The implementation from March 2021 added proper support for designators
but didn't model the brace levels correctly. In particular, it could
not handle additional braces or omitted braces. In such a case, lint
skipped the remaining initializers from the initialization. Due to
this, type errors in the remaining initializers went unnoticed. Another
effect was that arrays of unknown size were wrongly reported as having
size 0.
Both GCC and Clang recommend placing braces around each sub-type that is
initialized, such as a struct, union or array. Postfix does not follow
these recommendations, therefore lint had to be disabled in
external/ibm-public/postfix/Makefile.inc. This commit fixes the bugs
mentioned there.
This code paths is entered by kthreads marked MP-safe, not just from
autoconf.
I'm not sure this is sufficient -- it's not clear to me whether
anything prevents concurrently scanning the same target. Someone with
a better understanding of scsi(4) locking will have to audit this.
(For example, maybe it is guaranteed only to happen only either (a)
in autoconf, or (b) in a thread that doesn't start until autoconf is
done. But I don't know -- and if it is this, it should be asserted
so we can verify it.)
The newly added tests triggered the assertion in begin_designation since
for incomplete types the initialization is stopped before handling the
first brace.
The previous name could be mistaken to mean "increase the indentation of
the debug output". Instead, the function prints the current indentation.
In externs1.h, the macro definition was a duplicate, the macros were
sorted differently than the functions a few lines above.
No binary change.
synchronize_rcu() if the caller uses SLAB_TYPESAFE_BY_RCU. A
special pool allocator is not required in this case.
- Now that SLAB_TYPESAFE_BY_RCU does the right thing, no need to
call synchronize_rcu() in __i915_request_dtor().
a function to be called before the destructor for a batch of one or more
objects is called. This can be used as a synchronization point by
subsystems that rely on the type-stable nature of pool cache objects or
subsystems that use other forms of passive serialization.
915_sw_fence.c: for debug_fence_init_onstack
drm_mm.c: for rb_hole_size_to_node
intel_hdmi.c: for intel_hdmi_hdcp2_protocol
intel_guc_submission.c: for rq_prio
nouveau_nvkm_subdev_instmem_base.c: for fake_ioread32_native
nouveau_nvkm_subdev_mmu_vmmnv04.c: for nv04_vmm_pgt_pte
style to all of ehci_alloc_{sqh,sqtd,itd,sitd}.
While ehci can address memory >4GB for control structures
EHCI_CTRLDSSEGMENT is always set to zero so ehci assumes it's using the
low 4GB.
Nothing prevents a second worker from being queued when the first one
is about to do rnd_detach_source. Instead, just set a flag so future
requests don't bother running a new thread; if there's a concurrent
one that's already been scheduled on another CPU, well, too bad, we
get a couple extra log messages but that's fine.
A better way to do this would probably be to detect whether the tpm
is deactivated at attach time, but that requires reading more of the
tpm spec than I care to do when there are alternative ways to
procrastinate like scrubbing the toilet.
riastradh