non-interlocked CAS in the fast unlock path -- it is unsafe to test for
the waiters-bit while the owner thread is running, we have to spin for
the owner or its state change to be sure about the presence of the bit.
Split off the logic into the pthread__mutex_setwaiters() routine.
This is a partial fix to the named lockup problem (also see PR/44756).
It seems there is another race which can be reproduced on faster CPUs.
has two parts:
- in pthread_cond_timedwait() if the thread we are trying to unpark
exited, retry the the _lwp_park call without it.
- pthread_mutex() was affecting errno since it is calling _lwp_park()
from pthread_mutex_lock_slow(). preserve the original errno.
Note that the example problem still causes an occassional deadlock on machines
with many CPUs and it is the same deadlock we observe with named.
after call pthread_condattr_setclock(CLOCK_MONOTONIC)
_lwp_park(2) expects a realtime clock, and it gets passed a monotonic
one. Since monotonic < real, it never sleeps. This patch adjusts
the monotonic clock to be a real one before it passes is to
_lwp_park(2). This is the minimal hacky fix and it will be fixed
properly in _lwp_park(2) in the future.
XXX: pullup to 6.
- Fail if the dlopened libpthread does pthread_create(). From manu@
- Discussed at length in the mailing lists; approved by core@
- This was chosen as the least intrusive patch that will provide
the necessary functionality.
XXX: pullup to 6
and pthread_{g,s}etspecific functions, to one that invalidates
values of keys in other threads when pthread_key_delete() is called.
This fixes chromium, which expects pthread_key_delete() to do
cleanup in all threads.
.Vt type var No = Dv INITIALIZER;
to provide examples of static initialization in SYNOPSIS section.
.Va macro does the wrong thing (check groff PostScript output) and the
need for .Pp kludge (now dropped) should have been an indication too.
While here move static initialization example to be right after the
declaration of *_init().
the TLS pointer, therefore wrecking the pthread environement.
Some ports had _UC_TLSBASE flag or equivalent (_UC_UNIQUE on alpha)
that controlled whether setcontext() would change the TLS pointer.
This change let libpthread override setcontext() with its own version
that unsets _UC_TLSBASE, enabling safe usage of setcontext() with
-lpthread.
We also have the following required changes here:
- rename alpha's _UC_UNIQUE into _UC_TLSBASE
- add _UC_TLSBASE definition in header file for all ports
(powerpc, sh3, sparc and sparc64 lack the implementation for now)
- introduce a libc stub that can be overriden for setcontext()
- modify MD libcs swapcontext() implementations so that they use the
setcontext() libc stub instead of doing a plain system call.
While we are there:
- document various MD _UC_* flags in header file
- add libc and libpthread tests for swapcontext() behavior
(hopefully helpful to spot MD problems introduced with this change)
Future work:
- Deciding whether kernel support or _UC_TLSBASE should be added for
powerpc, sh3, sparc and sparc64 is left to portmasters
sparc64
Approved by core@
pthread__smt_pause. These are implemented using the ARM instructions
SEV (wake) and WFE (pause). These are treated as NOPs on ARM CPUs that
don't support them.
(this is not entirely correct because it can return ENOMEM which is
not mentioned in the spec, but there are other places in pthread_create
whete ENOMEM is returned -- it at all, this should be fixed everywhere)
