Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is good for up to %50 performance improvement of some test cases.
The problem has been the race conditions, and hopefully I've plugged
them all up here.

1) There was a serious race in switch_mm() wrt. lazy TLB
   switching to and from kernel threads.

   We could erroneously skip a tsb_context_switch() and thus
   use a stale TSB across a TSB grow event.

   There is a big comment now in that function describing
   exactly how it can happen.

2) All code paths that do something with the TSB need to be
   guarded with the mm->context.lock spinlock.  This makes
   page table flushing paths properly synchronize with both
   TSB growing and TLB context changes.

3) TSB growing events are moved to the end of successful fault
   processing.  Previously it was in update_mmu_cache() but
   that is deadlock prone.  At the end of do_sparc64_fault()
   we hold no spinlocks that could deadlock the TSB grow
   sequence.  We also have dropped the address space semaphore.

While we're here, add prefetching to the copy_tsb() routine
and put it in assembler into the tsb.S file.  This piece of
code is quite time critical.

There are some small negative side effects to this code which
can be improved upon.  In particular we grab the mm->context.lock
even for the tsb insert done by update_mmu_cache() now and that's
a bit excessive.  We can get rid of that locking, and the same
lock taking in flush_tsb_user(), by disabling PSTATE_IE around
the whole operation including the capturing of the tsb pointer
and tsb_nentries value.  That would work because anyone growing
the TSB won't free up the old TSB until all cpus respond to the
TSB change cross call.

I'm not quite so confident in that optimization to put it in
right now, but eventually we might be able to and the description
is here for reference.

This code seems very solid now.  It passes several parallel GCC
bootstrap builds, and our favorite "nut cruncher" stress test which is
a full "make -j8192" build of a "make allmodconfig" kernel.  That puts
about 256 processes on each cpu's run queue, makes lots of process cpu
migrations occur, causes lots of page table and TLB flushing activity,
incurs many context version number changes, and it swaps the machine
real far out to disk even though there is 16GB of ram on this test
system. :-)
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The page->flags manipulations done by the D-cache dirty
state tracking was broken because the constants were not
marked with "UL" to make them 64-bit, which means we were
clobbering the upper 32-bits of page->flags all the time.

This doesn't jive well with sparsemem which stores the
section and indexing information in the top 32-bits of
page->flags.

This is yet another sparc64 bug which has been with us
forever.

While we're here, tidy up some things in bootmem_init()
and paginig_init():

1) Pass min_low_pfn to init_bootmem_node(), it's identical
   to (phys_base >> PAGE_SHIFT) but we should use consistent
   with the variable names we print in CONFIG_BOOTMEM_DEBUG

2) max_mapnr, although no longer used, was being set
   inaccurately, we shouldn't subtract pfn_base any more.

3) All the games with phys_base in the zones_*[] arrays
   we pass to free_area_init_node() are no longer necessary.

Thanks to Josh Grebe and Fabbione for the bug reports
and testing.  Fix also verified locally on an SB2500
which had a memory layout that triggered the same problem.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This has been pending for a long time, and the fact
that we waste a ton of ram on some configurations
kind of pushed things over the edge.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

1) Always spin_lock_init() in init_context().  The caller essentially
   clears it out, or copies the mm info from the parent.  In both
   cases we need to explicitly initialize the spinlock.

2) Always do explicit IRQ disabling while taking mm->context.lock
   and ctx_alloc_lock.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If we were aligned, but didn't have at least 256MB left
to process, we would loop forever.

Thanks to fabbione for the report and testing the fix.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Don't try to avoid putting non-base page sized entries
into the user TSB.  It actually costs us more to check
this than it helps.

Eventually we'll have a multiple TSB scheme for user
processes.  Once a process starts using larger pages,
we'll allocate and use such a TSB.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It is totally wasted work, since we have no D-cache aliasing
issues on sun4v.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The context allocation scheme we use depends upon there being a 1<-->1
mapping from cpu to physical TLB for correctness.  Chips like Niagara
break this assumption.

So what we do is notify all cpus with a cross call when the context
version number changes, and if necessary this makes them allocate
a valid context for the address space they are running at the time.

Stress tested with make -j1024, make -j2048, and make -j4096 kernel
builds on a 32-strand, 8 core, T2000 with 16GB of ram.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

SBUS flash driver needs it.

Noticed by Fabbione.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It can map all of the linear kernel mappings with zero TSB hash
conflicts for systems with 16GB or less ram.  In such cases, on
SUN4V, once we load up this TSB the first time with all the
mappings, we never take a linear kernel mapping TLB miss ever
again, the hypervisor handles them all.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We use a bitmap, one bit for every 256MB of memory.  If the
bit is set we can use a 256MB PTE for linear mappings, else
we have to use a 4MB PTE.

SUN4V support is there, and we can very easily add support
for Panther cpu 256MB PTEs in the future.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

For drivers/media/*, noticed by Fabbione.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The SUN4V convention with non-shared TSBs is that the context
bit of the TAG is clear.  So we have to choose an "invalid"
bit and initialize new TSBs appropriately.  Otherwise a zero
TAG looks "valid".

Make sure, for the window fixup cases, that we use the right
global registers and that we don't potentially trample on
the live global registers in etrap/rtrap handling (%g2 and
%g6) and that we put the missing virtual address properly
in %g5.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It should be 1, not 0.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This handles the SUN4U vs SUN4V PTE layout differences
with near zero performance cost.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Yes, you heard it right, they changed the PTE layout for
SUN4V.  Ho hum...

This is the simple and inefficient way to support this.
It'll get optimized, don't worry.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We do this right after we take over the trap table from OBP.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is where the virtual address of the fault status
area belongs.

To set it up we don't make a hypervisor call, instead
we call OBP's SUNW,set-trap-table with the real address
of the fault status area as the second argument.  And
right before that call we write the virtual address into
ASI_SCRATCHPAD vaddr 0x0.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Function goes in %o5, args go in %o0 --> %o5.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We look for "SUNW,sun4v" in the 'compatible' property
of the root OBP device tree node.

Protect every %ver register access, to make sure it is
not touched on sun4v, as %ver is hyperprivileged there.

Lock kernel TLB entries using hypervisor calls instead of
calls into OBP.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

sun4v uses ASI_MMU instead of ASI_DMMU
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Things are a little tricky because, unlike sun4u, we have
to:

1) do a hypervisor trap to do the TLB load.
2) do the TSB lookup calculations by hand
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

And more consistently check cheetah{,_plus} instead
of assuming anything not spitfire is cheetah{,_plus}.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There are several tricky races involved with growing the TSB.  So just
use base-size TSBs for user contexts and we can revisit enabling this
later.

One part of the SMP problems is that tsb_context_switch() can see
partially updated TSB configuration state if tsb_grow() is running in
parallel.  That's easily solved with a seqlock taken as a writer by
tsb_grow() and taken as a reader to capture all the TSB config state
in tsb_context_switch().

Then there is flush_tsb_user() running in parallel with a tsb_grow().
In theory we could take the seqlock as a reader there too, and just
resample the TSB pointer and reflush but that looks really ugly.

Lastly, I believe there is a case with threads that results in a TSB
entry lock bit being set spuriously which will cause the next access
to that TSB entry to wedge the cpu (since the TSB entry lock bit will
never clear).  It's either copy_tsb() or some bug elsewhere in the TSB
assembly.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This way we don't need to lock the TSB into the TLB.
The trick is that every TSB load/store is registered into
a special instruction patch section.  The default uses
virtual addresses, and the patch instructions use physical
address load/stores.

We can't do this on all chips because only cheetah+ and later
have the physical variant of the atomic quad load.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

No longer used, and move extern declaration to a header file.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It is totally unnecessary complexity.  After we take over
the trap table, we handle all PROM tlb misses fully.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

As the RSS grows, grow the TSB in order to reduce the likelyhood
of hash collisions and thus poor hit rates in the TSB.

This definitely needs some serious tuning.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Taking a nod from the powerpc port.

With the per-cpu caching of both the page allocator and SLAB, the
pgtable quicklist scheme becomes relatively silly and primitive.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Unlike the virtual page tables, the new TSB scheme does not
require this ugly hack.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We now use the TSB hardware assist features of the UltraSPARC
MMUs.

SMP is currently knowingly broken, we need to find another place
to store the per-cpu base pointers.  We hid them away in the TSB
base register, and that obviously will not work any more :-)

Another known broken case is non-8KB base page size.

Also noticed that flush_tlb_all() is not referenced anywhere, only
the internal __flush_tlb_all() (local cpu only) is used by the
sparc64 port, so we can get rid of flush_tlb_all().

The kernel gets it's own 8KB TSB (swapper_tsb) and each address space
gets it's own private 8K TSB.  Later we can add code to dynamically
increase the size of per-process TSB as the RSS grows.  An 8KB TSB is
good enough for up to about a 4MB RSS, after which the TSB starts to
incur many capacity and conflict misses.

We even accumulate OBP translations into the kernel TSB.

Another area for refinement is large page size support.  We could use
a secondary address space TSB to handle those.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The sequence to move over to the Linux trap tables from
the firmware ones needs to be more air tight.  It turns
out that to be %100 safe we do need to be able to translate
OBP mappings in our TLB miss handlers early.

In order not to eat up a lot of kernel image memory with
static page tables, just use the translations array in
the OBP TLB miss handlers.  That solves the bulk of the
problem.

Furthermore, to make sure the OBP TLB miss path will work
even before the fixed MMU globals are loaded, explicitly
load %g1 to TLB_SFSR at the beginning of the i-TLB and
d-TLB miss handlers.

To ease the OBP TLB miss walking of the prom_trans[] array,
we sort it then delete all of the non-OBP entries in there
(for example, there are entries for the kernel image itself
which we're not interested in at all).

We also save about 32K of kernel image size with this change.
Not a bad side effect :-)

There are still some reasons why trampoline.S can't use the
setup_trap_table() yet.  The most noteworthy are:

1) OBP boots secondary processors with non-bias'd stack for
   some reason.  This is easily fixed by using a small bootup
   stack in the kernel image explicitly for this purpose.

2) Doing a firmware call via the normal C call prom_set_trap_table()
   goes through the whole OBP enter/exit sequence that saves and
   restores OBP and Linux kernel state in the MMUs.  This path
   unfortunately does a "flush %g6" while loading up the OBP locked
   TLB entries for the firmware call.

   If we setup the %g6 in the trampoline.S code properly, that
   is in the PAGE_OFFSET linear mapping, but we're not on the
   kernel trap table yet so those addresses won't translate properly.

   One idea is to do a by-hand firmware call like we do in the
   early bootup code and elsewhere here in trampoline.S  But this
   fails as well, as aparently the secondary processors are not
   booted with OBP's special locked TLB entries loaded.  These
   are necessary for the firwmare to processes TLB misses correctly
   up until the point where we take over the trap table.

This does need to be resolved at some point.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

By allocating early memory for the firmware page tables, we
can write over the beginning of the initrd image.

So what we do now is:

1) Read in firmware translations table while still on the
   firmware's trap table.
2) Switch to Linux trap table.
3) Init bootmem.
4) Build firmware page tables using __alloc_bootmem().

And this keeps the initrd from being clobbered.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>