platform/kernel/linux-rpi.git
10 years agopowerpc/eeh: Call opal_pci_reinit() on powernv for restoring config space
Gavin Shan [Fri, 3 Jan 2014 09:47:13 +0000 (17:47 +0800)]
powerpc/eeh: Call opal_pci_reinit() on powernv for restoring config space

The patch implements the EEH operation backend restore_config()
for PowerNV platform. That relies on OPAL API opal_pci_reinit()
where we reinitialize the error reporting properly after PE or
PHB reset.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/eeh: Add restore_config operation
Gavin Shan [Fri, 3 Jan 2014 09:47:12 +0000 (17:47 +0800)]
powerpc/eeh: Add restore_config operation

After reset on the specific PE or PHB, we never configure AER
correctly on PowerNV platform. We needn't care it on pSeries
platform. The patch introduces additional EEH operation eeh_ops::
restore_config() so that we have chance to configure AER correctly
for PowerNV platform.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Remove unnecessary assignment
Gavin Shan [Thu, 26 Dec 2013 01:29:40 +0000 (09:29 +0800)]
powerpc/powernv: Remove unnecessary assignment

We don't have IO ports on PHB3 and the assignment of variable
"iomap_off" on PHB3 is meaningless. The patch just removes the
unnecessary assignment to the variable. The code change should
have been part of commit c35d2a8c ("powerpc/powernv: Needn't IO
segment map for PHB3").

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agoRevert "pseries/iommu: Remove DDW on kexec"
Nishanth Aravamudan [Fri, 10 Jan 2014 23:10:41 +0000 (15:10 -0800)]
Revert "pseries/iommu: Remove DDW on kexec"

After reverting 25ebc45b93452d0bc60271f178237123c4b26808
("powerpc/pseries/iommu: remove default window before attempting DDW
manipulation"), we no longer remove the base window in enable_ddw.
Therefore, we no longer need to reset the DMA window state in
find_existing_ddw_windows(). We can instead go back to what was done
before, which simply reuses the previous configuration, if any. Further,
this removes the final caller of the reset-pe-dma-windows call, so
remove those functions.

This fixes an EEH on kdump with the ipr driver. The EEH occurs, because
the initcall removes the DDW configuration (64-bit DMA window), but
doesn't ensure the ops are via the IOMMU -- a DMA operation occurs
during probe (still investigating this) and we EEH.

This reverts commit 14b6f00f8a4fdec5ccd45a0710284de301a61628.

Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agoRevert "powerpc/pseries/iommu: remove default window before attempting DDW manipulation"
Nishanth Aravamudan [Fri, 10 Jan 2014 23:09:38 +0000 (15:09 -0800)]
Revert "powerpc/pseries/iommu: remove default window before attempting DDW manipulation"

Ben rightfully pointed out that there is a race in the "newer" DDW code.
Presuming we are running on recent enough firmware that supports the
"reset" DDW manipulation call, we currently always remove the base
32-bit DMA window in order to maximize the resources for Phyp when
creating the 64-bit window. However, this can be problematic for the
case where multiple functions are in the same PE (partitionable
endpoint), where some funtions might be 32-bit DMA only. All of a
sudden, the only functional DMA window for such functions is gone. We
will have serious errors in such situations. The best solution is simply
to revert the extension to the DDW code where we ever remove the base
DMA window.

This reverts commit 25ebc45b93452d0bc60271f178237123c4b26808.

Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Delete non-required instances of include <linux/init.h>
Paul Gortmaker [Thu, 9 Jan 2014 05:44:29 +0000 (00:44 -0500)]
powerpc: Delete non-required instances of include <linux/init.h>

None of these files are actually using any __init type directives
and hence don't need to include <linux/init.h>.  Most are just a
left over from __devinit and __cpuinit removal, or simply due to
code getting copied from one driver to the next.

The one instance where we add an include for init.h covers off
a case where that file was implicitly getting it from another
header which itself didn't need it.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Add vr save/restore functions
Andreas Schwab [Mon, 30 Dec 2013 14:31:17 +0000 (15:31 +0100)]
powerpc: Add vr save/restore functions

GCC 4.8 now generates out-of-line vr save/restore functions when
optimizing for size.  They are needed for the raid6 altivec support.

Signed-off-by: Andreas Schwab <schwab@linux-m68k.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agoMerge branch 'merge' into next
Benjamin Herrenschmidt [Mon, 30 Dec 2013 04:19:31 +0000 (15:19 +1100)]
Merge branch 'merge' into next

Merge a pile of fixes that went into the "merge" branch (3.13-rc's) such
as Anton Little Endian fixes.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix endian issues in power7/8 machine check handler
Anton Blanchard [Sun, 15 Dec 2013 23:47:54 +0000 (10:47 +1100)]
powerpc: Fix endian issues in power7/8 machine check handler

The SLB save area is shared with the hypervisor and is defined
as big endian, so we need to byte swap on little endian builds.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agoMerge remote-tracking branch 'agust/merge' into merge
Benjamin Herrenschmidt [Mon, 30 Dec 2013 03:48:27 +0000 (14:48 +1100)]
Merge remote-tracking branch 'agust/merge' into merge

Anatolij writes:

Please pull two DTS fixes for MPC5125 tower board. Without
them the v3.13-rcX kernels do not boot.

10 years agopowerpc/iommu: Update the generic code to use dynamic iommu page sizes
Alistair Popple [Mon, 9 Dec 2013 07:17:03 +0000 (18:17 +1100)]
powerpc/iommu: Update the generic code to use dynamic iommu page sizes

This patch updates the generic iommu backend code to use the
it_page_shift field to determine the iommu page size instead of
using hardcoded values.

Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/iommu: Add it_page_shift field to determine iommu page size
Alistair Popple [Mon, 9 Dec 2013 07:17:02 +0000 (18:17 +1100)]
powerpc/iommu: Add it_page_shift field to determine iommu page size

This patch adds a it_page_shift field to struct iommu_table and
initiliases it to 4K for all platforms.

Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/iommu: Update constant names to reflect their hardcoded page size
Alistair Popple [Mon, 9 Dec 2013 07:17:01 +0000 (18:17 +1100)]
powerpc/iommu: Update constant names to reflect their hardcoded page size

The powerpc iommu uses a hardcoded page size of 4K. This patch changes
the name of the IOMMU_PAGE_* macros to reflect the hardcoded values. A
future patch will use the existing names to support dynamic page
sizes.

Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Remove unused REDBOOT Kconfig parameter
Michael Opdenacker [Mon, 9 Dec 2013 05:27:40 +0000 (06:27 +0100)]
powerpc: Remove unused REDBOOT Kconfig parameter

This removes the REDBOOT Kconfig parameter,
which was no longer used anywhere in the source code
and Makefiles.

Signed-off-by: Michael Opdenacker <michael.opdenacker@free-electrons.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix "attempt to move .org backwards" error
Mahesh Salgaonkar [Mon, 9 Dec 2013 19:10:15 +0000 (00:40 +0530)]
powerpc: Fix "attempt to move .org backwards" error

With recent machine check patch series changes, The exception vectors
starting from 0x4300 are now overflowing with allyesconfig. Fix that by
moving machine_check_common and machine_check_handle_early code out of
that region to make enough room for exception vector area.

Fixes this build error reportes by Stephen:

arch/powerpc/kernel/exceptions-64s.S: Assembler messages:
arch/powerpc/kernel/exceptions-64s.S:958: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:959: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:983: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:984: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:1003: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:1013: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:1014: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:1015: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:1016: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:1017: Error: attempt to move .org backwards
arch/powerpc/kernel/exceptions-64s.S:1018: Error: attempt to move .org backwards

[Moved the code further down as it introduced link errors due to too long
 relative branches to the masked interrupts handlers from the exception
 prologs. Also removed the useless feature section --BenH
]

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Tested-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix alignment of secondary cpu spin vars
Olof Johansson [Sat, 28 Dec 2013 21:01:47 +0000 (13:01 -0800)]
powerpc: Fix alignment of secondary cpu spin vars

Commit 5c0484e25ec0 ('powerpc: Endian safe trampoline') resulted in
losing proper alignment of the spinlock variables used when booting
secondary CPUs, causing some quite odd issues with failing to boot on
PA Semi-based systems.

This showed itself on ppc64_defconfig, but not on pasemi_defconfig,
so it had gone unnoticed when I initially tested the LE patch set.

Fix is to add explicit alignment instead of relying on good luck. :)

[ It appears that there is a different issue with PA Semi systems
  however this fix is definitely correct so applying anyway -- BenH
]

Fixes: 5c0484e25ec0 ('powerpc: Endian safe trampoline')
Reported-by: Christian Zigotzky <chzigotzky@xenosoft.de>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=67811
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Align p_end
Anton Blanchard [Mon, 23 Dec 2013 01:19:51 +0000 (12:19 +1100)]
powerpc: Align p_end

p_end is an 8 byte value embedded in the text section. This means it
is only 4 byte aligned when it should be 8 byte aligned. Fix this
by adding an explicit alignment.

This fixes an issue where POWER7 little endian builds with
CONFIG_RELOCATABLE=y fail to boot.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowernv/eeh: Add buffer for P7IOC hub error data
Brian W Hart [Fri, 20 Dec 2013 19:06:01 +0000 (13:06 -0600)]
powernv/eeh: Add buffer for P7IOC hub error data

Prevent ioda_eeh_hub_diag() from clobbering itself when called by supplying
a per-PHB buffer for P7IOC hub diagnostic data.  Take care to inform OPAL of
the correct size for the buffer.

[Small style change to the use of sizeof -- BenH]

Signed-off-by: Brian W Hart <hartb@linux.vnet.ibm.com>
Acked-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowernv/eeh: Fix possible buffer overrun in ioda_eeh_phb_diag()
Brian W Hart [Thu, 19 Dec 2013 23:14:07 +0000 (17:14 -0600)]
powernv/eeh: Fix possible buffer overrun in ioda_eeh_phb_diag()

PHB diagnostic buffer may be smaller than PAGE_SIZE, especially when
PAGE_SIZE > 4KB.

Signed-off-by: Brian W Hart <hartb@linux.vnet.ibm.com>
Acked-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Make 64-bit non-VMX __copy_tofrom_user bi-endian
Paul E. McKenney [Tue, 17 Dec 2013 22:29:57 +0000 (09:29 +1100)]
powerpc: Make 64-bit non-VMX __copy_tofrom_user bi-endian

The powerpc 64-bit __copy_tofrom_user() function uses shifts to handle
unaligned invocations.  However, these shifts were designed for
big-endian systems: On little-endian systems, they must shift in the
opposite direction.

This commit relies on the C preprocessor to insert the correct shifts
into the assembly code.

[ This is a rare but nasty LE issue. Most of the time we use the POWER7
optimised __copy_tofrom_user_power7 loop, but when it hits an exception
we fall back to the base __copy_tofrom_user loop. - Anton ]

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Make unaligned accesses endian-safe for powerpc
Rajesh B Prathipati [Mon, 16 Dec 2013 07:58:22 +0000 (18:58 +1100)]
powerpc: Make unaligned accesses endian-safe for powerpc

The generic put_unaligned/get_unaligned macros were made endian-safe by
calling the appropriate endian dependent macros based on the endian type
of the powerpc processor.

Signed-off-by: Rajesh B Prathipati <rprathip@linux.vnet.ibm.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix bad stack check in exception entry
Michael Neuling [Mon, 16 Dec 2013 04:12:43 +0000 (15:12 +1100)]
powerpc: Fix bad stack check in exception entry

In EXCEPTION_PROLOG_COMMON() we check to see if the stack pointer (r1)
is valid when coming from the kernel.  If it's not valid, we die but
with a nice oops message.

Currently we allocate a stack frame (subtract INT_FRAME_SIZE) before we
check to see if the stack pointer is negative.  Unfortunately, this
won't detect a bad stack where r1 is less than INT_FRAME_SIZE.

This patch fixes the check to compare the modified r1 with
-INT_FRAME_SIZE.  With this, bad kernel stack pointers (including NULL
pointers) are correctly detected again.

Kudos to Paulus for finding this.

Signed-off-by: Michael Neuling <mikey@neuling.org>
cc: stable@vger.kernel.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/512x: dts: disable MPC5125 usb module
Matteo Facchinetti [Fri, 20 Dec 2013 09:16:22 +0000 (10:16 +0100)]
powerpc/512x: dts: disable MPC5125 usb module

At the moment the USB controller's pin muxing is not setup
correctly and causes a kernel panic upon system startup, so
disable the USB1 device tree node in the MPC5125 tower board
dts file.

The USB controller is connected to an USB3320 ULPI transceiver
and the device tree should receive an update to reflect correct
dependencies and required initialization data before the USB1
node can get re-enabled.

Signed-off-by: Matteo Facchinetti <matteo.facchinetti@sirius-es.it>
Signed-off-by: Anatolij Gustschin <agust@denx.de>
10 years agopowerpc/512x: dts: remove misplaced IRQ spec from 'soc' node (5125)
Gerhard Sittig [Tue, 10 Dec 2013 09:51:08 +0000 (10:51 +0100)]
powerpc/512x: dts: remove misplaced IRQ spec from 'soc' node (5125)

the 'soc' node in the MPC5125 "tower" board .dts has an '#interrupt-cells'
property although this node is not an interrupt controller

remove this erroneously placed property because starting with v3.13-rc1
lookup and resolution of 'interrupts' specs for peripherals gets misled
(tries to use the 'soc' as the interrupt parent which fails), emits
'no irq domain found' WARN() messages and breaks the boot process

[ best viewed with 'git diff -U5' to have DT node names in the context ]

Cc: Anatolij Gustschin <agust@denx.de>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: devicetree@vger.kernel.org
Signed-off-by: Gerhard Sittig <gsi@denx.de>
Signed-off-by: Anatolij Gustschin <agust@denx.de>
10 years agopowerpc/powernv: Fix OPAL LPC access in Little Endian
Benjamin Herrenschmidt [Fri, 13 Dec 2013 04:56:06 +0000 (15:56 +1100)]
powerpc/powernv: Fix OPAL LPC access in Little Endian

We are passing pointers to the firmware for reads, we need to properly
convert the result as OPAL is always BE.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Fix endian issue in opal_xscom_read
Anton Blanchard [Fri, 13 Dec 2013 04:53:43 +0000 (15:53 +1100)]
powerpc/powernv: Fix endian issue in opal_xscom_read

opal_xscom_read uses a pointer to return the data so we need
to byteswap it on LE builds.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix endian issues in crash dump code
Anton Blanchard [Thu, 12 Dec 2013 04:59:41 +0000 (15:59 +1100)]
powerpc: Fix endian issues in crash dump code

A couple more device tree properties that need byte swapping.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/pseries: Fix endian issues in MSI code
Anton Blanchard [Thu, 12 Dec 2013 04:59:40 +0000 (15:59 +1100)]
powerpc/pseries: Fix endian issues in MSI code

The MSI code is miscalculating quotas in little endian mode.
Add required byteswaps to fix this.

Before we claimed a quota of 65536, after the patch we
see the correct value of 256.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/pseries: Fix PCIE link speed endian issue
Anton Blanchard [Thu, 12 Dec 2013 04:59:39 +0000 (15:59 +1100)]
powerpc/pseries: Fix PCIE link speed endian issue

We need to byteswap ibm,pcie-link-speed-stats.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/pseries: Fix endian issues in nvram code
Anton Blanchard [Thu, 12 Dec 2013 04:59:38 +0000 (15:59 +1100)]
powerpc/pseries: Fix endian issues in nvram code

The NVRAM code has a number of endian issues. I noticed a very
confused error log count:

RTAS: 100663330 -------- RTAS event begin --------

100663330 == 0x06000022. 0x6 LE error logs and 0x22 BE error logs.

The pstore code has similar issues - if we write an oops in one
endian and attempt to read it in another we get junk.

Make both of these formats big endian, and byteswap as required.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/pseries: Fix endian issues in /proc/ppc64/lparcfg
Anton Blanchard [Thu, 12 Dec 2013 04:59:37 +0000 (15:59 +1100)]
powerpc/pseries: Fix endian issues in /proc/ppc64/lparcfg

Some obvious issues:

cat /proc/ppc64/lparcfg
...
partition_id=16777216
...
partition_potential_processors=268435456

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix topology core_id endian issue on LE builds
Anton Blanchard [Thu, 12 Dec 2013 04:59:36 +0000 (15:59 +1100)]
powerpc: Fix topology core_id endian issue on LE builds

cpu_to_core_id() is missing a byteswap:

cat /sys/devices/system/cpu/cpu63/topology/core_id
201326592

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix endian issue in setup-common.c
Anton Blanchard [Thu, 12 Dec 2013 04:59:35 +0000 (15:59 +1100)]
powerpc: Fix endian issue in setup-common.c

During on LE boot we see:

    Partition configured for 1073741824 cpus, operating system maximum is 2048.

Clearly missing a byteswap here.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: PTRACE_PEEKUSR always returns FPR0
Ulrich Weigand [Thu, 12 Dec 2013 04:59:34 +0000 (15:59 +1100)]
powerpc: PTRACE_PEEKUSR always returns FPR0

There is a bug in using ptrace to access FPRs via PTRACE_PEEKUSR /
PTRACE_POKEUSR. In effect, trying to access any of the FPRs always
really accesses FPR0, which does seriously break debugging :-)

The problem seems to have been introduced by commit 3ad26e5c4459d
(Merge branch 'for-kvm' into next).

[ It is indeed a merge conflict between Paul's FPU/VSX state rework
and my LE patches - Anton ]

Signed-off-by: Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix up the kdump base cap to 128M
Mahesh Salgaonkar [Mon, 9 Dec 2013 10:03:39 +0000 (15:33 +0530)]
powerpc: Fix up the kdump base cap to 128M

The current logic sets the kdump base to min of 2G or ppc64_rma_size/2.
On PowerNV kernel the first memory block 'memory@0' can be very large,
equal to the DIMM size with ppc64_rma_size value capped to 1G. Hence on
PowerNV, kdump base is set to 512M resulting kdump to fail while allocating
paca array. This is because, paca need its memory from RMA region capped
at 256M (see allocate_pacas()).

This patch lowers the kdump base cap to 128M so that kdump kernel can
successfully get memory below 256M for paca allocation.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowernv: Fix VFIO support with PHB3
Thadeu Lima de Souza Cascardo [Mon, 9 Dec 2013 16:41:01 +0000 (14:41 -0200)]
powernv: Fix VFIO support with PHB3

I have recently found out that no iommu_groups could be found under
/sys/ on a P8. That prevents PCI passthrough from working.

During my investigation, I found out there seems to be a missing
iommu_register_group for PHB3. The following patch seems to fix the
problem. After applying it, I see iommu_groups under
/sys/kernel/iommu_groups/, and can also bind vfio-pci to an adapter,
which gives me a device at /dev/vfio/.

Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/52xx: Re-enable bestcomm driver in defconfigs
Anatolij Gustschin [Mon, 9 Dec 2013 21:15:44 +0000 (22:15 +0100)]
powerpc/52xx: Re-enable bestcomm driver in defconfigs

The bestcomm driver has been moved to drivers/dma, so to select
this driver by default additionally CONFIG_DMADEVICES has to be
enabled. Currently it is not enabled in the config despite existing
CONFIG_PPC_BESTCOMM=y in the config files. Fix it.

Signed-off-by: Anatolij Gustschin <agust@denx.de>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/pasemi: Turn on devtmpfs in defconfig
Olof Johansson [Mon, 9 Dec 2013 23:40:20 +0000 (15:40 -0800)]
powerpc/pasemi: Turn on devtmpfs in defconfig

At least some distros expect it these days; turn it on. Also, random
churn from doing a savedefconfig for the first time in a year or so.

Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agooffb: Add palette hack for little endian
Cedric Le Goater [Wed, 4 Dec 2013 16:49:52 +0000 (17:49 +0100)]
offb: Add palette hack for little endian

The pseudo palette color entries need to be ajusted for little
endian.

This patch byteswaps the values in the pseudo palette depending
on the host endian order and the screen depth.

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agooffb: Little endian fixes
Cedric Le Goater [Wed, 4 Dec 2013 16:49:51 +0000 (17:49 +0100)]
offb: Little endian fixes

The "screen" properties : depth, width, height, linebytes need
to be converted to the host endian order when read from the device
tree.

The offb_init_palette_hacks() routine also made assumption on the
host endian order.

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix PTE page address mismatch in pgtable ctor/dtor
Hong H. Pham [Sat, 7 Dec 2013 14:06:33 +0000 (09:06 -0500)]
powerpc: Fix PTE page address mismatch in pgtable ctor/dtor

In pte_alloc_one(), pgtable_page_ctor() is passed an address that has
not been converted by page_address() to the newly allocated PTE page.

When the PTE is freed, __pte_free_tlb() calls pgtable_page_dtor()
with an address to the PTE page that has been converted by page_address().
The mismatch in the PTE's page address causes pgtable_page_dtor() to access
invalid memory, so resources for that PTE (such as the page lock) is not
properly cleaned up.

On PPC32, only SMP kernels are affected.

On PPC64, only SMP kernels with 4K page size are affected.

This bug was introduced by commit d614bb041209fd7cb5e4b35e11a7b2f6ee8f62b8
"powerpc: Move the pte free routines from common header".

On a preempt-rt kernel, a spinlock is dynamically allocated for each
PTE in pgtable_page_ctor().  When the PTE is freed, calling
pgtable_page_dtor() with a mismatched page address causes a memory leak,
as the pointer to the PTE's spinlock is bogus.

On mainline, there isn't any immediately obvious symptoms, but the
problem still exists here.

Fixes: d614bb041209fd7c "powerpc: Move the pte free routes from common header"
Cc: Paul Mackerras <paulus@samba.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: linux-stable <stable@vger.kernel.org> # v3.10+
Signed-off-by: Hong H. Pham <hong.pham@windriver.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/44x: Fix ocm_block allocation
Ilia Mirkin [Sat, 7 Dec 2013 00:43:37 +0000 (19:43 -0500)]
powerpc/44x: Fix ocm_block allocation

Allocate enough memory for the ocm_block structure, not just a pointer
to it.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix build break with PPC_EARLY_DEBUG_BOOTX=y
Michael Ellerman [Fri, 29 Nov 2013 03:55:22 +0000 (14:55 +1100)]
powerpc: Fix build break with PPC_EARLY_DEBUG_BOOTX=y

A kernel configured with PPC_EARLY_DEBUG_BOOTX=y but PPC_PMAC=n and
PPC_MAPLE=n will fail to link:

  btext.c:(.text+0x2d0fc): undefined reference to `.rmci_off'
  btext.c:(.text+0x2d214): undefined reference to `.rmci_on'

Fix it by making the build of rmci_on/off() depend on
PPC_EARLY_DEBUG_BOOTX, which also enable the only code that uses them.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agoMerge remote-tracking branch 'agust/merge' into merge
Benjamin Herrenschmidt [Tue, 10 Dec 2013 00:24:10 +0000 (11:24 +1100)]
Merge remote-tracking branch 'agust/merge' into merge

Anatolij Gustschin says:

<<
Please pull a device tree fix for v3.13. The booting on mpc512x
is broken since v3.13-rc1, this patch repairs it.
>>

10 years agopowerpc/powernv: Get FSP memory errors and plumb into memory poison infrastructure.
Mahesh Salgaonkar [Fri, 15 Nov 2013 04:20:57 +0000 (09:50 +0530)]
powerpc/powernv: Get FSP memory errors and plumb into memory poison infrastructure.

Get the memory errors reported by opal and plumb it into memory poison
infrastructure. This patch uses new messaging channel infrastructure to
pull the fsp memory errors to linux.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Add config option for hwpoisoning.
Mahesh Salgaonkar [Fri, 15 Nov 2013 04:20:50 +0000 (09:50 +0530)]
powerpc/powernv: Add config option for hwpoisoning.

Add config option to enable generic memory hwpoisoning infrastructure for
ppc64.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/mm: Enable _PAGE_NUMA for book3s
Aneesh Kumar K.V [Mon, 18 Nov 2013 09:28:13 +0000 (14:58 +0530)]
powerpc/mm: Enable _PAGE_NUMA for book3s

We steal the _PAGE_COHERENCE bit and use that for indicating NUMA ptes.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/mm: Only check for _PAGE_PRESENT in set_pte/pmd functions
Aneesh Kumar K.V [Mon, 18 Nov 2013 09:28:12 +0000 (14:58 +0530)]
powerpc/mm: Only check for _PAGE_PRESENT in set_pte/pmd functions

We want to make sure we don't use these function when updating a pte
or pmd entry that have a valid hpte entry, because these functions
don't invalidate them. So limit the check to _PAGE_PRESENT bit.
Numafault core changes use these functions for updating _PAGE_NUMA bits.
That should be ok because when _PAGE_NUMA is set we can be sure that
hpte entries are not present.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/mm: Free up _PAGE_COHERENCE for numa fault use later
Aneesh Kumar K.V [Mon, 18 Nov 2013 09:28:10 +0000 (14:58 +0530)]
powerpc/mm: Free up _PAGE_COHERENCE for numa fault use later

Set  memory coherence always on hash64 config. If
a platform cannot have memory coherence always set they
can infer that from _PAGE_NO_CACHE and _PAGE_WRITETHRU
like in lpar. So we dont' really need a separate bit
for tracking _PAGE_COHERENCE.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/mm: Use HPTE constants when updating hpte bits
Aneesh Kumar K.V [Mon, 18 Nov 2013 09:28:09 +0000 (14:58 +0530)]
powerpc/mm: Use HPTE constants when updating hpte bits

Even though we have same value for linux PTE bits and hash PTE pits
use the hash pte bits wen updating hash pte

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Dynamically allocate slb_shadow from memblock
Jeremy Kerr [Thu, 5 Dec 2013 03:42:40 +0000 (11:42 +0800)]
powerpc: Dynamically allocate slb_shadow from memblock

Currently, the slb_shadow buffer is our largest symbol:

  [jk@pablo linux]$ nm --size-sort -r -S obj/vmlinux | head -1
  c000000000da0000 0000000000040000 d slb_shadow

- we allocate 128 bytes per cpu; so 256k with NR_CPUS=2048. As we have
constant initialisers, it's allocated in .text, causing a larger vmlinux
image. We may also allocate unecessary slb_shadow buffers (> no. pacas),
since we use the build-time NR_CPUS rather than the run-time nr_cpu_ids.

We could move this to the bss, but then we still have the NR_CPUS vs
nr_cpu_ids potential for overallocation.

This change dynamically allocates the slb_shadow array, during
initialise_pacas(). At a cost of 104 bytes of text, we save 256k of
data:

  [jk@pablo linux]$ size obj/vmlinux{.orig,}
     text     data      bss       dec     hex filename
  9202795  5244676  1169576  15617047  ee4c17 obj/vmlinux.orig
  9202899  4982532  1169576  15355007  ea4c7f obj/vmlinux

Tested on pseries.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Make slb_shadow a local
Jeremy Kerr [Thu, 5 Dec 2013 03:31:08 +0000 (11:31 +0800)]
powerpc: Make slb_shadow a local

The only external user of slb_shadow is the pseries lpar code, and it
can access through the paca array instead.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/pci: Use dev_is_pci() to check whether it is pci device
Yijing Wang [Thu, 5 Dec 2013 12:01:20 +0000 (20:01 +0800)]
powerpc/pci: Use dev_is_pci() to check whether it is pci device

Use PCI standard marco dev_is_pci() instead of directly compare
pci_bus_type to check whether it is pci device.

Signed-off-by: Yijing Wang <wangyijing@huawei.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agomm: Move change_prot_numa outside CONFIG_ARCH_USES_NUMA_PROT_NONE
Aneesh Kumar K.V [Thu, 5 Dec 2013 18:38:22 +0000 (00:08 +0530)]
mm: Move change_prot_numa outside CONFIG_ARCH_USES_NUMA_PROT_NONE

change_prot_numa should work even if _PAGE_NUMA != _PAGE_PROTNONE.
On archs like ppc64 that don't use _PAGE_PROTNONE and also have
a separate page table outside linux pagetable, we just need to
make sure that when calling change_prot_numa we flush the
hardware page table entry so that next page access  result in a numa
fault.

We still need to make sure we use the numa faulting logic only
when CONFIG_NUMA_BALANCING is set. This implies the migrate-on-fault
(Lazy migration) via mbind will only work if CONFIG_NUMA_BALANCING
is set.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/512x: dts: remove misplaced IRQ spec from 'soc' node
Gerhard Sittig [Tue, 3 Dec 2013 10:56:52 +0000 (11:56 +0100)]
powerpc/512x: dts: remove misplaced IRQ spec from 'soc' node

the 'soc' node in the common .dtsi for MPC5121 has an '#interrupt-cells'
property although this node is not an interrupt controller

remove this erroneously placed property because starting with v3.13-rc1
lookup and resolution of 'interrupts' specs for peripherals gets misled,
emits 'no irq domain found' WARN() messages and breaks the boot process

  irq: no irq domain found for /soc@80000000 !
  ------------[ cut here ]------------
  WARNING: at drivers/of/platform.c:171
  Modules linked in:
  CPU: 0 PID: 1 Comm: swapper Tainted: G        W    3.13.0-rc1-00001-g8a66234 #8
  task: df823bb0 ti: df834000 task.ti: df834000
  NIP: c02b5190 LR: c02b5180 CTR: c01cf4e0
  REGS: df835c50 TRAP: 0700   Tainted: G        W     (3.13.0-rc1-00001-g8a66234)
  MSR: 00029032 <EE,ME,IR,DR,RI>  CR: 229a9d42  XER: 20000000

  GPR00: c02b5180 df835d00 df823bb0 00000000 00000000 df835b18 ffffffff 00000308
  GPR08: c0479cc0 c0480000 c0479cc0 00000308 00000308 00000000 c00040fc 00000000
  GPR16: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 df850880
  GPR24: df84d670 00000000 00000001 df8561a0 dffffccc df85089c 00000020 00000001
  NIP [c02b5190] of_device_alloc+0xf4/0x1a0
  LR [c02b5180] of_device_alloc+0xe4/0x1a0
  Call Trace:
  [df835d00] [c02b5180] of_device_alloc+0xe4/0x1a0 (unreliable)
  [df835d50] [c02b5278] of_platform_device_create_pdata+0x3c/0xc8
  [df835d70] [c02b53fc] of_platform_bus_create+0xf8/0x170
  [df835dc0] [c02b5448] of_platform_bus_create+0x144/0x170
  [df835e10] [c02b55a8] of_platform_bus_probe+0x98/0xe8
  [df835e30] [c0437508] mpc512x_init+0x28/0x1c4
  [df835e70] [c0435de8] ppc_init+0x4c/0x60
  [df835e80] [c0003b28] do_one_initcall+0x150/0x1a4
  [df835ef0] [c0432048] kernel_init_freeable+0x114/0x1c0
  [df835f30] [c0004114] kernel_init+0x18/0x124
  [df835f40] [c000e910] ret_from_kernel_thread+0x5c/0x64
  Instruction dump:
  409effd4 57c9103a 57de2834 7c89f050 7f83e378 7c972214 7f45d378 48001f55
  7c63d278 7c630034 5463d97e 687a0001 <0f1a00002f990000 387b0010 939b0098
  ---[ end trace 2257f10e5a20cbdd ]---

  ...
  irq: no irq domain found for /soc@80000000 !
  fsl-diu-fb 80002100.display: could not get DIU IRQ
  fsl-diu-fb: probe of 80002100.display failed with error -22
  irq: no irq domain found for /soc@80000000 !
  mpc512x_dma 80014000.dma: Error mapping IRQ!
  mpc512x_dma: probe of 80014000.dma failed with error -22
  ...
  irq: no irq domain found for /soc@80000000 !
  fs_enet: probe of 80002800.ethernet failed with error -22
  ...
  irq: no irq domain found for /soc@80000000 !
  mpc5121-rtc 80000a00.rtc: mpc5121_rtc_probe: could not request irq: 0
  mpc5121-rtc: probe of 80000a00.rtc failed with error -22
  ...

Cc: Anatolij Gustschin <agust@denx.de>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: devicetree@vger.kernel.org
Signed-off-by: Gerhard Sittig <gsi@denx.de>
Signed-off-by: Anatolij Gustschin <agust@denx.de>
10 years agopowernv: Remove get/set_rtc_time when they are not present
Michael Neuling [Fri, 29 Nov 2013 03:22:48 +0000 (14:22 +1100)]
powernv: Remove get/set_rtc_time when they are not present

Currently we continue to poll get/set_rtc_time even when we know they
are not working.

This changes it so that if it fails at boot time we remove the ppc_md
get/set_rtc_time hooks so that we don't end up polling known broken
calls.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Add real mode cache inhibited IO accessors
Michael Ellerman [Fri, 29 Nov 2013 02:27:18 +0000 (13:27 +1100)]
powerpc: Add real mode cache inhibited IO accessors

These accessors allow us to do cache inhibited accesses when in real
mode. They should only be used in real mode.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Increase EEH recovery timeout for SR-IOV
Brian King [Mon, 25 Nov 2013 22:27:54 +0000 (16:27 -0600)]
powerpc: Increase EEH recovery timeout for SR-IOV

In order to support concurrent adapter firmware download
to SR-IOV adapters on pSeries, each VF will see an EEH event
where the slot will remain in the unavailable state for
the duration of the adapter firmware update, which can take
as long as 5 minutes. Extend the EEH recovery timeout to
account for this.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Acked-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/eeh: Output PHB diag-data
Gavin Shan [Fri, 22 Nov 2013 08:28:46 +0000 (16:28 +0800)]
powerpc/eeh: Output PHB diag-data

When hitting frozen PE or fenced PHB, it's always indicative to
have dumped PHB diag-data for further analysis and diagnosis.
However, we never dump that for the cases. The patch intends to
dump PHB diag-data at the backend of eeh_ops::get_log() for PowerNV
platform.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Move PHB-diag dump functions around
Gavin Shan [Fri, 22 Nov 2013 08:28:45 +0000 (16:28 +0800)]
powerpc/powernv: Move PHB-diag dump functions around

Prior to the completion of PCI enumeration, we actively detects
EEH errors on PCI config cycles and dump PHB diag-data if necessary.
The EEH backend also dumps PHB diag-data in case of frozen PE or
fenced PHB. However, we are using different functions to dump the
PHB diag-data for those 2 cases.

The patch merges the functions for dumping PHB diag-data to one so
that we can avoid duplicate code. Also, we never dump PHB3 diag-data
during PCI config cycles with frozen PE. The patch fixes it as well.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agoPPC: POWERNV: move iommu_add_device earlier
Alexey Kardashevskiy [Thu, 21 Nov 2013 06:43:14 +0000 (17:43 +1100)]
PPC: POWERNV: move iommu_add_device earlier

The current implementation of IOMMU on sPAPR does not use iommu_ops
and therefore does not call IOMMU API's bus_set_iommu() which
1) sets iommu_ops for a bus
2) registers a bus notifier
Instead, PCI devices are added to IOMMU groups from
subsys_initcall_sync(tce_iommu_init) which does basically the same
thing without using iommu_ops callbacks.

However Freescale PAMU driver (https://lkml.org/lkml/2013/7/1/158)
implements iommu_ops and when tce_iommu_init is called, every PCI device
is already added to some group so there is a conflict.

This patch does 2 things:
1. removes the loop in which PCI devices were added to groups and
adds explicit iommu_add_device() calls to add devices as soon as they get
the iommu_table pointer assigned to them.
2. moves a bus notifier to powernv code in order to avoid conflict with
the notifier from Freescale driver.

iommu_add_device() and iommu_del_device() are public now.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Move SG list structure to header file
Vasant Hegde [Mon, 18 Nov 2013 11:09:22 +0000 (16:39 +0530)]
powerpc/powernv: Move SG list structure to header file

Move SG list and entry structure to header file so that
it can be used in other places as well.

Signed-off-by: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Infrastructure to read opal messages in generic format.
Mahesh Salgaonkar [Mon, 18 Nov 2013 10:05:58 +0000 (15:35 +0530)]
powerpc/powernv: Infrastructure to read opal messages in generic format.

Opal now has a new messaging infrastructure to push the messages to
linux in a generic format for different type of messages using only one
event bit. The format of the opal message is as below:

struct opal_msg {
        uint32_t msg_type;
uint32_t reserved;
uint64_t params[8];
};

This patch allows clients to subscribe for notification for specific
message type. It is upto the subscriber to decipher the messages who showed
interested in receiving specific message type.

The interface to subscribe for notification is:

int opal_message_notifier_register(enum OpalMessageType msg_type,
                                        struct notifier_block *nb)

The notifier will fetch the opal message when available and notify the
subscriber with message type and the opal message. It is subscribers
responsibility to copy the message data before returning from notifier
callback.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/windfarm: Remove superfluous name casts
Geert Uytterhoeven [Tue, 12 Nov 2013 19:07:14 +0000 (20:07 +0100)]
powerpc/windfarm: Remove superfluous name casts

wf_sensor.name is "const char *"

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Machine check exception handling.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:36:13 +0000 (20:06 +0530)]
powerpc/powernv: Machine check exception handling.

Add basic error handling in machine check exception handler.

- If MSR_RI isn't set, we can not recover.
- Check if disposition set to OpalMCE_DISPOSITION_RECOVERED.
- Check if address at fault is inside kernel address space, if not then send
  SIGBUS to process if we hit exception when in userspace.
- If address at fault is not provided then and if we get a synchronous machine
  check while in userspace then kill the task.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Remove machine check handling in OPAL.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:35:58 +0000 (20:05 +0530)]
powerpc/powernv: Remove machine check handling in OPAL.

Now that we are ready to handle machine check directly in linux, do not
register with firmware to handle machine check exception.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Queue up and process delayed MCE events.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:35:49 +0000 (20:05 +0530)]
powerpc/book3s: Queue up and process delayed MCE events.

When machine check real mode handler can not continue into host kernel
in V mode, it returns from the interrupt and we loose MCE event which
never gets logged. In such a situation queue up the MCE event so that
we can log it later when we get back into host kernel with r1 pointing to
kernel stack e.g. during syscall exit.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Decode and save machine check event.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:35:40 +0000 (20:05 +0530)]
powerpc/book3s: Decode and save machine check event.

Now that we handle machine check in linux, the MCE decoding should also
take place in linux host. This info is crucial to log before we go down
in case we can not handle the machine check errors. This patch decodes
and populates a machine check event which contain high level meaning full
MCE information.

We do this in real mode C code with ME bit on. The MCE information is still
available on emergency stack (in pt_regs structure format). Even if we take
another exception at this point the MCE early handler will allocate a new
stack frame on top of current one. So when we return back here we still have
our MCE information safe on current stack.

We use per cpu buffer to save high level MCE information. Each per cpu buffer
is an array of machine check event structure indexed by per cpu counter
mce_nest_count. The mce_nest_count is incremented every time we enter
machine check early handler in real mode to get the current free slot
(index = mce_nest_count - 1). The mce_nest_count is decremented once the
MCE info is consumed by virtual mode machine exception handler.

This patch provides save_mce_event(), get_mce_event() and release_mce_event()
generic routines that can be used by machine check handlers to populate and
retrieve the event. The routine release_mce_event() will free the event slot so
that it can be reused. Caller can invoke get_mce_event() with a release flag
either to release the event slot immediately OR keep it so that it can be
fetched again. The event slot can be also released anytime by invoking
release_mce_event().

This patch also updates kvm code to invoke get_mce_event to retrieve generic
mce event rather than paca->opal_mce_evt.

The KVM code always calls get_mce_event() with release flags set to false so
that event is available for linus host machine

If machine check occurs while we are in guest, KVM tries to handle the error.
If KVM is able to handle MC error successfully, it enters the guest and
delivers the machine check to guest. If KVM is not able to handle MC error, it
exists the guest and passes the control to linux host machine check handler
which then logs MC event and decides how to handle it in linux host. In failure
case, KVM needs to make sure that the MC event is available for linux host to
consume. Hence KVM always calls get_mce_event() with release flags set to false
and later it invokes release_mce_event() only if it succeeds to handle error.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Flush SLB/TLBs if we get SLB/TLB machine check errors on power8.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:35:26 +0000 (20:05 +0530)]
powerpc/book3s: Flush SLB/TLBs if we get SLB/TLB machine check errors on power8.

This patch handles the memory errors on power8. If we get a machine check
exception due to SLB or TLB errors, then flush SLBs/TLBs and reload SLBs to
recover.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Flush SLB/TLBs if we get SLB/TLB machine check errors on power7.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:35:11 +0000 (20:05 +0530)]
powerpc/book3s: Flush SLB/TLBs if we get SLB/TLB machine check errors on power7.

If we get a machine check exception due to SLB or TLB errors, then flush
SLBs/TLBs and reload SLBs to recover. We do this in real mode before turning
on MMU. Otherwise we would run into nested machine checks.

If we get a machine check when we are in guest, then just flush the
SLBs and continue. This patch handles errors for power7. The next
patch will handle errors for power8

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Add flush_tlb operation in cpu_spec.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:34:56 +0000 (20:04 +0530)]
powerpc/book3s: Add flush_tlb operation in cpu_spec.

This patch introduces flush_tlb operation in cpu_spec structure. This will
help us to invoke appropriate CPU-side flush tlb routine. This patch
adds the foundation to invoke CPU specific flush routine for respective
architectures. Currently this patch introduce flush_tlb for p7 and p8.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Introduce a early machine check hook in cpu_spec.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:34:40 +0000 (20:04 +0530)]
powerpc/book3s: Introduce a early machine check hook in cpu_spec.

This patch adds the early machine check function pointer in cputable for
CPU specific early machine check handling. The early machine handle routine
will be called in real mode to handle SLB and TLB errors. We can not reuse
the existing machine_check hook because it is always invoked in kernel
virtual mode and we would already be in trouble if we get SLB or TLB errors.
This patch just sets up a mechanism to invoke CPU specific handler. The
subsequent patches will populate the function pointer.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Return from interrupt if coming from evil context.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:34:31 +0000 (20:04 +0530)]
powerpc/book3s: Return from interrupt if coming from evil context.

We can get machine checks from any context. We need to make sure that
we handle all of them correctly. If we are coming from hypervisor user-space,
we can continue in host kernel in virtual mode to deliver the MC event.
If we got woken up from power-saving mode then we may come in with one of
the following state:
 a. No state loss
 b. Supervisor state loss
 c. Hypervisor state loss
For (a) and (b), we go back to nap again. State (c) is fatal, keep spinning.

For all other context which we not sure of queue up the MCE event and return
from the interrupt.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: handle machine check in Linux host.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:34:08 +0000 (20:04 +0530)]
powerpc/book3s: handle machine check in Linux host.

Move machine check entry point into Linux. So far we were dependent on
firmware to decode MCE error details and handover the high level info to OS.

This patch introduces early machine check routine that saves the MCE
information (srr1, srr0, dar and dsisr) to the emergency stack. We allocate
stack frame on emergency stack and set the r1 accordingly. This allows us to be
prepared to take another exception without loosing context. One thing to note
here that, if we get another machine check while ME bit is off then we risk a
checkstop. Hence we restrict ourselves to save only MCE information and
register saved on PACA_EXMC save are before we turn the ME bit on. We use
paca->in_mce flag to differentiate between first entry and nested machine check
entry which helps proper use of emergency stack. We increment paca->in_mce
every time we enter in early machine check handler and decrement it while
leaving. When we enter machine check early handler first time (paca->in_mce ==
0), we are sure nobody is using MC emergency stack and allocate a stack frame
at the start of the emergency stack. During subsequent entry (paca->in_mce >
0), we know that r1 points inside emergency stack and we allocate separate
stack frame accordingly. This prevents us from clobbering MCE information
during nested machine checks.

The early machine check handler changes are placed under CPU_FTR_HVMODE
section. This makes sure that the early machine check handler will get executed
only in hypervisor kernel.

This is the code flow:

Machine Check Interrupt
|
V
   0x200 vector   ME=0, IR=0, DR=0
|
V
+-----------------------------------------------+
|machine_check_pSeries_early: | ME=0, IR=0, DR=0
| Alloc frame on emergency stack |
| Save srr1, srr0, dar and dsisr on stack |
+-----------------------------------------------+
|
(ME=1, IR=0, DR=0, RFID)
|
V
machine_check_handle_early   ME=1, IR=0, DR=0
|
V
+-----------------------------------------------+
| machine_check_early (r3=pt_regs) | ME=1, IR=0, DR=0
| Things to do: (in next patches) |
| Flush SLB for SLB errors |
| Flush TLB for TLB errors |
| Decode and save MCE info |
+-----------------------------------------------+
|
(Fall through existing exception handler routine.)
|
V
machine_check_pSerie   ME=1, IR=0, DR=0
|
(ME=1, IR=1, DR=1, RFID)
|
V
machine_check_common   ME=1, IR=1, DR=1
.
.
.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Introduce exclusive emergency stack for machine check exception.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:34:00 +0000 (20:04 +0530)]
powerpc/book3s: Introduce exclusive emergency stack for machine check exception.

This patch introduces exclusive emergency stack for machine check exception.
We use emergency stack to handle machine check exception so that we can save
MCE information (srr1, srr0, dar and dsisr) before turning on ME bit and be
ready for re-entrancy. This helps us to prevent clobbering of MCE information
in case of nested machine checks.

The reason for using emergency stack over normal kernel stack is that the
machine check might occur in the middle of setting up a stack frame which may
result into improper use of kernel stack.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/book3s: Split the common exception prolog logic into two section.
Mahesh Salgaonkar [Wed, 30 Oct 2013 14:33:51 +0000 (20:03 +0530)]
powerpc/book3s: Split the common exception prolog logic into two section.

This patch splits the common exception prolog logic into three parts to
facilitate reuse of existing code in the next patch. This patch also
re-arranges few instructions in such a way that the second part now deals
with saving register values from paca save area to stack frame, and
the third part deals with saving current register values to stack frame.

The second and third part will be reused in the machine check exception
routine in the subsequent patch.

Please note that this patch does not introduce or change existing code
logic. Instead it is just a code movement and instruction re-ordering.

Patch Acked-by Paul. But made some minor modification (explained above) to
address Paul's comment in the later patch(3).

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/powernv: Replace CONFIG_POWERNV_MSI with just CONFIG_PPC_POWERNV
Michael Ellerman [Thu, 26 Sep 2013 07:52:17 +0000 (17:52 +1000)]
powerpc/powernv: Replace CONFIG_POWERNV_MSI with just CONFIG_PPC_POWERNV

We currently have a user visible CONFIG_POWERNV_MSI option, but it
doesn't actually disable MSI for powernv. The MSI code is always built,
what it does disable is the inclusion of the MSI bitmap code, which
leads to a build error.

eg, with PPC_POWERNV=y and POWERNV_MSI=n we get:

  arch/powerpc/platforms/built-in.o: In function `.pnv_teardown_msi_irqs':
  pci.c:(.text+0x3558): undefined reference to `.msi_bitmap_free_hwirqs'

We don't really need a POWERNV_MSI symbol, just have the MSI bitmap code
depend directly on PPC_POWERNV.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Reviewed-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/pseries: CONFIG_PSERIES_MSI should depend on PPC_PSERIES
Michael Ellerman [Thu, 26 Sep 2013 07:52:16 +0000 (17:52 +1000)]
powerpc/pseries: CONFIG_PSERIES_MSI should depend on PPC_PSERIES

Previously PSERIES_MSI depended on PPC_PSERIES via EEH. However in
commit 317f06d "powerpc/eeh: Move common part to kernel directory" we
made CONFIG_EEH selectable on POWERNV. That leaves us with PSERIES_MSI
being live even when PSERIES=n. Fix it by making PSERIES_MSI depend
directly on PPC_PSERIES.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Reviewed-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/kernel/sysfs: Cleanup set up macros for PMC/non-PMC SPRs
Madhavan Srinivasan [Thu, 3 Oct 2013 09:27:35 +0000 (14:57 +0530)]
powerpc/kernel/sysfs: Cleanup set up macros for PMC/non-PMC SPRs

Currently PMC (Performance Monitor Counter) setup macros are used
for other SPRs. Since not all SPRs are PMC related, this patch
modifies the exisiting macro and uses it to setup both PMC and
non PMC SPRs accordingly.

Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Acked-by: Olof Johansson <olof@lixom.net>
Acked-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Make irq_stat.timers_irqs counting more specific
fan.du [Wed, 23 Jan 2013 08:06:11 +0000 (16:06 +0800)]
powerpc: Make irq_stat.timers_irqs counting more specific

Current irq_stat.timers_irqs counting doesn't discriminate timer event handler
and other timer interrupt(like arch_irq_work_raise). Sometimes we need to know
exactly how much interrupts timer event handler fired, so let's be more specific
on this.

Signed-off-by: Fan Du <fan.du@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: purge all the prefetched instructions for the coherent icache flush
Kevin Hao [Thu, 22 Aug 2013 01:30:35 +0000 (09:30 +0800)]
powerpc: purge all the prefetched instructions for the coherent icache flush

As Benjamin Herrenschmidt has indicated, we still need a dummy icbi to
purge all the prefetched instructions from the ifetch buffers for the
snooping icache. We also need a sync before the icbi to order the
actual stores to memory that might have modified instructions with
the icbi.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: kernel: remove useless code which related with 'max_cpus'
Chen Gang [Mon, 22 Jul 2013 06:40:20 +0000 (14:40 +0800)]
powerpc: kernel: remove useless code which related with 'max_cpus'

Since not need 'max_cpus' after the related commit, the related code
are useless too, need be removed.

The related commit:

  c1aa687 powerpc: Clean up obsolete code relating to decrementer and timebase

The related warning:

  arch/powerpc/kernel/smp.c:323:43: warning: parameter ‘max_cpus’ set but not used [-Wunused-but-set-parameter]

Signed-off-by: Chen Gang <gang.chen@asianux.com>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/boot: Ignore .dtb files.
Ian Campbell [Mon, 3 Jun 2013 11:00:56 +0000 (12:00 +0100)]
powerpc/boot: Ignore .dtb files.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/dts/virtex440: Declare address/size-cells for phy device
Ian Campbell [Mon, 25 Nov 2013 09:42:26 +0000 (09:42 +0000)]
powerpc/dts/virtex440: Declare address/size-cells for phy device

This fixes a warning:

  DTC     arch/powerpc/boot/virtex440-ml507.dtb
Warning (reg_format): "reg" property in /plb@0/xps-ll-temac@81c00000/ethernet@81c00000/phy@7 has invalid length (4 bytes) (#address-cells == 2, #size-cells == 1)
Warning (avoid_default_addr_size): Relying on default #address-cells value for /plb@0/xps-ll-temac@81c00000/ethernet@81c00000/phy@7
Warning (avoid_default_addr_size): Relying on default #size-cells value for /plb@0/xps-ll-temac@81c00000/ethernet@81c00000/phy@7

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Gernot Vormayr <gvormayr@gmail.com>
Cc: devicetree@vger.kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/4xx: Fix warning in kilauea.dtb
Ian Campbell [Mon, 25 Nov 2013 09:40:37 +0000 (09:40 +0000)]
powerpc/4xx: Fix warning in kilauea.dtb

Currently I see:
  DTC     arch/powerpc/boot/kilauea.dtb
Warning (reg_format): "reg" property in /plb/ppc4xx-msi@C10000000 has invalid length (12 bytes) (#address-cells == 1, #size-cells == 1)

It appears that unlike the other platforms handled by 3fb7933850fa
"powerpc/4xx: Adding PCIe MSI support" this platform does not use address-cells=2.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Josh Boyer <jwboyer@gmail.com>
Cc: Rupjyoti Sarmah <rsarmah@apm.com>
Cc: Tirumala R Marri <tmarri@apm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: devicetree@vger.kernel.org (open list:OPEN FIRMWARE AND...)
Cc: linuxppc-dev@lists.ozlabs.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Use patch_exception to update the debug exception handler
Kevin Hao [Sat, 11 May 2013 23:26:23 +0000 (07:26 +0800)]
powerpc: Use patch_exception to update the debug exception handler

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Move the patch_exception to a common place
Kevin Hao [Sat, 11 May 2013 23:26:22 +0000 (07:26 +0800)]
powerpc: Move the patch_exception to a common place

So that it can be used by other codes. No function change.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/ps3: Remove inline marking of EXPORT_SYMBOL functions
Denis Efremov [Thu, 9 May 2013 10:36:58 +0000 (14:36 +0400)]
powerpc/ps3: Remove inline marking of EXPORT_SYMBOL functions

EXPORT_SYMBOL and inline directives are contradictory to each other.
The patch fixes this inconsistency.

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Denis Efremov <yefremov.denis@gmail.com>
Acked-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/windfarm: Fix XServe G5 fan control Makefile issue
Benjamin Herrenschmidt [Wed, 27 Nov 2013 00:35:47 +0000 (11:35 +1100)]
powerpc/windfarm: Fix XServe G5 fan control Makefile issue

We are missing building windfarm_max6690_sensor.o when building
CONFIG_WINDFARM_RM31. Usually all the windfarm drivers are built
and thus this isn't a problem but some more "tailored" setups
(Gentoo ?) building only that driver are not working because
the require sensor module is missing.

Reported-by: Stanislav Ponomarev <devhexorg@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agoarch/powerpc/kernel: Use %12.12s instead of %12s to avoid memory overflow
Chen Gang [Thu, 24 Jan 2013 04:14:52 +0000 (12:14 +0800)]
arch/powerpc/kernel: Use %12.12s instead of %12s to avoid memory overflow

for tmp_part->header.name:
    it is "Terminating null required only for names < 12 chars".
    so need to limit the %.12s for it in printk

  additional info:

    %12s  limit the width, not for the original string output length
          if name length is more than 12, it still can be fully displayed.
          if name length is less than 12, the ' ' will be filled before name.

    %.12s truly limit the original string output length (precision)

Signed-off-by: Chen Gang <gang.chen@asianux.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/signals: Improved mark VSX not saved with small contexts fix
Michael Neuling [Mon, 25 Nov 2013 00:12:20 +0000 (11:12 +1100)]
powerpc/signals: Improved mark VSX not saved with small contexts fix

In a recent patch:
  commit c13f20ac48328b05cd3b8c19e31ed6c132b44b42
  Author: Michael Neuling <mikey@neuling.org>
  powerpc/signals: Mark VSX not saved with small contexts

We fixed an issue but an improved solution was later discussed after the patch
was merged.

Firstly, this patch doesn't handle the 64bit signals case, which could also hit
this issue (but has never been reported).

Secondly, the original patch isn't clear what MSR VSX should be set to.  The
new approach below always clears the MSR VSX bit (to indicate no VSX is in the
context) and sets it only in the specific case where VSX is available (ie. when
VSX has been used and the signal context passed has space to provide the
state).

This reverts the original patch and replaces it with the improved solution.  It
also adds a 64 bit version.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Cc: stable@vger.kernel.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc/kdump: Adding symbols in vmcoreinfo to facilitate dump filtering
Hari Bathini [Fri, 15 Nov 2013 17:31:32 +0000 (23:01 +0530)]
powerpc/kdump: Adding symbols in vmcoreinfo to facilitate dump filtering

When CONFIG_SPARSEMEM_VMEMMAP option is used in kernel, makedumpfile fails
to filter vmcore dump as it fails to do vmemmap translations. So far
dump filtering on ppc64 never had to deal with vmemmap addresses seperately
as vmemmap regions where mapped in zone normal. But with the inclusion of
CONFIG_SPARSEMEM_VMEMMAP config option in kernel, this vmemmap address
translation support becomes necessary for dump filtering. For vmemmap adress
translation, few kernel symbols are needed by dump filtering tool. This patch
adds those symbols to vmcoreinfo, which a dump filtering tool can use for
filtering the kernel dump. Tested this changes successfully with makedumpfile
tool that supports vmemmap to physical address translation outside zone normal.

[ Removed unneeded #ifdef as suggested by Michael Ellerman --BenH ]

Signed-off-by: Hari Bathini <hbathini@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: allyesconfig should not select CONFIG_CPU_LITTLE_ENDIAN
Anton Blanchard [Thu, 21 Nov 2013 05:33:55 +0000 (16:33 +1100)]
powerpc: allyesconfig should not select CONFIG_CPU_LITTLE_ENDIAN

Stephen reported a failure in an allyesconfig build.
CONFIG_CPU_LITTLE_ENDIAN=y gets set but his toolchain is not
new enough to support little endian. We really want to
default to a big endian build; Ben suggested using a choice
which defaults to CPU_BIG_ENDIAN.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agopowerpc: Fix error when cross building TAGS & cscope
Michael Neuling [Thu, 21 Nov 2013 03:59:07 +0000 (14:59 +1100)]
powerpc: Fix error when cross building TAGS & cscope

Currently if I cross build TAGS or cscope from x86 I get this:
  % make ARCH=powerpc TAGS
  gcc-4.8.real: error: unrecognized command line option ‘-mbig-endian’
  GEN     TAGS
  %

I'm not setting CROSS_COMPILE= as logically I shouldn't need to and I
haven't needed to in the past when building TAGS or cscope.  Also, the
above completess correct as the error is not fatal to the build.

This was caused by:
    commit d72b08017161ab385d4ae080ea415c9eb7ceef83
    Author: Ian Munsie <imunsie@au1.ibm.com>
    powerpc: Add ability to build little endian kernels

The below fixes this by testing for the -mbig-endian option before
adding it.

I've not done the same thing in the little endian case as if
-mlittle-endian doesn't exist, we probably want to fail quickly as you
probably have an old big endian compiler.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
10 years agoMerge remote-tracking branch 'scott/master' into merge
Benjamin Herrenschmidt [Sun, 24 Nov 2013 23:04:27 +0000 (10:04 +1100)]
Merge remote-tracking branch 'scott/master' into merge

Scott wrote:

<<
The corenet64 patch fixes a regression introduced in 3.13-rc1 (commit
ef1313deafb7baa6d3382044e962d5ad5e8c8dd6, "powerpc: Add VMX optimised xor
for RAID5").

The 8xx patch fixes a regression introduced in 3.12 (commit
beb2dc0a7a84be003ce54e98b95d65cc66e6e536, "powerpc: Convert some
mftb/mftbu into mfspr").

The other two patches are fixes for minor, long standing bugs.
>>

10 years agopowerpc/booke: Only check for hugetlb in flush if vma != NULL
Scott Wood [Fri, 22 Nov 2013 00:26:42 +0000 (18:26 -0600)]
powerpc/booke: Only check for hugetlb in flush if vma != NULL

And in flush_hugetlb_page(), don't check whether vma is NULL after
we've already dereferenced it.

This was found by Dan using static analysis as described here:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2013-November/113161.html

We currently get away with this because the callers that currently pass
NULL for vma seem to be 32-bit-only (e.g. highmem, and
CONFIG_DEBUG_PGALLOC in pgtable_32.c)  Hugetlb is currently 64-bit only,
so we never saw a NULL vma here.

Signed-off-by: Scott Wood <scottwood@freescale.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
10 years agopowerpc/85xx: typo in dts: "interupt" (four devices)
Adam Borowski [Tue, 12 Nov 2013 16:56:51 +0000 (17:56 +0100)]
powerpc/85xx: typo in dts: "interupt" (four devices)

These lines were inoperative for four years, which puts some doubt into
their importance, and it's possible the fixed version will regress, but
at the very least they should be removed instead.

Signed-off-by: Adam Borowski <kilobyte@angband.pl>
Signed-off-by: Scott Wood <scottwood@freescale.com>
10 years agopowerpc/8xx: mfspr SPRN_TBRx in lieu of mftb/mftbu is not supported
LEROY Christophe [Fri, 22 Nov 2013 16:57:31 +0000 (17:57 +0100)]
powerpc/8xx: mfspr SPRN_TBRx in lieu of mftb/mftbu is not supported

Commit beb2dc0a7a84be003ce54e98b95d65cc66e6e536 breaks the MPC8xx which
seems to not support using mfspr SPRN_TBRx instead of mftb/mftbu
despite what is written in the reference manual.

This patch reverts to the use of mftb/mftbu when CONFIG_8xx is
selected.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Scott Wood <scottwood@freescale.com>
10 years agopowerpc/corenet64: compile with CONFIG_E{5,6}500_CPU well
Tiejun Chen [Fri, 22 Nov 2013 02:49:00 +0000 (10:49 +0800)]
powerpc/corenet64: compile with CONFIG_E{5,6}500_CPU well

If CONFIG_ALTIVEC is enabled for CoreNet64, and if we also
select CONFIG_E{5,6}500_CPU this may introduce -mcpu=e500mc64
into $CFLAGS. But Altivec option not allowed with e500mc64,
then some compiling errors occur like this:

      CC      arch/powerpc/lib/xor_vmx.o
    arch/powerpc/lib/xor_vmx.c:1:0: error: AltiVec not supported in this target
    make[1]: *** [arch/powerpc/lib/xor_vmx.o] Error 1
    make: *** [arch/powerpc/lib] Error 2

So we should restrict e500mc64 in altivec scenario.

Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
10 years agoLinux 3.13-rc1
Linus Torvalds [Fri, 22 Nov 2013 19:30:55 +0000 (11:30 -0800)]
Linux 3.13-rc1