GFP_THISNODE is for callers that implement their own clever fallback to
remote nodes. It restricts the allocation to the specified node and
does not invoke reclaim, assuming that the caller will take care of it
when the fallback fails, e.g. through a subsequent allocation request
without GFP_THISNODE set.
However, many current GFP_THISNODE users only want the node exclusive
aspect of the flag, without actually implementing their own fallback or
triggering reclaim if necessary. This results in things like page
migration failing prematurely even when there is easily reclaimable
memory available, unless kswapd happens to be running already or a
concurrent allocation attempt triggers the necessary reclaim.
Convert all callsites that don't implement their own fallback strategy
to __GFP_THISNODE. This restricts the allocation a single node too, but
at the same time allows the allocator to enter the slowpath, wake
kswapd, and invoke direct reclaim if necessary, to make the allocation
happen when memory is full.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Jan Stancek <jstancek@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The 64bit relocation code places a few symbols in the text segment.
These symbols are only 4 byte aligned where they need to be 8 byte
aligned. Add an explicit alignment.
Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: stable@vger.kernel.org
Tested-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
When we fork/clone we currently don't copy any of the TM state to the new
thread. This results in a TM bad thing (program check) when the new process is
switched in as the kernel does a tmrechkpt with TEXASR FS not set. Also, since
R1 is from userspace, we trigger the bad kernel stack pointer detection. So we
end up with something like this:
Bad kernel stack pointer 0 at c0000000000404fc
cpu 0x2: Vector: 700 (Program Check) at [c00000003ffefd40]
pc: c0000000000404fc: restore_gprs+0xc0/0x148
lr: 0000000000000000
sp: 0
msr: 9000000100201030
current = 0xc000001dd1417c30
paca = 0xc00000000fe00800 softe: 0 irq_happened: 0x01
pid = 0, comm = swapper/2
WARNING: exception is not recoverable, can't continue
The below fixes this by flushing the TM state before we copy the task_struct to
the clone. To do this we go through the tmreclaim patch, which removes the
checkpointed registers from the CPU and transitions the CPU out of TM suspend
mode. Hence we need to call tmrechkpt after to restore the checkpointed state
and the TM mode for the current task.
To make this fail from userspace is simply:
tbegin
li r0, 2
sc
<boom>
Kudos to Adhemerval Zanella Neto for finding this.
Signed-off-by: Michael Neuling <mikey@neuling.org>
cc: Adhemerval Zanella Neto <azanella@br.ibm.com>
cc: stable@vger.kernel.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
We need to unmangle the full address, not just the register
number, and we also need to support the real indirect bit
being set for in-kernel uses.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
CC: <stable@vger.kernel.org> [v3.13]
The OPAL firmware functions opal_xscom_read and opal_xscom_write
take a 64-bit argument for the XSCOM (PCB) address in order to
support the indirect mode on P8.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
CC: <stable@vger.kernel.org> [v3.13]
As Ben suggested, the patch prints PHB diag-data with multiple
fields in one line and omits the line if the fields of that
line are all zero.
With the patch applied, the PHB3 diag-data dump looks like:
PHB3 PHB#3 Diag-data (Version: 1)
brdgCtl: 00000002
RootSts: 0000000f 00400000 b0830008 00100147 00002000
nFir: 0000000000000000 0030006e00000000 0000000000000000
PhbSts: 0000001c00000000 0000000000000000
Lem: 0000000000100000 42498e327f502eae 0000000000000000
InAErr: 8000000000000000 8000000000000000 0402030000000000 0000000000000000
PE[ 8] A/B: 8480002b00000000 8000000000000000
[ The current diag data is so big that it overflows the printk
buffer pretty quickly in cases when we get a handful of errors
at once which can happen. --BenH
]
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
CC: <stable@vger.kernel.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The PHB diag-data is important to help locating the root cause for
EEH errors such as frozen PE or fenced PHB. However, the EEH core
enables IO path by clearing part of HW registers before collecting
this data causing it to be corrupted.
This patch fixes this by dumping the PHB diag-data immediately when
frozen/fenced state on PE or PHB is detected for the first time in
eeh_ops::get_state() or next_error() backend.
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
CC: <stable@vger.kernel.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The new ELFv2 little-endian ABI increases the stack redzone -- the
area below the stack pointer that can be used for storing data --
from 288 bytes to 512 bytes. This means that we need to allow more
space on the user stack when delivering a signal to a 64-bit process.
To make the code a bit clearer, we define new USER_REDZONE_SIZE and
KERNEL_REDZONE_SIZE symbols in ptrace.h. For now, we leave the
kernel redzone size at 288 bytes, since increasing it to 512 bytes
would increase the size of interrupt stack frames correspondingly.
Gcc currently only makes use of 288 bytes of redzone even when
compiling for the new little-endian ABI, and the kernel cannot
currently be compiled with the new ABI anyway.
In the future, hopefully gcc will provide an option to control the
amount of redzone used, and then we could reduce it even more.
This also changes the code in arch_compat_alloc_user_space() to
preserve the expanded redzone. It is not clear why this function would
ever be used on a 64-bit process, though.
Signed-off-by: Paul Mackerras <paulus@samba.org>
CC: <stable@vger.kernel.org> [v3.13]
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The branch target should be the func addr, not the addr of func_descr_t.
So using ppc_function_entry() to generate the right target addr.
Signed-off-by: Liu Ping Fan <pingfank@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
In copy_oldmem_page, the current check using max_pfn and min_low_pfn to
decide if the page is backed or not, is not valid when the memory layout is
not continuous.
This happens when running as a QEMU/KVM guest, where RTAS is mapped higher
in the memory. In that case max_pfn points to the end of RTAS, and a hole
between the end of the kdump kernel and RTAS is not backed by PTEs. As a
consequence, the kdump kernel is crashing in copy_oldmem_page when accessing
in a direct way the pages in that hole.
This fix relies on the memblock's service memblock_is_region_memory to
check if the read page is part or not of the directly accessible memory.
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Tested-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
CC: <stable@vger.kernel.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Currently we're storing a host endian RTAS token in
rtas_stop_self_args.token. We then pass that directly to rtas. This is
fine on big endian however on little endian the token is not what we
expect.
This will typically result in hitting:
panic("Alas, I survived.\n");
To fix this we always use the stop-self token in host order and always
convert it to be32 before passing this to rtas.
Signed-off-by: Tony Breeds <tony@bakeyournoodle.com>
Cc: stable@vger.kernel.org
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
We possiblly detect EEH errors during reboot, particularly in kexec
path, but it's impossible for device drivers and EEH core to handle
or recover them properly.
The patch registers one reboot notifier for EEH and disable EEH
subsystem during reboot. That means the EEH errors is going to be
cleared by hardware reset or second kernel during early stage of
PCI probe.
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The patch cleans up variable eeh_subsystem_enabled so that we needn't
refer the variable directly from external. Instead, we will use
function eeh_enabled() and eeh_set_enable() to operate the variable.
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
When doing reset in order to recover the affected PE, we issue
hot reset on PE primary bus if it's not root bus. Otherwise, we
issue hot or fundamental reset on root port or PHB accordingly.
For the later case, we didn't cover the situation where PE only
includes root port and it potentially causes kernel crash upon
EEH error to the PE.
The patch reworks the logic of EEH reset to improve the code
readability and also avoid the kernel crash.
Cc: stable@vger.kernel.org
Reported-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
We are seeing a lot of hits in the VDSO that are not resolved by perf.
A while(1) gettimeofday() loop shows the issue:
27.64% [vdso] [.] 0x000000000000060c
22.57% [vdso] [.] 0x0000000000000628
16.88% [vdso] [.] 0x0000000000000610
12.39% [vdso] [.] __kernel_gettimeofday
6.09% [vdso] [.] 0x00000000000005f8
3.58% test [.] 00000037.plt_call.gettimeofday@@GLIBC_2.18
2.94% [vdso] [.] __kernel_datapage_offset
2.90% test [.] main
We are using a stripped VDSO image which means only symbols with
relocation info can be resolved. There isn't a lot of point to
stripping the VDSO, the debug info is only about 1kB:
4680 arch/powerpc/kernel/vdso64/vdso64.so
5815 arch/powerpc/kernel/vdso64/vdso64.so.dbg
By using the unstripped image, we can resolve all the symbols in the
VDSO and the perf profile data looks much better:
76.53% [vdso] [.] __do_get_tspec
12.20% [vdso] [.] __kernel_gettimeofday
5.05% [vdso] [.] __get_datapage
3.20% test [.] main
2.92% test [.] 00000037.plt_call.gettimeofday@@GLIBC_2.18
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
perf is failing to resolve symbols in the VDSO. A while (1)
gettimeofday() loop shows:
93.99% [vdso] [.] 0x00000000000005e0
3.12% test [.] 00000037.plt_call.gettimeofday@@GLIBC_2.18
2.81% test [.] main
The reason for this is that we are linking our VDSO shared libraries
at 1MB, which is a little weird. Even though this is uncommon, Alan
points out that it is valid and we should probably fix perf userspace.
Regardless, I can't see a reason why we are doing this. The code
is all position independent and we never rely on the VDSO ending
up at 1M (and we never place it there on 64bit tasks).
Changing our link address to 0x0 fixes perf VDSO symbol resolution:
73.18% [vdso] [.] 0x000000000000060c
12.39% [vdso] [.] __kernel_gettimeofday
3.58% test [.] 00000037.plt_call.gettimeofday@@GLIBC_2.18
2.94% [vdso] [.] __kernel_datapage_offset
2.90% test [.] main
We still have some local symbol resolution issues that will be
fixed in a subsequent patch.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Archs like ppc64 doesn't do tlb flush in set_pte/pmd functions when using
a hash table MMU for various reasons (the flush is handled as part of
the PTE modification when necessary).
ppc64 thus doesn't implement flush_tlb_range for hash based MMUs.
Additionally ppc64 require the tlb flushing to be batched within ptl locks.
The reason to do that is to ensure that the hash page table is in sync with
linux page table.
We track the hpte index in linux pte and if we clear them without flushing
hash and drop the ptl lock, we can have another cpu update the pte and can
end up with duplicate entry in the hash table, which is fatal.
We also want to keep set_pte_at simpler by not requiring them to do hash
flush for performance reason. We do that by assuming that set_pte_at() is
never *ever* called on a PTE that is already valid.
This was the case until the NUMA code went in which broke that assumption.
Fix that by introducing a new pair of helpers to set _PAGE_NUMA in a
way similar to ptep/pmdp_set_wrprotect(), with a generic implementation
using set_pte_at() and a powerpc specific one using the appropriate
mechanism needed to keep the hash table in sync.
Acked-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
pte_update() is a powerpc-ism used to change the bits of a PTE
when the access permission is being restricted (a flush is
potentially needed).
It uses atomic operations on when needed and handles the hash
synchronization on hash based processors.
It is currently only used to clear PTE bits and so the current
implementation doesn't provide a way to also set PTE bits.
The new _PAGE_NUMA bit, when set, is actually restricting access
so it must use that function too, so this change adds the ability
for pte_update() to also set bits.
We will use this later to set the _PAGE_NUMA bit.
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Rev3 of the PCI Express Base Specification defines a Supported Link
Speeds Vector where the bit definitions within this field are:
Bit 0 - 2.5 GT/s
Bit 1 - 5.0 GT/s
Bit 2 - 8.0 GT/s
This vector definition is used by the platform firmware to export the
maximum and current link speeds of the PCI bus via the
"ibm,pcie-link-speed-stats" device-tree property.
This patch updates pseries_root_bridge_prepare() to detect Gen3
speed buses (defined by 0x04).
Signed-off-by: Kleber Sacilotto de Souza <klebers@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Commit 5091f0c (powerpc/pseries: Fix PCIE link speed endian issue)
introduced a regression on the PCI link speed detection using the
device-tree property. The ibm,pcie-link-speed-stats property is composed
of two 32-bit integers, the first one being the maxinum link speed and
the second the current link speed. The changes introduced by the
aforementioned commit are considering just the first integer.
Fix this issue by changing how the property is accessed, using the
helper functions to properly access the array of values. The explicit
byte swapping is not needed anymore here, since it's done by the helper
functions.
Signed-off-by: Kleber Sacilotto de Souza <klebers@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Guenter Roeck has got the following call trace on a p2020 board:
Kernel stack overflow in process eb3e5a00, r1=eb79df90
CPU: 0 PID: 2838 Comm: ssh Not tainted 3.13.0-rc8-juniper-00146-g19eca00 #4
task: eb3e5a00 ti: c0616000 task.ti: ef440000
NIP: c003a420 LR: c003a410 CTR: c0017518
REGS: eb79dee0 TRAP: 0901 Not tainted (3.13.0-rc8-juniper-00146-g19eca00)
MSR: 00029000 <CE,EE,ME> CR: 24008444 XER: 00000000
GPR00: c003a410 eb79df90 eb3e5a00 00000000 eb05d900 00000001 65d87646 00000000
GPR08: 00000000 020b8000 00000000 00000000 44008442
NIP [c003a420] __do_softirq+0x94/0x1ec
LR [c003a410] __do_softirq+0x84/0x1ec
Call Trace:
[eb79df90] [c003a410] __do_softirq+0x84/0x1ec (unreliable)
[eb79dfe0] [c003a970] irq_exit+0xbc/0xc8
[eb79dff0] [c000cc1c] call_do_irq+0x24/0x3c
[ef441f20] [c00046a8] do_IRQ+0x8c/0xf8
[ef441f40] [c000e7f4] ret_from_except+0x0/0x18
--- Exception: 501 at 0xfcda524
LR = 0x10024900
Instruction dump:
7c781b78 3b40000a 3a73b040 543c0024 3a800000 3b3913a0 7ef5bb78 48201bf9
5463103a 7d3b182e 7e89b92e 7c008146 <3ba00000> 7e7e9b78 48000014 57fff87f
Kernel panic - not syncing: kernel stack overflow
CPU: 0 PID: 2838 Comm: ssh Not tainted 3.13.0-rc8-juniper-00146-g19eca00 #4
Call Trace:
The reason is that we have used the wrong register to calculate the
ksp_limit in commit cbc9565ee8 (powerpc: Remove ksp_limit on ppc64).
Just fix it.
As suggested by Benjamin Herrenschmidt, also add the C prototype of the
function in the comment in order to avoid such kind of errors in the
future.
Cc: stable@vger.kernel.org # 3.12
Reported-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This patch adds the support for to create a direct iommu "bypass"
window on IODA2 bridges (such as Power8) allowing to bypass iommu
page translation completely for 64-bit DMA capable devices, thus
significantly improving DMA performances.
Additionally, this adds a hook to the struct iommu_table so that
the IOMMU API / VFIO can disable the bypass when external ownership
is requested, since in that case, the device will be used by an
environment such as userspace or a KVM guest which must not be
allowed to bypass translations.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
We expose a number of OF properties in the kexec and crash dump code
and these need to be big endian.
Cc: stable@vger.kernel.org # v3.13
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
We would allocate one specific exception stack for each kind of
non-base exceptions for every CPU. For ppc32 the CPU hard ID is
used as the subscript to get the specific exception stack for
one CPU. But for an UP kernel, there is only one element in the
each kind of exception stack array. We would get stuck if the
CPU hard ID is not equal to '0'. So in this case we should use the
subscript '0' no matter what the CPU hard ID is.
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Currently we set our cpu's bit in cpus_in_xmon, and then we take the
output lock and print the exception information.
This can race with the master cpu entering the command loop and printing
the backtrace. The result is that the backtrace gets garbled with
another cpu's exception print out.
Fix it by delaying the set of cpus_in_xmon until we are finished
printing.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
