]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
13 years agoShift responsibility for synchronizing access to the page's act_count
Alan Cox [Sun, 16 Jan 2011 18:01:39 +0000 (18:01 +0000)]
Shift responsibility for synchronizing access to the page's act_count
field to the object's lock.

Reviewed by: kib@

13 years agoClean up the start of vm_page_alloc(). In particular, eliminate an
Alan Cox [Sun, 16 Jan 2011 17:33:34 +0000 (17:33 +0000)]
Clean up the start of vm_page_alloc().  In particular, eliminate an
assertion that is no longer required.  Long ago, calls to vm_page_alloc()
from an interrupt handler had to specify VM_ALLOC_INTERRUPT so that
vm_page_alloc() would not attempt to reclaim a PQ_CACHE page from another vm
object.  Today, with the synchronization on a vm object's collection of
PQ_CACHE pages, this is no longer an issue.  In fact, VM_ALLOC_INTERRUPT now
reclaims PQ_CACHE pages just like VM_ALLOC_{NORMAL,SYSTEM}.

MFC after: 3 weeks

13 years agoAdd a missing word.
Marius Strobl [Sun, 16 Jan 2011 14:21:26 +0000 (14:21 +0000)]
Add a missing word.

Submitted by: gj

13 years agosh(1): Document changes to 'exit' from traps.
Jilles Tjoelker [Sun, 16 Jan 2011 14:11:50 +0000 (14:11 +0000)]
sh(1): Document changes to 'exit' from traps.

13 years agosh: If exit is used without args from a trap action, exit on the signal.
Jilles Tjoelker [Sun, 16 Jan 2011 13:56:41 +0000 (13:56 +0000)]
sh: If exit is used without args from a trap action, exit on the signal.

This is useful so that it is easier to exit on a signal than to reset the
trap to default and resend the signal. It matches ksh93. POSIX says that
'exit' without args from a trap action uses the exit status from the last
command before the trap, which is different from 'exit $?' and matches this
if the previous command is assumed to have exited on the signal.

If the signal is SIGSTOP, SIGTSTP, SIGTTIN or SIGTTOU, or if the default
action for the signal is to ignore it, a normal _exit(2) is done with exit
status 128+signal_number.

13 years agoAdd support for resource pooling to CMT.
Michael Tuexen [Sun, 16 Jan 2011 10:02:46 +0000 (10:02 +0000)]
Add support for resource pooling to CMT.
An original version of the patch was developed by Martin Becke
and Thomas Dreibholz.

MFC after: 3 months

13 years agoReference rgephy.4 in man pages of additional MAC drivers also known to
Marius Strobl [Sun, 16 Jan 2011 04:14:56 +0000 (04:14 +0000)]
Reference rgephy.4 in man pages of additional MAC drivers also known to
come in combination with these PHYs.

Submitted by: yongari
MFC after: 3 days

13 years agoAdd a manual page for rgephy(4) and reference it as appropriate. The
Marius Strobl [Sat, 15 Jan 2011 22:07:08 +0000 (22:07 +0000)]
Add a manual page for rgephy(4) and reference it as appropriate. The
motivation for having rgephy.4 is to document the special media option
add in r217415.

MFC after: 3 days

13 years agoFor consistency, use kernel_object instead of &kernel_object_store
Konstantin Belousov [Sat, 15 Jan 2011 21:56:38 +0000 (21:56 +0000)]
For consistency, use kernel_object instead of &kernel_object_store
when initializing the object mutex. Do the same for kmem_object.

Discussed with: alc
MFC after: 1 week

13 years agosh: Fix some things about -- in trap:
Jilles Tjoelker [Sat, 15 Jan 2011 21:09:00 +0000 (21:09 +0000)]
sh: Fix some things about -- in trap:
* Make 'trap --' do the same as 'trap' instead of nothing.
* Make '--' stop option processing (note that '-' action is not an option).

Side effect: The error message for an unknown option is different.

13 years agoDon't redefine MODINFOMD_BOOTINFO as MODINFOMD_DTBP. This
Marcel Moolenaar [Sat, 15 Jan 2011 20:18:08 +0000 (20:18 +0000)]
Don't redefine MODINFOMD_BOOTINFO as MODINFOMD_DTBP. This
breaks support for older loaders. Add MODINFOMD_DTBP as
a new tag instead.

13 years agoFor some time now, the kernel and kmem objects have been ordinary
Alan Cox [Sat, 15 Jan 2011 19:21:28 +0000 (19:21 +0000)]
For some time now, the kernel and kmem objects have been ordinary
OBJT_PHYS objects.  Thus, there is no need for handling them specially
in vm_fault().  In fact, this special case handling would have led to
an assertion failure just before the call to pmap_enter().

Reviewed by: kib@
MFC after: 6 weeks

13 years agoRemove unused variable. Spotted by a cppcheck
Andreas Tobler [Sat, 15 Jan 2011 19:16:56 +0000 (19:16 +0000)]
Remove unused variable. Spotted by a cppcheck
(devel/cppcheck, http://sourceforge.net/projects/cppcheck) run.

Approved by: nwhitehorn (mentor)

13 years agoRemove unused variables. Spotted by a cppcheck
Andreas Tobler [Sat, 15 Jan 2011 19:16:05 +0000 (19:16 +0000)]
Remove unused variables. Spotted by a cppcheck
(devel/cppcheck, http://sourceforge.net/projects/cppcheck) run.

Approved by: nwhitehorn (mentor)

13 years agoSome old WD SATA disks report supported and enabled device-initiated
Alexander Motin [Sat, 15 Jan 2011 09:43:25 +0000 (09:43 +0000)]
Some old WD SATA disks report supported and enabled device-initiated
interface power management, but return ABORT error on attempt to disable
it. Make CAM SATA probe sequence ignore this error, as it is not fatal.

13 years agoModify the experimental NFSv4 server so that it posts a SIGUSR2
Rick Macklem [Fri, 14 Jan 2011 23:30:35 +0000 (23:30 +0000)]
Modify the experimental NFSv4 server so that it posts a SIGUSR2
signal to the master nfsd daemon whenever the stable restart
file has been modified. This will allow the master nfsd daemon
to maintain an up to date backup copy of the file. This is
enabled via the nfssvc() syscall, so that older nfsd daemons
will not be signaled.

Reviewed by: jhb
MFC after: 1 week

13 years agosh: Make 'trap -l' look like 'kill -l'.
Jilles Tjoelker [Fri, 14 Jan 2011 21:30:27 +0000 (21:30 +0000)]
sh: Make 'trap -l' look like 'kill -l'.

13 years agoRemove redundant, bogus, and even harmful uses of setting TS bit in CR0.
Jung-uk Kim [Fri, 14 Jan 2011 21:09:01 +0000 (21:09 +0000)]
Remove redundant, bogus, and even harmful uses of setting TS bit in CR0.
It is done from fpstate_drop() when it is really necessary.

Reviewed by: kib
MFC after: 1 week

13 years ago- Even after masking the media with IFM_GMASK the result may have bits
Marius Strobl [Fri, 14 Jan 2011 20:26:59 +0000 (20:26 +0000)]
- Even after masking the media with IFM_GMASK the result may have bits
  besides the duplex ones set so just comparing it with IFM_FDX may lead
  to false negatives.
- Just let the default case handle all unsupported media types.
- In pnphy_status() don't unnecessarily read a register twice.
- Remove unnused macros.

MFC after: 1 week

13 years agoAdd two more features to cpuset(1):
John Baldwin [Fri, 14 Jan 2011 19:57:28 +0000 (19:57 +0000)]
Add two more features to cpuset(1):
- Add a new -C flag to create a new cpuset and move an existing pid into
 that set.
- Allow 'all' to be specified for a cpu list (e.g. cpuset -s 1 -l all)
 which maps to the list of all CPUs in the system.

MFC after: 2 weeks

13 years ago- Allow IFM_FLAG0 to be set indicating that auto-negotiation with manual
Marius Strobl [Fri, 14 Jan 2011 19:39:12 +0000 (19:39 +0000)]
- Allow IFM_FLAG0 to be set indicating that auto-negotiation with manual
  configuration, which is used to work around issues with certain setups
  (see r161237) by default, should not be triggered as it may in turn
  cause harm in some edge cases.
- Even after masking the media with IFM_GMASK the result may have bits
  besides the duplex ones set so just comparing it with IFM_FDX may lead
  to false negatives.
- Announce PAUSE support also for manually selected 1000BASE-T, but for
  all manually selected media types only in full-duplex mode. Announce
  asymmetric PAUSE support only for manually selected 1000BASE-T.
- Simplify setting the manual configuration bits to only once after we
  have figured them all out. This also means we no longer unnecessarily
  update the hardware along the road.
- Remove a stale comment.

Reviewed by: yongari (plus additional testing)
MFC after: 3 days

13 years ago- Don't announce PAUSE support for half-duplex when MIIF_FORCEPAUSE is
Marius Strobl [Fri, 14 Jan 2011 19:33:58 +0000 (19:33 +0000)]
- Don't announce PAUSE support for half-duplex when MIIF_FORCEPAUSE is
  set.
- Let mii_phy_auto() also announce PAUSE support for 10baseT-FDX.

MFC after: 1 week

13 years ago- Even after masking the media with IFM_GMASK the result may have bits
Marius Strobl [Fri, 14 Jan 2011 19:29:53 +0000 (19:29 +0000)]
- Even after masking the media with IFM_GMASK the result may have bits
  besides the duplex ones set so just comparing it with IFM_FDX may lead
  to false negatives.
- Simplify ciphy_service() to only set the manual configuration bits
  once after we have figured them all out. This also means we no longer
  unnecessarily update the hardware along the road.

MFC after: 1 week

13 years ago- Masking IFM_GMASK when also masking IFM_FDX is redundant and just
Marius Strobl [Fri, 14 Jan 2011 19:16:59 +0000 (19:16 +0000)]
- Masking IFM_GMASK when also masking IFM_FDX is redundant and just
  complicates the code.
- Don't let atphy_setmedia() announce PAUSE support for half-duplex when
  MIIF_FORCEPAUSE is set.
- Simplify e1000phy_service() and ip1000phy_service() to only set the
  manual configuration bits once after we have figured them all out. For
  ip1000phy_service() this also means we no longer unnecessarily update
  the hardware along the road.

MFC after: 1 week

13 years agoRework realtime priority support:
John Baldwin [Fri, 14 Jan 2011 17:06:54 +0000 (17:06 +0000)]
Rework realtime priority support:
- Move the realtime priority range up above kernel sleep priorities and
  just below interrupt thread priorities.
- Contract the interrupt and kernel sleep priority ranges a bit so that
  the timesharing priority band can be increased.  The new timeshare range
  is now slightly larger than the old realtime + timeshare ranges.
- Change the ULE scheduler to no longer use realtime priorities for
  interactive threads.  Instead, the larger timeshare range is now split
  into separate subranges for interactive and non-interactive ("batch")
  threads.  The end result is that interactive threads and non-interactive
  threads still use the same priority ranges as before, but realtime
  threads now have a separate, dedicated priority range.
- Do not modify the priority of non-timeshare threads in sched_sleep()
  or via cv_broadcastpri().  Realtime and idle priority threads will
  no longer have their priorities affected by sleeping in the kernel.

Reviewed by: jeff

13 years agoEnable shared page for the signal trampolines on PowerPC.
Konstantin Belousov [Fri, 14 Jan 2011 11:36:44 +0000 (11:36 +0000)]
Enable shared page for the signal trampolines on PowerPC.

Reviewed and tested by: nwhitehorn

13 years agoMark csu on PowerPC as not requiring executable stack.
Konstantin Belousov [Fri, 14 Jan 2011 11:34:58 +0000 (11:34 +0000)]
Mark csu on PowerPC as not requiring executable stack.

Reviewed and tested by: nwhitehorn

13 years agoMark assembler sources from libc as working with non-executable stack.
Konstantin Belousov [Fri, 14 Jan 2011 11:33:40 +0000 (11:33 +0000)]
Mark assembler sources from libc as working with non-executable stack.

Reviewed and tested by: nwhitehorn

13 years agoMark rtld on PowerPC as not requiring executable stack.
Konstantin Belousov [Fri, 14 Jan 2011 11:32:31 +0000 (11:32 +0000)]
Mark rtld on PowerPC as not requiring executable stack.

Reviewed and tested by: nwhitehorn

13 years agoConfigure gcc to emit .note.GNU-stack for both 32 and 64 bits PowerPC.
Konstantin Belousov [Fri, 14 Jan 2011 11:31:22 +0000 (11:31 +0000)]
Configure gcc to emit .note.GNU-stack for both 32 and 64 bits PowerPC.
Mark gcc-provided asm files as not requiring executable stack.

It seems that non-FreeBSD ABIs for powerpc64 claim stack non-executable.
Due to this, rs6000_elf_end_indicate_exec_stack() only emit the note for
32 bit target. I decided not to change FreeBSD ABI and patch
emit the notes for both variants.

Reviewed and tested by: nwhitehorn

13 years agoOn PowerPC64, linker emits a fake object into each linked object,
Konstantin Belousov [Fri, 14 Jan 2011 11:28:34 +0000 (11:28 +0000)]
On PowerPC64, linker emits a fake object into each linked object,
'linker stubs'. Add .note.GNU-stack for the stubs objects. Without this,
final binary will have RWE mode for PT_GNU_STACK regardless of the
actual requirements.

Tested by: nwhitehorn
Reviewed by: dim, nwhitehorn

13 years agoAdd a workaround for the warnings and sometimes errors due to missed
Konstantin Belousov [Fri, 14 Jan 2011 11:25:11 +0000 (11:25 +0000)]
Add a workaround for the warnings and sometimes errors due to missed
forward-declaration.

Reviewed by: dim

13 years agoMark libcompiler_rt as not needed executable stack on powerpc.
Konstantin Belousov [Fri, 14 Jan 2011 11:23:39 +0000 (11:23 +0000)]
Mark libcompiler_rt as not needed executable stack on powerpc.

Reviewed and tested by: nwhitehorn

13 years agoFix making a release. The old libdialog was renamed to libodialog.
Yoshihiro Takahashi [Fri, 14 Jan 2011 10:56:44 +0000 (10:56 +0000)]
Fix making a release.  The old libdialog was renamed to libodialog.

13 years agoCorrect bogus initialization. It should be noted that this change
Christian S.J. Peron [Fri, 14 Jan 2011 04:24:53 +0000 (04:24 +0000)]
Correct bogus initialization.  It should be noted that this change
has been corrected in the vendor branch, but for now, silence clang
warnings.

Found by: clang
Discussed with: mlaier
MFC after: 1 week

13 years agoIf driver is not able to allocate RX buffer, do not start driver.
Pyun YongHyeon [Thu, 13 Jan 2011 23:15:09 +0000 (23:15 +0000)]
If driver is not able to allocate RX buffer, do not start driver.
While I'm here move RX buffer allocation and descriptor
initialization up to not touch hardware registers in case of RX
buffer allocation failure.

13 years agoMake sure to check validity of dma maps before destroying.
Pyun YongHyeon [Thu, 13 Jan 2011 23:00:28 +0000 (23:00 +0000)]
Make sure to check validity of dma maps before destroying.

13 years agoThe (%esp & 0xf) == 0 should be true before the call instruction is
Konstantin Belousov [Thu, 13 Jan 2011 23:00:22 +0000 (23:00 +0000)]
The (%esp & 0xf) == 0 should be true before the call instruction is
executed, for the properly aligned stack.

Reported and tested by: rstone
Pointy hat to: kib
MFC after: 3 days

13 years agore_reset() should be called only after setting device specific
Pyun YongHyeon [Thu, 13 Jan 2011 22:52:57 +0000 (22:52 +0000)]
re_reset() should be called only after setting device specific
features.

13 years agoAllow TX/RX checksum offloading to be configured independently.
Pyun YongHyeon [Thu, 13 Jan 2011 22:49:10 +0000 (22:49 +0000)]
Allow TX/RX checksum offloading to be configured independently.

13 years ago- Move ether_ifdetach() earlier and remove now-unneeded IN_DETACH flag.
John Baldwin [Thu, 13 Jan 2011 21:49:14 +0000 (21:49 +0000)]
- Move ether_ifdetach() earlier and remove now-unneeded IN_DETACH flag.
- Expand locking in interrupt handler.

Reviewed by: yongari

13 years agoApply a workaround for a binutils issue with the .note.ABI-tag section
Dimitry Andric [Thu, 13 Jan 2011 20:44:31 +0000 (20:44 +0000)]
Apply a workaround for a binutils issue with the .note.ABI-tag section
generated from lib/csu/common/crtbrand.c (which ultimately ends up in
executables and shared libraries, via crt1.o, gcrt1.o or Scrt1.o).

For all arches except sparc, gcc emits the section directive for the
abitag struct in crtbrand.c with a PROGBITS type.  However, newer
versions of binutils (after 2.16.90) require the section to be of NOTE
type, to guarantee that the .note.ABI-tag section correctly ends up in
the first page of the final executable.

Unfortunately, there is no clean way to tell gcc to use another section
type, so crtbrand.c (or the C files that include it) must be compiled in
multiple steps:

- Compile the .c file to a .s file.
- Edit the .s file to change the 'progbits' type to 'note', for the section
  directive that defines the .note.ABI-tag section.
- Compile the .s file to an object file.

These steps are done in the invididual Makefiles for each applicable arch.

Reviewed by: kib

13 years ago- Add support for 64-byte contexts to XHCI driver.
Hans Petter Selasky [Thu, 13 Jan 2011 20:03:55 +0000 (20:03 +0000)]
- Add support for 64-byte contexts to XHCI driver.
- Remove some dead code.
- Fixed one instance of missing endian conversion.

Approved by: thompsa (mentor)

13 years agoRemove /usr/share/examples/[lib]dialog, missed in the new dialog import.
Nathan Whitehorn [Thu, 13 Jan 2011 18:52:13 +0000 (18:52 +0000)]
Remove /usr/share/examples/[lib]dialog, missed in the new dialog import.

Submitted by: dim

13 years agoOne more sysctl(9) type-safety that I missed before.
Matthew D Fleming [Thu, 13 Jan 2011 18:20:37 +0000 (18:20 +0000)]
One more sysctl(9) type-safety that I missed before.

13 years agoAdd a 64-bit hex-printed sysctl(9) since there is at least one place in
Matthew D Fleming [Thu, 13 Jan 2011 18:20:33 +0000 (18:20 +0000)]
Add a 64-bit hex-printed sysctl(9) since there is at least one place in
the code that wanted it.  It is named X64 rather than XQUAD since the
quad name is a historical abomination that should not be perpetuated.

13 years agoFix up a few more sysctl(9) mis-typing found in various LINT builds.
Matthew D Fleming [Thu, 13 Jan 2011 18:20:27 +0000 (18:20 +0000)]
Fix up a few more sysctl(9) mis-typing found in various LINT builds.

13 years agoRe-commit the zfs sysctl(9) type-safety changes.
Matthew D Fleming [Thu, 13 Jan 2011 18:20:19 +0000 (18:20 +0000)]
Re-commit the zfs sysctl(9) type-safety changes.

Thanks to dim and pjd for the pointer to zfs_context.h for building
userland.

13 years agoMerge ACPICA 20110112. Switch to BSD/GPLv2 dual license[1].
Jung-uk Kim [Thu, 13 Jan 2011 17:32:32 +0000 (17:32 +0000)]
Merge ACPICA 20110112.  Switch to BSD/GPLv2 dual license[1].

Discussed with: core [1]

13 years agoUpdate spkrtest(8) to use new dialog auto-sizing syntax.
Nathan Whitehorn [Thu, 13 Jan 2011 17:30:18 +0000 (17:30 +0000)]
Update spkrtest(8) to use new dialog auto-sizing syntax.

13 years agoUse a blocking malloc() to initialize the dummynet taskq.
John Baldwin [Thu, 13 Jan 2011 17:02:39 +0000 (17:02 +0000)]
Use a blocking malloc() to initialize the dummynet taskq.

Reviewed by: luigi

13 years agoIf an interrupt on an I/O APIC is moved to a different CPU after it has
John Baldwin [Thu, 13 Jan 2011 17:00:22 +0000 (17:00 +0000)]
If an interrupt on an I/O APIC is moved to a different CPU after it has
started to execute, it seems that the corresponding ISR bit in the "old"
local APIC can be cleared.  This causes the local APIC interrupt routine
to fail to find an interrupt to service.  Rather than panic'ing in this
case, simply return from the interrupt without sending an EOI to the
local APIC.  If there are any other pending interrupts in other ISR
registers, the local APIC will assert a new interrupt.

Tested by: steve

13 years agoUse auto-sizing syntax appropriate for the new dialog(1). This should
Nathan Whitehorn [Thu, 13 Jan 2011 16:42:16 +0000 (16:42 +0000)]
Use auto-sizing syntax appropriate for the new dialog(1). This should
probably be modified to use libdialog.

13 years agoDocuments OUTPUT formats. While I'm here cleans up styles and words.
Weongyo Jeong [Thu, 13 Jan 2011 16:37:12 +0000 (16:37 +0000)]
Documents OUTPUT formats.  While I'm here cleans up styles and words.

Reviewed by: brueffer@

13 years agoEmbed a quota error message (C string) into uprintf() fmt.
Sergey Kandaurov [Thu, 13 Jan 2011 16:29:27 +0000 (16:29 +0000)]
Embed a quota error message (C string) into uprintf() fmt.
While here, fix whitespaces.

Approved by: kib (mentor)

13 years agoImport ACPICA 20110112.
Jung-uk Kim [Thu, 13 Jan 2011 16:12:34 +0000 (16:12 +0000)]
Import ACPICA 20110112.

13 years agoSupport for 64 bit PTEs on n32 and n64 compilation.
Jayachandran C. [Thu, 13 Jan 2011 15:17:29 +0000 (15:17 +0000)]
Support for 64 bit PTEs on n32 and n64 compilation.

In n32 and n64, add support for physical address above 4GB by having
64 bit page table entries and physical addresses. Major changes are:
- param.h: update PTE sizes, masks and shift values to support 64 bit PTEs.
- param.h: remove DELAY(), mips_btop(same as atop), mips_ptob (same as
  ptoa), and reformat.
- param.h: remove casting to unsigned long in trunc_page and round_page
  since this will be used on physical addresses.
- _types.h: have 64 bit __vm_paddr_t for n32.
- pte.h: update TLB LO0/1 access macros to support 64 bit PTE
- pte.h: assembly macros for PTE operations.
- proc.h: md_upte is now 64 bit for n32 and n64.
- exception.S and swtch.S: use the new PTE macros for PTE operations.
- cpufunc.h: TLB_LO0/1 registers are 64bit for n32 and n64.
- xlr_machdep.c: Add memory segments above 4GB to phys_avail[] as they are
  supported now.

Reviewed by: jmallett (earlier version)

13 years ago- Add a locked variant of jme_start() and invoke it directly while holding
John Baldwin [Thu, 13 Jan 2011 14:42:43 +0000 (14:42 +0000)]
- Add a locked variant of jme_start() and invoke it directly while holding
  the lock instead of queueing it to a task.
- Do not invoke jme_rxintr() to reclaim any unprocessed but received
  packets when shutting down the interface.  Instead, just drop these
  packets to match the behavior of other drivers.
- Hold the driver lock in the interrupt handler to avoid races with
  ioctl requests to down the interface.

Reviewed by: yongari

13 years agoIntroduce two new helper macros to define the priority ranges used for
John Baldwin [Thu, 13 Jan 2011 14:22:27 +0000 (14:22 +0000)]
Introduce two new helper macros to define the priority ranges used for
interactive timeshare threads (PRI_*_INTERACTIVE) and non-interactive
timeshare threads (PRI_*_BATCH) and use these instead of PRI_*_REALTIME
and PRI_*_TIMESHARE.  No functional change.

Reviewed by: jeff

13 years agoUse software interrupt priorities for USB kthreads instead of hardware
John Baldwin [Thu, 13 Jan 2011 14:15:36 +0000 (14:15 +0000)]
Use software interrupt priorities for USB kthreads instead of hardware
interrupt priorities.

Reviewed by: hps
MFC after: 2 weeks

13 years agoForgot to remove unlock of the driver lock from age_start_locked() when
John Baldwin [Thu, 13 Jan 2011 13:04:49 +0000 (13:04 +0000)]
Forgot to remove unlock of the driver lock from age_start_locked() when
converting it to a locked variant.

PR: kern/153948

13 years agoCleanup physical address and PTE types on MIPS.
Jayachandran C. [Thu, 13 Jan 2011 06:48:43 +0000 (06:48 +0000)]
Cleanup physical address and PTE types on MIPS.

1. Use vm_paddr_t for physical addresses.

There are a few places in the MIPS platform code where vm_offset_t is
used for physical addresses, change these to use vm_paddr_t:
- phys_avail[], physmem_desc[] arrays
- pmap_mapdev(), page_is_managed(), is_cacheable_mem() pmap_map() args
- local variables of various pmap functions

2. Change init_pte_prot() return from int to pt_entry_t, as this can be
64 bit when using 64 bit TLB entries.

3. Update printing of pt_entry_t and of vm_paddr_t to use 'j' format with
uintmax_t. This will be useful later if we plan to use 64bit phsical addr
on 32 bit n32 compilation.

Reviewed by: imp

13 years agoo Somehow I revert Dd macro in the previous commit.
Maxim Konovalov [Thu, 13 Jan 2011 06:20:44 +0000 (06:20 +0000)]
o Somehow I revert Dd macro in the previous commit.
Re-revert it.

13 years agoo Start each sentence on a new line. No content changes.
Maxim Konovalov [Thu, 13 Jan 2011 06:18:00 +0000 (06:18 +0000)]
o Start each sentence on a new line.  No content changes.

Suggested by: jhb
MFC after: 1 week

13 years agoo Typo fixes.
Maxim Konovalov [Thu, 13 Jan 2011 06:10:47 +0000 (06:10 +0000)]
o Typo fixes.

PR: docs/153933
Submitted by: jpaetzel@, Warren Block
MFC after: 1 week

13 years agoFix handling of NX pages on capable CPUs. Thanks to kib for prodding me
Nathan Whitehorn [Thu, 13 Jan 2011 04:37:48 +0000 (04:37 +0000)]
Fix handling of NX pages on capable CPUs. Thanks to kib for prodding me
in the right direction.

13 years agoRevert to using bus_size_t for the bounce_zone's alignment member.
Matthew D Fleming [Thu, 13 Jan 2011 00:52:57 +0000 (00:52 +0000)]
Revert to using bus_size_t for the bounce_zone's alignment member.

Reuqested by: jhb

13 years agoIn the experimental NFS server, when converting an open-owner to a lock-owner,
Zack Kirsch [Wed, 12 Jan 2011 23:46:12 +0000 (23:46 +0000)]
In the experimental NFS server, when converting an open-owner to a lock-owner,
start at sequence id 1 instead of 0, to match up with both Solaris and Linux.

Reviewed by:    rmacklem
Approved by:    zml (mentor)

13 years agoClean up the experimental NFS server replay cache when the module is unloaded.
Zack Kirsch [Wed, 12 Jan 2011 23:34:09 +0000 (23:34 +0000)]
Clean up the experimental NFS server replay cache when the module is unloaded.

Reviewed by:    rmacklem
Approved by:    zml (mentor)

13 years agoFix formatting of cross-references.
Rebecca Cran [Wed, 12 Jan 2011 23:31:28 +0000 (23:31 +0000)]
Fix formatting of cross-references.

13 years agoUn-break the build: use the correct format specifier for sizeof()
Christian S.J. Peron [Wed, 12 Jan 2011 23:07:51 +0000 (23:07 +0000)]
Un-break the build: use the correct format specifier for sizeof()

13 years agoRevert cddl changes for sysctl(9) until I understand why this isn't
Matthew D Fleming [Wed, 12 Jan 2011 23:06:38 +0000 (23:06 +0000)]
Revert cddl changes for sysctl(9) until I understand why this isn't
building on universe.

13 years agoMake sure to invoke unlocked foo_start since the taskqueue does not
Pyun YongHyeon [Wed, 12 Jan 2011 22:24:07 +0000 (22:24 +0000)]
Make sure to invoke unlocked foo_start since the taskqueue does not
hold a driver lock. This should fix a regression introduced in
r216925.

PR: kern/153769

13 years agoFix a brain fart. Since this file is shared between i386 and amd64, a
Matthew D Fleming [Wed, 12 Jan 2011 21:08:49 +0000 (21:08 +0000)]
Fix a brain fart.  Since this file is shared between i386 and amd64, a
bus_size_t may be 32 or 64 bits.  Change the bounce_zone alignment field
to explicitly be 32 bits, as I can't really imagine a DMA device that
needs anything close to 2GB alignment of data.

13 years agoFix a typo. XLONG should allow pointers to unsigned and signed long.
Matthew D Fleming [Wed, 12 Jan 2011 20:38:55 +0000 (20:38 +0000)]
Fix a typo.  XLONG should allow pointers to unsigned and signed long.

Submitted by: <bf1783 AT gmail DOT com>

13 years agoFix sorting of cross-references.
Rebecca Cran [Wed, 12 Jan 2011 20:27:14 +0000 (20:27 +0000)]
Fix sorting of cross-references.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:54:19 +0000 (19:54 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the kernel changes.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:54:14 +0000 (19:54 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the security directory.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:54:07 +0000 (19:54 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the geom piece.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:53:56 +0000 (19:53 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the rest of the devices.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:53:50 +0000 (19:53 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the net* piece.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:53:44 +0000 (19:53 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the cxgb driver piece.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:53:39 +0000 (19:53 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the netgraph piece.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:53:30 +0000 (19:53 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the zfs piece.

13 years agosysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.
Matthew D Fleming [Wed, 12 Jan 2011 19:53:23 +0000 (19:53 +0000)]
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly.

Commit the Intel drivers.

13 years agoFix wrapping of 20110103 entry.
Rebecca Cran [Wed, 12 Jan 2011 19:28:52 +0000 (19:28 +0000)]
Fix wrapping of 20110103 entry.

13 years agoFix cross-reference to gvinum(8).
Rebecca Cran [Wed, 12 Jan 2011 19:20:30 +0000 (19:20 +0000)]
Fix cross-reference to gvinum(8).

13 years agoFix several bugs in the ARP code related to improperly formatted
George V. Neville-Neil [Wed, 12 Jan 2011 19:11:17 +0000 (19:11 +0000)]
Fix several bugs in the ARP code related to improperly formatted
packets.

*) Reject requests with a protocol length not equal to 4.  This is IPv4
and there is no reason to accept anything else.

*) Reject packets that have a multicast source hardware address.

*) Drop requests where the hardware address length is not equal
to the hardware address length of the interface.

Pointed out by: Rozhuk Ivan
MFC after: 1 week

13 years agoFix up the grammar.
Andrey V. Elsukov [Wed, 12 Jan 2011 19:06:59 +0000 (19:06 +0000)]
Fix up the grammar.

PR: docs/153933
MFC after: 3 days

13 years agoAdd type checking for static and dynamic sysctls using scalar types.
Matthew D Fleming [Wed, 12 Jan 2011 17:52:48 +0000 (17:52 +0000)]
Add type checking for static and dynamic sysctls using scalar types.
The code is turned off until the tree is fixed up so it compiles.
__FreeBSD_version was already bumped once today, so skip the bump, but
add an entry to UPDATING.

Note that __DESCR() is used in the SYSCTL_OID() macro and so is not
needed in macros that invoke it.  This use was inconsistent in the
file and I have made it consistent any lines already being changed.

Reviewed by: bde (previous version), -arch (previous version)

13 years agoexecve(2), not fork(2) resets signal handler to the default value (if it isn't
Pawel Jakub Dawidek [Wed, 12 Jan 2011 16:16:54 +0000 (16:16 +0000)]
execve(2), not fork(2) resets signal handler to the default value (if it isn't
ignored). Correct comment talking about that.

Pointed out by: kib
MFC after: 3 days

13 years agoUpdate dialog to version 20100428. This changes the license under which
Nathan Whitehorn [Wed, 12 Jan 2011 14:55:02 +0000 (14:55 +0000)]
Update dialog to version 20100428. This changes the license under which
dialog is distributed from GPLv2 to LGPLv2 and introduces a number of new
features and a new and better libdialog API. The existing libdialog will
be kept temporarily as libodialog for compatibility purposes until sade,
sysinstall and tzsetup have been either updated or replaced.

__FreeBSD_version is now 900030.

Discussed on: -current
Approved by: core
Obtained from: http://invisible-island.net/dialog

13 years agoAdd a note that when custom signal handler is installed for a signal,
Pawel Jakub Dawidek [Wed, 12 Jan 2011 14:38:17 +0000 (14:38 +0000)]
Add a note that when custom signal handler is installed for a signal,
signal action is restored to default in child after fork(2).
In this case there is no need to do anything with dummy SIGCHLD handler,
because after fork(2) it will be automatically reverted to SIG_IGN.

Obtained from: Wheel Systems Sp. z o.o. http://www.wheelsystems.com
MFC after: 3 days

13 years agoInstall default signal handlers before masking signals we want to handle.
Pawel Jakub Dawidek [Wed, 12 Jan 2011 14:35:29 +0000 (14:35 +0000)]
Install default signal handlers before masking signals we want to handle.
It is possible that the parent process ignores some of them and sigtimedwait()
will never see them, eventhough they are masked.

The most common situation for this to happen is boot process where init(8)
ignores SIGHUP before starting to execute /etc/rc. This in turn caused
hastd(8) to ignore SIGHUP.

Reported by: trasz
Obtained from: Wheel Systems Sp. z o.o. http://www.wheelsystems.com
MFC after: 3 days

13 years agoSector size can not be greater than MAXPHYS. Since GRAID3 calculates
Andrey V. Elsukov [Wed, 12 Jan 2011 13:55:01 +0000 (13:55 +0000)]
Sector size can not be greater than MAXPHYS. Since GRAID3 calculates
sector size from user-specified block size, report to user about
big blocksize.

PR: kern/147851
MFC after: 1 week

13 years agoThe message ring interrupt needs to be enabled for all cpus, not just
Jayachandran C. [Wed, 12 Jan 2011 13:16:35 +0000 (13:16 +0000)]
The message ring interrupt needs to be enabled for all cpus, not just
the ones which run the message ring handler.

Some bits of the interrupt mask are part of the status register which is
saved with the process context, and these bits are initialized from the
cpu on which the process is created. This means that all the processes
should have the same value for these interrupt mask bits, so that the
interrupt mask remains the same regardless of what thread is scheduled
on the cpu.

Submitted by: Sriram Gorti (srgorti at netlogicmicro dot com)

13 years agoSector size can not be greater than MAXPHYS.
Andrey V. Elsukov [Wed, 12 Jan 2011 12:26:10 +0000 (12:26 +0000)]
Sector size can not be greater than MAXPHYS.

MFC after: 1 week

13 years agoProvide up-to-date estimates that are pessimal today...
Warner Losh [Wed, 12 Jan 2011 04:59:29 +0000 (04:59 +0000)]
Provide up-to-date estimates that are pessimal today...

13 years agoFor re(4) controllers that uses new jumbo frame scheme(RTL8168C/D/E),
Pyun YongHyeon [Wed, 12 Jan 2011 03:43:47 +0000 (03:43 +0000)]
For re(4) controllers that uses new jumbo frame scheme(RTL8168C/D/E),
limit maximum RX buffer size to RE_RX_DESC_BUFLEN instead of
blindly configuring it to 16KB. Due to lack of documentation, re(4)
didn't allow jumbo frame on these controllers. However it seems
controller is confused with jumbo frame such that it can DMA the
received frame to wrong address instead of splitting it into
multiple RX buffers. Of course, this caused panic.

Since re(4) does not support jumbo frames on these controllers,
make controller drop frame that is longer than RE_RX_DESC_BUFLEN
sized frame. Fortunately RTL810x controllers, which do not support
jumbo frame, have no such issues but this change also limited
maximum RX buffer size allowed to RTL810x controllers. Allowing
16KB RX buffer for controllers that have no such capability is
meaningless.

MFC after: 3 days

13 years agoA couple problems discovered by Andrew Boyer:
Jack F Vogel [Wed, 12 Jan 2011 00:23:47 +0000 (00:23 +0000)]
A couple problems discovered by Andrew Boyer:
   - failure code in em_xmit got mangled along the way
     and was not properly handling errors.
   - local timer code had a leftover UNLOCK call that
     should be removed.

MFC after 3 days