Commit 694fc88c ("powerpc/string: Implement optimized
memset variants") added memset16(), memset32() and memset64()
for the 64 bits PPC.

On 32 bits, memset64() is not relevant, and as shown below,
the generic version of memset32() gives a good code, so only
memset16() is candidate for an optimised version.

000009c0 <memset32>:
 9c0:   2c 05 00 00     cmpwi   r5,0
 9c4:   39 23 ff fc     addi    r9,r3,-4
 9c8:   4d 82 00 20     beqlr
 9cc:   7c a9 03 a6     mtctr   r5
 9d0:   94 89 00 04     stwu    r4,4(r9)
 9d4:   42 00 ff fc     bdnz    9d0 <memset32+0x10>
 9d8:   4e 80 00 20     blr

The last part of memset() handling the not 4-bytes multiples
operates on bytes, making it unsuitable for handling word without
modification. As it would increase memset() complexity, it is
better to implement memset16() from scratch. In addition it
has the advantage of allowing a more optimised memset16() than what
we would have by using the memset() function.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Michael Ellerman reported that emulate_loadstore() was trying to
access element 32 of regs->gpr[], which doesn't exist, when
emulating a string store instruction.  This is because the string
load and store instructions (lswi, lswx, stswi and stswx) are
defined to wrap around from register 31 to register 0 if the number
of bytes being loaded or stored is sufficiently large.  This wrapping
was not implemented in the emulation code.  To fix it, we mask the
register number after incrementing it.
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Fixes: c9f6f4ed ("powerpc: Implement emulation of string loads and stores")
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds emulation for the lfiwax, lfiwzx and stfiwx instructions.
This necessitated adding a new flag to indicate whether a floating
point or an integer conversion was needed for LOAD_FP and STORE_FP,
so this moves the size field in op->type up 4 bits.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This replaces almost all of the instruction emulation code in
fix_alignment() with calls to analyse_instr(), emulate_loadstore()
and emulate_dcbz().  The only emulation code left is the SPE
emulation code; analyse_instr() etc. do not handle SPE instructions
at present.

One result of this is that we can now handle alignment faults on
all the new VSX load and store instructions that were added in POWER9.
VSX loads/stores will take alignment faults for unaligned accesses
to cache-inhibited memory.

Another effect is that we no longer rely on the DAR and DSISR values
set by the processor.

With this, we now need to include the instruction emulation code
unconditionally.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This moves the parts of emulate_step() that deal with emulating
load and store instructions into a new function called
emulate_loadstore().  This is to make it possible to reuse this
code in the alignment handler.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds code to the load and store emulation code to byte-swap
the data appropriately when the process being emulated is set to
the opposite endianness to that of the kernel.

This also enables the emulation for the multiple-register loads
and stores (lmw, stmw, lswi, stswi, lswx, stswx) to work for
little-endian.  In little-endian mode, the partial word at the
end of a transfer for lsw*/stsw* (when the byte count is not a
multiple of 4) is loaded/stored at the least-significant end of
the register.  Additionally, this fixes a bug in the previous
code in that it could call read_mem/write_mem with a byte count
that was not 1, 2, 4 or 8.

Note that this only works correctly on processors with "true"
little-endian mode, such as IBM POWER processors from POWER6 on, not
the so-called "PowerPC" little-endian mode that uses address swizzling
as implemented on the old 32-bit 603, 604, 740/750, 74xx CPUs.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds code to the instruction emulation code to set regs->dar
to the address of any memory access that fails.  This address is
not necessarily the same as the effective address of the instruction,
because if the memory access is unaligned, it might cross a page
boundary and fault on the second page.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds code to analyse_instr() and emulate_step() to understand the
dcbz (data cache block zero) instruction.  The emulate_dcbz() function
is made public so it can be used by the alignment handler in future.
(The apparently unnecessary cropping of the address to 32 bits is
there because it will be needed in that situation.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds lfdp[x] and stfdp[x] to the set of instructions that
analyse_instr() and emulate_step() understand.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds code to analyse_instr() and emulate_step() to handle the
vector element loads and stores:

lvebx, lvehx, lvewx, stvebx, stvehx, stvewx.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

At present, the analyse_instr/emulate_step code checks for the
relevant MSR_FP/VEC/VSX bit being set when a FP/VMX/VSX load
or store is decoded, but doesn't recheck the bit before reading or
writing the relevant FP/VMX/VSX register in emulate_step().

Since we don't have preemption disabled, it is possible that we get
preempted between checking the MSR bit and doing the register access.
If that happened, then the registers would have been saved to the
thread_struct for the current process.  Accesses to the CPU registers
would then potentially read stale values, or write values that would
never be seen by the user process.

Another way that the registers can become non-live is if a page
fault occurs when accessing user memory, and the page fault code
calls a copy routine that wants to use the VMX or VSX registers.

To fix this, the code for all the FP/VMX/VSX loads gets restructured
so that it forms an image in a local variable of the desired register
contents, then disables preemption, checks the MSR bit and either
sets the CPU register or writes the value to the thread struct.
Similarly, the code for stores checks the MSR bit, copies either the
CPU register or the thread struct to a local variable, then reenables
preemption and then copies the register image to memory.

If the instruction being emulated is in the kernel, then we must not
use the register values in the thread_struct.  In this case, if the
relevant MSR enable bit is not set, then emulate_step refuses to
emulate the instruction.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

At the moment, emulation of loads and stores of up to 8 bytes to
unaligned addresses on a little-endian system uses a sequence of
single-byte loads or stores to memory.  This is rather inefficient,
and the code is hard to follow because it has many ifdefs.
In addition, the Power ISA has requirements on how unaligned accesses
are performed, which are not met by doing all accesses as
sequences of single-byte accesses.

Emulation of VSX loads and stores uses __copy_{to,from}_user,
which means the emulation code has no control on the size of
accesses.

To simplify this, we add new copy_mem_in() and copy_mem_out()
functions for accessing memory.  These use a sequence of the largest
possible aligned accesses, up to 8 bytes (or 4 on 32-bit systems),
to copy memory between a local buffer and user memory.  We then
rewrite {read,write}_mem_unaligned and the VSX load/store
emulation using these new functions.

These new functions also simplify the code in do_fp_load() and
do_fp_store() for the unaligned cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The addpcis instruction puts the sum of the next instruction address
plus a constant into a register.  Since the result depends on the
address of the instruction, it will give an incorrect result if it
is single-stepped out of line, which is what the *probes subsystem
will currently do if a probe is placed on an addpcis instruction.
This fixes the problem by adding emulation of it to analyse_instr().
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The architecture shows the least-significant bit of the instruction
word as reserved for the popcnt[bwd], prty[wd] and bpermd
instructions, that is, these instructions never update CR0.
Therefore this changes the emulation of these instructions to
skip the CR0 update.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The case added for the isel instruction was added inside a switch
statement which uses the 10-bit minor opcode field in the 0x7fe
bits of the instruction word.  However, for the isel instruction,
the minor opcode field is only the 0x3e bits, and the 0x7c0 bits
are used for the "BC" field, which indicates which CR bit to use
to select the result.

Therefore, for the isel emulation to work correctly when BC != 0,
we need to match on ((instr >> 1) & 0x1f) == 15).  To do this, we
pull the isel case out of the switch statement and put it in an
if statement of its own.

Fixes: e27f71e5 ("powerpc/lib/sstep: Add isel instruction emulation")
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

When a 64-bit processor is executing in 32-bit mode, the update forms
of load and store instructions are required by the architecture to
write the full 64-bit effective address into the RA register, though
only the bottom 32 bits are used to address memory.  Currently,
the instruction emulation code writes the truncated address to the
RA register.  This fixes it by keeping the full 64-bit EA in the
instruction_op structure, truncating the address in emulate_step()
where it is used to address memory, rather than in the address
computations in analyse_instr().
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This extends the instruction emulation infrastructure in sstep.c to
handle all the load and store instructions defined in the Power ISA
v3.0, except for the atomic memory operations, ldmx (which was never
implemented), lfdp/stfdp, and the vector element load/stores.

The instructions added are:

Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx.,
lq, stq.

VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll,
lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx,
lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x,
stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv.

These instructions are handled both in the analyse_instr phase and in
the emulate_step phase.

The code for lxvd2ux and stxvd2ux has been taken out, as those
instructions were never implemented in any processor and have been
taken out of the architecture, and their opcodes have been reused for
other instructions in POWER9 (lxvb16x and stxvb16x).

The emulation for the VSX loads and stores uses helper functions
which don't access registers or memory directly, which can hopefully
be reused by KVM later.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This removes the checks for the FP/VMX/VSX enable bits in the MSR
from analyse_instr() and adds them to emulate_step() instead.

The reason for this is that we may want to use analyse_instr() in
a situation where the FP/VMX/VSX register values are stored in the
current thread_struct and the FP/VMX/VSX enable bits in the MSR
image in the pt_regs are zero.  Since analyse_instr() doesn't make
any changes to register state, it is reasonable for it to indicate
what the effect of an instruction would be even though the relevant
enable bit is off.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The analyse_instr function currently doesn't just work out what an
instruction does, it also executes those instructions whose effect
is only to update CPU registers that are stored in struct pt_regs.
This is undesirable because optprobes uses analyse_instr to work out
if an instruction could be successfully emulated in future.

This changes analyse_instr so it doesn't modify *regs; instead it
stores information in the instruction_op structure to indicate what
registers (GPRs, CR, XER, LR) would be set and what value they would
be set to.  A companion function called emulate_update_regs() can
then use that information to update a pt_regs struct appropriately.

As a minor cleanup, this replaces inline asm using the cntlzw and
cntlzd instructions with calls to __builtin_clz() and __builtin_clzl().
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The instruction code for xxlor that commit 0016a4cf ("powerpc:
Emulate most Book I instructions in emulate_step()", 2010-06-15)
added is actually the code for xxlnor.  It is used in get_vsr()
and put_vsr() and the effect of the error is that if emulate_step
is used to emulate a VSX load or store from any register other
than vsr0, the bitwise complement of the correct value will be
loaded or stored.  This corrects the error.

Fixes: 0016a4cf ("powerpc: Emulate most Book I instructions in emulate_step()")
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Anton noticed that if we fault part way through emulating an unaligned
instruction, we don't update the DAR to reflect that.

The DAR value is eventually reported back to userspace as the address
in the SEGV signal, and if userspace is using that value to demand
fault then it can be confused by us not setting the value correctly.

This patch is ugly as hell, but is intended to be the minimal fix and
back ports easily.

Cc: stable@vger.kernel.org
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Paul Mackerras <paulus@ozlabs.org>

Check if an LMB is assigned before attempting to call dlpar_acquire_drc
in order to avoid any unnecessary rtas calls. This substantially
reduces the running time of memory hot add on lpars with large amounts
of memory.

[mpe: We need to explicitly set rc to 0 in the success case, otherwise
 the compiler might think we use rc without initialising it.]

Fixes: c21f515c ("powerpc/pseries: Make the acquire/release of the drc for memory a seperate step")
Cc: stable@vger.kernel.org # v4.11+
Signed-off-by: John Allen <jallen@linux.vnet.ibm.com>
Reviewed-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

struct platform_suspend_ops are not supposed to change at runtime.
Functions suspend_set_ops working with const platform_suspend_ops. So
mark the non-const structs as const.
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

In previous generations of Power processors each core had a private L2
cache. The Power 9 processor has a slightly different design where the
L2 cache is shared among pairs of cores rather than being completely
private.

Making the scheduler aware of this cache sharing allows the scheduler to
make better migration decisions. For example, if two CPU heavy tasks
share a core then one task can be migrated to the paired core to improve
throughput. Under the existing three level topology the task could be
migrated to any core on the same chip, while with the new topology it
would be preferentially migrated to the paired core so it remains
cache-hot.
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

We want to add an extra level to the CPU scheduler topology to account
for cores which share a cache. To do this we need to build a cpumask
for each CPU that indicates which CPUs share this cache to use as an
input to the scheduler.
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The CPU scheduler topology is constructed from a number of per-cpu
cpumasks which describe which sets of logical CPUs are related in some
fashion. Current code that handles constructing these masks when CPUs
are hot(un)plugged can be simplified a bit by exploiting the fact that
the scheduler requires higher levels of the toplogy (e.g package level
groupings) to be supersets of the lower levels (e.g.  threas in a core).
This patch reworks the cpumask construction to be simpler and easier to
extend with extra topology levels.
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
[mpe: Fix CONFIG_HOTPLUG_CPU=n build]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

When building the CPU scheduler topology the kernel uses the ibm,chipid
property from the devicetree to group logical CPUs. Currently the DT
search for this property is open-coded in smp.c and this functionality
is a duplication of what's in cpu_to_chip_id() already. This patch
removes the existing search in favor of that.

It's worth mentioning that the semantics of the search are different
in cpu_to_chip_id(). When there is no ibm,chipid in the CPUs node it
will also search /cpus and / for the property, but this should not
effect the output topology.
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

cxl keeps a driver use count, which is used with the hash memory model
on p8 to know when to upgrade local TLBIs to global and to trigger
callbacks to manage the MMU for PSL8.

If a process opens a context and closes without attaching or fails the
attachment, the driver use count is never decremented. As a
consequence, TLB invalidations remain global, even if there are no
active cxl contexts.

We should increment the driver use count when the process is attaching
to the cxl adapter, and not on open. It's not needed before the
adapter starts using the context and the use count is decremented on
the detach path, so it makes more sense.

It affects only the user api. The kernel api is already doing The
Right Thing.
Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Cc: stable@vger.kernel.org # v4.2+
Fixes: 7bb5d91a ("cxl: Rework context lifetimes")
Acked-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Currently these tests won't build with a `--enable-default-pie`
compiler as they require r30 to be clobbered. This gives
an error:
  ptrace-tm-spd-gpr.c:41:2: error: PIC register clobbered by 'r30' in 'asm'

This forces these tests to be built no-pie.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

mpsc.c and mpc52xx-psc.c are platform-specific serial drivers, and
should be compiled for the respective platforms only.
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Torsten Duwe <duwe@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

.llong is an undocumented PPC specific directive. The generic
equivalent is .quad, but even better (because it's self describing) is
.8byte.

Convert all .llong directives to .8byte.
Signed-off-by: Tobin C. Harding <me@tobin.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Enable 64K page size and THP. I use ppc64le_defconfig when I need
a single config across guest and host, but having 4K page size
as default is not what I expect. I could move these over to
server.config and merge if ppc64_defconfig is meant for systems
that use 4k pages by default.
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

drivers/watchdog/wdrtas.c is of interest of linuxppc maintainers.
Signed-off-by: Murilo Opsfelder Araujo <mopsfelder@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Most (all?) distros turn these on, so it makes sense to enable them
for testing coverage, and they're also useful for developers.
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
Acked-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
[mpe: Reword change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Add support for printing the PIDR/TIDR for ISA 300 and PSSCR and PTCR
in ISA 3.0 hypervisor mode.

SPRN_PSSCR_PR is the privileged mode access and is used when we are
not in hypervisor mode.
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This patch adds support to xmon for dumping the AMR, UAMOR, AMOR and
IAMR SPRs based on their supported ISA revisions.
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

ISA 3.0 defines hypervisor decrementer to be 64 bits in length.
This patch extends the print format for to be 64 bits.
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Remove unneeded variables and assignments.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

When we map memory at boot we print out the ranges of real addresses
that we mapped and the page size that was used.

Currently it's a bit ugly:

  Mapped range 0x0 - 0x2000000000 with 0x40000000
  Mapped range 0x200000000000 - 0x202000000000 with 0x40000000

Pad the addresses so they line up, and print the page size using
actual units, eg:

  Mapped 0x0000000000000000-0x0000000001200000 with 64.0 KiB pages
  Mapped 0x0000000001200000-0x0000000040000000 with 2.00 MiB pages
  Mapped 0x0000000040000000-0x0000000100000000 with 1.00 GiB pages
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Make the printks look a bit nicer by adding a prefix.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>