platform/kernel/linux-rpi.git
4 years agokcsan: Move kcsan_{disable,enable}_current() to kcsan-checks.h
Marco Elver [Tue, 31 Mar 2020 19:32:32 +0000 (21:32 +0200)]
kcsan: Move kcsan_{disable,enable}_current() to kcsan-checks.h

Both affect access checks, and should therefore be in kcsan-checks.h.
This is in preparation to use these in compiler.h.

Acked-by: Will Deacon <will@kernel.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Introduce scoped ASSERT_EXCLUSIVE macros
Marco Elver [Wed, 25 Mar 2020 16:41:58 +0000 (17:41 +0100)]
kcsan: Introduce scoped ASSERT_EXCLUSIVE macros

Introduce ASSERT_EXCLUSIVE_*_SCOPED(), which provide an intuitive
interface to use the scoped-access feature, without having to explicitly
mark the start and end of the desired scope. Basing duration of the
checks on scope avoids accidental misuse and resulting false positives,
which may be hard to debug. See added comments for usage.

The macros are implemented using __attribute__((__cleanup__(func))),
which is supported by all compilers that currently support KCSAN.

Suggested-by: Boqun Feng <boqun.feng@gmail.com>
Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agoobjtool, kcsan: Add explicit check functions to uaccess whitelist
Marco Elver [Wed, 25 Mar 2020 16:41:57 +0000 (17:41 +0100)]
objtool, kcsan: Add explicit check functions to uaccess whitelist

Add explicitly invoked KCSAN check functions to objtool's uaccess
whitelist. This is needed in order to permit calling into
kcsan_check_scoped_accesses() from the fast-path, which in turn calls
__kcsan_check_access().  __kcsan_check_access() is the generic variant
of the already whitelisted specializations __tsan_{read,write}N.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Add support for scoped accesses
Marco Elver [Wed, 25 Mar 2020 16:41:56 +0000 (17:41 +0100)]
kcsan: Add support for scoped accesses

This adds support for scoped accesses, where the memory range is checked
for the duration of the scope. The feature is implemented by inserting
the relevant access information into a list of scoped accesses for
the current execution context, which are then checked (until removed)
on every call (through instrumentation) into the KCSAN runtime.

An alternative, more complex, implementation could set up a watchpoint for
the scoped access, and keep the watchpoint set up. This, however, would
require first exposing a handle to the watchpoint, as well as dealing
with cases such as accesses by the same thread while the watchpoint is
still set up (and several more cases). It is also doubtful if this would
provide any benefit, since the majority of delay where the watchpoint
is set up is likely due to the injected delays by KCSAN.  Therefore,
the implementation in this patch is simpler and avoids hurting KCSAN's
main use-case (normal data race detection); it also implicitly increases
scoped-access race-detection-ability due to increased probability of
setting up watchpoints by repeatedly calling __kcsan_check_access()
throughout the scope of the access.

The implementation required adding an additional conditional branch to
the fast-path. However, the microbenchmark showed a *speedup* of ~5%
on the fast-path. This appears to be due to subtly improved codegen by
GCC from moving get_ctx() and associated load of preempt_count earlier.

Suggested-by: Boqun Feng <boqun.feng@gmail.com>
Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Avoid blocking producers in prepare_report()
Marco Elver [Wed, 18 Mar 2020 17:38:45 +0000 (18:38 +0100)]
kcsan: Avoid blocking producers in prepare_report()

To avoid deadlock in case watchers can be interrupted, we need to ensure
that producers of the struct other_info can never be blocked by an
unrelated consumer. (Likely to occur with KCSAN_INTERRUPT_WATCHER.)

There are several cases that can lead to this scenario, for example:

1. A watchpoint A was set up by task T1, but interrupted by
   interrupt I1. Some other thread (task or interrupt) finds
   watchpoint A consumes it, and sets other_info. Then I1 also
   finds some unrelated watchpoint B, consumes it, but is blocked
   because other_info is in use. T1 cannot consume other_info
   because I1 never returns -> deadlock.

2. A watchpoint A was set up by task T1, but interrupted by
   interrupt I1, which also sets up a watchpoint B. Some other
   thread finds watchpoint A, and consumes it and sets up
   other_info with its information. Similarly some other thread
   finds watchpoint B and consumes it, but is then blocked because
   other_info is in use. When I1 continues it sees its watchpoint
   was consumed, and that it must wait for other_info, which
   currently contains information to be consumed by T1. However, T1
   cannot unblock other_info because I1 never returns -> deadlock.

To avoid this, we need to ensure that producers of struct other_info
always have a usable other_info entry. This is obviously not the case
with only a single instance of struct other_info, as concurrent
producers must wait for the entry to be released by some consumer (which
may be locked up as illustrated above).

While it would be nice if producers could simply call kmalloc() and
append their instance of struct other_info to a list, we are very
limited in this code path: since KCSAN can instrument the allocators
themselves, calling kmalloc() could lead to deadlock or corrupted
allocator state.

Since producers of the struct other_info will always succeed at
try_consume_watchpoint(), preceding the call into kcsan_report(), we
know that the particular watchpoint slot cannot simply be reused or
consumed by another potential other_info producer. If we move removal of
a watchpoint after reporting (by the consumer of struct other_info), we
can see a consumed watchpoint as a held lock on elements of other_info,
if we create a one-to-one mapping of a watchpoint to an other_info
element.

Therefore, the simplest solution is to create an array of struct
other_info that is as large as the watchpoints array in core.c, and pass
the watchpoint index to kcsan_report() for producers and consumers, and
change watchpoints to be removed after reporting is done.

With a default config on a 64-bit system, the array other_infos consumes
~37KiB. For most systems today this is not a problem. On smaller memory
constrained systems, the config value CONFIG_KCSAN_NUM_WATCHPOINTS can
be reduced appropriately.

Overall, this change is a simplification of the prepare_report() code,
and makes some of the checks (such as checking if at least one access is
a write) redundant.

Tested:
$ tools/testing/selftests/rcutorture/bin/kvm.sh \
--cpus 12 --duration 10 --kconfig "CONFIG_DEBUG_INFO=y \
CONFIG_KCSAN=y CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC=n \
CONFIG_KCSAN_REPORT_VALUE_CHANGE_ONLY=n \
CONFIG_KCSAN_REPORT_ONCE_IN_MS=100000 CONFIG_KCSAN_VERBOSE=y \
CONFIG_KCSAN_INTERRUPT_WATCHER=y CONFIG_PROVE_LOCKING=y" \
--configs TREE03
=> No longer hangs and runs to completion as expected.

Reported-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Introduce report access_info and other_info
Marco Elver [Wed, 18 Mar 2020 17:38:44 +0000 (18:38 +0100)]
kcsan: Introduce report access_info and other_info

Improve readability by introducing access_info and other_info structs,
and in preparation of the following commit in this series replaces the
single instance of other_info with an array of size 1.

No functional change intended.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Update API documentation in kcsan-checks.h
Marco Elver [Thu, 5 Mar 2020 14:21:09 +0000 (15:21 +0100)]
kcsan: Update API documentation in kcsan-checks.h

Update the API documentation for ASSERT_EXCLUSIVE_* macros and make them
generate readable documentation for the code examples.

All @variable short summaries were missing ':', which was updated for
the whole file.

Tested with "make htmldocs".

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Update Documentation/dev-tools/kcsan.rst
Marco Elver [Thu, 5 Mar 2020 14:21:08 +0000 (15:21 +0100)]
kcsan: Update Documentation/dev-tools/kcsan.rst

Extend and improve based on recent changes, and summarize important
bits that have been missing. Tested with "make htmldocs".

Signed-off-by: Marco Elver <elver@google.com>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Fix a typo in a comment
Qiujun Huang [Thu, 5 Mar 2020 14:21:07 +0000 (15:21 +0100)]
kcsan: Fix a typo in a comment

s/slots slots/slots/

Signed-off-by: Qiujun Huang <hqjagain@gmail.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
[elver: commit message]
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Add current->state to implicitly atomic accesses
Marco Elver [Tue, 25 Feb 2020 14:32:58 +0000 (15:32 +0100)]
kcsan: Add current->state to implicitly atomic accesses

Add volatile current->state to list of implicitly atomic accesses. This
is in preparation to eventually enable KCSAN on kernel/sched (which
currently still has KCSAN_SANITIZE := n).

Since accesses that match the special check in atomic.h are rare, it
makes more sense to move this check to the slow-path, avoiding the
additional compare in the fast-path. With the microbenchmark, a speedup
of ~6% is measured.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Add option for verbose reporting
Marco Elver [Fri, 21 Feb 2020 23:10:27 +0000 (00:10 +0100)]
kcsan: Add option for verbose reporting

Adds CONFIG_KCSAN_VERBOSE to optionally enable more verbose reports.
Currently information about the reporting task's held locks and IRQ
trace events are shown, if they are enabled.

Signed-off-by: Marco Elver <elver@google.com>
Suggested-by: Qian Cai <cai@lca.pw>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan: Add option to allow watcher interruptions
Marco Elver [Fri, 21 Feb 2020 22:02:09 +0000 (23:02 +0100)]
kcsan: Add option to allow watcher interruptions

Add option to allow interrupts while a watchpoint is set up. This can be
enabled either via CONFIG_KCSAN_INTERRUPT_WATCHER or via the boot
parameter 'kcsan.interrupt_watcher=1'.

Note that, currently not all safe per-CPU access primitives and patterns
are accounted for, which could result in false positives. For example,
asm-generic/percpu.h uses plain operations, which by default are
instrumented. On interrupts and subsequent accesses to the same
variable, KCSAN would currently report a data race with this option.

Therefore, this option should currently remain disabled by default, but
may be enabled for specific test scenarios.

To avoid new warnings, changes all uses of smp_processor_id() to use the
raw version (as already done in kcsan_found_watchpoint()). The exact SMP
processor id is for informational purposes in the report, and
correctness is not affected.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
4 years agokcsan, trace: Make KCSAN compatible with tracing
Marco Elver [Fri, 14 Feb 2020 21:10:35 +0000 (22:10 +0100)]
kcsan, trace: Make KCSAN compatible with tracing

Previously the system would lock up if ftrace was enabled together with
KCSAN. This is due to recursion on reporting if the tracer code is
instrumented with KCSAN.

To avoid this for all types of tracing, disable KCSAN instrumentation
for all of kernel/trace.

Furthermore, since KCSAN relies on udelay() to introduce delay, we have
to disable ftrace for udelay() (currently done for x86) in case KCSAN is
used together with lockdep and ftrace. The reason is that it may corrupt
lockdep IRQ flags tracing state due to a peculiar case of recursion
(details in Makefile comment).

Reported-by: Qian Cai <cai@lca.pw>
Tested-by: Qian Cai <cai@lca.pw>
Acked-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Introduce ASSERT_EXCLUSIVE_BITS(var, mask)
Marco Elver [Tue, 11 Feb 2020 16:04:23 +0000 (17:04 +0100)]
kcsan: Introduce ASSERT_EXCLUSIVE_BITS(var, mask)

This introduces ASSERT_EXCLUSIVE_BITS(var, mask).
ASSERT_EXCLUSIVE_BITS(var, mask) will cause KCSAN to assume that the
following access is safe w.r.t. data races (however, please see the
docbook comment for disclaimer here).

For more context on why this was considered necessary, please see:

  http://lkml.kernel.org/r/1580995070-25139-1-git-send-email-cai@lca.pw

In particular, before this patch, data races between reads (that use
@mask bits of an access that should not be modified concurrently) and
writes (that change ~@mask bits not used by the readers) would have been
annotated with "data_race()" (or "READ_ONCE()"). However, doing so would
then hide real problems: we would no longer be able to detect harmful
races between reads to @mask bits and writes to @mask bits.

Therefore, by using ASSERT_EXCLUSIVE_BITS(var, mask), we accomplish:

  1. Avoid proliferation of specific macros at the call sites: by
     including a single mask in the argument list, we can use the same
     macro in a wide variety of call sites, regardless of how and which
     bits in a field each call site actually accesses.

  2. The existing code does not need to be modified (although READ_ONCE()
     may still be advisable if we cannot prove that the data race is
     always safe).

  3. We catch bugs where the exclusive bits are modified concurrently.

  4. We document properties of the current code.

Acked-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Qian Cai <cai@lca.pw>
4 years agokcsan: Add kcsan_set_access_mask() support
Marco Elver [Tue, 11 Feb 2020 16:04:22 +0000 (17:04 +0100)]
kcsan: Add kcsan_set_access_mask() support

When setting up an access mask with kcsan_set_access_mask(), KCSAN will
only report races if concurrent changes to bits set in access_mask are
observed. Conveying access_mask via a separate call avoids introducing
overhead in the common-case fast-path.

Acked-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Introduce kcsan_value_change type
Marco Elver [Tue, 11 Feb 2020 16:04:21 +0000 (17:04 +0100)]
kcsan: Introduce kcsan_value_change type

Introduces kcsan_value_change type, which explicitly points out if we
either observed a value-change (TRUE), or we could not observe one but
cannot rule out a value-change happened (MAYBE). The MAYBE state can
either be reported or not, depending on configuration preferences.

A follow-up patch introduces the FALSE state, which should never be
reported.

No functional change intended.

Acked-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agocompiler.h, seqlock.h: Remove unnecessary kcsan.h includes
Marco Elver [Tue, 11 Feb 2020 16:04:20 +0000 (17:04 +0100)]
compiler.h, seqlock.h: Remove unnecessary kcsan.h includes

No we longer have to include kcsan.h, since the required KCSAN interface
for both compiler.h and seqlock.h are now provided by kcsan-checks.h.

Acked-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Move interfaces that affects checks to kcsan-checks.h
Marco Elver [Tue, 11 Feb 2020 16:04:19 +0000 (17:04 +0100)]
kcsan: Move interfaces that affects checks to kcsan-checks.h

This moves functions that affect state changing the behaviour of
kcsan_check_access() to kcsan-checks.h. Since these are likely used with
kcsan_check_access() it makes more sense to have them in kcsan-checks.h,
to avoid including all of 'include/linux/kcsan.h'.

No functional change intended.

Acked-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Fix misreporting if concurrent races on same address
Marco Elver [Mon, 10 Feb 2020 14:56:39 +0000 (15:56 +0100)]
kcsan: Fix misreporting if concurrent races on same address

If there are at least 4 threads racing on the same address, it can
happen that one of the readers may observe another matching reader in
other_info. To avoid locking up, we have to consume 'other_info'
regardless, but skip the report. See the added comment for more details.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Expose core configuration parameters as module params
Marco Elver [Fri, 7 Feb 2020 18:59:10 +0000 (19:59 +0100)]
kcsan: Expose core configuration parameters as module params

This adds early_boot, udelay_{task,interrupt}, and skip_watch as module
params. The latter parameters are useful to modify at runtime to tune
KCSAN's performance on new systems. This will also permit auto-tuning
these parameters to maximize overall system performance and KCSAN's race
detection ability.

None of the parameters are used in the fast-path and referring to them
via static variables instead of CONFIG constants will not affect
performance.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Qian Cai <cai@lca.pw>
4 years agokcsan: Add test to generate conflicts via debugfs
Marco Elver [Thu, 6 Feb 2020 15:46:26 +0000 (16:46 +0100)]
kcsan: Add test to generate conflicts via debugfs

Add 'test=<iters>' option to KCSAN's debugfs interface to invoke KCSAN
checks on a dummy variable. By writing 'test=<iters>' to the debugfs
file from multiple tasks, we can generate real conflicts, and trigger
data race reports.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Introduce ASSERT_EXCLUSIVE_*() macros
Marco Elver [Thu, 6 Feb 2020 15:46:25 +0000 (16:46 +0100)]
kcsan: Introduce ASSERT_EXCLUSIVE_*() macros

Introduces ASSERT_EXCLUSIVE_WRITER() and ASSERT_EXCLUSIVE_ACCESS(), which
may be used to assert properties of synchronization logic, where
violation cannot be detected as a normal data race.

Examples of the reports that may be generated:

    ==================================================================
    BUG: KCSAN: assert: race in test_thread / test_thread

    write to 0xffffffffab3d1540 of 8 bytes by task 466 on cpu 2:
     test_thread+0x8d/0x111
     debugfs_write.cold+0x32/0x44
     ...

    assert no writes to 0xffffffffab3d1540 of 8 bytes by task 464 on cpu 0:
     test_thread+0xa3/0x111
     debugfs_write.cold+0x32/0x44
     ...
    ==================================================================

    ==================================================================
    BUG: KCSAN: assert: race in test_thread / test_thread

    assert no accesses to 0xffffffffab3d1540 of 8 bytes by task 465 on cpu 1:
     test_thread+0xb9/0x111
     debugfs_write.cold+0x32/0x44
     ...

    read to 0xffffffffab3d1540 of 8 bytes by task 464 on cpu 0:
     test_thread+0x77/0x111
     debugfs_write.cold+0x32/0x44
     ...
    ==================================================================

Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Introduce KCSAN_ACCESS_ASSERT access type
Marco Elver [Thu, 6 Feb 2020 15:46:24 +0000 (16:46 +0100)]
kcsan: Introduce KCSAN_ACCESS_ASSERT access type

The KCSAN_ACCESS_ASSERT access type may be used to introduce dummy reads
and writes to assert certain properties of concurrent code, where bugs
could not be detected as normal data races.

For example, a variable that is only meant to be written by a single
CPU, but may be read (without locking) by other CPUs must still be
marked properly to avoid data races. However, concurrent writes,
regardless if WRITE_ONCE() or not, would be a bug. Using
kcsan_check_access(&x, sizeof(x), KCSAN_ACCESS_ASSERT) would allow
catching such bugs.

To support KCSAN_ACCESS_ASSERT the following notable changes were made:

  * If an access is of type KCSAN_ASSERT_ACCESS, disable various filters
    that only apply to data races, so that all races that KCSAN observes are
    reported.
  * Bug reports that involve an ASSERT access type will be reported as
    "KCSAN: assert: race in ..." instead of "data-race"; this will help
    more easily distinguish them.
  * Update a few comments to just mention 'races' where we do not always
    mean pure data races.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Fix 0-sized checks
Marco Elver [Wed, 5 Feb 2020 10:14:19 +0000 (11:14 +0100)]
kcsan: Fix 0-sized checks

Instrumentation of arbitrary memory-copy functions, such as user-copies,
may be called with size of 0, which could lead to false positives.

To avoid this, add a comparison in check_access() for size==0, which
will be optimized out for constant sized instrumentation
(__tsan_{read,write}N), and therefore not affect the common-case
fast-path.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Clean up the main KCSAN Kconfig option
Marco Elver [Tue, 4 Feb 2020 17:21:12 +0000 (18:21 +0100)]
kcsan: Clean up the main KCSAN Kconfig option

This patch cleans up the rules of the 'KCSAN' Kconfig option by:

  1. implicitly selecting 'STACKTRACE' instead of depending on it;
  2. depending on DEBUG_KERNEL, to avoid accidentally turning KCSAN on if
     the kernel is not meant to be a debug kernel;
  3. updating the short and long summaries.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Clarify Kconfig option KCSAN_IGNORE_ATOMICS
Marco Elver [Tue, 4 Feb 2020 17:21:11 +0000 (18:21 +0100)]
kcsan: Clarify Kconfig option KCSAN_IGNORE_ATOMICS

Clarify difference between options KCSAN_IGNORE_ATOMICS and
KCSAN_ASSUME_PLAIN_WRITES_ATOMIC in help text.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Add option to assume plain aligned writes up to word size are atomic
Marco Elver [Tue, 4 Feb 2020 17:21:10 +0000 (18:21 +0100)]
kcsan: Add option to assume plain aligned writes up to word size are atomic

This adds option KCSAN_ASSUME_PLAIN_WRITES_ATOMIC. If enabled, plain
aligned writes up to word size are assumed to be atomic, and also not
subject to other unsafe compiler optimizations resulting in data races.

This option has been enabled by default to reflect current kernel-wide
preferences.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Add docbook header for data_race()
Paul E. McKenney [Mon, 3 Feb 2020 22:42:18 +0000 (14:42 -0800)]
kcsan: Add docbook header for data_race()

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Marco Elver <elver@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
4 years agocopy_to_user, copy_from_user: Use generic instrumented.h
Marco Elver [Tue, 21 Jan 2020 16:05:12 +0000 (17:05 +0100)]
copy_to_user, copy_from_user: Use generic instrumented.h

This replaces the KASAN instrumentation with generic instrumentation,
implicitly adding KCSAN instrumentation support.

For KASAN no functional change is intended.

Suggested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agoiov_iter: Use generic instrumented.h
Marco Elver [Tue, 21 Jan 2020 16:05:11 +0000 (17:05 +0100)]
iov_iter: Use generic instrumented.h

This replaces the kasan instrumentation with generic instrumentation,
implicitly adding KCSAN instrumentation support.

For KASAN no functional change is intended.

Suggested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agoasm-generic, kcsan: Add KCSAN instrumentation for bitops
Marco Elver [Tue, 21 Jan 2020 16:05:10 +0000 (17:05 +0100)]
asm-generic, kcsan: Add KCSAN instrumentation for bitops

Add explicit KCSAN checks for bitops.

Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agoasm-generic, atomic-instrumented: Use generic instrumented.h
Marco Elver [Tue, 21 Jan 2020 16:05:09 +0000 (17:05 +0100)]
asm-generic, atomic-instrumented: Use generic instrumented.h

This switches atomic-instrumented.h to use the generic instrumentation
wrappers provided by instrumented.h.

No functional change intended.

Suggested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Marco Elver <elver@google.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agoinclude/linux: Add instrumented.h infrastructure
Marco Elver [Tue, 21 Jan 2020 16:05:08 +0000 (17:05 +0100)]
include/linux: Add instrumented.h infrastructure

This adds instrumented.h, which provides generic wrappers for memory
access instrumentation that the compiler cannot emit for various
sanitizers. Currently this unifies KASAN and KCSAN instrumentation. In
future this will also include KMSAN instrumentation.

Note that, copy_{to,from}_user should use special instrumentation, since
we should be able to instrument both source and destination memory
accesses if both are kernel memory.

The current patch only instruments the memory access where the address
is always in kernel space, however, both may in fact be kernel addresses
when a compat syscall passes an argument allocated in the kernel to a
real syscall. In a future change, both KASAN and KCSAN should check both
addresses in such cases, as well as KMSAN will make use of both
addresses. [It made more sense to provide the completed function
signature, rather than updating it and changing all locations again at a
later time.]

Suggested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Marco Elver <elver@google.com>
Acked-by: Alexander Potapenko <glider@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Address missing case with KCSAN_REPORT_VALUE_CHANGE_ONLY
Marco Elver [Wed, 29 Jan 2020 15:01:02 +0000 (16:01 +0100)]
kcsan: Address missing case with KCSAN_REPORT_VALUE_CHANGE_ONLY

Even with KCSAN_REPORT_VALUE_CHANGE_ONLY, KCSAN still reports data
races between reads and watchpointed writes, even if the writes wrote
values already present.  This commit causes KCSAN to unconditionally
skip reporting in this case.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Make KCSAN compatible with lockdep
Marco Elver [Wed, 15 Jan 2020 16:25:12 +0000 (17:25 +0100)]
kcsan: Make KCSAN compatible with lockdep

We must avoid any recursion into lockdep if KCSAN is enabled on utilities
used by lockdep. One manifestation of this is corruption of lockdep's
IRQ trace state (if TRACE_IRQFLAGS), resulting in spurious warnings
(see below).  This commit fixes this by:

1. Using raw_local_irq{save,restore} in kcsan_setup_watchpoint().
2. Disabling lockdep in kcsan_report().

Tested with:

  CONFIG_LOCKDEP=y
  CONFIG_DEBUG_LOCKDEP=y
  CONFIG_TRACE_IRQFLAGS=y

This fix eliminates spurious warnings such as the following one:

    WARNING: CPU: 0 PID: 2 at kernel/locking/lockdep.c:4406 check_flags.part.0+0x101/0x220
    Modules linked in:
    CPU: 0 PID: 2 Comm: kthreadd Not tainted 5.5.0-rc1+ #11
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
    RIP: 0010:check_flags.part.0+0x101/0x220
    <snip>
    Call Trace:
     lock_is_held_type+0x69/0x150
     freezer_fork+0x20b/0x370
     cgroup_post_fork+0x2c9/0x5c0
     copy_process+0x2675/0x3b40
     _do_fork+0xbe/0xa30
     ? _raw_spin_unlock_irqrestore+0x40/0x50
     ? match_held_lock+0x56/0x250
     ? kthread_park+0xf0/0xf0
     kernel_thread+0xa6/0xd0
     ? kthread_park+0xf0/0xf0
     kthreadd+0x321/0x3d0
     ? kthread_create_on_cpu+0x130/0x130
     ret_from_fork+0x3a/0x50
    irq event stamp: 64
    hardirqs last  enabled at (63): [<ffffffff9a7995d0>] _raw_spin_unlock_irqrestore+0x40/0x50
    hardirqs last disabled at (64): [<ffffffff992a96d2>] kcsan_setup_watchpoint+0x92/0x460
    softirqs last  enabled at (32): [<ffffffff990489b8>] fpu__copy+0xe8/0x470
    softirqs last disabled at (30): [<ffffffff99048939>] fpu__copy+0x69/0x470

Reported-by: Qian Cai <cai@lca.pw>
Signed-off-by: Marco Elver <elver@google.com>
Acked-by: Alexander Potapenko <glider@google.com>
Tested-by: Qian Cai <cai@lca.pw>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Rate-limit reporting per data races
Marco Elver [Fri, 10 Jan 2020 18:48:34 +0000 (19:48 +0100)]
kcsan: Rate-limit reporting per data races

KCSAN data-race reports can occur quite frequently, so much so as
to render the system useless.  This commit therefore adds support for
time-based rate-limiting KCSAN reports, with the time interval specified
by a new KCSAN_REPORT_ONCE_IN_MS Kconfig option.  The default is 3000
milliseconds, also known as three seconds.

Because KCSAN must detect data races in allocators and in other contexts
where use of allocation is ill-advised, a fixed-size array is used to
buffer reports during each reporting interval.  To reduce the number of
reports lost due to array overflow, this commit stores only one instance
of duplicate reports, which has the benefit of further reducing KCSAN's
console output rate.

Reported-by: Qian Cai <cai@lca.pw>
Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Show full access type in report
Marco Elver [Fri, 10 Jan 2020 18:48:33 +0000 (19:48 +0100)]
kcsan: Show full access type in report

This commit adds access-type information to KCSAN's reports as follows:
"read", "read (marked)", "write", and "write (marked)".

Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agokcsan: Prefer __always_inline for fast-path
Marco Elver [Tue, 7 Jan 2020 16:31:04 +0000 (17:31 +0100)]
kcsan: Prefer __always_inline for fast-path

Prefer __always_inline for fast-path functions that are called outside
of user_access_save, to avoid generating UACCESS warnings when
optimizing for size (CC_OPTIMIZE_FOR_SIZE). It will also avoid future
surprises with compiler versions that change the inlining heuristic even
when optimizing for performance.

Reported-by: Randy Dunlap <rdunlap@infradead.org>
Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: http://lkml.kernel.org/r/58708908-84a0-0a81-a836-ad97e33dbb62@infradead.org
4 years agoMerge branch 'linus' into locking/kcsan, to pick up fixes
Ingo Molnar [Sat, 21 Mar 2020 08:35:44 +0000 (09:35 +0100)]
Merge branch 'linus' into locking/kcsan, to pick up fixes

Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agoMerge branch 'x86/kdump' into locking/kcsan, to resolve conflicts
Ingo Molnar [Sat, 21 Mar 2020 08:23:40 +0000 (09:23 +0100)]
Merge branch 'x86/kdump' into locking/kcsan, to resolve conflicts

Conflicts:
arch/x86/purgatory/Makefile

Signed-off-by: Ingo Molnar <mingo@kernel.org>
4 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Fri, 20 Mar 2020 16:28:25 +0000 (09:28 -0700)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fixes from Will Deacon:

 - Fix panic() when it occurs during secondary CPU startup

 - Fix "kpti=off" when KASLR is enabled

 - Fix howler in compat syscall table for vDSO clock_getres() fallback

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: compat: Fix syscall number of compat_clock_getres
  arm64: kpti: Fix "kpti=off" when KASLR is enabled
  arm64: smp: fix crash_smp_send_stop() behaviour
  arm64: smp: fix smp_send_stop() behaviour

4 years agoMerge tag 'char-misc-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Fri, 20 Mar 2020 16:24:22 +0000 (09:24 -0700)]
Merge tag 'char-misc-5.6-rc7' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc driver fixes from Greg KH:
 "Here are some small different driver fixes for 5.6-rc7:

   - binderfs fix, yet again

   - slimbus new device id added

   - hwtracing bugfixes for reported issues and a new device id

  All of these have been in linux-next with no reported issues"

* tag 'char-misc-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  intel_th: pci: Add Elkhart Lake CPU support
  intel_th: Fix user-visible error codes
  intel_th: msu: Fix the unexpected state warning
  stm class: sys-t: Fix the use of time_after()
  slimbus: ngd: add v2.1.0 compatible
  binderfs: use refcount for binder control devices too

4 years agoMerge tag 'staging-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Fri, 20 Mar 2020 16:20:38 +0000 (09:20 -0700)]
Merge tag 'staging-5.6-rc7' of git://git./linux/kernel/git/gregkh/staging

Pull staging/IIO fixes from Greg KH:
 "Here are a number of small staging and IIO driver fixes for 5.6-rc7

  Nothing major here, just resolutions for some reported problems:
   - iio bugfixes for a number of different drivers
   - greybus loopback_test fixes
   - wfx driver fixes

  All of these have been in linux-next with no reported issues"

* tag 'staging-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: rtl8188eu: Add device id for MERCUSYS MW150US v2
  staging: greybus: loopback_test: fix potential path truncations
  staging: greybus: loopback_test: fix potential path truncation
  staging: greybus: loopback_test: fix poll-mask build breakage
  staging: wfx: fix RCU usage between hif_join() and ieee80211_bss_get_ie()
  staging: wfx: fix RCU usage in wfx_join_finalize()
  staging: wfx: make warning about pending frame less scary
  staging: wfx: fix lines ending with a comma instead of a semicolon
  staging: wfx: fix warning about freeing in-use mutex during device unregister
  staging/speakup: fix get_word non-space look-ahead
  iio: ping: set pa_laser_ping_cfg in of_ping_match
  iio: chemical: sps30: fix missing triggered buffer dependency
  iio: st_sensors: remap SMO8840 to LIS2DH12
  iio: light: vcnl4000: update sampling periods for vcnl4040
  iio: light: vcnl4000: update sampling periods for vcnl4200
  iio: accel: adxl372: Set iio_chan BE
  iio: magnetometer: ak8974: Fix negative raw values in sysfs
  iio: trigger: stm32-timer: disable master mode when stopping
  iio: adc: stm32-dfsdm: fix sleep in atomic context
  iio: adc: at91-sama5d2_adc: fix differential channels in triggered mode

4 years agoMerge tag 'usb-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Fri, 20 Mar 2020 16:16:35 +0000 (09:16 -0700)]
Merge tag 'usb-5.6-rc7' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are some small USB fixes for 5.6-rc7. And there's a thunderbolt
  driver fix thrown in for good measure as well.

  These fixes are:
   - new device ids for usb-serial drivers
   - thunderbolt error code fix
   - xhci driver fixes
   - typec fixes
   - cdc-acm driver fixes
   - chipidea driver fix
   - more USB quirks added for devices that need them.

  All of these have been in linux-next with no reported issues"

* tag 'usb-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  USB: cdc-acm: fix rounding error in TIOCSSERIAL
  USB: cdc-acm: fix close_delay and closing_wait units in TIOCSSERIAL
  usb: quirks: add NO_LPM quirk for RTL8153 based ethernet adapters
  usb: chipidea: udc: fix sleeping function called from invalid context
  USB: serial: pl2303: add device-id for HP LD381
  USB: serial: option: add ME910G1 ECM composition 0x110b
  usb: host: xhci-plat: add a shutdown
  usb: typec: ucsi: displayport: Fix a potential race during registration
  usb: typec: ucsi: displayport: Fix NULL pointer dereference
  USB: Disable LPM on WD19's Realtek Hub
  usb: xhci: apply XHCI_SUSPEND_DELAY to AMD XHCI controller 1022:145c
  xhci: Do not open code __print_symbolic() in xhci trace events
  thunderbolt: Fix error code in tb_port_is_width_supported()

4 years agoMerge tag 'tty-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
Linus Torvalds [Fri, 20 Mar 2020 16:13:35 +0000 (09:13 -0700)]
Merge tag 'tty-5.6-rc7' of git://git./linux/kernel/git/gregkh/tty

Pull tty fixes from Greg KH:
 "Here are three small tty_io bugfixes for reported issues that Eric has
  resolved for 5.6-rc7

  All of these have been in linux-next with no reported issues"

* tag 'tty-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty:
  tty: fix compat TIOCGSERIAL checking wrong function ptr
  tty: fix compat TIOCGSERIAL leaking uninitialized memory
  tty: drop outdated comments about release_tty() locking

4 years agoMerge tag 'sound-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Fri, 20 Mar 2020 16:10:29 +0000 (09:10 -0700)]
Merge tag 'sound-5.6-rc7' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "A few fixes covering the issues reported by syzkaller, a couple of
  fixes for the MIDI decoding bug, and a few usual HD-audio quirks.

  Some of them are about ALSA core stuff, but they are small fixes just
  for corner cases, and nothing thrilling"

* tag 'sound-5.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: hda/realtek - Enable the headset of Acer N50-600 with ALC662
  ALSA: hda/realtek - Enable headset mic of Acer X2660G with ALC662
  ALSA: seq: oss: Fix running status after receiving sysex
  ALSA: seq: virmidi: Fix running status after receiving sysex
  ALSA: pcm: oss: Remove WARNING from snd_pcm_plug_alloc() checks
  ALSA: hda/realtek: Fix pop noise on ALC225
  ALSA: line6: Fix endless MIDI read loop
  ALSA: pcm: oss: Avoid plugin buffer overflow

4 years agoMerge tag 'drm-fixes-2020-03-20' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 20 Mar 2020 16:03:54 +0000 (09:03 -0700)]
Merge tag 'drm-fixes-2020-03-20' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Hope you are well hiding out above the garage. A few amdgpu changes
  but nothing too major. I've had a wisdom tooth out this week so
  haven't been to on top of things, but all seems good.

  core:
   - fix lease warning

  i915:
   - Track active elements during dequeue
   - Fix failure to handle all MCR ranges
   - Revert unnecessary workaround

  amdgpu:
   - Pageflip fix
   - VCN clockgating fixes
   - GPR debugfs fix for umr
   - GPU reset fix
   - eDP fix for MBP
   - DCN2.x fix

  dw-hdmi:
   - fix AVI frame colorimetry

  komeda:
   - fix compiler warning

  bochs:
   - downgrade a binding failure to a warning"

* tag 'drm-fixes-2020-03-20' of git://anongit.freedesktop.org/drm/drm:
  drm/amd/display: Fix pageflip event race condition for DCN.
  drm/amdgpu: fix typo for vcn2.5/jpeg2.5 idle check
  drm/amdgpu: fix typo for vcn2/jpeg2 idle check
  drm/amdgpu: fix typo for vcn1 idle check
  drm/lease: fix WARNING in idr_destroy
  drm/i915: Handle all MCR ranges
  Revert "drm/i915/tgl: Add extra hdc flush workaround"
  drm/i915/execlists: Track active elements during dequeue
  drm/bochs: downgrade pci_request_region failure from error to warning
  drm/amd/display: Add link_rate quirk for Apple 15" MBP 2017
  drm/amdgpu: add fbdev suspend/resume on gpu reset
  drm/amd/amdgpu: Fix GPR read from debugfs (v2)
  drm/amd/display: fix typos for dcn20_funcs and dcn21_funcs struct
  drm/komeda: mark PM functions as __maybe_unused
  drm/bridge: dw-hdmi: fix AVI frame colorimetry

4 years agoMerge tag 'drm-intel-fixes-2020-03-19' of git://anongit.freedesktop.org/drm/drm-intel...
Dave Airlie [Fri, 20 Mar 2020 02:51:34 +0000 (12:51 +1000)]
Merge tag 'drm-intel-fixes-2020-03-19' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

drm/i915 fixes for v5.6-rc7:
- Track active elements during dequeue
- Fix failure to handle all MCR ranges
- Revert unnecessary workaround

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/877dzgepvu.fsf@intel.com
4 years agoMerge tag 'amd-drm-fixes-5.6-2020-03-19' of git://people.freedesktop.org/~agd5f/linux...
Dave Airlie [Fri, 20 Mar 2020 02:48:10 +0000 (12:48 +1000)]
Merge tag 'amd-drm-fixes-5.6-2020-03-19' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

amd-drm-fixes-5.6-2020-03-19:

amdgpu:
- Pageflip fix
- VCN clockgating fixes
- GPR debugfs fix for umr
- GPU reset fix
- eDP fix for MBP
- DCN2.x fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200319204054.1036478-1-alexander.deucher@amd.com
4 years agodrm/amd/display: Fix pageflip event race condition for DCN.
Mario Kleiner [Mon, 2 Mar 2020 06:17:32 +0000 (07:17 +0100)]
drm/amd/display: Fix pageflip event race condition for DCN.

Commit '16f17eda8bad ("drm/amd/display: Send vblank and user
events at vsartup for DCN")' introduces a new way of pageflip
completion handling for DCN, and some trouble.

The current implementation introduces a race condition, which
can cause pageflip completion events to be sent out one vblank
too early, thereby confusing userspace and causing flicker:

prepare_flip_isr():

1. Pageflip programming takes the ddev->event_lock.
2. Sets acrtc->pflip_status == AMDGPU_FLIP_SUBMITTED
3. Releases ddev->event_lock.

--> Deadline for surface address regs double-buffering passes on
    target pipe.

4. dc_commit_updates_for_stream() MMIO programs the new pageflip
   into hw, but too late for current vblank.

=> pflip_status == AMDGPU_FLIP_SUBMITTED, but flip won't complete
   in current vblank due to missing the double-buffering deadline
   by a tiny bit.

5. VSTARTUP trigger point in vblank is reached, VSTARTUP irq fires,
   dm_dcn_crtc_high_irq() gets called.

6. Detects pflip_status == AMDGPU_FLIP_SUBMITTED and assumes the
   pageflip has been completed/will complete in this vblank and
   sends out pageflip completion event to userspace and resets
   pflip_status = AMDGPU_FLIP_NONE.

=> Flip completion event sent out one vblank too early.

This behaviour has been observed during my testing with measurement
hardware a couple of time.

The commit message says that the extra flip event code was added to
dm_dcn_crtc_high_irq() to prevent missing to send out pageflip events
in case the pflip irq doesn't fire, because the "DCH HUBP" component
is clock gated and doesn't fire pflip irqs in that state. Also that
this clock gating may happen if no planes are active. This suggests
that the problem addressed by that commit can't happen if planes
are active.

The proposed solution is therefore to only execute the extra pflip
completion code iff the count of active planes is zero and otherwise
leave pflip completion handling to the pflip irq handler, for a
more race-free experience.

Note that i don't know if this fixes the problem the original commit
tried to address, as i don't know what the test scenario was. It
does fix the observed too early pageflip events though and points
out the problem introduced.

Fixes: 16f17eda8bad ("drm/amd/display: Send vblank and user events at vsartup for DCN")
Reviewed-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
4 years agoMerge tag 'mmc-v5.6-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
Linus Torvalds [Thu, 19 Mar 2020 19:45:14 +0000 (12:45 -0700)]
Merge tag 'mmc-v5.6-rc6' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:

 - rtsx_pci: Fix support for some various speed modes

 - sdhci-of-at91: Fix support for GPIO card detect on SAMA5D2

 - sdhci-cadence: Fix support for DDR52 speed mode for eMMC on UniPhier

 - sdhci-acpi: Fix broken WP support on Acer Aspire Switch 10

 - sdhci-acpi: Workaround FW bug for suspend on Lenovo Miix 320

* tag 'mmc-v5.6-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: rtsx_pci: Fix support for speed-modes that relies on tuning
  mmc: sdhci-of-at91: fix cd-gpios for SAMA5D2
  mmc: sdhci-cadence: set SDHCI_QUIRK2_PRESET_VALUE_BROKEN for UniPhier
  mmc: sdhci-acpi: Disable write protect detection on Acer Aspire Switch 10 (SW5-012)
  mmc: sdhci-acpi: Switch signal voltage back to 3.3V on suspend on external microSD on Lenovo Miix 320

4 years agoarm64: compat: Fix syscall number of compat_clock_getres
Vincenzo Frascino [Thu, 19 Mar 2020 14:11:38 +0000 (14:11 +0000)]
arm64: compat: Fix syscall number of compat_clock_getres

The syscall number of compat_clock_getres was erroneously set to 247
(__NR_io_cancel!) instead of 264. This causes the vDSO fallback of
clock_getres() to land on the wrong syscall for compat tasks.

Fix the numbering.

Cc: <stable@vger.kernel.org>
Fixes: 53c489e1dfeb6 ("arm64: compat: Add missing syscall numbers")
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Vincenzo Frascino <vincenzo.frascino@arm.com>
Signed-off-by: Will Deacon <will@kernel.org>
4 years agoMerge tag '5.6-rc6-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Thu, 19 Mar 2020 17:19:11 +0000 (10:19 -0700)]
Merge tag '5.6-rc6-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fixes from Steve French:
 "Three small smb3 fixes, two for stable"

* tag '5.6-rc6-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  CIFS: fiemap: do not return EINVAL if get nothing
  CIFS: Increment num_remote_opens stats counter even in case of smb2_query_dir_first
  cifs: potential unintitliazed error code in cifs_getattr()

4 years agoMerge tag 'kbuild-fixes-v5.6-3' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 19 Mar 2020 16:57:48 +0000 (09:57 -0700)]
Merge tag 'kbuild-fixes-v5.6-3' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - fix __uint128_t capability test in Kconfig when GCC that defaults to
   32-bit is used to build the 64-bit kernel

 - suppress new noisy Clang warnings -Wpointer-to-enum-cast

 - move the namespace field in Module.symvers for the backward
   compatibility reason for the depmod tool

 - use available compression for initramdisk when INTRAMFS_SOURCE is
   defined, which was the original behavior

 - fix modpost to handle correct large section numbers when it refers to
   modversion CRCs and module namespaces

 - fix comments and documents

* tag 'kbuild-fixes-v5.6-3' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  scripts/kallsyms: fix wrong kallsyms_relative_base
  modpost: Get proper section index by get_secindex() instead of st_shndx
  initramfs: restore default compression behavior
  modpost: move the namespace field in Module.symvers last
  kbuild: Disable -Wpointer-to-enum-cast
  kbuild: doc: fix references to other documents
  int128: fix __uint128_t compiler test in Kconfig
  kconfig: introduce m32-flag and m64-flag
  kbuild: Fix inconsistent comment

4 years agortc: max8907: add missing select REGMAP_IRQ
Corentin Labbe [Wed, 18 Mar 2020 15:26:49 +0000 (15:26 +0000)]
rtc: max8907: add missing select REGMAP_IRQ

I have hit the following build error:

  armv7a-hardfloat-linux-gnueabi-ld: drivers/rtc/rtc-max8907.o: in function `max8907_rtc_probe':
  rtc-max8907.c:(.text+0x400): undefined reference to `regmap_irq_get_virq'

max8907 should select REGMAP_IRQ

Fixes: 94c01ab6d7544 ("rtc: add MAX8907 RTC driver")
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoscripts/kallsyms: fix wrong kallsyms_relative_base
Mikhail Petrov [Wed, 11 Mar 2020 20:37:09 +0000 (23:37 +0300)]
scripts/kallsyms: fix wrong kallsyms_relative_base

There is the code in the read_symbol function in 'scripts/kallsyms.c':

if (is_ignored_symbol(name, type))
return NULL;

/* Ignore most absolute/undefined (?) symbols. */
if (strcmp(name, "_text") == 0)
_text = addr;

But the is_ignored_symbol function returns true for name="_text" and
type='A'. So the next condition is not executed and the _text variable
is always zero.

It makes the wrong kallsyms_relative_base symbol as a result of the code
(CONFIG_KALLSYMS_BASE_RELATIVE is defined):

if (base_relative) {
output_label("kallsyms_relative_base");
output_address(relative_base);
printf("\n");
}

Because the output_address function uses the _text variable.

So the kallsyms_lookup function and all related functions in the kernel
do not work properly. For example, the stack trace in oops:

 Call Trace:
 [aa095e58] [809feab8] kobj_ns_ops_tbl+0x7ff09ac8/0x7ff1c1c4 (unreliable)
 [aa095e98] [80002b64] kobj_ns_ops_tbl+0x7f50db74/0x80000010
 [aa095ef8] [809c3d24] kobj_ns_ops_tbl+0x7feced34/0x7ff1c1c4
 [aa095f28] [80002ed0] kobj_ns_ops_tbl+0x7f50dee0/0x80000010
 [aa095f38] [8000f238] kobj_ns_ops_tbl+0x7f51a248/0x80000010

The right stack trace:

 Call Trace:
 [aa095e58] [809feab8] module_vdu_video_init+0x2fc/0x3bc (unreliable)
 [aa095e98] [80002b64] do_one_initcall+0x40/0x1f0
 [aa095ef8] [809c3d24] kernel_init_freeable+0x164/0x1d8
 [aa095f28] [80002ed0] kernel_init+0x14/0x124
 [aa095f38] [8000f238] ret_from_kernel_thread+0x14/0x1c

[masahiroy@kernel.org:

This issue happens on binutils <= 2.22
The following commit fixed it:
https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=d2667025dd30611514810c28bee9709e4623012a

The symbol type of _text is 'T' on binutils >= 2.23
The minimal supported binutils version for the kernel build is 2.21
]

Signed-off-by: Mikhail Petrov <Mikhail.Petrov@mir.dev>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
4 years agoarm64: kpti: Fix "kpti=off" when KASLR is enabled
Will Deacon [Wed, 18 Mar 2020 20:38:29 +0000 (20:38 +0000)]
arm64: kpti: Fix "kpti=off" when KASLR is enabled

Enabling KASLR forces the use of non-global page-table entries for kernel
mappings, as this is a decision that we have to make very early on before
mapping the kernel proper. When used in conjunction with the "kpti=off"
command-line option, it is possible to use non-global kernel mappings but
with the kpti trampoline disabled.

Since commit 09e3c22a86f6 ("arm64: Use a variable to store non-global
mappings decision"), arm64_kernel_unmapped_at_el0() reflects only the use of
non-global mappings and does not take into account whether the kpti
trampoline is enabled. This breaks context switching of the TPIDRRO_EL0
register for 64-bit tasks, where the clearing of the register is deferred to
the ret-to-user code, but it also breaks the ARM SPE PMU driver which
helpfully recommends passing "kpti=off" on the command line!

Report whether or not KPTI is actually enabled in
arm64_kernel_unmapped_at_el0() and check the 'arm64_use_ng_mappings' global
variable directly when determining the protection flags for kernel mappings.

Cc: Mark Brown <broonie@kernel.org>
Reported-by: Hongbo Yao <yaohongbo@huawei.com>
Tested-by: Hongbo Yao <yaohongbo@huawei.com>
Fixes: 09e3c22a86f6 ("arm64: Use a variable to store non-global mappings decision")
Signed-off-by: Will Deacon <will@kernel.org>
4 years agoMerge tag 'drm-misc-fixes-2020-03-18-1' of git://anongit.freedesktop.org/drm/drm...
Dave Airlie [Thu, 19 Mar 2020 01:47:41 +0000 (11:47 +1000)]
Merge tag 'drm-misc-fixes-2020-03-18-1' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

- drm/lease: fix WARNING in idr_destroy
- Fix AVI frame colorimetry in the dw-hdmi bridge.
- Fix compiler warning in komeda by annotating functions as __maybe_unused.
- Downgrade bochs pci_request_region failure from error to warning to
  workaround firmware fb.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/7654ac39-deb8-c9ca-9fd5-ef77b2636380@linux.intel.com
4 years agodrm/amdgpu: fix typo for vcn2.5/jpeg2.5 idle check
James Zhu [Wed, 18 Mar 2020 21:12:12 +0000 (17:12 -0400)]
drm/amdgpu: fix typo for vcn2.5/jpeg2.5 idle check

fix typo for vcn2.5/jpeg2.5 idle check

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
4 years agodrm/amdgpu: fix typo for vcn2/jpeg2 idle check
James Zhu [Wed, 18 Mar 2020 21:10:56 +0000 (17:10 -0400)]
drm/amdgpu: fix typo for vcn2/jpeg2 idle check

fix typo for vcn2/jpeg2 idle check

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
4 years agodrm/amdgpu: fix typo for vcn1 idle check
James Zhu [Wed, 18 Mar 2020 21:09:05 +0000 (17:09 -0400)]
drm/amdgpu: fix typo for vcn1 idle check

fix typo for vcn1 idle check

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
4 years agolocks: reinstate locks_delete_block optimization
Linus Torvalds [Wed, 18 Mar 2020 11:52:21 +0000 (07:52 -0400)]
locks: reinstate locks_delete_block optimization

There is measurable performance impact in some synthetic tests due to
commit 6d390e4b5d48 (locks: fix a potential use-after-free problem when
wakeup a waiter). Fix the race condition instead by clearing the
fl_blocker pointer after the wake_up, using explicit acquire/release
semantics.

This does mean that we can no longer use the clearing of fl_blocker as
the wait condition, so switch the waiters over to checking whether the
fl_blocked_member list_head is empty.

Reviewed-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: NeilBrown <neilb@suse.de>
Fixes: 6d390e4b5d48 (locks: fix a potential use-after-free problem when wakeup a waiter)
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomodpost: Get proper section index by get_secindex() instead of st_shndx
Xiao Yang [Wed, 18 Mar 2020 10:34:16 +0000 (18:34 +0800)]
modpost: Get proper section index by get_secindex() instead of st_shndx

(uint16_t) st_shndx is limited to 65535(i.e. SHN_XINDEX) so sym_get_data() gets
wrong section index by st_shndx if requested symbol contains extended section
index that is more than 65535.  In this case, we need to get proper section index
by .symtab_shndx section.

Module.symvers generated by building kernel with "-ffunction-sections -fdata-sections"
shows the issue.

Fixes: 56067812d5b0 ("kbuild: modversions: add infrastructure for emitting relative CRCs")
Fixes: e84f9fbbece1 ("modpost: refactor namespace_from_kstrtabns() to not hard-code section name")
Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
4 years agomm: slub: be more careful about the double cmpxchg of freelist
Linus Torvalds [Tue, 17 Mar 2020 18:04:09 +0000 (11:04 -0700)]
mm: slub: be more careful about the double cmpxchg of freelist

This is just a cleanup addition to Jann's fix to properly update the
transaction ID for the slub slowpath in commit fd4d9c7d0c71 ("mm: slub:
add missing TID bump..").

The transaction ID is what protects us against any concurrent accesses,
but we should really also make sure to make the 'freelist' comparison
itself always use the same freelist value that we then used as the new
next free pointer.

Jann points out that if we do all of this carefully, we could skip the
transaction ID update for all the paths that only remove entries from
the lists, and only update the TID when adding entries (to avoid the ABA
issue with cmpxchg and list handling re-adding a previously seen value).

But this patch just does the "make sure to cmpxchg the same value we
used" rather than then try to be clever.

Acked-by: Jann Horn <jannh@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm: slub: add missing TID bump in kmem_cache_alloc_bulk()
Jann Horn [Tue, 17 Mar 2020 00:28:45 +0000 (01:28 +0100)]
mm: slub: add missing TID bump in kmem_cache_alloc_bulk()

When kmem_cache_alloc_bulk() attempts to allocate N objects from a percpu
freelist of length M, and N > M > 0, it will first remove the M elements
from the percpu freelist, then call ___slab_alloc() to allocate the next
element and repopulate the percpu freelist. ___slab_alloc() can re-enable
IRQs via allocate_slab(), so the TID must be bumped before ___slab_alloc()
to properly commit the freelist head change.

Fix it by unconditionally bumping c->tid when entering the slowpath.

Cc: stable@vger.kernel.org
Fixes: ebe909e0fdb3 ("slub: improve bulk alloc strategy")
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agodrm/lease: fix WARNING in idr_destroy
Qiujun Huang [Wed, 18 Mar 2020 07:53:50 +0000 (15:53 +0800)]
drm/lease: fix WARNING in idr_destroy

drm_lease_create takes ownership of leases. And leases will be released
by drm_master_put.

drm_master_put
    ->drm_master_destroy
            ->idr_destroy

So we needn't call idr_destroy again.

Reported-and-tested-by: syzbot+05835159fe322770fe3d@syzkaller.appspotmail.com
Signed-off-by: Qiujun Huang <hqjagain@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/1584518030-4173-1-git-send-email-hqjagain@gmail.com
4 years agotty: fix compat TIOCGSERIAL checking wrong function ptr
Eric Biggers [Mon, 24 Feb 2020 18:20:44 +0000 (10:20 -0800)]
tty: fix compat TIOCGSERIAL checking wrong function ptr

Commit 77654350306a ("take compat TIOC[SG]SERIAL treatment into
tty_compat_ioctl()") changed the compat version of TIOCGSERIAL to start
checking for the presence of the ->set_serial function pointer rather
than ->get_serial.  This appears to be a copy-and-paste error, since
->get_serial is the function pointer that is called as well as the
pointer that is checked by the non-compat version of TIOCGSERIAL.

Fix this by checking the correct function pointer.

Fixes: 77654350306a ("take compat TIOC[SG]SERIAL treatment into tty_compat_ioctl()")
Cc: <stable@vger.kernel.org> # v4.20+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Jiri Slaby <jslaby@suse.cz>
Link: https://lore.kernel.org/r/20200224182044.234553-3-ebiggers@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agotty: fix compat TIOCGSERIAL leaking uninitialized memory
Eric Biggers [Mon, 24 Feb 2020 18:20:43 +0000 (10:20 -0800)]
tty: fix compat TIOCGSERIAL leaking uninitialized memory

Commit 77654350306a ("take compat TIOC[SG]SERIAL treatment into
tty_compat_ioctl()") changed the compat version of TIOCGSERIAL to start
copying a whole 'serial_struct32' to userspace rather than individual
fields, but failed to initialize all padding and fields -- namely the
hole after the 'iomem_reg_shift' field, and the 'reserved' field.

Fix this by initializing the struct to zero.

[v2: use sizeof, and convert the adjacent line for consistency.]

Reported-by: syzbot+8da9175e28eadcb203ce@syzkaller.appspotmail.com
Fixes: 77654350306a ("take compat TIOC[SG]SERIAL treatment into tty_compat_ioctl()")
Cc: <stable@vger.kernel.org> # v4.20+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Jiri Slaby <jslaby@suse.cz>
Link: https://lore.kernel.org/r/20200224182044.234553-2-ebiggers@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agotty: drop outdated comments about release_tty() locking
Eric Biggers [Mon, 24 Feb 2020 07:33:59 +0000 (23:33 -0800)]
tty: drop outdated comments about release_tty() locking

The current version of the TTY code unlocks the tty_struct(s) before
release_tty() rather than after.  Moreover, tty_unlock_pair() no longer
exists.  Thus, remove the outdated comments regarding tty_unlock_pair().

Signed-off-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20200224073359.292795-1-ebiggers@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agommc: rtsx_pci: Fix support for speed-modes that relies on tuning
Ricky Wu [Mon, 16 Mar 2020 02:52:32 +0000 (10:52 +0800)]
mmc: rtsx_pci: Fix support for speed-modes that relies on tuning

The TX/RX register should not be treated the same way to allow for better
support of tuning. Fix this by using a default initial value for TX.

Signed-off-by: Ricky Wu <ricky_wu@realtek.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20200316025232.1167-1-ricky_wu@realtek.com
[Ulf: Updated changelog]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agointel_th: pci: Add Elkhart Lake CPU support
Alexander Shishkin [Tue, 17 Mar 2020 06:22:15 +0000 (08:22 +0200)]
intel_th: pci: Add Elkhart Lake CPU support

This adds support for the Trace Hub in Elkhart Lake CPU.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20200317062215.15598-7-alexander.shishkin@linux.intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agointel_th: Fix user-visible error codes
Alexander Shishkin [Tue, 17 Mar 2020 06:22:14 +0000 (08:22 +0200)]
intel_th: Fix user-visible error codes

There are a few places in the driver that end up returning ENOTSUPP to
the user, replace those with EINVAL.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Fixes: ba82664c134ef ("intel_th: Add Memory Storage Unit driver")
Cc: stable@vger.kernel.org # v4.4+
Link: https://lore.kernel.org/r/20200317062215.15598-6-alexander.shishkin@linux.intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agointel_th: msu: Fix the unexpected state warning
Alexander Shishkin [Tue, 17 Mar 2020 06:22:13 +0000 (08:22 +0200)]
intel_th: msu: Fix the unexpected state warning

The unexpected state warning should only warn on illegal state
transitions. Fix that.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Fixes: 615c164da0eb4 ("intel_th: msu: Introduce buffer interface")
Cc: stable@vger.kernel.org # v5.4+
Link: https://lore.kernel.org/r/20200317062215.15598-5-alexander.shishkin@linux.intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agostm class: sys-t: Fix the use of time_after()
Alexander Shishkin [Tue, 17 Mar 2020 06:22:11 +0000 (08:22 +0200)]
stm class: sys-t: Fix the use of time_after()

The operands of time_after() are in a wrong order in both instances in
the sys-t driver. Fix that.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Fixes: 39f10239df75 ("stm class: p_sys-t: Add support for CLOCKSYNC packets")
Fixes: d69d5e83110f ("stm class: Add MIPI SyS-T protocol support")
Cc: stable@vger.kernel.org # v4.20+
Link: https://lore.kernel.org/r/20200317062215.15598-3-alexander.shishkin@linux.intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoMerge tag 'iio-fixes-for-5.6a' of git://git.kernel.org/pub/scm/linux/kernel/git/jic23...
Greg Kroah-Hartman [Wed, 18 Mar 2020 10:20:42 +0000 (11:20 +0100)]
Merge tag 'iio-fixes-for-5.6a' of git://git./linux/kernel/git/jic23/iio into staging-linus

Jonathan writes:

First set of IIO fixes in the 5.6 cycle.

* adxl372
  - Fix marking of buffered values as big endian.
* ak8974
  - Fix wrong handling of negative values when read from sysfs.
* at91-sama5d2
  - Fix differential mode by ensuring configuration set correctly.
* ping
  - Use the write sensor type for of_ping_match table.
* sps30
  - Kconfig build dependency fix.
* st-sensors
  - Fix a wrong identification of which part the SMO8840 ACPI ID indicates.
* stm32-dsfdm
  - Fix a sleep in atomic issue by not using a trigger when it makes no sense.
* stm32-timer
  - Make sure master mode is disabled when stopping.
* vcnl400
  - Update some sampling periods based on new docs.

* tag 'iio-fixes-for-5.6a' of git://git.kernel.org/pub/scm/linux/kernel/git/jic23/iio:
  iio: ping: set pa_laser_ping_cfg in of_ping_match
  iio: chemical: sps30: fix missing triggered buffer dependency
  iio: st_sensors: remap SMO8840 to LIS2DH12
  iio: light: vcnl4000: update sampling periods for vcnl4040
  iio: light: vcnl4000: update sampling periods for vcnl4200
  iio: accel: adxl372: Set iio_chan BE
  iio: magnetometer: ak8974: Fix negative raw values in sysfs
  iio: trigger: stm32-timer: disable master mode when stopping
  iio: adc: stm32-dfsdm: fix sleep in atomic context
  iio: adc: at91-sama5d2_adc: fix differential channels in triggered mode

4 years agoMerge tag 'usb-serial-5.6-rc7' of https://git.kernel.org/pub/scm/linux/kernel/git...
Greg Kroah-Hartman [Wed, 18 Mar 2020 09:42:57 +0000 (10:42 +0100)]
Merge tag 'usb-serial-5.6-rc7' of https://git./linux/kernel/git/johan/usb-serial into usb-linus

Johan writes:

USB-serial fixes for 5.6-rc7

Here are a couple of new device ids for 5.6-rc.

All have been in linux-next with no reported issues.

* tag 'usb-serial-5.6-rc7' of https://git.kernel.org/pub/scm/linux/kernel/git/johan/usb-serial:
  USB: serial: pl2303: add device-id for HP LD381
  USB: serial: option: add ME910G1 ECM composition 0x110b

4 years agoarm64: smp: fix crash_smp_send_stop() behaviour
Cristian Marussi [Wed, 11 Mar 2020 17:12:45 +0000 (17:12 +0000)]
arm64: smp: fix crash_smp_send_stop() behaviour

On a system configured to trigger a crash_kexec() reboot, when only one CPU
is online and another CPU panics while starting-up, crash_smp_send_stop()
will fail to send any STOP message to the other already online core,
resulting in fail to freeze and registers not properly saved.

Moreover even if the proper messages are sent (case CPUs > 2)
it will similarly fail to account for the booting CPU when executing
the final stop wait-loop, so potentially resulting in some CPU not
been waited for shutdown before rebooting.

A tangible effect of this behaviour can be observed when, after a panic
with kexec enabled and loaded, on the following reboot triggered by kexec,
the cpu that could not be successfully stopped fails to come back online:

[  362.291022] ------------[ cut here ]------------
[  362.291525] kernel BUG at arch/arm64/kernel/cpufeature.c:886!
[  362.292023] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
[  362.292400] Modules linked in:
[  362.292970] CPU: 3 PID: 0 Comm: swapper/3 Kdump: loaded Not tainted 5.6.0-rc4-00003-gc780b890948a #105
[  362.293136] Hardware name: Foundation-v8A (DT)
[  362.293382] pstate: 200001c5 (nzCv dAIF -PAN -UAO)
[  362.294063] pc : has_cpuid_feature+0xf0/0x348
[  362.294177] lr : verify_local_elf_hwcaps+0x84/0xe8
[  362.294280] sp : ffff800011b1bf60
[  362.294362] x29: ffff800011b1bf60 x28: 0000000000000000
[  362.294534] x27: 0000000000000000 x26: 0000000000000000
[  362.294631] x25: 0000000000000000 x24: ffff80001189a25c
[  362.294718] x23: 0000000000000000 x22: 0000000000000000
[  362.294803] x21: ffff8000114aa018 x20: ffff800011156a00
[  362.294897] x19: ffff800010c944a0 x18: 0000000000000004
[  362.294987] x17: 0000000000000000 x16: 0000000000000000
[  362.295073] x15: 00004e53b831ae3c x14: 00004e53b831ae3c
[  362.295165] x13: 0000000000000384 x12: 0000000000000000
[  362.295251] x11: 0000000000000000 x10: 00400032b5503510
[  362.295334] x9 : 0000000000000000 x8 : ffff800010c7e204
[  362.295426] x7 : 00000000410fd0f0 x6 : 0000000000000001
[  362.295508] x5 : 00000000410fd0f0 x4 : 0000000000000000
[  362.295592] x3 : 0000000000000000 x2 : ffff8000100939d8
[  362.295683] x1 : 0000000000180420 x0 : 0000000000180480
[  362.296011] Call trace:
[  362.296257]  has_cpuid_feature+0xf0/0x348
[  362.296350]  verify_local_elf_hwcaps+0x84/0xe8
[  362.296424]  check_local_cpu_capabilities+0x44/0x128
[  362.296497]  secondary_start_kernel+0xf4/0x188
[  362.296998] Code: 52805001 72a00301 6b01001f 54000ec0 (d4210000)
[  362.298652] SMP: stopping secondary CPUs
[  362.300615] Starting crashdump kernel...
[  362.301168] Bye!
[    0.000000] Booting Linux on physical CPU 0x0000000003 [0x410fd0f0]
[    0.000000] Linux version 5.6.0-rc4-00003-gc780b890948a (crimar01@e120937-lin) (gcc version 8.3.0 (GNU Toolchain for the A-profile Architecture 8.3-2019.03 (arm-rel-8.36))) #105 SMP PREEMPT Fri Mar 6 17:00:42 GMT 2020
[    0.000000] Machine model: Foundation-v8A
[    0.000000] earlycon: pl11 at MMIO 0x000000001c090000 (options '')
[    0.000000] printk: bootconsole [pl11] enabled
.....
[    0.138024] rcu: Hierarchical SRCU implementation.
[    0.153472] its@2f020000: unable to locate ITS domain
[    0.154078] its@2f020000: Unable to locate ITS domain
[    0.157541] EFI services will not be available.
[    0.175395] smp: Bringing up secondary CPUs ...
[    0.209182] psci: failed to boot CPU1 (-22)
[    0.209377] CPU1: failed to boot: -22
[    0.274598] Detected PIPT I-cache on CPU2
[    0.278707] GICv3: CPU2: found redistributor 1 region 0:0x000000002f120000
[    0.285212] CPU2: Booted secondary processor 0x0000000001 [0x410fd0f0]
[    0.369053] Detected PIPT I-cache on CPU3
[    0.372947] GICv3: CPU3: found redistributor 2 region 0:0x000000002f140000
[    0.378664] CPU3: Booted secondary processor 0x0000000002 [0x410fd0f0]
[    0.401707] smp: Brought up 1 node, 3 CPUs
[    0.404057] SMP: Total of 3 processors activated.

Make crash_smp_send_stop() account also for the online status of the
calling CPU while evaluating how many CPUs are effectively online: this way
the right number of STOPs is sent and all other stopped-cores's registers
are properly saved.

Fixes: 78fd584cdec05 ("arm64: kdump: implement machine_crash_shutdown()")
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>
Signed-off-by: Will Deacon <will@kernel.org>
4 years agoarm64: smp: fix smp_send_stop() behaviour
Cristian Marussi [Wed, 11 Mar 2020 17:12:44 +0000 (17:12 +0000)]
arm64: smp: fix smp_send_stop() behaviour

On a system with only one CPU online, when another one CPU panics while
starting-up, smp_send_stop() will fail to send any STOP message to the
other already online core, resulting in a system still responsive and
alive at the end of the panic procedure.

[  186.700083] CPU3: shutdown
[  187.075462] CPU2: shutdown
[  187.162869] CPU1: shutdown
[  188.689998] ------------[ cut here ]------------
[  188.691645] kernel BUG at arch/arm64/kernel/cpufeature.c:886!
[  188.692079] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
[  188.692444] Modules linked in:
[  188.693031] CPU: 3 PID: 0 Comm: swapper/3 Not tainted 5.6.0-rc4-00001-g338d25c35a98 #104
[  188.693175] Hardware name: Foundation-v8A (DT)
[  188.693492] pstate: 200001c5 (nzCv dAIF -PAN -UAO)
[  188.694183] pc : has_cpuid_feature+0xf0/0x348
[  188.694311] lr : verify_local_elf_hwcaps+0x84/0xe8
[  188.694410] sp : ffff800011b1bf60
[  188.694536] x29: ffff800011b1bf60 x28: 0000000000000000
[  188.694707] x27: 0000000000000000 x26: 0000000000000000
[  188.694801] x25: 0000000000000000 x24: ffff80001189a25c
[  188.694905] x23: 0000000000000000 x22: 0000000000000000
[  188.694996] x21: ffff8000114aa018 x20: ffff800011156a38
[  188.695089] x19: ffff800010c944a0 x18: 0000000000000004
[  188.695187] x17: 0000000000000000 x16: 0000000000000000
[  188.695280] x15: 0000249dbde5431e x14: 0262cbe497efa1fa
[  188.695371] x13: 0000000000000002 x12: 0000000000002592
[  188.695472] x11: 0000000000000080 x10: 00400032b5503510
[  188.695572] x9 : 0000000000000000 x8 : ffff800010c80204
[  188.695659] x7 : 00000000410fd0f0 x6 : 0000000000000001
[  188.695750] x5 : 00000000410fd0f0 x4 : 0000000000000000
[  188.695836] x3 : 0000000000000000 x2 : ffff8000100939d8
[  188.695919] x1 : 0000000000180420 x0 : 0000000000180480
[  188.696253] Call trace:
[  188.696410]  has_cpuid_feature+0xf0/0x348
[  188.696504]  verify_local_elf_hwcaps+0x84/0xe8
[  188.696591]  check_local_cpu_capabilities+0x44/0x128
[  188.696666]  secondary_start_kernel+0xf4/0x188
[  188.697150] Code: 52805001 72a00301 6b01001f 54000ec0 (d4210000)
[  188.698639] ---[ end trace 3f12ca47652f7b72 ]---
[  188.699160] Kernel panic - not syncing: Attempted to kill the idle task!
[  188.699546] Kernel Offset: disabled
[  188.699828] CPU features: 0x00004,20c02008
[  188.700012] Memory Limit: none
[  188.700538] ---[ end Kernel panic - not syncing: Attempted to kill the idle task! ]---

[root@arch ~]# echo Helo
Helo
[root@arch ~]# cat /proc/cpuinfo | grep proce
processor : 0

Make smp_send_stop() account also for the online status of the calling CPU
while evaluating how many CPUs are effectively online: this way, the right
number of STOPs is sent, so enforcing a proper freeze of the system at the
end of panic even under the above conditions.

Fixes: 08e875c16a16c ("arm64: SMP support")
Reported-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>
Signed-off-by: Will Deacon <will@kernel.org>
4 years agoUSB: cdc-acm: fix rounding error in TIOCSSERIAL
Anthony Mallet [Thu, 12 Mar 2020 13:31:01 +0000 (14:31 +0100)]
USB: cdc-acm: fix rounding error in TIOCSSERIAL

By default, tty_port_init() initializes those parameters to a multiple
of HZ. For instance in line 69 of tty_port.c:
   port->close_delay = (50 * HZ) / 100;
https://github.com/torvalds/linux/blob/master/drivers/tty/tty_port.c#L69

With e.g. CONFIG_HZ = 250 (as this is the case for Ubuntu 18.04
linux-image-4.15.0-37-generic), the default setting for close_delay is
thus 125.

When ioctl(fd, TIOCGSERIAL, &s) is executed, the setting returned in
user space is '12' (125/10). When ioctl(fd, TIOCSSERIAL, &s) is then
executed with the same setting '12', the value is interpreted as '120'
which is different from the current setting and a EPERM error may be
raised by set_serial_info() if !CAP_SYS_ADMIN.
https://github.com/torvalds/linux/blob/master/drivers/usb/class/cdc-acm.c#L919

Fixes: ba2d8ce9db0a6 ("cdc-acm: implement TIOCSSERIAL to avoid blocking close(2)")
Signed-off-by: Anthony Mallet <anthony.mallet@laas.fr>
Cc: stable <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200312133101.7096-2-anthony.mallet@laas.fr
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoUSB: cdc-acm: fix close_delay and closing_wait units in TIOCSSERIAL
Anthony Mallet [Thu, 12 Mar 2020 13:31:00 +0000 (14:31 +0100)]
USB: cdc-acm: fix close_delay and closing_wait units in TIOCSSERIAL

close_delay and closing_wait are specified in hundredth of a second but stored
internally in jiffies. Use the jiffies_to_msecs() and msecs_to_jiffies()
functions to convert from each other.

Signed-off-by: Anthony Mallet <anthony.mallet@laas.fr>
Cc: stable <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200312133101.7096-1-anthony.mallet@laas.fr
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agousb: quirks: add NO_LPM quirk for RTL8153 based ethernet adapters
Hans de Goede [Fri, 13 Mar 2020 12:07:08 +0000 (13:07 +0100)]
usb: quirks: add NO_LPM quirk for RTL8153 based ethernet adapters

We have been receiving bug reports that ethernet connections over
RTL8153 based ethernet adapters stops working after a while with
errors like these showing up in dmesg when the ethernet stops working:

[12696.189484] r8152 6-1:1.0 enp10s0u1: Tx timeout
[12702.333456] r8152 6-1:1.0 enp10s0u1: Tx timeout
[12707.965422] r8152 6-1:1.0 enp10s0u1: Tx timeout

This has been reported on Dell WD15 docks, Belkin USB-C Express Dock 3.1
docks and with generic USB to ethernet dongles using the RTL8153
chipsets. Some users have tried adding usbcore.quirks=0bda:8153:k to
the kernel commandline and all users who have tried this report that
this fixes this.

Also note that we already have an existing NO_LPM quirk for the RTL8153
used in the Microsoft Surface Dock (where it uses a different usb-id).

This commit adds a NO_LPM quirk for the generic Realtek RTL8153
0bda:8153 usb-id, fixing the Tx timeout errors on these devices.

BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=198931
Cc: stable@vger.kernel.org
Cc: russianneuromancer@ya.ru
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Link: https://lore.kernel.org/r/20200313120708.100339-1-hdegoede@redhat.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agousb: chipidea: udc: fix sleeping function called from invalid context
Peter Chen [Mon, 16 Mar 2020 03:10:34 +0000 (11:10 +0800)]
usb: chipidea: udc: fix sleeping function called from invalid context

The code calls pm_runtime_get_sync with irq disabled, it causes below
warning:

BUG: sleeping function called from invalid context at
wer/runtime.c:1075
in_atomic(): 1, irqs_disabled(): 128, non_block: 0, pid:
er/u8:1
CPU: 1 PID: 37 Comm: kworker/u8:1 Not tainted
20200304-00181-gbebfd2a5be98 #1588
Hardware name: NVIDIA Tegra SoC (Flattened Device Tree)
Workqueue: ci_otg ci_otg_work
[<c010e8bd>] (unwind_backtrace) from [<c010a315>]
1/0x14)
[<c010a315>] (show_stack) from [<c0987d29>]
5/0x94)
[<c0987d29>] (dump_stack) from [<c013e77f>]
+0xeb/0x118)
[<c013e77f>] (___might_sleep) from [<c052fa1d>]
esume+0x75/0x78)
[<c052fa1d>] (__pm_runtime_resume) from [<c0627a33>]
0x23/0x74)
[<c0627a33>] (ci_udc_pullup) from [<c062fb93>]
nect+0x2b/0xcc)
[<c062fb93>] (usb_gadget_connect) from [<c062769d>]
_connect+0x59/0x104)
[<c062769d>] (ci_hdrc_gadget_connect) from [<c062778b>]
ssion+0x43/0x48)
[<c062778b>] (ci_udc_vbus_session) from [<c062f997>]
s_connect+0x17/0x9c)
[<c062f997>] (usb_gadget_vbus_connect) from [<c062634d>]
bd/0x128)
[<c062634d>] (ci_otg_work) from [<c0134719>]
rk+0x149/0x404)
[<c0134719>] (process_one_work) from [<c0134acb>]
0xf7/0x3bc)
[<c0134acb>] (worker_thread) from [<c0139433>]
x118)
[<c0139433>] (kthread) from [<c01010bd>]
(ret_from_fork+0x11/0x34)

Tested-by: Dmitry Osipenko <digetx@gmail.com>
Cc: <stable@vger.kernel.org> #v5.5
Fixes: 72dc8df7920f ("usb: chipidea: udc: protect usb interrupt enable")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Peter Chen <peter.chen@nxp.com>
Link: https://lore.kernel.org/r/20200316031034.17847-2-peter.chen@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoCIFS: fiemap: do not return EINVAL if get nothing
Murphy Zhou [Sat, 14 Mar 2020 03:38:31 +0000 (11:38 +0800)]
CIFS: fiemap: do not return EINVAL if get nothing

If we call fiemap on a truncated file with none blocks allocated,
it makes sense we get nothing from this call. No output means
no blocks have been counted, but the call succeeded. It's a valid
response.

Simple example reproducer:
xfs_io -f 'truncate 2M' -c 'fiemap -v' /cifssch/testfile
xfs_io: ioctl(FS_IOC_FIEMAP) ["/cifssch/testfile"]: Invalid argument

Signed-off-by: Murphy Zhou <jencce.kernel@gmail.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
CC: Stable <stable@vger.kernel.org>
4 years agoCIFS: Increment num_remote_opens stats counter even in case of smb2_query_dir_first
Shyam Prasad N [Mon, 9 Mar 2020 08:35:09 +0000 (01:35 -0700)]
CIFS: Increment num_remote_opens stats counter even in case of smb2_query_dir_first

The num_remote_opens counter keeps track of the number of open files which must be
maintained by the server at any point. This is a per-tree-connect counter, and the value
of this counter gets displayed in the /proc/fs/cifs/Stats output as a following...

Open files: 0 total (local), 1 open on server
                             ^^^^^^^^^^^^^^^^
As a thumb-rule, we want to increment this counter for each open/create that we
successfully execute on the server. Similarly, we should decrement the counter when
we successfully execute a close.

In this case, an increment was being missed in case of smb2_query_dir_first,
in case of successful open. As a result, we would underflow the counter and we
could even see the counter go to negative after sufficient smb2_query_dir_first calls.

I tested the stats counter for a bunch of filesystem operations with the fix.
And it looks like the counter looks correct to me.

I also check if we missed the increments and decrements elsewhere. It does not
seem so. Few other cases where an open is done and we don't increment the counter are
the compound calls where the corresponding close is also sent in the request.

Signed-off-by: Shyam Prasad N <nspmangalore@gmail.com>
CC: Stable <stable@vger.kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Aurelien Aptel <aaptel@suse.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
4 years agocifs: potential unintitliazed error code in cifs_getattr()
Dan Carpenter [Fri, 28 Feb 2020 09:22:59 +0000 (12:22 +0300)]
cifs: potential unintitliazed error code in cifs_getattr()

Smatch complains that "rc" could be uninitialized.

    fs/cifs/inode.c:2206 cifs_getattr() error: uninitialized symbol 'rc'.

Changing it to "return 0;" improves readability as well.

Fixes: cc1baf98c8f6 ("cifs: do not ignore the SYNC flags in getattr")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Acked-by: Ronnie Sahlberg <lsahlber@redhat.com>
4 years agoALSA: hda/realtek - Enable the headset of Acer N50-600 with ALC662
Jian-Hong Pan [Tue, 17 Mar 2020 08:28:09 +0000 (16:28 +0800)]
ALSA: hda/realtek - Enable the headset of Acer N50-600 with ALC662

A headset on the desktop like Acer N50-600 does not work, until quirk
ALC662_FIXUP_ACER_NITRO_HEADSET_MODE is applied.

Signed-off-by: Jian-Hong Pan <jian-hong@endlessm.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200317082806.73194-3-jian-hong@endlessm.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
4 years agoALSA: hda/realtek - Enable headset mic of Acer X2660G with ALC662
Jian-Hong Pan [Tue, 17 Mar 2020 08:28:07 +0000 (16:28 +0800)]
ALSA: hda/realtek - Enable headset mic of Acer X2660G with ALC662

The Acer desktop X2660G with ALC662 can't detect the headset microphone
until ALC662_FIXUP_ACER_X2660G_HEADSET_MODE quirk applied.

Signed-off-by: Jian-Hong Pan <jian-hong@endlessm.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200317082806.73194-2-jian-hong@endlessm.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
4 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid
Linus Torvalds [Tue, 17 Mar 2020 16:38:03 +0000 (09:38 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/hid/hid

Pull HID fixes from Jiri Kosina:

 - string buffer formatting fixes in picolcd and sensor drivers, from
   Takashi Iwai

 - two new device IDs from Chen-Tsung Hsieh and Tony Fischetti

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid:
  HID: add ALWAYS_POLL quirk to lenovo pixart mouse
  HID: google: add moonball USB id
  HID: hid-sensor-custom: Use scnprintf() for avoiding potential buffer overflow
  HID: hid-picolcd_fb: Use scnprintf() for avoiding potential buffer overflow

4 years agox86/purgatory: Fail the build if purgatory.ro has missing symbols
Hans de Goede [Tue, 17 Mar 2020 13:08:41 +0000 (14:08 +0100)]
x86/purgatory: Fail the build if purgatory.ro has missing symbols

Linking purgatory.ro with -r enables "incremental linking"; this means
no checks for unresolved symbols are done while linking purgatory.ro.

A change to the sha256 code has caused the purgatory in 5.4-rc1 to have
a missing symbol on memzero_explicit(), yet things still happily build.

Add an extra check for unresolved symbols by calling ld without -r
before running bin2c to generate kexec-purgatory.c.

This causes a build of 5.4-rc1 with this patch added to fail as it should:

    CHK     arch/x86/purgatory/purgatory.ro
  ld: arch/x86/purgatory/purgatory.ro: in function `sha256_transform':
  sha256.c:(.text+0x1c0c): undefined reference to `memzero_explicit'
  make[2]: *** [arch/x86/purgatory/Makefile:72:
      arch/x86/purgatory/kexec-purgatory.c] Error 1
  make[1]: *** [scripts/Makefile.build:509: arch/x86/purgatory] Error 2
  make: *** [Makefile:1650: arch/x86] Error 2

Also remove --no-undefined from LDFLAGS_purgatory.ro as that has no
effect.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200317130841.290418-2-hdegoede@redhat.com
4 years agox86/purgatory: Disable various profiling and sanitizing options
Hans de Goede [Tue, 17 Mar 2020 13:08:40 +0000 (14:08 +0100)]
x86/purgatory: Disable various profiling and sanitizing options

Since the purgatory is a special stand-alone binary, various profiling
and sanitizing options must be disabled. Having these options enabled
typically will cause dependencies on various special symbols exported by
special libs / stubs used by these frameworks. Since the purgatory is
special, it is not linked against these stubs causing missing symbols in
the purgatory if these options are not disabled.

Sync the set of disabled profiling and sanitizing options with that from
drivers/firmware/efi/libstub/Makefile, adding
-DDISABLE_BRANCH_PROFILING to the CFLAGS and setting:

  GCOV_PROFILE                    := n
  UBSAN_SANITIZE                  := n

This fixes broken references to ftrace_likely_update() when
CONFIG_TRACE_BRANCH_PROFILING is enabled and to __gcov_init() and
__gcov_exit() when CONFIG_GCOV_KERNEL is enabled.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200317130841.290418-1-hdegoede@redhat.com
4 years agostaging: rtl8188eu: Add device id for MERCUSYS MW150US v2
Michael Straube [Thu, 12 Mar 2020 09:36:52 +0000 (10:36 +0100)]
staging: rtl8188eu: Add device id for MERCUSYS MW150US v2

This device was added to the stand-alone driver on github.
Add it to the staging driver as well.

Link: https://github.com/lwfinger/rtl8188eu/commit/2141f244c3e7
Signed-off-by: Michael Straube <straube.linux@gmail.com>
Cc: stable <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200312093652.13918-1-straube.linux@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agostaging: greybus: loopback_test: fix potential path truncations
Johan Hovold [Thu, 12 Mar 2020 11:01:51 +0000 (12:01 +0100)]
staging: greybus: loopback_test: fix potential path truncations

Newer GCC warns about possible truncations of two generated path names as
we're concatenating the configurable sysfs and debugfs path prefixes
with a filename and placing the results in buffers of the same size as
the maximum length of the prefixes.

snprintf(d->name, MAX_STR_LEN, "gb_loopback%u", dev_id);

snprintf(d->sysfs_entry, MAX_SYSFS_PATH, "%s%s/",
 t->sysfs_prefix, d->name);

snprintf(d->debugfs_entry, MAX_SYSFS_PATH, "%sraw_latency_%s",
 t->debugfs_prefix, d->name);

Fix this by separating the maximum path length from the maximum prefix
length and reducing the latter enough to fit the generated strings.

Note that we also need to reduce the device-name buffer size as GCC
isn't smart enough to figure out that we ever only used MAX_STR_LEN
bytes of it.

Fixes: 6b0658f68786 ("greybus: tools: Add tools directory to greybus repo and add loopback")
Signed-off-by: Johan Hovold <johan@kernel.org>
Link: https://lore.kernel.org/r/20200312110151.22028-4-johan@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agostaging: greybus: loopback_test: fix potential path truncation
Johan Hovold [Thu, 12 Mar 2020 11:01:50 +0000 (12:01 +0100)]
staging: greybus: loopback_test: fix potential path truncation

Newer GCC warns about a possible truncation of a generated sysfs path
name as we're concatenating a directory path with a file name and
placing the result in a buffer that is half the size of the maximum
length of the directory path (which is user controlled).

loopback_test.c: In function 'open_poll_files':
loopback_test.c:651:31: warning: '%s' directive output may be truncated writing up to 511 bytes into a region of size 255 [-Wformat-truncation=]
  651 |   snprintf(buf, sizeof(buf), "%s%s", dev->sysfs_entry, "iteration_count");
      |                               ^~
loopback_test.c:651:3: note: 'snprintf' output between 16 and 527 bytes into a destination of size 255
  651 |   snprintf(buf, sizeof(buf), "%s%s", dev->sysfs_entry, "iteration_count");
      |   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Fix this by making sure the buffer is large enough the concatenated
strings.

Fixes: 6b0658f68786 ("greybus: tools: Add tools directory to greybus repo and add loopback")
Fixes: 9250c0ee2626 ("greybus: Loopback_test: use poll instead of inotify")
Signed-off-by: Johan Hovold <johan@kernel.org>
Link: https://lore.kernel.org/r/20200312110151.22028-3-johan@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agostaging: greybus: loopback_test: fix poll-mask build breakage
Johan Hovold [Thu, 12 Mar 2020 11:01:49 +0000 (12:01 +0100)]
staging: greybus: loopback_test: fix poll-mask build breakage

A scripted conversion from userland POLL* to kernel EPOLL* constants
mistakingly replaced the poll flags in the loopback_test tool, which
therefore no longer builds.

Fixes: a9a08845e9ac ("vfs: do bulk POLL* -> EPOLL* replacement")
Cc: stable <stable@vger.kernel.org> # 4.16
Signed-off-by: Johan Hovold <johan@kernel.org>
Link: https://lore.kernel.org/r/20200312110151.22028-2-johan@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agommc: sdhci-of-at91: fix cd-gpios for SAMA5D2
Michał Mirosław [Sun, 15 Mar 2020 16:44:25 +0000 (17:44 +0100)]
mmc: sdhci-of-at91: fix cd-gpios for SAMA5D2

SAMA5D2x doesn't drive CMD line if GPIO is used as CD line (at least
SAMA5D27 doesn't). Fix this by forcing card-detect in the module
if module-controlled CD is not used.

Fixed commit addresses the problem only for non-removable cards. This
amends it to also cover gpio-cd case.

Cc: stable@vger.kernel.org
Fixes: 7a1e3f143176 ("mmc: sdhci-of-at91: force card detect value for non removable devices")
Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/8d10950d9940468577daef4772b82a071b204716.1584290561.git.mirq-linux@rere.qmqm.pl
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
4 years agommc: sdhci-cadence: set SDHCI_QUIRK2_PRESET_VALUE_BROKEN for UniPhier
Masahiro Yamada [Thu, 12 Mar 2020 10:42:57 +0000 (19:42 +0900)]
mmc: sdhci-cadence: set SDHCI_QUIRK2_PRESET_VALUE_BROKEN for UniPhier

The SDHCI_PRESET_FOR_* registers are not set for the UniPhier platform
integration. (They are all read as zeros).

Set the SDHCI_QUIRK2_PRESET_VALUE_BROKEN quirk flag. Otherwise, the
High Speed DDR mode on the eMMC controller (MMC_TIMING_MMC_DDR52)
would not work.

I split the platform data to give no impact to other platforms,
although the UniPhier platform is currently only the upstream user
of this IP.

The SDHCI_QUIRK2_PRESET_VALUE_BROKEN flag is set if the compatible
string matches to "socionext,uniphier-sd4hc".

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20200312104257.21017-1-yamada.masahiro@socionext.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
4 years agommc: sdhci-acpi: Disable write protect detection on Acer Aspire Switch 10 (SW5-012)
Hans de Goede [Mon, 16 Mar 2020 18:47:53 +0000 (19:47 +0100)]
mmc: sdhci-acpi: Disable write protect detection on Acer Aspire Switch 10 (SW5-012)

On the Acer Aspire Switch 10 (SW5-012) microSD slot always reports the card
being write-protected even though microSD cards do not have a write-protect
switch at all.

Add a new DMI_QUIRK_SD_NO_WRITE_PROTECT quirk which when set sets
the MMC_CAP2_NO_WRITE_PROTECT flag on the controller for the external SD
slot; and add a DMI quirk table entry which selects this quirk for the
Acer SW5-012.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20200316184753.393458-2-hdegoede@redhat.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
4 years agommc: sdhci-acpi: Switch signal voltage back to 3.3V on suspend on external microSD...
Hans de Goede [Mon, 16 Mar 2020 18:47:52 +0000 (19:47 +0100)]
mmc: sdhci-acpi: Switch signal voltage back to 3.3V on suspend on external microSD on Lenovo Miix 320

Based on a sample of 7 DSDTs from Cherry Trail devices using an AXP288
PMIC depending on the design one of 2 possible LDOs on the PMIC is used
for the MMC signalling voltage, either DLDO3 or GPIO1LDO (GPIO1 pin in
low noise LDO mode).

The Lenovo Miix 320-10ICR uses GPIO1LDO in the SHC1 ACPI device's DSM
methods to set 3.3 or 1.8 signalling voltage and this appears to work
as advertised, so presumably the device is actually using GPIO1LDO for
the external microSD signalling voltage.

But this device has a bug in the _PS0 method of the SHC1 ACPI device,
the DSM remembers the last set signalling voltage and the _PS0 restores
this after a (runtime) suspend-resume cycle, but it "restores" the voltage
on DLDO3 instead of setting it on GPIO1LDO as the DSM method does. DLDO3
is used for the LCD and setting it to 1.8V causes the LCD to go black.

This commit works around this issue by calling the Intel DSM to reset the
signal voltage to 3.3V after the host has been runtime suspended.
This will make the _PS0 method reprogram the DLDO3 voltage to 3.3V, which
leaves it at its original setting fixing the LCD going black.

This commit adds and uses a DMI quirk mechanism to only trigger this
workaround on the Lenovo Miix 320 while leaving the behavior of the
driver unchanged on other devices.

BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=111294
BugLink: https://gitlab.freedesktop.org/drm/intel/issues/355
Reported-by: russianneuromancer <russianneuromancer@ya.ru>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20200316184753.393458-1-hdegoede@redhat.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
4 years agoinitramfs: restore default compression behavior
Eugeniy Paltsev [Mon, 16 Mar 2020 11:25:19 +0000 (14:25 +0300)]
initramfs: restore default compression behavior

Even though INITRAMFS_SOURCE kconfig option isn't set in most of
defconfigs it is used (set) extensively by various build systems.
Commit f26661e12765 ("initramfs: make initramfs compression choice
non-optional") has changed default compression mode. Previously we
compress initramfs using available compression algorithm. Now
we don't use any compression at all by default.
It significantly increases the image size in case of build system
chooses embedded initramfs. Initially I faced with this issue while
using buildroot.

As of today it's not possible to set preferred compression mode
in target defconfig as this option depends on INITRAMFS_SOURCE
being set. Modification of all build systems either doesn't look
like good option.

Let's instead rewrite initramfs compression mode choices list
the way that "INITRAMFS_COMPRESSION_NONE" will be the last option
in the list. In that case it will be chosen only if all other
options (which implements any compression) are not available.

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
4 years agomodpost: move the namespace field in Module.symvers last
Jessica Yu [Wed, 11 Mar 2020 17:01:20 +0000 (18:01 +0100)]
modpost: move the namespace field in Module.symvers last

In order to preserve backwards compatability with kmod tools, we have to
move the namespace field in Module.symvers last, as the depmod -e -E
option looks at the first three fields in Module.symvers to check symbol
versions (and it's expected they stay in the original order of crc,
symbol, module).

In addition, update an ancient comment above read_dump() in modpost that
suggested that the export type field in Module.symvers was optional. I
suspect that there were historical reasons behind that comment that are
no longer accurate. We have been unconditionally printing the export
type since 2.6.18 (commit bd5cbcedf44), which is over a decade ago now.

Fix up read_dump() to treat each field as non-optional. I suspect the
original read_dump() code treated the export field as optional in order
to support pre <= 2.6.18 Module.symvers (which did not have the export
type field). Note that although symbol namespaces are optional, the
field will not be omitted from Module.symvers if a symbol does not have
a namespace. In this case, the field will simply be empty and the next
delimiter or end of line will follow.

Cc: stable@vger.kernel.org
Fixes: cb9b55d21fe0 ("modpost: add support for symbol namespaces")
Tested-by: Matthias Maennich <maennich@google.com>
Reviewed-by: Matthias Maennich <maennich@google.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Jessica Yu <jeyu@kernel.org>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>