firefly-linux-kernel-4.4.55.git
11 years agopowerpc/pseries: Fix creation of loop in device node property list
Tyrel Datwyler [Thu, 15 Aug 2013 05:23:45 +0000 (22:23 -0700)]
powerpc/pseries: Fix creation of loop in device node property list

The update_dt_prop helper function fails to set the IN/OUT parameter prop to
NULL after a complete property has been parsed from the work area returned by
the ibm,update-properties rtas function. This results in the property list of
the device node being updated is corrupted and becomes a loop since the same
property structure is used repeatedly.

Signed-off-by: Tyrel Datwyler <tyreld@linux.vnet.ibm.com>
Acked-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Skip emulating & leave interrupts off for kernel program checks
Michael Ellerman [Thu, 15 Aug 2013 05:22:19 +0000 (15:22 +1000)]
powerpc: Skip emulating & leave interrupts off for kernel program checks

In the program check handler we handle some causes with interrupts off
and others with interrupts on.

We need to enable interrupts to handle the emulation cases, because they
access userspace memory and might sleep.

For faults in the kernel we don't want to do any emulation, and
emulate_instruction() enforces that. do_mathemu() doesn't but probably
should.

The other disadvantage of enabling interrupts for kernel faults is that
we may take another interrupt, and recurse. As seen below:

  --- Exception: e40 at c000000000004ee0 performance_monitor_relon_pSeries_1
  [link register   ] c00000000000f858 .arch_local_irq_restore+0x38/0x90
  [c000000fb185dc100000000000000000 (unreliable)
  [c000000fb185dc80c0000000007d8558 .program_check_exception+0x298/0x2d0
  [c000000fb185dd00c000000000002f40 emulation_assist_common+0x140/0x180
  --- Exception: e40 at c000000000004ee0 performance_monitor_relon_pSeries_1
  [link register   ] c00000000000f858 .arch_local_irq_restore+0x38/0x90
  [c000000fb185dff000000000008b9190 (unreliable)
  [c000000fb185e060c0000000007d8558 .program_check_exception+0x298/0x2d0

So avoid both problems by checking if the fault was in the kernel and
skipping the enable of interrupts and the emulation. Go straight to
delivering the SIGILL, which for kernel faults calls die() and so on,
dropping us in the debugger etc.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Add more exception trampolines for hypervisor exceptions
Michael Ellerman [Thu, 15 Aug 2013 05:22:18 +0000 (15:22 +1000)]
powerpc: Add more exception trampolines for hypervisor exceptions

This makes back traces and profiles easier to read.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Fix location and rename exception trampolines
Michael Ellerman [Thu, 15 Aug 2013 05:22:17 +0000 (15:22 +1000)]
powerpc: Fix location and rename exception trampolines

The symbols that name some of our exception trampolines are ahead of the
location they name. In most cases this is OK because the code is tightly
packed, but in some cases it means the symbol floats ahead of the
correct location, eg:

  c000000000000ea0 <performance_monitor_pSeries_1>:
          ...
  c000000000000f00:       7d b2 43 a6     mtsprg  2,r13

Fix them all by moving the symbol after the set of the location.

While we're moving them anyway, rename them to loose the camelcase and
to make it clear that they are trampolines.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Add more trap names to xmon
Michael Ellerman [Thu, 15 Aug 2013 05:22:16 +0000 (15:22 +1000)]
powerpc: Add more trap names to xmon

We haven't updated these for a while it seems, it's nice to have in the
oops output.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pseries: Add a warning in the case of cross-cpu VPA registration
Michael Ellerman [Thu, 15 Aug 2013 05:22:15 +0000 (15:22 +1000)]
powerpc/pseries: Add a warning in the case of cross-cpu VPA registration

The spec says it "may be problematic" if CPU x registers the VPA of
CPU y. Add a warning in case we ever do that.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Update the 00-Index in Documentation/powerpc
Michael Ellerman [Thu, 15 Aug 2013 05:22:14 +0000 (15:22 +1000)]
powerpc: Update the 00-Index in Documentation/powerpc

People have been dropping things in here without updating the index, do
it for them.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Never handle VSX alignment exceptions from kernel
Anton Blanchard [Tue, 20 Aug 2013 10:30:07 +0000 (20:30 +1000)]
powerpc: Never handle VSX alignment exceptions from kernel

The VSX alignment handler needs to write out the existing VSX
state to memory before operating on it (flush_vsx_to_thread()).
If we take a VSX alignment exception in the kernel bad things
will happen. It looks like we could write the kernel state out
to the user process, or we could handle the kernel exception
using data from the user process (depending if MSR_VSX is set
or not).

Worse still, if the code to read or write the VSX state causes an
alignment exception, we will recurse forever. I ended up with
hundreds of megabytes of kernel stack to look through as a result.

Floating point and SPE code have similar issues but already include
a user check. Add the same check to emulate_vsx().

With this patch any unaligned VSX loads and stores in the kernel
will show up as a clear oops rather than silent corruption of
kernel or userspace VSX state, or worse, corruption of a potentially
unlimited amount of kernel memory.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopseries: Move plpar_wrapper.h to powerpc common include/asm location.
Deepthi Dharwar [Thu, 22 Aug 2013 09:53:52 +0000 (15:23 +0530)]
pseries: Move plpar_wrapper.h to powerpc common include/asm location.

As a part of pseries_idle backend driver cleanup to make
the code common to both pseries and powernv platforms, it
is necessary to move the backend-driver code to drivers/cpuidle.

As a pre-requisite for that, it is essential to move plpar_wrapper.h
to include/asm.

Signed-off-by: Deepthi Dharwar <deepthi@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopseries/cpuidle: Remove dependency of pseries.h file
Deepthi Dharwar [Thu, 22 Aug 2013 09:53:42 +0000 (15:23 +0530)]
pseries/cpuidle: Remove dependency of pseries.h file

As a part of pseries_idle cleanup to make the backend driver
code common to both pseries and powernv.
Remove non-essential smt_snooze_delay declaration in pseries.h
header file and pseries.h file inclusion in
pseries/processor_idle.c

Signed-off-by: Deepthi Dharwar <deepthi@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Unaligned stores and stmw are broken in emulation code
Tom Musta [Thu, 22 Aug 2013 14:25:28 +0000 (09:25 -0500)]
powerpc: Unaligned stores and stmw are broken in emulation code

The stmw instruction was incorrectly decoded as an update form instruction
and thus the RA register was being clobbered.

Also, the utility routine to write memory to unaligned addresses breaks the
operation into smaller aligned accesses but was incorrectly incrementing
the address by only one; it needs to increment the address by the size of
the smaller aligned chunk.

Signed-off-by: Tom Musta <tmusta@us.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/mm: Mark Memory Resources as busy
Nathan Fontenot [Tue, 20 Aug 2013 02:52:20 +0000 (21:52 -0500)]
powerpc/mm: Mark Memory Resources as busy

Memory I/O resources need to be marked as busy or else we cannot remove
them when doing memory hot remove.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agoMerge remote-tracking branch 'scott/next' into next
Benjamin Herrenschmidt [Tue, 27 Aug 2013 03:49:32 +0000 (13:49 +1000)]
Merge remote-tracking branch 'scott/next' into next

Latest FSL updates from Scott:

<<
Highlights include changes in compiler flag settings
on e500 family cores, booke64 hibernation support, support for two new
boards, and an erratum workaround.
>>

11 years agopowerpc/85xx: Add C293PCIE board support
Mingkai Hu [Fri, 2 Aug 2013 06:39:11 +0000 (14:39 +0800)]
powerpc/85xx: Add C293PCIE board support

C293PCIE board is a series of Freescale PCIe add-in cards to perform
as public key crypto accelerator or secure key management module.

 - 512KB platform SRAM in addition to 512K L2 Cache/SRAM
 - 512MB soldered DDR3 32bit memory
 - CPLD System Logic
 - 64MB x16 NOR flash and 4GB x8 NAND flash
 - 16MB SPI flash

Signed-off-by: Mingkai Hu <Mingkai.Hu@freescale.com>
Signed-off-by: Po Liu <Po.Liu@freescale.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/85xx: Add silicon device tree for C293
Mingkai Hu [Fri, 2 Aug 2013 06:39:10 +0000 (14:39 +0800)]
powerpc/85xx: Add silicon device tree for C293

Signed-off-by: Mingkai Hu <Mingkai.Hu@freescale.com>
Signed-off-by: Po Liu <Po.Liu@freescale.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/85xx: Add SEC6.0 device tree
Mingkai Hu [Fri, 2 Aug 2013 06:39:09 +0000 (14:39 +0800)]
powerpc/85xx: Add SEC6.0 device tree

Add device tree for SEC 6.0 used on C29x silicon.

Signed-off-by: Mingkai Hu <Mingkai.Hu@freescale.com>
Signed-off-by: Po Liu <Po.Liu@freescale.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc: add Book E support to 64-bit hibernation
Wang Dongsheng [Thu, 8 Aug 2013 02:06:45 +0000 (10:06 +0800)]
powerpc: add Book E support to 64-bit hibernation

Update the 64-bit hibernation code to support Book E CPUs.
Some registers and instructions are not defined for Book3e
(SDR reg, tlbia instruction).

SDR: Storage Description Register. Book3S and Book3E have different
address translation mode, we do not need HTABORG & HTABSIZE to
translate virtual address to real address.

More registers are saved in BookE-64bit.(TCR, SPRG1)

Signed-off-by: Wang Dongsheng <dongsheng.wang@freescale.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/85xx: Add P1023RDB board support
Chunhe Lan [Tue, 30 Jul 2013 21:39:26 +0000 (05:39 +0800)]
powerpc/85xx: Add P1023RDB board support

P1023RDB Specification:
-----------------------
Memory subsystem:
   512MB DDR3 (Fixed DDR on board)
   64MB NOR flash
   128MB NAND flash

Ethernet:
   eTSEC1: Connected to Atheros AR8035 GETH PHY
   eTSEC2: Connected to Atheros AR8035 GETH PHY

PCIe:
   Three mini-PCIe slots

USB:
   Two USB2.0 Type A ports

I2C:
   AT24C08 8K Board EEPROM (8 bit address)

Signed-off-by: Chunhe Lan <Chunhe.Lan@freescale.com>
Cc: Scott Wood <scottwood@freescale.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/85xx: Add support for 85xx cpu type detection
Haijun.Zhang [Wed, 17 Jul 2013 05:12:27 +0000 (13:12 +0800)]
powerpc/85xx: Add support for 85xx cpu type detection

Add this file to help detect cpu type in runtime.
These macros will be more favorable for driver
to apply errata and workaround to specified cpu type.

Signed-off-by: Haijun Zhang <Haijun.Zhang@freescale.com>
Signed-off-by: Zhao Chenhui <chenhui.zhao@freescale.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/e500: Set -mcpu flag for 32-bit e500
Scott Wood [Fri, 16 Aug 2013 00:19:10 +0000 (19:19 -0500)]
powerpc/e500: Set -mcpu flag for 32-bit e500

Unlike 64-bit, we don't currently support multiplatform between e500
and non-e500, so the -mcpu is not configurable at this time.

-msoft-float is specified when testing for -mcpu=8540 because otherwise
some older toolchains will fail with "error: E500 and FPRs not
supported".

Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/booke64: Use appropriate -mcpu
Scott Wood [Wed, 21 Aug 2013 00:55:36 +0000 (19:55 -0500)]
powerpc/booke64: Use appropriate -mcpu

By default use -mcpu=powerpc64 rather than -mtune=power7

Add options for e5500/e6500, with fallbacks for older compilers.

Hide the POWER cpu options in booke configs.

Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/85xx: Remove -Wa,-me500
Scott Wood [Wed, 21 Aug 2013 00:53:10 +0000 (19:53 -0500)]
powerpc/85xx: Remove -Wa,-me500

This caused lwsync to be converted to sync on 64-bit (on 32-bit lwsync
is generated at runtime, and so wasn't affected).  Not using lwsync
caused a significant slowdown on certain workloads.

Setting this flag for any e500-enabled build is also not friendly to
multiplatform kernels.

Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc: Convert some mftb/mftbu into mfspr
Scott Wood [Wed, 21 Aug 2013 00:33:12 +0000 (19:33 -0500)]
powerpc: Convert some mftb/mftbu into mfspr

Some CPUs (such as e500v1/v2) don't implement mftb and will take a
trap.  mfspr should work on everything that has a timebase, and is the
preferred instruction according to ISA v2.06.

Currently we get away with mftb on 85xx because the assembler converts
it to mfspr due to -Wa,-me500.  However, that flag has other effects
that are undesireable for certain targets (e.g.  lwsync is converted to
sync), and is hostile to multiplatform kernels.  Thus we would like to
stop setting it for all e500-family builds.

mftb/mftbu instances which are in 85xx code or common code are
converted.  Instances which will never run on 85xx are left alone.

Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/fsl-booke: Work around erratum A-006958
Scott Wood [Wed, 24 Jul 2013 01:21:11 +0000 (20:21 -0500)]
powerpc/fsl-booke: Work around erratum A-006958

Erratum A-006598 says that 64-bit mftb is not atomic -- it's subject
to a similar race condition as doing mftbu/mftbl on 32-bit.  The lower
half of timebase is updated before the upper half; thus, we can share
the workaround for a similar bug on Cell.  This workaround involves
looping if the lower half of timebase is zero, thus avoiding the need
for a scratch register (other than CR0).  This workaround must be
avoided when the timebase is frozen, such as during the timebase sync
code.

This deals with kernel and vdso accesses, but other userspace accesses
will of course need to be fixed elsewhere.

Signed-off-by: Scott Wood <scottwood@freescale.com>
11 years agopowerpc/wsp: Fix early debug build
Benjamin Herrenschmidt [Fri, 16 Aug 2013 00:13:06 +0000 (10:13 +1000)]
powerpc/wsp: Fix early debug build

When reworking udbg_16550.c I forgot to remove the old and now useless
code for the CONFIG_PPC_EARLY_DEBUG_WSP case, which doesn't build as
a result. I also missed a cast.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make rwlocks endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:51 +0000 (02:01 +1000)]
powerpc: Make rwlocks endian safe

Our ppc64 spinlocks and rwlocks use a trick where a lock token and
the paca index are placed in the lock with a single store. Since we
are using two u16s they need adjusting for little endian.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Fix little endian coredumps
Anton Blanchard [Tue, 6 Aug 2013 16:01:50 +0000 (02:01 +1000)]
powerpc: Fix little endian coredumps

We need to set ELF_DATA correctly on LE coredumps.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pseries: Fix endian issues in H_GET_TERM_CHAR/H_PUT_TERM_CHAR
Anton Blanchard [Tue, 6 Aug 2013 16:01:49 +0000 (02:01 +1000)]
powerpc/pseries: Fix endian issues in H_GET_TERM_CHAR/H_PUT_TERM_CHAR

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Little endian SMP IPI demux
Anton Blanchard [Tue, 6 Aug 2013 16:01:48 +0000 (02:01 +1000)]
powerpc: Little endian SMP IPI demux

Add little endian support for demuxing SMP IPIs

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Emulate instructions in little endian mode
Anton Blanchard [Tue, 6 Aug 2013 16:01:47 +0000 (02:01 +1000)]
powerpc: Emulate instructions in little endian mode

Alistair noticed we got a SIGILL on userspace mfpvr instructions.

Remove the little endian check in the emulation code, it is
probably there to protect against the old pseudo little endian
implementations but doesn't make sense for real little endian.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Fix little endian lppaca, slb_shadow and dtl_entry
Anton Blanchard [Tue, 6 Aug 2013 16:01:46 +0000 (02:01 +1000)]
powerpc: Fix little endian lppaca, slb_shadow and dtl_entry

The lppaca, slb_shadow and dtl_entry hypervisor structures are
big endian, so we have to byte swap them in little endian builds.

LE KVM hosts will also need to be fixed but for now add an #error
to remind us.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Add endian annotations to lppaca, slb_shadow and dtl_entry
Anton Blanchard [Tue, 6 Aug 2013 16:01:45 +0000 (02:01 +1000)]
powerpc: Add endian annotations to lppaca, slb_shadow and dtl_entry

Add endian annotation to various hypervisor structures which
are defined as big endian.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make NUMA device node code endian safe
Alistair Popple [Tue, 6 Aug 2013 16:01:44 +0000 (02:01 +1000)]
powerpc: Make NUMA device node code endian safe

The device tree is big endian so make sure we byteswap on little
endian. We assume any pHyp calls also return big endian results in
memory.

Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Little endian fixes for legacy_serial.c
Alistair Popple [Tue, 6 Aug 2013 16:01:43 +0000 (02:01 +1000)]
powerpc: Little endian fixes for legacy_serial.c

Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make PCI device node device tree accesses endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:42 +0000 (02:01 +1000)]
powerpc: Make PCI device node device tree accesses endian safe

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make OF PCI device tree accesses endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:41 +0000 (02:01 +1000)]
powerpc: Make OF PCI device tree accesses endian safe

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make device tree accesses in VIO subsystem endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:40 +0000 (02:01 +1000)]
powerpc: Make device tree accesses in VIO subsystem endian safe

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make device tree accesses in HVC VIO console endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:39 +0000 (02:01 +1000)]
powerpc: Make device tree accesses in HVC VIO console endian safe

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make prom_init.c endian safe
Benjamin Herrenschmidt [Tue, 6 Aug 2013 16:01:38 +0000 (02:01 +1000)]
powerpc: Make prom_init.c endian safe

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make device tree accesses in cache info code endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:37 +0000 (02:01 +1000)]
powerpc: Make device tree accesses in cache info code endian safe

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: of_parse_dma_window should take a __be32 *dma_window
Anton Blanchard [Tue, 6 Aug 2013 16:01:36 +0000 (02:01 +1000)]
powerpc: of_parse_dma_window should take a __be32 *dma_window

We pass dma_window to of_parse_dma_window as a void * and then
run through hoops to cast it back to a u32 array. In the process
we lose endian annotation.

Simplify it by just passing a __be32 * down.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Fix some endian issues in xics code
Anton Blanchard [Tue, 6 Aug 2013 16:01:35 +0000 (02:01 +1000)]
powerpc: Fix some endian issues in xics code

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Add some endian annotations to time and xics code
Anton Blanchard [Tue, 6 Aug 2013 16:01:34 +0000 (02:01 +1000)]
powerpc: Add some endian annotations to time and xics code

Fix a couple of sparse warnings.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: More little endian fixes for setup-common.c
Alistair Popple [Tue, 6 Aug 2013 16:01:33 +0000 (02:01 +1000)]
powerpc: More little endian fixes for setup-common.c

Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make logical to real cpu mapping code endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:32 +0000 (02:01 +1000)]
powerpc: Make logical to real cpu mapping code endian safe

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make RTAS calls endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:31 +0000 (02:01 +1000)]
powerpc: Make RTAS calls endian safe

RTAS expects arguments in the call buffer to be big endian so we
need to byteswap on little endian builds

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make cache info device tree accesses endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:30 +0000 (02:01 +1000)]
powerpc: Make cache info device tree accesses endian safe

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make RTAS device tree accesses endian safe
Anton Blanchard [Tue, 6 Aug 2013 16:01:29 +0000 (02:01 +1000)]
powerpc: Make RTAS device tree accesses endian safe

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: More little endian fixes for prom.c
Alistair Popple [Tue, 6 Aug 2013 16:01:28 +0000 (02:01 +1000)]
powerpc: More little endian fixes for prom.c

Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make prom.c device tree accesses endian safe
Ian Munsie [Tue, 6 Aug 2013 16:01:27 +0000 (02:01 +1000)]
powerpc: Make prom.c device tree accesses endian safe

On PowerPC the device tree is always big endian, but the CPU could be
either, so add be32_to_cpu where appropriate and change the types of
device tree data to __be32 etc to allow sparse to locate endian issues.

Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Acked-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Avoid link stack corruption for MMU on exceptions
Michael Neuling [Tue, 13 Aug 2013 05:54:52 +0000 (15:54 +1000)]
powerpc: Avoid link stack corruption for MMU on exceptions

When we have MMU on exceptions (POWER8) and a relocatable kernel, we
need to branch from the initial exception vectors at 0x0 to up high
where the kernel might be located.  Currently we do this using the link
register.

Unfortunately this corrupts the link stack and instead we should use the
count register.  We did this for the syscall entry path in:
  6a40480 powerpc: Avoid link stack corruption in MMU on syscall entry path
but I stupidly forgot to do the same for other exceptions.

This patch changes the initial exception vectors to use the count
register instead of the link register when we need to branch up to the
relocated kernel.

I have a dodgy userspace test which loops calling a function that reads
the PVR (mfpvr in userspace will be emulated by the kernel via the
program check exception).  On POWER8 and with CONFIG_RELOCATABLE=y, I
get a ~10% performance improvement with my userspace test with this
patch.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make chip-id information available to userspace
Vasant Hegde [Mon, 12 Aug 2013 12:05:57 +0000 (17:35 +0530)]
powerpc: Make chip-id information available to userspace

So far "/sys/devices/system/cpu/cpuX/topology/physical_package_id"
was always default (-1) on ppc64 architecture.

Now, some systems have an ibm,chip-id property in the cpu nodes in
the device tree. On these systems, we now use this information to
display physical_package_id.

Signed-off-by: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>
Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/eeh: powerpc/eeh: Fix undefined variable
Mike Qiu [Mon, 12 Aug 2013 06:15:36 +0000 (02:15 -0400)]
powerpc/eeh: powerpc/eeh: Fix undefined variable

changes for V4:
- changes the type of frozen_pe_no from %d to %llu
  in pr_devel()

'pe_no' hasn't been defined, it should be an typo error,
it should be 'frozen_pe_no'.

Also '__func__' has missed in IODA_EEH_DBG(),

For safety reasons, use pr_devel() directly, instead
of use IODA_EEH_DBG()

Signed-off-by: Mike Qiu <qiudayu@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agoRevert "powerpc/e500: Update compilation flags with core specific options"
Benjamin Herrenschmidt [Wed, 14 Aug 2013 04:17:24 +0000 (14:17 +1000)]
Revert "powerpc/e500: Update compilation flags with core specific options"

This reverts commit c8db32c8669f7de05b820ee4934926405af52188.

The commit breaks the build of all my 64-bit embedded configs. It
looks like gcc-4.7.3 doesn't know about e5500. Additionally it
incorrectly does -mcpu=e5500 on a config that has both e5500 and A2
support enabled.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---

11 years agopowerpc: Use ibm, chip-id property to compute cpu_core_mask if available
Paul Mackerras [Mon, 12 Aug 2013 06:29:33 +0000 (16:29 +1000)]
powerpc: Use ibm, chip-id property to compute cpu_core_mask if available

Some systems have an ibm,chip-id property in the cpu nodes in the
device tree.  On these systems, we now use that to compute the
cpu_core_mask (i.e. the set of core siblings) rather than looking
at cache properties.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Tested-by: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Pull out cpu_core_mask updates into a separate function
Paul Mackerras [Mon, 12 Aug 2013 06:28:47 +0000 (16:28 +1000)]
powerpc: Pull out cpu_core_mask updates into a separate function

This factors out the details of updating cpu_core_mask into a separate
function, to make it easier to change how the mask is calculated later.
This makes no functional change.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Fix denormalized exception handler
Paul Mackerras [Mon, 12 Aug 2013 06:12:06 +0000 (16:12 +1000)]
powerpc: Fix denormalized exception handler

The denormalized exception handler (denorm_exception_hv) has a couple
of bugs.  If the CONFIG_PPC_DENORMALISATION option is not selected,
or the HSRR1_DENORM bit is not set in HSRR1, we don't test whether the
interrupt occurred within a KVM guest.  On the other hand, if the
HSRR1_DENORM bit is set and CONFIG_PPC_DENORMALISATION is enabled,
we corrupt the CFAR and PPR.

To correct these problems, this replaces the open-coded version of
EXCEPTION_PROLOG_1 that is there currently, and that is missing the
saving of PPR and CFAR values to the PACA, with an instance of
EXCEPTION_PROLOG_1.  This adds an explicit KVMTEST after testing
whether the exception is one we can handle, and adds code to restore
the CFAR on exit.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Ignore zImage.epapr
Mark Brown [Tue, 6 Aug 2013 17:41:52 +0000 (18:41 +0100)]
powerpc: Ignore zImage.epapr

This is another file we can generate so add it to the list.

Signed-off-by: Mark Brown <broonie@linaro.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/math-emu: Fix load/store indexed emulation
James Yang [Thu, 4 Jul 2013 21:18:44 +0000 (16:18 -0500)]
powerpc/math-emu: Fix load/store indexed emulation

Load/store indexed instructions where the index register RA=R0, such
as "lfdx f1,0,r3", are not illegal.

Load/store indexed with update instructions where the index register
RA=R0, such as "lfdux f1,0,r3", are invalid, and, to be consistent
with existing math-emu behavior for other invalid instruction forms,
will signal as illegal.

Signed-off-by: James Yang <James.Yang@freescale.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Remove the empty giveup_fpu() function on 32bit kernel
Kevin Hao [Sun, 14 Jul 2013 09:02:06 +0000 (17:02 +0800)]
powerpc: Remove the empty giveup_fpu() function on 32bit kernel

Instead of implementing an empty giveup_fpu() function for each
32bit processor type, replace them with an unique empty inline
function.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Make flush_fp_to_thread() nop when CONFIG_PPC_FPU is disabled
Kevin Hao [Sun, 14 Jul 2013 09:02:05 +0000 (17:02 +0800)]
powerpc: Make flush_fp_to_thread() nop when CONFIG_PPC_FPU is disabled

In the current kernel, the function flush_fp_to_thread() is not
dependent on CONFIG_PPC_FPU. So most invocations of this function
is not wrapped by CONFIG_PPC_FPU. Even through we don't really
save the FPRs to the thread struct if CONFIG_PPC_FPU is not enabled,
but there does have some runtime overhead such as the check for
tsk->thread.regs and preempt disable and enable. It really make
no sense to do that. So make it a nop when CONFIG_PPC_FPU is
disabled. Also remove the wrapped #ifdef CONFIG_PPC_FPU
when invoking this function.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Remove the redundant flush_fp_to_thread() in setup_sigcontext()
Kevin Hao [Sun, 14 Jul 2013 09:02:04 +0000 (17:02 +0800)]
powerpc: Remove the redundant flush_fp_to_thread() in setup_sigcontext()

In commit c6e6771b(powerpc: Introduce VSX thread_struct and CONFIG_VSX)
we add a invocation of flush_fp_to_thread() before copying the FPR or
VSR to users. But we already invoke the flush_fp_to_thread() in this
function. So remove one of them.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/mpc85xx: Only emulate the unimplemented FP instructions on corenet64
Kevin Hao [Tue, 16 Jul 2013 11:57:16 +0000 (19:57 +0800)]
powerpc/mpc85xx: Only emulate the unimplemented FP instructions on corenet64

We have split the math emulation into two parts. This makes it
possible to just emulate the unimplemented floating point instructions
on these boards.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: remove the unused function disable_kernel_fp()
Kevin Hao [Sun, 14 Jul 2013 09:02:03 +0000 (17:02 +0800)]
powerpc: remove the unused function disable_kernel_fp()

The only using of function disable_kernel_fp() was already dropped
in the commit 5daf9071 (powerpc: merge align.c).

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: split She math emulation into two parts
Kevin Hao [Tue, 16 Jul 2013 11:57:15 +0000 (19:57 +0800)]
powerpc: split She math emulation into two parts

For some SoC (such as the FSL BookE) even though there does have
a hardware FPU, but not all floating point instructions are
implemented. Unfortunately some versions of gcc do use these
unimplemented instructions. Then we have to enable the math emulation
to workaround this issue. It seems a little redundant to have the
support to emulate all the floating point instructions in this case.
So split the math emulation into two parts. One is for the SoC which
doesn't have FPU at all and the other for the SoC which does have the
hardware FPU and only need some special floating point instructions to
be emulated.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Introduce function emulate_math()
Kevin Hao [Sun, 14 Jul 2013 08:40:07 +0000 (16:40 +0800)]
powerpc: Introduce function emulate_math()

There are two invocations of do_mathemu() in traps.c. And the codes
in these two places are almost the same. Introduce a locale function
to eliminate the duplication. With this change we can also make sure
that in program_check_exception() the PPC_WARN_EMULATED is invoked for
the correctly emulated math instructions.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/math-emu: Move the flush FPU state function into do_mathemu
Kevin Hao [Sun, 14 Jul 2013 08:40:06 +0000 (16:40 +0800)]
powerpc/math-emu: Move the flush FPU state function into do_mathemu

By doing this we can make sure that the FPU state is only flushed to
the thread struct when it is really needed.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/85xx: Enable the math emulation for the corenet64_smp_defconfig
Kevin Hao [Wed, 10 Jul 2013 01:49:52 +0000 (09:49 +0800)]
powerpc/85xx: Enable the math emulation for the corenet64_smp_defconfig

I got the following error on my t4240qds board.
  ntpd[2713]: unhandled signal 4 at 0fd5b448 nip 0fd5b448 lr 0fd5b424 code 30001

The root cause is that the float point instruction 'fsqrt' is used.
But this instruction is not implemented on e6500 core. Even this
does seem a gcc bug, I would like to enable the math emulation
in the kernel to workaround this kind of issue.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/8xx: Remove last traces of 8XX_MINIMAL_FPEMU
Paul Bolle [Tue, 16 Jul 2013 08:56:38 +0000 (10:56 +0200)]
powerpc/8xx: Remove last traces of 8XX_MINIMAL_FPEMU

The Kconfig symbol 8XX_MINIMAL_FPEMU was removed in commit 968219fa33
("powerpc/8xx: Remove 8xx specific "minimal FPU emulation""). But that
commit didn't remove all code depending on that symbol. Do so now.

Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/math-emu: Remove the unneeded check for CONFIG_MATH_EMULATION in math.c
Kevin Hao [Wed, 10 Jul 2013 01:43:43 +0000 (09:43 +0800)]
powerpc/math-emu: Remove the unneeded check for CONFIG_MATH_EMULATION in math.c

The math.c is only built when CONFIG_MATH_EMULATION is enabled.
So the #ifdef check for CONFIG_MATH_EMULATION in it seems redundant.
Drop all of them.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/math-emu: Remove the dead code in math.c
Kevin Hao [Wed, 10 Jul 2013 01:43:42 +0000 (09:43 +0800)]
powerpc/math-emu: Remove the dead code in math.c

The math.c is only built when CONFIG_MATH_EMULATION is enabled.
So we would never get into the case that CONFIG_MATH_EMULATION
is not defined in this file.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Enable detection of legacy UARTs
Benjamin Herrenschmidt [Mon, 15 Jul 2013 03:03:15 +0000 (13:03 +1000)]
powerpc/powernv: Enable detection of legacy UARTs

Legacy UARTs can exist on PowerNV, memory-mapped ones on PCI
or IO based ones on the LPC bus.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Don't crash if there are no OPAL consoles
Benjamin Herrenschmidt [Mon, 15 Jul 2013 03:03:14 +0000 (13:03 +1000)]
powerpc/powernv: Don't crash if there are no OPAL consoles

Some machines might provide the console via a different mechanism
such as direct access to a UART from Linux, in which case OPAL
might not expose any console. In that case, the code would cause
a NULL dereference.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Check "status" property before adding legacy ISA serial ports
Benjamin Herrenschmidt [Mon, 15 Jul 2013 03:03:13 +0000 (13:03 +1000)]
powerpc: Check "status" property before adding legacy ISA serial ports

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Cleanup udbg_16550 and add support for LPC PIO-only UARTs
Benjamin Herrenschmidt [Mon, 15 Jul 2013 03:03:12 +0000 (13:03 +1000)]
powerpc: Cleanup udbg_16550 and add support for LPC PIO-only UARTs

The udbg_16550 code, which we use for our early consoles and debug
backends was fairly messy. Especially for the debug consoles, it
would re-implement the "high level" getc/putc/poll functions for
each access method. It also had code to configure the UART but only
for the straight MMIO method.

This changes it to instead abstract at the register accessor level,
and have the various functions and configuration routines use these.

The result is simpler and slightly smaller code, and free support
for non-MMIO mapped PIO UARTs, which such as the ones that can be
present on a POWER 8 LPC bus.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Add PIO accessors for Power8 LPC bus
Benjamin Herrenschmidt [Mon, 15 Jul 2013 03:03:11 +0000 (13:03 +1000)]
powerpc/powernv: Add PIO accessors for Power8 LPC bus

This uses the hooks provided by CONFIG_PPC_INDIRECT_PIO to
implement a set of hooks for IO port access to use the LPC
bus via OPAL calls for the first 64K of IO space

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Add helper to get ibm,chip-id of a node
Benjamin Herrenschmidt [Mon, 15 Jul 2013 03:03:10 +0000 (13:03 +1000)]
powerpc/powernv: Add helper to get ibm,chip-id of a node

This includes walking the parent nodes if necessary.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Update opal.h to add new LPC and XSCOM functions
Benjamin Herrenschmidt [Mon, 15 Jul 2013 03:03:09 +0000 (13:03 +1000)]
powerpc/powernv: Update opal.h to add new LPC and XSCOM functions

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Better split CONFIG_PPC_INDIRECT_PIO and CONFIG_PPC_INDIRECT_MMIO
Benjamin Herrenschmidt [Mon, 15 Jul 2013 03:03:08 +0000 (13:03 +1000)]
powerpc: Better split CONFIG_PPC_INDIRECT_PIO and CONFIG_PPC_INDIRECT_MMIO

Remove the generic PPC_INDIRECT_IO and ensure we only add overhead
to the right accessors. IE. If only CONFIG_PPC_INDIRECT_PIO is set,
we don't add overhead to all MMIO accessors.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/ppc64: Rename SOFT_DISABLE_INTS with RECONCILE_IRQ_STATE
Tiejun Chen [Tue, 16 Jul 2013 03:09:30 +0000 (11:09 +0800)]
powerpc/ppc64: Rename SOFT_DISABLE_INTS with RECONCILE_IRQ_STATE

The SOFT_DISABLE_INTS seems an odd name for something that updates the
software state to be consistent with interrupts being hard disabled, so
rename SOFT_DISABLE_INTS with RECONCILE_IRQ_STATE to avoid this confusion.

Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agomacintosh/ams: Replace strict_strtoul() with kstrtoul()
Jingoo Han [Fri, 19 Jul 2013 07:16:56 +0000 (16:16 +0900)]
macintosh/ams: Replace strict_strtoul() with kstrtoul()

The usage of strict_strtoul() is not preferred, because
strict_strtoul() is obsolete. Thus, kstrtoul() should be
used.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pmac: Early debug output on screen on 64-bit macs
Benjamin Herrenschmidt [Thu, 25 Jul 2013 02:12:32 +0000 (12:12 +1000)]
powerpc/pmac: Early debug output on screen on 64-bit macs

We have a bunch of CONFIG_PPC_EARLY_DEBUG_* options that are intended
for bringup/debug only. They hard wire a machine specific udbg backend
very early on (before we even probe the platform), and use whatever
tricks are available on each machine/cpu to be able to get some kind
of output out there early on.

So far, on powermac with no serial ports, we have CONFIG_PPC_EARLY_DEBUG_BOOTX
to use the low-level btext engine on the screen, but it doesn't do much, at
least on 64-bit. It only really gets enabled after the platform has been
probed and the MMU enabled.

This adds a way to enable it much earlier. From prom_init.c (while still
running with Open Firmware), we grab the screen details and set things up
using the physical address of the frame buffer.

Then btext itself uses the "rm_ci" feature of the 970 processor (Real
Mode Cache Inhibited) to access it while in real mode.

We need to do a little bit of reorg of the btext code to inline things
better, in order to limit how much we touch memory while in this mode as
the consequences might be ... interesting.

This successfully allowed me to debug problems early on with the G5
(related to gold being broken vs. ppc64 kernels).

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pci: Remove duplicate check in pcibios_fixup_bus()
Gavin Shan [Wed, 31 Jul 2013 08:43:56 +0000 (16:43 +0800)]
powerpc/pci: Remove duplicate check in pcibios_fixup_bus()

pci_read_bridge_bases() already checks if the PCI bus is root
bus or not, so we needn't do same check in pcibios_fixup_bus()
and just remove it.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Needn't IO segment map for PHB3
Gavin Shan [Wed, 31 Jul 2013 08:47:04 +0000 (16:47 +0800)]
powerpc/powernv: Needn't IO segment map for PHB3

PHB3 doesn't support IO ports and we needn't IO segment map for that.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Check primary PHB through ID
Gavin Shan [Wed, 31 Jul 2013 08:47:02 +0000 (16:47 +0800)]
powerpc/powernv: Check primary PHB through ID

The index of one specific PCI controller (struct pci_controller::
global_number) can tell that it's primary one or not. So we needn't
additional variable for that and just remove it.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Fetch PHB bus range from dev-tree
Gavin Shan [Wed, 31 Jul 2013 08:47:01 +0000 (16:47 +0800)]
powerpc/powernv: Fetch PHB bus range from dev-tree

The patch enables fetching bus range from device-tree for the
specific PHB. If we can't get that from device-tree, the default
range [0 255] will be used.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/powernv: Free PHB instance upon error
Gavin Shan [Wed, 31 Jul 2013 08:47:00 +0000 (16:47 +0800)]
powerpc/powernv: Free PHB instance upon error

We don't free PHB instance (struct pnv_phb) on error to creating
the associated PCI controler (struct pci_controller). The patch
fixes that. Also, the output messages have been cleaned for a bit
so that they looks unified for IODA_1/2 cases.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Fix VRSAVE handling
Paul Mackerras [Mon, 5 Aug 2013 04:13:16 +0000 (14:13 +1000)]
powerpc: Fix VRSAVE handling

Since 2002, the kernel has not saved VRSAVE on exception entry and
restored it on exit; rather, VRSAVE gets context-switched in _switch.
This means that when executing in process context in the kernel, the
userspace VRSAVE value is live in the VRSAVE register.

However, the signal code assumes that current->thread.vrsave holds
the current VRSAVE value, which is incorrect.  Therefore, this
commit changes it to use the actual VRSAVE register instead.  (It
still uses current->thread.vrsave as a temporary location to store
it in, as __get_user and __put_user can only transfer to/from a
variable, not an SPR.)

This also modifies the transactional memory code to save and restore
VRSAVE regardless of whether VMX is enabled in the MSR.  This is
because accesses to VRSAVE are not controlled by the MSR.VEC bit,
but can happen at any time.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Implement __get_user_pages_fast()
Paul Mackerras [Mon, 5 Aug 2013 04:11:23 +0000 (14:11 +1000)]
powerpc: Implement __get_user_pages_fast()

Other architectures have a __get_user_pages_fast(), in addition to the
regular get_user_pages_fast(), which doesn't call get_user_pages() on
failure, and thus doesn't attempt to fault pages in or COW them.  The
generic KVM code uses __get_user_pages_fast() to detect whether a page
for which we have only requested read access is actually writable.

This provides an implementation of __get_user_pages_fast() by
splitting the existing get_user_pages_fast() in two.  With this, the
generic KVM code will get the right answer instead of always
considering such pages non-writable.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agoselftests: Add test of PMU instruction counting on powerpc
Michael Ellerman [Tue, 6 Aug 2013 07:42:37 +0000 (17:42 +1000)]
selftests: Add test of PMU instruction counting on powerpc

This commit adds a test of instruction counting using the PMU on powerpc.

Although the bulk of the code is architecture agnostic, the code needs to
run a precisely sized loop which is implemented in assembler.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agoselftests: Add support files for powerpc tests
Michael Ellerman [Tue, 6 Aug 2013 07:42:36 +0000 (17:42 +1000)]
selftests: Add support files for powerpc tests

This commit adds support code used by upcoming powerpc tests.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agoselftests: Add infrastructure for powerpc selftests
Michael Ellerman [Tue, 6 Aug 2013 07:42:35 +0000 (17:42 +1000)]
selftests: Add infrastructure for powerpc selftests

This commit adds a powerpc subdirectory to tools/testing/selftests,
for tests that are powerpc specific.

On other architectures nothing is built. The makefile supports cross
compilation if the user sets ARCH and CROSS_COMPILE.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Convert platforms to smp_generic_cpu_bootable
Andy Fleming [Mon, 5 Aug 2013 19:58:35 +0000 (14:58 -0500)]
powerpc: Convert platforms to smp_generic_cpu_bootable

T4, Cell, powernv, and pseries had the same implementation, so switch
them to use a generic version. A2 apparently had a version, but
removed it at some point, so we remove the declaration, too.

Signed-off-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Add smp_generic_cpu_bootable
Andy Fleming [Mon, 5 Aug 2013 19:58:34 +0000 (14:58 -0500)]
powerpc: Add smp_generic_cpu_bootable

Cell and PSeries both implemented their own versions of a
cpu_bootable smp_op which do the same thing (well, the PSeries
one has support for more than 2 threads). Copy the PSeries one
to generic code, and rename it smp_generic_cpu_bootable.

Signed-off-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Remove the symbol __flush_icache_range
Kevin Hao [Tue, 6 Aug 2013 10:23:31 +0000 (18:23 +0800)]
powerpc: Remove the symbol __flush_icache_range

And now the function flush_icache_range() is just a wrapper which
only invoke the function __flush_icache_range() directly. So we
don't have reason to keep it anymore.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Move the testing of CPU_FTR_COHERENT_ICACHE into __flush_icache_range
Kevin Hao [Tue, 6 Aug 2013 10:23:30 +0000 (18:23 +0800)]
powerpc: Move the testing of CPU_FTR_COHERENT_ICACHE into __flush_icache_range

In function flush_icache_range(), we use cpu_has_feature() to test
the feature bit of CPU_FTR_COHERENT_ICACHE. But this seems not optimal
for two reasons:
 a) For ppc32, the function __flush_icache_range() already do this
    check with the macro END_FTR_SECTION_IFSET.
 b) Compare with the cpu_has_feature(), the method of using macro
    END_FTR_SECTION_IFSET will not introduce any runtime overhead.

[And while at it, add the missing required isync] -- BenH

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Stop using non-architected shared_proc field in lppaca
Anton Blanchard [Tue, 6 Aug 2013 16:01:26 +0000 (02:01 +1000)]
powerpc: Stop using non-architected shared_proc field in lppaca

Although the shared_proc field in the lppaca works today, it is
not architected. A shared processor partition will always have a non
zero yield_count so use that instead. Create a wrapper so users
don't have to know about the details.

In order for older kernels to continue to work on KVM we need
to set the shared_proc bit. While here, remove the ugly bitfield.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pci: Don't use bitfield for force_32bit_msi
Anton Blanchard [Tue, 6 Aug 2013 16:01:25 +0000 (02:01 +1000)]
powerpc/pci: Don't use bitfield for force_32bit_msi

Fix a sparse warning about force_32bit_msi being a one bit bitfield.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc: Fix a number of sparse warnings
Anton Blanchard [Tue, 6 Aug 2013 16:01:24 +0000 (02:01 +1000)]
powerpc: Fix a number of sparse warnings

Address some of the trivial sparse warnings in arch/powerpc.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pseries: Simplify H_GET_TERM_CHAR
Anton Blanchard [Tue, 6 Aug 2013 16:01:23 +0000 (02:01 +1000)]
powerpc/pseries: Simplify H_GET_TERM_CHAR

plpar_get_term_char is only used once and just adds a layer
of complexity to H_GET_TERM_CHAR. plpar_put_term_char isn't
used at all so we can remove it.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>