_setup() and moves the
access of SPRN_SDR1 into the mmio wrapper.  It also removes the now
unneeded member mfc_sdr_RW from struct spu_priv1_collapsed.
Signed-off-by: NMasato Noguchi <Masato.Noguchi@jp.sony.com>
Signed-off-by: NGeoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>

--
Signed-off-by: NPaul Mackerras <paulus@samba.org>

The SPU code will crash if CONFIG_NUMA is not set and SPUs are found on
a non-0 node. This workaround will ignore those SPEs and just print an
message in the kernel log.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Remove struct pt_regs * from remaining spu irq functions.
Signed-off-by: NOlaf Hering <olaf@aepfle.de>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

The properties we used traditionally in the device tree are somewhat
nonstandard.  This adds support for a more conventional format using
'interrupts' and 'reg' properties.

The interrupts are specified in three cells (class 0, 1 and 2) and
registered at the interrupt-parent.

The reg property contains either three or four register areas in the
order 'local-store', 'problem', 'priv2', and 'priv1', so the priv1 one
can be left out in case of hypervisor driven systems that access these
through hcalls.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Any firmware that still uses the 'spc' nodes already
stopped running for other reasons, so let's get rid of this.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This tries to fix spufs so we have an interface closer to what is
specified in the man page for events returned in the third argument of
spu_run.

Fortunately, libspe has never been using the returned contents of that
register, as they were the same as the return code of spu_run (duh!).

Unlike the specification that we never implemented correctly, we now
require a SPU_CREATE_EVENTS_ENABLED flag passed to spu_create, in
order to get the new behavior. When this flag is not passed, spu_run
will simply ignore the third argument now.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This patch adds NUMA support to the the spufs scheduler.

The new arch/powerpc/platforms/cell/spufs/sched.c is greatly
simplified, in an attempt to reduce complexity while adding
support for NUMA scheduler domains.  SPUs are allocated starting
from the calling thread's node, moving to others as supported by
current->cpus_allowed.  Preemption is gone as it was buggy, but
should be re-enabled in another patch when stable.

The new arch/powerpc/platforms/cell/spu_base.c maintains idle
lists on a per-node basis, and allows caller to specify which
node(s) an SPU should be allocated from, while passing -1 tells
spu_alloc() that any node is allowed.

Since the patch removes the currently implemented preemptive
scheduling, it is technically a regression, but practically
all users have since migrated to this version, as it is
part of the IBM SDK and the yellowdog distribution, so there
is not much point holding it back while the new preemptive
scheduling patch gets delayed further.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This patch reworks the cell iic interrupt handling so that:

 - Node ID is back in the interrupt number (only one IRQ host is created
for all nodes). This allows interrupts from sources on another node to
be routed non-locally. This will allow possibly one day to fix maxcpus=1
or 2 and still get interrupts from devices on BE 1. (A bit more fixing
is needed for that) and it will allow us to implement actual affinity
control of external interrupts.

 - Added handling of the IO exceptions interrupts (badly named, but I
re-used the name initially used by STI). Those are the interrupts
exposed by IIC_ISR and IIC_IRR, such as the IOC translation exception,
performance monitor, etc... Those get their special numbers in the IRQ
number space and are internally implemented as a cascade on unit 0xe,
class 1 of each node.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Now that get_property() returns a void *, there's no need to cast its
return value. Also, treat the return value as const, so we can
constify get_property later.

cell platform changes.

Built for cell_defconfig
Signed-off-by: NJeremy Kerr <jk@ozlabs.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This patch slightly reworks the new irq code to fix a small design error.  I
removed the passing of the trigger to the map() calls entirely, it was not a
good idea to have one call do two different things.  It also fixes a couple of
corner cases.

Mapping a linux virtual irq to a physical irq now does only that.  Setting the
trigger is a different action which has a different call.

The main changes are:

- I no longer call host->ops->map() for an already mapped irq, I just return
  the virtual number that was already mapped.  It was called before to give an
  opportunity to change the trigger, but that was causing issues as that could
  happen while the interrupt was in use by a device, and because of the
  trigger change, map would potentially muck around with things in a racy way.
   That was causing much burden on a given's controller implementation of
  map() to get it right.  This is much simpler now.  map() is only called on
  the initial mapping of an irq, meaning that you know that this irq is _not_
  being used.  You can initialize the hardware if you want (though you don't
  have to).

- Controllers that can handle different type of triggers (level/edge/etc...)
  now implement the standard irq_chip->set_type() call as defined by the
  generic code.  That means that you can use the standard set_irq_type() to
  configure an irq line manually if you wish or (though I don't like that
  interface), pass explicit trigger flags to request_irq() as defined by the
  generic kernel interfaces.  Also, using those interfaces guarantees that
  your controller set_type callback is called with the descriptor lock held,
  thus providing locking against activity on the same interrupt (including
  mask/unmask/etc...) automatically.  A result is that, for example, MPIC's
  own map() implementation calls irq_set_type(NONE) to configure the hardware
  to the default triggers.

- To allow the above, the irq_map array entry for the new mapped interrupt
  is now set before map() callback is called for the controller.

- The irq_create_of_mapping() (also used by irq_of_parse_and_map()) function
  for mapping interrupts from the device-tree now also call the separate
  set_irq_type(), and only does so if there is a change in the trigger type.

- While I was at it, I changed pci_read_irq_line() (which is the helper I
  would expect most archs to use in their pcibios_fixup() to get the PCI
  interrupt routing from the device tree) to also handle a fallback when the
  DT mapping fails consisting of reading the PCI_INTERRUPT_PIN to know wether
  the device has an interrupt at all, and the the PCI_INTERRUPT_LINE to get an
  interrupt number from the device.  That number is then mapped using the
  default controller, and the trigger is set to level low.  That default
  behaviour works for several platforms that don't have a proper interrupt
  tree like Pegasos.  If it doesn't work for your platform, then either
  provide a proper interrupt tree from the firmware so that fallback isn't
  needed, or don't call pci_read_irq_line()

- Add back a bit that got dropped by my main rework patch for properly
  clearing pending IPIs on pSeries when using a kexec
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This adds the new irq remapper core and removes the old one.  Because
there are some fundamental conflicts with the old code, like the value
of NO_IRQ which I'm now setting to 0 (as per discussions with Linus),
etc..., this commit also changes the relevant platform and driver code
over to use the new remapper (so as not to cause difficulties later
in bisecting).

This patch removes the old pre-parsing of the open firmware interrupt
tree along with all the bogus assumptions it made to try to renumber
interrupts according to the platform. This is all to be handled by the
new code now.

For the pSeries XICS interrupt controller, a single remapper host is
created for the whole machine regardless of how many interrupt
presentation and source controllers are found, and it's set to match
any device node that isn't a 8259.  That works fine on pSeries and
avoids having to deal with some of the complexities of split source
controllers vs. presentation controllers in the pSeries device trees.

The powerpc i8259 PIC driver now always requests the legacy interrupt
range. It also has the feature of being able to match any device node
(including NULL) if passed no device node as an input. That will help
porting over platforms with broken device-trees like Pegasos who don't
have a proper interrupt tree.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Use the new IRQF_ constants and remove the SA_INTERRUPT define
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

The class zero interrupt handling for spus was confusing alignment and
error interrupts, so swap them.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

SPU interrupt status must be cleared before handle it.
Otherwise, kernel may drop some interrupt packet.

Currently, class2 interrupt treated like:
 1) call callback to wake up waiting process
 2) mask raised mailbox interrupt
 3) clear interrupt status

I changed like:
 1) mask raised mailbox interrupt
 2) clear interrupt status
 3) call callback to wake up waiting process

Clearing status before masking will make spurious interrupt.
Thus, it is necessary to hold by steps I described above, I think.
Signed-off-by: NMasato Noguchi <Masato.Noguchi@jp.sony.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This patch remove 'stop_code' -- discarded member of struct spu.
It is written at initialize and interrupt, but never read
in current implementation.
Signed-off-by: NMasato Noguchi <Masato.Noguchi@jp.sony.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This changes the hypervisor abstraction of setting cpu affinity to a
higher level to avoid platform dependent interrupt controller
routines.  I replaced spu_priv1_ops:spu_int_route_set() with a
new routine spu_priv1_ops:spu_cpu_affinity_set().

As a by-product, this change eliminated what looked like an
existing bug in the set affinity code where spu_int_route_set()
mistakenly called int_stat_get().
Signed-off-by: NGeoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

To support muti-platform binaries the spu hypervisor accessor
routines must have runtime binding.

I removed the existing statically linked routines in spu.h
and spu_priv1_mmio.c and created new accessor routines in spu_priv1.h
that operate indirectly through an ops struct spu_priv1_ops.
spu_priv1_mmio.c contains the instance of the accessor routines
for running on raw hardware.
Signed-off-by: NGeoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Clean up create_spu() a little by using kzalloc instead of kmalloc +
assignments.
Signed-off-by: NJeremy Kerr <jk@ozlabs.org>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

spufs currently knows only 4k pages and 16M hugetlb
pages. Make it use the regular methods for deciding on
the SLB bits.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

SPUs are registered as system devices, exposing attributes through
sysfs. Since the sysdev includes a kref, we can remove the one in
struct spu (it isn't used at the moment anyway).

Currently only the interrupt source and numa node attributes are added.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Add an nid member to the spu structure, and store the numa id of the spu there
on creation.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Based on an older patch from  Mike Kravetz <kravetz@us.ibm.com>

We need to have a mem_map for high addresses in order to make fops->no_page
work on spufs mem and register files.  So far, we have used the
memory_present() function during early bootup, but that did not work when
CONFIG_NUMA was enabled.

We now use the __add_pages() function to add the mem_map when loading the
spufs module, which is a lot nicer.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>