firefly-linux-kernel-4.4.55.git
12 years agoMerge remote-tracking branch 'agust/next' into next
Benjamin Herrenschmidt [Thu, 27 Sep 2012 02:47:17 +0000 (12:47 +1000)]
Merge remote-tracking branch 'agust/next' into next

mpc5xxx updates from Anatolij

12 years agoMerge remote-tracking branch 'kumar/next' into next
Benjamin Herrenschmidt [Tue, 18 Sep 2012 06:04:33 +0000 (16:04 +1000)]
Merge remote-tracking branch 'kumar/next' into next

12 years agopowerpc/eeh: Fix crash on converting OF node to edev
Gavin Shan [Mon, 17 Sep 2012 04:34:28 +0000 (04:34 +0000)]
powerpc/eeh: Fix crash on converting OF node to edev

The kernel crash was reported by Alexy. He was testing some feature
with private kernel, in which Alexy added some code in pci_pm_reset()
to read the CSR after writting it. The bug could be reproduced on
Fiber Channel card (Fibre Channel: Emulex Corporation Saturn-X:
LightPulse Fibre Channel Host Adapter (rev 03)) by the following
commands.

# echo 1 > /sys/devices/pci0004:01/0004:01:00.0/reset
# rmmod lpfc
# modprobe lpfc

The history behind the test case is that those additional config
space reading operations in pci_pm_reset() would cause EEH error,
but we didn't detect EEH error until "modprobe lpfc". For the case,
all the PCI devices on PCI bus (0004:01) were removed and added after
PE reset. Then the EEH devices would be figured out again based on
the OF nodes. Unfortunately, there were some child OF nodes under
PCI device (0004:01:00.0), but they didn't have attached PCI_DN since
they're invisible from PCI domain. However, we were still trying to
convert OF node to EEH device without checking on the attached PCI_DN.
Eventually, it caused the kernel crash as follows:

Unable to handle kernel paging request for data at address 0x00000030
Faulting instruction address: 0xc00000000004d888
cpu 0x0: Vector: 300 (Data Access) at [c000000fc797b950]
    pc: c00000000004d888: .eeh_add_device_tree_early+0x78/0x140
    lr: c00000000004d880: .eeh_add_device_tree_early+0x70/0x140
    sp: c000000fc797bbd0
   msr: 8000000000009032
   dar: 30
 dsisr: 40000000
  current = 0xc000000fc78d9f70
  paca    = 0xc00000000edb0000   softe: 0        irq_happened: 0x00
    pid   = 2951, comm = eehd
enter ? for help
[c000000fc797bc50c00000000004d848 .eeh_add_device_tree_early+0x38/0x140
[c000000fc797bcd0c00000000004d848 .eeh_add_device_tree_early+0x38/0x140
[c000000fc797bd50c000000000051b54 .pcibios_add_pci_devices+0x34/0x190
[c000000fc797bde0c00000000004fb10 .eeh_reset_device+0x100/0x160
[c000000fc797be70c0000000000502dc .eeh_handle_event+0x19c/0x300
[c000000fc797bf00c000000000050570 .eeh_event_handler+0x130/0x1a0
[c000000fc797bf90c000000000020138 .kernel_thread+0x54/0x70

The patch changes of_node_to_eeh_dev() and just returns NULL if the
passed OF node doesn't have attached PCI_DN.

Cc: stable@vger.kernel.org
Reported-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Lock module while handling EEH event
Gavin Shan [Mon, 17 Sep 2012 04:34:27 +0000 (04:34 +0000)]
powerpc/eeh: Lock module while handling EEH event

The EEH core is talking with the PCI device driver to determine the
action (purely reset, or PCI device removal). During the period, the
driver might be unloaded and in turn causes kernel crash as follows:

EEH: Detected PCI bus error on PHB#4-PE#10000
EEH: This PCI device has failed 3 times in the last hour
lpfc 0004:01:00.0: 0:2710 PCI channel disable preparing for reset
Unable to handle kernel paging request for data at address 0x00000490
Faulting instruction address: 0xd00000000e682c90
cpu 0x1: Vector: 300 (Data Access) at [c000000fc75ffa20]
    pc: d00000000e682c90: .lpfc_io_error_detected+0x30/0x240 [lpfc]
    lr: d00000000e682c8c: .lpfc_io_error_detected+0x2c/0x240 [lpfc]
    sp: c000000fc75ffca0
   msr: 8000000000009032
   dar: 490
 dsisr: 40000000
  current = 0xc000000fc79b88b0
  paca    = 0xc00000000edb0380  softe: 0  irq_happened: 0x00
    pid   = 3386, comm = eehd
enter ? for help
[c000000fc75ffca0c000000fc75ffd30 (unreliable)
[c000000fc75ffd30c00000000004fd3c .eeh_report_error+0x7c/0xf0
[c000000fc75ffdc0c00000000004ee00 .eeh_pe_dev_traverse+0xa0/0x180
[c000000fc75ffe70c00000000004ffd8 .eeh_handle_event+0x68/0x300
[c000000fc75fff00c0000000000503a0 .eeh_event_handler+0x130/0x1a0
[c000000fc75fff90c000000000020138 .kernel_thread+0x54/0x70
1:mon>

The patch increases the reference of the corresponding driver modules
while EEH core does the negotiation with PCI device driver so that the
corresponding driver modules can't be unloaded during the period and
we're safe to refer the callbacks.

Cc: stable@vger.kernel.org
Reported-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/kprobe: Don't emulate store when kprobe stwu r1
Tiejun Chen [Sun, 16 Sep 2012 23:54:31 +0000 (23:54 +0000)]
powerpc/kprobe: Don't emulate store when kprobe stwu r1

We don't do the real store operation for kprobing 'stwu Rx,(y)R1'
since this may corrupt the exception frame, now we will do this
operation safely in exception return code after migrate current
exception frame below the kprobed function stack.

So we only update gpr[1] here and trigger a thread flag to mask
this.

Note we should make sure if we trigger kernel stack over flow.

Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/kprobe: Complete kprobe and migrate exception frame
Tiejun Chen [Sun, 16 Sep 2012 23:54:30 +0000 (23:54 +0000)]
powerpc/kprobe: Complete kprobe and migrate exception frame

We can't emulate stwu since that may corrupt current exception stack.
So we will have to do real store operation in the exception return code.

Firstly we'll allocate a trampoline exception frame below the kprobed
function stack and copy the current exception frame to the trampoline.
Then we can do this real store operation to implement 'stwu', and reroute
the trampoline frame to r1 to complete this exception migration.

Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/kprobe: Introduce a new thread flag
Tiejun Chen [Sun, 16 Sep 2012 23:54:29 +0000 (23:54 +0000)]
powerpc/kprobe: Introduce a new thread flag

We need to add a new thread flag, TIF_EMULATE_STACK_STORE,
for emulating stack store operation while exiting exception.

Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Remove unused __get_user64() and __put_user64()
Bharat Bhushan [Tue, 11 Sep 2012 00:19:45 +0000 (00:19 +0000)]
powerpc: Remove unused __get_user64() and __put_user64()

__get_user64()  and __put_user64() are not used.

Signed-off-by: Bharat Bhushan <bharat.bhushan@freescale.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Global mutex to protect PE tree
Gavin Shan [Tue, 11 Sep 2012 19:16:18 +0000 (19:16 +0000)]
powerpc/eeh: Global mutex to protect PE tree

We have missed lots of situations where the PE hierarchy tree need
protection through the EEH global mutex. The patch fixes that for
those public APIs implemented in eeh_pe.c. The only exception is
eeh_pe_restore_bars() because it calls eeh_pe_dev_traverse(), which
has been protected by the mutex.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Remove EEH PE for normal PCI hotplug
Gavin Shan [Tue, 11 Sep 2012 19:16:17 +0000 (19:16 +0000)]
powerpc/eeh: Remove EEH PE for normal PCI hotplug

Function eeh_rmv_from_parent_pe() could be called by the path of
either normal PCI hotplug, or EEH recovery. For the former case,
we need purge the corresponding PE on removal of the associated
PE bus.

The patch tries to cover that by passing more information to function
pcibios_remove_pci_devices() so that we know if the corresponding PE
needs to be purged or be marked as "invalid".

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_PE_INVALID type PE
Gavin Shan [Tue, 11 Sep 2012 19:16:16 +0000 (19:16 +0000)]
powerpc/eeh: Introduce EEH_PE_INVALID type PE

When EEH error happens on the PE whose PCI devices don't have
attached drivers. In function eeh_handle_event(), the default
value PCI_ERS_RESULT_NONE will be returned after iterating all
drivers of those PCI devices belonging to the PE. Actually, we
don't have installed drivers for the PCI devices. Under the
circumstance, we will remove the corresponding PCI bus of the PE,
including the associated EEH devices and PE instance. However,
we still need the information stored in the PE instance to do PE
reset after that. So it's unsafe to free the PE instance.

The patch introduces EEH_PE_INVALID type PE to address the issue.
When the PCI bus and the corresponding attached EEH devices are
removed, we will mark the PE as EEH_PE_INVALID. At later point,
the PE will be changed to EEH_PE_DEVICE or EEH_PE_BUS when the
corresponding EEH devices are attached again.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Add an xmon command to dump one or all pacas
Michael Ellerman [Thu, 13 Sep 2012 23:01:31 +0000 (23:01 +0000)]
powerpc: Add an xmon command to dump one or all pacas

This was originally motivated by a desire to see the mapping between
logical and hardware cpu numbers.

But it seemed that it made more sense to just add a command to dump
(most of) the paca.

With no arguments "dp" will dump the paca for the current cpu.

It also takes an argument, eg. "dp 3" which is the logical cpu number
in hex. This form does not check if the cpu is possible, but displays
the paca regardless, as well as the cpu's state in the possible, present
and online masks.

Thirdly, "dpa" will display the paca for all possible cpus. If there are
no possible cpus, like early in boot, it will tell you that.

Sample output, number in brackets is the offset into the struct:

2:mon> dp 3
paca for cpu 0x3 @ c00000000ff20a80:
 possible         = yes
 present          = yes
 online           = yes
 lock_token       = 0x8000             (0x8)
 paca_index       = 0x3                (0xa)
 kernel_toc       = 0xc00000000144f990 (0x10)
 kernelbase       = 0xc000000000000000 (0x18)
 kernel_msr       = 0xb000000000001032 (0x20)
 stab_real        = 0x0                (0x28)
 stab_addr        = 0x0                (0x30)
 emergency_sp     = 0xc00000003ffe4000 (0x38)
 data_offset      = 0xa40000           (0x40)
 hw_cpu_id        = 0x9                (0x50)
 cpu_start        = 0x1                (0x52)
 kexec_state      = 0x0                (0x53)
 __current        = 0xc00000007e568680 (0x218)
 kstack           = 0xc00000007e5a3e30 (0x220)
 stab_rr          = 0x1a               (0x228)
 saved_r1         = 0xc00000007e7cb450 (0x230)
 trap_save        = 0x0                (0x240)
 soft_enabled     = 0x0                (0x242)
 irq_happened     = 0x0                (0x243)
 io_sync          = 0x0                (0x244)
 irq_work_pending = 0x0                (0x245)
 nap_state_lost   = 0x0                (0x246)

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/powernv: Remove unused functions
Gavin Shan [Mon, 20 Aug 2012 03:49:21 +0000 (03:49 +0000)]
powerpc/powernv: Remove unused functions

We don't need them anymore. The patch removes those functions.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Richard Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/powernv: Using PCI core to do resource assignment
Gavin Shan [Mon, 20 Aug 2012 03:49:20 +0000 (03:49 +0000)]
powerpc/powernv: Using PCI core to do resource assignment

Currently, the PCI probe flags "PCI_PROBE_ONLY | PCI_REASSIGN_ALL_RSRC"
used on powernv platform. That means the platform has to do the PCI
resource assignment by itself.

The patch changes the PCI probe flag to "PCI_REASSIGN_ALL_RSRC" so
that the PCI core will do the resource assignment. Also, the I/O
and MMIO minimal alignment for P2P bridges have been configured
while doing fixup for the PHBs.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Richard Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/powernv: Fix overrunning segment tracing array
Gavin Shan [Mon, 20 Aug 2012 03:49:19 +0000 (03:49 +0000)]
powerpc/powernv: Fix overrunning segment tracing array

There're 2 arrays introduced to trace which PE has occupied the
corresponding resource (I/O or MMIO) segment. However, we didn't
allocate enough memory for them and that possiblly leads to PE
descriptor corruption.

The patch fixes that by allocating enough memory for those 2 arrays.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Richard Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/powernv: Skip check on PE if necessary
Gavin Shan [Mon, 20 Aug 2012 03:49:18 +0000 (03:49 +0000)]
powerpc/powernv: Skip check on PE if necessary

While the device driver or PCI core tries to enable PCI device, the
platform dependent callback "ppc_md.pcibios_enable_device_hook" will
be called to check if there has one associated PE for the PCI device.
If we don't have the associated PE for the PCI device, it's not allowed
to enable the PCI device. Unfortunately, there might have some cases
we have to enable the PCI device (e.g. P2P bridge), but the PEs have
not been created yet.

The patch handles the unfortunate cases. Each PHB (struct pnv_phb)
has one field "initialized" to trace if the PEs have been created
and configured or not. When the PEs are not available, we won't check
the associated PE for the PCI device to be enabled.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Richard Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/powernv: Initialize DMA for PEs
Gavin Shan [Mon, 20 Aug 2012 03:49:17 +0000 (03:49 +0000)]
powerpc/powernv: Initialize DMA for PEs

The patch introduces additional wrapper function to call the original
implementation so that the DMA can be configured for all existing PEs.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Richard Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/powernv: I/O and MMIO resource assignment for PEs
Gavin Shan [Mon, 20 Aug 2012 03:49:16 +0000 (03:49 +0000)]
powerpc/powernv: I/O and MMIO resource assignment for PEs

There're 2 types of PCI bus sensitive PEs: (A) The PE includes
single PCI bus. (B) The PE includes the PCI bus and all the subordinate
PCI buses, and the patch tries to assign I/O and MMIO resources
based on created PEs. Fortunately, we figured out unified scheme
to do resource assignment for all types of PCI bus based PEs according
to Ben's idea:

        - Resource assignment based on PE from top to bottom.
        - The soureces, either I/O or MMIO, of the PE are figured out
          from the assigned PCI bus.
        - The occupied resource by parent PE could possibilly be overrided
          by children PEs.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Richard Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/powernv: PE list based on creation order
Gavin Shan [Mon, 20 Aug 2012 03:49:15 +0000 (03:49 +0000)]
powerpc/powernv: PE list based on creation order

The resource (I/O and MMIO) will be assigned on basis of PE from
top to bottom so that we can implement the trick here: the resource
that has been assigned to parent PE could be taken by child PE if
necessary.

The current implementation already has PE list per PHB basis, but
the list doesn't meet our requirment: tracing PE based on their
cration time from top to bottom. So the patch does rename for the
DMA based PE list and introduces the list to trace the PEs sequentially
based on their creation time.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Richard Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/powernv: Create bus sensitive PEs
Gavin Shan [Mon, 20 Aug 2012 03:49:14 +0000 (03:49 +0000)]
powerpc/powernv: Create bus sensitive PEs

Basically, there're 2 types of PCI bus sensitive PEs: (A) The PE
includes single PCI bus. (B) The PE includes the PCI bus and all
the subordinate PCI buses. At present, we'd like to put PCI bus
originated by PCI-e link to form PE that contains single PCI bus,
and the PCIe-to-PCI bridge will form the 2nd type of PE. We don't
figure out to detect PLX bridge yet. Once we can detect PLX bridge
some day, we have to put PCI buses originated from the downstream
port of PLX bridge to the 2nd type of PE.

The patch changes the original implementation for a little bit
to support 2 types of PCI bus sensitive PEs described as above.
Also, the function used to retrieve the corresponding PE according
to the given PCI device has been changed based on that because each
PCI device should trace the directly associated PE.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Richard Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/trace: Fix interrupt tracepoints vs. RCU
Li Zhong [Mon, 10 Sep 2012 15:37:43 +0000 (15:37 +0000)]
powerpc/trace: Fix interrupt tracepoints vs. RCU

There are a few tracepoints in the interrupt code path, which is before
irq_enter(), or after irq_exit(), like
trace_irq_entry()/trace_irq_exit() in do_IRQ(),
trace_timer_interrupt_entry()/trace_timer_interrupt_exit() in
timer_interrupt().

If the interrupt is from idle(), and because tracepoint contains RCU
read-side critical section, we could see following suspicious RCU usage
reported:

[  145.127743] ===============================
[  145.127747] [ INFO: suspicious RCU usage. ]
[  145.127752] 3.6.0-rc3+ #1 Not tainted
[  145.127755] -------------------------------
[  145.127759] /root/.workdir/linux/arch/powerpc/include/asm/trace.h:33
suspicious rcu_dereference_check() usage!
[  145.127765]
[  145.127765] other info that might help us debug this:
[  145.127765]
[  145.127771]
[  145.127771] RCU used illegally from idle CPU!
[  145.127771] rcu_scheduler_active = 1, debug_locks = 0
[  145.127777] RCU used illegally from extended quiescent state!
[  145.127781] no locks held by swapper/0/0.
[  145.127785]
[  145.127785] stack backtrace:
[  145.127789] Call Trace:
[  145.127796] [c00000000108b530] [c000000000013c40] .show_stack
+0x70/0x1c0 (unreliable)
[  145.127806] [c00000000108b5e0]
[c0000000000f59d8] .lockdep_rcu_suspicious+0x118/0x150
[  145.127813] [c00000000108b680] [c00000000000fc58] .do_IRQ+0x498/0x500
[  145.127820] [c00000000108b750] [c000000000003950]
hardware_interrupt_common+0x150/0x180
[  145.127828] --- Exception: 501 at .plpar_hcall_norets+0x84/0xd4
[  145.127828]     LR = .check_and_cede_processor+0x38/0x70
[  145.127836] [c00000000108bab0] [c0000000000665dc] .shared_cede_loop
+0x5c/0x100
[  145.127844] [c00000000108bb70] [c000000000588ab0] .cpuidle_enter
+0x30/0x50
[  145.127850] [c00000000108bbe0]
[c000000000588b0c] .cpuidle_enter_state+0x3c/0xb0
[  145.127857] [c00000000108bc60] [c000000000589730] .cpuidle_idle_call
+0x150/0x6c0
[  145.127863] [c00000000108bd30] [c000000000058440] .pSeries_idle
+0x10/0x40
[  145.127870] [c00000000108bda0] [c00000000001683c] .cpu_idle
+0x18c/0x2d0
[  145.127876] [c00000000108be60] [c00000000000b434] .rest_init
+0x124/0x1b0
[  145.127884] [c00000000108bef0] [c0000000009d0d28] .start_kernel
+0x568/0x588
[  145.127890] [c00000000108bf90] [c000000000009660] .start_here_common
+0x20/0x40

This is because the RCU usage in interrupt context should be used in
area marked by rcu_irq_enter()/rcu_irq_exit(), called in
irq_enter()/irq_exit() respectively.

Move them into the irq_enter()/irq_exit() area to avoid the reporting.

Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Make some of the PGTABLE_RANGE dependency explicit
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:57 +0000 (02:52 +0000)]
powerpc/mm: Make some of the PGTABLE_RANGE dependency explicit

slice array size and slice mask size depend on PGTABLE_RANGE.

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Update VSID allocation documentation
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:56 +0000 (02:52 +0000)]
powerpc/mm: Update VSID allocation documentation

This update the proto-VSID and VSID scramble related information
to be more generic by using names instead of current values.

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Add 64TB support
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:55 +0000 (02:52 +0000)]
powerpc/mm: Add 64TB support

Increase max addressable range to 64TB. This is not tested on
real hardware yet.

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Use 32bit array for slb cache
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:54 +0000 (02:52 +0000)]
powerpc/mm: Use 32bit array for slb cache

With larger vsid we need to track more bits of ESID in slb cache
for slb invalidate.

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Use the required number of VSID bits in slbmte
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:53 +0000 (02:52 +0000)]
powerpc/mm: Use the required number of VSID bits in slbmte

ASM_VSID_SCRAMBLE can leave non-zero bits in the high 28 bits of the result
for 256MB segment (40 bits for 1T segment). Properly mask them before using
the values in slbmte

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Increase the slice range to 64TB
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:52 +0000 (02:52 +0000)]
powerpc/mm: Increase the slice range to 64TB

This patch makes the high psizes mask as an unsigned char array
so that we can have more than 16TB. Currently we support upto
64TB

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Make KERN_VIRT_SIZE not dependend on PGTABLE_RANGE
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:51 +0000 (02:52 +0000)]
powerpc/mm: Make KERN_VIRT_SIZE not dependend on PGTABLE_RANGE

As we keep increasing PGTABLE_RANGE we need not increase the virual
map area for kernel.

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Convert virtual address to vpn
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:50 +0000 (02:52 +0000)]
powerpc/mm: Convert virtual address to vpn

This patch convert different functions to take virtual page number
instead of virtual address. Virtual page number is virtual address
shifted right by VPN_SHIFT (12) bits. This enable us to have an
address range of upto 76 bits.

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Simplify hpte_decode
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:49 +0000 (02:52 +0000)]
powerpc/mm: Simplify hpte_decode

This patch simplify hpte_decode for easy switching of virtual address to
virtual page number in the later patch

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Use hpt_va to compute virtual address
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:48 +0000 (02:52 +0000)]
powerpc/mm: Use hpt_va to compute virtual address

Don't open code the same

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Replace open coded CONTEXT_BITS value
Aneesh Kumar K.V [Mon, 10 Sep 2012 02:52:47 +0000 (02:52 +0000)]
powerpc/mm: Replace open coded CONTEXT_BITS value

To clarify the meaning for future readers, replace the open coded
19 with CONTEXT_BITS

Reviewed-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mm: Fix typo in PTRS_PER_PUD
Scott Wood [Wed, 12 Sep 2012 13:00:09 +0000 (13:00 +0000)]
powerpc/mm: Fix typo in PTRS_PER_PUD

PTRS_PER_PUD should be based on PUD_INDEX_SIZE, not PMD_INDEX_SIZE.  We
got away with it because PUD and PMD had the same index size, but this is
no longer true with Aneesh's patchset to support a 46-bit user effective
address space.

Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Add denormalisation exception handling for POWER6/7
Michael Neuling [Mon, 10 Sep 2012 00:35:26 +0000 (00:35 +0000)]
powerpc: Add denormalisation exception handling for POWER6/7

On POWER6 and POWER7 if the input operand to an instruction is a
denormalised single precision binary floating point value we can take
a denormalisation exception where it's expected that the hypervisor
(HV=1) will fix up the inputs before the instruction is run.

This adds code to handle this denormalisation exception for POWER6 and
POWER7.

It also add a CONFIG_PPC_DENORMALISATION option and sets it in
pseries/ppc64_defconfig.

This is useful on bare metal systems only.  Based on patch from Milton
Miller.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agoMerge remote-tracking branch 'pci/pci/gavin-window-alignment' into next
Benjamin Herrenschmidt [Mon, 17 Sep 2012 06:07:43 +0000 (16:07 +1000)]
Merge remote-tracking branch 'pci/pci/gavin-window-alignment' into next

Merge Gavin patches from the PCI tree as subsequent powerpc
patches are going to depend on them

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/mpc52xx_lpbfifo: optionally defer fifo transfer start
Anatolij Gustschin [Wed, 29 Aug 2012 21:33:28 +0000 (21:33 +0000)]
powerpc/mpc52xx_lpbfifo: optionally defer fifo transfer start

Currently fifo transfer is started when submitting a transfer
request. Add posibility to defer the fifo transfer and start it
later by calling additional function. This change is backward
compatible, the behaviour of mpc52xx_lpbfifo_submit() is the same
for previous driver users, so there is no need to adapt them.

Signed-off-by: Anatolij Gustschin <agust@denx.de>
12 years agodt/misc: add bindings documentation for ifm camera sensor interface
Anatolij Gustschin [Fri, 24 Aug 2012 05:39:39 +0000 (05:39 +0000)]
dt/misc: add bindings documentation for ifm camera sensor interface

IFM O2D cameras use special sensor bus interface glue-logic to
connect camera sensors to mpc5200 LocalPlus bus. Add device tree
bindings documentation for it.

Signed-off-by: Anatolij Gustschin <agust@denx.de>
12 years agopowerpc/mpc5200: add dts files for ifm camera machines
Anatolij Gustschin [Wed, 29 Aug 2012 21:31:08 +0000 (21:31 +0000)]
powerpc/mpc5200: add dts files for ifm camera machines

Add common o2d dtsi file to reuse it for other configurations.
Add machine compatible string to mpc5200 simple platform file.
Add dts files for O2D, O2I, O2MNT, O2DNT2, O2D300 and O3DNT boards.

Signed-off-by: Anatolij Gustschin <agust@denx.de>
12 years agoMerge commit 'v3.6-rc5' into pci/gavin-window-alignment
Bjorn Helgaas [Thu, 13 Sep 2012 21:54:57 +0000 (15:54 -0600)]
Merge commit 'v3.6-rc5' into pci/gavin-window-alignment

* commit 'v3.6-rc5': (1098 commits)
  Linux 3.6-rc5
  HID: tpkbd: work even if the new Lenovo Keyboard driver is not configured
  Remove user-triggerable BUG from mpol_to_str
  xen/pciback: Fix proper FLR steps.
  uml: fix compile error in deliver_alarm()
  dj: memory scribble in logi_dj
  Fix order of arguments to compat_put_time[spec|val]
  xen: Use correct masking in xen_swiotlb_alloc_coherent.
  xen: fix logical error in tlb flushing
  xen/p2m: Fix one-off error in checking the P2M tree directory.
  powerpc: Don't use __put_user() in patch_instruction
  powerpc: Make sure IPI handlers see data written by IPI senders
  powerpc: Restore correct DSCR in context switch
  powerpc: Fix DSCR inheritance in copy_thread()
  powerpc: Keep thread.dscr and thread.dscr_inherit in sync
  powerpc: Update DSCR on all CPUs when writing sysfs dscr_default
  powerpc/powernv: Always go into nap mode when CPU is offline
  powerpc: Give hypervisor decrementer interrupts their own handler
  powerpc/vphn: Fix arch_update_cpu_topology() return value
  ARM: gemini: fix the gemini build
  ...

Conflicts:
drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
drivers/rapidio/devices/tsi721.c

12 years agopowerpc/fsl-pci: Unify pci/pcie initialization code
Jia Hongtao [Tue, 28 Aug 2012 07:44:08 +0000 (15:44 +0800)]
powerpc/fsl-pci: Unify pci/pcie initialization code

We unified the Freescale pci/pcie initialization by changing the fsl_pci
to a platform driver. In previous PCI code architecture the initialization
routine is called at board_setup_arch stage. Now the initialization is done
in probe function which is architectural better. Also It's convenient for
adding PM support for PCI controller in later patch.

Now we registered pci controllers as platform devices. So we combine two
initialization code as one platform driver.

Signed-off-by: Jia Hongtao <B38951@freescale.com>
Signed-off-by: Li Yang <leoli@freescale.com>
Signed-off-by: Chunhe Lan <Chunhe.Lan@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/pci: Use PCIe IP block revision register instead of compatible
Roy Zang [Mon, 3 Sep 2012 09:22:10 +0000 (17:22 +0800)]
powerpc/pci: Use PCIe IP block revision register instead of compatible

Freescale PCIe IP block revision bigger than rev2.2 will also need
redefine the sequence of inbound windows. So change to use IP block
revision instead of compatible for the judgment.

Signed-off-by: Roy Zang <tie-fei.zang@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/pci: Add IP revision register define for Freescale PCIe controller
Roy Zang [Mon, 3 Sep 2012 09:22:09 +0000 (17:22 +0800)]
powerpc/pci: Add IP revision register define for Freescale PCIe controller

Signed-off-by: Roy Zang <tie-fei.zang@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/p5040: fix dtb build warning of p5040ds.dtb
Shaohui Xie [Wed, 12 Sep 2012 10:36:03 +0000 (18:36 +0800)]
powerpc/p5040: fix dtb build warning of p5040ds.dtb

Device node adt7461 was wrongly added in p5040ds.dts, it should be added
into i2c instead of localbus, when build p5040ds.dtb, a warning will dump:

Warning (reg_format): "reg" property in
/localbus@ffe124000/nand@2,0/adt7461@4c has invalid length (4 bytes)
(#address-cells == 1, #size-cells == 1)

This was introduced by:

commit ea6b1ba692bcb5f6e39f409a78cf8b04fdf23baa
Author: Jia Hongtao <B38951@freescale.com>
Date:   Tue Aug 28 10:00:55 2012 +0800

    powerpc: add adt7461 thermal monitor support to applicable boards

    Add thermal monitor support to following boards:
    P1022DS, MPC8536DS, P2041RDB, P3041DS, P4080DS, P5020DS, P5040DS

Signed-off-by: Shaohui Xie <Shaohui.Xie@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/8544ds: add partition table for norflash
Wang Dongsheng [Wed, 29 Aug 2012 03:00:49 +0000 (11:00 +0800)]
powerpc/8544ds: add partition table for norflash

create partition table for norflash.

Signed-off-by: Wang Dongsheng <Dongsheng.Wang@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc: add adt7461 thermal monitor support to applicable boards
Jia Hongtao [Tue, 28 Aug 2012 02:00:55 +0000 (10:00 +0800)]
powerpc: add adt7461 thermal monitor support to applicable boards

Add thermal monitor support to following boards:
P1022DS, MPC8536DS, P2041RDB, P3041DS, P4080DS, P5020DS, P5040DS

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/mpc85xx: Add new ext fields to Integrated FLash Controller
Prabhakar Kushwaha [Thu, 16 Aug 2012 03:58:22 +0000 (09:28 +0530)]
powerpc/mpc85xx: Add new ext fields to Integrated FLash Controller

Freescale's Integrated Flash controller(IFC) v1.1.0 supports 40 bit
address bus width.
In case more than 32 bit address is used, the EXT registers should be set.

Add support of ext registers.

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: York Sun <yorksun@freescale.com>
Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc: Update Integrated Flash controller device tree bindings
Prabhakar Kushwaha [Wed, 15 Aug 2012 15:13:52 +0000 (20:43 +0530)]
powerpc: Update Integrated Flash controller device tree bindings

Freescale's Integrated Flash controller (IFC) may have one or two
interrupts. In case of single interrupt line, it will cover all IFC
interrupts.

Update this information in IFC device tree bindings

Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/e5500: Add Power ISA properties to comply with ePAPR 1.1
Olivia Yin [Thu, 9 Aug 2012 07:42:36 +0000 (15:42 +0800)]
powerpc/e5500: Add Power ISA properties to comply with ePAPR 1.1

power-isa-version and power-isa-* are cpu node general properties defined
in ePAPR.

If the power-isa-version property exists, then for each category from the
Categories section of Book I of the Power ISA version indicated, the
existence of a property named power-isa-[CAT], where [CAT] is the
abbreviated category name with all uppercase letters converted to
lowercase, indicates that the category is supported by the implementation.

This patch update all the e5500 platforms.

Signed-off-by: Liu Yu <yu.liu@freescale.com>
Signed-off-by: Olivia Yin <hong-hua.yin@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/e500mc: Add Power ISA properties to comply with ePAPR 1.1
Olivia Yin [Thu, 9 Aug 2012 07:42:35 +0000 (15:42 +0800)]
powerpc/e500mc: Add Power ISA properties to comply with ePAPR 1.1

power-isa-version and power-isa-* are cpu node general properties defined
in ePAPR.

If the power-isa-version property exists, then for each category from the
Categories section of Book I of the Power ISA version indicated, the
existence of a property named power-isa-[CAT], where [CAT] is the
abbreviated category name with all uppercase letters converted to
lowercase, indicates that the category is supported by the implementation.

The patch update all the e500mc platforms.

Signed-off-by: Liu Yu <yu.liu@freescale.com>
Signed-off-by: Olivia Yin <hong-hua.yin@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/e500v2: Add Power ISA properties to comply with ePAPR 1.1
Olivia Yin [Thu, 9 Aug 2012 07:42:34 +0000 (15:42 +0800)]
powerpc/e500v2: Add Power ISA properties to comply with ePAPR 1.1

power-isa-version and power-isa-* are cpu node general properties defined
in ePAPR.

If the power-isa-version property exists, then for each category from the
Categories section of Book I of the Power ISA version indicated, the
existence of a property named power-isa-[CAT], where [CAT] is the
abbreviated category name with all uppercase letters converted to
lowercase, indicates that the category is supported by the implementation.

The patch update all e500v2 platforms.

Signed-off-by: Liu Yu <yu.liu@freescale.com>
Signed-off-by: Olivia Yin <hong-hua.yin@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/mpic: FSL MPIC error interrupt support.
Varun Sethi [Wed, 8 Aug 2012 04:06:09 +0000 (09:36 +0530)]
powerpc/mpic: FSL MPIC error interrupt support.

All SOC device error interrupts are muxed and delivered to the core
as a single MPIC error interrupt. Currently all the device drivers
requiring access to device errors have to register for the MPIC error
interrupt as a shared interrupt.

With this patch we add interrupt demuxing capability in the mpic driver,
allowing device drivers to register for their individual error interrupts.
This is achieved by handling error interrupts in a cascaded fashion.

MPIC error interrupt is handled by the "error_int_handler", which
subsequently demuxes it using the EISR and delivers it to the respective
drivers.

The error interrupt capability is dependent on the MPIC EIMR register,
which was introduced in FSL MPIC version 4.1 (P4080 rev2). So, error
interrupt demuxing capability is dependent on the MPIC version and can
be used for versions >= 4.1.

Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Bogdan Hamciuc <bogdan.hamciuc@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/swiotlb: Enable at early stage and disable if not necessary
Jia Hongtao [Fri, 3 Aug 2012 10:14:10 +0000 (18:14 +0800)]
powerpc/swiotlb: Enable at early stage and disable if not necessary

Remove the dependency on PCI initialization for SWIOTLB initialization.
So that PCI can be initialized at proper time.

SWIOTLB is partly determined by PCI inbound/outbound map which is assigned
in PCI initialization. But swiotlb_init() should be done at the stage of
mem_init() which is much earlier than PCI initialization. So we reserve the
memory for SWIOTLB first and free it if not necessary.

All boards are converted to fit this change.

Signed-off-by: Jia Hongtao <B38951@freescale.com>
Signed-off-by: Li Yang <leoli@freescale.com>
Acked-by: Tony Breeds <tony@bakeyournoodle.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/booke: Add CPU_FTR_EMB_HV check for e5500.
Varun Sethi [Mon, 9 Jul 2012 13:04:01 +0000 (18:34 +0530)]
powerpc/booke: Add CPU_FTR_EMB_HV check for e5500.

Added CPU_FTR_EMB_HV feature check for e5500.

Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Mihai Caraman <mihai.caraman@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/booke: Separate out restore_e5500/setup_e5500 routines.
Varun Sethi [Mon, 9 Jul 2012 13:01:51 +0000 (18:31 +0530)]
powerpc/booke: Separate out restore_e5500/setup_e5500 routines.

For the 64 bit case separate out e5500 cpu_setup and cpu_restore functions.
The cpu_setup function (for the primary core) is passed the cpu_spec
pointer, which is not there in case of the cpu_restore function. Also, in
our case we will have to manipulate the CPU_FTR_EMB_HV flag on the primary
core.

Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/booke: Merge the 32 bit e5500/e500mc cpu setup code.
Varun Sethi [Mon, 9 Jul 2012 12:58:21 +0000 (18:28 +0530)]
powerpc/booke: Merge the 32 bit e5500/e500mc cpu setup code.

Merge the 32 bit cpu setup code for e500mc/e5500 and define the
"cpu_restore" routine (for e5500/e6500) only for the 64 bit case. The
cpu_restore routine is used in the 64 bit case for setting up the secondary
cores.

Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowepc/booke: Separate out E.HV check and ivor setup code.
Varun Sethi [Mon, 9 Jul 2012 12:55:31 +0000 (18:25 +0530)]
powepc/booke: Separate out E.HV check and ivor setup code.

Move the E.HV check and CPU_FTR_EMB_HV flag manipulation to the cpu setup
code.  Create a separate routine for E.HV ivors setup.

Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: add HOTPLUG_CPU support
Zhao Chenhui [Fri, 20 Jul 2012 12:42:36 +0000 (20:42 +0800)]
powerpc/85xx: add HOTPLUG_CPU support

Add support to disable and re-enable individual cores at runtime on
MPC85xx/QorIQ SMP machines. Currently support e500v1/e500v2 core.

MPC85xx machines use ePAPR spin-table in boot page for CPU kick-off.  This
patch uses the boot page from bootloader to boot core at runtime.  It
supports 32-bit and 36-bit physical address.

Signed-off-by: Li Yang <leoli@freescale.com>
Signed-off-by: Jin Qing <b24347@freescale.com>
Signed-off-by: Zhao Chenhui <chenhui.zhao@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: implement hardware timebase sync
Zhao Chenhui [Fri, 20 Jul 2012 12:42:35 +0000 (20:42 +0800)]
powerpc/85xx: implement hardware timebase sync

Do hardware timebase sync. Firstly, stop all timebases, and transfer the
timebase value of the boot core to the other core. Finally, start all
timebases.

Only apply to dual-core chips, such as MPC8572, P2020, etc.

Signed-off-by: Zhao Chenhui <chenhui.zhao@freescale.com>
Signed-off-by: Li Yang <leoli@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/smp: add generic_set_cpu_up() to set cpu_state as CPU_UP_PREPARE
Zhao Chenhui [Fri, 20 Jul 2012 12:42:34 +0000 (20:42 +0800)]
powerpc/smp: add generic_set_cpu_up() to set cpu_state as CPU_UP_PREPARE

In the case of cpu hotplug, the cpu_state should be set to CPU_UP_PREPARE
when kicking cpu.  Otherwise, the cpu_state is always CPU_DEAD after
calling generic_set_cpu_dead(), which makes the delay in generic_cpu_die()
not happen.

Signed-off-by: Zhao Chenhui <chenhui.zhao@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Replace epapr spin table macros/defines with a struct
Zhao Chenhui [Fri, 20 Jul 2012 12:42:33 +0000 (20:42 +0800)]
powerpc/85xx: Replace epapr spin table macros/defines with a struct

Signed-off-by: Zhao Chenhui <chenhui.zhao@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/mpic: finish supporting timer group B on Freescale chips
Varun Sethi [Mon, 9 Jul 2012 08:45:42 +0000 (14:15 +0530)]
powerpc/mpic: finish supporting timer group B on Freescale chips

Previously, these interrupts would be mapped, but the offset calculation
was broken, and only the first group was initialized.

Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: introduce support for the Freescale / iVeia P1022RDK
Timur Tabi [Mon, 23 Jul 2012 23:12:29 +0000 (18:12 -0500)]
powerpc/85xx: introduce support for the Freescale / iVeia P1022RDK

The Freescale / iVeia P1022RDK reference board is a small-factor board
with a Freescale P1022 SOC.  It includes:

1) 512 MB 64-bit DDR3-800 (max) memory
2) 8MB SPI serial flash memory for boot loader
3) Bootable 4-bit SD/MMC port
4) Two 10/100/1000 Ethernet connectors
5) One SATA port
6) Two USB ports
7) One PCIe x4 slot
8) DVI video connector
9) Audio input and output jacks, powered by a Wolfson WM8960 codec.

Unlike the P1022DS, the P1022RDK does not have any localbus devices,
presumably because of the localbus / DIU multiplexing restriction of
the P1022 SOC.

Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: Add support for P5040DS board
Timur Tabi [Thu, 26 Jul 2012 15:08:54 +0000 (10:08 -0500)]
powerpc/85xx: Add support for P5040DS board

Add support for the Freescale P5040DS Reference Board ("Superhydra"), which
is similar to the P5020DS.  Features of the P5040 are listed below, but
not all of these features (e.g. DPAA networking) are currently supported.

Four P5040 single-threaded e5500 cores built
    Up to 2.4 GHz with 64-bit ISA support
    Three levels of instruction: user, supervisor, hypervisor
CoreNet platform cache (CPC)
    2.0 MB configures as dual 1 MB blocks hierarchical interconnect fabric
Two 64-bit DDR3/3L SDRAM memory controllers with ECC and interleaving
 support Up to 1600MT/s
    Memory pre-fetch engine
DPAA incorporating acceleration for the following functions
    Packet parsing, classification, and distribution (FMAN)
    Queue management for scheduling, packet sequencing and
congestion management (QMAN)
    Hardware buffer management for buffer allocation and
de-allocation (BMAN)
    Cryptography acceleration (SEC 5.0) at up to 40 Gbps SerDes
    20 lanes at up to 5 Gbps
    Supports SGMII, XAUI, PCIe rev1.1/2.0, SATA Ethernet interfaces
    Two 10 Gbps Ethernet MACs
    Ten 1 Gbps Ethernet MACs
High-speed peripheral interfaces
    Two PCI Express 2.0/3.0 controllers
Additional peripheral interfaces
    Two serial ATA (SATA 2.0) controllers
    Two high-speed USB 2.0 controllers with integrated PHY
    Enhanced secure digital host controller (SD/MMC/eMMC)
    Enhanced serial peripheral interface (eSPI)
    Two I2C controllers
    Four UARTs
    Integrated flash controller supporting NAND and NOR flash
DMA
    Dual four channel
Support for hardware virtualization and partitioning enforcement
    Extra privileged level for hypervisor support
QorIQ Trust Architecture 1.1
    Secure boot, secure debug, tamper detection, volatile key storage

Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: add Freescale P5040 SOC and SEC v5.2 device trees
Kim Phillips [Thu, 26 Jul 2012 15:08:53 +0000 (10:08 -0500)]
powerpc/85xx: add Freescale P5040 SOC and SEC v5.2 device trees

Add device tree (dtsi) files for the Freescale P5040 SOC.  Since this
SOC introduces SEC v5.2, add the dtsi file for that also.

Signed-off-by: Kim Phillips <kim.phillips@freescale.com>
Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/fsl-pci: add fsl,qoriq-pcie-v2.4 compatible string
Timur Tabi [Thu, 26 Jul 2012 15:08:52 +0000 (10:08 -0500)]
powerpc/fsl-pci: add fsl,qoriq-pcie-v2.4 compatible string

The PCI controller on the Freescale P5040 is v2.4.

Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: remove P1020RDB and P2020RDB CAMP device trees
Timur Tabi [Fri, 13 Jul 2012 22:40:43 +0000 (17:40 -0500)]
powerpc/85xx: remove P1020RDB and P2020RDB CAMP device trees

We only need two examples of CAMP device trees in the upstream kernel.

Co-operative Asymmetric Multi-Processing (CAMP) is a technique where two
or more operating systems (typically multiple copies of the same Linux
kernel) are loaded into memory, and each kernel is given a subset of the
available cores to execute on.  For example, on a four-core system, one
kernel runs on cores 0 and 1, and the other runs on cores 2 and 3.

The devices are also partitioned among the operating systems, and this is
done with customized device trees.  Each kernel gets its own device tree
that has only the devices that it should know about.

Unfortunately, this approach is very hackish.  The kernels are trusted to
only access devices in their respective device trees, and the partitioning
only works for devices that can be handled.  Crafting the device trees is a
tricky process, and getting U-Boot to load and start all kernels is
cumbersome.

But most importantly, each CAMP setup is very application-specific, since
the actual partitioning of resources is done in the DTS by the system
designer.  Therefore, it doesn't make a lot of sense to have a lot of CAMP
device trees, since we only expect them to be used as examples.

Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/85xx: L2sram - Add compatible string to the device id list
Tang Yuantian [Fri, 13 Jul 2012 02:27:34 +0000 (10:27 +0800)]
powerpc/85xx: L2sram - Add compatible string to the device id list

The following platforms are supported:
mpc8544, mpc8572, mpc8536, p1021, p1025, p1024, p1010.

Signed-off-by: Tang Yuantian <Yuantian.Tang@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
12 years agopowerpc/powernv: I/O and memory alignment for P2P bridges
Gavin Shan [Tue, 11 Sep 2012 22:59:47 +0000 (16:59 -0600)]
powerpc/powernv: I/O and memory alignment for P2P bridges

The patch implements ppc_md.pcibios_window_alignment for powernv
platform so that the resource reassignment in PCI core will be
done according to the I/O and memory alignment returned from
powernv platform. The alignments returned from powernv platform
is closely depending on the scheme for PE segmenting. Besides,
the patch isn't useful for now, but the subsequent patches will
be working based on it.

[bhelgaas: use pci_pcie_type() since pci_dev.pcie_type was removed]
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
12 years agopowerpc/PCI: Override pcibios_window_alignment()
Gavin Shan [Tue, 11 Sep 2012 22:59:46 +0000 (16:59 -0600)]
powerpc/PCI: Override pcibios_window_alignment()

This patch implements pcibios_window_alignment() so powerpc platforms can
force P2P bridge windows to be at larger alignments than the PCI spec
requires.

[bhelgaas: changelog]
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
12 years agoPCI: Refactor pbus_size_mem()
Gavin Shan [Tue, 11 Sep 2012 22:59:46 +0000 (16:59 -0600)]
PCI: Refactor pbus_size_mem()

The original idea comes from Ram Pai.  This patch puts the chunk of
code for calculating the minimal alignment of memory window into a
separate inline function.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
12 years agoPCI: Align P2P windows using pcibios_window_alignment()
Gavin Shan [Tue, 11 Sep 2012 22:59:46 +0000 (16:59 -0600)]
PCI: Align P2P windows using pcibios_window_alignment()

This patch changes pbus_size_io() and pbus_size_mem() to do window (I/O,
memory and prefetchable memory) reassignment based on the minimal
alignments for the P2P bridge, which was retrieved by window_alignment().

[bhelgaas: changelog]
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
12 years agoPCI: Add weak pcibios_window_alignment() interface
Gavin Shan [Tue, 11 Sep 2012 22:59:45 +0000 (16:59 -0600)]
PCI: Add weak pcibios_window_alignment() interface

This patch implements a weak function to return the default I/O or memory
window alignment for a P2P bridge.  By default, I/O windows are aligned to
4KiB or 1KiB and memory windows are aligned to 4MiB.  Some platforms, e.g.,
powernv, have special alignment requirements and can override
pcibios_window_alignment().

[bhelgaas: changelog]
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
12 years agopowerpc/mm: Match variable types to API
Joe MacDonald [Tue, 21 Aug 2012 08:22:28 +0000 (08:22 +0000)]
powerpc/mm: Match variable types to API

sys_subpage_prot() takes an unsigned long for 'addr' then does some stuff
with it and the result is stored in a signed int, i, which is eventually
used as the size parameter in a copy_from_user call.  Update 'i' to be an
unsigned long as well and since 'nw' is used in a size_t context which,
depending on whether this is 32- or 64-bit may be unsigned int or unsigned
long, switch that to a size_t and always be right.

Finally, since we're in the neighbourhood, make the same changes to
subpage_prot_clear().

Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Joe MacDonald <joe.macdonald@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/iommu: Add ppc_md.tce_get() callback for use by VFIO
Alexey Kardashevskiy [Tue, 4 Sep 2012 15:19:35 +0000 (15:19 +0000)]
powerpc/iommu: Add ppc_md.tce_get() callback for use by VFIO

The upcoming VFIO support requires a way to know which
entry in the TCE map is not empty in order to do cleanup
at QEMU exit/crash. This patch adds such functionality
to POWERNV platform code.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: cleanup old DABRX #defines
Michael Neuling [Thu, 6 Sep 2012 21:24:58 +0000 (21:24 +0000)]
powerpc: cleanup old DABRX #defines

These are no longer used so get rid of them

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Dynamically calculate the dabrx based on kernel/user/hypervisor
Michael Neuling [Thu, 6 Sep 2012 21:24:57 +0000 (21:24 +0000)]
powerpc: Dynamically calculate the dabrx based on kernel/user/hypervisor

Currently we mark the DABRX to interrupt on all matches
(hypervisor/kernel/user and then filter in software.  We can be a lot
smarter now that we can set the DABRX dynamically.

This sets the DABRX based on the flags passed by the user.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Rework set_dabr so it can take a DABRX value as well
Michael Neuling [Thu, 6 Sep 2012 21:24:56 +0000 (21:24 +0000)]
powerpc: Rework set_dabr so it can take a DABRX value as well

Rework set_dabr to take a DABRX value as well.

Both the pseries and PS3 hypervisors do some checks on the DABRX
values that are passed in the hcall.  This patch stops bogus values
from being passed to hypervisor.  Also, in the case where we are
clearing the breakpoint, where DABR and DABRX are zero, we modify the
DABRX value to make it valid so that the hcall won't fail.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Cleanup on EEH PCI address cache
Gavin Shan [Fri, 7 Sep 2012 22:44:23 +0000 (22:44 +0000)]
powerpc/eeh: Cleanup on EEH PCI address cache

The patch does cleanup on EEH PCI address cache based on the fact
EEH core is the only user of the component.

        * Cleanup on function names so that they all have prefix
          "eeh" and looks more short.
        * Function printk() has been replaced with pr_debug() or
          pr_warning() accordingly.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Trace eeh device from I/O cache
Gavin Shan [Fri, 7 Sep 2012 22:44:22 +0000 (22:44 +0000)]
powerpc/eeh: Trace eeh device from I/O cache

The idea comes from Benjamin Herrenschmidt. The eeh cache helps
fetching the pci device according to the given I/O address. Since
the eeh cache is serving for eeh, it's reasonable for eeh cache
to trace eeh device except pci device.

The patch make eeh cache to trace eeh device. Also, the major
eeh entry function eeh_dn_check_failure has been renamed to
eeh_dev_check_failure since it will take eeh device as input
parameter.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Probe mode support
Gavin Shan [Fri, 7 Sep 2012 22:44:21 +0000 (22:44 +0000)]
powerpc/eeh: Probe mode support

While EEH module is installed, PCI devices is checked one by one
to see if it supports eeh. On different platforms, the PCI devices
are referred through different ways when the EEH module is loaded.
For example, on pSeries platform, that is done by OF node. However,
we would do that by real PCI devices (struct pci_dev) on PowerNV
platform in future. So we needs some mechanism to differentiate
those cases by classifying them to probe modes, either from OF
nodes or real PCI devices.

The patch implements the support to eeh probe mode. Also, the
EEH on pSeries has set it into EEH_PROBE_MODE_DEVTREE. That means
the probe will be done based on OF nodes on pSeries platform.

In addition, On pSeries platform, it's done by OF nodes. The patch
moves the the probe function from EEH core to platform dependent
backend and some cleanup applied.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Move stats to PE
Gavin Shan [Fri, 7 Sep 2012 22:44:20 +0000 (22:44 +0000)]
powerpc/eeh: Move stats to PE

The patch removes the eeh related statistics for eeh device since
they have been maintained by the corresponding eeh PE. Also, the
flags used to trace the state of eeh device and PE have been reworked
for a little bit.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Handle EEH error based on PE
Gavin Shan [Fri, 7 Sep 2012 22:44:19 +0000 (22:44 +0000)]
powerpc/eeh: Handle EEH error based on PE

The patch reworks the current implementation so that the eeh errors
will be handled basing on PE instead of eeh device.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Make EEH handler PE sensitive
Gavin Shan [Fri, 7 Sep 2012 22:44:18 +0000 (22:44 +0000)]
powerpc/eeh: Make EEH handler PE sensitive

Once eeh error is found, eeh event will be created and put it into
the global linked list. At the mean while, kernel thread will be
started to process it. The handler for the kernel thread originally
was eeh device sensitive.

The patch reworks the handler of the kernel thread so that it's PE
sensitive.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Do reset based on PE
Gavin Shan [Fri, 7 Sep 2012 22:44:17 +0000 (22:44 +0000)]
powerpc/eeh: Do reset based on PE

The patch implements reset based on PE instead of eeh device. Also,
The functions used to retrieve the reset type, either hot or fundamental
reset, have been reworked for a little bit. More specificly, it's
implemented based the the eeh device traverse function.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: I/O enable and log retrival based on PE
Gavin Shan [Fri, 7 Sep 2012 22:44:16 +0000 (22:44 +0000)]
powerpc/eeh: I/O enable and log retrival based on PE

The patch refactors the original implementation in order to enable
I/O and retrieve EEH log based on PE.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Device bars restore based on PE
Gavin Shan [Fri, 7 Sep 2012 22:44:15 +0000 (22:44 +0000)]
powerpc/eeh: Device bars restore based on PE

The patch introduces the function to traverse the devices of the
specified PE and its child PEs. Also, the restore on device bars
is implemented based on the traverse function.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Make EEH operations based on PE
Gavin Shan [Fri, 7 Sep 2012 22:44:14 +0000 (22:44 +0000)]
powerpc/eeh: Make EEH operations based on PE

Originally, all the EEH operations were implemented based on OF node.

Actually, it explicitly breaks the rules that the operation target
is PE instead of device. Therefore, the patch makes all the operations
based on PE instead of device.

Unfortunately, the backend for config space has to be kept as original
because it doesn't depend on PE.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Trace error based on PE from beginning
Gavin Shan [Fri, 7 Sep 2012 22:44:13 +0000 (22:44 +0000)]
powerpc/eeh: Trace error based on PE from beginning

There're 2 conditions to trigger EEH error detection: invalid value
returned from reading I/O or config space. On each case, the function
eeh_dn_check_failure will be called to initialize EEH event and put
it into the poll for further processing.

The patch changes the function for a little bit so that the EEH error
will be traced based on PE instead of EEH device any more. Also, the
function eeh_find_device_pe() has been removed since the eeh device
is tracing the PE by struct eeh_dev::pe.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Trace EEH state based on PE
Gavin Shan [Fri, 7 Sep 2012 22:44:12 +0000 (22:44 +0000)]
powerpc/eeh: Trace EEH state based on PE

Since we've introduced dedicated struct to trace individual PEs,
it's reasonable to trace its state through the dedicated struct
instead of using "eeh_dev" any more.

The patches implements the state tracing based on PE. It's notable
that the PE state will be applied to the specified PE as well as
its child PEs. That complies with the rule that problematic parent
PE will prevent those child PEs from working properly.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Build EEH event based on PE
Gavin Shan [Fri, 7 Sep 2012 22:44:11 +0000 (22:44 +0000)]
powerpc/eeh: Build EEH event based on PE

The original implementation builds EEH event based on EEH device.
We already had dedicated struct to depict PE. It's reasonable to
build EEH event based on PE.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Remove PE at appropriate time
Gavin Shan [Fri, 7 Sep 2012 22:44:10 +0000 (22:44 +0000)]
powerpc/eeh: Remove PE at appropriate time

During PCI hotplug and EEH recovery, the PE hierarchy tree might be
changed due to the PCI topology changes. At later point when the
PCI device is added, the PE will be created dynamically again.

The patch introduces new function to remove EEH devices from the
associated PE. That also can cause that the parent PE is removed
from the PE tree if the parent PE doesn't include valid EEH devices
and child PEs.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Create PEs duing EEH initialization
Gavin Shan [Fri, 7 Sep 2012 22:44:09 +0000 (22:44 +0000)]
powerpc/eeh: Create PEs duing EEH initialization

The patch creates PEs and associated the newly created PEs with
it parent/silbing as well as EEH devices. It would become more
straight to trace EEH errors and recover them accordingly.

Once the EEH functionality on one PCI IOA has been enabled, we
tries to create PE against it. If there's existing PE, to which
the current PCI IOA should be attached, the existing PE will be
converted from "device" type to "bus" type accordingly.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Search PE based on requirement
Gavin Shan [Fri, 7 Sep 2012 22:44:08 +0000 (22:44 +0000)]
powerpc/eeh: Search PE based on requirement

The patch implements searching PE based on the following
requirements:

 * Search PE according to PE address, which is traditional
   PE address that is composed of PCI bus/device/function
   number, or unified PE address assigned by firmware or
   platform.
 * Search parent PE according to the given EEH device. It's
   useful when creating new PE and put it into right position.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Create PEs for PHBs
Gavin Shan [Fri, 7 Sep 2012 22:44:07 +0000 (22:44 +0000)]
powerpc/eeh: Create PEs for PHBs

For one particular PE, it's only meaningful in the ancestor PHB
domain. Therefore, each PHB should have its own PE hierarchy tree
to trace those PEs created against the PHB.

The patch creates PEs for the PHBs and put those PEs into the
global link list traced by "eeh_phb_pe". The link list of PEs
would be first level of overall PE hierarchy tree across the
system.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Introduce global mutex
Gavin Shan [Fri, 7 Sep 2012 22:44:06 +0000 (22:44 +0000)]
powerpc/eeh: Introduce global mutex

The patch introduces global mutex for EEH so that the core data
structures can be protected by that. Also, 2 inline functions
are exported for that: eeh_lock() and eeh_unlock().

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_pe struct
Gavin Shan [Fri, 7 Sep 2012 22:44:05 +0000 (22:44 +0000)]
powerpc/eeh: Introduce eeh_pe struct

As defined in PAPR 2.4, Partitionable Endpoint (PE) is an I/O subtree
that can be treated as a unit for the purposes of partitioning and error
recovery. Therefore, eeh core should be aware of PE. With eeh_pe struct,
we can support PE explicitly. Further more, it makes all the stuff much
more data centralized. Another important reason is for eeh core to support
multiple platforms. Some of them like pSeries figures out PEs through
OF nodes while others like powernv have to do that through PCI bus/device
tree. With explicit PE support, eeh core will be implemented based on
the centrialized data and platform dependent implementations figure it
out by their feasible ways.

When the struct is designed, following factors are taken in account:
  * Reflecting the relationships of PEs. PE might have parent
    as well children.
  * Reflecting the association of PE and (eeh) devices.
  * PEs have PHB boundary.
  * PE should have unique address assigned in the corresponding
    PHB domain.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: More logs for EEH initialization
Gavin Shan [Fri, 7 Sep 2012 22:44:04 +0000 (22:44 +0000)]
powerpc/eeh: More logs for EEH initialization

The patch adds more logs to EEH initialization functions for
debugging purpose. Also, the machine type (pSeries) is checked
in the platform initialization to assure it's the correct platform
to invoke it.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Use slab to allocate eeh devices
Gavin Shan [Fri, 7 Sep 2012 22:44:03 +0000 (22:44 +0000)]
powerpc/eeh: Use slab to allocate eeh devices

The EEH initialization functions have been postponed until slab/slub
are ready. So we use slab/slub to allocate the memory chunks for newly
creatd EEH devices. That would save lots of memory.

The patch also does cleanup to replace "kmalloc" with "kzalloc" so
that we needn't clear the allocated memory chunk explicitly.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc/eeh: Move EEH initialization around
Gavin Shan [Fri, 7 Sep 2012 22:44:02 +0000 (22:44 +0000)]
powerpc/eeh: Move EEH initialization around

Currently, we have 3 phases for EEH initialization on pSeries platform.
All of them are done through builtin functions: platform initialization,
EEH device creation, and EEH subsystem enablement. All of them are done
no later than ppc_md.setup_arch. That means that the slab/slub isn't ready
yet, so we have to allocate memory chunks on basis of PAGE_SIZE for those
dynamically created EEH devices. That's pretty expensive.

In order to utilize slab/slub for memory allocation, we have to move the EEH
initialization functions around, but all of them should be called after slab
is ready.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agopowerpc: Initialise paca.data_offset with poison
Michael Ellerman [Fri, 7 Sep 2012 15:31:44 +0000 (15:31 +0000)]
powerpc: Initialise paca.data_offset with poison

It's possible for the cpu_possible_mask to change between the time we
initialise the pacas and the time we setup per_cpu areas.

Obviously impossible cpus shouldn't ever be running, but stranger things
have happened. So be paranoid and initialise data_offset with a poison
value in case we don't set it up later.

Based on a patch from Anton Blanchard.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>