as normal device interrupts. Because of this, we won't get IPIs while
servicing such interrupts. This can lead to the following deadlock
scenario as reported by Bill Sommerfeld:
- Process runs on cpu1, but has FP state on cpu0.
- Process executes FP-using insn, causing an FP trap, which causes
the kernel lock to be acquired.
- At roughly the same time, cpu0 receives a device interrupt, and attempts
to acquire the kernel lock, which blocks since cpu1 already has it.
- cpu1 sends cpu0 a SYNCH FPU IPI, and waits for cpu0 to release its
FP state.
- Since cpu0 cannot notice the IPI until it has processed the device
interrupt, which it cannot do because it cannot acquire the kernel
lock, we have deadlock.
Solve the problem by adding a spinlock interlock release hook which
checks for pending IPIs and processes them.
Idea from Bill Sommerfeld.
- MB_LEN_MAX is increased to 32.
- To ensure binary compatibility for old executables
under multibyte locale, versioned setlocale is added.
- __mb_len_cur definision is added in setlocale.c
and enable it in stdlib.h .
It is also important for multibyte locale stuffs,
but I just forgot.
"off_t" and the return value is a "paddr_t" to allow mappings
at offsets past 2^31 bytes. Somewhat inspired by FreeBSD, which
only changed the offset to a "vm_offset_t".
Includes updates for the i386, pc532 and sh3 mmmmap from Jason Thorpe.
"return (v0);" where v0 is an unsigned long and the function return
type is int. Stops lint warnings when this file is included.
Fix from Jason Thorpe.
macro in the MULTIPROCESSOR case (hardclock() wants it).
- Implement __GENERIC_SOFT_INTERRUPTS, and redefine the legacy
software interrupts in terms of it. Garbage-collect setsoftserial().
marked SONPROC.
- Always make curproc, fpcurproc, astpending, and want_resched per-CPU
variables in struct cpu_info. Restructure code accordingly, and trim
a few instructions from a few spots in various places in locore.
state into global and per-CPU scheduler state:
- Global state: sched_qs (run queues), sched_whichqs (bitmap
of non-empty run queues), sched_slpque (sleep queues).
NOTE: These may collectively move into a struct schedstate
at some point in the future.
- Per-CPU state, struct schedstate_percpu: spc_runtime
(time process on this CPU started running), spc_flags
(replaces struct proc's p_schedflags), and
spc_curpriority (usrpri of processes on this CPU).
- Every platform must now supply a struct cpu_info and
a curcpu() macro. Simplify existing cpu_info declarations
where appropriate.
- All references to per-CPU scheduler state now made through
curcpu(). NOTE: this will likely be adjusted in the future
after further changes to struct proc are made.
Tested on i386 and Alpha. Changes are mostly mechanical, but apologies
in advance if it doesn't compile on a particular platform.
contains the values __SIMPLELOCK_LOCKED and __SIMPLELOCK_UNLOCKED, which
replace the old SIMPLELOCK_LOCKED and SIMPLELOCK_UNLOCKED. These files
are also required to supply inline functions __cpu_simple_lock(),
__cpu_simple_lock_try(), and __cpu_simple_unlock() if locking is to be
supported on that platform (i.e. if MULTIPROCESSOR is defined in the
_KERNEL case). Change these functions to take an int * (&alp->lock_data)
rather than the struct simplelock * itself.
These changes make it possible for userland to use the locking primitives
by including <machine/lock.h>.
an extern for 'struct platform platform' further down this file anyway,
and this would have declared a 'platform' variable in any userland code
including this file.
- Add a bitmask for the CPUs which need an isync before this pmap returns
to userspace on that CPU.
- Define PMAP_USERRET(), a utility macro for userret() to use to process
the deferred isync, and call it as appropriate in userret().
- Add sysarch methods for "get bus window count", "get bus window",
and "pci conf read/write".
These are a hack, but they're what's necessary in order to make
XFree86 work in its current state.
processors. Instead, allocate separate idle PCBs for them (including
the primary -- don't use proc0's for its idle context).
- Use SysValue to store the cpu_info for each processor.
- alpha_rpcc(), alpha_mb(), alpha_wmb() -- these are instructions, and
we win by inlining them: rpcc is generally used for profiling, and
the memory barriers really should execute as quickly as possible with
minimal side-effects (like additional loads/stores required to call the
functions!)
- alpha_pal_imb(), alpha_pal_rdps(), alpha_pal_swpipl(), alpha_pal_tbi(),
alpha_pal_whami() -- these are PALcode ops. We must specify some register
clobbers for these.
We have a very decent size savings as a result. My test system:
text data bss dec hex filename
2671724 235848 377016 3284588 321e6c /netbsd.bak
2617708 235736 377016 3230460 314afc /netbsd
Most of this comes from fewer register saves/restores around spl*() calls
(now that alpha_pal_rdps() and alpha_pal_swpipl() are inlined).
Note that alpha_pal_rdps() and alpha_pal_swpipl() remain in pal.s to
maintain binary compatibility with LKMs that may use spl*() functions.
add an IPI which causes the target CPU to perform AST processing when
it returns to userspace.
- Add a way to get/set a private pointer in the shared interrupt header.
longer need to lock the kernel pmap in pmap_kenter_pa() and pmap_kremove().
- Since locking the kernel pmap in interrupt context is no longer required,
don't go to splimp() when locking the kernel pmap.
- Implement a new pmap_remove() function, not yet enabled by default. It
is structured like pmap_protect, and should be *much* faster. This was
actually written quite some time ago, but never committed because it
didn't work properly. Given the recent bugfix to pmap_protect(), "duh,
of course it didn't work properly before...". It seems to work fine now
(have done several builds and run the UVM regression tests using the new
code), but it is currently run-time optional so that some performance
measurements can be easily run against the old and new code.
big-endian. i386, pc532 and vax still include <machine/byte_swap.h>
and define macros for the {n,h}to{h,n}*() functions. mips also
defines some endian-independent assembly-code aliases for unaligned
memory accesses.
that is priority is rasied. Add a new spllowersoftclock() to provide the
atomic drop-to-softclock semantics that the old splsoftclock() provided,
and update calls accordingly.
This fixes a problem with using the "rnd" pseudo-device from within
interrupt context to extract random data (e.g. from within the softnet
interrupt) where doing so would incorrectly unblock interrupts (causing
all sorts of lossage).
XXX 4 platforms do not have priority-raising capability: newsmips, sparc,
XXX sparc64, and VAX. This platforms still have this bug until their
XXX spl*() functions are fixed.
context, so we must block interrupts which may cause memory allocation
before asserting the kernel pmap's lock. Put this all in PMAP_LOCK()
and PMAP_UNLOCK() macros to make it easier.
* Implement fpgetsticky() for alpha.
* Direct fpsetsticky() and fp{get,set}mask() into alpha kernel via sysarch(2).
* Define new sysarch(2) stub for above and install and distribute sysarch.h
for alpha. (The fpcr IS user mode r/w, but for reasons beyond the scope
of a commit message kernel calls are needed.) And much kernel Magick is
required before these do anything, but this way programs compiled under
1.4 will DTRT on future snapshots and releases.
allocated from a pool, and the MIPS and Alpha use KSEG to map pool
pages. So, mb_map wasn't actually being used. Saves around 4MB of
kernel virtual address space in a typical configuration.
Garbage-collect the related VM_MBUF_SIZE constant.
thorpej