Pull RCU updates from Ingo Molnad:
"The main RCU related changes in this cycle were:
- Removal of spin_unlock_wait()
- SRCU updates
- RCU torture-test updates
- RCU Documentation updates
- Extend the sys_membarrier() ABI with the MEMBARRIER_CMD_PRIVATE_EXPEDITED variant
- Miscellaneous RCU fixes
- CPU-hotplug fixes"
* 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (63 commits)
arch: Remove spin_unlock_wait() arch-specific definitions
locking: Remove spin_unlock_wait() generic definitions
drivers/ata: Replace spin_unlock_wait() with lock/unlock pair
ipc: Replace spin_unlock_wait() with lock/unlock pair
exit: Replace spin_unlock_wait() with lock/unlock pair
completion: Replace spin_unlock_wait() with lock/unlock pair
doc: Set down RCU's scheduling-clock-interrupt needs
doc: No longer allowed to use rcu_dereference on non-pointers
doc: Add RCU files to docbook-generation files
doc: Update memory-barriers.txt for read-to-write dependencies
doc: Update RCU documentation
membarrier: Provide expedited private command
rcu: Remove exports from rcu_idle_exit() and rcu_idle_enter()
rcu: Add warning to rcu_idle_enter() for irqs enabled
rcu: Make rcu_idle_enter() rely on callers disabling irqs
rcu: Add assertions verifying blocked-tasks list
rcu/tracing: Set disable_rcu_irq_enter on rcu_eqs_exit()
rcu: Add TPS() protection for _rcu_barrier_trace strings
rcu: Use idle versions of swait to make idle-hack clear
swait: Add idle variants which don't contribute to load average
...
There is some work that should be done after setting the personality.
Currently it's done in the macro, which is not the best idea.
In this patch new arch_setup_new_exec() routine is introduced, and all
setup code is moved there, as suggested by Catalin:
https://lkml.org/lkml/2017/8/4/494
Cc: Pratyush Anand <panand@redhat.com>
Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
[catalin.marinas@arm.com: comments changed or removed]
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
With 16KB pages and a kernel Image larger than 16MB, the current
kaslr_early_init() logic for avoiding mappings across swapper table
boundaries fails since increasing the offset by kimg_sz just moves the
problem to the next boundary.
This patch rounds the offset down to (1 << SWAPPER_TABLE_SHIFT) if the
Image crosses a PMD_SIZE boundary.
Fixes: afd0e5a876 ("arm64: kaslr: Fix up the kernel image alignment")
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
In the KASLR setup routine, we ensure that the early virtual mapping
of the kernel image does not cover more than a single table entry at
the level above the swapper block level, so that the assembler routines
involved in setting up this mapping can remain simple.
In this calculation we add the proposed KASLR offset to the values of
the _text and _end markers, and reject it if they would end up falling
in different swapper table sized windows.
However, when taking the addresses of _text and _end, the modulo offset
(the physical displacement modulo 2 MB) is already accounted for, and
so adding it again results in incorrect results. So disregard the modulo
offset from the calculation.
Fixes: 08cdac619c ("arm64: relocatable: deal with physically misaligned ...")
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Tested-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
There are some tricky dependencies between the different stages of
flushing the FPSIMD register state during exec, and these can race
with context switch in ways that can cause the old task's regs to
leak across. In particular, a context switch during the memset() can
cause some of the task's old FPSIMD registers to reappear.
Disabling preemption for this small window would be no big deal for
performance: preemption is already disabled for similar scenarios
like updating the FPSIMD registers in sigreturn.
So, instead of rearranging things in ways that might swap existing
subtle bugs for new ones, this patch just disables preemption
around the FPSIMD state flushing so that races of this type can't
occur here. This brings fpsimd_flush_thread() into line with other
code paths.
Cc: stable@vger.kernel.org
Fixes: 674c242c93 ("arm64: flush FP/SIMD state correctly after execve()")
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Currently mm->context.flags field uses thread_info flags which is not
the best idea for many reasons. For example, mm_context_t doesn't need
most of thread_info flags. And it would be difficult to add new mm-related
flag if needed because it may easily interfere with TIF ones.
To deal with it, the new MMCF_AARCH32 flag is introduced for
mm_context_t->flags, where MMCF prefix stands for mm_context_t flags.
Also, mm_context_t flag doesn't require atomicity and ordering of the
access, so using set/clear_bit() is replaced with simple masks.
Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Commit 0ee5941 : (x86/panic: replace smp_send_stop() with kdump friendly
version in panic path) introduced crash_smp_send_stop() which is a weak
function and can be overridden by architecture codes to fix the side effect
caused by commit f06e515 : (kernel/panic.c: add "crash_kexec_post_
notifiers" option).
ARM64 architecture uses the weak version function and the problem is that
the weak function simply calls smp_send_stop() which makes other CPUs
offline and takes away the chance to save crash information for nonpanic
CPUs in machine_crash_shutdown() when crash_kexec_post_notifiers kernel
option is enabled.
Calling smp_send_crash_stop() in machine_crash_shutdown() is useless
because all nonpanic CPUs are already offline by smp_send_stop() in this
case and smp_send_crash_stop() only works against online CPUs.
The result is that secondary CPUs registers are not saved by
crash_save_cpu() and the vmcore file misreports these CPUs as being
offline.
crash_smp_send_stop() is implemented to fix this problem by replacing the
existing smp_send_crash_stop() and adding a check for multiple calling to
the function. The function (strong symbol version) saves crash information
for nonpanic CPUs and machine_crash_shutdown() tries to save crash
information for nonpanic CPUs only when crash_kexec_post_notifiers kernel
option is disabled.
* crash_kexec_post_notifiers : false
panic()
__crash_kexec()
machine_crash_shutdown()
crash_smp_send_stop() <= save crash dump for nonpanic cores
* crash_kexec_post_notifiers : true
panic()
crash_smp_send_stop() <= save crash dump for nonpanic cores
__crash_kexec()
machine_crash_shutdown()
crash_smp_send_stop() <= just return.
Signed-off-by: Hoeun Ryu <hoeun.ryu@gmail.com>
Reviewed-by: James Morse <james.morse@arm.com>
Tested-by: James Morse <james.morse@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Currently PTE_RDONLY is treated as a hardware only bit and not handled
by the pte_mkwrite(), pte_wrprotect() or the user PAGE_* definitions.
The set_pte_at() function is responsible for setting this bit based on
the write permission or dirty state. This patch moves the PTE_RDONLY
handling out of set_pte_at into the pte_mkwrite()/pte_wrprotect()
functions. The PAGE_* definitions to need to be updated to explicitly
include PTE_RDONLY when !PTE_WRITE.
The patch also removes the redundant PAGE_COPY(_EXEC) definitions as
they are identical to the corresponding PAGE_READONLY(_EXEC).
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
* 'for-next/perf' of git://git.kernel.org/pub/scm/linux/kernel/git/will/linux:
arm64: perf: add support for Cortex-A35
arm64: perf: add support for Cortex-A73
arm64: perf: Remove redundant entries from CPU-specific event maps
arm64: perf: Connect additional events to pmu counters
arm64: perf: Allow standard PMUv3 events to be extended by the CPU type
perf: xgene: Remove unnecessary managed resources cleanup
arm64: perf: Allow more than one cycle counter to be used
The percpu variables efi_fpsimd_state and efi_fpsimd_state_used,
used by the FPSIMD save/restore routines for EFI calls, are
unintentionally global.
There's no reason for anything outside fpsimd.c to touch these, so
this patch makes them static (as they should have been in the first
place).
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Implement MEMBARRIER_CMD_PRIVATE_EXPEDITED with IPIs using cpumask built
from all runqueues for which current thread's mm is the same as the
thread calling sys_membarrier. It executes faster than the non-expedited
variant (no blocking). It also works on NOHZ_FULL configurations.
Scheduler-wise, it requires a memory barrier before and after context
switching between processes (which have different mm). The memory
barrier before context switch is already present. For the barrier after
context switch:
* Our TSO archs can do RELEASE without being a full barrier. Look at
x86 spin_unlock() being a regular STORE for example. But for those
archs, all atomics imply smp_mb and all of them have atomic ops in
switch_mm() for mm_cpumask(), and on x86 the CR3 load acts as a full
barrier.
* From all weakly ordered machines, only ARM64 and PPC can do RELEASE,
the rest does indeed do smp_mb(), so there the spin_unlock() is a full
barrier and we're good.
* ARM64 has a very heavy barrier in switch_to(), which suffices.
* PPC just removed its barrier from switch_to(), but appears to be
talking about adding something to switch_mm(). So add a
smp_mb__after_unlock_lock() for now, until this is settled on the PPC
side.
Changes since v3:
- Properly document the memory barriers provided by each architecture.
Changes since v2:
- Address comments from Peter Zijlstra,
- Add smp_mb__after_unlock_lock() after finish_lock_switch() in
finish_task_switch() to add the memory barrier we need after storing
to rq->curr. This is much simpler than the previous approach relying
on atomic_dec_and_test() in mmdrop(), which actually added a memory
barrier in the common case of switching between userspace processes.
- Return -EINVAL when MEMBARRIER_CMD_SHARED is used on a nohz_full
kernel, rather than having the whole membarrier system call returning
-ENOSYS. Indeed, CMD_PRIVATE_EXPEDITED is compatible with nohz_full.
Adapt the CMD_QUERY mask accordingly.
Changes since v1:
- move membarrier code under kernel/sched/ because it uses the
scheduler runqueue,
- only add the barrier when we switch from a kernel thread. The case
where we switch from a user-space thread is already handled by
the atomic_dec_and_test() in mmdrop().
- add a comment to mmdrop() documenting the requirement on the implicit
memory barrier.
CC: Peter Zijlstra <peterz@infradead.org>
CC: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
CC: Boqun Feng <boqun.feng@gmail.com>
CC: Andrew Hunter <ahh@google.com>
CC: Maged Michael <maged.michael@gmail.com>
CC: gromer@google.com
CC: Avi Kivity <avi@scylladb.com>
CC: Benjamin Herrenschmidt <benh@kernel.crashing.org>
CC: Paul Mackerras <paulus@samba.org>
CC: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Dave Watson <davejwatson@fb.com>
Both unwind_frame() and dump_backtrace() try to check whether a stack
address is sane to access, with very similar logic. Both will need
updating in order to handle overflow stacks.
Factor out this logic into a helper, so that we can avoid further
duplication when we add overflow stacks.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
This patch enables arm64 to be built with vmap'd task and IRQ stacks.
As vmap'd stacks are mapped at page granularity, stacks must be a multiple of
PAGE_SIZE. This means that a 64K page kernel must use stacks of at least 64K in
size.
To minimize the increase in Image size, IRQ stacks are dynamically allocated at
boot time, rather than embedding the boot CPU's IRQ stack in the kernel image.
This patch was co-authored by Ard Biesheuvel and Mark Rutland.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
We allocate our IRQ stacks using a percpu array. This allows us to generate our
IRQ stack pointers with adr_this_cpu, but bloats the kernel Image with the boot
CPU's IRQ stack. Additionally, these are packed with other percpu variables,
and aren't guaranteed to have guard pages.
When we enable VMAP_STACK we'll want to vmap our IRQ stacks also, in order to
provide guard pages and to permit more stringent alignment requirements. Doing
so will require that we use a percpu pointer to each IRQ stack, rather than
allocating a percpu IRQ stack in the kernel image.
This patch updates our IRQ stack code to use a percpu pointer to the base of
each IRQ stack. This will allow us to change the way the stack is allocated
with minimal changes elsewhere. In some cases we may try to backtrace before
the IRQ stack pointers are initialised, so on_irq_stack() is updated to account
for this.
In testing with cyclictest, there was no measureable difference between using
adr_this_cpu (for irq_stack) and ldr_this_cpu (for irq_stack_ptr) in the IRQ
entry path.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
In subsequent patches, we will detect stack overflow in our exception
entry code, by verifying the SP after it has been decremented to make
space for the exception regs.
This verification code is small, and we can minimize its impact by
placing it directly in the vectors. To avoid redundant modification of
the SP, we also need to move the initial decrement of the SP into the
vectors.
As a preparatory step, this patch introduces kernel_ventry, which
performs this decrement, and updates the entry code accordingly.
Subsequent patches will fold SP verification into kernel_ventry.
There should be no functional change as a result of this patch.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
[Mark: turn into prep patch, expand commit msg]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Currently we define SEGMENT_ALIGN directly in our vmlinux.lds.S.
This is unfortunate, as the EFI stub currently open-codes the same
number, and in future we'll want to fiddle with this.
This patch moves the definition to our <asm/memory.h>, where it can be
used by both vmlinux.lds.S and the EFI stub code.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Before we add yet another stack to the kernel, it would be nice to
ensure that we consistently organise stack definitions and related
helper functions.
This patch moves the basic IRQ stack defintions to <asm/memory.h> to
live with their task stack counterparts. Helpers used for unwinding are
moved into <asm/stacktrace.h>, where subsequent patches will add helpers
for other stacks. Includes are fixed up accordingly.
This patch is a pure refactoring -- there should be no functional
changes as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
For historical reasons, we leave the top 16 bytes of our task and IRQ
stacks unused, a practice used to ensure that the SP can always be
masked to find the base of the current stack (historically, where
thread_info could be found).
However, this is not necessary, as:
* When an exception is taken from a task stack, we decrement the SP by
S_FRAME_SIZE and stash the exception registers before we compare the
SP against the task stack. In such cases, the SP must be at least
S_FRAME_SIZE below the limit, and can be safely masked to determine
whether the task stack is in use.
* When transitioning to an IRQ stack, we'll place a dummy frame onto the
IRQ stack before enabling asynchronous exceptions, or executing code
we expect to trigger faults. Thus, if an exception is taken from the
IRQ stack, the SP must be at least 16 bytes below the limit.
* We no longer mask the SP to find the thread_info, which is now found
via sp_el0. Note that historically, the offset was critical to ensure
that cpu_switch_to() found the correct stack for new threads that
hadn't yet executed ret_from_fork().
Given that, this initial offset serves no purpose, and can be removed.
This brings us in-line with other architectures (e.g. x86) which do not
rely on this masking.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
[Mark: rebase, kill THREAD_START_SP, commit msg additions]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Our __die() implementation tries to dump the stack memory, in addition
to a backtrace, which is problematic.
For contemporary 16K stacks, this can be a lot of data, which can take a
long time to dump, and can push other useful context out of the kernel's
printk ringbuffer (and/or a user's scrollback buffer on an attached
console).
Additionally, the code implicitly assumes that the SP is on the task's
stack, and tries to dump everything between the SP and the highest task
stack address. When the SP points at an IRQ stack (or is corrupted),
this makes the kernel attempt to dump vast amounts of VA space. With
vmap'd stacks, this may result in erroneous accesses to peripherals.
This patch removes the memory dump, leaving us to rely on the backtrace,
and other means of dumping stack memory such as kdump.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
The Cortex-A35 uses some implementation defined perf events.
The Cortex-A35 derives from the Cortex-A53 core, using the same event mapings
based on Cortex-A35 TRM r0p2, section C2.3 - Performance monitoring events
(pages C2-562 to C2-565).
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The Cortex-A73 uses some implementation defined perf events.
This patch sets up the necessary mapping for Cortex-A73.
Mappings are based on Cortex-A73 TRM r0p2, section 11.9 Events
(pages 11-457 to 11-460).
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Now that the event mapping code always looks into the PMUv3 events
before any extended mappings, the extended mappings can be reduced to
only those events that are not discoverable through the PMCEID registers.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Last level caches and node events were almost never connected in current
supported cores.
We connect last level caches to the actual last level within the core and
node events are connected to bus accesses.
Signed-off-by: Julien Thierry <julien.thierry@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The unwind code sets the sp member of struct stackframe to
'frame pointer + 0x10' unconditionally, without regard for whether
doing so produces a legal value. So let's simply remove it now that
we have stopped using it anyway.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
As it turns out, the unwind code is slightly broken, and probably has
been for a while. The problem is in the dumping of the exception stack,
which is intended to dump the contents of the pt_regs struct at each
level in the call stack where an exception was taken and routed to a
routine marked as __exception (which means its stack frame is right
below the pt_regs struct on the stack).
'Right below the pt_regs struct' is ill defined, though: the unwind
code assigns 'frame pointer + 0x10' to the .sp member of the stackframe
struct at each level, and dump_backtrace() happily dereferences that as
the pt_regs pointer when encountering an __exception routine. However,
the actual size of the stack frame created by this routine (which could
be one of many __exception routines we have in the kernel) is not known,
and so frame.sp is pretty useless to figure out where struct pt_regs
really is.
So it seems the only way to ensure that we can find our struct pt_regs
when walking the stack frames is to put it at a known fixed offset of
the stack frame pointer that is passed to such __exception routines.
The simplest way to do that is to put it inside pt_regs itself, which is
the main change implemented by this patch. As a bonus, doing this allows
us to get rid of a fair amount of cruft related to walking from one stack
to the other, which is especially nice since we intend to introduce yet
another stack for overflow handling once we add support for vmapped
stacks. It also fixes an inconsistency where we only add a stack frame
pointing to ELR_EL1 if we are executing from the IRQ stack but not when
we are executing from the task stack.
To consistly identify exceptions regs even in the presence of exceptions
taken from entry code, we must check whether the next frame was created
by entry text, rather than whether the current frame was crated by
exception text.
To avoid backtracing using PCs that fall in the idmap, or are controlled
by userspace, we must explcitly zero the FP and LR in startup paths, and
must ensure that the frame embedded in pt_regs is zeroed upon entry from
EL0. To avoid these NULL entries showin in the backtrace, unwind_frame()
is updated to avoid them.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
[Mark: compare current frame against .entry.text, avoid bogus PCs]
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
vDSO VMA address is saved in mm_context for the purpose of using
restorer from vDSO page to return to userspace after signal handling.
In Checkpoint Restore in Userspace (CRIU) project we place vDSO VMA
on restore back to the place where it was on the dump.
With the exception for x86 (where there is API to map vDSO with
arch_prctl()), we move vDSO inherited from CRIU task to restoree
position by mremap().
CRIU does support arm64 architecture, but kernel doesn't update
context.vdso pointer after mremap(). Which results in translation
fault after signal handling on restored application:
https://github.com/xemul/criu/issues/288
Make vDSO code track the VMA address by supplying .mremap() fops
the same way it's done for x86 and arm32 by:
commit b059a453b1 ("x86/vdso: Add mremap hook to vm_special_mapping")
commit 280e87e98c ("ARM: 8683/1: ARM32: Support mremap() for sigpage/vDSO").
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Cc: Christopher Covington <cov@codeaurora.org>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Dmitry Safonov <dsafonov@virtuozzo.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Add a clean-to-point-of-persistence cache maintenance helper, and wire
up the basic architectural support for the pmem driver based on it.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
[catalin.marinas@arm.com: move arch_*_pmem() functions to arch/arm64/mm/flush.c]
[catalin.marinas@arm.com: change dmb(sy) to dmb(osh)]
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Cache clean to PoP is subject to the same access controls as to PoC, so
if we are trapping userspace cache maintenance with SCTLR_EL1.UCI, we
need to be prepared to handle it. To avoid getting into complicated
fights with binutils about ARMv8.2 options, we'll just cheat and use the
raw SYS instruction rather than the 'proper' DC alias.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
The ARMv8.2-DCPoP feature introduces persistent memory support to the
architecture, by defining a point of persistence in the memory
hierarchy, and a corresponding cache maintenance operation, DC CVAP.
Expose the support via HWCAP and MRS emulation.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
__inval_cache_range() is already the odd one out among our data cache
maintenance routines as the only remaining range-based one; as we're
going to want an invalidation routine to call from C code for the pmem
API, let's tweak the prototype and name to bring it in line with the
clean operations, and to make its relationship with __dma_inv_area()
neatly mirror that of __clean_dcache_area_poc() and __dma_clean_area().
The loop clearing the early page tables gets mildly massaged in the
process for the sake of consistency.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Rather than continue adding CPU-specific event maps, instead look up by
default in the PMUv3 event map and only fallback to the CPU-specific maps
if either the event isn't described by PMUv3, or it is described but
the PMCEID registers say that it is unsupported by the current CPU.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Currently, when unwinding the call stack, we validate the frame pointer
of each frame against frame.sp, whose value is not clearly defined, and
which makes it more difficult to link stack frames together across
different stacks. It is far better to simply check whether the frame
pointer itself points into a valid stack.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Our IRQ_STACK_PTR() and on_irq_stack() helpers both take a cpu argument,
used to generate a percpu address. In all cases, they are passed
{raw_,}smp_processor_id(), so this parameter is redundant.
Since {raw_,}smp_processor_id() use a percpu variable internally, this
approach means we generate a percpu offset to find the current cpu, then
use this to index an array of percpu offsets, which we then use to find
the current CPU's IRQ stack pointer. Thus, most of the work is
redundant.
Instead, we can consistently use raw_cpu_ptr() to generate the CPU's
irq_stack pointer by simply adding the percpu offset to the irq_stack
address, which is simpler in both respects.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Currently, cpu_switch_to and ret_from_fork both live in .entry.text,
though neither form the critical path for an exception entry.
In subsequent patches, we will require that code in .entry.text is part
of the critical path for exception entry, for which we can assume
certain properties (e.g. the presence of exception regs on the stack).
Neither cpu_switch_to nor ret_from_fork will meet these requirements, so
we must move them out of .entry.text. To ensure that neither are kprobed
after being moved out of .entry.text, we must explicitly blacklist them,
requiring a new NOKPROBE() asm helper.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
In most cases, our exception entry assembly branches to C handlers with
a BL instruction, but in cases where we do not expect to return, we use
B instead.
While this is correct today, it means that backtraces for fatal
exceptions miss the entry assembly (as the LR is stale at the point we
call C code), while non-fatal exceptions have the entry assembly in the
LR. In subsequent patches, we will need the LR to be set in these cases
in order to backtrace reliably.
This patch updates these sites to use a BL, ensuring consistency, and
preparing for backtrace rework. An ASM_BUG() is added after each of
these new BLs, which both catches unexpected returns, and ensures that
the LR value doesn't point to another function label.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Currently:
$ perf stat -e cycles:u -e cycles:k true
Performance counter stats for 'true':
2,24,699 cycles:u
<not counted> cycles:k (0.00%)
0.000788087 seconds time elapsed
We can not count more than one cycle counter in one instance,because we
allow to map cycle counter into PMCCNTR_EL0 only. However, if I did not
miss anything then specification do not prohibit to use PMEVCNTR<n>_EL0
for cycle count as well.
Modify the code so that it still prefers to use PMCCNTR_EL0 for cycle
counter, however allow to use PMEVCNTR<n>_EL0 if PMCCNTR_EL0 is already
in use.
After this patch:
$ perf stat -e cycles:u -e cycles:k true
Performance counter stats for 'true':
2,17,310 cycles:u
7,40,009 cycles:k
0.000764149 seconds time elapsed
Signed-off-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
may_use_simd() can be invoked from loadable modules and it accesses
kernel_neon_busy. Make sure that the latter is exported.
Fixes: cb84d11e16 ("arm64: neon: Remove support for nested or hardirq kernel-mode NEON")
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
The -1 "no syscall" value is written in various ways, shared with
the user ABI in some places, and generally obscure.
This patch attempts to make things a little more consistent and
readable by replacing all these uses with a single #define. A
couple of symbolic helpers are provided to clarify the intent
further.
Because the in-syscall check in do_signal() is changed from >= 0 to
!= NO_SYSCALL by this patch, different behaviour may be observable
if syscallno is set to values less than -1 by a tracer. However,
this is not different from the behaviour that is already observable
if a tracer sets syscallno to a value >= __NR_(compat_)syscalls.
It appears that this can cause spurious syscall restarting, but
that is not a new behaviour either, and does not appear harmful.
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
The upper 32 bits of the syscallno field in thread_struct are
handled inconsistently, being sometimes zero extended and sometimes
sign-extended. In fact, only the lower 32 bits seem to have any
real significance for the behaviour of the code: it's been OK to
handle the upper bits inconsistently because they don't matter.
Currently, the only place I can find where those bits are
significant is in calling trace_sys_enter(), which may be
unintentional: for example, if a compat tracer attempts to cancel a
syscall by passing -1 to (COMPAT_)PTRACE_SET_SYSCALL at the
syscall-enter-stop, it will be traced as syscall 4294967295
rather than -1 as might be expected (and as occurs for a native
tracer doing the same thing). Elsewhere, reads of syscallno cast
it to an int or truncate it.
There's also a conspicuous amount of code and casting to bodge
around the fact that although semantically an int, syscallno is
stored as a u64.
Let's not pretend any more.
In order to preserve the stp x instruction that stores the syscall
number in entry.S, this patch special-cases the layout of struct
pt_regs for big endian so that the newly 32-bit syscallno field
maps onto the low bits of the stored value. This is not beautiful,
but benchmarking of the getpid syscall on Juno suggests indicates a
minor slowdown if the stp is split into an stp x and stp w.
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Support for kernel-mode NEON to be nested and/or used in hardirq
context adds significant complexity, and the benefits may be
marginal. In practice, kernel-mode NEON is not used in hardirq
context, and is rarely used in softirq context (by certain mac80211
drivers).
This patch implements an arm64 may_use_simd() function to allow
clients to check whether kernel-mode NEON is usable in the current
context, and simplifies kernel_neon_{begin,end}() to handle only
saving of the task FPSIMD state (if any). Without nesting, there
is no other state to save.
The partial fpsimd save/restore functions become redundant as a
result of these changes, so they are removed too.
The save/restore model is changed to operate directly on
task_struct without additional percpu storage. This simplifies the
code and saves a bit of memory, but means that softirqs must now be
disabled when manipulating the task fpsimd state from task context:
correspondingly, preempt_{en,dis}sable() calls are upgraded to
local_bh_{en,dis}able() as appropriate. fpsimd_thread_switch()
already runs with hardirqs disabled and so is already protected
from softirqs.
These changes should make it easier to support kernel-mode NEON in
the presence of the Scalable Vector extension in the future.
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
In order to be able to cope with kernel-mode NEON being unavailable
in hardirq/nmi context and non-nestable, we need special handling
for EFI runtime service calls that may be made during an interrupt
that interrupted a kernel_neon_begin()..._end() block. This will
occur if the kernel tries to write diagnostic data to EFI
persistent storage during a panic triggered by an NMI for example.
EFI runtime services specify an ABI that clobbers the FPSIMD state,
rather than being able to use it optionally as an accelerator.
This means that EFI is really a special case and can be handled
specially.
To enable EFI calls from interrupts, this patch creates dedicated
__efi_fpsimd_{begin,end}() helpers solely for this purpose, which
save/restore to a separate percpu buffer if called in a context
where kernel_neon_begin() is not usable.
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
__this_cpu_ ops are not used consistently with regard to this_cpu_
ops in a couple of places in fpsimd.c.
Since preemption is explicitly disabled in
fpsimd_restore_current_state() and fpsimd_update_current_state(),
this patch converts this_cpu_ ops in those functions to __this_cpu_
ops. This doesn't save cost on arm64, but benefits from additional
assertions in the core code.
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Multiple architectures define this as a trivial function, and I'm adding
another one as part of the RISC-V port. Add a __weak version of
pcibios_align_resource() and delete the now-obselete ones in a handful of
ports.
The only functional change should be that a handful of ports used to export
pcibios_fixup_bus(). Only some architectures export this, so I just
dropped it.
Signed-off-by: Palmer Dabbelt <palmer@dabbelt.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Multiple architectures define this as an empty function, and I'm adding
another one as part of the RISC-V port. Add a __weak version of
pcibios_fixup_bus() and delete the now-obselete ones in a handful of
ports.
The only functional change should be that microblaze used to export
pcibios_fixup_bus(). None of the other architectures exports this, so I
just dropped it.
Signed-off-by: Palmer Dabbelt <palmer@dabbelt.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
