- In case of ailure in the child, kill the child and wakeup the parent
- Do call child_return
- The src and dst args were mixed up when copying regs 0-31 to the trapframe
This makes pthread_create(3) fully working (test program works).
mach_port_move_member, mach_port_set_attributes, mach_task_set_special_port,
(none do anything)
Added mach_thread_create_running, which creates a new Mach thread. It
provides the register context of the new thread. We use it in a child
function provided to fork1(). The child function is machine dependent and
is not yet implemented for i386.
The new thread crashes quickly, but at least it starts.
the passed device number to only contain lower 16 bits; older glibc
appears to pass some junk in upper bits sometimes, resulting in incorrect
device entries being created
Problem initially analyzed by Thor Lancelot Simon.
and make the sleep length depend on value of variable forkfsleep;
it's set to zero by default (no sleep)
this is a preparation for making the sleep length settable via sysctl
These are of use to userland code which previously depended on the
hard-coded values of LABELSECTOR and LABELOFFSET to figure out the
location of the disklabel for a particular platform.
With the introduction of umbrella ports such as evbarm, evbmips, etc,
the location of the disklabel may vary between kernels for the same
MACHINE. This sysctl will allow userland programs to remain independent
of the particular flavour of MACHINE in such cases.
to sleep. Define UVM_KMF_NOWAIT in terms of UVM_FLAG_NOWAIT.
From Manuel Bouyer. Fixes a problem where any mapping with
read protection was created in a "nowait" context, causing
spurious failures.
The driver puts the adapter in promisc mode to receive multicast addresses.
At last set the IFF_PROMISC flag so that the upper layer filters frames
that are not for us.
Sure, the real fix would be to get multicast filters working ...
must be shared. This requires drivers that request the installation of
a `fast trap' handler to supply a regular interrupt handler as well to fall
back on.
The `fast trap' routine (if present) is now passed as an additional
argument to intr_establish().
least 1us. Documentation I've found for the simple (SPP) parallel port
mode says that data should be stable 500ns before STROBE, STROBE should
be pulsed for no less than 500ns, and that data should be stable another
500ns after STROBE has been de-asserted.
Makes lpt@ebus on my Sun Ultra5 work with my HP DeskJet 712C, at least in
polled mode. Thanks to Martin for astutely noting it was probably a bug
with STROBE being pulsed too quickly.
1. Reduce debugging level to sane levels
2. Fix bugs in alloc_data_map related to allocing whole bytes of bitmap at
a time.
At this point the driver is functional. It talks to a local drive here and
can label/newfs. Performance is...lacking at the moment as its chewing cpu
heavily (probably due to the number of memcpy's) and will be the next area
attacked.
To make this work, we now have to use separate handler lists for hardware
and software interrupts as the soft interrupt handlers do not return
an `interrupt handled' status.
Thanks to Matt Fredette for providing an initial set of patches on port-sparc.
the same method to access TOD clock, so move chipset-dependent mcclock
routines to appropriate files.
This should also fix Express RISCserver and RISCserver2200 (aka R96) support
which has been broken since "struct platform" was introduced.
The first board to be included here is the port to the 405GP-based
Walnut evaluation board, which up until now lived in arch/walnut.
arch/walnut will go away soon, once all the remaining walnut-isms
in the tree have been dealt with.
support up to the max ohci descriptor segments. Then attempt to fill it in
via a load. Use the number of segments filled in as the basis for figuring
out how many descriptors to supply to the ohci DMA engine.
XXX: Still needs to account for requests which because of splits may overflow
the 6 dma descriptors available today. For now this fixes cases where a
single 512 byte write was getting split into 2 dma segments from 1 incoming
iov request
two ways:
- the child gets its pid as retval[0] (userland stub will turn it into a 0),
retval[1] is 1 and it is 0 in the parent.
- in the child, the fork syscall is successful, hence we must skip the next
instruction.
- The pcibus is now called 'pcib' and the isabus 'isab'. Their attributes are
'pcibus' and 'isabus' respectively. This makes the underlying busses
attach again.
- Initialize the parent structure in the config_console() function.
Otherwise we end up without a console.
given that PROM maps just 4 or 16 this is not going to be a bottle
neck). Doesn't really affect normal kernels, need it for the changed
kernel base address (uncommitted) hack for broken javastation OFW.
Ok by pk.
The definitions were not the same between the scsi_messages file and this
definition so simply removing it here and letting the other one be used
results in incorrect behavior (regardless of whether it made the code
compile....)
compat/common, so that they can be shared by several emulations, and use
them for Darwin.
This removes the ugly dependance on FreeBSD freebsd_file.c for COMPAT_DARWIN
used to get and set the thread user value, which is an opaque pointer to
a per thread structure stored in userland. cthread_self() is used by Darwin
as an implementation for pthread_self(), which return the thread id.
We use the p_emuldata field of struct proc in order to keep track of the
thread user value. For now the value is per-process, but we will make it
per-thread when we will take care of threading.
While we are there, do some KNF
- J96A (Express5800/240 R4400 EISA) requires different method to access
todclock from other NEC machines (it's similar with magnum),
so handle it in p_nec_j96a.c.
Now my NEC Express5800/240 works. (and now I can test 53c700 SCSI.)
- Call usbd_transfer_complete at splusb.
- Fix a botched for loop in ohci_rem_ed.
- In ohci_close_pipe, wait 1ms after removing an ED to avoid possible race
condition.
raise the ipl in the interrupt handlers to the appropriate level. This avoids
interrupt handler interference if one of the devices actually interrupts at
a lower hardware level than the maximum level assined to a device class.
Based on code from Art Grabowski in openbsd.
and use home-grown buffer structure.
Handle display orientation (normal or upside-down) appropriately
by making use of devinfo of base device.
If the LCD is at an expansion slot of a base unit
whose di_connector_direction == MAPLE_CONN_TOP,
the driver automatically rotates the bitmap.
You need not rotate the bitmap before passing to the driver (spec change).
under some circumstances, leave turds in the icache following vmspace
teardown.
It's not yet clear if this is a pmap bug or a toolchain problem since
the hack is unecessary when the kernel is compiled with -O0. Of course
that could just be masking the problem due to increased icache pressure...
the standard flags. See also PR#19163.
Before:
cpu0: AMD Athlon XP 1800+ (686-class), 1532.11 MHz
cpu0: features 383f9ff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,SEP,MTRR>
cpu0: features 383f9ff<PGE,MCA,CMOV,FGPAT,PSE36,MMX>
cpu0: features 383f9ff<FXSR,SSE>
After:
cpu0: AMD Athlon XP 1800+ (686-class), 1532.11 MHz
cpu0: features c3cbf9ff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,SEP,MTRR>
cpu0: features c3cbf9ff<PGE,MCA,CMOV,PAT,PSE36,MPC,MMXX,MMX>
cpu0: features c3cbf9ff<FXSR,SSE,3DNOW2,3DNOW>
While I'm here, amd_cpuid_cpu_cacheinfo() is an info function rather
than a probe function.
and seems like generally sensible (more sensible than not doing so), so done
in generic code rather than compat glue only
Change proposed in PR kern/18767 by Emmanuel Dreyfus.
- leave 5 processes for root-only use, the previous value of 1
was unsufficient to execute additional commands once logged, and
perhaps also not enough to actually login remotely with recent (open)sshd
- protect the log of "proc: table full" with ratecheck(), so that
the message is only logged once per 10 seconds; though syslogd normally
doesn't pass the repeated messages through, this avoids flooding
syslogd and potentially also screen/logs
- If the process hits either system limit of number of processes in system,
or user's limit of same, force the process to sleep for 0.5 seconds
before returning failure. This turns 2000 rampaging fork monsters into
2000 harmlessly snoozing fork monsters.
The sleep is intentionally uninterruptible by signals.
These are not intended as ultimate protection agains fork-bombs.
Determined attacker can eat CPU differently than via repeating
fork() calls. But this is good enough to help protect against
programming mistakes or simple-minded tests.
Based on FreeBSD kern_fork.c change in revision 1.132 by
Mike Silbersack <silby at FreeBSD org>
Change also discussed on tech-kern@NetBSD.org, thread
'Fork bomb protection patch'.
to read sector 18, and fallback to 1.44MB drive geometry if that fails.
This allows to boot from 1.44MB floppy disk in 2.88MB drive.
Tested with 2.88MB drive in IBM PS/2 model 95 donated
by 'Yokotashi' <lhc at kanal ucw cz> and Pavel Cahyna
<pavel.cahyna at st ms mff cuni cz>
Bump biosboot version.
Fixes PR kern/3418 by Keith Moore.
Change okayed by Frank van den Linden.
