]> git.karo-electronics.de Git - karo-tx-linux.git/log
karo-tx-linux.git
11 years agoarm64: KVM: 32bit specific register world switch
Marc Zyngier [Thu, 7 Feb 2013 10:52:10 +0000 (10:52 +0000)]
arm64: KVM: 32bit specific register world switch

Allow registers specific to 32bit guests to be saved/restored
during the world switch.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: CPU specific 32bit coprocessor access
Marc Zyngier [Thu, 7 Feb 2013 10:50:18 +0000 (10:50 +0000)]
arm64: KVM: CPU specific 32bit coprocessor access

Enable handling of CPU specific 32bit coprocessor access. Not much
here either.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: 32bit handling of coprocessor traps
Marc Zyngier [Thu, 7 Feb 2013 10:32:33 +0000 (10:32 +0000)]
arm64: KVM: 32bit handling of coprocessor traps

Provide the necessary infrastructure to trap coprocessor accesses that
occur when running 32bit guests.

Also wire SMC and HVC trapped in 32bit mode while were at it.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: 32bit conditional execution emulation
Marc Zyngier [Wed, 6 Feb 2013 19:54:04 +0000 (19:54 +0000)]
arm64: KVM: 32bit conditional execution emulation

As conditional instructions can trap on AArch32, add the thinest
possible emulation layer to keep 32bit guests happy.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: 32bit GP register access
Marc Zyngier [Wed, 6 Feb 2013 19:40:29 +0000 (19:40 +0000)]
arm64: KVM: 32bit GP register access

Allow access to the 32bit register file through the usual API.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: define 32bit specific registers
Marc Zyngier [Wed, 6 Feb 2013 19:17:50 +0000 (19:17 +0000)]
arm64: KVM: define 32bit specific registers

Define the 32bit specific registers (SPSRs, cp15...).

Most CPU registers are directly mapped to a 64bit register
(r0->x0...). Only the SPSRs have separate registers.

cp15 registers are also mapped into their 64bit counterpart in most
cases.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: Build system integration
Marc Zyngier [Mon, 10 Dec 2012 16:41:44 +0000 (16:41 +0000)]
arm64: KVM: Build system integration

Only the Makefile is plugged in. The Kconfig stuff is in a separate
patch to allow for an easier merge process.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: PSCI implementation
Marc Zyngier [Wed, 12 Dec 2012 18:52:05 +0000 (18:52 +0000)]
arm64: KVM: PSCI implementation

Wire the PSCI backend into the exit handling code.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: Plug the arch timer
Marc Zyngier [Fri, 7 Dec 2012 17:52:03 +0000 (17:52 +0000)]
arm64: KVM: Plug the arch timer

Add support for the in-kernel timer emulation.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoARM: KVM: timer: allow DT matching for ARMv8 cores
Marc Zyngier [Thu, 30 May 2013 17:31:28 +0000 (18:31 +0100)]
ARM: KVM: timer: allow DT matching for ARMv8 cores

ARMv8 cores have the exact same timer as ARMv7 cores. Make sure the
KVM timer code can match it in the device tree.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: Plug the VGIC
Marc Zyngier [Fri, 7 Dec 2012 17:54:54 +0000 (17:54 +0000)]
arm64: KVM: Plug the VGIC

Add support for the in-kernel GIC emulation.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: Exit handling
Marc Zyngier [Mon, 10 Dec 2012 16:40:41 +0000 (16:40 +0000)]
arm64: KVM: Exit handling

Handle the exit of a VM, decoding the exit reason from HYP mode
and calling the corresponding handler.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: HYP mode world switch implementation
Marc Zyngier [Mon, 10 Dec 2012 16:40:18 +0000 (16:40 +0000)]
arm64: KVM: HYP mode world switch implementation

The HYP mode world switch in all its glory.

Implements save/restore of host/guest registers, EL2 trapping,
IPA resolution, and additional services (tlb invalidation).

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: hypervisor initialization code
Marc Zyngier [Mon, 17 Dec 2012 17:07:52 +0000 (17:07 +0000)]
arm64: KVM: hypervisor initialization code

Provide EL2 with page tables and stack, and set the vectors
to point to the full blown world-switch code.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: guest one-reg interface
Marc Zyngier [Mon, 10 Dec 2012 16:37:02 +0000 (16:37 +0000)]
arm64: KVM: guest one-reg interface

Let userspace play with the guest registers.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: MMIO access backend
Marc Zyngier [Mon, 10 Dec 2012 16:29:50 +0000 (16:29 +0000)]
arm64: KVM: MMIO access backend

Define the necessary structures to perform an MMIO access.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: kvm_arch and kvm_vcpu_arch definitions
Marc Zyngier [Mon, 10 Dec 2012 16:29:28 +0000 (16:29 +0000)]
arm64: KVM: kvm_arch and kvm_vcpu_arch definitions

Provide the architecture dependent structures for VM and
vcpu abstractions.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: virtual CPU reset
Marc Zyngier [Mon, 10 Dec 2012 16:23:59 +0000 (16:23 +0000)]
arm64: KVM: virtual CPU reset

Provide the reset code for a virtual CPU booted in 64bit mode.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: CPU specific system registers handling
Marc Zyngier [Wed, 6 Feb 2013 17:30:48 +0000 (17:30 +0000)]
arm64: KVM: CPU specific system registers handling

Add the support code for CPU specific system registers. Not much
here yet.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: system register handling
Marc Zyngier [Mon, 10 Dec 2012 16:15:34 +0000 (16:15 +0000)]
arm64: KVM: system register handling

Provide 64bit system register handling, modeled after the cp15
handling for ARM.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: user space interface
Marc Zyngier [Mon, 10 Dec 2012 16:29:28 +0000 (16:29 +0000)]
arm64: KVM: user space interface

Provide the kvm.h file that defines the user space visible
interface.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: architecture specific MMU backend
Marc Zyngier [Mon, 10 Dec 2012 15:35:24 +0000 (15:35 +0000)]
arm64: KVM: architecture specific MMU backend

Define the arm64 specific MMU backend:
- HYP/kernel VA offset
- S2 4/64kB definitions
- S2 page table populating and flushing
- icache cleaning

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: fault injection into a guest
Marc Zyngier [Mon, 17 Dec 2012 12:27:42 +0000 (12:27 +0000)]
arm64: KVM: fault injection into a guest

Implement the injection of a fault (undefined, data abort or
prefetch abort) into a 64bit guest.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: Basic ESR_EL2 helpers and vcpu register access
Marc Zyngier [Mon, 10 Dec 2012 13:27:52 +0000 (13:27 +0000)]
arm64: KVM: Basic ESR_EL2 helpers and vcpu register access

Implements helpers for dealing with the EL2 syndrome register as
well as accessing the vcpu registers.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: system register definitions for 64bit guests
Marc Zyngier [Mon, 10 Dec 2012 11:16:40 +0000 (11:16 +0000)]
arm64: KVM: system register definitions for 64bit guests

Define the saved/restored registers for 64bit guests.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: EL2 register definitions
Marc Zyngier [Mon, 10 Dec 2012 10:46:47 +0000 (10:46 +0000)]
arm64: KVM: EL2 register definitions

Define all the useful bitfields for EL2 registers.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: HYP mode idmap support
Marc Zyngier [Fri, 7 Dec 2012 18:40:43 +0000 (18:40 +0000)]
arm64: KVM: HYP mode idmap support

Add the necessary infrastructure for identity-mapped HYP page
tables. Idmap-ed code must be in the ".hyp.idmap.text" linker
section.

The rest of the HYP ends up in ".hyp.text".

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoarm64: KVM: define HYP and Stage-2 translation page flags
Marc Zyngier [Fri, 7 Dec 2012 18:35:41 +0000 (18:35 +0000)]
arm64: KVM: define HYP and Stage-2 translation page flags

Add HYP and S2 page flags, for both normal and device memory.

Reviewed-by: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
11 years agoLinux 3.10-rc4 v3.10-rc4
Linus Torvalds [Sun, 2 Jun 2013 08:11:17 +0000 (17:11 +0900)]
Linux 3.10-rc4

11 years agoMerge branch 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/paris...
Linus Torvalds [Sat, 1 Jun 2013 21:24:54 +0000 (06:24 +0900)]
Merge branch 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux

Pull parisc fixes from Helge Deller:
 "This patcheset includes fixes for:

   - the PCI/LBA which brings back the stifb graphics framebuffer
     console
   - possible memory overflows in parisc kernel init code
   - parport support on older GSC machines
   - avoids that users by mistake enable PARPORT_PC_SUPERIO on parisc
   - MAINTAINERS file list updates for parisc."

* 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: parport0: fix this legacy no-device port driver!
  parport_pc: disable PARPORT_PC_SUPERIO on parisc architecture
  parisc/PCI: lba: fix: convert to pci_create_root_bus() for correct root bus resources (v2)
  parisc/PCI: Set type for LBA bus_num resource
  MAINTAINERS: update parisc architecture file list
  parisc: kernel: using strlcpy() instead of strcpy()
  parisc: rename "CONFIG_PA7100" to "CONFIG_PA7000"
  parisc: fix kernel BUG at arch/parisc/include/asm/mmzone.h:50
  parisc: memory overflow, 'name' length is too short for using

11 years agoparisc: parport0: fix this legacy no-device port driver!
Helge Deller [Thu, 30 May 2013 21:06:39 +0000 (21:06 +0000)]
parisc: parport0: fix this legacy no-device port driver!

Fix the above kernel error from parport_announce_port() on 32bit GSC
machines (e.g. B160L). The parport driver requires now a pointer to the
device struct.

Signed-off-by: Helge Deller <deller@gmx.de>
11 years agoparport_pc: disable PARPORT_PC_SUPERIO on parisc architecture
Helge Deller [Thu, 30 May 2013 16:24:46 +0000 (16:24 +0000)]
parport_pc: disable PARPORT_PC_SUPERIO on parisc architecture

If enabled, CONFIG_PARPORT_PC_SUPERIO scans on PC-like hardware for
various super-io chips by accessing i/o ports in a range which will
crash any parisc hardware at once.

In addition, parisc has it's own incompatible superio chip
(CONFIG_SUPERIO), so if we disable PARPORT_PC_SUPERIO completely for
parisc we can avoid that people by accident enable the parport_pc
superio option too.

Signed-off-by: Helge Deller <deller@gmx.de>
11 years agoparisc/PCI: lba: fix: convert to pci_create_root_bus() for correct root bus resources...
Helge Deller [Fri, 31 May 2013 22:24:58 +0000 (22:24 +0000)]
parisc/PCI: lba: fix: convert to pci_create_root_bus() for correct root bus resources (v2)

commit dc7dce280a
Author: Bjorn Helgaas <bhelgaas@google.com>
Date:   Fri Oct 28 16:27:27 2011 -0600
   parisc/PCI: lba: convert to pci_create_root_bus() for correct root bus
                    resources

  Supply root bus resources to pci_create_root_bus() so they're correct
  immediately.  This fixes the problem of "early" and "header" quirks seeing
  incorrect root bus resources.

added tests for elmmio_space.start while it should use
elmmio_space.flags.  This for example led to incorrect resource
assignments and a non-working stifb framebuffer on most parisc machines.

LBA 10:1: PCI host bridge to bus 0000:01
pci_bus 0000:01: root bus resource [io  0x12000-0x13fff] (bus address [0x2000-0x3fff])
pci_bus 0000:01: root bus resource [mem 0xfffffffffa000000-0xfffffffffbffffff] (bus address [0xfa000000-0xfbffffff])
pci_bus 0000:01: root bus resource [mem 0xfffffffff4800000-0xfffffffff4ffffff] (bus address [0xf4800000-0xf4ffffff])
pci_bus 0000:01: root bus resource [??? 0x00000001 flags 0x0]

Signed-off-by: Helge Deller <deller@gmx.de>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
11 years agoparisc/PCI: Set type for LBA bus_num resource
Bjorn Helgaas [Thu, 30 May 2013 17:45:39 +0000 (11:45 -0600)]
parisc/PCI: Set type for LBA bus_num resource

The non-PAT resource probing code failed to set the type of the LBA bus_num
resource (30aa80da43 "parisc/PCI: register busn_res for root buses" did
the corresponding thing for the PAT case).

This causes incorrect resource assignments and a non-working stifb
framebuffer on most parisc machines.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Helge Deller <deller@gmx.de>
11 years agoMAINTAINERS: update parisc architecture file list
Helge Deller [Thu, 30 May 2013 13:48:07 +0000 (13:48 +0000)]
MAINTAINERS: update parisc architecture file list

Signed-off-by: Helge Deller <deller@gmx.de>
11 years agoparisc: kernel: using strlcpy() instead of strcpy()
Chen Gang [Thu, 30 May 2013 01:18:43 +0000 (01:18 +0000)]
parisc: kernel: using strlcpy() instead of strcpy()

'boot_args' is an input args, and 'boot_command_line' has a fix length.
So use strlcpy() instead of strcpy() to avoid memory overflow.

Signed-off-by: Chen Gang <gang.chen@asianux.com>
Acked-by: Kyle McMartin <kyle@mcmartin.ca>
Signed-off-by: Helge Deller <deller@gmx.de>
11 years agoparisc: rename "CONFIG_PA7100" to "CONFIG_PA7000"
Paul Bolle [Wed, 29 May 2013 09:56:58 +0000 (09:56 +0000)]
parisc: rename "CONFIG_PA7100" to "CONFIG_PA7000"

There's a Makefile line setting cflags for CONFIG_PA7100. But that
Kconfig macro doesn't exist. There is a Kconfig symbol PA7000, which
covers both PA7000 and PA7100 processors. So let's use the corresponding
Kconfig macro.

Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Helge Deller <deller@gmx.de>
11 years agoparisc: fix kernel BUG at arch/parisc/include/asm/mmzone.h:50
Helge Deller [Tue, 28 May 2013 20:35:54 +0000 (20:35 +0000)]
parisc: fix kernel BUG at arch/parisc/include/asm/mmzone.h:50

With CONFIG_DISCONTIGMEM=y and multiple physical memory areas,
cat /proc/kpageflags triggers this kernel bug:

kernel BUG at arch/parisc/include/asm/mmzone.h:50!
CPU: 2 PID: 7848 Comm: cat Tainted: G      D W 3.10.0-rc3-64bit #44
 IAOQ[0]: kpageflags_read0x128/0x238
 IAOQ[1]: kpageflags_read0x12c/0x238
 RP(r2): proc_reg_read0xbc/0x130
Backtrace:
 [<00000000402ca2d4>] proc_reg_read0xbc/0x130
 [<0000000040235bcc>] vfs_read0xc4/0x1d0
 [<0000000040235f0c>] SyS_read0x94/0xf0
 [<0000000040105fc0>] syscall_exit0x0/0x14

kpageflags_read() walks through the whole memory, even if some memory
areas are physically not available. So, we should better not BUG on an
unavailable pfn in pfn_to_nid() but just return the expected value -1 or
0.

Signed-off-by: Helge Deller <deller@gmx.de>
11 years agoparisc: memory overflow, 'name' length is too short for using
Chen Gang [Mon, 27 May 2013 04:57:09 +0000 (04:57 +0000)]
parisc: memory overflow, 'name' length is too short for using

'path.bc[i]' can be asigned by PCI_SLOT() which can '> 10', so sizeof(6
* "%u:" + "%u" + '\0') may be 21.

Since 'name' length is 20, it may be memory overflow.

And 'path.bc[i]' is 'unsigned char' for printing, we can be sure the
max length of 'name' must be less than 28.

So simplify thinking, we can use 28 instead of 20 directly, and do not
think of whether 'patchc.bc[i]' can '> 100'.

Signed-off-by: Chen Gang <gang.chen@asianux.com>
Signed-off-by: Helge Deller <deller@gmx.de>
11 years agoMerge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc
Linus Torvalds [Sat, 1 Jun 2013 11:13:16 +0000 (20:13 +0900)]
Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc

Pull powerpc fixes from Ben Herrenschmidt:
 "Here are a few more fixes for powerpc 3.10.  It's a bit more than I
  would have liked this late in the game but I suppose that's what
  happens with a brand new chip generation coming out.

  A few regression fixes, some last minute fixes for new P8 features
  such as transactional memory,...

  There's also one powerpc KVM patch that I requested that adds two
  missing functions to our in-kernel interrupt controller support which
  is itself a new 3.10 feature.  These are defined by the base
  hypervisor specification.  We didn't implement them originally because
  Linux doesn't use them but they are simple and I'm not comfortable
  having a half-implemented interface in 3.10 and having to deal with
  versionning etc...  later when something starts needing those calls.
  They cannot be emulated in qemu when using in-kernel interrupt
  controller (not enough shared state).

  Just added a last minute patch to fix a typo introducing a breakage in
  our cputable for Power7+ processors, sorry about that, but the
  regression it fixes just hurt me :-)"

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc:
  powerpc/cputable: Fix typo on P7+ cputable entry
  powerpc/perf: Add missing SIER support
  powerpc/perf: Revert to original NO_SIPR logic
  powerpc/pci: Remove the unused variables in pci_process_bridge_OF_ranges
  powerpc/pci: Remove the stale comments of pci_process_bridge_OF_ranges
  powerpc/pseries: Always enable CONFIG_HOTPLUG_CPU on PSERIES SMP
  powerpc/kvm/book3s: Add support for H_IPOLL and H_XIRR_X in XICS emulation
  powerpc/32bit:Store temporary result in r0 instead of r8
  powerpc/mm: Always invalidate tlb on hpte invalidate and update
  powerpc/pseries: Improve stream generation comments in copypage/user
  powerpc/pseries: Kill all prefetch streams on context switch
  powerpc/cputable: Fix oprofile_cpu_type on power8
  powerpc/mpic: Fix irq distribution problem when MPIC_SINGLE_DEST_CPU
  powerpc/tm: Fix userspace stack corruption on signal delivery for active transactions
  powerpc/tm: Move TM abort cause codes to uapi
  powerpc/tm: Abort on emulation and alignment faults
  powerpc/tm: Update cause codes documentation
  powerpc/tm: Make room for hypervisor in abort cause codes

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending
Linus Torvalds [Sat, 1 Jun 2013 11:05:20 +0000 (20:05 +0900)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending

Pull scsi target fixes from Nicholas Bellinger:
 "The highlights include:

   - Re-instate sess->wait_list in target_wait_for_sess_cmds() for
     active I/O shutdown handling in fabrics using se_cmd->cmd_kref
   - Make ib_srpt call target_sess_cmd_list_set_waiting() during session
     shutdown
   - Fix FILEIO off-by-one READ_CAPACITY bug for !S_ISBLK export
   - Fix iscsi-target login error heap buffer overflow (Kees)
   - Fix iscsi-target active I/O shutdown handling regression in
     v3.10-rc1

  A big thanks to Kees Cook for fixing a long standing login error
  buffer overflow bug.

  All patches are CC'ed to stable with the exception of the v3.10-rc1
  specific regression + other minor target cleanup."

* git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending:
  iscsi-target: Fix iscsit_free_cmd() se_cmd->cmd_kref shutdown handling
  target: Propigate up ->cmd_kref put return via transport_generic_free_cmd
  iscsi-target: fix heap buffer overflow on error
  target/file: Fix off-by-one READ_CAPACITY bug for !S_ISBLK export
  ib_srpt: Call target_sess_cmd_list_set_waiting during shutdown_session
  target: Re-instate sess_wait_list for target_wait_for_sess_cmds
  target: Remove unused wait_for_tasks bit in target_wait_for_sess_cmds

11 years agoMerge tag 'clk-fixes-for-linus' of git://git.linaro.org/people/mturquette/linux
Linus Torvalds [Sat, 1 Jun 2013 10:55:26 +0000 (19:55 +0900)]
Merge tag 'clk-fixes-for-linus' of git://git.linaro.org/people/mturquette/linux

Pull clock subsystem fixes from Mike Turquette:
 "A mix of small fixes affecting mostly ARM platforms as well as a
  discrete clock expander chip.  Most fixes are corrections to lousy
  clock data of one form or another."

* tag 'clk-fixes-for-linus' of git://git.linaro.org/people/mturquette/linux:
  clk: mxs: Include clk mxs header file
  clk: vt8500: Fix unbalanced spinlock in vt8500_dclk_set_rate()
  clk: si5351: Set initial clkout rate when defined in platform data.
  clk: si5351: Fix clkout rate computation.
  clk: samsung: Add CLK_IGNORE_UNUSED flag for the sysreg clocks
  clk: ux500: clk-sysctrl: handle clocks with no parents
  clk: ux500: Provide device enumeration number suffix for SMSC911x

11 years agoMerge tag 'fbdev-for-3.10-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/plagn...
Linus Torvalds [Sat, 1 Jun 2013 10:53:41 +0000 (19:53 +0900)]
Merge tag 'fbdev-for-3.10-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/plagnioj/linux-fbdev

Pull fbdev fixes from Jean-Christophe PLAGNIOL-VILLARD:
 "This contains some small fixes

   - Atmel LCDC: fix blank the backlight on remove
   - ps3fb: fix compile warning
   - OMAPDSS: Fix crash with DT boot"

* tag 'fbdev-for-3.10-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/plagnioj/linux-fbdev:
  atmel_lcdfb: blank the backlight on remove
  trivial: atmel_lcdfb: add missing error message
  OMAPDSS: Fix crash with DT boot
  fbdev/ps3fb: fix compile warning

11 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Sat, 1 Jun 2013 10:51:52 +0000 (19:51 +0900)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs

Pull assorted fixes from Al Viro:
 "There'll be more - I'm trying to dig out from under the pile of mail
  (a couple of weeks of something flu-like ;-/) and there's several more
  things waiting for review; this is just the obvious stuff."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  zoran: racy refcount handling in vm_ops ->open()/->close()
  befs_readdir(): do not increment ->f_pos if filldir tells us to stop
  hpfs: deadlock and race in directory lseek()
  qnx6: qnx6_readdir() has a braino in pos calculation
  fix buffer leak after "scsi: saner replacements for ->proc_info()"
  vfs: Fix invalid ida_remove() call

11 years agoMerge tag 'nfs-for-3.10-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Sat, 1 Jun 2013 10:48:59 +0000 (19:48 +0900)]
Merge tag 'nfs-for-3.10-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull two NFS client fixes from Trond Myklebust:
 - Fix a regression that broke NFS mounting using klibc and busybox
 - Stable fix to check access modes correctly on NFSv4 delegated open()

* tag 'nfs-for-3.10-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  NFS: Fix security flavor negotiation with legacy binary mounts
  NFSv4: Fix a thinko in nfs4_try_open_cached

11 years agopowerpc/cputable: Fix typo on P7+ cputable entry
Will Schmidt [Mon, 20 May 2013 05:04:18 +0000 (05:04 +0000)]
powerpc/cputable: Fix typo on P7+ cputable entry

Fix a typo in setting COMMON_USER2_POWER7 bits to .cpu_user_features2
cpu specs table.

Signed-off-by: Will Schmidt <will_schmidt@vnet.ibm.com>
Acked-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/perf: Add missing SIER support
Michael Ellerman [Wed, 15 May 2013 20:19:31 +0000 (20:19 +0000)]
powerpc/perf: Add missing SIER support

Commit 8f61aa3 "Add support for SIER" missed updates to siar_valid()
and perf_get_data_addr().

In both cases we need to check the SIER instead of mmcra.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/perf: Revert to original NO_SIPR logic
Michael Ellerman [Wed, 15 May 2013 20:19:30 +0000 (20:19 +0000)]
powerpc/perf: Revert to original NO_SIPR logic

This is a revert and then some of commit 860aad7 "Add regs_no_sipr()".
This workaround was only needed on early chip versions.

As before NO_SIPR becomes a static flag of the PMU struct.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pci: Remove the unused variables in pci_process_bridge_OF_ranges
Kevin Hao [Thu, 16 May 2013 20:58:42 +0000 (20:58 +0000)]
powerpc/pci: Remove the unused variables in pci_process_bridge_OF_ranges

The codes which ever used these two variables have gone. Throw away
them too.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pci: Remove the stale comments of pci_process_bridge_OF_ranges
Kevin Hao [Thu, 16 May 2013 20:58:41 +0000 (20:58 +0000)]
powerpc/pci: Remove the stale comments of pci_process_bridge_OF_ranges

These comments already don't apply to the current code. So just remove
them.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pseries: Always enable CONFIG_HOTPLUG_CPU on PSERIES SMP
Srivatsa S. Bhat [Tue, 21 May 2013 09:32:48 +0000 (09:32 +0000)]
powerpc/pseries: Always enable CONFIG_HOTPLUG_CPU on PSERIES SMP

Adam Lackorzynski reported the following build failure on
!CONFIG_HOTPLUG_CPU configuration:

  CC      arch/powerpc/kernel/rtas.o
arch/powerpc/kernel/rtas.c: In function â€˜rtas_cpu_state_change_mask’:
arch/powerpc/kernel/rtas.c:843:4: error: implicit declaration of function â€˜cpu_down’ [-Werror=implicit-function-declaration]
cc1: all warnings being treated as errors
make[1]: *** [arch/powerpc/kernel/rtas.o] Error 1
make: *** [arch/powerpc/kernel] Error 2

The build fails because cpu_down() is defined only under CONFIG_HOTPLUG_CPU.

Looking further, the mobility code in pseries is one of the call-sites which
uses rtas_ibm_suspend_me(), which in turn calls rtas_cpu_state_change_mask().
And the mobility code is unconditionally compiled-in (it does not fall under
any Kconfig option). And commit 120496ac (powerpc: Bring all threads online
prior to migration/hibernation) which introduced this build regression is
critical for the proper functioning of the migration code. So it appears
that the only solution to this problem is to enable CONFIG_HOTPLUG_CPU if
SMP is enabled on PPC_PSERIES platforms. So make that change in the Kconfig.

Reported-by: Adam Lackorzynski <adam@os.inf.tu-dresden.de>
Cc: stable@vger.kernel.org
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/kvm/book3s: Add support for H_IPOLL and H_XIRR_X in XICS emulation
Paul Mackerras [Thu, 23 May 2013 15:42:21 +0000 (15:42 +0000)]
powerpc/kvm/book3s: Add support for H_IPOLL and H_XIRR_X in XICS emulation

This adds the remaining two hypercalls defined by PAPR for manipulating
the XICS interrupt controller, H_IPOLL and H_XIRR_X.  H_IPOLL returns
information about the priority and pending interrupts for a virtual
cpu, without changing any state.  H_XIRR_X is like H_XIRR in that it
reads and acknowledges the highest-priority pending interrupt, but it
also returns the timestamp (timebase register value) from when the
interrupt was first received by the hypervisor.  Currently we just
return the current time, since we don't do any software queueing of
virtual interrupts inside the XICS emulation code.

These hcalls are not currently used by Linux guests, but may be in
future.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/32bit:Store temporary result in r0 instead of r8
Priyanka Jain [Fri, 31 May 2013 01:20:02 +0000 (01:20 +0000)]
powerpc/32bit:Store temporary result in r0 instead of r8

Commit a9c4e541ea9b22944da356f2a9258b4eddcc953b
"powerpc/kprobe: Complete kprobe and migrate exception frame"
introduced a regression:

While returning from exception handling in case of PREEMPT enabled,
_TIF_NEED_RESCHED bit is checked in TI_FLAGS (thread_info flag) of current
task. Only if this bit is set, it should continue with the process of
calling preempt_schedule_irq() to schedule highest priority task if
available.

Current code assumes that r8 contains TI_FLAGS and check this for
_TIF_NEED_RESCHED, but as r8 is modified in the code which executes before
this check, r8 no longer contains the expected TI_FLAGS information.

As a result check for comparison with _TIF_NEED_RESCHED was failing even if
NEED_RESCHED bit is set in the current thread_info flag. Due to this,
preempt_schedule_irq() and in turn scheduler was not getting called even if
highest priority task is ready for execution.

So, store temporary results in r0 instead of r8 to prevent r8 from getting
modified as subsequent code is dependent on its value.

Signed-off-by: Priyanka Jain <Priyanka.Jain@freescale.com>
CC: <stable@vger.kernel.org> [v3.7+]
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/mm: Always invalidate tlb on hpte invalidate and update
Aneesh Kumar K.V [Fri, 31 May 2013 01:03:24 +0000 (01:03 +0000)]
powerpc/mm: Always invalidate tlb on hpte invalidate and update

If a hash bucket gets full, we "evict" a more/less random entry from it.
When we do that we don't invalidate the TLB (hpte_remove) because we assume
the old translation is still technically "valid". This implies that when
we are invalidating or updating pte, even if HPTE entry is not valid
we should do a tlb invalidate.

This was a regression introduced by b1022fbd293564de91596b8775340cf41ad5214c

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pseries: Improve stream generation comments in copypage/user
Michael Neuling [Wed, 29 May 2013 19:34:29 +0000 (19:34 +0000)]
powerpc/pseries: Improve stream generation comments in copypage/user

No code changes, just documenting what's happening a little better.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/pseries: Kill all prefetch streams on context switch
Michael Neuling [Wed, 29 May 2013 19:34:27 +0000 (19:34 +0000)]
powerpc/pseries: Kill all prefetch streams on context switch

On context switch, we should have no prefetch streams leak from one
userspace process to another.  This frees up prefetch resources for the
next process.

Based on patch from Milton Miller.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/cputable: Fix oprofile_cpu_type on power8
Nishanth Aravamudan [Tue, 28 May 2013 10:39:50 +0000 (10:39 +0000)]
powerpc/cputable: Fix oprofile_cpu_type on power8

Maynard informed me that neither the oprofile kernel module nor oprofile
userspace has been updated to support that "legacy" oprofile module
interface for power8, which is indicated by "ppc64/power8." This results
in no samples. The solution is to default to the "timer" type, instead.
The raw entry also should be updated, as "ppc64/ibm-compat-v1" indicates
to oprofile userspace to use "compatibility events" which are obsolete
in ISA 2.07.

Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/mpic: Fix irq distribution problem when MPIC_SINGLE_DEST_CPU
chenhui zhao [Mon, 27 May 2013 21:59:43 +0000 (21:59 +0000)]
powerpc/mpic: Fix irq distribution problem when MPIC_SINGLE_DEST_CPU

For the mpic with a flag MPIC_SINGLE_DEST_CPU, only one bit should be
set in interrupt destination registers.

The code is applicable to 64-bit platforms as well as 32-bit.

Signed-off-by: Zhao Chenhui <chenhui.zhao@freescale.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/tm: Fix userspace stack corruption on signal delivery for active transactions
Michael Neuling [Sun, 26 May 2013 18:09:41 +0000 (18:09 +0000)]
powerpc/tm: Fix userspace stack corruption on signal delivery for active transactions

When in an active transaction that takes a signal, we need to be careful with
the stack.  It's possible that the stack has moved back up after the tbegin.
The obvious case here is when the tbegin is called inside a function that
returns before a tend.  In this case, the stack is part of the checkpointed
transactional memory state.  If we write over this non transactionally or in
suspend, we are in trouble because if we get a tm abort, the program counter
and stack pointer will be back at the tbegin but our in memory stack won't be
valid anymore.

To avoid this, when taking a signal in an active transaction, we need to use
the stack pointer from the checkpointed state, rather than the speculated
state.  This ensures that the signal context (written tm suspended) will be
written below the stack required for the rollback.  The transaction is aborted
becuase of the treclaim, so any memory written between the tbegin and the
signal will be rolled back anyway.

For signals taken in non-TM or suspended mode, we use the
normal/non-checkpointed stack pointer.

Tested with 64 and 32 bit signals

Signed-off-by: Michael Neuling <mikey@neuling.org>
Cc: <stable@vger.kernel.org> # v3.9
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/tm: Move TM abort cause codes to uapi
Michael Neuling [Sun, 26 May 2013 18:30:56 +0000 (18:30 +0000)]
powerpc/tm: Move TM abort cause codes to uapi

These cause codes are usable by userspace, so let's export to uapi.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Cc: <stable@vger.kernel.org> # v3.9
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/tm: Abort on emulation and alignment faults
Michael Neuling [Sun, 26 May 2013 18:09:39 +0000 (18:09 +0000)]
powerpc/tm: Abort on emulation and alignment faults

If we are emulating an instruction inside an active user transaction that
touches memory, the kernel can't emulate it as it operates in transactional
suspend context.  We need to abort these transactions and send them back to
userspace for the hardware to rollback.

We can service these if the user transaction is in suspend mode, since the
kernel will operate in the same suspend context.

This adds a check to all alignment faults and to specific instruction
emulations (only string instructions for now).  If the user process is in an
active (non-suspended) transaction, we abort the transaction go back to
userspace allowing the HW to roll back the transaction and tell the user of the
failure.  This also adds new tm abort cause codes to report the reason of the
persistent error to the user.

Crappy test case here http://neuling.org/devel/junkcode/aligntm.c

Signed-off-by: Michael Neuling <mikey@neuling.org>
Cc: <stable@vger.kernel.org> # v3.9
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/tm: Update cause codes documentation
Michael Neuling [Sun, 26 May 2013 18:09:38 +0000 (18:09 +0000)]
powerpc/tm: Update cause codes documentation

Signed-off-by: Michael Neuling <mikey@neuling.org>
Cc: <stable@vger.kernel.org> # 3.9 only
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agopowerpc/tm: Make room for hypervisor in abort cause codes
Michael Neuling [Sun, 26 May 2013 18:09:37 +0000 (18:09 +0000)]
powerpc/tm: Make room for hypervisor in abort cause codes

PAPR carves out 0xff-0xe0 for hypervisor use of transactional memory software
abort cause codes.  Unfortunately we don't respect this currently.

Below fixes this to move our cause codes to below this region.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Cc: <stable@vger.kernel.org> # 3.9 only
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
11 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs
Linus Torvalds [Fri, 31 May 2013 21:59:14 +0000 (06:59 +0900)]
Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs

Pull reiserfs fixes from Jan Kara:
 "Three reiserfs fixes.  They fix real problems spotted by users so I
  hope they are ok even at this stage."

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs:
  reiserfs: fix deadlock with nfs racing on create/lookup
  reiserfs: fix problems with chowning setuid file w/ xattrs
  reiserfs: fix spurious multiple-fill in reiserfs_readdir_dentry

11 years agoMerge tag 'for-linus-v3.10-rc4-crc-xattr-fixes' of git://oss.sgi.com/xfs/xfs
Linus Torvalds [Fri, 31 May 2013 21:56:21 +0000 (06:56 +0900)]
Merge tag 'for-linus-v3.10-rc4-crc-xattr-fixes' of git://oss.sgi.com/xfs/xfs

Pull xfs extended attribute fixes for CRCs from Ben Myers:
 "Here are several fixes that are relevant on CRC enabled XFS
  filesystems.  They are followed by a rework of the remote attribute
  code so that each block of the attribute contains a header with a CRC.

  Previously there was a CRC header per extent in the remote attribute
  code, but this was untenable because it was not possible to know how
  many extents would be allocated for the attribute until after the
  allocation has completed, due to the fragmentation of free space.
  This became complicated because the size of the headers needs to be
  added to the length of the payload to get the overall length required
  for the allocation.  With a header per block, things are less
  complicated at the cost of a little space.

  I would have preferred to defer this and the rest of the CRC queue to
  3.11 to mitigate risk for existing non-crc users in 3.10.  Doing so
  would require setting a feature bit for the on-disk changes, and so I
  have been pressured into sending this pull request by Eric Sandeen and
  David Chinner from Red Hat.  I'll send another pull request or two
  with the rest of the CRC queue next week.

   - Remove assert on count of remote attribute CRC headers
   - Fix the number of blocks read in for remote attributes
   - Zero remote attribute tails properly
   - Fix mapping of remote attribute buffers to have correct length
   - initialize temp leaf properly in xfs_attr3_leaf_unbalance, and
     xfs_attr3_leaf_compact
   - Rework remote atttributes to have a header per block"

* tag 'for-linus-v3.10-rc4-crc-xattr-fixes' of git://oss.sgi.com/xfs/xfs:
  xfs: rework remote attr CRCs
  xfs: fully initialise temp leaf in xfs_attr3_leaf_compact
  xfs: fully initialise temp leaf in xfs_attr3_leaf_unbalance
  xfs: correctly map remote attr buffers during removal
  xfs: remote attribute tail zeroing does too much
  xfs: remote attribute read too short
  xfs: remote attribute allocation may be contiguous

11 years agoMerge tag 'for-linus-v3.10-rc4' of git://oss.sgi.com/xfs/xfs
Linus Torvalds [Fri, 31 May 2013 21:50:16 +0000 (06:50 +0900)]
Merge tag 'for-linus-v3.10-rc4' of git://oss.sgi.com/xfs/xfs

Pull xfs fixes from Ben Myers:
 - Fix nested transactions in xfs_qm_scall_setqlim
 - Clear suid/sgid bits when we truncate with size update
 - Fix recovery for split buffers
 - Fix block count on remote symlinks
 - Add fsgeom flag for v5 superblock support
 - Disable XFS_IOC_SWAPEXT for CRC enabled filesystems
 - Fix dirv3 freespace block corruption

* tag 'for-linus-v3.10-rc4' of git://oss.sgi.com/xfs/xfs:
  xfs: fix dir3 freespace block corruption
  xfs: disable swap extents ioctl on CRC enabled filesystems
  xfs: add fsgeom flag for v5 superblock support.
  xfs: fix incorrect remote symlink block count
  xfs: fix split buffer vector log recovery support
  xfs: kill suid/sgid through the truncate path.
  xfs: avoid nesting transactions in xfs_qm_scall_setqlim()

11 years agoMerge tag 'please-pull-aertracefix' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 31 May 2013 21:47:04 +0000 (06:47 +0900)]
Merge tag 'please-pull-aertracefix' of git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras

Pull aer error logging fix from Tony Luck:
 "Can't call pci_get_domain_bus_and_slot() from interupt context"

* tag 'please-pull-aertracefix' of git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras:
  aerdrv: Move cper_print_aer() call out of interrupt context

11 years agoMerge tag 'arm64-stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmarinas...
Linus Torvalds [Fri, 31 May 2013 21:45:10 +0000 (06:45 +0900)]
Merge tag 'arm64-stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmarinas/linux-aarch64

Pull arm64 fixes from Catalin Marinas:
 - Module compilation issues (symbol not exported).
 - Plug a hole where user space can bring the kernel down.

* tag 'arm64-stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmarinas/linux-aarch64:
  arm64: don't kill the kernel on a bad esr from el0
  arm64: treat unhandled compat el0 traps as undef
  arm64: Do not report user faults for handled signals
  arm64: kernel: compiling issue, need 'EXPORT_SYMBOL(clear_page)'

11 years agoreiserfs: fix deadlock with nfs racing on create/lookup
Jeff Mahoney [Fri, 31 May 2013 19:51:17 +0000 (15:51 -0400)]
reiserfs: fix deadlock with nfs racing on create/lookup

Reiserfs is currently able to be deadlocked by having two NFS clients
where one has removed and recreated a file and another is accessing the
file with an open file handle.

If one client deletes and recreates a file with timing such that the
recreated file obtains the same [dirid, objectid] pair as the original
file while another client accesses the file via file handle, the create
and lookup can race and deadlock if the lookup manages to create the
in-memory inode first.

The create thread, in insert_inode_locked4, will hold the write lock
while waiting on the other inode to be unlocked. The lookup thread,
anywhere in the iget path, will release and reacquire the write lock while
it schedules. If it needs to reacquire the lock while the create thread
has it, it will never be able to make forward progress because it needs
to reacquire the lock before ultimately unlocking the inode.

This patch drops the write lock across the insert_inode_locked4 call so
that the ordering of inode_wait -> write lock is retained. Since this
would have been the case before the BKL push-down, this is safe.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Jan Kara <jack@suse.cz>
11 years agoreiserfs: fix problems with chowning setuid file w/ xattrs
Jeff Mahoney [Fri, 31 May 2013 19:54:17 +0000 (15:54 -0400)]
reiserfs: fix problems with chowning setuid file w/ xattrs

reiserfs_chown_xattrs() takes the iattr struct passed into ->setattr
and uses it to iterate over all the attrs associated with a file to change
ownership of xattrs (and transfer quota associated with the xattr files).

When the setuid bit is cleared during chown, ATTR_MODE and iattr->ia_mode
are passed to all the xattrs as well. This means that the xattr directory
will have S_IFREG added to its mode bits.

This has been prevented in practice by a missing IS_PRIVATE check
in reiserfs_acl_chmod, which caused a double-lock to occur while holding
the write lock. Since the file system was completely locked up, the
writeout of the corrupted mode never happened.

This patch temporarily clears everything but ATTR_UID|ATTR_GID for the
calls to reiserfs_setattr and adds the missing IS_PRIVATE check.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Jan Kara <jack@suse.cz>
11 years agoreiserfs: fix spurious multiple-fill in reiserfs_readdir_dentry
Jeff Mahoney [Fri, 31 May 2013 19:07:52 +0000 (15:07 -0400)]
reiserfs: fix spurious multiple-fill in reiserfs_readdir_dentry

After sleeping for filldir(), we check to see if the file system has
changed and research. The next_pos pointer is updated but its value
isn't pushed into the key used for the search itself. As a result,
the search returns the same item that the last cycle of the loop did
and filldir() is called multiple times with the same data.

The end result is that the buffer can contain the same name multiple
times. This can be returned to userspace or used internally in the
xattr code where it can manifest with the following warning:

jdm-20004 reiserfs_delete_xattrs: Couldn't delete all xattrs (-2)

reiserfs_for_each_xattr uses reiserfs_readdir_dentry to iterate over
the xattr names and ends up trying to unlink the same name twice. The
second attempt fails with -ENOENT and the error is returned. At some
point I'll need to add support into reiserfsck to remove the orphaned
directories left behind when this occurs.

The fix is to push the value into the key before researching.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Jan Kara <jack@suse.cz>
11 years agozoran: racy refcount handling in vm_ops ->open()/->close()
Al Viro [Thu, 23 May 2013 08:38:22 +0000 (04:38 -0400)]
zoran: racy refcount handling in vm_ops ->open()/->close()

worse, we lock ->resource_lock too late when we are destroying the
final clonal VMA; the check for lack of other mappings of the same
opened file can race with mmap().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agoatmel_lcdfb: blank the backlight on remove
Richard Genoud [Fri, 31 May 2013 15:49:35 +0000 (15:49 +0000)]
atmel_lcdfb: blank the backlight on remove

When removing atmel_lcdfb module, the backlight is unregistered but not
blanked. (only for CONFIG_BACKLIGHT_ATMEL_LCDC case).
This can result in the screen going full white depending on how the PWM
is wired.

Signed-off-by: Richard Genoud <richard.genoud@gmail.com>
Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
11 years agotrivial: atmel_lcdfb: add missing error message
Richard Genoud [Fri, 31 May 2013 14:28:38 +0000 (14:28 +0000)]
trivial: atmel_lcdfb: add missing error message

When a too small framebuffer is given, the atmel_lcdfb_check_var
silently fails.
Adding an error message will save some head scratching.

Signed-off-by: Richard Genoud <richard.genoud@gmail.com>
Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
11 years agobefs_readdir(): do not increment ->f_pos if filldir tells us to stop
Al Viro [Wed, 22 May 2013 17:41:26 +0000 (13:41 -0400)]
befs_readdir(): do not increment ->f_pos if filldir tells us to stop

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agohpfs: deadlock and race in directory lseek()
Al Viro [Sat, 18 May 2013 06:38:52 +0000 (02:38 -0400)]
hpfs: deadlock and race in directory lseek()

For one thing, there's an ABBA deadlock on hpfs fs-wide lock and i_mutex
in hpfs_dir_lseek() - there's a lot of methods that grab the former with
the caller already holding the latter, so it must take i_mutex first.

For another, locking the damn thing, carefully validating the offset,
then dropping locks and assigning the offset is obviously racy.

Moreover, we _must_ do hpfs_add_pos(), or the machinery in dnode.c
won't modify the sucker on B-tree surgeries.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agoqnx6: qnx6_readdir() has a braino in pos calculation
Al Viro [Fri, 17 May 2013 19:21:56 +0000 (15:21 -0400)]
qnx6: qnx6_readdir() has a braino in pos calculation

We want to mask lower 5 bits out, not leave only those and clear the
rest...  As it is, we end up always starting to read from the beginning
of directory, no matter what the current position had been.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agofix buffer leak after "scsi: saner replacements for ->proc_info()"
Jan Beulich [Wed, 29 May 2013 12:26:53 +0000 (13:26 +0100)]
fix buffer leak after "scsi: saner replacements for ->proc_info()"

That patch failed to set proc_scsi_fops' .release method.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agovfs: Fix invalid ida_remove() call
Takashi Iwai [Fri, 10 May 2013 12:04:11 +0000 (14:04 +0200)]
vfs: Fix invalid ida_remove() call

When the group id of a shared mount is not allocated, the umount still
tries to call mnt_release_group_id(), which eventually hits a kernel
warning at ida_remove() spewing a message like:
  ida_remove called for id=0 which is not allocated.

This patch fixes the bug simply checking the group id in the caller.

Reported-by: Cristian Rodríguez <crrodriguez@opensuse.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agoarm64: don't kill the kernel on a bad esr from el0
Mark Rutland [Tue, 28 May 2013 14:54:15 +0000 (15:54 +0100)]
arm64: don't kill the kernel on a bad esr from el0

Rather than completely killing the kernel if we receive an esr value we
can't deal with in the el0 handlers, send the process a SIGILL and log
the esr value in the hope that we can debug it. If we receive a bad esr
from el1, we'll die() as before.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: stable@vger.kernel.org
11 years agoarm64: treat unhandled compat el0 traps as undef
Mark Rutland [Fri, 24 May 2013 11:02:35 +0000 (12:02 +0100)]
arm64: treat unhandled compat el0 traps as undef

Currently, if a compat process reads or writes from/to a disabled
cp15/cp14 register, the trap is not handled by the el0_sync_compat
handler, and the kernel will head to bad_mode, where it will die(), and
oops(). For 64 bit processes, disabled system register accesses are
currently treated as unhandled instructions.

This patch modifies entry.S to treat these unhandled traps as undefined
instructions, sending a SIGILL to userspace. This gives processes a
chance to handle this and stop using inaccessible registers, and
prevents further issues in the kernel as a result of the die().

Reported-by: Johannes Jensen <Johannes.Jensen@arm.com>
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
11 years agoiscsi-target: Fix iscsit_free_cmd() se_cmd->cmd_kref shutdown handling
Nicholas Bellinger [Fri, 31 May 2013 07:49:41 +0000 (00:49 -0700)]
iscsi-target: Fix iscsit_free_cmd() se_cmd->cmd_kref shutdown handling

With the introduction of target_get_sess_cmd() referencing counting for
ISCSI_OP_SCSI_CMD processing with iser-target, iscsit_free_cmd() usage
in traditional iscsi-target driver code now needs to be aware of the
active I/O shutdown case when a remaining se_cmd->cmd_kref reference may
exist after transport_generic_free_cmd() completes, requiring a final
target_put_sess_cmd() to release iscsi_cmd descriptor memory.

This patch changes iscsit_free_cmd() to invoke __iscsit_free_cmd() before
transport_generic_free_cmd() -> target_put_sess_cmd(), and also avoids
aquiring the per-connection queue locks for typical fast-path calls
during normal ISTATE_REMOVE operation.

Also update iscsit_free_cmd() usage throughout iscsi-target to
use the new 'bool shutdown' parameter.

This patch fixes a regression bug introduced during v3.10-rc1 in
commit 3e1c81a95, that was causing the following WARNING to appear:

[  257.235153] ------------[ cut here]------------
[  257.240314] WARNING: at kernel/softirq.c:160 local_bh_enable_ip+0x3c/0x86()
[  257.248089] Modules linked in: vhost_scsi ib_srpt ib_cm ib_sa ib_mad ib_core tcm_qla2xxx tcm_loop
tcm_fc libfc iscsi_target_mod target_core_pscsi target_core_file
target_core_iblock target_core_mod configfs ipv6 iscsi_tcp libiscsi_tcp
libiscsi scsi_transport_iscsi loop acpi_cpufreq freq_table mperf
kvm_intel kvm crc32c_intel button ehci_pci pcspkr joydev i2c_i801
microcode ext3 jbd raid10 raid456 async_pq async_xor xor async_memcpy
async_raid6_recov raid6_pq async_tx raid1 raid0 linear igb hwmon
i2c_algo_bit i2c_core ptp ata_piix libata qla2xxx uhci_hcd ehci_hcd
mlx4_core scsi_transport_fc scsi_tgt pps_core
[  257.308748] CPU: 1 PID: 3295 Comm: iscsi_ttx Not tainted 3.10.0-rc2+ #103
[  257.316329] Hardware name: Intel Corporation S5520HC/S5520HC, BIOS S5500.86B.01.00.0057.031020111721 03/10/2011
[  257.327597]  ffffffff814c24b7 ffff880458331b58 ffffffff8138eef2 ffff880458331b98
[  257.335892]  ffffffff8102c052 ffff880400000008 0000000000000000 ffff88085bdf0000
[  257.344191]  ffff88085bdf00d8 ffff88085bdf00e0 ffff88085bdf00f8 ffff880458331ba8
[  257.352488] Call Trace:
[  257.355223]  [<ffffffff8138eef2>] dump_stack+0x19/0x1f
[  257.360963]  [<ffffffff8102c052>] warn_slowpath_common+0x62/0x7b
[  257.367669]  [<ffffffff8102c080>] warn_slowpath_null+0x15/0x17
[  257.374181]  [<ffffffff81032345>] local_bh_enable_ip+0x3c/0x86
[  257.380697]  [<ffffffff813917fd>] _raw_spin_unlock_bh+0x10/0x12
[  257.387311]  [<ffffffffa029069c>] iscsit_free_r2ts_from_list+0x5e/0x67 [iscsi_target_mod]
[  257.396438]  [<ffffffffa02906c5>] iscsit_release_cmd+0x20/0x223 [iscsi_target_mod]
[  257.404893]  [<ffffffffa02977a4>] lio_release_cmd+0x3a/0x3e [iscsi_target_mod]
[  257.412964]  [<ffffffffa01d59a1>] target_release_cmd_kref+0x7a/0x7c [target_core_mod]
[  257.421712]  [<ffffffffa01d69bc>] target_put_sess_cmd+0x5f/0x7f [target_core_mod]
[  257.430071]  [<ffffffffa01d6d6d>] transport_release_cmd+0x59/0x6f [target_core_mod]
[  257.438625]  [<ffffffffa01d6eb4>] transport_put_cmd+0x131/0x140 [target_core_mod]
[  257.446985]  [<ffffffffa01d6192>] ? transport_wait_for_tasks+0xfa/0x1d5 [target_core_mod]
[  257.456121]  [<ffffffffa01d6f11>] transport_generic_free_cmd+0x4e/0x52 [target_core_mod]
[  257.465159]  [<ffffffff81050537>] ? __migrate_task+0x110/0x110
[  257.471674]  [<ffffffffa02904ba>] iscsit_free_cmd+0x46/0x55 [iscsi_target_mod]
[  257.479741]  [<ffffffffa0291edb>] iscsit_immediate_queue+0x301/0x353 [iscsi_target_mod]
[  257.488683]  [<ffffffffa0292f7e>] iscsi_target_tx_thread+0x1c6/0x2a8 [iscsi_target_mod]
[  257.497623]  [<ffffffff81047486>] ? wake_up_bit+0x25/0x25
[  257.503652]  [<ffffffffa0292db8>] ? iscsit_ack_from_expstatsn+0xd5/0xd5 [iscsi_target_mod]
[  257.512882]  [<ffffffff81046f89>] kthread+0xb0/0xb8
[  257.518329]  [<ffffffff81046ed9>] ? kthread_freezable_should_stop+0x60/0x60
[  257.526105]  [<ffffffff81396fec>] ret_from_fork+0x7c/0xb0
[  257.532133]  [<ffffffff81046ed9>] ? kthread_freezable_should_stop+0x60/0x60
[  257.539906] ---[ end trace 5520397d0f2e0800 ]---

Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
11 years agotarget: Propigate up ->cmd_kref put return via transport_generic_free_cmd
Nicholas Bellinger [Fri, 31 May 2013 07:46:11 +0000 (00:46 -0700)]
target: Propigate up ->cmd_kref put return via transport_generic_free_cmd

Go ahead and propigate up the ->cmd_kref put return value from
target_put_sess_cmd() -> transport_release_cmd() -> transport_put_cmd()
-> transport_generic_free_cmd().

This is useful for certain fabrics when determining the active I/O
shutdown case with SCF_ACK_KREF where a final target_put_sess_cmd()
is still required by the caller.

Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
11 years agoMerge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Fri, 31 May 2013 06:04:05 +0000 (16:04 +1000)]
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "One qxl 32-bit warning fix, the rest is a bunch of radeon fixes from
  Alex for some issues we've been seeing."

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm/qxl: fix build warnings on 32-bit
  radeon: use max_bus_speed to activate gen2 speeds
  drm/radeon: narrow scope of Apple re-POST hack
  drm/radeon: don't check crtcs in card_posted() on cards without DCE
  drm/radeon: fix card_posted check for newer asics
  drm/radeon: fix typo in cu_per_sh on verde
  drm/radeon: UVD block on SUMO2 is the same as on SUMO

11 years agodrm/qxl: fix build warnings on 32-bit
Dave Airlie [Fri, 31 May 2013 02:45:09 +0000 (12:45 +1000)]
drm/qxl: fix build warnings on 32-bit

Just the usual printk related warnings.

Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoclk: mxs: Include clk mxs header file
Fabio Estevam [Mon, 27 May 2013 15:28:25 +0000 (12:28 -0300)]
clk: mxs: Include clk mxs header file

Fix the following sparse warnings:

drivers/clk/mxs/clk-imx28.c:72:5: warning: symbol 'mxs_saif_clkmux_select' was not declared. Should it be static?
drivers/clk/mxs/clk-imx28.c:156:12: warning: symbol 'mx28_clocks_init' was not declared. Should it be static?

Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Acked-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Mike Turquette <mturquette@linaro.org>
[mturquette@linaro.org: fixed $SUBJECT line]

11 years agoiscsi-target: fix heap buffer overflow on error
Kees Cook [Thu, 23 May 2013 17:32:17 +0000 (10:32 -0700)]
iscsi-target: fix heap buffer overflow on error

If a key was larger than 64 bytes, as checked by iscsi_check_key(), the
error response packet, generated by iscsi_add_notunderstood_response(),
would still attempt to copy the entire key into the packet, overflowing
the structure on the heap.

Remote preauthentication kernel memory corruption was possible if a
target was configured and listening on the network.

CVE-2013-2850

Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: stable@vger.kernel.org
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
11 years agoMerge branch 'for-3.10' of git://linux-nfs.org/~bfields/linux
Linus Torvalds [Fri, 31 May 2013 00:48:56 +0000 (09:48 +0900)]
Merge branch 'for-3.10' of git://linux-nfs.org/~bfields/linux

Pull nfsd fixes from Bruce Fields:
 "A couple minor fixes for the (new to 3.10) gss-proxy code.

  And one regression from user-namespace changes.  (XBMC clients were
  doing something admittedly weird--sending -1 gid's--but something that
  we used to allow.)"

* 'for-3.10' of git://linux-nfs.org/~bfields/linux:
  svcrpc: fix failures to handle -1 uid's and gid's
  svcrpc: implement O_NONBLOCK behavior for use-gss-proxy
  svcauth_gss: fix error code in use_gss_proxy()

11 years agotarget/file: Fix off-by-one READ_CAPACITY bug for !S_ISBLK export
Nicholas Bellinger [Thu, 30 May 2013 04:35:23 +0000 (21:35 -0700)]
target/file: Fix off-by-one READ_CAPACITY bug for !S_ISBLK export

This patch fixes a bug where FILEIO was incorrectly reporting the number
of logical blocks (+ 1) when using non struct block_device export mode.

It changes fd_get_blocks() to follow all other backend ->get_blocks() cases,
and reduces the calculated dev_size by one dev->dev_attrib.block_size
number of bytes, and also fixes initial fd_block_size assignment at
fd_configure_device() time introduced in commit 0fd97ccf4.

Reported-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Reported-by: Badari Pulavarty <pbadari@us.ibm.com>
Tested-by: Badari Pulavarty <pbadari@us.ibm.com>
Cc: stable@vger.kernel.org
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
11 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 31 May 2013 00:44:10 +0000 (09:44 +0900)]
Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 fixes from Peter Anvin:

 - Three EFI-related fixes

 - Two early memory initialization fixes

 - build fix for older binutils

 - fix for an eager FPU performance regression -- currently we don't
   allow the use of the FPU at interrupt time *at all* in eager mode,
   which is clearly wrong.

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86: Allow FPU to be used at interrupt time even with eagerfpu
  x86, crc32-pclmul: Fix build with older binutils
  x86-64, init: Fix a possible wraparound bug in switchover in head_64.S
  x86, range: fix missing merge during add range
  x86, efi: initial the local variable of DataSize to zero
  efivar: fix oops in efivar_update_sysfs_entries() caused by memory reuse
  efivarfs: Never return ENOENT from firmware again

11 years agox86: Allow FPU to be used at interrupt time even with eagerfpu
Pekka Riikonen [Mon, 13 May 2013 12:32:07 +0000 (14:32 +0200)]
x86: Allow FPU to be used at interrupt time even with eagerfpu

With the addition of eagerfpu the irq_fpu_usable() now returns false
negatives especially in the case of ksoftirqd and interrupted idle task,
two common cases for FPU use for example in networking/crypto.  With
eagerfpu=off FPU use is possible in those contexts.  This is because of
the eagerfpu check in interrupted_kernel_fpu_idle():

...
  * For now, with eagerfpu we will return interrupted kernel FPU
  * state as not-idle. TBD: Ideally we can change the return value
  * to something like __thread_has_fpu(current). But we need to
  * be careful of doing __thread_clear_has_fpu() before saving
  * the FPU etc for supporting nested uses etc. For now, take
  * the simple route!
...
  if (use_eager_fpu())
  return 0;

As eagerfpu is automatically "on" on those CPUs that also have the
features like AES-NI this patch changes the eagerfpu check to return 1 in
case the kernel_fpu_begin() has not been said yet.  Once it has been the
__thread_has_fpu() will start returning 0.

Notice that with eagerfpu the __thread_has_fpu is always true initially.
FPU use is thus always possible no matter what task is under us, unless
the state has already been saved with kernel_fpu_begin().

[ hpa: this is a performance regression, not a correctness regression,
  but since it can be quite serious on CPUs which need encryption at
  interrupt time I am marking this for urgent/stable. ]

Signed-off-by: Pekka Riikonen <priikone@iki.fi>
Link: http://lkml.kernel.org/r/alpine.GSO.2.00.1305131356320.18@git.silcnet.org
Cc: <stable@vger.kernel.org> v3.7+
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
11 years agox86, crc32-pclmul: Fix build with older binutils
Jan Beulich [Wed, 29 May 2013 12:43:54 +0000 (13:43 +0100)]
x86, crc32-pclmul: Fix build with older binutils

binutils prior to 2.18 (e.g. the ones found on SLE10) don't support
assembling PEXTRD, so a macro based approach like the one for PCLMULQDQ
in the same file should be used.

This requires making the helper macros capable of recognizing 32-bit
general purpose register operands.

[ hpa: tagging for stable as it is a low risk build fix ]

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Link: http://lkml.kernel.org/r/51A6142A02000078000D99D8@nat28.tlf.novell.com
Cc: Alexander Boyko <alexander_boyko@xyratex.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Huang Ying <ying.huang@intel.com>
Cc: <stable@vger.kernel.org> v3.9
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
11 years agoxfs: rework remote attr CRCs
Dave Chinner [Tue, 21 May 2013 08:02:08 +0000 (18:02 +1000)]
xfs: rework remote attr CRCs

Note: this changes the on-disk remote attribute format. I assert
that this is OK to do as CRCs are marked experimental and the first
kernel it is included in has not yet reached release yet. Further,
the userspace utilities are still evolving and so anyone using this
stuff right now is a developer or tester using volatile filesystems
for testing this feature. Hence changing the format right now to
save longer term pain is the right thing to do.

The fundamental change is to move from a header per extent in the
attribute to a header per filesytem block in the attribute. This
means there are more header blocks and the parsing of the attribute
data is slightly more complex, but it has the advantage that we
always know the size of the attribute on disk based on the length of
the data it contains.

This is where the header-per-extent method has problems. We don't
know the size of the attribute on disk without first knowing how
many extents are used to hold it. And we can't tell from a
mapping lookup, either, because remote attributes can be allocated
contiguously with other attribute blocks and so there is no obvious
way of determining the actual size of the atribute on disk short of
walking and mapping buffers.

The problem with this approach is that if we map a buffer
incorrectly (e.g. we make the last buffer for the attribute data too
long), we then get buffer cache lookup failure when we map it
correctly. i.e. we get a size mismatch on lookup. This is not
necessarily fatal, but it's a cache coherency problem that can lead
to returning the wrong data to userspace or writing the wrong data
to disk. And debug kernels will assert fail if this occurs.

I found lots of niggly little problems trying to fix this issue on a
4k block size filesystem, finally getting it to pass with lots of
fixes. The thing is, 1024 byte filesystems still failed, and it was
getting really complex handling all the corner cases that were
showing up. And there were clearly more that I hadn't found yet.

It is complex, fragile code, and if we don't fix it now, it will be
complex, fragile code forever more.

Hence the simple fix is to add a header to each filesystem block.
This gives us the same relationship between the attribute data
length and the number of blocks on disk as we have without CRCs -
it's a linear mapping and doesn't require us to guess anything. It
is simple to implement, too - the remote block count calculated at
lookup time can be used by the remote attribute set/get/remove code
without modification for both CRC and non-CRC filesystems. The world
becomes sane again.

Because the copy-in and copy-out now need to iterate over each
filesystem block, I moved them into helper functions so we separate
the block mapping and buffer manupulations from the attribute data
and CRC header manipulations. The code becomes much clearer as a
result, and it is a lot easier to understand and debug. It also
appears to be much more robust - once it worked on 4k block size
filesystems, it has worked without failure on 1k block size
filesystems, too.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit ad1858d77771172e08016890f0eb2faedec3ecee)

11 years agoxfs: fully initialise temp leaf in xfs_attr3_leaf_compact
Dave Chinner [Tue, 21 May 2013 08:02:06 +0000 (18:02 +1000)]
xfs: fully initialise temp leaf in xfs_attr3_leaf_compact

xfs_attr3_leaf_compact() uses a temporary buffer for compacting the
the entries in a leaf. It copies the the original buffer into the
temporary buffer, then zeros the original buffer completely. It then
copies the entries back into the original buffer.  However, the
original buffer has not been correctly initialised, and so the
movement of the entries goes horribly wrong.

Make sure the zeroed destination buffer is fully initialised, and
once we've set up the destination incore header appropriately, write
is back to the buffer before starting to move entries around.

While debugging this, the _d/_s prefixes weren't sufficient to
remind me what buffer was what, so rename then all _src/_dst.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit d4c712bcf26a25c2b67c90e44e0b74c7993b5334)

11 years agoxfs: fully initialise temp leaf in xfs_attr3_leaf_unbalance
Dave Chinner [Tue, 21 May 2013 08:02:05 +0000 (18:02 +1000)]
xfs: fully initialise temp leaf in xfs_attr3_leaf_unbalance

xfs_attr3_leaf_unbalance() uses a temporary buffer for recombining
the entries in two leaves when the destination leaf requires
compaction. The temporary buffer ends up being copied back over the
original destination buffer, so the header in the temporary buffer
needs to contain all the information that is in the destination
buffer.

To make sure the temporary buffer is fully initialised, once we've
set up the temporary incore header appropriately, write is back to
the temporary buffer before starting to move entries around.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit 8517de2a81da830f5d90da66b4799f4040c76dc9)

11 years agoxfs: correctly map remote attr buffers during removal
Dave Chinner [Tue, 21 May 2013 08:02:04 +0000 (18:02 +1000)]
xfs: correctly map remote attr buffers during removal

If we don't map the buffers correctly (same as for get/set
operations) then the incore buffer lookup will fail. If a block
number matches but a length is wrong, then debug kernels will ASSERT
fail in _xfs_buf_find() due to the length mismatch. Ensure that we
map the buffers correctly by basing the length of the buffer on the
attribute data length rather than the remote block count.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit 6863ef8449f1908c19f43db572e4474f24a1e9da)

11 years agoxfs: remote attribute tail zeroing does too much
Dave Chinner [Tue, 21 May 2013 08:02:03 +0000 (18:02 +1000)]
xfs: remote attribute tail zeroing does too much

When an attribute data does not fill then entire remote block, we
zero the remaining part of the buffer. This, however, needs to take
into account that the buffer has a header, and so the offset where
zeroing starts and the length of zeroing need to take this into
account. Otherwise we end up with zeros over the end of the
attribute value when CRCs are enabled.

While there, make sure we only ask to map an extent that covers the
remaining range of the attribute, rather than asking every time for
the full length of remote data. If the remote attribute blocks are
contiguous with other parts of the attribute tree, it will map those
blocks as well and we can potentially zero them incorrectly. We can
also get buffer size mistmatches when trying to read or remove the
remote attribute, and this can lead to not finding the correct
buffer when looking it up in cache.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit 4af3644c9a53eb2f1ecf69cc53576561b64be4c6)

11 years agoxfs: remote attribute read too short
Dave Chinner [Tue, 21 May 2013 08:02:02 +0000 (18:02 +1000)]
xfs: remote attribute read too short

Reading a maximally size remote attribute fails when CRCs are
enabled with this verification error:

XFS (vdb): remote attribute header does not match required off/len/owner)

There are two reasons for this, the first being that the
length of the buffer being read is determined from the
args->rmtblkcnt which doesn't take into account CRC headers. Hence
the mapped length ends up being too short and so we need to
calculate it directly from the value length.

The second is that the byte count of valid data within a buffer is
capped by the length of the data and so doesn't take into account
that the buffer might be longer due to headers. Hence we need to
calculate the data space in the buffer first before calculating the
actual byte count of data.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit 913e96bc292e1bb248854686c79d6545ef3ee720)

11 years agoxfs: remote attribute allocation may be contiguous
Dave Chinner [Tue, 21 May 2013 08:02:01 +0000 (18:02 +1000)]
xfs: remote attribute allocation may be contiguous

When CRCs are enabled, there may be multiple allocations made if the
headers cause a length overflow. This, however, does not mean that
the number of headers required increases, as the second and
subsequent extents may be contiguous with the previous extent. Hence
when we map the extents to write the attribute data, we may end up
with less extents than allocations made. Hence the assertion that we
consume the number of headers we calculated in the allocation loop
is incorrect and needs to be removed.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit 90253cf142469a40f89f989904abf0a1e500e1a6)

11 years agoxfs: fix dir3 freespace block corruption
Dave Chinner [Tue, 28 May 2013 08:37:17 +0000 (18:37 +1000)]
xfs: fix dir3 freespace block corruption

When the directory freespace index grows to a second block (2017
4k data blocks in the directory), the initialisation of the second
new block header goes wrong. The write verifier fires a corruption
error indicating that the block number in the header is zero. This
was being tripped by xfs/110.

The problem is that the initialisation of the new block is done just
fine in xfs_dir3_free_get_buf(), but the caller then users a dirv2
structure to zero on-disk header fields that xfs_dir3_free_get_buf()
has already zeroed. These lined up with the block number in the dir
v3 header format.

While looking at this, I noticed that the struct xfs_dir3_free_hdr()
had 4 bytes of padding in it that wasn't defined as padding or being
zeroed by the initialisation. Add a pad field declaration and fully
zero the on disk and in-core headers in xfs_dir3_free_get_buf() so
that this is never an issue in the future. Note that this doesn't
change the on-disk layout, just makes the 32 bits of padding in the
layout explicit.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit 5ae6e6a401957698f2bd8c9f4a86d86d02199fea)