A typical backtrace acquired from ftraced function currently looks like
the following (e.g. for "path_openat"):
arch_stack_walk+0x15c/0x2d8
stack_trace_save+0x50/0x68
stack_trace_call+0x15a/0x3b8
ftrace_graph_caller+0x0/0x1c
0x3e0007e3c98 <- ftraced function caller (should be do_filp_open+0x7c/0xe8)
do_open_execat+0x70/0x1b8
__do_execve_file.isra.0+0x7d8/0x860
__s390x_sys_execve+0x56/0x68
system_call+0xdc/0x2d8
Note random "0x3e0007e3c98" stack value as ftraced function caller. This
value causes either imprecise unwinder result or unwinding failure.
That "0x3e0007e3c98" comes from r14 of ftraced function stack frame, which
it haven't had a chance to initialize since the very first instruction
calls ftrace code ("ftrace_caller"). (ftraced function might never
save r14 as well). Nevertheless according to s390 ABI any function
is called with stack frame allocated for it and r14 contains return
address. "ftrace_caller" itself is called with "brasl %r0,ftrace_caller".
So, to fix this issue simply always save traced function caller onto
ftraced function stack frame.
Reported-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Consider reaching user mode pt_regs at the bottom of irq stack graceful
unwinder termination. This is the case when irq/mcck/ext interrupt arrives
while in user mode.
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
diag 0x44 is a voluntary undirected yield of a virtual CPU. This has
caused a lot of performance issues in the past.
There is only one caller left, and that one is only executed if diag
0x9c (directed yield) is not present. Given that all hypervisors
implement diag 0x9c anyway, remove the last diag 0x44 to avoid that
more callers will be added.
Worst case that could happen now, if diag 0x9c is not present, is that
a virtual CPU would loop a bit instead of giving its time slice up.
diag 0x44 statistics in debugfs are kept and will always be zero, so
that user space can tell that there are no calls.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
The s390 CPU Measurement sampling facility has an overflow condition
which fires when all entries in a SBD are used.
The measurement alert interrupt is triggered and reads out all samples
in this SDB. It then tests the successor SDB, if this SBD is not full,
the interrupt handler does not read any samples at all from this SDB
The design waits for the hardware to fill this SBD and then trigger
another meassurement alert interrupt.
This scheme works nicely until
an perf_event_overflow() function call discards the sample due to
a too high sampling rate.
The interrupt handler has logic to read out a partially filled SDB
when the perf event overflow condition in linux common code is met.
This causes the CPUM sampling measurement hardware and the PMU
device driver to operate on the same SBD's trailer entry.
This should not happen.
This can be seen here using this trace:
cpumsf_pmu_add: tear:0xb5286000
hw_perf_event_update: sdbt 0xb5286000 full 1 over 0 flush_all:0
hw_perf_event_update: sdbt 0xb5286008 full 0 over 0 flush_all:0
above shows 1. interrupt
hw_perf_event_update: sdbt 0xb5286008 full 1 over 0 flush_all:0
hw_perf_event_update: sdbt 0xb5286008 full 0 over 0 flush_all:0
above shows 2. interrupt
... this goes on fine until...
hw_perf_event_update: sdbt 0xb5286068 full 1 over 0 flush_all:0
perf_push_sample1: overflow
one or more samples read from the IRQ handler are rejected by
perf_event_overflow() and the IRQ handler advances to the next SDB
and modifies the trailer entry of a partially filled SDB.
hw_perf_event_update: sdbt 0xb5286070 full 0 over 0 flush_all:1
timestamp: 14:32:52.519953
Next time the IRQ handler is called for this SDB the trailer entry shows
an overflow count of 19 missed entries.
hw_perf_event_update: sdbt 0xb5286070 full 1 over 19 flush_all:1
timestamp: 14:32:52.970058
Remove access to a follow on SDB when event overflow happened.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Function perf_event_ever_overflow() and perf_event_account_interrupt()
are called every time samples are processed by the interrupt handler.
However function perf_event_account_interrupt() has checks to avoid being
flooded with interrupts (more then 1000 samples are received per
task_tick). Samples are then dropped and a PERF_RECORD_THROTTLED is
added to the perf data. The perf subsystem limit calculation is:
maximum sample frequency := 100000 --> 1 samples per 10 us
task_tick = 10ms = 10000us --> 1000 samples per task_tick
The work flow is
measurement_alert() uses SDBT head and each SBDT points to 511
SDB pages, each with 126 sample entries. After processing 8 SBDs
and for each valid sample calling:
perf_event_overflow()
perf_event_account_interrupts()
there is a considerable amount of samples being dropped, especially when
the sample frequency is very high and near the 100000 limit.
To avoid the high amount of samples being dropped near the end of a
task_tick time frame, increment the sampling interval in case of
dropped events. The CPU Measurement sampling facility on the s390
supports only intervals, specifiing how many CPU cycles have to be
executed before a sample is generated. Increase the interval when the
samples being generated hit the task_tick limit.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
- Make stack unwinder reliable and suitable for livepatching. Add unwinder
testing module.
- Fixes for CALL_ON_STACK helper used for stack switching.
- Fix unwinding from bpf code.
- Fix getcpu and remove compat support in vdso code.
- Fix address space control registers initialization.
- Save KASLR offset for early dumps.
- Handle new FILTERED_BY_HYPERVISOR reply code in crypto code.
- Minor perf code cleanup and potential memory leak fix.
- Add couple of error messages for corner cases during PCI device
creation.
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEE3QHqV+H2a8xAv27vjYWKoQLXFBgFAl3mUEMACgkQjYWKoQLX
FBgHBgf/Ui3sgKGozvIAwy2kQ3oPtCdsmnTKEhLdhYT0cKMWNkA/jc13vn37ZqSk
vMhawMjgjHhn4CLSjxKRGCprYViXIgnF2XrCywTDsBoj87QwB6/dME1gXJRW+/Rm
OPvO+8D+210Ow0Xip3xXSRIPNFsUINCQeCCEtQCOuhGMdQPC0VIKgYtgvk1TAo1E
+DycHbZ0e+uEp6zvVSsoP9wrkXw/L9krTDnjHncQ7FULJAYnBhY+qaeNTek09QAT
j3Ywh5/fYR11c62W6fjb1lQHLb75L0aeK7Q5r5WspxG5LwiR2ncYWOQ4BQPZoUXq
GjdNvwRmvEkB3IbnpLp/ft7sqsPn2w==
=CoqQ
-----END PGP SIGNATURE-----
Merge tag 's390-5.5-2' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Pull more s390 updates from Vasily Gorbik:
- Make stack unwinder reliable and suitable for livepatching. Add
unwinder testing module.
- Fixes for CALL_ON_STACK helper used for stack switching.
- Fix unwinding from bpf code.
- Fix getcpu and remove compat support in vdso code.
- Fix address space control registers initialization.
- Save KASLR offset for early dumps.
- Handle new FILTERED_BY_HYPERVISOR reply code in crypto code.
- Minor perf code cleanup and potential memory leak fix.
- Add couple of error messages for corner cases during PCI device
creation.
* tag 's390-5.5-2' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (33 commits)
s390: remove compat vdso code
s390/livepatch: Implement reliable stack tracing for the consistency model
s390/unwind: add stack pointer alignment sanity checks
s390/unwind: filter out unreliable bogus %r14
s390/unwind: start unwinding from reliable state
s390/test_unwind: add program check context tests
s390/test_unwind: add irq context tests
s390/test_unwind: print verbose unwinding results
s390/test_unwind: add CALL_ON_STACK tests
s390: fix register clobbering in CALL_ON_STACK
s390/test_unwind: require that unwinding ended successfully
s390/unwind: add a test for the internal API
s390/unwind: always inline get_stack_pointer
s390/pci: add error message on device number limit
s390/pci: add error message for UID collision
s390/cpum_sf: Check for SDBT and SDB consistency
s390/cpum_sf: Use TEAR_REG macro consistantly
s390/cpum_sf: Remove unnecessary check for pending SDBs
s390/cpum_sf: Replace function name in debug statements
s390/kaslr: store KASLR offset for early dumps
...
Remove compat vdso code, since there is hardly any compat user space
left. Still existing compat user space will have to use system calls
instead.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
The livepatch consistency model requires reliable stack tracing
architecture support in order to work properly. In order to achieve
this, two main issues have to be solved. First, reliable and consistent
call chain backtracing has to be ensured. Second, the unwinder needs to
be able to detect stack corruptions and return errors.
The "zSeries ELF Application Binary Interface Supplement" says:
"The stack pointer points to the first word of the lowest allocated
stack frame. If the "back chain" is implemented this word will point to
the previously allocated stack frame (towards higher addresses), except
for the first stack frame, which shall have a back chain of zero (NULL).
The stack shall grow downwards, in other words towards lower addresses."
"back chain" is optional. GCC option -mbackchain enables it. Quoting
Martin Schwidefsky [1]:
"The compiler is called with the -mbackchain option, all normal C
function will store the backchain in the function prologue. All
functions written in assembler code should do the same, if you find one
that does not we should fix that. The end result is that a task that
*voluntarily* called schedule() should have a proper backchain at all
times.
Dependent on the use case this may or may not be enough. Asynchronous
interrupts may stop the CPU at the beginning of a function, if kernel
preemption is enabled we can end up with a broken backchain. The
production kernels for IBM Z are all compiled *without* kernel
preemption. So yes, we might get away without the objtool support.
On a side-note, we do have a line item to implement the ORC unwinder for
the kernel, that includes the objtool support. Once we have that we can
drop the -mbackchain option for the kernel build. That gives us a nice
little performance benefit. I hope that the change from backchain to the
ORC unwinder will not be too hard to implement in the livepatch tools."
Since -mbackchain is enabled by default when the kernel is compiled, the
call chain backtracing should be currently ensured and objtool should
not be necessary for livepatch purposes.
Regarding the second issue, stack corruptions and non-reliable states
have to be recognized by the unwinder. Mainly it means to detect
preemption or page faults, the end of the task stack must be reached,
return addresses must be valid text addresses and hacks like function
graph tracing and kretprobes must be properly detected.
Unwinding a running task's stack is not a problem, because there is a
livepatch requirement that every checked task is blocked, except for the
current task. Due to that, the implementation can be much simpler
compared to the existing non-reliable infrastructure. We can consider a
task's kernel/thread stack only and skip the other stacks.
[1] 20180912121106.31ffa97c@mschwideX1 [not archived on lore.kernel.org]
Link: https://lkml.kernel.org/r/20191106095601.29986-5-mbenes@suse.cz
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Tested-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Currently unwinder unconditionally returns %r14 from the first frame
pointed by %r15 from pt_regs. A task could be interrupted when a function
already allocated this frame (if it needs it) for its callees or to
store local variables. In that case this frame would contain random
values from stack or values stored there by a callee. As we are only
interested in %r14 to get potential return address, skip bogus return
addresses which doesn't belong to kernel text.
This helps to avoid duplicating filtering logic in unwider users, most
of which use unwind_get_return_address() and would choke on bogus 0
address returned by it otherwise.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
A comment in arch/s390/include/asm/unwind.h says:
> If 'first_frame' is not zero unwind_start skips unwind frames until it
> reaches the specified stack pointer.
> The end of the unwinding is indicated with unwind_done, this can be true
> right after unwind_start, e.g. with first_frame!=0 that can not be found.
> unwind_next_frame skips to the next frame.
> Once the unwind is completed unwind_error() can be used to check if there
> has been a situation where the unwinder could not correctly understand
> the tasks call chain.
With this change backchain unwinder now comply with behaviour
described. As well as matches orc unwinder implementation. Now unwinder
starts from reliable state, i.e. __unwind_start own stack frame is
taken or stack frame generated by __switch_to (ksp) - both known to be
valid. In case of pt_regs %r15 is better match for pt_regs psw, than
sometimes random "sp" caller passed.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Add stack name, sp and reliable information into test unwinding
results. Also consider ip outside of kernel text as failure if the
state is reported reliable.
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Each SBDT is located at a 4KB page and contains 512 entries.
Each entry of a SDBT points to a SDB, a 4KB page containing
sampled data. The last entry is a link to another SDBT page.
When an event is created the function sequence executed is:
__hw_perf_event_init()
+--> allocate_buffers()
+--> realloc_sampling_buffers()
+---> alloc_sample_data_block()
Both functions realloc_sampling_buffers() and
alloc_sample_data_block() allocate pages and the allocation
can fail. This is handled correctly and all allocated
pages are freed and error -ENOMEM is returned to the
top calling function. Finally the event is not created.
Once the event has been created, the amount of initially
allocated SDBT and SDB can be too low. This is detected
during measurement interrupt handling, where the amount
of lost samples is calculated. If the number of lost samples
is too high considering sampling frequency and already allocated
SBDs, the number of SDBs is enlarged during the next execution
of cpumsf_pmu_enable().
If more SBDs need to be allocated, functions
realloc_sampling_buffers()
+---> alloc-sample_data_block()
are called to allocate more pages. Page allocation may fail
and the returned error is ignored. A SDBT and SDB setup
already exists.
However the modified SDBTs and SDBs might end up in a situation
where the first entry of an SDBT does not point to an SDB,
but another SDBT, basicly an SBDT without payload.
This can not be handled by the interrupt handler, where an SDBT
must have at least one entry pointing to an SBD.
Add a check to avoid SDBTs with out payload (SDBs) when enlarging
the buffer setup.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
The macro TEAR_REG() saves the last used SDBT address
in the perf_hw_event structure. This is also done
by function hw_reset_registers() which is a one-liner
and simply uses macro TEAR_REG(). Remove function
hw_reset_registers(), which is only used one time and use
macro TEAR_REG() instead. This macro is used throughout
the code anyway.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
In interrupt handling the function extend_sampling_buffer()
is called after checking for a possibly extension.
This check is not necessary as the called function itself
performs this check again.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Replace hard coded function names in debug statements
by the "%s ...", __func__ construct suggested by checkpatch.pl
script. Use consistent debug print format of the form variable
blank value. Also add leading 0x for all hex values.
Print allocated page addresses consistantly as hex numbers
with leading 0x.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
The KASLR offset is added to vmcoreinfo in arch_crash_save_vmcoreinfo(),
so that it can be found by crash when processing kernel dumps.
However, arch_crash_save_vmcoreinfo() is called during a subsys_initcall,
so if the kernel crashes before that, we have no vmcoreinfo and no KASLR
offset.
Fix this by storing the KASLR offset in the lowcore, where the vmcore_info
pointer will be stored, and where it can be found by crash. In order to
make it distinguishable from a real vmcore_info pointer, mark it as uneven
(KASLR offset itself is aligned to THREAD_SIZE).
When arch_crash_save_vmcoreinfo() stores the real vmcore_info pointer in
the lowcore, it overwrites the KASLR offset. At that point, the KASLR
offset is not yet added to vmcoreinfo, so we also need to move the
mem_assign_absolute() behind the vmcoreinfo_append_str().
Fixes: b2d24b97b2 ("s390/kernel: add support for kernel address space layout randomization (KASLR)")
Cc: <stable@vger.kernel.org> # v5.2+
Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Consider reaching task pt_regs graceful unwinder termination. Task
pt_regs itself never contains a valid state to which a task might return
within the kernel context (user task pt_regs is a special case). Since
we already avoid printing user task pt_regs and in most cases we don't
even bother filling task pt_regs psw and r15 with something reasonable
simply skip task pt_regs altogether. With this change unwind_error() now
accurately represent whether unwinder reached task pt_regs successfully
or failed along the way.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Add missing allocation of pt_regs at the bottom of the stack. This
makes it consistent with other stack setup cases and also what stack
unwinder expects.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Currently unwinder yields 2 entries when pt_regs are met:
sp="address of pt_regs itself" ip=pt_regs->psw
sp=pt_regs->gprs[15] ip="r14 from stack frame pointed by pt_regs->gprs[15]"
And neither of those 2 states (combination of sp and ip) ever happened.
reuse_sp has been introduced by commit a1d863ac3e ("s390/unwind: fix
mixing regs and sp"). reuse_sp=true makes unwinder keen to produce the
following result, when pt_regs are given (as an arg to unwind_start):
sp=pt_regs->gprs[15] ip=pt_regs->psw
sp=pt_regs->gprs[15] ip="r14 from stack frame pointed by pt_regs->gprs[15]"
The first state is an actual state in which a task was when pt_regs were
collected. The second state is marked unreliable and is for debugging
purposes to cover the case when a task has been interrupted in between
stack frame allocation and writing back_chain - in this case r14 might
show an actual caller.
Make unwinder behaviour enabled via reuse_sp=true default and drop the
special case handling.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
If unwinder is looking at pt_regs which is not on stack then something
went wrong and an error has to be reported rather than successful
unwinding termination.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
CALL_ON_STACK is intended to be used for temporary stack switching with
potential return to the caller.
When CALL_ON_STACK is misused to switch from nodat stack to task stack
back_chain information would later lead stack unwinder from task stack into
(per cpu) nodat stack which is reused for other purposes. This would
yield confusing unwinding result or errors.
To avoid that introduce CALL_ON_STACK_NORETURN to be used instead. It
makes sure that back_chain is zeroed and unwinder finishes gracefully
ending up at task pt_regs.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Avoid mixture of task == NULL and task == current meaning the same
thing and simply always initialize task with current in unwind_start.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Make sure preemption is disabled when temporary switching to nodat
stack with CALL_ON_STACK helper, because nodat stack is per cpu.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
getcpu reads the required values for cpu and node with two
instructions. This might lead to an inconsistent result if user space
gets preempted and migrated to a different CPU between the two
instructions.
Fix this by using just a single instruction to read both values at
once.
This is currently rather a theoretical bug, since there is no real
NUMA support available (except for NUMA emulation).
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
When a secondary CPU is brought up it must initialize its control
registers. CPU A which triggers that a secondary CPU B is brought up
stores its control register contents into the lowcore of new CPU B,
which then loads these values on startup.
This is problematic in various ways: the control register which
contains the home space ASCE will correctly contain the kernel ASCE;
however control registers for primary and secondary ASCEs are
initialized with whatever values were present in CPU A.
Typically:
- the primary ASCE will contain the user process ASCE of the process
that triggered onlining of CPU B.
- the secondary ASCE will contain the percpu VDSO ASCE of CPU A.
Due to lazy ASCE handling we may also end up with other combinations.
When then CPU B switches to a different process (!= idle) it will
fixup the primary ASCE. However the problem is that the (wrong) ASCE
from CPU A was loaded into control register 1: as soon as an ASCE is
attached (aka loaded) a CPU is free to generate TLB entries using that
address space.
Even though it is very unlikey that CPU B will actually generate such
entries, this could result in TLB entries of the address space of the
process that ran on CPU A. These entries shouldn't exist at all and
could cause problems later on.
Furthermore the secondary ASCE of CPU B will not be updated correctly.
This means that processes may see wrong results or even crash if they
access VDSO data on CPU B. The correct VDSO ASCE will eventually be
loaded on return to user space as soon as the kernel executed a call
to strnlen_user or an atomic futex operation on CPU B.
Fix both issues by intializing the to be loaded control register
contents with the correct ASCEs and also enforce (re-)loading of the
ASCEs upon first context switch and return to user space.
Fixes: 0aaba41b58 ("s390: remove all code using the access register mode")
Cc: stable@vger.kernel.org # v4.15+
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Pull scheduler updates from Ingo Molnar:
"The biggest changes in this cycle were:
- Make kcpustat vtime aware (Frederic Weisbecker)
- Rework the CFS load_balance() logic (Vincent Guittot)
- Misc cleanups, smaller enhancements, fixes.
The load-balancing rework is the most intrusive change: it replaces
the old heuristics that have become less meaningful after the
introduction of the PELT metrics, with a grounds-up load-balancing
algorithm.
As such it's not really an iterative series, but replaces the old
load-balancing logic with the new one. We hope there are no
performance regressions left - but statistically it's highly probable
that there *is* going to be some workload that is hurting from these
chnages. If so then we'd prefer to have a look at that workload and
fix its scheduling, instead of reverting the changes"
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (46 commits)
rackmeter: Use vtime aware kcpustat accessor
leds: Use all-in-one vtime aware kcpustat accessor
cpufreq: Use vtime aware kcpustat accessors for user time
procfs: Use all-in-one vtime aware kcpustat accessor
sched/vtime: Bring up complete kcpustat accessor
sched/cputime: Support other fields on kcpustat_field()
sched/cpufreq: Move the cfs_rq_util_change() call to cpufreq_update_util()
sched/fair: Add comments for group_type and balancing at SD_NUMA level
sched/fair: Fix rework of find_idlest_group()
sched/uclamp: Fix overzealous type replacement
sched/Kconfig: Fix spelling mistake in user-visible help text
sched/core: Further clarify sched_class::set_next_task()
sched/fair: Use mul_u32_u32()
sched/core: Simplify sched_class::pick_next_task()
sched/core: Optimize pick_next_task()
sched/core: Make pick_next_task_idle() more consistent
sched/fair: Better document newidle_balance()
leds: Use vtime aware kcpustat accessor to fetch CPUTIME_SYSTEM
cpufreq: Use vtime aware kcpustat accessor to fetch CPUTIME_SYSTEM
procfs: Use vtime aware kcpustat accessor to fetch CPUTIME_SYSTEM
...
Pull x86 asm updates from Ingo Molnar:
"The main changes in this cycle were:
- Cross-arch changes to move the linker sections for NOTES and
EXCEPTION_TABLE into the RO_DATA area, where they belong on most
architectures. (Kees Cook)
- Switch the x86 linker fill byte from x90 (NOP) to 0xcc (INT3), to
trap jumps into the middle of those padding areas instead of
sliding execution. (Kees Cook)
- A thorough cleanup of symbol definitions within x86 assembler code.
The rather randomly named macros got streamlined around a
(hopefully) straightforward naming scheme:
SYM_START(name, linkage, align...)
SYM_END(name, sym_type)
SYM_FUNC_START(name)
SYM_FUNC_END(name)
SYM_CODE_START(name)
SYM_CODE_END(name)
SYM_DATA_START(name)
SYM_DATA_END(name)
etc - with about three times of these basic primitives with some
label, local symbol or attribute variant, expressed via postfixes.
No change in functionality intended. (Jiri Slaby)
- Misc other changes, cleanups and smaller fixes"
* 'x86-asm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (67 commits)
x86/entry/64: Remove pointless jump in paranoid_exit
x86/entry/32: Remove unused resume_userspace label
x86/build/vdso: Remove meaningless CFLAGS_REMOVE_*.o
m68k: Convert missed RODATA to RO_DATA
x86/vmlinux: Use INT3 instead of NOP for linker fill bytes
x86/mm: Report actual image regions in /proc/iomem
x86/mm: Report which part of kernel image is freed
x86/mm: Remove redundant address-of operators on addresses
xtensa: Move EXCEPTION_TABLE to RO_DATA segment
powerpc: Move EXCEPTION_TABLE to RO_DATA segment
parisc: Move EXCEPTION_TABLE to RO_DATA segment
microblaze: Move EXCEPTION_TABLE to RO_DATA segment
ia64: Move EXCEPTION_TABLE to RO_DATA segment
h8300: Move EXCEPTION_TABLE to RO_DATA segment
c6x: Move EXCEPTION_TABLE to RO_DATA segment
arm64: Move EXCEPTION_TABLE to RO_DATA segment
alpha: Move EXCEPTION_TABLE to RO_DATA segment
x86/vmlinux: Move EXCEPTION_TABLE to RO_DATA segment
x86/vmlinux: Actually use _etext for the end of the text segment
vmlinux.lds.h: Allow EXCEPTION_TABLE to live in RO_DATA
...
- Adjust PMU device drivers registration to avoid WARN_ON and few other
perf improvements.
- Enhance tracing in vfio-ccw.
- Few stack unwinder fixes and improvements, convert get_wchan custom
stack unwinding to generic api usage.
- Fixes for mm helpers issues uncovered with tests validating architecture
page table helpers.
- Fix noexec bit handling when hardware doesn't support it.
- Fix memleak and unsigned value compared with zero bugs in crypto
code. Minor code simplification.
- Fix crash during kdump with kasan enabled kernel.
- Switch bug and alternatives from asm to asm_inline to improve inlining
decisions.
- Use 'depends on cc-option' for MARCH and TUNE options in Kconfig,
add z13s and z14 ZR1 to TUNE descriptions.
- Minor head64.S simplification.
- Fix physical to logical CPU map for SMT.
- Several cleanups in qdio code.
- Other minor cleanups and fixes all over the code.
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEE3QHqV+H2a8xAv27vjYWKoQLXFBgFAl3ahGYACgkQjYWKoQLX
FBguwAgAig+FNos8zkd7Sr2wg4DPL2IlYERVP40fOLXfGuVUOnMLg8OTO6yDWDpH
5+cKAQS1wWgyvlfjWRUJ6anXLBAsgKRD1nyFIZTpn/wArGk/duCbnl/VFriDgrST
8KTQDJpZ9w9nXtQ7lA2QWaw5U2WG8I2T2JuQJCdLXze7RXi0bDVe8e6131NMaJ42
LLxqOqm8d8XDnd8oDVP04LT5IfhuI2cILoGBP/GyI2fqQk9Ems6M2gxuISq1COmy
WORDLfwWyCLeF7gWKKjxf8Vo1HYcyoFvdXnxWiHb0TDZesQZJr/LLELTP03fbCW9
U4jbXncnnPA7kT4tlC95jT5M69yK5w==
=+FxG
-----END PGP SIGNATURE-----
Merge tag 's390-5.5-1' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Pull s390 updates from Vasily Gorbik:
- Adjust PMU device drivers registration to avoid WARN_ON and few other
perf improvements.
- Enhance tracing in vfio-ccw.
- Few stack unwinder fixes and improvements, convert get_wchan custom
stack unwinding to generic api usage.
- Fixes for mm helpers issues uncovered with tests validating
architecture page table helpers.
- Fix noexec bit handling when hardware doesn't support it.
- Fix memleak and unsigned value compared with zero bugs in crypto
code. Minor code simplification.
- Fix crash during kdump with kasan enabled kernel.
- Switch bug and alternatives from asm to asm_inline to improve
inlining decisions.
- Use 'depends on cc-option' for MARCH and TUNE options in Kconfig, add
z13s and z14 ZR1 to TUNE descriptions.
- Minor head64.S simplification.
- Fix physical to logical CPU map for SMT.
- Several cleanups in qdio code.
- Other minor cleanups and fixes all over the code.
* tag 's390-5.5-1' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (41 commits)
s390/cpumf: Adjust registration of s390 PMU device drivers
s390/smp: fix physical to logical CPU map for SMT
s390/early: move access registers setup in C code
s390/head64: remove unnecessary vdso_per_cpu_data setup
s390/early: move control registers setup in C code
s390/kasan: support memcpy_real with TRACE_IRQFLAGS
s390/crypto: Fix unsigned variable compared with zero
s390/pkey: use memdup_user() to simplify code
s390/pkey: fix memory leak within _copy_apqns_from_user()
s390/disassembler: don't hide instruction addresses
s390/cpum_sf: Assign error value to err variable
s390/cpum_sf: Replace function name in debug statements
s390/cpum_sf: Use consistant debug print format for sampling
s390/unwind: drop unnecessary code around calling ftrace_graph_ret_addr()
s390: add error handling to perf_callchain_kernel
s390: always inline current_stack_pointer()
s390/mm: add mm_pxd_folded() checks to pxd_free()
s390/mm: properly clear _PAGE_NOEXEC bit when it is not supported
s390/mm: simplify page table helpers for large entries
s390/mm: make pmd/pud_bad() report large entries as bad
...
Linux-next commit titled "perf/core: Optimize perf_init_event()"
changed the semantics of PMU device driver registration.
It was done to speed up the lookup/handling of PMU device driver
specific events. It also enforces that only one PMU device
driver will be registered of type PERF_EVENT_RAW.
This change added these line in function perf_pmu_register():
...
+ ret = idr_alloc(&pmu_idr, pmu, max, 0, GFP_KERNEL);
+ if (ret < 0)
goto free_pdc;
+
+ WARN_ON(type >= 0 && ret != type);
The warn_on generates a message. We have 3 PMU device drivers,
each registered as type PERF_TYPE_RAW.
The cf_diag device driver (arch/s390/kernel/perf_cpumf_cf_diag.c)
always hits the WARN_ON because it is the second PMU device driver
(after sampling device driver arch/s390/kernel/perf_cpumf_sf.c)
which is registered as type 4 (PERF_TYPE_RAW).
So when the sampling device driver is registered, ret has value 4.
When cf_diag device driver is registered with type 4,
ret has value of 5 and WARN_ON fires.
Adjust the PMU device drivers for s390 to support the new
semantics required by perf_pmu_register().
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
If an SMT capable system is not IPL'ed from the first CPU the setup of
the physical to logical CPU mapping is broken: the IPL core gets CPU
number 0, but then the next core gets CPU number 1. Correct would be
that all SMT threads of CPU 0 get the subsequent logical CPU numbers.
This is important since a lot of code (like e.g. the CPU topology
code) assumes that CPU maps are setup like this. If the mapping is
broken the system will not IPL due to broken topology masks:
[ 1.716341] BUG: arch topology broken
[ 1.716342] the SMT domain not a subset of the MC domain
[ 1.716343] BUG: arch topology broken
[ 1.716344] the MC domain not a subset of the BOOK domain
This scenario can usually not happen since LPARs are always IPL'ed
from CPU 0 and also re-IPL is intiated from CPU 0. However older
kernels did initiate re-IPL on an arbitrary CPU. If therefore a re-IPL
from an old kernel into a new kernel is initiated this may lead to
crash.
Fix this by setting up the physical to logical CPU mapping correctly.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
vdso_per_cpu_data lowcore value is only needed for fully functional
exception handlers, which are activated in setup_lowcore_dat_off. The
same function does init vdso_per_cpu_data via vdso_alloc_boot_cpu.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Due to kptr_restrict, JITted BPF code is now displayed like this:
000000000b6ed1b2: ebdff0800024 stmg %r13,%r15,128(%r15)
000000004cde2ba0: 41d0f040 la %r13,64(%r15)
00000000fbad41b0: a7fbffa0 aghi %r15,-96
Leaking kernel addresses to dmesg is not a concern in this case, because
this happens only when JIT debugging is explicitly activated, which only
root can do.
Use %px in this particular instance, and also to print an instruction
address in show_code and PCREL (e.g. brasl) arguments in print_insn.
While at present functionally equivalent to %016lx, %px is recommended
by Documentation/core-api/printk-formats.rst for such cases.
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
When starting the CPU Measurement sampling facility using
qsi() function, this function may return an error value.
This error value is referenced in the else part of the
if statement to dump its value in a debug statement.
Right now this value is always zero because it has not been
assigned a value.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Replace hard coded function names in debug statements
by the "%s ...", __func__ construct suggested by checkpatch.pl
script.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Use consistant debug print format of the form variable
blank value. Also add leading 0x for all hex values.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
The current code around calling ftrace_graph_ret_addr() is ifdeffed and
also tests if ftrace redirection is present on stack.
ftrace_graph_ret_addr() however performs the test internally and there
is a version for !CONFIG_FUNCTION_GRAPH_TRACER as well. The unnecessary
code can thus be dropped.
Link: http://lkml.kernel.org/r/20191029143904.24051-2-mbenes@suse.cz
Signed-off-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
The idle time reported in /proc/stat sometimes incorrectly contains
huge values on s390. This is caused by a bug in arch_cpu_idle_time().
The kernel tries to figure out when a different cpu entered idle by
accessing its per-cpu data structure. There is an ordering problem: if
the remote cpu has an idle_enter value which is not zero, and an
idle_exit value which is zero, it is assumed it is idle since
"now". The "now" timestamp however is taken before the idle_enter
value is read.
Which in turn means that "now" can be smaller than idle_enter of the
remote cpu. Unconditionally subtracting idle_enter from "now" can thus
lead to a negative value (aka large unsigned value).
Fix this by moving the get_tod_clock() invocation out of the
loop. While at it also make the code a bit more readable.
A similar bug also exists for show_idle_time(). Fix this is as well.
Cc: <stable@vger.kernel.org>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
unwind_for_each_frame stops after the first frame if regs->gprs[15] <=
sp.
The reason is that in case regs are specified, the first frame should be
regs->psw.addr and the second frame should be sp->gprs[8]. However,
currently the second frame is regs->gprs[15], which confuses
outside_of_stack().
Fix by introducing a flag to distinguish this special case from
unwinding the interrupt handler, for which the current behavior is
appropriate.
Fixes: 78c98f9074 ("s390/unwind: introduce stack unwind API")
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Cc: stable@vger.kernel.org # v5.2+
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
perf_callchain_kernel stops neither when it encounters a garbage
address, nor when it runs out of space. Fix both issues using x86
version as an inspiration.
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
