This changes vmemmap to use a different region (region 0xf) of the
address space, and to configure the page size of that region
dynamically at boot.
The problem with the current approach of always using 16M pages is that
it's not well suited to machines that have small amounts of memory such
as small partitions on pseries, or PS3's.
In fact, on the PS3, failure to allocate the 16M page backing vmmemmap
tends to prevent hotplugging the HV's "additional" memory, thus limiting
the available memory even more, from my experience down to something
like 80M total, which makes it really not very useable.
The logic used by my match to choose the vmemmap page size is:
- If 16M pages are available and there's 1G or more RAM at boot,
use that size.
- Else if 64K pages are available, use that
- Else use 4K pages
I've tested on a POWER6 (16M pages) and on an iSeries POWER3 (4K pages)
and it seems to work fine.
Note that I intend to change the way we organize the kernel regions &
SLBs so the actual region will change from 0xf back to something else at
one point, as I simplify the SLB miss handler, but that will be for a
later patch.
Signed-off-by: Paul Mackerras <paulus@samba.org>
This fixes a regression reported by Kamalesh Bulabel where a POWER4
machine would crash because of an SLB miss at a point where the SLB
miss exception was unrecoverable. This regression is tracked at:
http://bugzilla.kernel.org/show_bug.cgi?id=10082
SLB misses at such points shouldn't happen because the kernel stack is
the only memory accessed other than things in the first segment of the
linear mapping (which is mapped at all times by entry 0 of the SLB).
The context switch code ensures that SLB entry 2 covers the kernel
stack, if it is not already covered by entry 0. None of entries 0
to 2 are ever replaced by the SLB miss handler.
Where this went wrong is that the context switch code assumes it
doesn't have to write to SLB entry 2 if the new kernel stack is in the
same segment as the old kernel stack, since entry 2 should already be
correct. However, when we start up a secondary cpu, it calls
slb_initialize, which doesn't set up entry 2. This is correct for
the boot cpu, where we will be using a stack in the kernel BSS at this
point (i.e. init_thread_union), but not necessarily for secondary
cpus, whose initial stack can be allocated anywhere. This doesn't
cause any immediate problem since the SLB miss handler will just
create an SLB entry somewhere else to cover the initial stack.
In fact it's possible for the cpu to go quite a long time without SLB
entry 2 being valid. Eventually, though, the entry created by the SLB
miss handler will get overwritten by some other entry, and if the next
access to the stack is at an unrecoverable point, we get the crash.
This fixes the problem by making slb_initialize create a suitable
entry for the kernel stack, if we are on a secondary cpu and the stack
isn't covered by SLB entry 0. This requires initializing the
get_paca()->kstack field earlier, so I do that in smp_create_idle
where the current field is initialized. This also abstracts a bit of
the computation that mk_esid_data in slb.c does so that it can be used
in slb_initialize.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Arrange for a syntax check to always be done on the powerpc/mm/slb.c
DBG() macro by defining it to pr_debug() for non-debug builds.
Also, fix these related compile warnings:
slb.c:273: warning: format '%04x' expects type 'unsigned int', but argument 2 has type 'long unsigned int
slb.c:274: warning: format '%04x' expects type 'unsigned int', but argument 2 has type 'long unsigned int'
Signed-off-by: Geoff Levand <geoffrey.levand@am.sony.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Since the PMU is an NMI now, it can come at any time we are only soft
disabled. We must hard disable around the two places we allow the kernel
stack SLB and r1 to go out of sync. Otherwise the PMU exception can
force a kernel stack SLB into another slot, which can lead to it
getting evicted, which can lead to a nasty unrecoverable SLB miss
in the exception entry code.
Signed-off-by: Anton Blanchard <anton@samba.org>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Commit 473980a993 added a call to clear
the SLB shadow buffer before registering it. Unfortunately this means
that we clear out the entries that slb_initialize has previously set in
there. On POWER6, the hypervisor uses the SLB shadow buffer when doing
partition switches, and that means that after the next partition switch,
each non-boot CPU has no SLB entries to map the kernel text and data,
which causes it to crash.
This fixes it by reverting most of 473980a9 and instead clearing the
3rd entry explicitly in slb_initialize. This fixes the problem that
473980a9 was trying to solve, but without breaking POWER6.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Before we register the SLB shadow buffer, we need to invalidate the
entries in the buffer, otherwise we can end up stale entries from when
we previously offlined the CPU.
This does this invalidate as well as unregistering the buffer with
PHYP before we offline the cpu. Tested and fixes crashes seen on
970MP (thanks to tonyb) and POWER5.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Currently we hardwire the number of SLBs to 64, but PAPR says we
should use the ibm,slb-size property to obtain the number of SLB
entries. This uses this property instead of assuming 64. If no
property is found, we assume 64 entries as before.
This soft patches the SLB handler, so it shouldn't change performance
at all.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Now that we have 1TB segment size support, we need to be using the
GET_ESID_1T macro when comparing ESID values for pc, stack, and
unmapped_base within switch_slb(). A new helper function called
esids_match() contains the logic for deciding when to call GET_ESID
and GET_ESID_1T.
This fixes a duplicate-slb-entry inspired machine-check exception I
was seeing when trying to run java on a power6 partition.
Tested on power6 and power5.
Signed-off-by: Will Schmidt <will_schmidt@vnet.ibm.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This fixes the error
error: implicit declaration of function "udbg_printf"
We have a few spots where we reference udbg_printf() without #including
udbg.h. These are within #ifdef DEBUG blocks, so unnoticed until we do
a #define DEBUG or #define DEBUG_LOW nearby.
Signed-off-by: Will Schmidt <will_schmidt@vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
PA6T has a bug where the slbie instruction does not honor the large
segment bit. As a result, we have to always use slbia when switching
context.
We don't have to worry about changing the slbie's during fault processing,
since they should never be replacing one VSID with another using the
same ESID. I.e. there's no risk for inserting duplicate entries due to a
failed slbie of the old entry. So as long as we clear it out on context
switch we should be fine.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This makes the kernel use 1TB segments for all kernel mappings and for
user addresses of 1TB and above, on machines which support them
(currently POWER5+, POWER6 and PA6T).
We detect that the machine supports 1TB segments by looking at the
ibm,processor-segment-sizes property in the device tree.
We don't currently use 1TB segments for user addresses < 1T, since
that would effectively prevent 32-bit processes from using huge pages
unless we also had a way to revert to using 256MB segments. That
would be possible but would involve extra complications (such as
keeping track of which segment size was used when HPTEs were inserted)
and is not addressed here.
Parts of this patch were originally written by Ben Herrenschmidt.
Signed-off-by: Paul Mackerras <paulus@samba.org>
After talking to an IBM POWER hypervisor (PHYP) design and development
guy, there seems to be no need for memory barriers when updating the SLB
shadow buffer provided we only update it from the current CPU, which we
do.
Also, these guys see no need in the future for these barriers.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This partially reverts edd0622bd2.
It turns out that the part of that commit that aimed to ensure that we
created an SLB entry for the kernel stack on secondary CPUs when
starting the CPU didn't achieve its aim, and in fact caused a
regression, because get_paca()->kstack is not initialized at the point
where slb_initialize is called.
This therefore just reverts that part of that commit, while keeping
the change to slb_flush_and_rebolt, which is correct and necessary.
Signed-off-by: Paul Mackerras <paulus@samba.org>
We were getting a duplicate entry in the SLB shadow buffer in
slb_flush_and_rebolt() if the kernel stack was in the same segment
as PAGE_OFFSET, which on POWER6 causes the hypervisor to terminate
the partition with an error. This fixes it.
Also we were not creating an SLB entry (or an SLB shadow buffer
entry) for the kernel stack on secondary CPUs when starting the
CPU. This isn't a major problem, since an appropriate entry will
be created on demand, but this fixes that also for consistency.
Signed-off-by: Paul Mackerras <paulus@samba.org>
On a machine with hardware 64kB pages and a kernel configured for a
64kB base page size, we need to change the vmalloc segment from 64kB
pages to 4kB pages if some driver creates a non-cacheable mapping in
the vmalloc area. However, we never updated with SLB shadow buffer.
This fixes it. Thanks to paulus for finding this.
Also added some write barriers to ensure the shadow buffer contents
are always consistent.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
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: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This adds a shadow buffer for the SLBs and regsiters it with PHYP.
Only the bolted SLB entries (top 3) are shadowed.
The SLB shadow buffer tells the hypervisor what the kernel needs to
have in the SLB for the kernel to be able to function. The hypervisor
can use this information to speed up partition context switches.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Some POWER5+ machines can do 64k hardware pages for normal memory but
not for cache-inhibited pages. This patch lets us use 64k hardware
pages for most user processes on such machines (assuming the kernel
has been configured with CONFIG_PPC_64K_PAGES=y). User processes
start out using 64k pages and get switched to 4k pages if they use any
non-cacheable mappings.
With this, we use 64k pages for the vmalloc region and 4k pages for
the imalloc region. If anything creates a non-cacheable mapping in
the vmalloc region, the vmalloc region will get switched to 4k pages.
I don't know of any driver other than the DRM that would do this,
though, and these machines don't have AGP.
When a region gets switched from 64k pages to 4k pages, we do not have
to clear out all the 64k HPTEs from the hash table immediately. We
use the _PAGE_COMBO bit in the Linux PTE to indicate whether the page
was hashed in as a 64k page or a set of 4k pages. If hash_page is
trying to insert a 4k page for a Linux PTE and it sees that it has
already been inserted as a 64k page, it first invalidates the 64k HPTE
before inserting the 4k HPTE. The hash invalidation routines also use
the _PAGE_COMBO bit, to determine whether to look for a 64k HPTE or a
set of 4k HPTEs to remove. With those two changes, we can tolerate a
mix of 4k and 64k HPTEs in the hash table, and they will all get
removed when the address space is torn down.
Signed-off-by: Paul Mackerras <paulus@samba.org>
The pgdir field in the paca was a leftover from the dynamic VSIDs
patch, and is not used in the current kernel code. This removes it.
Signed-off-by: Paul Mackerras <paulus@samba.org>
On ppc64, we independently define VMALLOCBASE and VMALLOC_START to be
the same thing: the start of the vmalloc() area at 0xd000000000000000.
VMALLOC_START is used much more widely, including in generic code, so
this patch gets rid of the extraneous VMALLOCBASE.
This does require moving the definitions of region IDs from page_64.h
to pgtable.h, but they don't clearly belong in the former rather than
the latter, anyway. While we're moving them, clean up the definitions
of the REGION_IDs:
- Abolish REGION_SIZE, it was only used once, to define
REGION_MASK anyway
- Define the specific region ids in terms of the REGION_ID()
macro.
- Define KERNEL_REGION_ID in terms of PAGE_OFFSET rather than
KERNELBASE. It amounts to the same thing, but conceptually this is
about the region of the linear mapping (which starts at PAGE_OFFSET)
rather than of the kernel text itself (which is at KERNELBASE).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This patch separates usage of KERNELBASE and PAGE_OFFSET. I haven't
looked at any of the PPC32 code, if we ever want to support Kdump on
PPC we'll have to do another audit, ditto for iSeries.
This patch makes PAGE_OFFSET the constant, it'll always be 0xC * 1
gazillion for 64-bit.
To get a physical address from a virtual one you subtract PAGE_OFFSET,
_not_ KERNELBASE.
KERNELBASE is the virtual address of the start of the kernel, it's
often the same as PAGE_OFFSET, but _might not be_.
If you want to know something's offset from the start of the kernel
you should subtract KERNELBASE.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
There's a bunch of code that compares an address with KERNELBASE to see if
it's a "kernel address", ie. >= KERNELBASE. The proper test is actually to
compare with PAGE_OFFSET, since we're going to change KERNELBASE soon.
So replace all of them with an is_kernel_addr() macro that does that.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Adds a new CONFIG_PPC_64K_PAGES which, when enabled, changes the kernel
base page size to 64K. The resulting kernel still boots on any
hardware. On current machines with 4K pages support only, the kernel
will maintain 16 "subpages" for each 64K page transparently.
Note that while real 64K capable HW has been tested, the current patch
will not enable it yet as such hardware is not released yet, and I'm
still verifying with the firmware architects the proper to get the
information from the newer hypervisors.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This moves the remaining files in arch/ppc64/mm to arch/powerpc/mm,
and arranges that we use them when compiling with ARCH=ppc64.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Currently, we set the class bit in kernel SLB entries, and clear it on
user SLB entries. On POWER5, ERAT entries created in real mode have
the class bit clear. So to avoid flushing kernel ERAT entries on each
context switch, this patch inverts our usage of the class bit, setting
it on user SLB entries and clearing it on kernel SLB entries.
Booted on POWER5 and G5.
Signed-off-by: David Gibson <dwg@au1.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Remove vsid argument to create_slbe, since it's no longer used.
Spotted by R Sharada.
Signed-off-by: Olof Johansson <olof@austin.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
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!
Benjamin Herrenschmidt