There is an explicit wait-type violation in debug_object_fill_pool()
for PREEMPT_RT=n kernels which allows them to more easily fill the
object pool and reduce the chance of allocation failures.
Lockdep's wait-type checks are designed to check the PREEMPT_RT
locking rules even for PREEMPT_RT=n kernels and object to this, so
create a lockdep annotation to allow this to stand.
Specifically, create a 'lock' type that overrides the inner wait-type
while it is held -- allowing one to temporarily raise it, such that
the violation is hidden.
Reported-by: Vlastimil Babka <[email protected]>
Reported-by: Qi Zheng <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Qi Zheng <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
Sync with the urgent patches; in particular:
a53ce18cac ("sched/fair: Sanitize vruntime of entity being migrated")
Signed-off-by: Peter Zijlstra <[email protected]>
Fundamentally semaphores are a counted primitive, but
DEFINE_SEMAPHORE() does not expose this and explicitly creates a
binary semaphore.
Change DEFINE_SEMAPHORE() to take a number argument and use that in the
few places that open-coded it using __SEMAPHORE_INITIALIZER().
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
[mcgrof: add some tribal knowledge about why some folks prefer
binary sempahores over mutexes]
Reviewed-by: Sergey Senozhatsky <[email protected]>
Reviewed-by: Davidlohr Bueso <[email protected]>
Signed-off-by: Luis Chamberlain <[email protected]>
Thomas reported that offlining CPUs spends a lot of time in
synchronize_rcu() as called from perf_pmu_migrate_context() even though
he's not actually using uncore events.
Turns out, the thing is unconditionally waiting for RCU, even if there's
no actual events to migrate.
Fixes: 0cda4c0231 ("perf: Introduce perf_pmu_migrate_context()")
Reported-by: Thomas Gleixner <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Thomas Gleixner <[email protected]>
Reviewed-by: Thomas Gleixner <[email protected]>
Reviewed-by: Paul E. McKenney <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
(Ab)use the trace_ipi_send_cpu*() family to trace all
smp_function_call*() invocations, not only those that result in an
actual IPI.
The queued entries log their callback function while the actual IPIs
are traced on generic_smp_call_function_single_interrupt().
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Things like ALTERNATIVE_{2,3}() generate multiple alternatives on the
same place, objtool would override the first orig_alt_group with the
second (or third), failing to check the CFI among all the different
variants.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Acked-by: Josh Poimboeuf <[email protected]>
Tested-by: Nathan Chancellor <[email protected]> # build only
Tested-by: Thomas Weißschuh <[email protected]> # compile and run
Link: https://lore.kernel.org/r/[email protected]
Tetsuo-San noted that commit f5d39b0208 ("freezer,sched: Rewrite
core freezer logic") broke call_usermodehelper_exec() for the KILLABLE
case.
Specifically it was missed that the second, unconditional,
wait_for_completion() was not optional and ensures the on-stack
completion is unused before going out-of-scope.
Fixes: f5d39b0208 ("freezer,sched: Rewrite core freezer logic")
Reported-by: [email protected]
Reported-by: Tetsuo Handa <[email protected]>
Debugged-by: Tetsuo Handa <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
For mysterious raisins I listed the new __asan_mem*() functions as
being uaccess safe, this is giving objtool fails on KASAN builds
because these functions call out to the actual __mem*() functions
which are not marked uaccess safe.
Removing it doesn't make the robots unhappy.
Fixes: 69d4c0d321 ("entry, kasan, x86: Disallow overriding mem*() functions")
Reported-by: "Paul E. McKenney" <[email protected]>
Bisected-by: Josh Poimboeuf <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lkml.kernel.org/r/20230126182302.GA687063@paulmck-ThinkPad-P17-Gen-1
Clang likes to create conditional tail calls like:
0000000000000350 <amd_pmu_add_event>:
350: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) 351: R_X86_64_NONE __fentry__-0x4
355: 48 83 bf 20 01 00 00 00 cmpq $0x0,0x120(%rdi)
35d: 0f 85 00 00 00 00 jne 363 <amd_pmu_add_event+0x13> 35f: R_X86_64_PLT32 __SCT__amd_pmu_branch_add-0x4
363: e9 00 00 00 00 jmp 368 <amd_pmu_add_event+0x18> 364: R_X86_64_PLT32 __x86_return_thunk-0x4
Where 0x35d is a static call site that's turned into a conditional
tail-call using the Jcc class of instructions.
Teach the in-line static call text patching about this.
Notably, since there is no conditional-ret, in that case patch the Jcc
to point at an empty stub function that does the ret -- or the return
thunk when needed.
Reported-by: "Erhard F." <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Reviewed-by: Masami Hiramatsu (Google) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
In order to re-write Jcc.d32 instructions text_poke_bp() needs to be
taught about them.
The biggest hurdle is that the whole machinery is currently made for 5
byte instructions and extending this would grow struct text_poke_loc
which is currently a nice 16 bytes and used in an array.
However, since text_poke_loc contains a full copy of the (s32)
displacement, it is possible to map the Jcc.d32 2 byte opcodes to
Jcc.d8 1 byte opcode for the int3 emulation.
This then leaves the replacement bytes; fudge that by only storing the
last 5 bytes and adding the rule that 'length == 6' instruction will
be prefixed with a 0x0f byte.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Reviewed-by: Masami Hiramatsu (Google) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
In order to use sched_clock() from noinstr code, mark it and all it's
implenentations noinstr.
The whole pvclock thing (used by KVM/Xen) is a bit of a pain,
since it calls out to watchdogs, create a
pvclock_clocksource_read_nowd() variant doesn't do that and can be
noinstr.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
In order to avoid WARN/BUG from generating nested or even recursive
warnings, force rcu_is_watching() true during
WARN/lockdep_rcu_suspicious().
Notably things like unwinding the stack can trigger rcu_dereference()
warnings, which then triggers more unwinding which then triggers more
warnings etc..
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
The recent cpuidle changes started triggering RCU splats on
Juno development boards:
| =============================
| WARNING: suspicious RCU usage
| -----------------------------
| include/trace/events/ipi.h:19 suspicious rcu_dereference_check() usage!
Fix cpuidle on ARM64:
- ... by introducing a new 'is_rcu' flag to the cpuidle helpers & make
ARM64 use it, as ARM64 wants to keep RCU active longer and wants to
do the ct_cpuidle_enter()/exit() dance itself.
- Also update the PSCI driver accordingly.
- This also removes the last known RCU_NONIDLE() user as a bonus.
Reported-by: Mark Rutland <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Sudeep Holla <[email protected]>
Tested-by: Mark Rutland <[email protected]>
Reviewed-by: Mark Rutland <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
--
For all cpuidle drivers that do not use CPUIDLE_FLAG_RCU_IDLE (iow,
the simple ones) make sure all the functions are marked __cpuidle.
( due to lack of noinstr validation on these platforms it is entirely
possible this isn't complete )
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
For all cpuidle drivers that use CPUIDLE_FLAG_RCU_IDLE, ensure that
all functions that call ct_cpuidle_enter() are marked __cpuidle.
( due to lack of noinstr validation on these platforms it is entirely
possible this isn't complete )
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Tracing (kprobes included) and other compiler instrumentation relies
on a normal kernel runtime. Therefore all functions that disable RCU
should be noinstr, as should all functions that are called while RCU
is disabled.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
arch_cpu_idle() is a very simple idle interface and exposes only a
single idle state and is expected to not require RCU and not do any
tracing/instrumentation.
As such, omap2_pm_idle() is not a valid implementation. Replace it
with a simple (shallow) omap2_do_wfi() call.
Omap2 doesn't have a cpuidle driver; but adding one would be the
recourse to (re)gain the other idle states.
Suggested-by: Tony Lindgren <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
arch_cpu_idle() is a very simple idle interface and exposes only a
single idle state and is expected to not require RCU and not do any
tracing/instrumentation.
As such, omap_sram_idle() is not a valid implementation. Replace it
with the simple (shallow) omap3_do_wfi() call. Leaving the more
complicated idle states for the cpuidle driver.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Acked-by: Tony Lindgren <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Robot reported that trace_hardirqs_{on,off}() tickle the forbidden
_rcuidle() tracepoint through local_irq_{en,dis}able().
For 'sane' configs, these calls will only happen with RCU enabled and
as such can use the regular tracepoint. This also means it's possible
to trace them from NMI context again.
Reported-by: kernel test robot <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
objtool found that the mds_idle_clear_cpu_buffers() method got
uninlined by the compiler where it called out into instrumentation:
vmlinux.o: warning: objtool: mwait_idle+0x47: call to mds_idle_clear_cpu_buffers() leaves .noinstr.text section
vmlinux.o: warning: objtool: acpi_processor_ffh_cstate_enter+0xa2: call to mds_idle_clear_cpu_buffers() leaves .noinstr.text section
vmlinux.o: warning: objtool: intel_idle+0x91: call to mds_idle_clear_cpu_buffers() leaves .noinstr.text section
vmlinux.o: warning: objtool: intel_idle_s2idle+0x8c: call to mds_idle_clear_cpu_buffers() leaves .noinstr.text section
vmlinux.o: warning: objtool: intel_idle_irq+0xaa: call to mds_idle_clear_cpu_buffers() leaves .noinstr.text section
Solve this by marking mds_idle_clear_cpu_buffers() as __always_inline.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
objtool found a few cases where this code called out into instrumented
code:
vmlinux.o: warning: objtool: acpi_idle_enter_s2idle+0xde: call to wbinvd() leaves .noinstr.text section
vmlinux.o: warning: objtool: default_idle+0x4: call to arch_safe_halt() leaves .noinstr.text section
vmlinux.o: warning: objtool: xen_safe_halt+0xa: call to HYPERVISOR_sched_op.constprop.0() leaves .noinstr.text section
Solve this by:
- marking arch_safe_halt(), wbinvd(), native_wbinvd() and
HYPERVISOR_sched_op() as __always_inline().
- Explicitly uninlining xen_safe_halt() and pv_native_wbinvd() [they were
already uninlined by the compiler on use as function pointers] and
annotating them as 'noinstr'.
- Annotating pv_native_safe_halt() as 'noinstr'.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Reviewed-by: Srivatsa S. Bhat (VMware) <[email protected]>
Reviewed-by: Juergen Gross <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
objtool found a few cases where this code called out into instrumented
code:
vmlinux.o: warning: objtool: intel_idle_s2idle+0x6e: call to __monitor.constprop.0() leaves .noinstr.text section
vmlinux.o: warning: objtool: intel_idle_irq+0x8c: call to __monitor.constprop.0() leaves .noinstr.text section
vmlinux.o: warning: objtool: intel_idle+0x73: call to __monitor.constprop.0() leaves .noinstr.text section
vmlinux.o: warning: objtool: mwait_idle+0x88: call to clflush() leaves .noinstr.text section
Fix it by marking the affected methods as __always_inline.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
None of these functions should ever be ran with RCU disabled anymore.
Specifically, do_handle_IPI() is only called from handle_IPI() which
explicitly does irq_enter()/irq_exit() which ensures RCU is watching.
The problem with smp_cross_call() was, per commit description:
7c64cc0531 ("arm: Use _rcuidle for smp_cross_call() tracepoints")
... that cpuidle_enter_state_coupled() already had RCU disabled, but that's
long been fixed by commit:
1098582a0f ("sched,idle,rcu: Push rcu_idle deeper into the idle path")
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Reviewed-by: Ulf Hansson <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Current arch_cpu_idle() is called with IRQs disabled, but will return
with IRQs enabled.
However, the very first thing the generic code does after calling
arch_cpu_idle() is raw_local_irq_disable(). This means that
architectures that can idle with IRQs disabled end up doing a
pointless 'enable-disable' dance.
Therefore, push this IRQ disabling into the idle function, meaning
that those architectures can avoid the pointless IRQ state flipping.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Reviewed-by: Gautham R. Shenoy <[email protected]>
Acked-by: Mark Rutland <[email protected]> [arm64]
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Guo Ren <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
So objtool found this bug:
vmlinux.o: warning: objtool: intel_idle_irq+0x10c: call to trace_hardirqs_off() leaves .noinstr.text section
As per commit 32d4fd5751 ("cpuidle,intel_idle: Fix CPUIDLE_FLAG_IRQ_ENABLE"):
"must not have tracing in idle functions"
Clearly people can't read and tinker along until splat dissapears.
This straight up reverts commit d295ad34f2 ("intel_idle: Fix false
positive RCU splats due to incorrect hardirqs state").
It doesn't re-introduce the problem because preceding patches fixed it
properly.
Fixes: d295ad34f2 ("intel_idle: Fix false positive RCU splats due to incorrect hardirqs state")
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
The whole disable-RCU, enable-IRQS dance is very intricate since
changing IRQ state is traced, which depends on RCU.
Add two helpers for the cpuidle case that mirror the entry code:
ct_cpuidle_enter()
ct_cpuidle_exit()
And fix all the cases where the enter/exit dance was buggy.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
Tested-by: Ulf Hansson <[email protected]>
Acked-by: Rafael J. Wysocki <[email protected]>
Acked-by: Frederic Weisbecker <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
With 'GNU assembler (GNU Binutils for Debian) 2.39.90.20221231' the
build now reports:
arch/x86/realmode/rm/../../boot/bioscall.S: Assembler messages:
arch/x86/realmode/rm/../../boot/bioscall.S:35: Warning: found `movsd'; assuming `movsl' was meant
arch/x86/realmode/rm/../../boot/bioscall.S:70: Warning: found `movsd'; assuming `movsl' was meant
arch/x86/boot/bioscall.S: Assembler messages:
arch/x86/boot/bioscall.S:35: Warning: found `movsd'; assuming `movsl' was meant
arch/x86/boot/bioscall.S:70: Warning: found `movsd'; assuming `movsl' was meant
Which is due to:
PR gas/29525
Note that with the dropped CMPSD and MOVSD Intel Syntax string insn
templates taking operands, mixed IsString/non-IsString template groups
(with memory operands) cannot occur anymore. With that
maybe_adjust_templates() becomes unnecessary (and is hence being
removed).
More details: https://sourceware.org/bugzilla/show_bug.cgi?id=29525
Borislav Petkov further explains:
" the particular problem here is is that the 'd' suffix is
"conflicting" in the sense that you can have SSE mnemonics like movsD %xmm...
and the same thing also for string ops (which is the case here) so apparently
the agreement in binutils land is to use the always accepted suffixes 'l' or 'q'
and phase out 'd' slowly... "
Fixes: 7a734e7dd9 ("x86, setup: "glove box" BIOS calls -- infrastructure")
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Acked-by: Borislav Petkov (AMD) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
There is a cludge in change_page_attr_set_clr() that inhibits
propagating NX changes to the aliases (directmap and highmap) -- this
is a cludge twofold:
- it also inhibits the primary checks in __change_page_attr();
- it hard depends on single bit changes.
The introduction of set_memory_rox() triggered this last issue for
clearing both _PAGE_RW and _PAGE_NX.
Explicitly ignore _PAGE_NX in cpa_process_alias() instead.
Fixes: b38994948567 ("x86/mm: Implement native set_memory_rox()")
Reported-by: kernel test robot <[email protected]>
Debugged-by: Dave Hansen <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lkml.kernel.org/r/20221110125544.594991716%40infradead.org
The .checkalias argument to __change_page_attr_set_clr() is overloaded
and serves two different purposes:
- it inhibits the call to cpa_process_alias() -- as suggested by the
name; however,
- it also serves as 'primary' indicator for __change_page_attr()
( which in turn also serves as a recursion terminator for
cpa_process_alias() ).
Untangle these by extending the use of CPA_NO_CHECK_ALIAS to all
callsites that currently use .checkalias=0 for this purpose.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lkml.kernel.org/r/20221110125544.527267183%40infradead.org
Given that ptep_get_and_clear() uses cmpxchg8b, and that should be by
far the most common case, there's no point in having an optimized
variant for pmd/pud.
Introduce the pxx_xchg64() helper to implement the common logic once.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lkml.kernel.org/r/20221022114425.103392961%40infradead.org
Provide a native implementation of set_memory_rox(), avoiding the
double set_memory_ro();set_memory_x(); calls.
Suggested-by: Linus Torvalds <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Peter Zijlstra