When PCI Device pass-through is enabled via VFIO, KVM-PPC will
pin pages using get_user_pages_fast(). One of the downsides of
the pinning is that the page could be in CMA region. The CMA
region is used for other allocations like the hash page table.
Ideally we want the pinned pages to be from non CMA region.

This patch (currently only for KVM PPC with VFIO) forcefully
migrates the pages out (huge pages are omitted for the moment).
There are more efficient ways of doing this, but that might
be elaborate and might impact a larger audience beyond just
the kvm ppc implementation.

The magic is in new_iommu_non_cma_page() which allocates the
new page from a non CMA region.

I've tested the patches lightly at my end. The full solution
requires migration of THP pages in the CMA region. That work
will be done incrementally on top of this.
Signed-off-by: NBalbir Singh <bsingharora@gmail.com>
Acked-by: NAlexey Kardashevskiy <aik@ozlabs.ru>
[mpe: Merged via powerpc tree as that's where the changes are]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This supports PCI surprise hotplug. The design is highlighted as
below:

   * The PCI slot's surprise hotplug capability is exposed through
     device node property "ibm,slot-surprise-pluggable", meaning
     PCI surprise hotplug will be disabled if skiboot doesn't support
     it yet.
   * The interrupt because of presence or link state change is raised
     on surprise hotplug event. One event is allocated and queued to
     the PCI slot for workqueue to pick it up and process in serialized
     fashion. The code flow for surprise hotplug is same to that for
     managed hotplug except: the affected PEs are put into frozen state
     to avoid unexpected EEH error reporting in surprise hot remove path.
Signed-off-by: NGavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This removes likely() and unlikely() in pnv_php.c as the code isn't
running in hot path. Those macros to affect CPU's branch stream don't
help a lot for performance. I used them to identify the cases are
likely or unlikely to happen. No logical changes introduced.
Signed-off-by: NGavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This unfreezes PE when it's initialized because the PE might be put
into frozen state in the last hot remove path. It's not harmful to
do so if the PE is already in unfrozen state.
Signed-off-by: NGavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This exports eeh_pe_state_mark(). It will be used to mark the surprise
hot removed PE as isolated to avoid unexpected EEH error reporting in
surprise remove path.
Signed-off-by: NGavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This exports @confirm_error_lock so that eeh_serialize_{lock, unlock}()
can be used to freeze the affected PE in PCI surprise hot remove path.
Signed-off-by: NGavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Function eeh_pe_set_option() is used to apply the requested options
(enable, disable, unfreeze) in EEH virtualization path. The semantics
of this function isn't complete until freezing is supported.

This allows to freeze the indicated PE. The new semantics is going to
be used in PCI surprise hot remove path, to freeze removed PCI devices
(PE) to avoid unexpected EEH error reporting.
Signed-off-by: NGavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When issuing PHB reset, OPAL API opal_pci_poll() is called to drive
the state machine in OPAL forward. However, we needn't always call
the function under some circumstances like reset deassert.

This avoids calling opal_pci_poll() when OPAL_SUCCESS is returned
from opal_pci_reset(). Except the overhead introduced by additional
one unnecessary OPAL call, I didn't run into real issue because of
this.
Reported-by: NPridhiviraj Paidipeddi <ppaiddipe@in.ibm.com>
Signed-off-by: NGavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This patch adds an option to use XZ compression for the kernel image.

Currently this is only enabled for 64-bit Book3S targets, which is
roughly equivalent to the platforms that use the kernel's zImage
wrapper, and that have been tested.

The bulk of the 32-bit platforms and 64-bit BookE use uboot images,
which relies on uboot implementing XZ. In future we can enable XZ
support for those targets once someone has tested it.
Signed-off-by: NOliver O'Halloran <oohall@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This modifies the wrapper script so that the -Z option takes an argument
to specify the compression type. It can either be 'gz', 'xz' or 'none'.

The legazy --no-gzip and -z options are still supported and will set the
compression to none and gzip respectively, but they are not documented.

Only XZ -6 is used for compression rather than XZ -9. Using compression
levels higher than 6 requires the decompressor to build a large (64MB)
dictionary when decompressing and some environments cannot satisfy such
large allocations (e.g. POWER 6 LPAR partition firmware).
Signed-off-by: NOliver O'Halloran <oohall@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This code is no longer used and can be removed.
Signed-off-by: NOliver O'Halloran <oohall@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Currently the powerpc boot wrapper has its own wrapper around zlib to
handle decompressing gzipped kernels. The kernel decompressor library
functions now provide a generic interface that can be used in the
pre-boot environment. This allows boot wrappers to easily support
different compression algorithms. This patch converts the wrapper to use
this new API, but does not add support for using new algorithms.
Signed-off-by: NOliver O'Halloran <oohall@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Most architectures allow the compression algorithm used to produced the
vmlinuz image to be selected as a kernel config option. In preperation
for supporting algorithms other than gzip in the powerpc boot wrapper
the makefile needs to be modified to use these config options.
Signed-off-by: NOliver O'Halloran <oohall@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The powerpc boot wrapper is potentially compiled with a separate
toolchain and/or toolchain flags than the rest of the kernel. The usual
case is a 64-bit big endian kernel builds a 32-bit big endian wrapper.

The main problem with this is that the wrapper does not have access to
the kernel headers (without a lot of gross hacks). To get around this
the required headers are copied into the build directory via several sed
scripts which rewrite problematic includes. This patch moves these
fixups out of the makefile into a separate .sed script file to clean up
makefile slightly.
Signed-off-by: NOliver O'Halloran <oohall@gmail.com>
[mpe: Reword first paragraph of change log a little]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

It might be nice to compile selftests against older kernels and
headers but which may not have HWCAP2.
Signed-off-by: NCyril Bur <cyrilbur@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The same logic appears twice and should probably be pulled out into a
function.
Suggested-by: NMichael Ellerman <mpe@ellerman.id.au>
Signed-off-by: NRui Teng <rui.teng@linux.vnet.ibm.com>
[mpe: Rename to tm_flush_hash_page() and move comment into the function]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Signed-off-by: NAndrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

CLR_TOP32() is defined as blank. Last useful instance of CLR_TOP32()
was removed by commit 40ef8cbc ("powerpc: Get 64-bit configs to
compile with ARCH=powerpc") in 2005.
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

On some CPUs like the 8xx, _PAGE_RW hence _PAGE_WRITE is defined
as 0 and _PAGE_RO has to be set when a page is not writable

_PAGE_RO is defined by default in pte-common.h, however BOOK3S/64
doesn't include that file so _PAGE_RO has to be defined explicitly
in book3s/64/pgtable.h

Fixes: a7b9f671 ("powerpc32: adds handling of _PAGE_RO")
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

In eeh_handle_special_event(), eeh_pe_bus_get() is called before calling
eeh_report_failure() on every device under a PE.  If a PE was missing a
bus for some reason, the error would occur before reporting failure, even
though eeh_report_failure() doesn't require a bus.

Fix this by moving the bus retrieval and error check after the
eeh_report_failure() calls.
Signed-off-by: NRussell Currey <ruscur@russell.cc>
Reviewed-by: NAndrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When the PE used in pnv_eeh_reset() is that of a VF,
pnv_eeh_reset_vf_pe() is used.  Unlike the other reset functions called
in pnv_eeh_reset(), the VF reset doesn't require a bus, and if a bus was
missing the function would error out before resetting the VF PE.

To avoid this, reorder the VF reset function to occur before finding and
checking the bus.
Signed-off-by: NRussell Currey <ruscur@russell.cc>
Reviewed-by: NAndrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

eeh_pe_bus_get() can return NULL if a PCI bus isn't found for a given PE.
Some callers don't check this, and can cause a null pointer dereference
under certain circumstances.

Fix this by checking NULL everywhere eeh_pe_bus_get() is called.

Fixes: 8a6b1bc7 ("powerpc/eeh: EEH core to handle special event")
Cc: stable@vger.kernel.org # v3.11+
Signed-off-by: NRussell Currey <ruscur@russell.cc>
Reviewed-by: NAndrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When we originally added the ability to split the exception vectors from
the kernel (commit 1f6a93e4 ("powerpc: Make it possible to move the
interrupt handlers away from the kernel" 2008-09-15)), the LOAD_HANDLER() macro
used an addi instruction to compute the offset of the common handler
from the kernel base address.

Using addi meant the handler had to be within 32K of the kernel base
address, due to the addi instruction taking a signed immediate value.
That necessitated creating a trampoline for the system call handler,
because system_call_common (in entry64.S) is not linked within 32K of
the kernel base address.

Later in commit 61e2390e ("powerpc: Make load_hander handle upto 64k
offset" 2012-11-15) we changed LOAD_HANDLER to take a 64K offset, by
changing it to use ori.

Although system_call_common is not in head_64.S or exceptions-64s.S, it
is included in head-y, which causes it to be linked early in the kernel
text, so in practice it ends up below 64K. Additionally if it can't be
placed below 64K the linker will fail to build with a "relocation
truncated to fit" error.

So remove the trampoline.

Newer toolchains are able to work out that the ori in LOAD_HANDLER only
takes a 16 bit offset, and so they generate a 16 bit relocation. Older
toolchains (binutils 2.22 at least) are not so smart, so we have to add
the @l annotation to tell the assembler to generate a 16 bit relocation.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The 0xf80 hv_facility_unavailable trampoline branches to the 0xf60
handler. This works because they both do the same thing, but it should
be fixed.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

On EEH events the kernel will print a dump of relevant registers.
If EEH is unavailable (i.e. CONFIG_EEH is disabled, a new platform
doesn't have EEH support, etc) this information isn't readily available.

Add a new debugfs handler to trigger a PHB register dump, so that this
information can be made available on demand.
Signed-off-by: NRussell Currey <ruscur@russell.cc>
Reviewed-by: NGavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The only difference is now the TCE table check which doesn't need
to be ifdef'ed out, it will basically do nothing on BookE (it is
only useful for ancient IBM machines).
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Currently we turn the MMU off after copying the image, and we make
sure there is no overlap between the hash table and the target pages
in that case.

That doesn't work for Radix however. In that case, the page tables
are scattered and we can't really enforce that the target of the
image isn't overlapping one of them.

So instead, let's turn the MMU off before copying the image in radix
mode. Thankfully, in radix mode, even under a hypervisor, we know we
don't have the same kind of RMA limitations that hash mode has.

While at it, also turn the MMU off early when using hash in non-LPAR
mode, that way we can get rid of the collision check completely.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NBalbir Singh <bsingharora@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Just using the hash ops won't work anymore since radix will have
NULL in there. Instead create an mmu_cleanup_all() function which
will do the right thing based on the MMU mode.

For Radix, for now I clear UPRT and the PTCR, effectively switching
back to Radix with no partition table setup.

Currently set it to NULL on BookE thought it might be a good idea
to wipe the TLB there (Scott ?)
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NBalbir Singh <bsingharora@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

With Radix, it can be NULL even on !BOOKE these days so replace
the ifdef with a NULL check which is cleaner anyway.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NBalbir Singh <bsingharora@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

NO_IRQ has been == 0 on powerpc for just over ten years (since commit
0ebfff14 ("[POWERPC] Add new interrupt mapping core and change
platforms to use it")). It's also 0 on most other arches.

Although it's fairly harmless, every now and then it causes confusion
when a driver is built on powerpc and another arch which doesn't define
NO_IRQ. There's at least 6 definitions of NO_IRQ in drivers/, at least
some of which are to work around that problem.

So we'd like to remove it. This is fairly trivial in the arch code, we
just convert:

    if (irq == NO_IRQ)	to	if (!irq)
    if (irq != NO_IRQ)	to	if (irq)
    irq = NO_IRQ;	to	irq = 0;
    return NO_IRQ;	to	return 0;

And a few other odd cases as well.

At least for now we keep the #define NO_IRQ, because there is driver
code that uses NO_IRQ and the fixes to remove those will go via other
trees.

Note we also change some occurrences in PPC sound drivers, drivers/ps3,
and drivers/macintosh.
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Fred has taken over the cxl maintenance I was doing.  This updates the
MAINTAINERS file to reflect this.

It also removes a duplicate entry in the files covered and adds an entry
for the CXL PCI code in arch/powerpc.
Signed-off-by: NMichael Neuling <mikey@neuling.org>
Reviewed-by: NAndrew Donnellan <andrew.donnellan@au1.ibm.com>
Reviewed-by: NAndrew Donnellan <andrew.donnellan@au1.ibm.com>
Acked-by: NFrederic Barrat <fbarrat@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

If we fail to allocate work, we don't end up using hp_errlog_copy. Free it
in the error path.
Signed-off-by: NAndrew Donnellan <andrew.donnellan@au1.ibm.com>
Reviewed-by: NNathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When we merge two contiguous partitions whose signatures are marked
NVRAM_SIG_FREE, We need update prev's length and checksum, then write it
to nvram, not cur's. So lets fix this mistake now.

Also use memset instead of strncpy to set the partition's name. It's
more readable if we want to fill up with duplicate chars .

Fixes: fa2b4e54 ("powerpc/nvram: Improve partition removal")
Signed-off-by: NPan Xinhui <xinhui.pan@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

If kmemdup fails, We need kfree *buff* first then return -ENOMEM.
Otherwise there is a memory leak.
Signed-off-by: NPan Xinhui <xinhui.pan@linux.vnet.ibm.com>
Reviewed-by: NNathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

mtmsrd with L=1 only affects MSR_EE and MSR_RI bits, and we always
know what state those bits are, so the kernel MSR does not need to be
loaded when modifying them.

mtmsrd is often in the critical execution path, so avoiding dependency
on even L1 load is noticable. On a POWER8 this saves about 3 cycles
from the syscall path, and possibly a few from other exception returns
(not measured).
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The mflr r10 instruction was left over from when the code used LR to
branch to system_call_entry from the exception handler. That was
changed by commit 6a404806 ("powerpc: Avoid link stack corruption in
MMU on syscall entry path") to use the count register. The value is
never used now, so mflr can be removed, and r10 can be used for storage
rather than spilling to the SPR scratch register.

The scratch register spill causes a long pipeline stall due to the SPR
read after write. This change brings getppid syscall cost from 406 to
376 cycles on POWER8. getppid for non-relocatable case is 371 cycles.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Acked-by: NBalbir Singh <bsingharora@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

For hugetlb to work with 4K page size, we need MAX_ORDER to be 13 or
more. When switching from a 64K page size to 4K linux page size using
make oldconfig, we end up with a CONFIG_FORCE_MAX_ZONEORDER value of 9.
This results in a 16M hugepage beiing considered as a gigantic huge page
which in turn results in failure to setup hugepages if gigantic hugepage
support is not enabled.

This also results in kernel crash with 4K radix configuration. We
hit the below BUG_ON on radix:

  kernel BUG at mm/huge_memory.c:364!
  Oops: Exception in kernel mode, sig: 5 [#1]
  SMP NR_CPUS=2048 NUMA PowerNV
  CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.8.0-rc1-00006-gbae9cc6 #1
  task: c0000000f1af8000 task.stack: c0000000f1aec000
  NIP: c000000000c5fa0c LR: c000000000c5f9d8 CTR: c000000000c5f9a4
  REGS: c0000000f1aef920 TRAP: 0700   Not tainted (4.8.0-rc1-00006-gbae9cc6)
  MSR: 9000000102029033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE,TM[E]>  CR: 24000844  XER: 00000000
  CFAR: c000000000c5f9e0 SOFTE: 1
  ....
  NIP [c000000000c5fa0c] hugepage_init+0x68/0x238
  LR [c000000000c5f9d8] hugepage_init+0x34/0x238

Fixes: a7ee5395 ("powerpc/Kconfig: Update config option based on page size")
Cc: stable@vger.kernel.org # v4.7+
Reported-by: NSanthosh <santhog4@linux.vnet.ibm.com>
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NBalbir Singh <bsingharora@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Add some documentation for the 64-bit syscall ABI, which doesn't seem
to be documented elsewhere.

This attempts to document existing practice. The only small discrepancy
is glibc clobbers not quite matching the kernel (e.g., xer, some
vsyscalls trash cr1 whereas glibc only clobbers cr0). These will be
resolved after this document is merged.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The HMI (Hypervisor Maintenance Interrupt) is defined by the
architecture to be higher priority than other maskable interrupts, so
replay it first, as a best-effort to replay according to hardware
priorities.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>