very early (SI_SUB_TUNABLES - 1) and is responsible for setting mp_maxid.
cpu_mp_probe() is now called at SI_SUB_CPU and determines if SMP is
actually present and sets mp_ncpus and all_cpus. Splitting these up
allows an architecture to probe CPUs later than SI_SUB_TUNABLES by just
setting mp_maxid to MAXCPU in cpu_mp_setmaxid(). This could allow the
CPU probing code to live in a module, for example, since modules
sysinit's in modules cannot be invoked prior to SI_SUB_KLD. This is
needed to re-enable the ACPI module on i386.
- For the alpha SMP probing code, use LOCATE_PCS() instead of duplicating
its contents in a few places. Also, add a smp_cpu_enabled() function
to avoid duplicating some code. There is room for further code
reduction later since much of this code is also present in cpu_mp_start().
- All archs besides i386 still set mp_maxid to the same values they set it
to before this change. i386 now sets mp_maxid to MAXCPU.
Tested on: alpha, amd64, i386, ia64, sparc64
Approved by: re (scottl)
delete of objects. Also revert our temporary workaround in dsmthdat.c
that always copied objects. This is the correct fix for errors
evaluating _BST (and GBST) on IBM Thinkpads where an argument (Arg3)
was returned to the caller and the object was freed while still in use.
This will be in a future ACPI-CA dist.
Thanks to: kochi@netbsd.org, shaohua.li@intel.com
fixes an interrupt storm for certain users. This is done on the vendor
branch since the code is already in the 20031029 ACPI-CA dist and will
be imported after 5.2R.
Tested by: sebastian ssmoller <sebastian.ssmoller@gmx.net>
PR: i386/57909
Approved by: re (jhb)
different kernel to boot with kernel="NAME" would load the kernel and
loader.conf-selected modules from /boot/NAME, but it would not change
module_path. So, for instance, the automatically loaded acpi.ko would come
from /boot/kernel/acpi.ko, *always*.
Mind you, this happened for unassisted boot. If you interrupted, typed
"unload" and then "boot NAME", it would Do The Right Thing.
The source of the problem is the double initialization with beastie's
loader.rc. One would happen inside "start", and would load the kernel. The
next one would happen later in the loader.rc script, resetting module_path.
Because module_path is set to the Right Value by the functions in support.4th
that actually load the kernel, when beastie.4th proceeded to boot
module_path would remain wrong, as the kernel was already loaded.
This can be corrected by removing either initialization, and also by changing
the command used by beastie.4th from "boot" to "boot-conf", which makes sure
you use the right kernel and modules.
I chose to remove the second initialization, since this let you interrupt
(or confirm) boot before beastie even comes up. I avoid also doing the
boot-conf change because that would simply cause the kernel and modules to
be loaded twice (in fact, that was my original patch, until, in writing this
very commit message, I saw the error of my ways).
This commit changes the semantics of module loading when using the beastie
menu. Now it does what one would expect it to, but not what it was actually
doing, so something may break for unusual setups depending on broken
behavior. As our japanese friends so nicely put it, shikata ga nakatta. :-)
Approved by: re (scottl)
known samples of broken chipsets that needed mixed mode in the first place
are so broken (ie: locks up) that we can't use IO APIC mode at all and it
needs to be turned off in the bios. So, the MIXED_MODE penalty on the
good chipsets gained nothing.
Approved by: re (scottl)
the compiler having to parse and optimize the PCPU_GET(curthread) so often.
__curthread() is an inline optimized version of PCPU_GET(curthread) that
knows that pc_curthread is at offset zero in the pcpu struct. Add a
CTASSERT() to catch any possible changes to this. This accounts for
just over a 1% wall clock speedup for total kernel compile/link time,
and 20% compile time speedup on some specific files depending on which
compile options are used.
Approved by: re (jhb)
the routing table. Move all usage and references in the tcp stack
from the routing table metrics to the tcp hostcache.
It caches measured parameters of past tcp sessions to provide better
initial start values for following connections from or to the same
source or destination. Depending on the network parameters to/from
the remote host this can lead to significant speedups for new tcp
connections after the first one because they inherit and shortcut
the learning curve.
tcp_hostcache is designed for multiple concurrent access in SMP
environments with high contention and is hash indexed by remote
ip address.
It removes significant locking requirements from the tcp stack with
regard to the routing table.
Reviewed by: sam (mentor), bms
Reviewed by: -net, -current, core@kame.net (IPv6 parts)
Approved by: re (scottl)
boot-disabled devices instead of skipping the last interrupt. This is
especially important for devices that only have one interrupt as this
bug was keeping any interrupt from being tried at all.
Reviewed by: msmith
Approved by: re (scottl)
the routing table. Move all usage and references in the tcp stack
from the routing table metrics to the tcp hostcache.
It caches measured parameters of past tcp sessions to provide better
initial start values for following connections from or to the same
source or destination. Depending on the network parameters to/from
the remote host this can lead to significant speedups for new tcp
connections after the first one because they inherit and shortcut
the learning curve.
tcp_hostcache is designed for multiple concurrent access in SMP
environments with high contention and is hash indexed by remote
ip address.
It removes significant locking requirements from the tcp stack with
regard to the routing table.
Reviewed by: sam (mentor), bms
Reviewed by: -net, -current, core@kame.net (IPv6 parts)
Approved by: re (scottl)
accordingly. The define is left intact for ABI compatibility
with userland.
This is a pre-step for the introduction of tcp_hostcache. The
network stack remains fully useable with this change.
Reviewed by: sam (mentor), bms
Reviewed by: -net, -current, core@kame.net (IPv6 parts)
Approved by: re (scottl)
on SMP systems has a chance of working. This was a loose end of the
implementation of the ACPI Cx idle states. Since our logical CPU Id
is the ACPI processor Id, we do not need to jump through hoops to
obtain it.
Approved: re@ (jhb)
happen in interrupt context; 1) sleep locks, and 2) malloc/free
calls.
1) is fixed by using spin locks instead.
2) is fixed by preallocating a FIFO (implemented with a STAILQ)
and using elements from this FIFO instead. This turns out
to be rather fast.
OK'ed by: re (scottl)
Thanks to: peter, jhb, rwatson, jake
Apologies to: *
an acpi_cpu method for shutdown that disables entry to acpi_cpu_idle
and then IPIs/waits for threads to exit. This fixes a panic late in
reboot in the SMP case.
* In the !SMP case, don't use the processor id filled out by the MADT
since there can only be one processor. This was causing a panic in
acpi_cpu_idle if the id was 1 since the data was being dereferenced from
cpu_softc[1] even though the actual data was in cpu_softc[0] (which is
correct).
* Rework the initialization functions so that cpu_idle_hook is written
late in the boot process.
* Make the P_BLK, P_BLK_LEN, and cpu_cx_count all softc-local variables.
This will help SMP boxes that have _CST or multiple P_BLKs. No such
boxes are known at this time.
* Always allocate the C1 state, even if the P_BLK is invalid. This means
we will always take over idling if enabled. Remove the value -1 as
valid for cx_lowest since this is redundant with machdep.cpu_idle_hlt.
* Reduce locking for the throttle initialization case to around the write
to the smi_cmd port. Add disabled code to write the CST_CNT. It will
be enabled once _CST re-evaluation is tested (post 5.2R).
Thank you: dfr, imp, jhb, marcel, peter
Tested by: rwatson, Harald Schmalzbauer <h@schmalzbauer.de>
Approved by: re (rwatson)
occurs when kmem_malloc() fails to allocate a sufficient number of vm
pages. Specifically, we avoid the lock-order reversal by not grabbing
Giant around pmap_remove() if the map is the kmem_map.
Approved by: re (jhb)
Reported by: Eugene <eugene3@web.de>
- turn on SMP in generic
- add 'device atpic' - this is unconditional on i386, but certain nvidia
based systems need to disable acpi because the reference bios seems to be
hosed. If acpi is disabled, we won't find the apic. amd64 has the
mptable code in a seperate compile option as well.
- turn sym back on, it doesn't fail to compile anymore.
Approved by: re
source count pointers at them so that intr_execute_handlers() won't
choke when it tries to handle an unregisterd ATPIC interrupt source.
- Install the low-level ATPIC interrupt handlers when we first program the
ATPIC in atpic_startup() rather than at SI_SUB_INTR. This is only
necessary to work around buggy code that enables interrupts too early
in the boot process (namely, the vm86 code).
Approved by: re (rwatson)
regocnized as such at the time. Now that the other bogons in the
tree have been fixed, we can remove this ugly kludge.
o Remove stale/bogus opt_foo.h files. These are left over from
by-gone resources. And they point to the need, yet again, to
improve the build system so meta information is only in one place.
Submitted by: ru
Reviewed by: bde
Approved by: re@ (jhb)
purpose and the resulting vattr structure was ignored. In addition,
the VOP_GETATTR call was made with no vnode lock held, resulting in
vnode locking violation panic with debug kernels.
Reported by: truckman
Approved by: re@ (rwatson)
In practice it seems that in situations of high packet loss the ACK
timeout seems to hit this maximum (perhaps inappropriately, but the
estimation algorithm is not perfect, so apparently it happens). In
any case, 10 seconds is way too high a value so lower to 1 second.
MFC after: 3 days
the "old" SYSINIT. This makes sure things happen in the right order.
XXX: md(4) needs to be fully geom-ified and in particluar /dev/md.ctl
should be abandonded for the GEOM OaM api.
Approved by: re@
rather than right before and right after. This allows these routines
to manipulate the mesh.
KASSERT that nobody creates a geom on an alien class.
Assert topology in g_valid_obj().
Approved by: re@
TOC's for the same media!! that borks up GEOM.
Although this looks like bad HW the following patch removes the
chance for GEOM panic'ing.
Approved by: re@
the MAC label referenced from 'struct socket' in the IPv4 and
IPv6-based protocols. This permits MAC labels to be checked during
network delivery operations without dereferencing inp->inp_socket
to get to so->so_label, which will eventually avoid our having to
grab the socket lock during delivery at the network layer.
This change introduces 'struct inpcb' as a labeled object to the
MAC Framework, along with the normal circus of entry points:
initialization, creation from socket, destruction, as well as a
delivery access control check.
For most policies, the inpcb label will simply be a cache of the
socket label, so a new protocol switch method is introduced,
pr_sosetlabel() to notify protocols that the socket layer label
has been updated so that the cache can be updated while holding
appropriate locks. Most protocols implement this using
pru_sosetlabel_null(), but IPv4/IPv6 protocols using inpcbs use
the the worker function in_pcbsosetlabel(), which calls into the
MAC Framework to perform a cache update.
Biba, LOMAC, and MLS implement these entry points, as do the stub
policy, and test policy.
Reviewed by: sam, bms
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
o Each source gets its own queue, which is a FIFO, not a ring buffer.
The FIFOs are implemented with the sys/queue.h macros. The separation
is so that a low entropy/high rate source can't swamp the harvester
with low-grade entropy and destroy the reseeds.
o Each FIFO is limited to 256 (set as a macro, so adjustable) events
queueable. Full FIFOs are ignored by the harvester. This is to
prevent memory wastage, and helps to keep the kernel thread CPU
usage within reasonable limits.
o There is no need to break up the event harvesting into ${burst}
sized chunks, so retire that feature.
o Break the device away from its roots with the memory device, and
allow it to get its major number automagically.
to see_other_uids but with the logical conversion. This is based
on (but not identical to) the patch submitted by Samy Al Bahra.
Submitted by: Samy Al Bahra <samy@kerneled.com>
more than one sf_buf for one vm_page. To accomplish this, we add
a global hash table mapping vm_pages to sf_bufs and a reference
count to each sf_buf. (This is similar to the patches for RELENG_4
at http://www.cs.princeton.edu/~yruan/debox/.)
For the uninitiated, an sf_buf is nothing more than a kernel virtual
address that is used for temporary virtual-to-physical mappings by
sendfile(2) and zero-copy sockets. As such, there is no reason for
one vm_page to have several sf_bufs mapping it. In fact, using more
than one sf_buf for a single vm_page increases the likelihood that
sendfile(2) blocks, hurting throughput.
(See http://www.cs.princeton.edu/~yruan/debox/.)
- This is heavily derived from John Baldwin's apic/pci cleanup on i386.
- I have completely rewritten or drastically cleaned up some other parts.
(in particular, bootstrap)
- This is still a WIP. It seems that there are some highly bogus bioses
on nVidia nForce3-150 boards. I can't stress how broken these boards
are. I have a workaround in mind, but right now the Asus SK8N is broken.
The Gigabyte K8NPro (nVidia based) is also mind-numbingly hosed.
- Most of my testing has been with SCHED_ULE. SCHED_4BSD works.
- the apic and acpi components are 'standard'.
- If you have an nVidia nForce3-150 board, you are stuck with 'device
atpic' in addition, because they somehow managed to forget to connect the
8254 timer to the apic, even though its in the same silicon! ARGH!
This directly violates the ACPI spec.
with multiple ports on a shared interrupt demultiplexed by the puc_intr()
handler.
siointr1() first read as much input as possible and then checked all
possibly-relevant status registers, partly for robustness and partly
for historical reasons. This is very bad if it is called for every
port sharing an interrupt like puc_intr() does. It can spend too long
reading all the input for some ports when the interrupt is for a more
urgent event on another, or just too long checking all the status
registers when there are lots of ports. The inter-character time is
too long for reading all the input even when the interrupt is for a
transmitter interrupt on the same port, and at 921600 bps the inter-char
time is 10.85 usec and was often exceeded with just 2 ports, leaving
the transmitters idle for about 6% of the time.
The tweak is to break out of the read loop after reading 1 char if
output can be done. This avoids most of the idle transmitter time for
2 active ports at 921600 bps bidirectional on the test system. It
also reduces overhead by about 20%. More complete fixes use the
programmable tx low watermark on 16950's and reduce overhead by another
65%.
do not have mh_nextpkt initialized. Somtimes what's there is "1", and the
ip_input() code pukes trying to m_free() it, rendering divert sockets and
such broken.
This really underscores the need to get rid of MT_TAG.
Reviewed by: rwatson
is the warning that points to the bug in `(char *)malloc(...)' where
malloc() is implicitly declared as returning int. We do similar things
here, but they work because u_int is the same as uintptr_t on i386's.)
system calls, and prefer these calls over getsockopt()/setsockopt()
for ABI reasons. When addressing UNIX domain sockets, these calls
retrieve and modify the socket label, not the label of the
rendezvous vnode.
- Create mac_copy_socket_label() entry point based on
mac_copy_pipe_label() entry point, intended to copy the socket
label into temporary storage that doesn't require a socket lock
to be held (currently Giant).
- Implement mac_copy_socket_label() for various policies.
- Expose socket label allocation, free, internalize, externalize
entry points as non-static from mac_net.c.
- Use mac_socket_label_set() in __mac_set_fd().
MAC-aware applications may now use mac_get_fd(), mac_set_fd(), and
mac_get_peer() to retrieve and set various socket labels without
directly invoking the getsockopt() interface.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
will now need editing except for spot checks.
Changed this buffer from a circular one to a linear one. This is more
useful for some cases and the sysctl that prints it doesn't support
circular buffers.
Fixed (output) formatting bugs in this sysctl. An off by 1 error caused
a garbage byte to be returned after annotation of large deltas, and
a race with the writer sometimes caused premature string termination.
o when compiling lint, undefine certain things and redefine them so that the
driver doesn't #error out. Since lint kernels aren't supposed to be
bootable, I'm no troubled by this breakage.
This fixes the tinderbox
Suggested by: rwatson
Approved by: bms
SO_PEERLABEL. This provides an interface to query the label of a
socket peer without embedding implementation details of mac_t in
the application. Previously, sizeof(*mac_t) had to be specified
by an application when performing getsockopt().
Document mac_get_peer(3), and expand documentation of the other
mac_get(3) functions. Note that it's possible to get EINVAL back
from mac_get_fd(3) when pointing it at an inappropriate object.
NOTE: mac_get_fd() and mac_set_fd() support for sockets will
follow shortly, so the documentation is slightly ahead of the
code.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
mac_setsockopt_label() into mac_socket_label_set(); make it non-static
so that it can be invoked from kern_mac.c for mac_set_fd().
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
While we end up the same place, we end up with two different CS register
values after the jump and 0xf000 is compatible with the hardware reset
value.
This makes a difference if the BIOS does a near jump before a far jump.
Detective work and patch by: Adrian Steinmann <ast@marabu.ch>
- improve sysinfo(2) syscall;
- add dummy fadvise64(2) syscall;
- add dummy *xattr(2) family of syscalls;
- add protos for the syscalls 222-225, 238-249 and 253-267;
- add exit_group(2) syscall, which is currently just wired to exit(2).
Obtained from: OpenBSD
MFC after: 2 weeks
Its restoration in rev.1.102 was mistranslated to the equivalent of
setsofttty() in rev.1.105. This increased overheads by causing a
context switch to the SWI handler after almost every interrupt. The
increase was approx. 50% on a Celeron 366 (from 23 usec to 34 usec
per interrupt).
I'm having bad luck with different parts of the sys tree being checked
out at slightly different times. Back it out, noting it doesn't cause
harm in any case. Tinderbox also makes these things more fun.
of newfs, to signify the newfs operation has not yet completed. Re-
write the superblock with the correct magic number once all of the
cylinder groups have been created to show the operation has finished.
Sponsored by: St. Bernard Software
physical mapping.
- Move the sf_buf API to its own header file; make struct sf_buf's
definition machine dependent. In this commit, we remove an
unnecessary field from struct sf_buf on the alpha, amd64, and ia64.
Ultimately, we may eliminate struct sf_buf on those architecures
except as an opaque pointer that references a vm page.
