platform/kernel/linux-rpi.git
2 years agopowerpc/64: use 32-bit immediate for STACK_FRAME_REGS_MARKER
Nicholas Piggin [Mon, 26 Sep 2022 03:40:53 +0000 (13:40 +1000)]
powerpc/64: use 32-bit immediate for STACK_FRAME_REGS_MARKER

Using a 32-bit constant for this marker allows it to be loaded with
two ALU instructions, like 32-bit. This avoids a TOC entry and a
TOC load that depends on the r2 value that has just been loaded from
the PACA.

This changes the value for 32-bit as well, so both have the same
value in the low 4 bytes and 64-bit has 0 in the top bytes.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926034057.2360083-2-npiggin@gmail.com
2 years agopowerpc/64s: POWER10 CPU Kconfig build option
Nicholas Piggin [Fri, 23 Sep 2022 03:30:04 +0000 (13:30 +1000)]
powerpc/64s: POWER10 CPU Kconfig build option

This adds basic POWER10_CPU option, which builds with -mcpu=power10.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220923033004.536127-1-npiggin@gmail.com
2 years agopowerpc/pseries: Move vas_migration_handler early during migration
Haren Myneni [Thu, 22 Sep 2022 08:27:07 +0000 (01:27 -0700)]
powerpc/pseries: Move vas_migration_handler early during migration

When the migration is initiated, the hypervisor changes VAS
mappings as part of pre-migration event. Then the OS gets the
migration event which closes all VAS windows before the migration
starts. NX generates continuous faults until windows are closed
and the user space can not differentiate these NX faults coming
from the actual migration. So to reduce this time window, close
VAS windows first in pseries_migrate_partition().

Signed-off-by: Haren Myneni <haren@linux.ibm.com>
Reviewed-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d8efade91dda831c9ed4abb226dab627da594c5f.camel@linux.ibm.com
2 years agopowerpc/64/irq: tidy soft-masked irq replay and improve documentation
Nicholas Piggin [Mon, 26 Sep 2022 05:43:05 +0000 (15:43 +1000)]
powerpc/64/irq: tidy soft-masked irq replay and improve documentation

irq replay is quite complicated because of softirq processing which
itself enables and disables irqs. Several considerations need to be
accounted for due to this, and they are not clearly documented.

Refactor the irq replay code a bit to tidy and deduplicate some common
functions. Add comments, debug checks.

This has a minor functional change that irq tracing enable/disable is
done after each interrupt replayed, rather than after a batch. It also
re-sets state to IRQS_ALL_DISABLED after an interrupt, which doesn't
matter much because interrupts are hard disabled at this point, but it
is more consistent with how interrupt handlers are called.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926054305.2671436-8-npiggin@gmail.com
2 years agopowerpc/64/interrupt: avoid BUG/WARN recursion in interrupt entry
Nicholas Piggin [Mon, 26 Sep 2022 05:43:04 +0000 (15:43 +1000)]
powerpc/64/interrupt: avoid BUG/WARN recursion in interrupt entry

BUG/WARN are handled with a program interrupt which can turn into an
infinite recursion when there are bugs in interrupt handler entry
(which can be irritated by bugs in other parts of the code).

There is one feeble attempt to avoid this recursion, but it misses
several cases. Make a tidier macro for this and switch most bugs in
the interrupt entry wrapper over to use it.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926054305.2671436-7-npiggin@gmail.com
2 years agopowerpc/64s/interrupt: masked handler debug check for previous hard disable
Nicholas Piggin [Mon, 26 Sep 2022 05:43:03 +0000 (15:43 +1000)]
powerpc/64s/interrupt: masked handler debug check for previous hard disable

Prior changes eliminated cases of masked PACA_IRQ_MUST_HARD_MASK
interrupts that re-fire due to MSR[EE] being enabled while they are
pending. Add a debug check in the masked interrupt handler to catch
if this occurs.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926054305.2671436-6-npiggin@gmail.com
2 years agopowerpc/64s: Fix irq state management in runlatch functions
Nicholas Piggin [Mon, 26 Sep 2022 05:43:02 +0000 (15:43 +1000)]
powerpc/64s: Fix irq state management in runlatch functions

When irqs are soft-disabled, MSR[EE] is volatile and can change from
1 to 0 asynchronously (if a PACA_IRQ_MUST_HARD_MASK interrupt hits).
So it can not be used to check hard IRQ enabled status, except to
confirm it is disabled.

ppc64_runlatch_on/off functions use MSR this way to decide whether to
re-enable MSR[EE] after disabling it, which leads to MSR[EE] being
enabled when it shouldn't be (when a PACA_IRQ_MUST_HARD_MASK had
disabled it between reading the MSR and clearing EE).

This has been tolerated in the kernel previously, and it doesn't seem
to cause a problem, but it is unexpected and may trip warnings or cause
other problems as we tighten up this state management. Fix this by only
re-enabling if PACA_IRQ_HARD_DIS is clear.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926054305.2671436-5-npiggin@gmail.com
2 years agopowerpc/64/interrupt: Fix return to masked context after hard-mask irq becomes pending
Nicholas Piggin [Mon, 26 Sep 2022 05:43:01 +0000 (15:43 +1000)]
powerpc/64/interrupt: Fix return to masked context after hard-mask irq becomes pending

If a synchronous interrupt (e.g., hash fault) is taken inside an
irqs-disabled region which has MSR[EE]=1, then an asynchronous interrupt
that is PACA_IRQ_MUST_HARD_MASK (e.g., PMI) is taken inside the
synchronous interrupt handler, then the synchronous interrupt will
return with MSR[EE]=1 and the asynchronous interrupt fires again.

If the asynchronous interrupt is a PMI and the original context does not
have PMIs disabled (only Linux IRQs), the asynchronous interrupt will
fire despite having the PMI marked soft pending. This can confuse the
perf code and cause warnings.

This patch changes the interrupt return so that irqs-disabled MSR[EE]=1
contexts will be returned to with MSR[EE]=0 if a PACA_IRQ_MUST_HARD_MASK
interrupt has become pending in the meantime.

The longer explanation for what happens:
1. local_irq_disable()
2. Hash fault interrupt fires, do_hash_fault handler runs
3. interrupt_enter_prepare() sets IRQS_ALL_DISABLED
4. interrupt_enter_prepare() sets MSR[EE]=1
5. PMU interrupt fires, masked handler runs
6. Masked handler marks PMI pending
7. Masked handler returns with PACA_IRQ_HARD_DIS set, MSR[EE]=0
8. do_hash_fault interrupt return handler runs
9. interrupt_exit_kernel_prepare() clears PACA_IRQ_HARD_DIS
10. interrupt returns with MSR[EE]=1
11. PMU interrupt fires, perf handler runs

Fixes: 4423eb5ae32e ("powerpc/64/interrupt: make normal synchronous interrupts enable MSR[EE] if possible")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926054305.2671436-4-npiggin@gmail.com
2 years agopowerpc/64: mark irqs hard disabled in boot paca
Nicholas Piggin [Mon, 26 Sep 2022 05:43:00 +0000 (15:43 +1000)]
powerpc/64: mark irqs hard disabled in boot paca

This prevents interrupts in early boot (e.g., program check) from
enabling MSR[EE], potentially causing endian mismatch or other
crashes when reporting early boot traps.

Fixes: 4423eb5ae32ec ("powerpc/64/interrupt: make normal synchronous interrupts enable MSR[EE] if possible")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926054305.2671436-3-npiggin@gmail.com
2 years agopowerpc/64/interrupt: Fix false warning in context tracking due to idle state
Nicholas Piggin [Mon, 26 Sep 2022 05:42:59 +0000 (15:42 +1000)]
powerpc/64/interrupt: Fix false warning in context tracking due to idle state

Commit 171476775d32 ("context_tracking: Convert state to atomic_t")
added a CONTEXT_IDLE state which can be encountered by interrupts from
kernel mode in the idle thread, causing a false positive warning.

Fixes: 171476775d32 ("context_tracking: Convert state to atomic_t")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926054305.2671436-2-npiggin@gmail.com
2 years agopowerpc/64s: Enable KFENCE on book3s64
Nicholas Miehlbradt [Mon, 26 Sep 2022 07:57:26 +0000 (07:57 +0000)]
powerpc/64s: Enable KFENCE on book3s64

KFENCE support was added for ppc32 in commit 90cbac0e995d
("powerpc: Enable KFENCE for PPC32").
Enable KFENCE on ppc64 architecture with hash and radix MMUs.
It uses the same mechanism as debug pagealloc to
protect/unprotect pages. All KFENCE kunit tests pass on both
MMUs.

KFENCE memory is initially allocated using memblock but is
later marked as SLAB allocated. This necessitates the change
to __pud_free to ensure that the KFENCE pages are freed
appropriately.

Based on previous work by Christophe Leroy and Jordan Niethe.

Signed-off-by: Nicholas Miehlbradt <nicholas@linux.ibm.com>
Reviewed-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926075726.2846-4-nicholas@linux.ibm.com
2 years agopowerpc/64s: Allow double call of kernel_[un]map_linear_page()
Christophe Leroy [Mon, 26 Sep 2022 07:57:25 +0000 (07:57 +0000)]
powerpc/64s: Allow double call of kernel_[un]map_linear_page()

If the page is already mapped resp. already unmapped, bail out.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Nicholas Miehlbradt <nicholas@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926075726.2846-3-nicholas@linux.ibm.com
2 years agopowerpc/64s: Remove unneeded #ifdef CONFIG_DEBUG_PAGEALLOC in hash_utils
Christophe Leroy [Mon, 26 Sep 2022 07:57:24 +0000 (07:57 +0000)]
powerpc/64s: Remove unneeded #ifdef CONFIG_DEBUG_PAGEALLOC in hash_utils

debug_pagealloc_enabled() is always defined and constant folds to
'false' when CONFIG_DEBUG_PAGEALLOC is not enabled.

Remove the #ifdefs, the code and associated static variables will
be optimised out by the compiler when CONFIG_DEBUG_PAGEALLOC is
not defined.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Nicholas Miehlbradt <nicholas@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926075726.2846-2-nicholas@linux.ibm.com
2 years agopowerpc/64s: Add DEBUG_PAGEALLOC for radix
Nicholas Miehlbradt [Mon, 26 Sep 2022 07:57:23 +0000 (07:57 +0000)]
powerpc/64s: Add DEBUG_PAGEALLOC for radix

There is support for DEBUG_PAGEALLOC on hash but not on radix.
Add support on radix.

Signed-off-by: Nicholas Miehlbradt <nicholas@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926075726.2846-1-nicholas@linux.ibm.com
2 years agopowerpc/64s: update cpu selection options
Nicholas Piggin [Wed, 21 Sep 2022 01:41:03 +0000 (11:41 +1000)]
powerpc/64s: update cpu selection options

Update the 64s GENERIC_CPU option. POWER4 support has been dropped, so
make that clear in the option name. The POWER5_CPU option is dropped
because it's uncommon, and GENERIC_CPU covers it.

-mtune= before power8 is dropped because the minimum gcc version
supports power8, and tuning is made consistent between big and little
endian.

A 970 option is added for PowerPC 970 / G5 because they still have a
user base, and -mtune=power8 does not generate good code for the 970.

This also updates the ISA versions document to add Power4/Power4+
because I didn't realise Power4+ used 2.01.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921014103.587954-2-npiggin@gmail.com
2 years agopowerpc/64s: Fix GENERIC_CPU build flags for PPC970 / G5
Nicholas Piggin [Wed, 21 Sep 2022 01:41:02 +0000 (11:41 +1000)]
powerpc/64s: Fix GENERIC_CPU build flags for PPC970 / G5

Big-endian GENERIC_CPU supports 970, but builds with -mcpu=power5.
POWER5 is ISA v2.02 whereas 970 is v2.01 plus Altivec. 2.02 added
the popcntb instruction which a compiler might use.

Use -mcpu=power4.

Fixes: 471d7ff8b51b ("powerpc/64s: Remove POWER4 support")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921014103.587954-1-npiggin@gmail.com
2 years agopowerpc/64s: Make POWER10 and later use pause_short in cpu_relax loops
Nicholas Piggin [Tue, 20 Sep 2022 12:22:59 +0000 (22:22 +1000)]
powerpc/64s: Make POWER10 and later use pause_short in cpu_relax loops

We want to move away from using SMT priority updates for cpu_relax, and
use a 'wait' instruction which is similar to x86. As well as being a
much better fit for what everybody else uses and tests with, priority
nops are stateful which is nasty (interrupts have to consider they might
be taken at a different priority), and they're expensive to execute,
similar to a mtSPR which can effect other threads in the pipe.

This has shown to give results that are less affected by code alignment
on benchmarks that cause a lot of spin waiting (e.g., rwsem contention
on unixbench filesystem benchmarks) on POWER10.

QEMU TCG only supports this instruction correctly since v7.1, versions
without the fix may cause hangs whne running POWER10 CPUs.

Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Fix checkpatch warnings RE the macros]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220920122259.363092-2-npiggin@gmail.com
2 years agopowerpc: add ISA v3.0 / v3.1 wait opcode macro
Nicholas Piggin [Tue, 20 Sep 2022 12:22:58 +0000 (22:22 +1000)]
powerpc: add ISA v3.0 / v3.1 wait opcode macro

The wait instruction encoding changed between ISA v2.07 and ISA v3.0.
In v3.1 the instruction gained a new field.

Update the PPC_WAIT macro to the current encoding. Rename the older
incompatible one with a _v203 suffix as it was introduced in v2.03
(the WC field was introduced in v2.07 but the kernel only uses WC=0).

Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220920122259.363092-1-npiggin@gmail.com
2 years agopowerpc/time: avoid programming DEC at the start of the timer interrupt
Nicholas Piggin [Fri, 9 Sep 2022 14:24:57 +0000 (00:24 +1000)]
powerpc/time: avoid programming DEC at the start of the timer interrupt

Setting DEC to maximum at the start of the timer interrupt is not
necessary and can be avoided for performance when MSR[EE] is not
enabled during the handler as explained in commit 0faf20a1ad16
("powerpc/64s/interrupt: Don't enable MSR[EE] in irq handlers unless
perf is in use"), where this change was first attempted.

The idea is that the timer interrupt runs with MSR[EE]=0, and at the end
of the interrupt DEC is programmed to the next timer interval, so there
is no need to clear the decrementer exception before then.

When the above commit was merged, that was not quite true. The low res
timer subsystem had some cases in the oneshot timer code where if the
tick was to be stopped and no timers active, the clock device would not
get the ->set_state_oneshot_stopped() call, so DEC would not get
reprogrammed, and this would hang taking continual timer interrupts.

So this was reverted in commit d2b9be1f4af5 ("powerpc/time: Always set
decrementer in timer_interrupt()"), which was a partial revert of the
above commit.

Commit 62c1256d5447 ("timers/nohz: Switch to ONESHOT_STOPPED in the
low-res handler when the tick is stopped") was later merged to fix this
missing case in the timer subsystem, so now the behaviour can be
restored.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220909142457.278032-1-npiggin@gmail.com
2 years agopowerpc: Add support for early debugging via Serial 16550 console
Pali Rohár [Mon, 22 Aug 2022 23:15:01 +0000 (01:15 +0200)]
powerpc: Add support for early debugging via Serial 16550 console

Currently powerpc early debugging contains lot of platform specific
options, but does not support standard UART / serial 16550 console.

Later legacy_serial.c code supports registering UART as early debug console
from device tree but it is not early during booting, but rather later after
machine description code finishes.

So for real early debugging via UART is current code unsuitable.

Add support for new early debugging option CONFIG_PPC_EARLY_DEBUG_16550
which enable Serial 16550 console on address defined by new option
CONFIG_PPC_EARLY_DEBUG_16550_PHYSADDR and by stride by option
CONFIG_PPC_EARLY_DEBUG_16550_STRIDE.

With this change it is possible to debug powerpc machine descriptor code.
For example this early debugging code can print on serial console also
"No suitable machine description found" error which is done before
legacy_serial.c code.

Signed-off-by: Pali Rohár <pali@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220822231501.16827-1-pali@kernel.org
2 years agopowerpc/64/kdump: Limit kdump base to 512MB
Hari Bathini [Mon, 12 Sep 2022 06:50:31 +0000 (12:20 +0530)]
powerpc/64/kdump: Limit kdump base to 512MB

Since commit e641eb03ab2b0 ("powerpc: Fix up the kdump base cap to
128M") memory for kdump kernel has been reserved at an offset of 128MB.
This held up well for a long time before running into boot failure on
LPARs having a lot of cores. Commit 7c5ed82b800d8 ("powerpc: Set
crashkernel offset to mid of RMA region") fixed this boot failure by
moving the offset to mid of RMA region. This change meant the offset is
either 256MB or 512MB on LPARs as ppc64_rma_size was 512MB or 1024MB
owing to commit 103a8542cb35b ("powerpc/book3s64/ radix: Fix boot
failure with large amount of guest memory").

But ppc64_rma_size can be larger as well with newer f/w. So, limit
crashkernel reservation offset to 512MB to avoid running into boot
failures during kdump kernel boot, due to RTAS or other allocation
restrictions.

Also, while here, use SZ_128M instead of opening coding it.

Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
Tested-by: Sachin Sant <sachinp@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220912065031.57416-1-hbathini@linux.ibm.com
2 years agopowerpc: Provide syscall wrapper
Rohan McLure [Wed, 21 Sep 2022 06:56:01 +0000 (16:56 +1000)]
powerpc: Provide syscall wrapper

Implement syscall wrapper as per s390, x86, arm64. When enabled
cause handlers to accept parameters from a stack frame rather than
from user scratch register state. This allows for user registers to be
safely cleared in order to reduce caller influence on speculation
within syscall routine. The wrapper is a macro that emits syscall
handler symbols that call into the target handler, obtaining its
parameters from a struct pt_regs on the stack.

As registers are already saved to the stack prior to calling
system_call_exception, it appears that this function is executed more
efficiently with the new stack-pointer convention than with parameters
passed by registers, avoiding the allocation of a stack frame for this
method. On a 32-bit system, we see >20% performance increases on the
null_syscall microbenchmark, and on a Power 8 the performance gains
amortise the cost of clearing and restoring registers which is
implemented at the end of this series, seeing final result of ~5.6%
performance improvement on null_syscall.

Syscalls are wrapped in this fashion on all platforms except for the
Cell processor as this commit does not provide SPU support. This can be
quickly fixed in a successive patch, but requires spu_sys_callback to
allocate a pt_regs structure to satisfy the wrapped calling convention.

Co-developed-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmai.com>
[mpe: Make incompatible with COMPAT to retain clearing of high bits of args]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-22-rmclure@linux.ibm.com
2 years agopowerpc: Change system_call_exception calling convention
Rohan McLure [Wed, 21 Sep 2022 06:56:00 +0000 (16:56 +1000)]
powerpc: Change system_call_exception calling convention

Change system_call_exception arguments to pass a pointer to a stack
frame container caller state, as well as the original r0, which
determines the number of the syscall. This has been observed to yield
improved performance to passing them by registers, circumventing the
need to allocate a stack frame.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Retain clearing of high bits of args for compat tasks]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-21-rmclure@linux.ibm.com
2 years agopowerpc: Use common syscall handler type
Rohan McLure [Wed, 21 Sep 2022 06:55:58 +0000 (16:55 +1000)]
powerpc: Use common syscall handler type

Cause syscall handlers to be typed as follows when called indirectly
throughout the kernel. This is to allow for better type checking.

typedef long (*syscall_fn)(unsigned long, unsigned long, unsigned long,
                           unsigned long, unsigned long, unsigned long);

Since both 32 and 64-bit abis allow for at least the first six
machine-word length parameters to a function to be passed by registers,
even handlers which admit fewer than six parameters may be viewed as
having the above type.

Coercing syscalls to syscall_fn requires a cast to void* to avoid
-Wcast-function-type.

Fixup comparisons in VDSO to avoid pointer-integer comparison. Introduce
explicit cast on systems with SPUs.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-19-rmclure@linux.ibm.com
2 years agopowerpc: Enable compile-time check for syscall handlers
Rohan McLure [Wed, 21 Sep 2022 06:55:57 +0000 (16:55 +1000)]
powerpc: Enable compile-time check for syscall handlers

The table of syscall handlers and registered compatibility syscall
handlers has in past been produced using assembly, with function
references resolved at link time. This moves link-time errors to
compile-time, by rewriting systbl.S in C, and including the
linux/syscalls.h, linux/compat.h and asm/syscalls.h headers for
prototypes.

Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-18-rmclure@linux.ibm.com
2 years agopowerpc: Include all arch-specific syscall prototypes
Rohan McLure [Wed, 21 Sep 2022 06:55:56 +0000 (16:55 +1000)]
powerpc: Include all arch-specific syscall prototypes

Forward declare all syscall handler prototypes where a generic prototype
is not provided in either linux/syscalls.h or linux/compat.h in
asm/syscalls.h. This is required for compile-time type-checking for
syscall handlers, which is implemented later in this series.

32-bit compatibility syscall handlers are expressed in terms of types in
ppc32.h. Expose this header globally.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Acked-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Use standard include guard naming for syscalls_32.h]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-17-rmclure@linux.ibm.com
2 years agopowerpc: Adopt SYSCALL_DEFINE for arch-specific syscall handlers
Rohan McLure [Wed, 21 Sep 2022 06:55:55 +0000 (16:55 +1000)]
powerpc: Adopt SYSCALL_DEFINE for arch-specific syscall handlers

Arch-specific implementations of syscall handlers are currently used
over generic implementations for the following reasons:

1. Semantics unique to powerpc
2. Compatibility syscalls require 'argument padding' to comply with
   64-bit argument convention in ELF32 abi.
3. Parameter types or order is different in other architectures.

These syscall handlers have been defined prior to this patch series
without invoking the SYSCALL_DEFINE or COMPAT_SYSCALL_DEFINE macros with
custom input and output types. We remove every such direct definition in
favour of the aforementioned macros.

Also update syscalls.tbl in order to refer to the symbol names generated
by each of these macros. Since ppc64_personality can be called by both
64 bit and 32 bit binaries through compatibility, we must generate both
both compat_sys_ and sys_ symbols for this handler.

As an aside:
A number of architectures including arm and powerpc agree on an
alternative argument order and numbering for most of these arch-specific
handlers. A future patch series may allow for asm/unistd.h to signal
through its defines that a generic implementation of these syscall
handlers with the correct calling convention be emitted, through the
__ARCH_WANT_COMPAT_SYS_... convention.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-16-rmclure@linux.ibm.com
2 years agopowerpc: Provide do_ppc64_personality helper
Rohan McLure [Wed, 21 Sep 2022 06:55:54 +0000 (16:55 +1000)]
powerpc: Provide do_ppc64_personality helper

Avoid duplication in future patch that will define the ppc64_personality
syscall handler in terms of the SYSCALL_DEFINE and COMPAT_SYSCALL_DEFINE
macros, by extracting the common body of ppc64_personality into a helper
function.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-15-rmclure@linux.ibm.com
2 years agopowerpc: Remove direct call to mmap2 syscall handlers
Rohan McLure [Wed, 21 Sep 2022 06:55:53 +0000 (16:55 +1000)]
powerpc: Remove direct call to mmap2 syscall handlers

Syscall handlers should not be invoked internally by their symbol names,
as these symbols defined by the architecture-defined SYSCALL_DEFINE
macro. Move the compatibility syscall definition for mmap2 to
syscalls.c, so that all mmap implementations can share a helper function.

Remove 'inline' on static mmap helper.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Fix compat_sys_mmap2() prototype and offset handling]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-14-rmclure@linux.ibm.com
2 years agopowerpc: Remove direct call to personality syscall handler
Rohan McLure [Wed, 21 Sep 2022 06:55:52 +0000 (16:55 +1000)]
powerpc: Remove direct call to personality syscall handler

Syscall handlers should not be invoked internally by their symbol names,
as these symbols defined by the architecture-defined SYSCALL_DEFINE
macro. Fortunately, in the case of ppc64_personality, its call to
sys_personality can be replaced with an invocation to the
equivalent ksys_personality inline helper in <linux/syscalls.h>.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-13-rmclure@linux.ibm.com
2 years agopowerpc/32: Remove powerpc select specialisation
Rohan McLure [Wed, 21 Sep 2022 06:55:51 +0000 (16:55 +1000)]
powerpc/32: Remove powerpc select specialisation

Syscall #82 has been implemented for 32-bit platforms in a unique way on
powerpc systems. This hack will in effect guess whether the caller is
expecting new select semantics or old select semantics. It does so via a
guess, based off the first parameter. In new select, this parameter
represents the length of a user-memory array of file descriptors, and in
old select this is a pointer to an arguments structure.

The heuristic simply interprets sufficiently large values of its first
parameter as being a call to old select. The following is a discussion
on how this syscall should be handled.

As discussed in this thread, the existence of such a hack suggests that for
whatever powerpc binaries may predate glibc, it is most likely that they
would have taken use of the old select semantics. x86 and arm64 both
implement this syscall with oldselect semantics.

Remove the powerpc implementation, and update syscall.tbl to refer to emit
a reference to sys_old_select and compat_sys_old_select
for 32-bit binaries, in keeping with how other architectures support
syscall #82.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/lkml/13737de5-0eb7-e881-9af0-163b0d29a1a0@csgroup.eu/
Link: https://lore.kernel.org/r/20220921065605.1051927-12-rmclure@linux.ibm.com
2 years agopowerpc: Use generic fallocate compatibility syscall
Rohan McLure [Wed, 21 Sep 2022 06:55:50 +0000 (16:55 +1000)]
powerpc: Use generic fallocate compatibility syscall

The powerpc fallocate compat syscall handler is identical to the
generic implementation provided by commit 59c10c52f573f ("riscv:
compat: syscall: Add compat_sys_call_table implementation"), and as
such can be removed in favour of the generic implementation.

A future patch series will replace more architecture-defined syscall
handlers with generic implementations, dependent on introducing generic
implementations that are compatible with powerpc and arm's parameter
reorderings.

Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-11-rmclure@linux.ibm.com
2 years agoasm-generic: compat: Support BE for long long args in 32-bit ABIs
Rohan McLure [Wed, 21 Sep 2022 06:55:49 +0000 (16:55 +1000)]
asm-generic: compat: Support BE for long long args in 32-bit ABIs

32-bit ABIs support passing 64-bit integers by registers via argument
translation. Commit 59c10c52f573 ("riscv: compat: syscall: Add
compat_sys_call_table implementation") implements the compat_arg_u64
macro for efficiently defining little endian compatibility syscalls.

Architectures supporting big endianness may benefit from reciprocal
argument translation, but are welcome also to implement their own.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Arnd Bergmann <arnd@anrdb.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-10-rmclure@linux.ibm.com
2 years agopowerpc: Fix fallocate and fadvise64_64 compat parameter combination
Rohan McLure [Wed, 21 Sep 2022 06:55:48 +0000 (16:55 +1000)]
powerpc: Fix fallocate and fadvise64_64 compat parameter combination

As reported[1] by Arnd, the arch-specific fadvise64_64 and fallocate
compatibility handlers assume parameters are passed with 32-bit
big-endian ABI. This affects the assignment of odd-even parameter pairs
to the high or low words of a 64-bit syscall parameter.

Fix fadvise64_64 fallocate compat handlers to correctly swap upper/lower
32 bits conditioned on endianness.

A future patch will replace the arch-specific compat fallocate with an
asm-generic implementation. This patch is intended for ease of
back-port.

[1]: https://lore.kernel.org/all/be29926f-226e-48dc-871a-e29a54e80583@www.fastmail.com/

Fixes: 57f48b4b74e7 ("powerpc/compat_sys: swap hi/lo parts of 64-bit syscall args in LE mode")
Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-9-rmclure@linux.ibm.com
2 years agopowerpc/64s: Fix comment on interrupt handler prologue
Rohan McLure [Wed, 21 Sep 2022 06:55:47 +0000 (16:55 +1000)]
powerpc/64s: Fix comment on interrupt handler prologue

Interrupt handlers on 64s systems will often need to save register state
from the interrupted process to make space for loading special purpose
registers or for internal state.

Fix a comment documenting a common code path macro in the beginning of
interrupt handlers where r10 is saved to the PACA to afford space for
the value of the CFAR. Comment is currently written as if r10-r12 are
saved to PACA, but in fact only r10 is saved, with r11-r12 saved much
later. The distance in code between these saves has grown over the many
revisions of this macro. Fix this by signalling with a comment where
r11-r12 are saved to the PACA.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reported-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-8-rmclure@linux.ibm.com
2 years agopowerpc/64e: Clarify register saves and clears with {SAVE,ZEROIZE}_GPRS
Rohan McLure [Wed, 21 Sep 2022 06:55:46 +0000 (16:55 +1000)]
powerpc/64e: Clarify register saves and clears with {SAVE,ZEROIZE}_GPRS

The common interrupt handler prologue macro and the bad_stack
trampolines include consecutive sequences of register saves, and some
register clears. Neaten such instances by expanding use of the SAVE_GPRS
macro and employing the ZEROIZE_GPR macro when appropriate.

Also simplify an invocation of SAVE_GPRS targetting all non-volatile
registers to SAVE_NVGPRS.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reported-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-7-rmclure@linux.ibm.com
2 years agopowerpc/32: Clarify interrupt restores with REST_GPR macro in entry_32.S
Rohan McLure [Wed, 21 Sep 2022 06:55:45 +0000 (16:55 +1000)]
powerpc/32: Clarify interrupt restores with REST_GPR macro in entry_32.S

Restoring the register state of the interrupted thread involves issuing
a large number of predictable loads to the kernel stack frame. Issue the
REST_GPR{,S} macros to clearly signal when this is happening, and bunch
together restores at the end of the interrupt handler where the saved
value is not consumed earlier in the handler code.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reported-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-6-rmclure@linux.ibm.com
2 years agopowerpc/64s: Use {ZEROIZE,SAVE,REST}_GPRS macros in sc, scv 0 handlers
Rohan McLure [Wed, 21 Sep 2022 06:55:44 +0000 (16:55 +1000)]
powerpc/64s: Use {ZEROIZE,SAVE,REST}_GPRS macros in sc, scv 0 handlers

Use the convenience macros for saving/clearing/restoring gprs in keeping
with syscall calling conventions. The plural variants of these macros
can store a range of registers for concision.

This works well when the user gpr value we are hoping to save is still
live. In the syscall interrupt handlers, user register state is
sometimes juggled between registers. Hold-off from issuing the SAVE_GPR
macro for applicable neighbouring lines to highlight the delicate
register save logic.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-5-rmclure@linux.ibm.com
2 years agopowerpc: Add ZEROIZE_GPRS macros for register clears
Rohan McLure [Wed, 21 Sep 2022 06:55:43 +0000 (16:55 +1000)]
powerpc: Add ZEROIZE_GPRS macros for register clears

Provide register zeroing macros, following the same convention as
existing register stack save/restore macros, to be used in later
change to concisely zero a sequence of consecutive gprs.

The resulting macros are called ZEROIZE_GPRS and ZEROIZE_NVGPRS, keeping
with the naming of the accompanying restore and save macros, and usage
of zeroize to describe this operation elsewhere in the kernel.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-4-rmclure@linux.ibm.com
2 years agopowerpc: Save caller r3 prior to system_call_exception
Rohan McLure [Wed, 21 Sep 2022 06:55:42 +0000 (16:55 +1000)]
powerpc: Save caller r3 prior to system_call_exception

This reverts commit 8875f47b7681 ("powerpc/syscall: Save r3 in regs->orig_r3
").

Save caller's original r3 state to the kernel stackframe before entering
system_call_exception. This allows for user registers to be cleared by
the time system_call_exception is entered, reducing the influence of
user registers on speculation within the kernel.

Prior to this commit, orig_r3 was saved at the beginning of
system_call_exception. Instead, save orig_r3 while the user value is
still live in r3.

Also replicate this early save in 32-bit. A similar save was removed in
commit 6f76a01173cc ("powerpc/syscall: implement system call entry/exit
logic in C for PPC32") when 32-bit adopted system_call_exception. Revert
its removal of orig_r3 saves.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-3-rmclure@linux.ibm.com
2 years agopowerpc: Remove asmlinkage from syscall handler definitions
Rohan McLure [Wed, 21 Sep 2022 06:55:41 +0000 (16:55 +1000)]
powerpc: Remove asmlinkage from syscall handler definitions

The asmlinkage macro has no special meaning in powerpc, and prior to
this patch is used sporadically on some syscall handler definitions. On
architectures that do not define asmlinkage, it resolves to extern "C"
for C++ compilers and a nop otherwise. The current invocations of
asmlinkage provide far from complete support for C++ toolchains, and so
the macro serves no purpose in powerpc.

Remove all invocations of asmlinkage in arch/powerpc. These incidentally
only occur in syscall definitions and prototypes.

Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220921065605.1051927-2-rmclure@linux.ibm.com
2 years agopowerpc/irq: Refactor irq_soft_mask_{set,or}_return()
Christophe Leroy [Tue, 20 Sep 2022 06:41:08 +0000 (08:41 +0200)]
powerpc/irq: Refactor irq_soft_mask_{set,or}_return()

This partialy reapply commit ef5b570d3700 ("powerpc/irq: Don't
open code irq_soft_mask helpers") which was reverted by
commit 684c68d92e2e ("Revert "powerpc/irq: Don't open code
irq_soft_mask helpers"")

irq_soft_mask_set_return() and irq_soft_mask_or_return()
are overset of irq_soft_mask_set().

Have them use irq_soft_mask_set() instead of duplicating it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/18473da42362ee8f07bce36b9caef8ba77d7633f.1663656054.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Remove impossible mmu_psize_defs[] on nohash
Christophe Leroy [Mon, 19 Sep 2022 17:01:43 +0000 (19:01 +0200)]
powerpc: Remove impossible mmu_psize_defs[] on nohash

Today there is:

  if e500 or 8xx
    if e500
      mmu_psize_defs[] =
    else if 8xx
      mmu_psize_defs[] =
    else
      mmu_psize_defs[] =
    endif
  endif

The else leg is dead definition.

Drop that else leg and rewrite as:

  if e500
    mmu_psize_defs[] =
  endif
  if 8xx
    mmu_psize_defs[] =
  endif

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/030a843449f348c0b709ca5349640624f36a016f.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Cleanup idle for e500
Christophe Leroy [Mon, 19 Sep 2022 17:01:42 +0000 (19:01 +0200)]
powerpc: Cleanup idle for e500

e500 idle setup is a bit messy.

e500_idle() is used for PPC32 while book3e_idle() is used for PPC64.
As they are mutually exclusive, call them all e500_idle().

Use CONFIG_MPC_85xx instead of PPC32 + E500 in Makefile and rename
idle_e500.c to idle_85xx.c .

Rename idle_book3e.c to idle_64e.c and remove #ifdef PPC64 in
as it's only built on PPC64.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/8039301334e948974c85ec5ef2db37751075185b.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Simplify redundant Kconfig tests
Christophe Leroy [Mon, 19 Sep 2022 17:01:41 +0000 (19:01 +0200)]
powerpc: Simplify redundant Kconfig tests

PPC_85xx implies PPC32 so no need to check PPC32 in addition.

PPC64 && !PPC_BOOK3E_64 means PPC_BOOK3S_64.

PPC_BOOK3E_64 implies PPC_E500.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/244cce3e603f2b79796314c0c1c46cab927b9adc.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Replace PPC_85xx || PPC_BOOKE_64 by PPC_E500
Christophe Leroy [Mon, 19 Sep 2022 17:01:40 +0000 (19:01 +0200)]
powerpc: Replace PPC_85xx || PPC_BOOKE_64 by PPC_E500

PPC_E500 is the same as PPC_85xx || PPC_BOOKE_64

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/af79696f8cb8536fb4e20c0d98a6bf159a9e371b.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Remove CONFIG_PPC_BOOK3E_MMU
Christophe Leroy [Mon, 19 Sep 2022 17:01:39 +0000 (19:01 +0200)]
powerpc: Remove CONFIG_PPC_BOOK3E_MMU

CONFIG_PPC_BOOK3E_MMU is redundant with CONFIG_PPC_E500.

Remove it.

Also rename mmu-book3e.h to mmu-e500.h

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c5549cd59a131204ff94ab909cad2e2dad4ddf2f.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Remove CONFIG_PPC_FSL_BOOK3E
Christophe Leroy [Mon, 19 Sep 2022 17:01:38 +0000 (19:01 +0200)]
powerpc: Remove CONFIG_PPC_FSL_BOOK3E

CONFIG_PPC_FSL_BOOK3E is redundant with CONFIG_PPC_E500.

Remove it.

And rename five files accordingly.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Rename include guards to match new file names]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/795cb93b88c9a0279289712e674f39e3b108a1b4.1663606876.git.christophe.leroy@csgroup.eu
2 years agowatchdog: booke_wdt: Replace PPC_FSL_BOOK3E by PPC_E500
Christophe Leroy [Mon, 19 Sep 2022 17:01:37 +0000 (19:01 +0200)]
watchdog: booke_wdt: Replace PPC_FSL_BOOK3E by PPC_E500

CONFIG_PPC_FSL_BOOK3E is redundant with CONFIG_PPC_E500.

Replace it so that CONFIG_PPC_FSL_BOOK3E can be removed later.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/01a9132d51d3d8d9c74576d3da4d9d1fa5a88bde.1663606876.git.christophe.leroy@csgroup.eu
2 years agoDocumentation: Rename PPC_FSL_BOOK3E to PPC_E500
Christophe Leroy [Mon, 19 Sep 2022 17:01:36 +0000 (19:01 +0200)]
Documentation: Rename PPC_FSL_BOOK3E to PPC_E500

CONFIG_PPC_FSL_BOOK3E is redundant with CONFIG_PPC_E500.

Rename it so that CONFIG_PPC_FSL_BOOK3E can be removed later.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d3d42b395c09e66b0705fda1e51779f33e13ac38.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Change CONFIG_E500 to CONFIG_PPC_E500
Christophe Leroy [Mon, 19 Sep 2022 17:01:35 +0000 (19:01 +0200)]
powerpc: Change CONFIG_E500 to CONFIG_PPC_E500

It will be used outside arch/powerpc, make it clear its a
powerpc configuration item.

And we already have CONFIG_PPC_E500MC, so that will make
it more consistent.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e63b22083c11c4300f4a82d3123a46e5fdd54fa6.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Remove redundant selection of E500 and E500MC
Christophe Leroy [Mon, 19 Sep 2022 17:01:34 +0000 (19:01 +0200)]
powerpc: Remove redundant selection of E500 and E500MC

PPC_85xx and PPC_BOOK3E_64 already select E500 so no need
to select it again by PPC_QEMU_E500 and CORENET_GENERIC
as they depend on PPC_85xx || PPC_BOOK3E_64.

PPC_BOOK3E_64 already selects E500MC so no need to
select it again by PPC_QEMU_E500 if PPC64, PPC_BOOK3E_64
is the only way into PPC_QEMU_E500 with PPC64.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/44f03fa1506892fabf626dceb2f47a049908b6af.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Remove CONFIG_PPC_BOOK3E
Christophe Leroy [Mon, 19 Sep 2022 17:01:33 +0000 (19:01 +0200)]
powerpc: Remove CONFIG_PPC_BOOK3E

CONFIG_PPC_BOOK3E is redundant with CONFIG_PPC_BOOK3E_64.

The later is more explicit about the fact that it's a 64 bits target.

Remove CONFIG_PPC_BOOK3E.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5d0891490813c19cdcfc04678f512ea68cba3e64.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc/cputable: Split cpu_specs[] for mpc85xx and e500mc
Christophe Leroy [Mon, 19 Sep 2022 17:01:32 +0000 (19:01 +0200)]
powerpc/cputable: Split cpu_specs[] for mpc85xx and e500mc

e500v1/v2 and e500mc are said to be mutually exclusive in Kconfig.

Split e500 cpu_specs[] and then restrict the non e500mc to PPC32
which is then 85xx.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Tweak formatting]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/553b901ea91e393df231103da4b018e9b251b0e9.1663606876.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Remove CONFIG_FSL_BOOKE
Christophe Leroy [Mon, 19 Sep 2022 17:01:31 +0000 (19:01 +0200)]
powerpc: Remove CONFIG_FSL_BOOKE

PPC_85xx is PPC32 only.
PPC_85xx always selects E500 and is the only PPC32 that
selects E500.
FSL_BOOKE is selected when E500 and PPC32 are selected.

So FSL_BOOKE is redundant with PPC_85xx.

Remove FSL_BOOKE.

And rename four files accordingly.

cpu_setup_fsl_booke.S is not renamed because it is linked to
PPC_FSL_BOOK3E and not to FSL_BOOKE as suggested by its name.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/08e3e15594e66d63b9e89c5b4f9c35153913c28f.1663606875.git.christophe.leroy@csgroup.eu
2 years agopowerpc/cputable: Split cpu_specs[] out of cputable.h
Christophe Leroy [Mon, 19 Sep 2022 17:01:30 +0000 (19:01 +0200)]
powerpc/cputable: Split cpu_specs[] out of cputable.h

cpu_specs[] is full of #ifdefs depending on the different
types of CPU.

CPUs are mutually exclusive, it is therefore possible to split
cpu_specs[] into smaller more readable pieces.

Create cpu_specs_XXX.h that will each be dedicated on one
of the following mutually exclusive families:
- 40x
- 44x
- 47x
- 8xx
- e500
- book3s/32
- book3s/64

In book3s/32, the block for 603 has been moved in front in order
to not have two 604 blocks.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Fix CONFIG_47x to be CONFIG_PPC_47x, tweak some formatting]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/a44b865e0318286155273b10cdf524ab697928c1.1663606875.git.christophe.leroy@csgroup.eu
2 years agopowerpc/cputable: Move __cpu_setup() prototypes out of cputable.h
Christophe Leroy [Mon, 19 Sep 2022 17:01:29 +0000 (19:01 +0200)]
powerpc/cputable: Move __cpu_setup() prototypes out of cputable.h

Move all prototypes out of cputable.h

For that rename cpu_setup_power.h to cpu_setup.h and move all
prototypes in it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Standardise cpu_spec *spec formatting]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f45118489ee450db654db8bbcdfd8f5907337c22.1663606875.git.christophe.leroy@csgroup.eu
2 years agopowerpc/cputable: Remove __machine_check_early_realmode_p{7/8/9} prototypes
Christophe Leroy [Mon, 19 Sep 2022 17:01:28 +0000 (19:01 +0200)]
powerpc/cputable: Remove __machine_check_early_realmode_p{7/8/9} prototypes

__machine_check_early_realmode_p{7/8/9} are already in mce.h
which is included. Remove them from cputable.c

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b77fc0f90e3a9c065324cbff549b718ccf0809f8.1663606875.git.christophe.leroy@csgroup.eu
2 years agopowerpc/64e: Remove unnecessary #ifdef CONFIG_PPC_FSL_BOOK3E
Christophe Leroy [Mon, 19 Sep 2022 17:01:27 +0000 (19:01 +0200)]
powerpc/64e: Remove unnecessary #ifdef CONFIG_PPC_FSL_BOOK3E

CONFIG_PPC_BOOK3E_64 implies CONFIG_PPC_FSL_BOOK3E so no need of
additional #ifdefs in files built exclusively for CONFIG_PPC_BOOK3E_64.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/df16255c13b63b0221c9be63b94a6864bed22c12.1663606875.git.christophe.leroy@csgroup.eu
2 years agopowerpc/64e: Tie PPC_BOOK3E_64 to PPC_E500MC
Christophe Leroy [Mon, 19 Sep 2022 17:01:26 +0000 (19:01 +0200)]
powerpc/64e: Tie PPC_BOOK3E_64 to PPC_E500MC

The only 64-bit Book3E CPUs we support require the selection
of CONFIG_PPC_E500MC.

However our Kconfig allows configurating a kernel that has 64-bit
Book3E support, but without CONFIG_PPC_E500MC enabled. Such a kernel
would never boot, it doesn't know about any CPUs.

To fix this, force CONFIG_PPC_E500MC to be selected whenever we are
building a 64-bit Book3E kernel.

And add a test to detect future situations where cpu_specs is empty.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ae5d8b8b3ccc346e61d2ec729767f92766273f0b.1663606875.git.christophe.leroy@csgroup.eu
2 years agopowerpc/Kconfig: Fix non existing CONFIG_PPC_FSL_BOOKE
Christophe Leroy [Mon, 19 Sep 2022 17:01:25 +0000 (19:01 +0200)]
powerpc/Kconfig: Fix non existing CONFIG_PPC_FSL_BOOKE

CONFIG_PPC_FSL_BOOKE doesn't exist. Should be CONFIG_FSL_BOOKE.

Fixes: 49e3d8ea6248 ("powerpc/fsl_booke: Enable STRICT_KERNEL_RWX")
Cc: stable@vger.kernel.org
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/828f6a64eeb51ce9abfa1d4e84c521a02fecebb8.1663606875.git.christophe.leroy@csgroup.eu
2 years agopowerpc: dts: turris1x.dts: Fix labels in DSA cpu port nodes
Pali Rohár [Sat, 27 Aug 2022 13:15:38 +0000 (15:15 +0200)]
powerpc: dts: turris1x.dts: Fix labels in DSA cpu port nodes

DSA cpu port node has to be marked with "cpu" label.
So fix it for both cpu port nodes.

Fixes: 54c15ec3b738 ("powerpc: dts: Add DTS file for CZ.NIC Turris 1.x routers")
Signed-off-by: Pali Rohár <pali@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220827131538.14577-1-pali@kernel.org
2 years agopowerpc: dts: turris1x.dts: Fix NOR partitions labels
Pali Rohár [Tue, 30 Aug 2022 22:55:00 +0000 (00:55 +0200)]
powerpc: dts: turris1x.dts: Fix NOR partitions labels

Partition partition@20000 contains generic kernel image and it does not
have to be used only for rescue purposes. Partition partition@1c0000
contains bootable rescue system and partition partition@340000 contains
factory image/data for restoring to NAND. So change partition labels to
better fit their purpose by removing possible misleading substring "rootfs"
from these labels.

Fixes: 54c15ec3b738 ("powerpc: dts: Add DTS file for CZ.NIC Turris 1.x routers")
Signed-off-by: Pali Rohár <pali@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220830225500.8856-1-pali@kernel.org
2 years agopowerpc: Always select HAVE_EFFICIENT_UNALIGNED_ACCESS
Michael Ellerman [Fri, 16 Sep 2022 13:15:23 +0000 (23:15 +1000)]
powerpc: Always select HAVE_EFFICIENT_UNALIGNED_ACCESS

Currently powerpc selects HAVE_EFFICIENT_UNALIGNED_ACCESS in all cases
but one. The exception is if the kernel is being built little endian and
explicitly targeted for Power7.

The combination of Power7 and little endian was never commercially
supported, or widely used. It was only ever possible on bare metal
machines, using unofficial firmware, or in qemu guests hosted on those
machines.

The bare metal firmware support for Power7 was removed in 2019, see
skiboot commit 16b7ae64 ("Remove POWER7 and POWER7+ support").

Little endian kernel builds were switched to target Power8 or later in
2018, in commit a73657ea19ae ("powerpc/64: Add GENERIC_CPU support for
little endian"). Since then it's only been possible to boot a Power7/LE
kernel by explicitly building for Power7.

So drop the exception and always select HAVE_EFFICIENT_UNALIGNED_ACCESS.
If anyone does still have a Power7/LE machine it should hopefully
continue to boot, just with some performance penality, and if not they
can report a bug.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916131523.319123-1-mpe@ellerman.id.au
2 years agopowerpc/highmem: Properly handle fragmented memory
Christophe Leroy [Tue, 20 Sep 2022 17:36:42 +0000 (19:36 +0200)]
powerpc/highmem: Properly handle fragmented memory

In addition to checking whether a page is reserved before allocating
it to highmem, verify that it is valid memory.

Otherwise the kernel Oopses as below:

  mem auto-init: stack:off, heap alloc:off, heap free:off
  Kernel attempted to read user page (7df58) - exploit attempt? (uid: 0)
  BUG: Unable to handle kernel data access on read at 0x0007df58
  Faulting instruction address: 0xc01c8348
  Oops: Kernel access of bad area, sig: 11 [#1]
  BE PAGE_SIZE=4K SMP NR_CPUS=2 P2020RDB-PC
  Modules linked in:
  CPU: 0 PID: 0 Comm: swapper Not tainted 6.0.0-rc2-0caacb197b677410bdac81bc34f05235+ #121
  NIP:  c01c8348 LR: c01cb2bc CTR: 0000000a
  REGS: c10d7e20 TRAP: 0300   Not tainted  (6.0.0-rc2-0caacb197b677410bdac81bc34f05235+)
  MSR:  00021000 <CE,ME>  CR: 48044224  XER: 00000000
  DEAR: 0007df58 ESR: 00000000
  GPR00: c01cb294 c10d7f10 c1045340 00000001 00000004 c112bcc0 00000015 eedf1000
  GPR08: 00000003 0007df58 00000000 f0000000 28044228 00000200 00000000 00000000
  GPR16: 00000000 00000000 00000000 0275cb7a c0000000 00000001 0000075f 00000000
  GPR24: c1031004 00000000 00000000 00000001 c10f0000 eedf1000 00080000 00080000
  NIP free_unref_page_prepare.part.93+0x48/0x60
  LR  free_unref_page+0x84/0x4b8
  Call Trace:
    0xeedf1000 (unreliable)
    free_unref_page+0x5c/0x4b8
    mem_init+0xd0/0x194
    start_kernel+0x4c0/0x6d0
    set_ivor+0x13c/0x178

Reported-by: Pali Rohár <pali@kernel.org>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Fixes: b0e0d68b1c52 ("powerpc/32: Allow fragmented physical memory")
Tested-by: Pali Rohár <pali@kernel.org>
[mpe: Trim oops]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f08cca5c46d67399c53262eca48e015dcf1841f9.1663695394.git.christophe.leroy@csgroup.eu
2 years agopowerpc/prom_init: drop PROM_BUG()
David Hildenbrand [Tue, 20 Sep 2022 12:23:01 +0000 (14:23 +0200)]
powerpc/prom_init: drop PROM_BUG()

Unused, let's drop it.

Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220920122302.99195-3-david@redhat.com
2 years agopowerpc/mm: Make PAGE_KERNEL_xxx macros grep-friendly
Christophe Leroy [Wed, 7 Sep 2022 10:05:21 +0000 (12:05 +0200)]
powerpc/mm: Make PAGE_KERNEL_xxx macros grep-friendly

Avoid multi-lines to help getting a complete view when using
grep. They still remain under the 100 chars limit.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/3bc3f5a51949ee7f52dba36677db23d4337c7995.1662544980.git.christophe.leroy@csgroup.eu
2 years agopowerpc/mm: Reduce redundancy in pgtable.h
Christophe Leroy [Wed, 7 Sep 2022 10:05:01 +0000 (12:05 +0200)]
powerpc/mm: Reduce redundancy in pgtable.h

PAGE_KERNEL_TEXT, PAGE_KERNEL_EXEC and PAGE_AGP are the same
for all powerpcs.

Remove duplicated definitions.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/92254499430d13d99e4a4d7e9ad8e8284cb35380.1662544974.git.christophe.leroy@csgroup.eu
2 years agopowerpc/book3s: Inline first level of update_mmu_cache()
Christophe Leroy [Mon, 5 Sep 2022 09:38:25 +0000 (11:38 +0200)]
powerpc/book3s: Inline first level of update_mmu_cache()

update_mmu_cache() voids when hash page tables are not used.
On PPC32 that means when MMU_FTR_HPTE_TABLE is not defined.
On PPC64 that means when RADIX is enabled.

Rename core part of update_mmu_cache() as __update_mmu_cache()
and include the initial verification in an inlined caller.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/bea5ad0de7f83eff256116816d46c84fa0a444de.1662370698.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Rely on generic definition of hugepd_t and is_hugepd when unused
Christophe Leroy [Wed, 7 Sep 2022 09:34:45 +0000 (11:34 +0200)]
powerpc: Rely on generic definition of hugepd_t and is_hugepd when unused

CONFIG_ARCH_HAS_HUGEPD is used to tell core mm when huge page
directories are used.

When they are not used, no need to provide hugepd_t or is_hugepd(),
just rely on the core mm fallback definition.

For that, change core mm behaviour so that CONFIG_ARCH_HAS_HUGEPD
is used instead of indirect is_hugepd macro existence.

powerpc being the only user of huge page directories, there is no
impact on other architectures.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/da81462d93069bb90fe5e762dd3283a644318937.1662543243.git.christophe.leroy@csgroup.eu
2 years agopowerpc/nohash: Remove pgd_huge() stub
Christophe Leroy [Wed, 7 Sep 2022 09:34:44 +0000 (11:34 +0200)]
powerpc/nohash: Remove pgd_huge() stub

linux/hugetlb.h has a fallback pgd_huge() macro for when
pgd_huge is not defined.

Remove the powerpc redundant definitions.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ae6aa7fce84f7abcbf67f534271a4a6dd7949b0d.1662543243.git.christophe.leroy@csgroup.eu
2 years agopowerpc/mm/64s: Drop p4d_leaf()
Michael Ellerman [Sat, 3 Sep 2022 12:36:40 +0000 (22:36 +1000)]
powerpc/mm/64s: Drop p4d_leaf()

Because 64-bit Book3S uses pgtable-nop4d.h, the P4D is folded into the
PGD. So P4D entries are actually PGD entries, or vice versa.

The other way to think of it is that the P4D is a single entry page
table below the PGD. Zero bits of the address are needed to index into
the P4D, therefore a P4D entry maps the same size address space as a PGD
entry.

As explained in the previous commit, there are no huge page sizes
supported directly at the PGD level on 64-bit Book3S, so there are also
no huge page sizes supported at the P4D level.

Therefore p4d_is_leaf() can never be true, so drop the definition and
fallback to the default implementation that always returns false.

Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220903123640.719846-2-mpe@ellerman.id.au
2 years agopowerpc/mm/64s: Drop pgd_huge()
Michael Ellerman [Sat, 3 Sep 2022 12:36:39 +0000 (22:36 +1000)]
powerpc/mm/64s: Drop pgd_huge()

On powerpc there are two ways for huge pages to be represented in the
top level page table, aka PGD (Page Global Directory).

If the address space mapped by an individual PGD entry does not
correspond to a given huge page size, then the PGD entry points to a
non-standard page table, known as a "hugepd" (Huge Page Directory).
The hugepd contains some number of huge page PTEs sufficient to map the
address space with the given huge page size.

On the other hand, if the address space mapped by an individual PGD
entry does correspond exactly to a given huge page size, that PGD entry
is used to directly encode the huge page PTE in place. In this case the
pgd_huge() wrapper indicates to generic code that the PGD entry is
actually a huge page PTE.

This commit deals with the pgd_huge() case only, it does nothing with
respect to the hugepd case.

Over time the size of the virtual address space supported on powerpc has
increased several times, which means the location at which huge pages
can sit in the tree has also changed. There have also been new huge page
sizes added, with the introduction of the Radix MMU.

On Power9 and later with the Radix MMU, the largest huge page size in
any implementation is 1GB.

Since the introduction of Radix, 1GB entries have been supported at the
PUD level, with both 4K and 64K base page size. Radix has never had a
supported huge page size at the PGD level.

On Power8 or earlier, which uses the Hash MMU, or Power9 or later with
the Hash MMU enabled, the largest huge page size is 16GB.

Using the Hash MMU and a base page size of 4K, 16GB has never been a
supported huge page size at the PGD level, due to the geometry being
incompatible. The two supported huge page sizes (16M & 16GB) both use
the hugepd format.

Using the Hash MMU and a base page size of 64K, 16GB pages were
supported in the past at the PGD level.

However in commit ba95b5d03596 ("powerpc/mm/book3s/64: Rework page table
geometry for lower memory usage") the page table layout was reworked to
shrink the size of the PGD.

As a result the 16GB page size now fits at the PUD level when using 64K
base page size.

Therefore there are no longer any supported configurations where
pgd_huge() can be true, so drop the definitions for pgd_huge(), and
fallback to the generic definition which is always false.

Fixes: ba95b5d03596 ("powerpc/mm/book3s/64: Rework page table geometry for lower memory usage")
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220903123640.719846-1-mpe@ellerman.id.au
2 years agopowerpc/microwatt: Remove unused early debug code
Michael Ellerman [Mon, 19 Sep 2022 05:27:55 +0000 (15:27 +1000)]
powerpc/microwatt: Remove unused early debug code

The original microwatt submission[1] included some early debug code for
using the Microwatt "potato" UART.

The series that was eventually merged switched to using a standard UART,
and so doesn't need any special early debug handling. But some of the
original code was merged accidentally under the non-existent
CONFIG_PPC_EARLY_DEBUG_MICROWATT.

Drop the unused code.

1: https://lore.kernel.org/linuxppc-dev/20200509050340.GD1464954@thinks.paulus.ozlabs.org/

Fixes: 48b545b8018d ("powerpc/microwatt: Use standard 16550 UART for console")
Reported-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Reviewed-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220919052755.800907-1-mpe@ellerman.id.au
2 years agopowerpc/64: Remove unused SYS_CALL_TABLE symbol
Michael Ellerman [Tue, 13 Sep 2022 12:45:45 +0000 (22:45 +1000)]
powerpc/64: Remove unused SYS_CALL_TABLE symbol

In interrupt_64.S, formerly entry_64.S, there are two toc entries
created for sys_call_table and compat_sys_call_table.

These are no longer used, since the system call entry was converted from
asm to C, so remove them.

Fixes: 68b34588e202 ("powerpc/64/sycall: Implement syscall entry/exit logic in C")
Acked-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913124545.2817825-1-mpe@ellerman.id.au
2 years agopowerpc/build: put sys_call_table in .data.rel.ro if RELOCATABLE
Nicholas Piggin [Fri, 16 Sep 2022 04:07:55 +0000 (14:07 +1000)]
powerpc/build: put sys_call_table in .data.rel.ro if RELOCATABLE

Const function pointers by convention live in .data.rel.ro if they need
to be relocated. Now that .data.rel.ro is linked into the read-only
region, put them in the right section. This doesn't make much practical
difference, but it will make the C conversion of sys_call_table a
smaller change as far as linking goes.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916040755.2398112-8-npiggin@gmail.com
2 years agopowerpc/64/build: merge .got and .toc input sections
Nicholas Piggin [Fri, 16 Sep 2022 04:07:54 +0000 (14:07 +1000)]
powerpc/64/build: merge .got and .toc input sections

Follow the binutils ld internal linker script and merge .got and .toc
input sections in the .got output section.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916040755.2398112-7-npiggin@gmail.com
2 years agopowerpc/64/build: only include .opd with ELFv1
Nicholas Piggin [Fri, 16 Sep 2022 04:07:53 +0000 (14:07 +1000)]
powerpc/64/build: only include .opd with ELFv1

ELFv2 does not use function descriptors so .opd is not required.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916040755.2398112-6-npiggin@gmail.com
2 years agopowerpc/build: move .data.rel.ro, .sdata2 to read-only
Nicholas Piggin [Fri, 16 Sep 2022 04:07:52 +0000 (14:07 +1000)]
powerpc/build: move .data.rel.ro, .sdata2 to read-only

.sdata2 is a readonly small data section for ppc32, and .data.rel.ro
is data that needs relocating but is read-only after that so these
can both be moved to the read only memory region.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916040755.2398112-5-npiggin@gmail.com
2 years agopowerpc/build: move got, toc, plt, branch_lt sections to read-only
Nicholas Piggin [Fri, 16 Sep 2022 04:07:51 +0000 (14:07 +1000)]
powerpc/build: move got, toc, plt, branch_lt sections to read-only

This moves linker-related tables from .data to read-only area.
Relocations are performed at early boot time before memory is protected,
after which there should be no modifications required.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Don't use SPECIAL as reported by lkp@intel.com]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916040755.2398112-4-npiggin@gmail.com
2 years agopowerpc/32/build: move got1/got2 sections out of text
Nicholas Piggin [Fri, 16 Sep 2022 04:07:50 +0000 (14:07 +1000)]
powerpc/32/build: move got1/got2 sections out of text

Following the example from the binutils default linker script, move
.got1 and .got2 out of .text, to just after RO_DATA.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916040755.2398112-3-npiggin@gmail.com
2 years agopowerpc: move __end_rodata to cover arch read-only sections
Nicholas Piggin [Fri, 16 Sep 2022 04:07:49 +0000 (14:07 +1000)]
powerpc: move __end_rodata to cover arch read-only sections

powerpc has a number of read-only sections and tables that are put after
RO_DATA(). Move the __end_rodata symbol to cover these as well.

Setting memory to read-only at boot is done using __init_begin, change
that to use __end_rodata.

This makes is_kernel_rodata() exactly cover the read-only region, as
well as other things using __end_rodata (e.g., kernel/dma/debug.c).
Boot dmesg also prints the rodata size more accurately.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916040755.2398112-2-npiggin@gmail.com
2 years agopowerpc/vmlinux.lds: Add an explicit symbol for the SRWX boundary
Michael Ellerman [Fri, 16 Sep 2022 04:41:24 +0000 (14:41 +1000)]
powerpc/vmlinux.lds: Add an explicit symbol for the SRWX boundary

Currently __init_begin is used as the boundary for strict RWX between
executable/read-only text and data, and non-executable (after boot) code
and data.

But that's a little subtle, so add an explicit symbol to document that
the SRWX boundary lies there, and add a comment making it clear that
__init_begin must also begin there.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916131422.318752-2-mpe@ellerman.id.au
2 years agopowerpc/vmlinux.lds: Ensure STRICT_ALIGN_SIZE is at least page aligned
Michael Ellerman [Fri, 16 Sep 2022 04:40:57 +0000 (14:40 +1000)]
powerpc/vmlinux.lds: Ensure STRICT_ALIGN_SIZE is at least page aligned

Add a check that STRICT_ALIGN_SIZE is aligned to at least PAGE_SIZE.

That then makes the alignment to PAGE_SIZE immediately after the
alignment to STRICT_ALIGN_SIZE redundant, so remove it.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220916131422.318752-1-mpe@ellerman.id.au
2 years agopowerpc/64: Remove unused prom_init_toc symbols
Michael Ellerman [Sat, 17 Sep 2022 06:36:47 +0000 (16:36 +1000)]
powerpc/64: Remove unused prom_init_toc symbols

Commit 24d33ac5b8ff ("powerpc/64s: Make prom_init require RELOCATABLE")
made prom_init depend on CONFIG_RELOCATABLE.

But it missed cleaning up a case in the linker script for RELOCATABLE=n,
and associated symbols. Remove them now.

Fixes: 24d33ac5b8ff ("powerpc/64s: Make prom_init require RELOCATABLE")
Reported-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220920131157.1032707-1-mpe@ellerman.id.au
2 years agopowerpc/pseries: move hcall_tracepoint_refcount out of .toc
Nicholas Piggin [Mon, 26 Sep 2022 05:38:23 +0000 (15:38 +1000)]
powerpc/pseries: move hcall_tracepoint_refcount out of .toc

The .toc section is not really intended for arbitrary data. Writable
data in particular prevents making the TOC read-only after relocation.
Move hcall_tracepoint_refcount into the .data section.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220926053823.2668799-1-npiggin@gmail.com
2 years agoMerge branch 'fixes' into next
Michael Ellerman [Fri, 23 Sep 2022 14:40:14 +0000 (00:40 +1000)]
Merge branch 'fixes' into next

Merge our fixes branch to bring in a few things that new feature patches
rely on or conflict with.

2 years agoKVM: PPC: remove orphan declarations from kvm_ppc.h
Gaosheng Cui [Tue, 13 Sep 2022 07:50:29 +0000 (15:50 +0800)]
KVM: PPC: remove orphan declarations from kvm_ppc.h

Remove the following orphan declarations from kvm_ppc.h:

1. kvmppc_mmu_priv_switch() has been removed since
commit dd9ebf1f9435 ("KVM: PPC: e500: Add shadow PID
support").
2. kvmppc_core_destroy_mmu() has been removed since
commit ecc0981ff07c ("KVM: ppc: cosmetic changes to mmu
hook names").
3. kvmppc_prepare_vrma() has been removed since
commit aa04b4cc5be6 ("KVM: PPC: Allocate RMAs (Real Mode
Areas) at boot for use by guests").

So remove the declarations for them from header file.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-10-cuigaosheng1@huawei.com
2 years agopowerpc/ps3: remove orphan declarations from ps3av.h
Gaosheng Cui [Tue, 13 Sep 2022 07:50:28 +0000 (15:50 +0800)]
powerpc/ps3: remove orphan declarations from ps3av.h

Remove the following orphan declarations from ps3av.h:
1. ps3av_dev_open()
2. ps3av_dev_close()

They have been removed since commit 13a5e30cf740 ("[POWERPC] PS3:
Rework AV settings driver"), so remove them.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Acked-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-9-cuigaosheng1@huawei.com
2 years agopowerpc/sysdev: remove unused xics_ipi_dispatch() declaration
Gaosheng Cui [Tue, 13 Sep 2022 07:50:27 +0000 (15:50 +0800)]
powerpc/sysdev: remove unused xics_ipi_dispatch() declaration

xics_ipi_dispatch() has been removed since
commit 23d72bfd8f9f ("powerpc: Consolidate ipi message mux
and demux"), so remove it.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-8-cuigaosheng1@huawei.com
2 years agopowerpc/powernv: remove orphan declarations from opal.h
Gaosheng Cui [Tue, 13 Sep 2022 07:50:26 +0000 (15:50 +0800)]
powerpc/powernv: remove orphan declarations from opal.h

Remove the following orphan declarations from opal.h:
1. opal_notifier_register()
2. opal_notifier_unregister()
3. opal_notifier_update_evt()
4. opal_notifier_enable()
5. opal_notifier_disable()

They have been removed since commit 81f2f7ce4c5b ("opal:
Remove events notifier"), so remove them.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-7-cuigaosheng1@huawei.com
2 years agopowerpc/mm: remove orphan declarations from mmu_context.h
Gaosheng Cui [Tue, 13 Sep 2022 07:50:25 +0000 (15:50 +0800)]
powerpc/mm: remove orphan declarations from mmu_context.h

Remove the following orphan declarations from mmu_context.h:
1. switch_cop() and drop_cop() have been removed since
commit 6ff4d3e96652 ("powerpc: Remove old unused icswx
based coprocessor support").

2. mm_iommu_cleanup() has been removed since
commit 4b6fad7097f8 ("powerpc/mm/iommu, vfio/spapr: Put
pages on VFIO container shutdown").

So remove the declarations for them from header file.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-6-cuigaosheng1@huawei.com
2 years agopowerpc: remove unused udbg_init_debug_beat() declaration
Gaosheng Cui [Tue, 13 Sep 2022 07:50:24 +0000 (15:50 +0800)]
powerpc: remove unused udbg_init_debug_beat() declaration

udbg_init_debug_beat() has been removed since
commit bf4981a00636 ("powerpc: Remove the celleb support"),
so remove it.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-5-cuigaosheng1@huawei.com
2 years agopowerpc: remove unused chrp_event_scan() declaration
Gaosheng Cui [Tue, 13 Sep 2022 07:50:23 +0000 (15:50 +0800)]
powerpc: remove unused chrp_event_scan() declaration

chrp_event_scan() has been removed since
commit 3d541c4b7f6e ("powerpc/chrp: Use the same RTAS
daemon as pSeries"), so remove it.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-4-cuigaosheng1@huawei.com
2 years agopowerpc/spufs: remove orphan declarations from spufs.h
Gaosheng Cui [Tue, 13 Sep 2022 07:50:22 +0000 (15:50 +0800)]
powerpc/spufs: remove orphan declarations from spufs.h

Remove the following orphan declarations from spufs.h:
1. spufs_coredump_calls has been removed since
commit 48cad41f7ee7 ("[POWERPC] spufs: Combine
spufs_coredump_calls with spufs_calls").

2. spufs_coredump_num_notes has been removed since
commit 936d5bf1d7dc ("[POWERPC] spufs: Get rid of spufs_coredump_num_notes,
it's not needed if we NULL terminate").

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-3-cuigaosheng1@huawei.com
2 years agopowerpc/xmon: remove unused ppc_parse_cpu() declaration
Gaosheng Cui [Tue, 13 Sep 2022 07:50:21 +0000 (15:50 +0800)]
powerpc/xmon: remove unused ppc_parse_cpu() declaration

ppc_parse_cpu() has been removed since
commit 5b102782c7f4 ("powerpc/xmon: Enable disassembly
files (compilation changes)"), so remove it.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220913075029.682327-2-cuigaosheng1@huawei.com
2 years agoRevert "powerpc/rtas: Implement reentrant rtas call"
Nathan Lynch [Wed, 7 Sep 2022 22:01:11 +0000 (17:01 -0500)]
Revert "powerpc/rtas: Implement reentrant rtas call"

At the time this was submitted by Leonardo, I confirmed -- or thought
I had confirmed -- with PowerVM partition firmware development that
the following RTAS functions:

- ibm,get-xive
- ibm,int-off
- ibm,int-on
- ibm,set-xive

were safe to call on multiple CPUs simultaneously, not only with
respect to themselves as indicated by PAPR, but with arbitrary other
RTAS calls:

https://lore.kernel.org/linuxppc-dev/875zcy2v8o.fsf@linux.ibm.com/

Recent discussion with firmware development makes it clear that this
is not true, and that the code in commit b664db8e3f97 ("powerpc/rtas:
Implement reentrant rtas call") is unsafe, likely explaining several
strange bugs we've seen in internal testing involving DLPAR and
LPM. These scenarios use ibm,configure-connector, whose internal state
can be corrupted by the concurrent use of the "reentrant" functions,
leading to symptoms like endless busy statuses from RTAS.

Fixes: b664db8e3f97 ("powerpc/rtas: Implement reentrant rtas call")
Cc: stable@vger.kernel.org # v5.8+
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Laurent Dufour <laurent.dufour@fr.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220907220111.223267-1-nathanl@linux.ibm.com
2 years agopowerpc/powernv: add missing of_node_put() in opal_export_attrs()
Zheng Yongjun [Tue, 6 Sep 2022 14:17:03 +0000 (14:17 +0000)]
powerpc/powernv: add missing of_node_put() in opal_export_attrs()

After using 'np' returned by of_find_node_by_path(), of_node_put()
need be called to decrease the refcount.

Fixes: 11fe909d2362 ("powerpc/powernv: Add OPAL exports attributes to sysfs")
Signed-off-by: Zheng Yongjun <zhengyongjun3@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220906141703.118192-1-zhengyongjun3@huawei.com
2 years agopowerpc/85xx: Fix fall-through warning for Clang
Gustavo A. R. Silva [Tue, 6 Sep 2022 21:32:13 +0000 (22:32 +0100)]
powerpc/85xx: Fix fall-through warning for Clang

Fix the following fallthrough warning:

arch/powerpc/platforms/85xx/mpc85xx_cds.c:161:3: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough]

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://github.com/KSPP/linux/issues/198
Link: https://lore.kernel.org/lkml/202209061224.KxORRGVg-lkp@intel.com/
Link: https://lore.kernel.org/r/Yxe8XTY5C9qJLd0Z@work
2 years agopowerpc/64s: add pte_needs_flush and huge_pmd_needs_flush
Nicholas Piggin [Thu, 1 Sep 2022 11:03:34 +0000 (21:03 +1000)]
powerpc/64s: add pte_needs_flush and huge_pmd_needs_flush

Allow PTE changes to avoid flushing the TLB when access permissions are
being relaxed, the dirty bit is being set, and the accessed bit is being
changed.

Relaxing access permissions and setting dirty and accessed bits do not
require a flush because the MMU will re-load the PTE and notice the
updates (it may also cause a spurious fault).

Clearing the accessed bit does not require a flush because of the
imprecise PTE accessed bit accounting that is already performed, as
documented in ptep_clear_flush_young().

This reduces TLB flushing for some mprotect(2) calls.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Link: https://lore.kernel.org/r/20220901110334.1618913-1-npiggin@gmail.com