cascardo/linux.git
8 years agoselftests/powerpc: Make context_switch touch FP/altivec/vector by default
Michael Ellerman [Wed, 2 Dec 2015 09:44:11 +0000 (20:44 +1100)]
selftests/powerpc: Make context_switch touch FP/altivec/vector by default

Simply because it touches more code paths that way, and therefore tests
more things.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Anton Blanchard <anton@samba.org>
8 years agoselftests/powerpc: Make context_switch do something with no args
Michael Ellerman [Wed, 2 Dec 2015 09:44:10 +0000 (20:44 +1100)]
selftests/powerpc: Make context_switch do something with no args

For ease of use make the context_switch test do something useful when
called with no arguments.

Default to a 30 second run, using threads, doing yield, and use any
online cpu. Make it print out what it's doing to avoid confusion.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Anton Blanchard <anton@samba.org>
8 years agoselftests/powerpc: Import Anton's context_switch2 benchmark
Michael Ellerman [Wed, 2 Dec 2015 09:44:09 +0000 (20:44 +1100)]
selftests/powerpc: Import Anton's context_switch2 benchmark

This gets referred to a lot in commit messages, so let's pull it into
the selftests.

Almost vanilla from: http://ozlabs.org/~anton/junkcode/context_switch2.c

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Anton Blanchard <anton@samba.org>
8 years agoselftests/powerpc: Move pick_online_cpu() up into utils.c
Michael Ellerman [Wed, 16 Dec 2015 07:59:31 +0000 (18:59 +1100)]
selftests/powerpc: Move pick_online_cpu() up into utils.c

We want to use this in another test, so make it available at the top of
the powerpc selftests tree.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Remove broken GregorianDay()
Daniel Axtens [Tue, 15 Dec 2015 07:09:14 +0000 (18:09 +1100)]
powerpc: Remove broken GregorianDay()

GregorianDay() is supposed to calculate the day of the week
(tm->tm_wday) for a given day/month/year. In that calcuation it
indexed into an array called MonthOffset using tm->tm_mon-1. However
tm_mon is zero-based, not one-based, so this is off-by-one. It also
means that every January, GregoiranDay() will access element -1 of
the MonthOffset array.

It also doesn't appear to be a correct algorithm either: see in
contrast kernel/time/timeconv.c's time_to_tm function.

It's been broken forever, which suggests no-one in userland uses
this. It looks like no-one in the kernel uses tm->tm_wday either
(see e.g. drivers/rtc/rtc-ds1305.c:319).

tm->tm_wday is conventionally set to -1 when not available in
hardware so we can simply set it to -1 and drop the function.
(There are over a dozen other drivers in drivers/rtc that do
this.)

Found using UBSAN.

Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Andrew Morton <akpm@linux-foundation.org> # as an example of what UBSan finds.
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Cc: rtc-linux@googlegroups.com
Signed-off-by: Daniel Axtens <dja@axtens.net>
Acked-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoselftests/powerpc: Add test to check if VSRs are corrupted
Rashmica Gupta [Thu, 10 Dec 2015 09:49:33 +0000 (20:49 +1100)]
selftests/powerpc: Add test to check if VSRs are corrupted

When a transaction is aborted, VSR values should rollback to the
checkpointed values before the transaction began. VSRs used elsewhere in
the kernel during a transaction, or while the transaction is suspended
should not affect the checkpointed values.

Prior to the bug fix in commit d31626f70b61 ("powerpc: Don't corrupt
transactional state when using FP/VMX in kernel") when VMX was requested
by the kernel the .vr_state (which held the checkpointed state of VSRs
before the transaction) was overwritten with the current state from
outside the transation. Thus if the transaction did not complete, the
VSR values would be "rolled back" to potentially incorrect values.

Signed-off-by: Rashmica Gupta <rashmicy@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/xmon: Append linux_banner to exception information in xmon.
Rashmica Gupta [Wed, 25 Nov 2015 02:46:25 +0000 (13:46 +1100)]
powerpc/xmon: Append linux_banner to exception information in xmon.

Currently if you are in xmon without an oops etc. to view the kernel
version you have to type "d $linux_banner" - not necessarily obvious. As
this is useful information, append to the output of "e" command.

Example output:
  $mon> e
  cpu 0x1: Vector: 0  at [c0000000f879ba80]
      pc: c000000000081718: sysrq_handle_xmon+0x68/0x80
      lr: c000000000081718: sysrq_handle_xmon+0x68/0x80
      sp: c0000000f879bbe0
     msr: 8000000000009033
    current = 0xc0000000f604d5c0
    paca    = 0xc00000000fdc0480  softe: 0  irq_happened: 0x01
      pid   = 2467, comm = bash
  Linux version 4.4.0-rc2-00008-gc51af91c3ab3-dirty (rashmica@circle) (gcc
  version 5.1.1 20150629 (GCC) ) #45 SMP Wed Nov 25 10:25:12 AEDT 2015

Signed-off-by: Rashmica Gupta <rashmicy@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/cell: Remove the Cell QPACE code
Rashmica Gupta [Tue, 1 Dec 2015 03:51:38 +0000 (14:51 +1100)]
powerpc/cell: Remove the Cell QPACE code

All users of QPACE have upgraded to QPACE2 so remove the Cell QPACE code.

Signed-off-by: Rashmica Gupta <rashmicy@gmail.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/pseries: Limit EPOW reset event warnings
Vipin K Parashar [Tue, 1 Dec 2015 11:13:42 +0000 (16:43 +0530)]
powerpc/pseries: Limit EPOW reset event warnings

Kernel prints respective warnings about various EPOW events for
user information/action after parsing EPOW interrupts. At times
below EPOW reset event warning is seen to be flooding kernel log
over a period of time.

May 25 03:46:34 alp kernel: Non critical power or cooling issue cleared
May 25 03:46:52 alp kernel: Non critical power or cooling issue cleared
May 25 03:53:48 alp kernel: Non critical power or cooling issue cleared
May 25 03:55:46 alp kernel: Non critical power or cooling issue cleared
May 25 03:56:34 alp kernel: Non critical power or cooling issue cleared
May 25 03:59:04 alp kernel: Non critical power or cooling issue cleared
May 25 04:02:01 alp kernel: Non critical power or cooling issue cleared

These EPOW reset events are spurious in nature and are triggered by
firmware without an actual EPOW event being reset. This patch avoids these
multiple EPOW reset warnings by using a counter variable. This variable
is incremented every time an EPOW event is reported. Upon receiving a EPOW
reset event the same variable is checked to filter out spurious events and
decremented accordingly.

This patch also improves log messages to better describe EPOW event being
reported. Merged adjacent log messages into single one to reduce number of
lines printed per event.

Signed-off-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
Signed-off-by: Vipin K Parashar <vipin@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoselftests/powerpc: Add TM signal with invalid stack test
Michael Neuling [Fri, 20 Nov 2015 04:15:34 +0000 (15:15 +1100)]
selftests/powerpc: Add TM signal with invalid stack test

Test the kernels signal generation code to ensure it can handle an
invalid stack pointer when transactional.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Tested-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
[mpe: Skip if we don't have TM]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoselftests/powerpc: Add TM signal return test
Michael Neuling [Fri, 20 Nov 2015 04:15:33 +0000 (15:15 +1100)]
selftests/powerpc: Add TM signal return test

Test the kernel's signal return code to ensure that it doesn't crash
when both the transactional and suspend MSR bits are set in the signal
context.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Tested-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
[mpe: Skip if we don't have TM]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoselftests/powerpc: Skip tm-resched-dscr if we don't have TM
Michael Ellerman [Wed, 2 Dec 2015 05:00:04 +0000 (16:00 +1100)]
selftests/powerpc: Skip tm-resched-dscr if we don't have TM

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoselftests/powerpc: Move TM helpers into tm.h
Michael Ellerman [Tue, 24 Nov 2015 02:05:40 +0000 (13:05 +1100)]
selftests/powerpc: Move TM helpers into tm.h

Move have_htm_nosc() into a new tm.h, and add a new helper, have_htm()
which we'll use in the next patch.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoselftests/powerpc: Add have_hwcap2() helper
Michael Ellerman [Tue, 24 Nov 2015 02:05:39 +0000 (13:05 +1100)]
selftests/powerpc: Add have_hwcap2() helper

We already do this twice and want to add another so add a helper.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoselftests/powerpc: Move get_auxv_entry() into utils.c
Michael Ellerman [Tue, 24 Nov 2015 02:05:38 +0000 (13:05 +1100)]
selftests/powerpc: Move get_auxv_entry() into utils.c

This doesn't really belong in harness.c, it's a helper function. So move
it into utils.c.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoMerge tag 'powerpc-4.4-3' into next
Michael Ellerman [Mon, 14 Dec 2015 09:40:32 +0000 (20:40 +1100)]
Merge tag 'powerpc-4.4-3' into next

Merge the two TM fixes we merged in 4.4. We are about to merge selftests
for these, and without the fixes the selftests will oops.

powerpc fixes for 4.4 #2

 - tm: Block signal return from setting invalid MSR state from Michael Neuling
 - tm: Check for already reclaimed tasks from Michael Neuling

8 years agopowerpc: Print MSR TM bits in oops messages
Michael Neuling [Fri, 20 Nov 2015 04:15:32 +0000 (15:15 +1100)]
powerpc: Print MSR TM bits in oops messages

Print MSR TM bits in oops messages.  This appends them to the end
like this:

    MSR: 8000000502823031 <SF,VEC,VSX,FP,ME,IR,DR,LE,TM[TE]>

You get the TM[] only if at least one TM MSR bit is set.  Inside the
TM[], E means Enabled (bit 32), S means Suspended (bit 33), and T
means Transactional (bit 34)

If no bits are set, you get no TM[] output.

Include rework of printbits() to handle this case.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Make {cmp}xchg* and their atomic_ versions fully ordered
Boqun Feng [Mon, 2 Nov 2015 01:30:32 +0000 (09:30 +0800)]
powerpc: Make {cmp}xchg* and their atomic_ versions fully ordered

According to memory-barriers.txt, xchg*, cmpxchg* and their atomic_
versions all need to be fully ordered, however they are now just
RELEASE+ACQUIRE, which are not fully ordered.

So also replace PPC_RELEASE_BARRIER and PPC_ACQUIRE_BARRIER with
PPC_ATOMIC_ENTRY_BARRIER and PPC_ATOMIC_EXIT_BARRIER in
__{cmp,}xchg_{u32,u64} respectively to guarantee fully ordered semantics
of atomic{,64}_{cmp,}xchg() and {cmp,}xchg(), as a complement of commit
b97021f85517 ("powerpc: Fix atomic_xxx_return barrier semantics")

This patch depends on patch "powerpc: Make value-returning atomics fully
ordered" for PPC_ATOMIC_ENTRY_BARRIER definition.

Cc: stable@vger.kernel.org # 3.2+
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Make value-returning atomics fully ordered
Boqun Feng [Mon, 2 Nov 2015 01:30:31 +0000 (09:30 +0800)]
powerpc: Make value-returning atomics fully ordered

According to memory-barriers.txt:

> Any atomic operation that modifies some state in memory and returns
> information about the state (old or new) implies an SMP-conditional
> general memory barrier (smp_mb()) on each side of the actual
> operation ...

Which mean these operations should be fully ordered. However on PPC,
PPC_ATOMIC_ENTRY_BARRIER is the barrier before the actual operation,
which is currently "lwsync" if SMP=y. The leading "lwsync" can not
guarantee fully ordered atomics, according to Paul Mckenney:

https://lkml.org/lkml/2015/10/14/970

To fix this, we define PPC_ATOMIC_ENTRY_BARRIER as "sync" to guarantee
the fully-ordered semantics.

This also makes futex atomics fully ordered, which can avoid possible
memory ordering problems if userspace code relies on futex system call
for fully ordered semantics.

Fixes: b97021f85517 ("powerpc: Fix atomic_xxx_return barrier semantics")
Cc: stable@vger.kernel.org # 3.2+
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Don't open code pgtable_t size
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:37:00 +0000 (09:07 +0530)]
powerpc/mm: Don't open code pgtable_t size

The slot information of base page size hash pte is stored in the
pgtable_t w.r.t transparent hugepage. We need to make sure we don't
index beyond pgtable_t size.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Use H_READ with H_READ_4
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:59 +0000 (09:06 +0530)]
powerpc/mm: Use H_READ with H_READ_4

This will bulk read 4 hash pte slot entries and should reduce the loop

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/nohash: we don't use real_pte_t for nohash
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:58 +0000 (09:06 +0530)]
powerpc/nohash: we don't use real_pte_t for nohash

Remove the related functions and #defines

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/nohash: Update 64K nohash config to have 32 pte fragement
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:57 +0000 (09:06 +0530)]
powerpc/nohash: Update 64K nohash config to have 32 pte fragement

They don't need to track 4k subpage slot details and hence don't need
second half of pgtable_t.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Don't hardcode the hash pte slot shift
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:56 +0000 (09:06 +0530)]
powerpc/mm: Don't hardcode the hash pte slot shift

Use the #define instead of open-coding the same

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Don't hardcode page table size
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:55 +0000 (09:06 +0530)]
powerpc/mm: Don't hardcode page table size

pte and pmd table size are dependent on config items. Don't
hard code the same. This make sure we use the right value
when masking pmd entries and also while checking pmd_bad

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Add a _PAGE_PTE bit
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:54 +0000 (09:06 +0530)]
powerpc/mm: Add a _PAGE_PTE bit

For a pte entry we will have _PAGE_PTE set. Our pte page
address have a minimum alignment requirement of HUGEPD_SHIFT_MASK + 1.
We use the lower 7 bits to indicate hugepd. ie.

For pmd and pgd we can find:
1) _PAGE_PTE set pte -> indicate PTE
2) bits [2..6] non zero -> indicate hugepd.
   They also encode the size. We skip bit 1 (_PAGE_PRESENT).
3) othewise pointer to next table.

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Move THP headers around
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:53 +0000 (09:06 +0530)]
powerpc/mm: Move THP headers around

We support THP only with book3s_64 and 64K page size. Move
THP details to hash64-64k.h to clarify the same.

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Move hugetlb related headers
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:52 +0000 (09:06 +0530)]
powerpc/mm: Move hugetlb related headers

W.r.t hugetlb, we support two format for pmd. With book3s_64 and
64K linux page size, we can have pte at the pmd level. Hence we
don't need to support hugepd there. For everything else hugepd
is supported and pmd_huge is (0).

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Move WIMG update to helper.
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:51 +0000 (09:06 +0530)]
powerpc/mm: Move WIMG update to helper.

Only difference here is, we apply the WIMG mapping early, so rflags
passed to updatepp will also be changed.

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Add helper for converting pte bit to hpte bits
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:50 +0000 (09:06 +0530)]
powerpc/mm: Add helper for converting pte bit to hpte bits

Instead of open coding it in multiple code paths, export the helper
and add more documentation. Also make sure we don't make assumption
regarding pte bit position

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Convert 4k insert from asm to C
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:49 +0000 (09:06 +0530)]
powerpc/mm: Convert 4k insert from asm to C

This is similar to 64K insert. May be we want to consolidate

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Convert __hash_page_64K to C
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:48 +0000 (09:06 +0530)]
powerpc/mm: Convert __hash_page_64K to C

Convert from asm to C

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Increase the width of #define
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:47 +0000 (09:06 +0530)]
powerpc/mm: Increase the width of #define

No real change, only style changes

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Remove pte_val usage for the second half of pgtable_t
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:46 +0000 (09:06 +0530)]
powerpc/mm: Remove pte_val usage for the second half of pgtable_t

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Don't track subpage valid bit in pte_t
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:45 +0000 (09:06 +0530)]
powerpc/mm: Don't track subpage valid bit in pte_t

This free up 11 bits in pte_t. In the later patch we also change
the pte_t format so that we can start supporting migration pte
at pmd level. We now track 4k subpage valid bit as below

If we have _PAGE_COMBO set, we override the _PAGE_F_GIX_SHIFT
and _PAGE_F_SECOND. Together we have 4 bits, each of them
used to indicate whether any of the 4 4k subpage in that group
is valid. ie,

[ group 1 bit ]   [ group 2 bit ]  ..... [ group 4 ]
[ subpage 1 - 4]  [ subpage 5- 8]  ..... [ subpage 13 - 16]

We still track each 4k subpage slot number and secondary hash
information in the second half of pgtable_t. Removing the subpage
tracking have some significant overhead on aim9 and ebizzy benchmark and
to support THP with 4K subpage, we do need a pgtable_t of 4096 bytes.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Remove the dependency on pte bit position in asm code
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:44 +0000 (09:06 +0530)]
powerpc/mm: Remove the dependency on pte bit position in asm code

We should not expect pte bit position in asm code. Simply
by moving part of that to C

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Convert 4k hash insert to C
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:43 +0000 (09:06 +0530)]
powerpc/mm: Convert 4k hash insert to C

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/booke: Move nohash headers
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:38 +0000 (09:06 +0530)]
powerpc/booke: Move nohash headers

Move the booke related headers below booke/32 or booke/64

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Move PTE bits from generic functions to hash64 functions.
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:37 +0000 (09:06 +0530)]
powerpc/mm: Move PTE bits from generic functions to hash64 functions.

functions which operate on pte bits are moved to hash*.h and other
generic functions are moved to pgtable.h

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Move hash64 PTE bits from book3s/64/pgtable.h to hash.h
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:36 +0000 (09:06 +0530)]
powerpc/mm: Move hash64 PTE bits from book3s/64/pgtable.h to hash.h

This enables us to keep hash64 related bits together, and makes it easy
to follow.

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Don't use pmd_val, pud_val and pgd_val as lvalue
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:35 +0000 (09:06 +0530)]
powerpc/mm: Don't use pmd_val, pud_val and pgd_val as lvalue

We convert them static inline function here as we did with pte_val in
the previous patch

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Don't use pte_val as lvalue
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:34 +0000 (09:06 +0530)]
powerpc/mm: Don't use pte_val as lvalue

We also convert few #define to static inline in this patch for better
type checking

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Drop pte-common.h from BOOK3S 64
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:33 +0000 (09:06 +0530)]
powerpc/mm: Drop pte-common.h from BOOK3S 64

We copy only needed PTE bits define from pte-common.h to respective
hash related header. This should greatly simply later patches in which
we are going to change the pte format for hash config

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Don't have generic headers introduce functions touching pte bits
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:32 +0000 (09:06 +0530)]
powerpc/mm: Don't have generic headers introduce functions touching pte bits

We are going to drop pte_common.h in the later patch. The idea is to
enable hash code not require to define all PTE bits. Having PTE bits
defined in pte_common.h made the code unnecessarily complex.

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Delete booke bits from book3s
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:31 +0000 (09:06 +0530)]
powerpc/mm: Delete booke bits from book3s

We also move __ASSEMBLY__ towards the end of header. This avoid
having #ifndef __ASSEMBLY___ all over the header

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Move hash specific pte width and other defines to book3s
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:30 +0000 (09:06 +0530)]
powerpc/mm: Move hash specific pte width and other defines to book3s

This further make a copy of pte defines to book3s/64/hash*.h. This
remove the dependency on pgtable-ppc64-4k.h and pgtable-ppc64-64k.h

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: make a separate copy for book3s
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:28 +0000 (09:06 +0530)]
powerpc/mm: make a separate copy for book3s

In this patch we do:
cp pgtable-ppc32.h book3s/32/pgtable.h
cp pgtable-ppc64.h book3s/64/pgtable.h

This enable us to do further changes to hash specific config.
We will change the page table format for 64bit hash in later patches.

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: move pte headers to book3s directory
Aneesh Kumar K.V [Tue, 1 Dec 2015 03:36:26 +0000 (09:06 +0530)]
powerpc/mm: move pte headers to book3s directory

Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/mm: Fix infinite loop in hash fault with 4K page size
Aneesh Kumar K.V [Sat, 28 Nov 2015 17:09:33 +0000 (22:39 +0530)]
powerpc/mm: Fix infinite loop in hash fault with 4K page size

This is the same bug we fixed as part of 09567e7fd44291bfc08accfdd67ad8f467842332
("powerpc/mm: Check paca psize is up to date for huge mappings"). Please
check that for details. The difference here is that faults were
happening on a 4K page at an address previously mapped by hugetlb.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Fix DSCR inheritance over fork()
Anton Blanchard [Wed, 9 Dec 2015 09:11:47 +0000 (20:11 +1100)]
powerpc: Fix DSCR inheritance over fork()

Two DSCR tests have a hack in them:

/*
 * XXX: Force a context switch out so that DSCR
 * current value is copied into the thread struct
 * which is required for the child to inherit the
 * changed value.
 */
sleep(1);

We should not be working around this in the testcase, it is a kernel bug.
Fix it by copying the current DSCR to the child, instead of what we
had in the thread struct at last context switch.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Call restore_sprs() before _switch()
Anton Blanchard [Thu, 10 Dec 2015 09:44:39 +0000 (20:44 +1100)]
powerpc: Call restore_sprs() before _switch()

commit 152d523e6307 ("powerpc: Create context switch helpers save_sprs()
and restore_sprs()") moved the restore of SPRs after the call to _switch().

There is an issue with this approach - new tasks do not return through
_switch(), they are set up by copy_thread() to directly return through
ret_from_fork() or ret_from_kernel_thread(). This means restore_sprs() is
not getting called for new tasks.

Fix this by moving restore_sprs() before _switch().

Fixes: 152d523e6307 ("powerpc: Create context switch helpers save_sprs() and restore_sprs()")
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Call check_if_tm_restore_required() in enable_kernel_*()
Anton Blanchard [Thu, 10 Dec 2015 09:04:05 +0000 (20:04 +1100)]
powerpc: Call check_if_tm_restore_required() in enable_kernel_*()

Commit a0e72cf12b1a ("powerpc: Create msr_check_and_{set,clear}()")
removed a call to check_if_tm_restore_required() in the
enable_kernel_*() functions. Add them back in.

Fixes: a0e72cf12b1a ("powerpc: Create msr_check_and_{set,clear}()")
Reported-by: Rashmica Gupta <rashmicy@gmail.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: clean up asm/switch_to.h
Anton Blanchard [Thu, 29 Oct 2015 00:44:11 +0000 (11:44 +1100)]
powerpc: clean up asm/switch_to.h

Remove a bunch of unnecessary fallback functions and group
things in a more logical way.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Rearrange __switch_to()
Anton Blanchard [Thu, 29 Oct 2015 00:44:10 +0000 (11:44 +1100)]
powerpc: Rearrange __switch_to()

Most of __switch_to() is housekeeping, TLB batching, timekeeping etc.
Move these away from the more complex and critical context switching
code.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: create flush_all_to_thread()
Anton Blanchard [Thu, 29 Oct 2015 00:44:09 +0000 (11:44 +1100)]
powerpc: create flush_all_to_thread()

Create a single function that flushes everything (FP, VMX, VSX, SPE).
Doing this all at once means we only do one MSR write.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: create giveup_all()
Anton Blanchard [Thu, 29 Oct 2015 00:44:08 +0000 (11:44 +1100)]
powerpc: create giveup_all()

Create a single function that gives everything up (FP, VMX, VSX, SPE).
Doing this all at once means we only do one MSR write.

A context switch microbenchmark using yield():

http://ozlabs.org/~anton/junkcode/context_switch2.c

./context_switch2 --test=yield --fp --altivec --vector 0 0

shows an improvement of 3% on POWER8.

Signed-off-by: Anton Blanchard <anton@samba.org>
[mpe: giveup_all() needs to be EXPORT_SYMBOL'ed]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Remove fp_enable() and vec_enable(), use msr_check_and_{set, clear}()
Anton Blanchard [Thu, 29 Oct 2015 00:44:07 +0000 (11:44 +1100)]
powerpc: Remove fp_enable() and vec_enable(), use msr_check_and_{set, clear}()

More consolidation of our MSR available bit handling.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Add ppc_strict_facility_enable boot option
Anton Blanchard [Thu, 29 Oct 2015 00:44:06 +0000 (11:44 +1100)]
powerpc: Add ppc_strict_facility_enable boot option

Add a boot option that strictly manages the MSR unavailable bits.
This catches kernel uses of FP/Altivec/SPE that would otherwise
corrupt user state.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Create disable_kernel_{fp,altivec,vsx,spe}()
Anton Blanchard [Thu, 29 Oct 2015 00:44:05 +0000 (11:44 +1100)]
powerpc: Create disable_kernel_{fp,altivec,vsx,spe}()

The enable_kernel_*() functions leave the relevant MSR bits enabled
until we exit the kernel sometime later. Create disable versions
that wrap the kernel use of FP, Altivec VSX or SPE.

While we don't want to disable it normally for performance reasons
(MSR writes are slow), it will be used for a debug boot option that
does this and catches bad uses in other areas of the kernel.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Create msr_check_and_{set,clear}()
Anton Blanchard [Thu, 29 Oct 2015 00:44:04 +0000 (11:44 +1100)]
powerpc: Create msr_check_and_{set,clear}()

Create helper functions to set and clear MSR bits after first
checking if they are already set. Grouping them will make it
easy to avoid the MSR writes in a subsequent optimisation.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agocrypto: vmx: Only call enable_kernel_vsx()
Anton Blanchard [Thu, 29 Oct 2015 00:44:03 +0000 (11:44 +1100)]
crypto: vmx: Only call enable_kernel_vsx()

With the recent change to enable_kernel_vsx(), we no longer need
to call enable_kernel_fp() and enable_kernel_altivec().

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Move part of giveup_vsx into c
Anton Blanchard [Thu, 29 Oct 2015 00:44:02 +0000 (11:44 +1100)]
powerpc: Move part of giveup_vsx into c

Move the MSR modification into c. Removing it from the assembly
function will allow us to avoid costly MSR writes by batching them
up.

Check the FP and VMX bits before calling the relevant giveup_*()
function. This makes giveup_vsx() and flush_vsx_to_thread() perform
more like their sister functions, and allows us to use
flush_vsx_to_thread() in the signal code.

Move the check_if_tm_restore_required() check in.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Move part of giveup_fpu,altivec,spe into c
Anton Blanchard [Thu, 29 Oct 2015 00:44:01 +0000 (11:44 +1100)]
powerpc: Move part of giveup_fpu,altivec,spe into c

Move the MSR modification into new c functions. Removing it from
the low level functions will allow us to avoid costly MSR writes
by batching them up.

Move the check_if_tm_restore_required() check into these new functions.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Remove NULL task struct pointer checks in FP and vector code
Anton Blanchard [Thu, 29 Oct 2015 00:44:00 +0000 (11:44 +1100)]
powerpc: Remove NULL task struct pointer checks in FP and vector code

We used to allow giveup_*() to be called with a NULL task struct
pointer. Now those cases are handled in the caller we can remove
the checks. We can also remove giveup_altivec_notask() which is also
unused.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Create mtmsrd_isync()
Anton Blanchard [Thu, 29 Oct 2015 00:43:59 +0000 (11:43 +1100)]
powerpc: Create mtmsrd_isync()

mtmsrd_isync() will do an mtmsrd followed by an isync on older
processors. On newer processors we avoid the isync via a feature fixup.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Simplify TM restore checks
Anton Blanchard [Thu, 29 Oct 2015 00:43:58 +0000 (11:43 +1100)]
powerpc: Simplify TM restore checks

Instead of having multiple giveup_*_maybe_transactional() functions,
separate out the TM check into a new function called
check_if_tm_restore_required().

This will make it easier to optimise the giveup_*() functions in a
subsequent patch.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Remove UP only lazy floating point and vector optimisations
Anton Blanchard [Thu, 29 Oct 2015 00:43:57 +0000 (11:43 +1100)]
powerpc: Remove UP only lazy floating point and vector optimisations

The UP only lazy floating point and vector optimisations were written
back when SMP was not common, and neither glibc nor gcc used vector
instructions. Now SMP is very common, glibc aggressively uses vector
instructions and gcc autovectorises.

We want to add new optimisations that apply to both UP and SMP, but
in preparation for that remove these UP only optimisations.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Remove redundant mflr in _switch
Anton Blanchard [Thu, 29 Oct 2015 00:43:56 +0000 (11:43 +1100)]
powerpc: Remove redundant mflr in _switch

No need to execute mflr twice.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Create context switch helpers save_sprs() and restore_sprs()
Anton Blanchard [Thu, 29 Oct 2015 00:43:55 +0000 (11:43 +1100)]
powerpc: Create context switch helpers save_sprs() and restore_sprs()

Move all our context switch SPR save and restore code into two
helpers. We do a few optimisations:

- Group all mfsprs and all mtsprs. In many cases an mtspr sets a
scoreboarding bit that an mfspr waits on, so the current practise of
mfspr A; mtspr A; mfpsr B; mtspr B is the worst scheduling we can
do.

- SPR writes are slow, so check that the value is changing before
writing it.

A context switch microbenchmark using yield():

http://ozlabs.org/~anton/junkcode/context_switch2.c

./context_switch2 --test=yield 0 0

shows an improvement of almost 10% on POWER8.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Don't disable MSR bits in do_load_up_transact_*() functions
Anton Blanchard [Thu, 29 Oct 2015 00:43:54 +0000 (11:43 +1100)]
powerpc: Don't disable MSR bits in do_load_up_transact_*() functions

Similar to the non TM load_up_*() functions, don't disable the MSR
bits on the way out.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Don't disable kernel FP/VMX/VSX MSR bits on context switch
Anton Blanchard [Thu, 29 Oct 2015 00:43:53 +0000 (11:43 +1100)]
powerpc: Don't disable kernel FP/VMX/VSX MSR bits on context switch

Writing the MSR is slow, so we want to avoid it whenever possible.

A subsequent patch will add a debug option that strictly manages the
FP/VMX/VSX unavailable bits. For now just remove it, matching what
we do in other areas of the kernel (eg enable_kernel_altivec()).

A context switch microbenchmark using yield():

http://ozlabs.org/~anton/junkcode/context_switch2.c

./context_switch2 --test=yield --fp 0 0

shows an improvement of almost 3% on POWER8.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/64: Include KVM guest test in all interrupt vectors
Paul Mackerras [Thu, 12 Nov 2015 05:44:42 +0000 (16:44 +1100)]
powerpc/64: Include KVM guest test in all interrupt vectors

Currently, if HV KVM is configured but PR KVM isn't, we don't include
a test to see whether we were interrupted in KVM guest context for the
set of interrupts which get delivered directly to the guest by hardware
if they occur in the guest.  This includes things like program
interrupts.

However, the recent bug where userspace could set the MSR for a VCPU
to have an illegal value in the TS field, and thus cause a TM Bad Thing
type of program interrupt on the hrfid that enters the guest, showed that
we can never be completely sure that these interrupts can never occur
in the guest entry/exit code.  If one of these interrupts does happen
and we have HV KVM configured but not PR KVM, then we end up trying to
run the handler in the host with the MMU set to the guest MMU context,
which generally ends badly.

Thus, for robustness it is better to have the test in every interrupt
vector, so that if some way is found to trigger some interrupt in the
guest entry/exit path, we can handle it without immediately crashing
the host.

This means that the distinction between KVMTEST and KVMTEST_PR goes
away.  Thus we delete KVMTEST_PR and associated macros and use KVMTEST
everywhere that we previously used either KVMTEST_PR or KVMTEST.  It
also means that SOFTEN_TEST_HV_201 becomes the same as SOFTEN_TEST_PR,
so we deleted SOFTEN_TEST_HV_201 and use SOFTEN_TEST_PR instead.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/axonram: Fix module autoload for OF platform driver
Luis de Bethencourt [Tue, 20 Oct 2015 15:04:13 +0000 (16:04 +0100)]
powerpc/axonram: Fix module autoload for OF platform driver

This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.

Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/rackmeter: Fix module autoload for OF platform driver
Luis de Bethencourt [Tue, 20 Oct 2015 15:13:53 +0000 (16:13 +0100)]
powerpc/rackmeter: Fix module autoload for OF platform driver

This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.

Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Add rN aliases to the pt_regs_offset table.
Rashmica Gupta [Sat, 21 Nov 2015 06:08:16 +0000 (17:08 +1100)]
powerpc: Add rN aliases to the pt_regs_offset table.

It is common practice with powerpc to use 'rN' to refer to register 'N'. However
when using the pt_regs_offset table we have to use 'gprN'.

So add aliases such that both 'rN' and 'gprN' can be used.

For example, we can currently do:
  $ su -
  $ cd /sys/kernel/debug/tracing
  $ echo "p:probe/sys_fchownat sys_fchownat %gpr3:s32 +0(%gpr4):string %gpr5:s32 %gpr6:s32 %gpr7:s32" > kprobe_events
  $ echo 1 > events/probe/sys_fchownat/enable
  $ touch /tmp/foo
  $ chown root /tmp/foo
  $ echo 0 > events/enable
  $ cat trace
    chown-2925  [014] d...    76.160657: sys_fchownat: (SyS_fchownat+0x8/0x1a0) arg1=-100 arg2="/tmp/foo" arg3=0 arg4=-1 arg5=0

Instead we'd like to be able to use:
 $ echo "p:probe/sys_fchownat sys_fchownat %r3:s32 +0(%r4):string %r5:s32 %r6:s32 %r7:s32" > kprobe_events

Signed-off-by: Rashmica Gupta <rashmicy@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Standardise on NR_syscalls rather than __NR_syscalls.
Rashmica Gupta [Thu, 19 Nov 2015 06:04:53 +0000 (17:04 +1100)]
powerpc: Standardise on NR_syscalls rather than __NR_syscalls.

Most architectures use NR_syscalls as the #define for the number of syscalls.

We use __NR_syscalls, and then define NR_syscalls as __NR_syscalls.

__NR_syscalls is not used outside arch code, whereas NR_syscalls is. So as
NR_syscalls must be defined and __NR_syscalls does not, replace __NR_syscalls
with NR_syscalls.

Signed-off-by: Rashmica Gupta <rashmicy@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc: Remove unused function trace_syscall()
Rashmica Gupta [Thu, 19 Nov 2015 03:26:28 +0000 (14:26 +1100)]
powerpc: Remove unused function trace_syscall()

This function has been unused since commit 14cf11af6cf6 ("powerpc: Merge enough
to start building in arch/powerpc."), so remove it.

Signed-off-by: Rashmica Gupta <rashmicy@gmail.com>
Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Reviewed-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/boot: allow wrapper to work on non-english system
Laurent Vivier [Thu, 5 Nov 2015 11:31:34 +0000 (12:31 +0100)]
powerpc/boot: allow wrapper to work on non-english system

if the language is not english objdump output is not parsed correctly
and format is "". Later, "ld -m $format" fails.

This patch adds "LANG=C" to force english output for objdump.

Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/powermac: IRQF_NO_SUSPEND not IRQF_TIMER for non-timer
John Ogness [Wed, 11 Nov 2015 14:15:03 +0000 (15:15 +0100)]
powerpc/powermac: IRQF_NO_SUSPEND not IRQF_TIMER for non-timer

Since gpio1 is not a timer, it also should not use IRQF_TIMER.

Similar to commit ba461f094bab ("powerpc: Use IRQF_NO_SUSPEND not
IRQF_TIMER for non-timer interrupts").

Signed-off-by: John Ogness <john.ogness@linutronix.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/powermac: set IRQF_NO_THREAD for xmon/cascade handlers
John Ogness [Wed, 11 Nov 2015 13:48:50 +0000 (14:48 +0100)]
powerpc/powermac: set IRQF_NO_THREAD for xmon/cascade handlers

The xmon and cascade irq handlers must not run as threads.
pmac_pic_lock is already a raw_spinlock, but the irq flag
IRQF_NO_THREAD needs to be set as well.

Signed-off-by: John Ogness <john.ogness@linutronix.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/powernv: Drop owner assignment from platform_driver
Krzysztof Kozlowski [Thu, 19 Nov 2015 04:00:39 +0000 (13:00 +0900)]
powerpc/powernv: Drop owner assignment from platform_driver

platform_driver does not need to set an owner because
platform_driver_register() will set it.

Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agocxl: use correct operator when writing pcie config space values
Andrew Donnellan [Wed, 4 Nov 2015 02:24:09 +0000 (13:24 +1100)]
cxl: use correct operator when writing pcie config space values

When writing a value to config space, cxl_pcie_write_config() calls
cxl_pcie_config_info() to obtain a mask and shift value, shifts the new
value accordingly, then uses the mask to combine the shifted value with the
existing value at the address as part of a read-modify-write pattern.

Currently, we use a logical OR operator rather than a bitwise OR operator,
which means any use of this function results in an incorrect value being
written. Replace the logical OR operator with a bitwise OR operator so the
value is written correctly.

Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Cc: stable@vger.kernel.org
Fixes: 6f7f0b3df6d4 ("cxl: Add AFU virtual PHB and kernel API")
Signed-off-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agocxl: Fix possible idr warning when contexts are released
Vaibhav Jain [Mon, 16 Nov 2015 04:03:45 +0000 (09:33 +0530)]
cxl: Fix possible idr warning when contexts are released

An idr warning is reported when a context is release after the capi card
is unbound from the cxl driver via sysfs. Below are the steps to
reproduce:

1. Create multiple afu contexts in an user-space application using libcxl.
2. Unbind capi card from cxl using command of form
   echo <capi-card-pci-addr> > /sys/bus/pci/drivers/cxl-pci/unbind
3. Exit/kill the application owning afu contexts.

After above steps a warning message is usually seen in the kernel logs
of the form "idr_remove called for id=<context-id> which is not
allocated."

This is caused by the function cxl_release_afu which destroys the
contexts_idr table. So when a context is release no entry for context pe
is found in the contexts_idr table and idr code prints this warning.

This patch fixes this issue by increasing & decreasing the ref-count on
the afu device when a context is initialized or when its freed
respectively. This prevents the afu from being released until all the
afu contexts have been released. The patch introduces two new functions
namely cxl_afu_get/put that manage the ref-count on the afu device.

Also the patch removes code inside cxl_dev_context_init that increases ref
on the afu device as its guaranteed to be alive during this function.

Reported-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Vaibhav Jain <vaibhav@linux.vnet.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/tm: Check for already reclaimed tasks
Michael Neuling [Thu, 19 Nov 2015 04:44:45 +0000 (15:44 +1100)]
powerpc/tm: Check for already reclaimed tasks

Currently we can hit a scenario where we'll tm_reclaim() twice.  This
results in a TM bad thing exception because the second reclaim occurs
when not in suspend mode.

The scenario in which this can happen is the following.  We attempt to
deliver a signal to userspace.  To do this we need obtain the stack
pointer to write the signal context.  To get this stack pointer we
must tm_reclaim() in case we need to use the checkpointed stack
pointer (see get_tm_stackpointer()).  Normally we'd then return
directly to userspace to deliver the signal without going through
__switch_to().

Unfortunatley, if at this point we get an error (such as a bad
userspace stack pointer), we need to exit the process.  The exit will
result in a __switch_to().  __switch_to() will attempt to save the
process state which results in another tm_reclaim().  This
tm_reclaim() now causes a TM Bad Thing exception as this state has
already been saved and the processor is no longer in TM suspend mode.
Whee!

This patch checks the state of the MSR to ensure we are TM suspended
before we attempt the tm_reclaim().  If we've already saved the state
away, we should no longer be in TM suspend mode.  This has the
additional advantage of checking for a potential TM Bad Thing
exception.

Found using syscall fuzzer.

Fixes: fb09692e71f1 ("powerpc: Add reclaim and recheckpoint functions for context switching transactional memory processes")
Cc: stable@vger.kernel.org # v3.9+
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agopowerpc/tm: Block signal return setting invalid MSR state
Michael Neuling [Thu, 19 Nov 2015 04:44:44 +0000 (15:44 +1100)]
powerpc/tm: Block signal return setting invalid MSR state

Currently we allow both the MSR T and S bits to be set by userspace on
a signal return.  Unfortunately this is a reserved configuration and
will cause a TM Bad Thing exception if attempted (via rfid).

This patch checks for this case in both the 32 and 64 bit signals
code.  If both T and S are set, we mark the context as invalid.

Found using a syscall fuzzer.

Fixes: 2b0a576d15e0 ("powerpc: Add new transactional memory state to the signal context")
Cc: stable@vger.kernel.org # v3.9+
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agoLinux 4.4-rc2 v4.4-rc2
Linus Torvalds [Mon, 23 Nov 2015 00:45:59 +0000 (16:45 -0800)]
Linux 4.4-rc2

8 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sun, 22 Nov 2015 23:21:40 +0000 (15:21 -0800)]
Merge branch 'akpm' (patches from Andrew)

Merge slub bulk allocator updates from Andrew Morton:
 "This missed the merge window because I was waiting for some repairs to
  come in.  Nothing actually uses the bulk allocator yet and the changes
  to other code paths are pretty small.  And the net guys are waiting
  for this so they can start merging the client code"

More comments from Jesper Dangaard Brouer:
 "The kmem_cache_alloc_bulk() call, in mm/slub.c, were included in
  previous kernel.  The present version contains a bug.  Vladimir
  Davydov noticed it contained a bug, when kernel is compiled with
  CONFIG_MEMCG_KMEM (see commit 03ec0ed57ffc: "slub: fix kmem cgroup
  bug in kmem_cache_alloc_bulk").  Plus the mem cgroup counterpart in
  kmem_cache_free_bulk() were missing (see commit 033745189b1b "slub:
  add missing kmem cgroup support to kmem_cache_free_bulk").

  I don't consider the fix stable-material because there are no in-tree
  users of the API.

  But with known bugs (for memcg) I cannot start using the API in the
  net-tree"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  slab/slub: adjust kmem_cache_alloc_bulk API
  slub: add missing kmem cgroup support to kmem_cache_free_bulk
  slub: fix kmem cgroup bug in kmem_cache_alloc_bulk
  slub: optimize bulk slowpath free by detached freelist
  slub: support for bulk free with SLUB freelists

8 years agoMerge tag 'tty-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
Linus Torvalds [Sun, 22 Nov 2015 23:10:57 +0000 (15:10 -0800)]
Merge tag 'tty-4.4-rc2' of git://git./linux/kernel/git/gregkh/tty

Pull tty/serial fixes from Greg KH:
 "Here are a few small tty/serial driver fixes for 4.4-rc2 that resolve
  some reported problems.

  All have been in linux-next, full details are in the shortlog below"

* tag 'tty-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty:
  serial: export fsl8250_handle_irq
  serial: 8250_mid: Add missing dependency
  tty: audit: Fix audit source
  serial: etraxfs-uart: Fix crash
  serial: fsl_lpuart: Fix earlycon support
  bcm63xx_uart: Use the device name when registering an interrupt
  tty: Fix direct use of tty buffer work
  tty: Fix tty_send_xchar() lock order inversion

8 years agoMerge tag 'staging-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 22 Nov 2015 21:26:24 +0000 (13:26 -0800)]
Merge tag 'staging-4.4-rc2' of git://git./linux/kernel/git/gregkh/staging

Pull staging/IIO fixes from Greg KH:
 "Here are some staging and iio driver fixes for 4.4-rc2.  All of these
  are in response to issues that have been reported and have been in
  linux-next for a while"

* tag 'staging-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  Revert "Staging: wilc1000: coreconfigurator: Drop unneeded wrapper functions"
  iio: adc: xilinx: Fix VREFN scale
  iio: si7020: Swap data byte order
  iio: adc: vf610_adc: Fix division by zero error
  iio:ad7793: Fix ad7785 product ID
  iio: ad5064: Fix ad5629/ad5669 shift
  iio:ad5064: Make sure ad5064_i2c_write() returns 0 on success
  iio: lpc32xx_adc: fix warnings caused by enabling unprepared clock
  staging: iio: select IRQ_WORK for IIO_DUMMY_EVGEN
  vf610_adc: Fix internal temperature calculation

8 years agoMerge tag 'usb-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sun, 22 Nov 2015 21:15:05 +0000 (13:15 -0800)]
Merge tag 'usb-4.4-rc2' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are a number of USB fixes and new device ids for 4.4-rc2.  All
  have been in linux-next and the details are in the shortlog"

* tag 'usb-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb: (28 commits)
  usblp: do not set TASK_INTERRUPTIBLE before lock
  USB: MAINTAINERS: cxacru
  usb: kconfig: fix warning of select USB_OTG
  USB: option: add XS Stick W100-2 from 4G Systems
  xhci: Fix a race in usb2 LPM resume, blocking U3 for usb2 devices
  usb: xhci: fix checking ep busy for CFC
  xhci: Workaround to get Intel xHCI reset working more reliably
  usb: chipidea: imx: fix a possible NULL dereference
  usb: chipidea: usbmisc_imx: fix a possible NULL dereference
  usb: chipidea: otg: gadget module load and unload support
  usb: chipidea: debug: disable usb irq while role switch
  ARM: dts: imx27.dtsi: change the clock information for usb
  usb: chipidea: imx: refine clock operations to adapt for all platforms
  usb: gadget: atmel_usba_udc: Expose correct device speed
  usb: musb: enable usb_dma parameter
  usb: phy: phy-mxs-usb: fix a possible NULL dereference
  usb: dwc3: gadget: let us set lower max_speed
  usb: musb: fix tx fifo flush handling
  usb: gadget: f_loopback: fix the warning during the enumeration
  usb: dwc2: host: Fix remote wakeup when not in DWC2_L2
  ...

8 years agoMerge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
Linus Torvalds [Sun, 22 Nov 2015 20:59:46 +0000 (12:59 -0800)]
Merge branch 'upstream' of git://git.linux-mips.org/ralf/upstream-linus

Pull MIPS fixes from Ralf Baechle:

 - Fix a flood of annoying build warnings

 - A number of fixes for Atheros 79xx platforms

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus:
  MIPS: ath79: Add a machine entry for booting OF machines
  MIPS: ath79: Fix the size of the MISC INTC registers in ar9132.dtsi
  MIPS: ath79: Fix the DDR control initialization on ar71xx and ar934x
  MIPS: Fix flood of warnings about comparsion being always true.

8 years agoMerge branch 'parisc-4.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Sun, 22 Nov 2015 20:50:58 +0000 (12:50 -0800)]
Merge branch 'parisc-4.4-2' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc update from Helge Deller:
 "This patchset adds Huge Page and HUGETLBFS support for parisc"

Honestly, the hugepage support should have gone through in the merge
window, and is not really an rc-time fix.  But it only touches
arch/parisc, and I cannot find it in myself to care.  If one of the
three parisc users notices a breakage, I will point at Helge and make
rude farting noises.

* 'parisc-4.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Map kernel text and data on huge pages
  parisc: Add Huge Page and HUGETLBFS support
  parisc: Use long branch to do_syscall_trace_exit
  parisc: Increase initial kernel mapping to 32MB on 64bit kernel
  parisc: Initialize the fault vector earlier in the boot process.
  parisc: Add defines for Huge page support
  parisc: Drop unused MADV_xxxK_PAGES flags from asm/mman.h
  parisc: Drop definition of start_thread_som for HP-UX SOM binaries
  parisc: Fix wrong comment regarding first pmd entry flags

8 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 22 Nov 2015 20:37:20 +0000 (12:37 -0800)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf tool fixes from Thomas Gleixner:
 "A couple of fixes for perf tools:

   - Build system updates

   - Plug a memory leak in an error path of perf probe

   - Tear down probes correctly when adding fails

   - Fixes to the perf symbol handling

   - Fix ordering of event processing in buildid-list

   - Fix per DSO filtering in the histogram browser"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf probe: Clear probe_trace_event when add_probe_trace_event() fails
  perf probe: Fix memory leaking on failure by clearing all probe_trace_events
  perf inject: Also re-pipe lost_samples event
  perf buildid-list: Requires ordered events
  perf symbols: Fix dso lookup by long name and missing buildids
  perf symbols: Allow forcing reading of non-root owned files by root
  perf hists browser: The dso can be obtained from popup_action->ms.map->dso
  perf hists browser: Fix 'd' hotkey action to filter by DSO
  perf symbols: Rebuild rbtree when adjusting symbols for kcore
  tools: Add a "make all" rule
  tools: Actually install tmon in the install rule

8 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 22 Nov 2015 20:00:12 +0000 (12:00 -0800)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "This update contains:

   - MPX updates for handling 32bit processes

   - A fix for a long standing bug in 32bit signal frame handling
     related to FPU/XSAVE state

   - Handle get_xsave_addr() correctly in KVM

   - Fix SMAP check under paravirtualization

   - Add a comment to the static function trace entry to avoid further
     confusion about the difference to dynamic tracing"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/cpu: Fix SMAP check in PVOPS environments
  x86/ftrace: Add comment on static function tracing
  x86/fpu: Fix get_xsave_addr() behavior under virtualization
  x86/fpu: Fix 32-bit signal frame handling
  x86/mpx: Fix 32-bit address space calculation
  x86/mpx: Do proper get_user() when running 32-bit binaries on 64-bit kernels

8 years agoslab/slub: adjust kmem_cache_alloc_bulk API
Jesper Dangaard Brouer [Fri, 20 Nov 2015 23:57:58 +0000 (15:57 -0800)]
slab/slub: adjust kmem_cache_alloc_bulk API

Adjust kmem_cache_alloc_bulk API before we have any real users.

Adjust API to return type 'int' instead of previously type 'bool'.  This
is done to allow future extension of the bulk alloc API.

A future extension could be to allow SLUB to stop at a page boundary, when
specified by a flag, and then return the number of objects.

The advantage of this approach, would make it easier to make bulk alloc
run without local IRQs disabled.  With an approach of cmpxchg "stealing"
the entire c->freelist or page->freelist.  To avoid overshooting we would
stop processing at a slab-page boundary.  Else we always end up returning
some objects at the cost of another cmpxchg.

To keep compatible with future users of this API linking against an older
kernel when using the new flag, we need to return the number of allocated
objects with this API change.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Vladimir Davydov <vdavydov@virtuozzo.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoslub: add missing kmem cgroup support to kmem_cache_free_bulk
Jesper Dangaard Brouer [Fri, 20 Nov 2015 23:57:55 +0000 (15:57 -0800)]
slub: add missing kmem cgroup support to kmem_cache_free_bulk

Initial implementation missed support for kmem cgroup support in
kmem_cache_free_bulk() call, add this.

If CONFIG_MEMCG_KMEM is not enabled, the compiler should be smart enough
to not add any asm code.

Incoming bulk free objects can belong to different kmem cgroups, and
object free call can happen at a later point outside memcg context.  Thus,
we need to keep the orig kmem_cache, to correctly verify if a memcg object
match against its "root_cache" (s->memcg_params.root_cache).

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Reviewed-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoslub: fix kmem cgroup bug in kmem_cache_alloc_bulk
Jesper Dangaard Brouer [Fri, 20 Nov 2015 23:57:52 +0000 (15:57 -0800)]
slub: fix kmem cgroup bug in kmem_cache_alloc_bulk

The call slab_pre_alloc_hook() interacts with kmemgc and is not allowed to
be called several times inside the bulk alloc for loop, due to the call to
memcg_kmem_get_cache().

This would result in hitting the VM_BUG_ON in __memcg_kmem_get_cache.

As suggested by Vladimir Davydov, change slab_post_alloc_hook() to be able
to handle an array of objects.

A subtle detail is, loop iterator "i" in slab_post_alloc_hook() must have
same type (size_t) as size argument.  This helps the compiler to easier
realize that it can remove the loop, when all debug statements inside loop
evaluates to nothing.  Note, this is only an issue because the kernel is
compiled with GCC option: -fno-strict-overflow

In slab_alloc_node() the compiler inlines and optimizes the invocation of
slab_post_alloc_hook(s, flags, 1, &object) by removing the loop and access
object directly.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Reported-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Suggested-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Reviewed-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoslub: optimize bulk slowpath free by detached freelist
Jesper Dangaard Brouer [Fri, 20 Nov 2015 23:57:49 +0000 (15:57 -0800)]
slub: optimize bulk slowpath free by detached freelist

This change focus on improving the speed of object freeing in the
"slowpath" of kmem_cache_free_bulk.

The calls slab_free (fastpath) and __slab_free (slowpath) have been
extended with support for bulk free, which amortize the overhead of
the (locked) cmpxchg_double.

To use the new bulking feature, we build what I call a detached
freelist.  The detached freelist takes advantage of three properties:

 1) the free function call owns the object that is about to be freed,
    thus writing into this memory is synchronization-free.

 2) many freelist's can co-exist side-by-side in the same slab-page
    each with a separate head pointer.

 3) it is the visibility of the head pointer that needs synchronization.

Given these properties, the brilliant part is that the detached
freelist can be constructed without any need for synchronization.  The
freelist is constructed directly in the page objects, without any
synchronization needed.  The detached freelist is allocated on the
stack of the function call kmem_cache_free_bulk.  Thus, the freelist
head pointer is not visible to other CPUs.

All objects in a SLUB freelist must belong to the same slab-page.
Thus, constructing the detached freelist is about matching objects
that belong to the same slab-page.  The bulk free array is scanned is
a progressive manor with a limited look-ahead facility.

Kmem debug support is handled in call of slab_free().

Notice kmem_cache_free_bulk no longer need to disable IRQs. This
only slowed down single free bulk with approx 3 cycles.

Performance data:
 Benchmarked[1] obj size 256 bytes on CPU i7-4790K @ 4.00GHz

SLUB fastpath single object quick reuse: 47 cycles(tsc) 11.931 ns

To get stable and comparable numbers, the kernel have been booted with
"slab_merge" (this also improve performance for larger bulk sizes).

Performance data, compared against fallback bulking:

bulk -  fallback bulk            - improvement with this patch
   1 -  62 cycles(tsc) 15.662 ns - 49 cycles(tsc) 12.407 ns- improved 21.0%
   2 -  55 cycles(tsc) 13.935 ns - 30 cycles(tsc) 7.506 ns - improved 45.5%
   3 -  53 cycles(tsc) 13.341 ns - 23 cycles(tsc) 5.865 ns - improved 56.6%
   4 -  52 cycles(tsc) 13.081 ns - 20 cycles(tsc) 5.048 ns - improved 61.5%
   8 -  50 cycles(tsc) 12.627 ns - 18 cycles(tsc) 4.659 ns - improved 64.0%
  16 -  49 cycles(tsc) 12.412 ns - 17 cycles(tsc) 4.495 ns - improved 65.3%
  30 -  49 cycles(tsc) 12.484 ns - 18 cycles(tsc) 4.533 ns - improved 63.3%
  32 -  50 cycles(tsc) 12.627 ns - 18 cycles(tsc) 4.707 ns - improved 64.0%
  34 -  96 cycles(tsc) 24.243 ns - 23 cycles(tsc) 5.976 ns - improved 76.0%
  48 -  83 cycles(tsc) 20.818 ns - 21 cycles(tsc) 5.329 ns - improved 74.7%
  64 -  74 cycles(tsc) 18.700 ns - 20 cycles(tsc) 5.127 ns - improved 73.0%
 128 -  90 cycles(tsc) 22.734 ns - 27 cycles(tsc) 6.833 ns - improved 70.0%
 158 -  99 cycles(tsc) 24.776 ns - 30 cycles(tsc) 7.583 ns - improved 69.7%
 250 - 104 cycles(tsc) 26.089 ns - 37 cycles(tsc) 9.280 ns - improved 64.4%

Performance data, compared current in-kernel bulking:

bulk - curr in-kernel  - improvement with this patch
   1 -  46 cycles(tsc) - 49 cycles(tsc) - improved (cycles:-3) -6.5%
   2 -  27 cycles(tsc) - 30 cycles(tsc) - improved (cycles:-3) -11.1%
   3 -  21 cycles(tsc) - 23 cycles(tsc) - improved (cycles:-2) -9.5%
   4 -  18 cycles(tsc) - 20 cycles(tsc) - improved (cycles:-2) -11.1%
   8 -  17 cycles(tsc) - 18 cycles(tsc) - improved (cycles:-1) -5.9%
  16 -  18 cycles(tsc) - 17 cycles(tsc) - improved (cycles: 1)  5.6%
  30 -  18 cycles(tsc) - 18 cycles(tsc) - improved (cycles: 0)  0.0%
  32 -  18 cycles(tsc) - 18 cycles(tsc) - improved (cycles: 0)  0.0%
  34 -  78 cycles(tsc) - 23 cycles(tsc) - improved (cycles:55) 70.5%
  48 -  60 cycles(tsc) - 21 cycles(tsc) - improved (cycles:39) 65.0%
  64 -  49 cycles(tsc) - 20 cycles(tsc) - improved (cycles:29) 59.2%
 128 -  69 cycles(tsc) - 27 cycles(tsc) - improved (cycles:42) 60.9%
 158 -  79 cycles(tsc) - 30 cycles(tsc) - improved (cycles:49) 62.0%
 250 -  86 cycles(tsc) - 37 cycles(tsc) - improved (cycles:49) 57.0%

Performance with normal SLUB merging is significantly slower for
larger bulking.  This is believed to (primarily) be an effect of not
having to share the per-CPU data-structures, as tuning per-CPU size
can achieve similar performance.

bulk - slab_nomerge   -  normal SLUB merge
   1 -  49 cycles(tsc) - 49 cycles(tsc) - merge slower with cycles:0
   2 -  30 cycles(tsc) - 30 cycles(tsc) - merge slower with cycles:0
   3 -  23 cycles(tsc) - 23 cycles(tsc) - merge slower with cycles:0
   4 -  20 cycles(tsc) - 20 cycles(tsc) - merge slower with cycles:0
   8 -  18 cycles(tsc) - 18 cycles(tsc) - merge slower with cycles:0
  16 -  17 cycles(tsc) - 17 cycles(tsc) - merge slower with cycles:0
  30 -  18 cycles(tsc) - 23 cycles(tsc) - merge slower with cycles:5
  32 -  18 cycles(tsc) - 22 cycles(tsc) - merge slower with cycles:4
  34 -  23 cycles(tsc) - 22 cycles(tsc) - merge slower with cycles:-1
  48 -  21 cycles(tsc) - 22 cycles(tsc) - merge slower with cycles:1
  64 -  20 cycles(tsc) - 48 cycles(tsc) - merge slower with cycles:28
 128 -  27 cycles(tsc) - 57 cycles(tsc) - merge slower with cycles:30
 158 -  30 cycles(tsc) - 59 cycles(tsc) - merge slower with cycles:29
 250 -  37 cycles(tsc) - 56 cycles(tsc) - merge slower with cycles:19

Joint work with Alexander Duyck.

[1] https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/mm/slab_bulk_test01.c

[akpm@linux-foundation.org: BUG_ON -> WARN_ON;return]
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoslub: support for bulk free with SLUB freelists
Jesper Dangaard Brouer [Fri, 20 Nov 2015 23:57:46 +0000 (15:57 -0800)]
slub: support for bulk free with SLUB freelists

Make it possible to free a freelist with several objects by adjusting API
of slab_free() and __slab_free() to have head, tail and an objects counter
(cnt).

Tail being NULL indicate single object free of head object.  This allow
compiler inline constant propagation in slab_free() and
slab_free_freelist_hook() to avoid adding any overhead in case of single
object free.

This allows a freelist with several objects (all within the same
slab-page) to be free'ed using a single locked cmpxchg_double in
__slab_free() and with an unlocked cmpxchg_double in slab_free().

Object debugging on the free path is also extended to handle these
freelists.  When CONFIG_SLUB_DEBUG is enabled it will also detect if
objects don't belong to the same slab-page.

These changes are needed for the next patch to bulk free the detached
freelists it introduces and constructs.

Micro benchmarking showed no performance reduction due to this change,
when debugging is turned off (compiled with CONFIG_SLUB_DEBUG).

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoparisc: Map kernel text and data on huge pages
Helge Deller [Sat, 21 Nov 2015 23:07:44 +0000 (00:07 +0100)]
parisc: Map kernel text and data on huge pages

Adjust the linker script and map_pages() to map kernel text and data on
physical 1MB huge/large pages.

Signed-off-by: Helge Deller <deller@gmx.de>