platform/kernel/linux-rpi.git
12 years agopowerpc: Move GE PIC drivers
Martyn Welch [Mon, 12 Mar 2012 17:12:59 +0000 (17:12 +0000)]
powerpc: Move GE PIC drivers

Move the GE PIC drivers to allow these to be used by non-86xx boards.

Signed-off-by: Martyn Welch <martyn.welch@ge.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agogpio: Move GE GPIO driver to reside within GPIO subsystem
Martyn Welch [Mon, 12 Mar 2012 17:12:58 +0000 (17:12 +0000)]
gpio: Move GE GPIO driver to reside within GPIO subsystem

The GE GPIO driver provides basic support (set direction, read/write state)
for the GPIO provided on some GE single board computers. This patch moves
the driver from the 86xx specific platform directrory to the GPIO subsystem
so that it can be used on non-86xx boards.

Signed-off-by: Martyn Welch <martyn.welch@ge.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc: Add GE FPGA config option
Martyn Welch [Mon, 12 Mar 2012 17:12:57 +0000 (17:12 +0000)]
powerpc: Add GE FPGA config option

This patch adds the GE_FPGA configuration option. This is being carried
out as ground work to allow the PIC and GPIO drivers to be move from the
powerpc 86xx platform directory to more general locations to allow them to
be used on non-86xx boards and to reduce churn when further boards using
these drivers are added.

Signed-off-by: Martyn Welch <martyn.welch@ge.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add dts for p1020rdb-pc board
Zhicheng Fan [Fri, 10 Feb 2012 06:48:16 +0000 (14:48 +0800)]
powerpc/85xx: Add dts for p1020rdb-pc board

P1020RDB-PC Overview
------------------
1Gbyte DDR3 SDRAM
32 Mbyte NAND flash
10 16Mbyte NOR flash
16 Mbyte SPI flash
SD connector to interface with the SD memory card
Real-time clock on I2C bus

PCIe:
- x1 PCIe slot
- x1 mini-PCIe slot

10/100/1000 BaseT Ethernet ports:
- eTSEC1, RGMII: one 10/100/1000 port using VitesseTM VSC7385 L2 switch
- eTSEC2, SGMII: one 10/100/1000 port using VitesseTM VSC8221
- eTSEC3, RGMII: one 10/100/1000 port using AtherosTM AR8021

USB 2.0 port:
- Two USB2.0 Type A receptacles
- One USB2.0 signal to Mini PCIe slot

Dual RJ45 UART ports:
- DUART interface: supports two UARTs up to 115200 bps for console display

Signed-off-by: Zhicheng Fan <b32736@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add p1020rdb-pc platform support
Zhicheng Fan [Fri, 10 Feb 2012 06:48:15 +0000 (14:48 +0800)]
powerpc/85xx: Add p1020rdb-pc platform support

Signed-off-by: Zhicheng Fan <b32736@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: add P1020UTM-PC platform support
Jerry Huang [Wed, 14 Mar 2012 09:08:28 +0000 (17:08 +0800)]
powerpc/85xx: add P1020UTM-PC platform support

The p1020utm-pc has the similar feature as the p1020rdb.
Therefore, p1020utm-pc use the same platform file as the p1/p2 rdb board.
Overview of P1020UTM-PC platform:
        - DDR3 1GB
        - NOR flash 32MB
        - I2C EEPROM 256Kb
        - eTSEC1 (RGMII PHY Atheros AR8021)
        - eTSEC2 (SGMII PHY Vitesse VSC8221)
        - eTSEC3 (RGMII PHY Atheros AR8021)
        - SDHC
        - 2 USB ports
        - PCIe (Lane1 to dual SATA controller)

Signed-off-by: Jerry Huang <Chang-Ming.Huang@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: add P1020MBG-PC platform support
Jerry Huang [Wed, 14 Mar 2012 09:08:27 +0000 (17:08 +0800)]
powerpc/85xx: add P1020MBG-PC platform support

The p1020mbg-pc has the similar feature as the p1020rdb.
Therefore, p1020mbg-pc use the same platform file as the p1/p2 rdb board.
Overview of P1020MBG-PC platform:
        - DDR3 2GB
        - NOR flash 64MB
        - I2C EEPROM 256Kb
        - eTSEC1 (RGMII PHY) connected to VSC7385 L2 switch
        - eTSEC2 (SGMII PHY)
        - eTSEC3 (RGMII PHY)
        - SDHC
        - 2 USB ports
        - 4 TDM ports
        - PCIe (Lane1 to dual SATA controller)

Signed-off-by: Jerry Huang <Chang-Ming.Huang@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agoNAND Machine support for Integrated Flash Controller
Prabhakar Kushwaha [Thu, 15 Mar 2012 05:34:23 +0000 (11:04 +0530)]
NAND Machine support for Integrated Flash Controller

Integrated Flash Controller(IFC) can be used to hook NAND Flash
chips using NAND Flash Machine available on it.

Signed-off-by: Dipen Dudhat <Dipen.Dudhat@freescale.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Li Yang <leoli@freescale.com>
Signed-off-by: Liu Shuo <b35362@freescale.com>
Signed-off-by: Poonam Aggrwal <poonam.aggrwal@freescale.com>
Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Clean up partition nodes in dts for MPC8572DS
Jia Hongtao [Tue, 21 Feb 2012 02:11:23 +0000 (10:11 +0800)]
powerpc/85xx: Clean up partition nodes in dts for MPC8572DS

Signed-off-by: Jin Qing <b24347@freescale.com>
Signed-off-by: Jia Hongtao <B38951@freescale.com>
Signed-off-by: Li Yang <leoli@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: p1022ds: disable the NOR flash node if video is enabled
Timur Tabi [Thu, 16 Feb 2012 00:25:47 +0000 (18:25 -0600)]
powerpc/85xx: p1022ds: disable the NOR flash node if video is enabled

The Freescale P1022 has a unique pin muxing "feature" where the DIU video
controller's video signals are muxed with 24 of the local bus address signals.
When the DIU is enabled, the bulk of the local bus is disabled, preventing
access to memory-mapped devices like NOR flash and the pixis FPGA.

Therefore, if the DIU is going to be enabled, then memory-mapped devices on
the localbus, like NOR flash, need to be disabled.

This also means that the localbus is not a 'simple-bus' any more, so remove
that string from the compatible node.

Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: create 32-bit DTS for the P1022DS
Timur Tabi [Thu, 16 Feb 2012 00:25:48 +0000 (18:25 -0600)]
powerpc/85xx: create 32-bit DTS for the P1022DS

Create a 32-bit address space version of p1022ds.dts.  To avoid confusion,
p1022ds.dts is renamed to p1022ds_36b.dts.  We also create p1022ds.dtsi
to store some common nodes.

Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add magic-packet properties for etsec
Xie Xiaobo [Tue, 17 Jan 2012 09:59:51 +0000 (17:59 +0800)]
powerpc/85xx: Add magic-packet properties for etsec

The properties indicates that the hardware supports waking up via magic
packet.

Signed-off-by: Xie Xiaobo <X.Xie@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add some DTS nodes and attributes for mpc8536ds
Xie Xiaobo [Tue, 17 Jan 2012 09:59:50 +0000 (17:59 +0800)]
powerpc/85xx: Add some DTS nodes and attributes for mpc8536ds

Add partitions for NOR and NAND Flash.

Signed-off-by: Xie Xiaobo <X.Xie@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/fsl_msi: return proper error value when ioremap failed.
Liu Shuo [Thu, 8 Mar 2012 22:47:37 +0000 (14:47 -0800)]
powerpc/fsl_msi: return proper error value when ioremap failed.

Signed-off-by: Liu Shuo <soniccat.liu@gmail.com>
Acked-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: fix typo in p1010rdb.dtsi
Gustavo Zacarias [Tue, 28 Feb 2012 19:43:08 +0000 (16:43 -0300)]
powerpc/85xx: fix typo in p1010rdb.dtsi

Fix typo introduced by "powerpc: Add TBI PHY node to first MDIO bus"
from Andy Fleming.
It's device_type rather than device-type, which causes the mdio probe to
fail thus making all gianfar ethernet interfaces unusable.

Signed-off-by: Gustavo Zacarias <gustavo@zacarias.com.ar>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: p2020rdb & p1010rdb - lower spi flash freq to 40Mhz
Sebastian Andrzej Siewior [Thu, 15 Mar 2012 17:40:28 +0000 (18:40 +0100)]
powerpc/85xx: p2020rdb & p1010rdb - lower spi flash freq to 40Mhz

This is here most likely since the FSL bsp. Back in the FSL bsp it was
set to 50Mhz and working. However the driver divided the SoC freq. only
by 2. According to the TRM the platform clock (which the manual refers
in its formula) is the system clock divided by two. So in the end it has
to divide by 4 and this is what the fsl-spi driver in tree is doing.
Since then the flash is not wokring I guess. After chaning the freq from
50Mhz to 40Mhz like others do then I can access the flash.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: p2020rdb - move the NAND address.
Sebastian Andrzej Siewior [Thu, 15 Mar 2012 17:40:27 +0000 (18:40 +0100)]
powerpc/85xx: p2020rdb - move the NAND address.

It is not at 0xffa00000. According to current u-boot source the NAND
controller is always at 0xff800000 and it is either at CS0 or CS1
depending on NAND or NAND+NOR mode. In 36bit mode it is shifted to
0xfff800000 but it has always an eight there and never an A.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/srio: Fix the compile errors when building with 64bit
Liu Gang [Fri, 9 Mar 2012 08:10:38 +0000 (16:10 +0800)]
powerpc/srio: Fix the compile errors when building with 64bit

For the file "arch/powerpc/sysdev/fsl_rmu.c", there will be some compile
errors while using the corenet64_smp_defconfig:

.../fsl_rmu.c:315: error: cast from pointer to integer of different size
.../fsl_rmu.c:320: error: cast to pointer from integer of different size
.../fsl_rmu.c:320: error: cast to pointer from integer of different size
.../fsl_rmu.c:320: error: cast to pointer from integer of different size
.../fsl_rmu.c:330: error: cast to pointer from integer of different size
.../fsl_rmu.c:332: error: cast to pointer from integer of different size
.../fsl_rmu.c:339: error: cast to pointer from integer of different size
.../fsl_rmu.c:340: error: cast to pointer from integer of different size
.../fsl_rmu.c:341: error: cast to pointer from integer of different size
.../fsl_rmu.c:348: error: cast to pointer from integer of different size
.../fsl_rmu.c:348: error: cast to pointer from integer of different size
.../fsl_rmu.c:348: error: cast to pointer from integer of different size
.../fsl_rmu.c:659: error: cast from pointer to integer of different size
.../fsl_rmu.c:659: error: format '%8.8x' expects type 'unsigned int',
                   but argument 5 has type 'size_t'
.../fsl_rmu.c:985: error: cast from pointer to integer of different size
.../fsl_rmu.c:997: error: cast to pointer from integer of different size

Rewrote the corresponding code with the support of 64bit building.

Signed-off-by: Liu Gang <Gang.Liu@freescale.com>
Signed-off-by: Shaohui Xie <Shaohui.Xie@freescale.com>
Reported-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/srio: Fix the relocation errors when building with 64bit
Liu Gang [Tue, 6 Mar 2012 02:58:12 +0000 (10:58 +0800)]
powerpc/srio: Fix the relocation errors when building with 64bit

For the file "arch/powerpc/sysdev/fsl_rio.c", there will be some relocation
errors while using the corenet64_smp_defconfig:

WARNING: modpost: Found 6 section mismatch(es).
To see full details build your kernel with:
'make CONFIG_DEBUG_SECTION_MISMATCH=y'
  GEN     .version
  CHK     include/generated/compile.h
  UPD     include/generated/compile.h
  CC      init/version.o
  LD      init/built-in.o
  LD      .tmp_vmlinux1
arch/powerpc/sysdev/built-in.o:(__ex_table+0x0):
relocation truncated to fit: R_PPC64_ADDR16 against `.text'+3208
arch/powerpc/sysdev/built-in.o:(__ex_table+0x2):
relocation truncated to fit: R_PPC64_ADDR16 against `.fixup'
arch/powerpc/sysdev/built-in.o:(__ex_table+0x4):
relocation truncated to fit: R_PPC64_ADDR16 against `.text'+3230
arch/powerpc/sysdev/built-in.o:(__ex_table+0x6):
relocation truncated to fit: R_PPC64_ADDR16 against `.fixup'+c
arch/powerpc/sysdev/built-in.o:(__ex_table+0x8):
relocation truncated to fit: R_PPC64_ADDR16 against `.text'+3250
arch/powerpc/sysdev/built-in.o:(__ex_table+0xa):
relocation truncated to fit: R_PPC64_ADDR16 against `.fixup'+18

Rewrote the corresponding code with the support of 64bit building.

Signed-off-by: Liu Gang <Gang.Liu@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add dts for p1025rdb board
Zhicheng Fan [Mon, 13 Feb 2012 22:06:23 +0000 (22:06 +0000)]
powerpc/85xx: Add dts for p1025rdb board

P1025RDB Overview
------------------
1Gbyte DDR3 SDRAM
32 Mbyte NAND flash
16Mbyte NOR flash
16 Mbyte SPI flash
SD connector to interface with the SD memory card
Real-time clock on I2C bus

PCIe:
- x1 PCIe slot
- x1 mini-PCIe slot

10/100/1000 BaseT Ethernet ports:
- eTSEC1, RGMII: one 10/100/1000 port using AtherosTM AR8021
- eTSEC2, SGMII: one 10/100/1000 port using VitesseTM VSC8221
- eTSEC3, RGMII: one 10/100/1000 port using AtherosTM AR8021

USB 2.0 port:
- Two USB2.0 Type A receptacles
- One USB2.0 signal to Mini PCIe slot

Dual RJ45 UART ports:
- DUART interface: supports two UARTs up to 115200 bps for console display

Signed-off-by: Zhicheng Fan <b32736@freescale.com>
Acked-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add p1025rdb platform support
Zhicheng Fan [Mon, 13 Feb 2012 22:06:22 +0000 (22:06 +0000)]
powerpc/85xx: Add p1025rdb platform support

Signed-off-by: Zhicheng Fan <b32736@freescale.com>
Acked-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add usb controller version info
Ramneek Mehresh [Wed, 18 Jan 2012 05:40:48 +0000 (11:10 +0530)]
powerpc/85xx: Add usb controller version info

Add usb controller version info for the following:
MPC8536, P1010, P1020, P1021, P1022, P1023, P2020, P2041,
P3041, P3060, P5020

Signed-off-by: Ramneek Mehresh <ramneek.mehresh@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add p2020rdb-pc dts support
Tang Yuantian [Thu, 9 Feb 2012 21:59:57 +0000 (21:59 +0000)]
powerpc/85xx: Add p2020rdb-pc dts support

Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Signed-off-by: Poonam Aggrwal <poonam.aggrwal@freescale.com>
Signed-off-by: Tang Yuantian <b29983@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Adds Support for P2020RDB-PC board
Tang Yuantian [Wed, 28 Dec 2011 03:41:47 +0000 (11:41 +0800)]
powerpc/85xx: Adds Support for P2020RDB-PC board

P2020RDB-PC Board shares the same design(PCB) as P102x RDB style platforms.
The difference between this platform and the already existing P2020RDB
is mainly with respect to DDR. The P2020RDB-PC has a DDR3 memory.
The P2020RDB-PC also has a CPLD device connected to local bus.

The main differences from the P102x RDB-PC is 64-bit DDR and SYSCLK of
100Mhz.

Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Signed-off-by: Poonam Aggrwal <poonam.aggrwal@freescale.com>
Signed-off-by: Tang Yuantian <b29983@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Added P1021RDB-PC Platform support
Xu Jiucheng [Tue, 17 Jan 2012 08:01:30 +0000 (16:01 +0800)]
powerpc/85xx: Added P1021RDB-PC Platform support

Signed-off-by: Xu Jiucheng <B37781@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Added dts for P1021RDB-PC board
Xu Jiucheng [Tue, 17 Jan 2012 08:01:29 +0000 (16:01 +0800)]
powerpc/85xx: Added dts for P1021RDB-PC board

P1021RDB-PC Overview
-----------------
1Gbyte DDR3 (on board DDR)
16Mbyte NOR flash
32Mbyte eSLC NAND Flash
256 Kbit M24256 I2C EEPROM
128 Mbit SPI Flash memory
Real-time clock on I2C bus
SD/MMC connector to interface with the SD memory card
PCIex
    - x1 PCIe slot or x1 PCIe to dual SATA controller
    - x1 mini-PCIe slot
USB 2.0
    - ULPI PHY interface: SMSC USB3300 USB PHY and Genesys Logic’s GL850A
    - Two USB2.0 Type A receptacles
    - One USB2.0 signal to Mini PCIe slot
eTSEC1: Connected to RGMII PHY VSC7385
eTSEC2: Connected to SGMII PHY VSC8221
eTSEC3: Connected to SGMII PHY AR8021
DUART interface: supports two UARTs up to 115200 bps for console display

Signed-off-by: Matthew McClintock <msm@freescale.com>
Signed-off-by: Xu Jiucheng <B37781@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc: Add initial e6500 cpu support
Kumar Gala [Sun, 6 Nov 2011 17:51:07 +0000 (11:51 -0600)]
powerpc: Add initial e6500 cpu support

Add basic support for e6500 core in its single threaded mode.

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/fsl-booke: Fixup calc_cam_sz to support MMU v2
Kumar Gala [Thu, 5 Jan 2012 18:37:16 +0000 (12:37 -0600)]
powerpc/fsl-booke: Fixup calc_cam_sz to support MMU v2

The registers that describe size supported by TLB are different on MMU
v2 as well as we support power of two page sizes.  For now we continue
to assume that FSL variable size array supports all page sizes up to the
maximum one reported in TLB1PS.

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: fix Kconfig warning about missing 8250 dependency
Paul Gortmaker [Fri, 20 Jan 2012 01:23:20 +0000 (20:23 -0500)]
powerpc/85xx: fix Kconfig warning about missing 8250 dependency

The SERIAL_8250_EXTENDED option just enables access to other
less regularly used options, like SERIAL_8250_SHARE_IRQ.
Select it to get rid of this warning when selecting the child
option living underneath it.

  warning: (FSL_SOC_BOOKE && SERIAL_8250_RM9K) selects
  SERIAL_8250_SHARE_IRQ which has unmet direct dependencies
  (HAS_IOMEM && SERIAL_8250_EXTENDED)

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agoMerge branch 'eeh' into next
Benjamin Herrenschmidt [Mon, 12 Mar 2012 23:15:35 +0000 (10:15 +1100)]
Merge branch 'eeh' into next

12 years agopowerpc: Rework lazy-interrupt handling
Benjamin Herrenschmidt [Tue, 6 Mar 2012 07:27:59 +0000 (18:27 +1100)]
powerpc: Rework lazy-interrupt handling

The current implementation of lazy interrupts handling has some
issues that this tries to address.

We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.

The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.

Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.

This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.

The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.

When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.

We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).

This removes the need to play with the decrementer to try to create
fake interrupts, among others.

In addition, this adds a few refinements:

 - We no longer  hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.

 - Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.

 - On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)

 - We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.

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

v2:

- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
  to retrigger an interrupt without preventing hard-enable

v3:

 - Fix or vs. ori bug on Book3E
 - Fix enabling of interrupts for some exceptions on Book3E

v4:

 - Fix resend of doorbells on return from interrupt on Book3E

v5:

 - Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.

v6:
 - 32-bit compile fix
 - more compile fixes with various .config combos
 - factor out the asm code to soft-disable interrupts
 - remove the C wrapper around preempt_schedule_irq

v7:
 - Fix a bug with hard irq state tracking on native power7

12 years agopowerpc/eeh: pseries platform config space access in EEH
Gavin Shan [Mon, 27 Feb 2012 20:04:11 +0000 (20:04 +0000)]
powerpc/eeh: pseries platform config space access in EEH

With the original EEH implementation, the access to config space of
the corresponding PCI device is done by RTAS sensitive function. That
depends on pci_dn heavily. That would limit EEH extension to other
platforms like powernv because other platforms might have different
ways to access PCI config space.

The patch splits those functions used to access PCI config space
and implement them in platform related EEH component. It would be
helpful to support EEH on multiple platforms simutaneously in future.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Introduce struct eeh_stats for EEH
Gavin Shan [Wed, 29 Feb 2012 15:47:45 +0000 (15:47 +0000)]
powerpc/eeh: Introduce struct eeh_stats for EEH

With the original EEH implementation, the EEH global statistics
are maintained by individual global variables. That makes the
code a little hard to maintain.

The patch introduces extra struct eeh_stats for the EEH global
statistics so that it can be maintained in collective fashion.

It's the rework on the corresponding v5 patch. According to
the comments from David Laight, the EEH global statistics have
been changed for a litte bit so that they have fixed-type of
"u64". Also, the format used to print them has been changed to
"%llu" based on David's suggestion. Also, the output format of
EEH global statistics should be kept as intacted according to
Michael's suggestion that there might be tools parsing them.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Replace pci_dn with eeh_dev for EEH on pSeries
Gavin Shan [Mon, 27 Feb 2012 20:04:09 +0000 (20:04 +0000)]
powerpc/eeh: Replace pci_dn with eeh_dev for EEH on pSeries

The pci_dn has been replaced with eeh_dev. In order to comply with
the rule, the EEH platform implementation on pSeries should also
be adjusted for a little bit so that it will depend on eeh_dev instead
of pci_dn.

The patch replaces pci_dn with eeh_dev. The corresponding information
will be retrieved from eeh_dev instead of pci_dn.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Replace pci_dn with eeh_dev for EEH aux components
Gavin Shan [Mon, 27 Feb 2012 20:04:08 +0000 (20:04 +0000)]
powerpc/eeh: Replace pci_dn with eeh_dev for EEH aux components

The original EEH implementation is heavily depending on struct pci_dn.
We have to put EEH related information to pci_dn. Actually, we could
split struct pci_dn so that the EEH sensitive information to form an
individual struct, then EEH looks more independent.

The patch replaces pci_dn with eeh_dev for EEH aux components like
event and driver. Also, the eeh_event struct has been adjusted for
a little bit since eeh_dev has linked the associated FDT (Flat Device
Tree) node and PCI device. It's not necessary for eeh_event struct to
trace FDT node and PCI device. We can just simply to trace eeh_dev in
eeh_event.

The patch also renames function pcid_name() to eeh_pcid_name(), which
should be missed in the previous patch where the EEH aux components
have been cleaned up.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Replace pci_dn with eeh_dev for EEH core
Gavin Shan [Mon, 27 Feb 2012 20:04:07 +0000 (20:04 +0000)]
powerpc/eeh: Replace pci_dn with eeh_dev for EEH core

The original EEH implementation is heavily depending on struct pci_dn.
We have to put EEH related information to pci_dn. Actually, we could
split struct pci_dn so that the EEH sensitive information to form an
individual struct, then EEH looks more independent.

The patch replaces pci_dn with eeh_dev for EEH core.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Replace pci_dn with eeh_dev for EEH address cache
Gavin Shan [Mon, 27 Feb 2012 20:04:06 +0000 (20:04 +0000)]
powerpc/eeh: Replace pci_dn with eeh_dev for EEH address cache

With original EEH implementation, struct pci_dn is used while building
PCI I/O address cache, which helps on searching the corresponding
PCI device according to the given physical I/O address. Besides, pci_dn
is associated with the corresponding PCI device while building its
I/O cache.

The patch replaces struct pci_dn with struct eeh_dev so that EEH address
cache won't depend on struct pci_dn. That will help EEH to become an
independent module in future. Besides, the binding of eeh_dev and PCI
device is done while building PCI device I/O cache.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Replace pci_dn with eeh_dev for EEH sysfs
Gavin Shan [Mon, 27 Feb 2012 20:04:05 +0000 (20:04 +0000)]
powerpc/eeh: Replace pci_dn with eeh_dev for EEH sysfs

With original EEH implementation, all EEH related statistics have
been put into struct pci_dn. We've introduced struct eeh_dev to
replace struct pci_dn in EEH core components, including EEH sysfs
component.

The patch shows EEH statistics from struct eeh_dev instead of struct
pci_dn in EEH sysfs component. Besides, it also fixed the EEH device
retrieval from PCI device, which was introduced by the previous patch
in the series of patch.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Introduce EEH device
Gavin Shan [Mon, 27 Feb 2012 20:04:04 +0000 (20:04 +0000)]
powerpc/eeh: Introduce EEH device

Original EEH implementation depends on struct pci_dn heavily. However,
EEH shouldn't depend on that actually because EEH needn't share much
information with other PCI components. That's to say, EEH should have
worked independently.

The patch introduces struct eeh_dev so that EEH core components needn't
be working based on struct pci_dn in future. Also, struct pci_dn, struct
eeh_dev instances are created in dynamic fasion and the binding with EEH
device, OF node, PCI device is implemented as well.

The EEH devices are created after PHBs are detected and initialized, but
PCI emunation hasn't started yet. Apart from that, PHB might be created
dynamically through DLPAR component and the EEH devices should be creatd
as well. Another case might be OF node is created dynamically by DR
(Dynamic Reconfiguration), which has been defined by PAPR. For those OF
nodes created by DR, EEH devices should be also created accordingly. The
binding between EEH device and OF node is done while the EEH device is
initially created.

The binding between EEH device and PCI device should be done after PCI
emunation is done. Besides, PCI hotplug also needs the binding so that
the EEH devices could be traced from the newly coming PCI buses or PCI
devices.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Cleanup function names in EEH aux components
Gavin Shan [Mon, 27 Feb 2012 20:04:03 +0000 (20:04 +0000)]
powerpc/eeh: Cleanup function names in EEH aux components

The patch does some cleanup on the function names of EEH
aux components. Currently, only couple of function names from
eeh_cache have been adjusted so that:

        * The function name has prefix "eeh_addr_cache".
        * Move around pci_addr_cache_build() in the header file
          to reflect function call sequence.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/pseries: Cleanup comments in EEH aux components
Gavin Shan [Mon, 27 Feb 2012 20:04:02 +0000 (20:04 +0000)]
powerpc/pseries: Cleanup comments in EEH aux components

There're several EEH aux components and the patch does some cleanup
for them so that they look more clean.

        * Duplicated comments have been removed from the header file.
        * Comments have been reorganized so that it looks more clean.
        * The leading comments of functions are adjusted for a little
          bit so that the result of "make pdfdocs" would be more
          unified.
        * Function calls "xxx ()" has been replaced by "xxx()".

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: pseries platform EEH configure bridge
Gavin Shan [Mon, 27 Feb 2012 20:04:01 +0000 (20:04 +0000)]
powerpc/eeh: pseries platform EEH configure bridge

In order to enable particular PCI device, which has been included
in the parent PE. The involved PCI bridges should be enabled explicitly
if there has. On pSeries platform, there're dedicated RTAS calls
to fulfil the purpose.

The patch implements the function of configuring PCI bridges through
the dedicated RTAS calls. Besides, the function has been abstracted
by struct eeh_ops::configure_bridge so that the EEH core components
could support multiple platforms in future.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: pseries platform EEH error log retrieval
Gavin Shan [Mon, 27 Feb 2012 20:04:00 +0000 (20:04 +0000)]
powerpc/eeh: pseries platform EEH error log retrieval

On RTAS compliant pSeries platform, one dedicated RTAS call has
been introduced to retrieve EEH temporary or permanent error log.

The patch implements the function of retriving EEH error log through
RTAS call. Besides, it has been abstracted by struct eeh_ops::get_log
so that EEH core components could support multiple platforms in future.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: pseries platform EEH reset PE
Gavin Shan [Mon, 27 Feb 2012 20:03:59 +0000 (20:03 +0000)]
powerpc/eeh: pseries platform EEH reset PE

On RTAS compliant pSeries platform, there is a dedicated RTAS call
(ibm,set-slot-reset) to reset the specified PE. Furthermore, two
types of resets are supported: hot and fundamental. the type of
reset is to be used actually depends on the included PCI device's
requirements.

The patch implements resetting PE on pSeries platform through RTAS
call. Besides, it has been abstracted through struct eeh_ops::reset
so that EEH core components could support multiple platforms in future.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: pseries platform EEH wait PE state
Gavin Shan [Mon, 27 Feb 2012 20:03:58 +0000 (20:03 +0000)]
powerpc/eeh: pseries platform EEH wait PE state

On pSeries platform, the PE state might be temporarily unavailable.
In that case, the firmware will return the corresponding wait time.
That means the kernel has to wait for appropriate time in order to
get the PE state.

The patch does the implementation for that. Besides, the function
has been abstracted through struct eeh_ops::wait_state so that EEH core
components could support multiple platforms in future.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: pseries platform PE state retrieval
Gavin Shan [Mon, 27 Feb 2012 20:03:57 +0000 (20:03 +0000)]
powerpc/eeh: pseries platform PE state retrieval

On pSeries platform, there're 2 dedicated RTAS calls introduced to
retrieve the corresponding PE's state: ibm,read-slot-reset-state and
ibm,read-slot-reset-state2.

The patch implements the retrieval of PE's state according to the
given PE address. Besides, the implementation has been abstracted by
struct eeh_ops::get_state so that EEH core components could support
multiple platforms in future.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: pseries platform EEH PE address retrieval
Gavin Shan [Mon, 27 Feb 2012 20:03:56 +0000 (20:03 +0000)]
powerpc/eeh: pseries platform EEH PE address retrieval

There're 2 types of addresses used for EEH operations. The first
one would be BDF (Bus/Device/Function) address which is retrieved
from the reg property of the corresponding FDT node. Another one
is PE address that should be enquired from firmware through RTAS
call on pSeries platform. When issuing EEH operation, the PE address
has precedence over BDF address.

The patch implements retrieving PE address according to the given
BDF address on pSeries platform. Also, the struct eeh_early_enable_info
has been removed since the information can be figured out from
dn->pdn->phb->buid directly and that simplifies the code.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: pseries platform EEH operations
Gavin Shan [Mon, 27 Feb 2012 20:03:55 +0000 (20:03 +0000)]
powerpc/eeh: pseries platform EEH operations

There're 4 EEH operations that are covered by the dedicated RTAS
call <ibm,set-eeh-option>: enable or disable EEH, enable MMIO and
enable DMA. At early stage of system boot, the EEH would be tried
to enable on PCI device related device node. MMIO and DMA for
particular PE should be enabled when doing recovery on EEH errors
so that the PE could function properly again.

The patch implements it and abstract that through struct
eeh_ops::set_eeh. It would be help for EEH to support multiple
platforms in future.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: pseries platform EEH initialization
Gavin Shan [Mon, 27 Feb 2012 20:03:54 +0000 (20:03 +0000)]
powerpc/eeh: pseries platform EEH initialization

The platform specific EEH operations have been abstracted by
struct eeh_ops. The individual platroms, including pSeries, needs
doing necessary initialization before the platform dependent EEH
operations work properly.

The patch is addressing that and do necessary platform initialization
for pSeries platform. More specificly, it will figure out the tokens
of EEH related RTAS calls.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Platform dependent EEH operations
Gavin Shan [Mon, 27 Feb 2012 20:03:53 +0000 (20:03 +0000)]
powerpc/eeh: Platform dependent EEH operations

EEH has been implemented on RTAS-compliant pSeries platform.
That's to say, the EEH operations will be implemented through RTAS
calls eventually. The situation limited feasible extension on EEH.
In order to support EEH on multiple platforms like pseries and powernv
simutaneously. We have to split the platform dependent EEH options
up out of current implementation.

The patch addresses supporting EEH on multiple platforms. The pseries
platform dependent EEH operations will be abstracted by struct eeh_ops.
EEH core components will be built based on the registered EEH operations.
With the mechanism, what the individual platform needs to do is implement
platform dependent EEH operations.

For now, the pseries platform is covered under the mechanism. That means
we have to think about other platforms to support EEH, like powernv.
Besides, we only have framework for the mechanism and we have to implement
it for pseries platform later.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Cleanup function names in the EEH core
Gavin Shan [Mon, 27 Feb 2012 20:03:52 +0000 (20:03 +0000)]
powerpc/eeh: Cleanup function names in the EEH core

The EEH has been implemented on pSeries platform. The original
code looks a little bit nasty. The patch does cleanup on the
current EEH implementation so that it looks more clean.

        * Try adding prefix "eeh" for functions.
        * Some function names have been adjusted so that they looks
          shorter and meaningful.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Cleanup comments in the EEH core
Gavin Shan [Mon, 27 Feb 2012 20:03:51 +0000 (20:03 +0000)]
powerpc/eeh: Cleanup comments in the EEH core

The EEH has been implemented on pSeries platform. The original
code looks a little bit nasty. The patch does cleanup on the
current EEH implementation so that it looks more clean.

        * Duplicated comments have been removed from the corresponding
          header files.
        * Comments have been reorganized so that it looks more clean.
        * The leading comments of functions are adjusted for a little
          bit so that the result of "make pdfdocs" would be more
          unified.
        * Function definitions and calls have unified format as "xxx()".
          That means the format "xxx ()" has been replaced by "xxx()".
        * There're multiple functions implemented for resetting PE. The
          position of those functions have been move around so that they
          are adjacent to each other to reflect their relationship.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Replace mfmsr instructions with load from PACA kernel_msr field
Benjamin Herrenschmidt [Fri, 2 Mar 2012 00:33:52 +0000 (11:33 +1100)]
powerpc: Replace mfmsr instructions with load from PACA kernel_msr field

On 64-bit, the mfmsr instruction can be quite slow, slower
than loading a field from the cache-hot PACA, which happens
to already contain the value we want in most cases.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Fix 64-bit BookE FP unavailable exceptions
Benjamin Herrenschmidt [Sun, 4 Mar 2012 23:55:04 +0000 (10:55 +1100)]
powerpc: Fix 64-bit BookE FP unavailable exceptions

We were using CR0.EQ after EXCEPTION_COMMON, hoping it still
contained whether we came from userspace or kernel space.

However, under some circumstances, EXCEPTION_COMMON will
call C code and clobber non-volatile registers, so we really
need to re-load the previous MSR from the stackframe and
re-test.

While there, invert the condition to make the fast path more
obvious and remove the BUG_OPCODE which was a debugging
leftover and call .ret_from_except as we should.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Fix register clobbering when accumulating stolen time
Benjamin Herrenschmidt [Fri, 2 Mar 2012 00:01:31 +0000 (11:01 +1100)]
powerpc: Fix register clobbering when accumulating stolen time

When running under a hypervisor that supports stolen time accounting,
we may call C code from the macro EXCEPTION_PROLOG_COMMON in the
exception entry path, which clobbers CR0.

However, the FPU and vector traps rely on CR0 indicating whether we
are coming from userspace or kernel to decide what to do.

So we need to restore that value after the C call

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/xmon: Add display of soft & hard irq states
Benjamin Herrenschmidt [Thu, 1 Mar 2012 23:10:09 +0000 (10:10 +1100)]
powerpc/xmon: Add display of soft & hard irq states

Also use local_paca instead of get_paca() to avoid getting into
the smp_processor_id() debugging code from the debugger

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Add support for page fault retry and fatal signals
Benjamin Herrenschmidt [Thu, 1 Mar 2012 07:14:45 +0000 (18:14 +1100)]
powerpc: Add support for page fault retry and fatal signals

Other architectures such as x86 and ARM have been growing
new support for features like retrying page faults after
dropping the mm semaphore to break contention, or being
able to return from a stuck page fault when a SIGKILL is
pending.

This refactors our implementation of do_page_fault() to
move the error handling out of line in a way similar to
x86 and adds support for those two features.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Disable interrupts in 64-bit kernel FP and vector faults
Benjamin Herrenschmidt [Thu, 1 Mar 2012 04:47:44 +0000 (15:47 +1100)]
powerpc: Disable interrupts in 64-bit kernel FP and vector faults

If we get a floating point, altivec or vsx unavaible interrupt in
kernel, we trigger a kernel error. There is no point preserving
the interrupt state, in fact, that can even make debugging harder
as the processor state might change (we may even preempt) between
taking the exception and landing in a debugger.

So just make those 3 disable interrupts unconditionally.

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

v2: On BookE only disable when hitting the kernel unavailable
    path, otherwise it will fail to restore softe as
    fast_exception_return doesn't do it.

12 years agopowerpc: Call do_page_fault() with interrupts off
Benjamin Herrenschmidt [Wed, 7 Mar 2012 05:48:45 +0000 (16:48 +1100)]
powerpc: Call do_page_fault() with interrupts off

We currently turn interrupts back to their previous state before
calling do_page_fault(). This can be annoying when debugging as
a bad fault will potentially have lost some processor state before
getting into the debugger.

We also end up calling some generic code with interrupts enabled
such as notify_page_fault() with interrupts enabled, which could
be unexpected.

This changes our code to behave more like other architectures,
and make the assembly entry code call into do_page_faults() with
interrupts disabled. They are conditionally re-enabled from
within do_page_fault() in the same spot x86 does it.

While there, add the might_sleep() test in the case of a successful
trylock of the mmap semaphore, again like x86.

Also fix a bug in the existing assembly where r12 (_MSR) could get
clobbered by C calls (the DTL accounting in the exception common
macro and DISABLE_INTS) in some cases.

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

v2. Add the r12 clobber fix

12 years agopowerpc: Improve behaviour of irq tracing on 64-bit exception entry
Benjamin Herrenschmidt [Thu, 1 Mar 2012 04:42:56 +0000 (15:42 +1100)]
powerpc: Improve behaviour of irq tracing on 64-bit exception entry

Some exceptions would unconditionally disable interrupts on entry,
which is fine, but calling lockdep every time not only adds more
overhead than strictly needed, but also means we get quite a few
"redudant" disable logged, which makes it hard to spot the really
bad ones.

So instead, split the macro used by the exception code into a
normal one and a separate one used when CONFIG_TRACE_IRQFLAGS is
enabled, and make the later skip th tracing if interrupts were
already disabled.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Improve 64-bit syscall entry/exit
Benjamin Herrenschmidt [Thu, 1 Mar 2012 04:40:23 +0000 (15:40 +1100)]
powerpc: Improve 64-bit syscall entry/exit

We unconditionally hard enable interrupts. This is unnecessary as
syscalls are expected to always be called with interrupts enabled.

While at it, we add a WARN_ON if that is not the case and
CONFIG_TRACE_IRQFLAGS is enabled (we don't want to add overhead
to the fast path when this is not set though).

Thus let's remove the enabling (and associated irq tracing) from
the syscall entry path. Also on Book3S, replace a few mfmsr
instructions with loads of PACAMSR from the PACA, which should be
faster & schedule better.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Rework runlatch code
Benjamin Herrenschmidt [Thu, 1 Mar 2012 01:45:27 +0000 (12:45 +1100)]
powerpc: Rework runlatch code

This moves the inlines into system.h and changes the runlatch
code to use the thread local flags (non-atomic) rather than
the TIF flags (atomic) to keep track of the latch state.

The code to turn it back on in an asynchronous interrupt is
now simplified and partially inlined.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Use the same interrupt prolog for perfmon as other interrupts
Benjamin Herrenschmidt [Wed, 29 Feb 2012 23:52:01 +0000 (10:52 +1100)]
powerpc: Use the same interrupt prolog for perfmon as other interrupts

The perfmon interrupt is the sole user of a special variant of the
interrupt prolog which differs from the one used by external and timer
interrupts in that it saves the non-volatile GPRs and doesn't turn the
runlatch on.

The former is unnecessary and the later is arguably incorrect, so
let's clean that up by using the same prolog. While at it we rename
that prolog to use the _ASYNC prefix.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Remove legacy iSeries bits from assembly files
Benjamin Herrenschmidt [Tue, 28 Feb 2012 02:44:58 +0000 (13:44 +1100)]
powerpc: Remove legacy iSeries bits from assembly files

This removes the various bits of assembly in the kernel entry,
exception handling and SLB management code that were specific
to running under the legacy iSeries hypervisor which is no
longer supported.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: clean up vio.c
Stephen Rothwell [Wed, 7 Mar 2012 18:43:10 +0000 (18:43 +0000)]
powerpc: clean up vio.c

This cleans up vio.c after the removal of the legacy iSeries platform.
It also removes some no longer referenced include files.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agodriver-core: remove legacy iSeries hack
Stephen Rothwell [Wed, 7 Mar 2012 18:41:09 +0000 (18:41 +0000)]
driver-core: remove legacy iSeries hack

The PowerPC legacy iSeries plateform is being removed along with the
"one looney iseries driver", so this code can now be removed as well.

cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agotty: powerpc: remove SERIAL_ICOM dependency on PPC_ISERIES
Stephen Rothwell [Wed, 7 Mar 2012 18:39:31 +0000 (18:39 +0000)]
tty: powerpc: remove SERIAL_ICOM dependency on PPC_ISERIES

The PowerPC legacy iSeries platform is being removed so this is no
longer selectable.

Cc: Alan Cox <alan@linux.intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: linux-serial@vger.kernel.org
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agotty: powerpc: remove hvc_iseries
Stephen Rothwell [Wed, 7 Mar 2012 18:37:40 +0000 (18:37 +0000)]
tty: powerpc: remove hvc_iseries

The PowerPC legacy iSeries platform is being removed, so this code is no
longer needed.

Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: remove the legacy iSeries part of ibmvscsi
Stephen Rothwell [Wed, 7 Mar 2012 18:35:38 +0000 (18:35 +0000)]
powerpc: remove the legacy iSeries part of ibmvscsi

The PowerPC legacy iSeries platform is being removed and this code is
no longer selectable.  There is more clean up that can be done, but this
just gets the old code out of the way.

Cc: "James E.J. Bottomley" <JBottomley@parallels.com>
Cc: Brian King <brking@linux.vnet.ibm.com>
Cc: linux-scsi@vger.kernel.org
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agonet: powerpc: remove the legacy iSeries ethernet driver
Stephen Rothwell [Wed, 7 Mar 2012 18:33:53 +0000 (18:33 +0000)]
net: powerpc: remove the legacy iSeries ethernet driver

This driver is specific to the PowerPC legcay iSeries platform which is
being removed.

Cc: David Miller <davem@davemloft.net>
Cc: <netdev@vger.kernel.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Remove the main legacy iSerie platform code
Stephen Rothwell [Wed, 7 Mar 2012 17:02:07 +0000 (17:02 +0000)]
powerpc: Remove the main legacy iSerie platform code

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/pmac: Use string library in nvram code
Akinobu Mita [Fri, 27 Jan 2012 04:24:48 +0000 (04:24 +0000)]
powerpc/pmac: Use string library in nvram code

- Use memchr_inv to check if the data contains all 0xFF bytes.
  It is faster than looping for each byte.

- Use memcmp to compare memory areas

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Make SPARSE_IRQ required
Grant Likely [Mon, 30 Jan 2012 08:02:19 +0000 (08:02 +0000)]
powerpc: Make SPARSE_IRQ required

All IRQs on powerpc are managed via irq_domain anyway, there isn't really
any advantage to turning SPARSE_IRQ off, and it's the direction we want
to take the kernel design anyway.  This patch makes powerpc always use
SPARSE_IRQ.

On pseries_defconfig, SPARSE_IRQ adds only about 0x300 bytes to the
.text sections, and removes about 0x20000 from the data section for the
static irq_desc table.

Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
Cc: Rob Herring <rob.herring@calxeda.com>
Cc: Ben Herrenschmidt <benh@kernel.crashing.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/prom: Remove limit on maximum size of properties
Nishanth Aravamudan [Mon, 27 Feb 2012 08:55:15 +0000 (08:55 +0000)]
powerpc/prom: Remove limit on maximum size of properties

On a 16TB system (using AMS/CMO), I get:

WARNING: ignoring large property [/ibm,dynamic-reconfiguration-memory] ibm,dynamic-memory length 0x000000000017ffec

and significantly less memory is thus shown to the partition. As far as
I can tell, the constant used is arbitrary. Ben Herrenschmidt provided
additional background that

> The limit was originally set because of Apple machines carrying ROM
> images in the device-tree, at a time where we were much more memory
> constrained than we are now.

and that it is likely not very useful any longer.

Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Use set_current_blocked() and block_sigmask()
Matt Fleming [Tue, 14 Feb 2012 01:40:59 +0000 (01:40 +0000)]
powerpc: Use set_current_blocked() and block_sigmask()

As described in e6fa16ab ("signal: sigprocmask() should do
retarget_shared_pending()") the modification of current->blocked is
incorrect as we need to check whether the signal we're about to block
is pending in the shared queue.

Also, use the new helper function introduced in commit 5e6292c0f28f
("signal: add block_sigmask() for adding sigmask to current->blocked")
which centralises the code for updating current->blocked after
successfully delivering a signal and reduces the amount of duplicate
code across architectures. In the past some architectures got this
code wrong, so using this helper function should stop that from
happening again.

Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Use vsprintf extention %pf with builtin_return_address
Joe Perches [Tue, 28 Feb 2012 08:49:34 +0000 (08:49 +0000)]
powerpc: Use vsprintf extention %pf with builtin_return_address

Emit the function name not the address when possible.

builtin_return_address() gives an address.  When building
a kernel with CONFIG_KALLSYMS, emit the actual function
name not the address.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/icswx: Fix race condition with IPI setting ACOP
Jimi Xenidis [Tue, 28 Feb 2012 13:27:07 +0000 (13:27 +0000)]
powerpc/icswx: Fix race condition with IPI setting ACOP

There is a race where a thread causes a coprocessor type to be valid
in its own ACOP _and_ in the current context, but it does not
propagate to the ACOP register of other threads in time for them to
use it.  The original code tries to solve this by sending an IPI to
all threads on the system, which is heavy handed, but unfortunately
still provides a window where the icswx is issued by other threads and
the ACOP is not up to date.

This patch detects that the ACOP DSI fault was a "false positive" and
syncs the ACOP and causes the icswx to be replayed.

Signed-off-by: Jimi Xenidis <jimix@pobox.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/atomic: Implement atomic*_inc_not_zero
Anton Blanchard [Wed, 29 Feb 2012 21:12:16 +0000 (21:12 +0000)]
powerpc/atomic: Implement atomic*_inc_not_zero

Implement atomic_inc_not_zero and atomic64_inc_not_zero. At the
moment we use atomic*_add_unless which requires us to put 0 and
1 constants into registers. We can also avoid a subtract by
saving the original value in a second temporary.

This removes 3 instructions from fget:

c0000000001b63c0:       39 00 00 00     li      r8,0
c0000000001b63c4:       39 40 00 01     li      r10,1
...
c0000000001b63e8:       7c 0a 00 50     subf    r0,r10,r0

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agoatomic: Allow atomic_inc_not_zero to be overridden
Anton Blanchard [Wed, 29 Feb 2012 21:09:53 +0000 (21:09 +0000)]
atomic: Allow atomic_inc_not_zero to be overridden

We want to implement a ppc64 specific version of atomic_inc_not_zero
so wrap it in an ifdef to allow it to be overridden.

Signed-off-by: Anton Blanchard <anton@samba.org>
Acked-by: Mike Frysinger <vapier@gentoo.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agocarma-fpga: fix race between data dumping and DMA callback
Ira Snyder [Thu, 26 Jan 2012 11:00:14 +0000 (11:00 +0000)]
carma-fpga: fix race between data dumping and DMA callback

When the system is under heavy load, we occasionally saw a problem where
the system would get a legitimate interrupt when they should be
disabled.

This was caused by the data_dma_cb() DMA callback unconditionally
re-enabling FPGA interrupts even when data dumping is disabled. When
data dumping was re-enabled, the irq handler would fire while a DMA was
in progress. The "BUG_ON(priv->inflight != NULL);" during the second
invocation of the DMA callback caused the system to crash.

To fix the issue, the priv->enabled boolean is moved under the
protection of the priv->lock spinlock. The DMA callback checks the
boolean to know whether to re-enable FPGA interrupts before it returns.

Now that it is fixed, the driver keeps FPGA interrupts disabled when it
expects that they are disabled, fixing the bug.

Signed-off-by: Ira W. Snyder <iws@ovro.caltech.edu>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agocarma-fpga: fix lockdep warning
Ira Snyder [Thu, 26 Jan 2012 10:59:54 +0000 (10:59 +0000)]
carma-fpga: fix lockdep warning

Lockdep occasionally complains with the message:
INFO: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected

This is caused by calling videobuf_dma_unmap() under spin_lock_irq(). To
fix the warning, we drop the lock before unmapping and freeing the
buffer.

Signed-off-by: Ira W. Snyder <iws@ovro.caltech.edu>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agomacintosh: Fix typo in mediabay.c
Masanari Iida [Mon, 23 Jan 2012 07:26:36 +0000 (07:26 +0000)]
macintosh: Fix typo in mediabay.c

Fix typo "unsuported" to "unsupported" in
drivers/machintosh/mediabay.c

Signed-off-by: Masanari Iida<standby24x7@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agoarch/powerpc/platforms/powernv/setup.c: included asm/xics.h twice
Danny Kukawka [Thu, 16 Feb 2012 03:56:03 +0000 (03:56 +0000)]
arch/powerpc/platforms/powernv/setup.c: included asm/xics.h twice

arch/powerpc/platforms/powernv/setup.c: included 'asm/xics.h' twice,
remove the duplicate.

Signed-off-by: Danny Kukawka <danny.kukawka@bisect.de>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agoarch/powerpc/kvm/book3s_hv.c: included linux/sched.h twice
Danny Kukawka [Thu, 16 Feb 2012 03:55:54 +0000 (03:55 +0000)]
arch/powerpc/kvm/book3s_hv.c: included linux/sched.h twice

arch/powerpc/kvm/book3s_hv.c: included 'linux/sched.h' twice,
remove the duplicate.

Signed-off-by: Danny Kukawka <danny.kukawka@bisect.de>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: remove CONFIG_PPC_ISERIES from the architecture Kconfig files
Stephen Rothwell [Wed, 22 Feb 2012 14:10:12 +0000 (14:10 +0000)]
powerpc: remove CONFIG_PPC_ISERIES from the architecture Kconfig files

After this, we can remove the legacy iSeries code more easily.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mpic: Fix allocation of reverse-map for multi-ISU mpics
Benjamin Herrenschmidt [Wed, 22 Feb 2012 13:50:13 +0000 (13:50 +0000)]
powerpc/mpic: Fix allocation of reverse-map for multi-ISU mpics

When using a multi-ISU MPIC, we can interrupts up to
isu_size * MPIC_MAX_ISU, not just isu_size, so allocate
the right size reverse map.

Without this, the code will constantly fallback to
a linear search.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agoMerge remote-tracking branch 'origin/master' into next
Benjamin Herrenschmidt [Sun, 26 Feb 2012 23:50:11 +0000 (10:50 +1100)]
Merge remote-tracking branch 'origin/master' into next

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Sun, 26 Feb 2012 20:47:17 +0000 (12:47 -0800)]
Merge git://git./linux/kernel/git/davem/net

1) ICMP sockets leave err uninitialized but we try to return it for the
   unsupported MSG_OOB case, reported by Dave Jones.

2) Add new Zaurus device ID entries, from Dave Jones.

3) Pointer calculation in hso driver memset is wrong, from Dan
   Carpenter.

4) ks8851_probe() checks unsigned value as negative, fix also from Dan
   Carpenter.

5) Fix crashes in atl1c driver due to TX queue handling, from Eric
   Dumazet.  I anticipate some TX side locking fixes coming in the near
   future for this driver as well.

6) The inline directive fix in Bluetooth which was breaking the build
   only with very new versions of GCC, from Johan Hedberg.

7) Fix crashes in the ATP CLIP code due to ARP cleanups this merge
   window, reported by Meelis Roos and fixed by Eric Dumazet.

8) JME driver doesn't flush RX FIFO correctly, from Guo-Fu Tseng.

9) Some ip6_route_output() callers test the return value for NULL, but
   this never happens as the convention is to return a dst entry with
   dst->error set.  Fixes from RonQing Li.

10) Logitech Harmony 900 should be handled by zaurus driver not
   cdc_ether, update white lists and black lists accordingly.  From
   Scott Talbert.

11) Receiving from certain kinds of devices there won't be a MAC header,
   so there is no MAC header to fixup in the IPSEC code, and if we try
   to do it we'll crash.  Fix from Eric Dumazet.

12) Port type array indexing off-by-one in mlx4 driver, fix from Yevgeny
   Petrilin.

13) Fix regression in link-down handling in davinci_emac which causes
   all RX descriptors to be freed up and therefore RX to wedge
   completely, from Christian Riesch.

14) It took two attempts, but ctnetlink soft lockups seem to be
   cured now, from Pablo Neira Ayuso.

15) Endianness bug fix in ENIC driver, from Santosh Nayak.

16) The long ago conversion of the PPP fragmentation code over to
   abstracted SKB list handling wasn't perfect, once we get an
   out of sequence SKB we don't flush the rest of them like we
   should.  From Ben McKeegan.

17) Fix regression of ->ip_summed initialization in sfc driver.
   From Ben Hutchings.

18) Bluetooth timeout mistakenly using msecs instead of jiffies,
   from Andrzej Kaczmarek.

19) Using _sync variant of work cancellation results in deadlocks,
   use the non _sync variants instead.  From Andre Guedes.

20) Bluetooth rfcomm code had reference counting problems leading
   to crashes, fix from Octavian Purdila.

21) The conversion of netem over to classful qdisc handling added
   two bugs to netem_dequeue(), fixes from Eric Dumazet.

22) Missing pci_iounmap() in ATM Solos driver.  Fix from Julia Lawall.

23) b44_pci_exit() should not have __exit tag since it's invoked from
   non-__exit code.  From Nikola Pajkovsky.

24) The conversion of the neighbour hash tables over to RCU added a
   race, fixed here by adding the necessary reread of tbl->nht, fix
   from Michel Machado.

25) When we added VF (virtual function) attributes for network device
   dumps, this potentially bloats up the size of the dump of one
   network device such that the dump size is too large for the buffer
   allocated by properly written netlink applications.

   In particular, if you add 255 VFs to a network device, parts of
   GLIBC stop working.

   To fix this, we add an attribute that is used to turn on these
   extended portions of the network device dump.  Sophisticaed
   applications like 'ip' that want to see this stuff  will be changed
   to set the attribute, whereas things like GLIBC that don't care
   about VFs simply will not, and therefore won't be busted by the
   mere presence of VFs on a network device.

   Thanks to the tireless work of Greg Rose on this fix.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (53 commits)
  sfc: Fix assignment of ip_summed for pre-allocated skbs
  ppp: fix 'ppp_mp_reconstruct bad seq' errors
  enic: Fix endianness bug.
  gre: fix spelling in comments
  netfilter: ctnetlink: fix soft lockup when netlink adds new entries (v2)
  Revert "netfilter: ctnetlink: fix soft lockup when netlink adds new entries"
  davinci_emac: Do not free all rx dma descriptors during init
  mlx4_core: Fixing array indexes when setting port types
  phy: IC+101G and PHY_HAS_INTERRUPT flag
  netdev/phy/icplus: Correct broken phy_init code
  ipsec: be careful of non existing mac headers
  Move Logitech Harmony 900 from cdc_ether to zaurus
  hso: memsetting wrong data in hso_get_count()
  netfilter: ip6_route_output() never returns NULL.
  ethernet/broadcom: ip6_route_output() never returns NULL.
  ipv6: ip6_route_output() never returns NULL.
  jme: Fix FIFO flush issue
  atm: clip: remove clip_tbl
  ipv4: ping: Fix recvmsg MSG_OOB error handling.
  rtnetlink: Fix problem with buffer allocation
  ...

12 years agoFix autofs compile without CONFIG_COMPAT
Linus Torvalds [Sun, 26 Feb 2012 17:44:55 +0000 (09:44 -0800)]
Fix autofs compile without CONFIG_COMPAT

The autofs compat handling fix caused a compile failure when
CONFIG_COMPAT isn't defined.

Instead of adding random #ifdef'fery in autofs, let's just make the
compat helpers earlier to use: without CONFIG_COMPAT, is_compat_task()
just hardcodes to zero.

We could probably do something similar for a number of other cases where
we have #ifdef's in code, but this is the low-hanging fruit.

Reported-and-tested-by: Andreas Schwab <schwab@linux-m68k.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoLinux 3.3-rc5
Linus Torvalds [Sat, 25 Feb 2012 20:18:16 +0000 (12:18 -0800)]
Linux 3.3-rc5

12 years agoMerge tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck...
Linus Torvalds [Sat, 25 Feb 2012 20:12:08 +0000 (12:12 -0800)]
Merge tag 'hwmon-for-linus' of git://git./linux/kernel/git/groeck/linux-staging

Couple of minor driver fixes.

* tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  hwmon: (max34440) Fix resetting temperature history
  hwmon: (f75375s) Fix register write order when setting fans to full speed
  hwmon: (ads1015) Fix file leak in probe function
  hwmon: (max6639) Fix PPR register initialization to set both channels
  hwmon: (max6639) Fix FAN_FROM_REG calculation

12 years agoMerge branch 'rc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild
Linus Torvalds [Sat, 25 Feb 2012 20:11:25 +0000 (12:11 -0800)]
Merge branch 'rc-fixes' of git://git./linux/kernel/git/mmarek/kbuild

three kbuild fixes for 3.3:
 - make deb-pkg symlink race fix.
 - make coccicheck fix.
 - Dropping the check for modutils.  This is not a regression, but
   allows the module-init-tools replacement kmod work with the 3.3
   kernel.

* 'rc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild:
  coccicheck: change handling of C={1,2} when M= is set
  builddeb: Don't create files in /tmp with predictable names
  kbuild: do not check for ancient modutils tools

12 years agoautofs: work around unhappy compat problem on x86-64
Ian Kent [Wed, 22 Feb 2012 12:45:44 +0000 (20:45 +0800)]
autofs: work around unhappy compat problem on x86-64

When the autofs protocol version 5 packet type was added in commit
5c0a32fc2cd0 ("autofs4: add new packet type for v5 communications"), it
obvously tried quite hard to be word-size agnostic, and uses explicitly
sized fields that are all correctly aligned.

However, with the final "char name[NAME_MAX+1]" array at the end, the
actual size of the structure ends up being not very well defined:
because the struct isn't marked 'packed', doing a "sizeof()" on it will
align the size of the struct up to the biggest alignment of the members
it has.

And despite all the members being the same, the alignment of them is
different: a "__u64" has 4-byte alignment on x86-32, but native 8-byte
alignment on x86-64.  And while 'NAME_MAX+1' ends up being a nice round
number (256), the name[] array starts out a 4-byte aligned.

End result: the "packed" size of the structure is 300 bytes: 4-byte, but
not 8-byte aligned.

As a result, despite all the fields being in the same place on all
architectures, sizeof() will round up that size to 304 bytes on
architectures that have 8-byte alignment for u64.

Note that this is *not* a problem for 32-bit compat mode on POWER, since
there __u64 is 8-byte aligned even in 32-bit mode.  But on x86, 32-bit
and 64-bit alignment is different for 64-bit entities, and as a result
the structure that has exactly the same layout has different sizes.

So on x86-64, but no other architecture, we will just subtract 4 from
the size of the structure when running in a compat task.  That way we
will write the properly sized packet that user mode expects.

Not pretty.  Sadly, this very subtle, and unnecessary, size difference
has been encoded in user space that wants to read packets of *exactly*
the right size, and will refuse to touch anything else.

Reported-and-tested-by: Thomas Meyer <thomas@m3y3r.de>
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoMerge tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Sat, 25 Feb 2012 04:03:14 +0000 (20:03 -0800)]
Merge tag 'rdma-for-linus' of git://git./linux/kernel/git/roland/infiniband

One InfiniBand/RDMA regression fix for 3.3:

 - mlx4 SR-IOV changes added static exported functions, which doesn't
   build on powerpc at least.  Fix from Doug Ledford for this.

* tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
  mlx4_core: Exported functions can't be static

12 years agoMerge branch 'sfc-3.3' of git://git.kernel.org/pub/scm/linux/kernel/git/bwh/sfc
David S. Miller [Sat, 25 Feb 2012 03:12:44 +0000 (22:12 -0500)]
Merge branch 'sfc-3.3' of git://git./linux/kernel/git/bwh/sfc

12 years agosfc: Fix assignment of ip_summed for pre-allocated skbs
Ben Hutchings [Sat, 25 Feb 2012 00:03:10 +0000 (00:03 +0000)]
sfc: Fix assignment of ip_summed for pre-allocated skbs

When pre-allocating skbs for received packets, we set ip_summed =
CHECKSUM_UNNCESSARY.  We used to change it back to CHECKSUM_NONE when
the received packet had an incorrect checksum or unhandled protocol.

Commit bc8acf2c8c3e43fcc192762a9f964b3e9a17748b ('drivers/net: avoid
some skb->ip_summed initializations') mistakenly replaced the latter
assignment with a DEBUG-only assertion that ip_summed ==
CHECKSUM_NONE.  This assertion is always false, but it seems no-one
has exercised this code path in a DEBUG build.

Fix this by moving our assignment of CHECKSUM_UNNECESSARY into
efx_rx_packet_gro().

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
12 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi...
Linus Torvalds [Sat, 25 Feb 2012 00:08:51 +0000 (16:08 -0800)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi-rc-fixes-2.6

SCSI fixes on 20120224:
 "This is a set of assorted bug fixes for power management, mpt2sas,
  ipr, the rdac device handler and quite a big chunk for qla2xxx (plus a
  use after free of scsi_host in scsi_scan.c). "

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6:
  [SCSI] scsi_dh_rdac: Fix for unbalanced reference count
  [SCSI] scsi_pm: Fix bug in the SCSI power management handler
  [SCSI] scsi_scan: Fix 'Poison overwritten' warning caused by using freed 'shost'
  [SCSI] qla2xxx: Update version number to 8.03.07.13-k.
  [SCSI] qla2xxx: Proper detection of firmware abort error code for ISP82xx.
  [SCSI] qla2xxx: Remove resetting memory during device initialization for ISP82xx.
  [SCSI] qla2xxx: Complete mailbox command timedout to avoid initialization failures during next reset cycle.
  [SCSI] qla2xxx: Remove check for null fcport from host reset handler.
  [SCSI] qla2xxx: Correct out of bounds read of ISP2200 mailbox registers.
  [SCSI] qla2xxx: Remove errant clearing of MBX_INTERRUPT flag during CT-IOCB processing.
  [SCSI] qla2xxx: Clear options-flags while issuing stop-firmware mbx command.
  [SCSI] qla2xxx: Add an "is reset active" helper.
  [SCSI] qla2xxx: Add check for null fcport references in qla2xxx_queuecommand.
  [SCSI] qla2xxx: Propagate up abort failures.
  [SCSI] isci: Fix NULL ptr dereference when no firmware is being loaded
  [SCSI] ipr: fix eeh recovery for 64-bit adapters
  [SCSI] mpt2sas: Fix mismatch in mpt2sas_base_hard_reset_handler() mutex lock-unlock

12 years agoppp: fix 'ppp_mp_reconstruct bad seq' errors
Ben McKeegan [Fri, 24 Feb 2012 06:33:56 +0000 (06:33 +0000)]
ppp: fix 'ppp_mp_reconstruct bad seq' errors

This patch fixes a (mostly cosmetic) bug introduced by the patch
'ppp: Use SKB queue abstraction interfaces in fragment processing'
found here: http://www.spinics.net/lists/netdev/msg153312.html

The above patch rewrote and moved the code responsible for cleaning
up discarded fragments but the new code does not catch every case
where this is necessary.  This results in some discarded fragments
remaining in the queue, and triggering a 'bad seq' error on the
subsequent call to ppp_mp_reconstruct.  Fragments are discarded
whenever other fragments of the same frame have been lost.
This can generate a lot of unwanted and misleading log messages.

This patch also adds additional detail to the debug logging to
make it clearer which fragments were lost and which other fragments
were discarded as a result of losses. (Run pppd with 'kdebug 1'
option to enable debug logging.)

Signed-off-by: Ben McKeegan <ben@netservers.co.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agoenic: Fix endianness bug.
Santosh Nayak [Fri, 24 Feb 2012 06:56:39 +0000 (06:56 +0000)]
enic: Fix endianness bug.

Sparse complaints the endian bug.

Signed-off-by: Santosh Nayak <santoshprasadnayak@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agococcicheck: change handling of C={1,2} when M= is set
Greg Dietsche [Fri, 20 Jan 2012 23:10:35 +0000 (17:10 -0600)]
coccicheck: change handling of C={1,2} when M= is set

This patch reverts a portion of d0bc1fb4 so that coccicheck will
work properly when C=1 or C=2.

Reported-and-tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Greg Dietsche <Gregory.Dietsche@cuw.edu>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Michal Marek <mmarek@suse.cz>