The remaining ones occur before we have imported the
device tree.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In particular, move the IRQ probing out to sparc32/sparc64
arch specific code where it belongs.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is the long overdue conversion of sparc64 over to
the generic IRQ layer.

The kernel image is slightly larger, but the BSS is ~60K
smaller due to the reduced size of struct ino_bucket.

A lot of IRQ implementation details, including ino_bucket,
were moved out of asm-sparc64/irq.h and are now private to
arch/sparc64/kernel/irq.c, and most of the code in irq.c
totally disappeared.

One thing that's different at the moment is IRQ distribution,
we do it at enable_irq() time.  If the cpu mask is ALL then
we round-robin using a global rotating cpu counter, else
we pick the first cpu in the mask to support single cpu
targetting.  This is similar to what powerpc's XICS IRQ
support code does.

This works fine on my UP SB1000, and the SMP build goes
fine and runs on that machine, but lots of testing on
different setups is needed.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Inspired by PowerPC XICS interrupt support code.

All IRQs are virtualized in order to keep NR_IRQS from needing
to be too large.  Interrupts on sparc64 are arbitrary 11-bit
values, but we don't need to define NR_IRQS to 2048 if we
virtualize the IRQs.

As PCI and SBUS controller drivers build device IRQs, we divy
out virtual IRQ numbers incrementally starting at 1.  Zero is
a special virtual IRQ used for the timer interrupt.

So device drivers all see virtual IRQs, and all the normal
interfaces such as request_irq(), enable_irq(), etc. translate
that into a real IRQ number in order to configure the IRQ.

At this point knowledge of the struct ino_bucket is almost
entirely contained within arch/sparc64/kernel/irq.c  There are
a few small bits in the PCI controller drivers that need to
be swept away before we can remove ino_bucket's definition
out of asm-sparc64/irq.h and privately into kernel/irq.c
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

And reuse that struct member for virt_irq, which will
be used in future changesets for the implementation of
mapping between real and virtual IRQ numbers.

This nicely kills off a ton of SBUS and PCI controller
PIL assignment code which is no longer necessary.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Don't piggy back the SMP receive signal code to do the
context version change handling.

Instead allocate another fixed PIL number for this
asynchronous cross-call.  We can't use smp_call_function()
because this thing is invoked with interrupts disabled
and a few spinlocks held.

Also, fix smp_call_function_mask() to count "cpus" correctly.
There is no guarentee that the local cpu is in the mask
yet that is exactly what this code was assuming.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Earlier I unifdefed PageCompound, so that snd_pcm_mmap_control_nopage and
others can give out a 0-order component of a higher-order page, which won't
be mistakenly freed when zap_pte_range unmaps it.  But many Bad page states
reported a PG_reserved was freed after all: I had missed that we need to
say __GFP_COMP to get compound page behaviour.

Some of these higher-order pages are allocated by snd_malloc_pages, some by
snd_malloc_dev_pages; or if SBUS, by sbus_alloc_consistent - but that has
no gfp arg, so add __GFP_COMP into its sparc32/64 implementations.

I'm still rather puzzled that DRM seems not to need a similar change.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Use ARRAY_SIZE macro instead of sizeof(x)/sizeof(x[0]) and remove a
duplicate of ARRAY_SIZE which is never used anyways.
Signed-off-by: NTobias Klauser <tklauser@nuerscht.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It appears that a memory barrier soon after a mispredicted
branch, not just in the delay slot, can cause the hang
condition of this cpu errata.

So move them out-of-line, and explicitly put them into
a "branch always, predict taken" delay slot which should
fully kill this problem.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Firstly, if the direction is TODEVICE, then dirty data in the
streaming cache is impossible so we can elide the flush-flag
synchronization in that case.

Next, the context allocator is broken.  It is highly likely
that contexts get used multiple times for different dma
mappings, which confuses the strbuf flushing code and makes
it run inefficiently.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The recent change to add a timeout to strbuf flushing had
a negative performance impact.  The udelay()'s are too long,
and they were done in the wrong order wrt. the register read
checks.  Fix both, and things are happy again.

There are more possible improvements in this area.  In fact,
PCI streaming buffer flushing seems to be part of the bottleneck
in network receive performance on my SunBlade1000 box.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If some hardware error occurs and the flush flag never updates,
we will hang forever in these routines.  Add a timeout, and
print out a diagnostic if it is reached.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.

Let it rip!