]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
14 years agoDo not force verbose and single mode in non-metadata boot case.
Rafal Jaworowski [Wed, 24 Feb 2010 20:31:00 +0000 (20:31 +0000)]
Do not force verbose and single mode in non-metadata boot case.

We want to go multi-user by default also in case of booting without loader(8).

14 years agoImprove VESA mode switching via loader tunable `hint.sc.0.vesa_mode'.
Jung-uk Kim [Wed, 24 Feb 2010 20:13:34 +0000 (20:13 +0000)]
Improve VESA mode switching via loader tunable `hint.sc.0.vesa_mode'.
The most notable change is history buffer is fully saved/restored now.

14 years agoUpdate the commented out option for omitting the sysctl descriptions; it
Rebecca Cran [Wed, 24 Feb 2010 19:28:15 +0000 (19:28 +0000)]
Update the commented out option for omitting the sysctl descriptions; it
was committed as NO_SYSCTL_DESCR.

Approved by: rrs (mentor)

14 years agoUse the DB_SHOW_ALL_COMMAND() macro to register the formerly 'show ifnets'
Bjoern A. Zeeb [Wed, 24 Feb 2010 15:54:24 +0000 (15:54 +0000)]
Use the DB_SHOW_ALL_COMMAND() macro to register the formerly 'show ifnets'
in the db_show_all_table as 'show all ifnets' and with that follow the
convention for showing complete lists.

Submitted by: thompsa
MFC after: 3 days

14 years agoDon't inforce an upper bound on kern.ngroups. The INT_MAX-1 limit was
Brooks Davis [Wed, 24 Feb 2010 15:52:18 +0000 (15:52 +0000)]
Don't inforce an upper bound on kern.ngroups.  The INT_MAX-1 limit was
too high due to several overflows.  The actual limit is somewhere in the
neighborhood of INT_MAX/4 on 64-bit machines, but most systems could not
support such a limit due to a lack of memory and the cost of duplicate
credentials.

Reported by: bde

14 years agoFix expansion of \W in prompt strings when the working directory is "/".
Jaakko Heinonen [Wed, 24 Feb 2010 14:19:56 +0000 (14:19 +0000)]
Fix expansion of \W in prompt strings when the working directory is "/".
The prompt string was truncated after \W when the working directory was "/".

PR: bin/89410
Submitted by: Dr Balwinder Singh Dheeman
MFC after: 1 week

14 years agoFix spelling of EACCES.
John Baldwin [Wed, 24 Feb 2010 13:13:29 +0000 (13:13 +0000)]
Fix spelling of EACCES.

PR: docs/144112
Submitted by: Alexander Best
MFC after: 3 days

14 years agoThere is no need to test __FreeBSD_version for features that have
Navdeep Parhar [Wed, 24 Feb 2010 10:16:18 +0000 (10:16 +0000)]
There is no need to test __FreeBSD_version for features that have
been around for a long time now (7.1-ish or even earlier); assume
they are present.  These includes MSI, TSO, LRO, VLAN, INTR_FILTERS,
FIRMWARE, etc.

Also, eliminate some dead code and clean up in other places as part
of this quick once-over.

MFC after: 1 week

14 years agoAccessing an mbuf after it has been handed off to the hardware is a bad
Navdeep Parhar [Wed, 24 Feb 2010 01:44:39 +0000 (01:44 +0000)]
Accessing an mbuf after it has been handed off to the hardware is a bad
race as it could already have been tx'd and freed by that time.  Place
the bpf tap just _before_ writing the gen bit.

This fixes a panic when running tcpdump on a cxgb interface.

14 years agoAdd the ability to set SMU-based machines to restart automatically after
Nathan Whitehorn [Wed, 24 Feb 2010 01:27:36 +0000 (01:27 +0000)]
Add the ability to set SMU-based machines to restart automatically after
power loss.

14 years agoUse dcbz instead of word stores for page zeroing, providing a factor of
Nathan Whitehorn [Wed, 24 Feb 2010 00:55:55 +0000 (00:55 +0000)]
Use dcbz instead of word stores for page zeroing, providing a factor of
3-4 speedup.

14 years agoClose a race involving the OEA64 scratchpage. When the scratch page's
Nathan Whitehorn [Wed, 24 Feb 2010 00:54:37 +0000 (00:54 +0000)]
Close a race involving the OEA64 scratchpage. When the scratch page's
physical address is changed, there is a brief window during which its PTE
is invalid. Since moea64_set_scratchpage_pa() does not and cannot hold
the page table lock, it was possible for another CPU to insert a new PTE
into the scratch page's PTEG slot during this interval, corrupting both
mappings.

Solve this by creating a new flag, LPTE_LOCKED, such that
moea64_pte_insert will avoid claiming locked PTEG slots even if they
are invalid. This change also incorporates some additional paranoia
added to solve things I thought might be this bug.

Reported by: linimon

14 years agoAllow cxgbtool to build with WARNS=6
Navdeep Parhar [Tue, 23 Feb 2010 23:01:41 +0000 (23:01 +0000)]
Allow cxgbtool to build with WARNS=6

MFC after: 1 week

14 years agoYet another attempt to make palette loading more safer:
Jung-uk Kim [Tue, 23 Feb 2010 21:51:14 +0000 (21:51 +0000)]
Yet another attempt to make palette loading more safer:

- Add a separate palette data for 8-bit DAC mode when SC_PIXEL_MODE is set
and fill it up with default gray-scale palette data for text.  Now we don't
have to set `hint.sc.0.vesa_mode' to get the default palette data.
- Add a new adapter flag, V_ADP_DAC8 to track whether the controller is
using 8-bit palette format and load correct palette when switching modes.
- Set 8-bit DAC mode only for non-VGA compatible graphics mode.

14 years agoMinor diff reduction with Adaptec's driver: in aac_release_command() set
Ed Maste [Tue, 23 Feb 2010 21:41:13 +0000 (21:41 +0000)]
Minor diff reduction with Adaptec's driver: in aac_release_command() set
cm_queue to AAC_ADAP_NORM_CMD_QUEUE by default.  In every place it was set,
it was set to AAC_ADAP_NORM_CMD_QUEUE anyhow.

14 years agoo adds sysctl variables to show device statistics.
Weongyo Jeong [Tue, 23 Feb 2010 19:55:54 +0000 (19:55 +0000)]
o adds sysctl variables to show device statistics.
o records RTS success/fail statistics.

Pointed by: imp

14 years agofixes a compile error; invalid type argument of '->'.
Weongyo Jeong [Tue, 23 Feb 2010 19:44:51 +0000 (19:44 +0000)]
fixes a compile error; invalid type argument of '->'.

14 years agoMake xpt_rescan() more correct, as it was before r197208: do not use
Alexander Motin [Tue, 23 Feb 2010 18:42:07 +0000 (18:42 +0000)]
Make xpt_rescan() more correct, as it was before r197208: do not use
XPT_SCAN_LUN for wildcard LUN, fall back to XPT_SCAN_BUS.

14 years agoFix recursive lock attempt on hot-plug event in non-ATA_CAM mode.
Alexander Motin [Tue, 23 Feb 2010 16:39:53 +0000 (16:39 +0000)]
Fix recursive lock attempt on hot-plug event in non-ATA_CAM mode.

14 years agoThe New and Improved snake_server - Service Pack 1: now even more
Ivan Voras [Tue, 23 Feb 2010 15:27:07 +0000 (15:27 +0000)]
The New and Improved snake_server - Service Pack 1: now even more
sensitive to load average variations!

14 years agoUpgrade the "snake" syscons screensaver to the new, multimedia version!
Ivan Voras [Tue, 23 Feb 2010 15:12:41 +0000 (15:12 +0000)]
Upgrade the "snake" syscons screensaver to the new, multimedia version!
Now, with color! And system load averages!

Amused by it: gnn

14 years agoFix two bugs in this manpage: 'System' is not abbreviated with 'P',
Nathan Whitehorn [Tue, 23 Feb 2010 14:56:49 +0000 (14:56 +0000)]
Fix two bugs in this manpage: 'System' is not abbreviated with 'P',
and a non-useful version of smu(4) was in 8.0, so modify the history
to reflect that.

Submitted by: gavin

14 years agoxpt_rescan only honors a wildcard in the target field. Revert the previous
Matt Jacob [Tue, 23 Feb 2010 14:35:44 +0000 (14:35 +0000)]
xpt_rescan only honors a wildcard in the target field. Revert the previous
change and have isp_make_here scan the whole bus which will then scan all
luns.

I think xpt_rescan needs to be fixed, but that's a separable issue.

Suggested by: Alexander

14 years agoFix compilation problems with INVARIANTS.
Warner Losh [Tue, 23 Feb 2010 05:02:10 +0000 (05:02 +0000)]
Fix compilation problems with INVARIANTS.

# also limit RX decryption attempted messages to 50

Reviewed by: weongyo

14 years agoAdd a manpage for smu(4) now that it does something useful.
Nathan Whitehorn [Tue, 23 Feb 2010 04:56:06 +0000 (04:56 +0000)]
Add a manpage for smu(4) now that it does something useful.

14 years agoRe-add accidentally removed pixel format for direct memory model.
Jung-uk Kim [Tue, 23 Feb 2010 01:02:11 +0000 (01:02 +0000)]
Re-add accidentally removed pixel format for direct memory model.

14 years agoFix FBIO_ADPINFO ioctl on amd64.
Jung-uk Kim [Tue, 23 Feb 2010 01:00:19 +0000 (01:00 +0000)]
Fix FBIO_ADPINFO ioctl on amd64.

14 years agoWhen we rescan, just scan from logical unit 0. In other words, don't
Matt Jacob [Tue, 23 Feb 2010 00:34:20 +0000 (00:34 +0000)]
When we rescan, just scan from logical unit 0. In other words, don't
specify a wildcard lun here.

This unbreaks disk re-arrival.

MFC after: 2 days

14 years agoRemove Tx mbuf parsing code for VLAN in TSO path. Controller does
Pyun YongHyeon [Mon, 22 Feb 2010 23:35:35 +0000 (23:35 +0000)]
Remove Tx mbuf parsing code for VLAN in TSO path. Controller does
not support TSO over VLAN if VLAN hardware tagging is disabled so
there is no need to check VLAN here.

14 years agoAdd TSO support on VLANs. Also make sure to update TSO capability
Pyun YongHyeon [Mon, 22 Feb 2010 22:41:27 +0000 (22:41 +0000)]
Add TSO support on VLANs. Also make sure to update TSO capability
whenever jumbo frame is configured.
While I'm here remove unnecessary check of VLAN hardware checksum
offloading. vlan(4) already takes care of this.

14 years agoPOSIX patch(1) would treat -b as different meaning (the functionality
Xin LI [Mon, 22 Feb 2010 22:32:24 +0000 (22:32 +0000)]
POSIX patch(1) would treat -b as different meaning (the functionality
is to be provided by --suffix).  Looking at the usage here in diffutils,
it seems that we can just get rid of the -b .orig stuff.  This resolves
a problem that can triggered if we move toward to a more permissively
licensed patch(1) program.

14 years agoAdd TSO support on VLANs. jme(4) controllers do not require VLAN
Pyun YongHyeon [Mon, 22 Feb 2010 22:05:49 +0000 (22:05 +0000)]
Add TSO support on VLANs. jme(4) controllers do not require VLAN
hardware tagging to make TSO work over VLANs.

14 years agoRemove Tx mbuf parsing code for VLAN in TSO path. Controller does
Pyun YongHyeon [Mon, 22 Feb 2010 21:03:15 +0000 (21:03 +0000)]
Remove Tx mbuf parsing code for VLAN in TSO path. Controller does
not support TSO over VLAN if VLAN hardware tagging is disabled so
there is no need to check VLAN here.
While I'm here make sure to pullup IP/TCP headers in the first
buffer.

14 years agoAccording to the Linux sungem driver, in case of Apple (K2) GMACs
Marius Strobl [Mon, 22 Feb 2010 21:01:08 +0000 (21:01 +0000)]
According to the Linux sungem driver, in case of Apple (K2) GMACs
GEM_MIF_CONFIG_MDI0 cannot be trusted when the firmware has powered
down the chip so the internal transceiver has to be hardcoded. This
is also in line with the AppleGMACEthernet driver, which just doesn't
distinguish between internal/external transceiver and MDIO/MDI1
respectively in the first place. Tested by: Andreas Tobler

MFC after: 1 week

14 years agoVirtualize transport part of periph announcement.
Alexander Motin [Mon, 22 Feb 2010 19:17:17 +0000 (19:17 +0000)]
Virtualize transport part of periph announcement.

14 years agoAdd TSO on VLANs. Because re(4) has a TSO limitation for jumbo
Pyun YongHyeon [Mon, 22 Feb 2010 18:51:14 +0000 (18:51 +0000)]
Add TSO on VLANs. Because re(4) has a TSO limitation for jumbo
frame, make sure to update VLAN capabilities whenever jumbo frame
is configured.
While I'm here rearrange interface capabilities configuration. The
controller requires VLAN hardware tagging to make TSO work on VLANs
so explicitly check this requirement.

14 years agoProvide a new useless feature: an led(4) interface for the system's sleep
Nathan Whitehorn [Mon, 22 Feb 2010 18:49:43 +0000 (18:49 +0000)]
Provide a new useless feature: an led(4) interface for the system's sleep
LED.

14 years agoOops! Wrong word order. :(
Alexander Motin [Mon, 22 Feb 2010 17:34:35 +0000 (17:34 +0000)]
Oops! Wrong word order. :(

14 years agoEnforce stronger semantics for bus-dma alignment (currently only on amd64).
Justin T. Gibbs [Mon, 22 Feb 2010 17:03:45 +0000 (17:03 +0000)]
Enforce stronger semantics for bus-dma alignment (currently only on amd64).
Now all contiguous regions returned from bus-dma will be aligned to the
alignment constraint and all but the last region are guaranteed to be
a multiple of the alignment in length.  This also means that the relative
alignment of two adjacent bytes in the I/O stream have a difference of 1
even if they are not physically contiguous.

The old code, when needing to perform a copy in order to align data, only
copied the amount of data needed to reach the next page boundary.  This
often left an unaligned end to the segment.  Drivers such as Xen's blkfront
can't deal with such segments.

The downside to this approach is that, once an unaligned region is encountered,
the remainder of the I/O will be bounced.  However, bouncing should be rare.
It is typically caused by non-performance critical userland programs that
don't bother to align their I/O buffers (e.g. bsdlabel).  In-kernel I/O
buffers are always aligned to at least a page boundary.

Reviewed by: scottl
MFC after:      2 weeks

14 years agoFix some typos.
Bernhard Schmidt [Mon, 22 Feb 2010 17:03:26 +0000 (17:03 +0000)]
Fix some typos.

Approved by: rpaulo (mentor)
MFC after: 1 week

14 years agoUpdate mxge to support IFCAP_VLAN_HWTSO.
Andrew Gallatin [Mon, 22 Feb 2010 16:57:03 +0000 (16:57 +0000)]
Update mxge to support IFCAP_VLAN_HWTSO.

Note: If/when FreeBSD supports TSO over IPv6, the minimal mxge fw
rev to enable IFCAP_VLAN_HWTSO will need to be increased to 1.4.37

14 years agoSupport the extended PLT format used when objects have more than 8192
Nathan Whitehorn [Mon, 22 Feb 2010 16:49:45 +0000 (16:49 +0000)]
Support the extended PLT format used when objects have more than 8192
PLT relocations on PPC32.

14 years agoAdd Intel PCH SATA controller IDs.
Alexander Motin [Mon, 22 Feb 2010 16:27:47 +0000 (16:27 +0000)]
Add Intel PCH SATA controller IDs.

14 years agoFix constant assignment for netisr protocol information sysctl.
Robert Watson [Mon, 22 Feb 2010 16:16:16 +0000 (16:16 +0000)]
Fix constant assignment for netisr protocol information sysctl.

MFC after: 1 week
Spotted by: bz

14 years agoUpdate date on netstat(1) for -Q.
Robert Watson [Mon, 22 Feb 2010 16:05:45 +0000 (16:05 +0000)]
Update date on netstat(1) for -Q.

Suggested by: bz
MFC after: 1 week

14 years agoCross-reference netstat(1) from netisr(9) since you can now use it to
Robert Watson [Mon, 22 Feb 2010 16:02:17 +0000 (16:02 +0000)]
Cross-reference netstat(1) from netisr(9) since you can now use it to
monitor netisr status.

MFC after: 1 week
Sponsored by: Juniper Networks

14 years agoRemove write-only variable.
Konstantin Belousov [Mon, 22 Feb 2010 16:00:56 +0000 (16:00 +0000)]
Remove write-only variable.

MFC after: 3 days

14 years agoTeach netstat(1) to print out netisr statistics when given the -Q argument.
Robert Watson [Mon, 22 Feb 2010 15:57:36 +0000 (15:57 +0000)]
Teach netstat(1) to print out netisr statistics when given the -Q argument.
Currently supports only reporting on live systems via sysctl, kmem support
needs to be edded.

MFC after: 1 week
Sponsored by: Juniper Networks

14 years agoExport netisr configuration and statistics to userspace via sysctl(9).
Robert Watson [Mon, 22 Feb 2010 15:03:16 +0000 (15:03 +0000)]
Export netisr configuration and statistics to userspace via sysctl(9).

MFC after: 1 week
Sponsored by: Juniper Networks

14 years agoFix previous commit: led_func() doesn't exist, it should be fled_cb().
Rui Paulo [Mon, 22 Feb 2010 14:49:52 +0000 (14:49 +0000)]
Fix previous commit: led_func() doesn't exist, it should be fled_cb().

Pointed out by: bz

14 years agoAllow user programs to execute mfpvr instructions. Linux allows this, and
Nathan Whitehorn [Mon, 22 Feb 2010 14:17:23 +0000 (14:17 +0000)]
Allow user programs to execute mfpvr instructions. Linux allows this, and
some math-related software like GMP expects to be able to use it to pick
a target appropriately.

MFC after: 1 week

14 years agoImprove output for controllers that doesn't report SATA speed.
Alexander Motin [Mon, 22 Feb 2010 10:45:40 +0000 (10:45 +0000)]
Improve output for controllers that doesn't report SATA speed.

14 years agoUse KERNLOADADDR instead of hardcoded address. This is required to be able to
Neel Natu [Mon, 22 Feb 2010 01:48:38 +0000 (01:48 +0000)]
Use KERNLOADADDR instead of hardcoded address. This is required to be able to
build the trampoline kernel that has symbol/string tables embedded within it.

14 years agoUse mf and not mf.a. The latter doesn't force memory ordering and
Marcel Moolenaar [Mon, 22 Feb 2010 01:24:34 +0000 (01:24 +0000)]
Use mf and not mf.a. The latter doesn't force memory ordering and
applies to sequential memory.

14 years agoPrefer I-units and M-units for nop instructions. This works around
Marcel Moolenaar [Mon, 22 Feb 2010 01:23:41 +0000 (01:23 +0000)]
Prefer I-units and M-units for nop instructions. This works around
McKinley flaws. It also avoids using the F-unit in the kernel for
no reason.

14 years agoNormalize nop instructions: Only use 0 for the immediate operand.
Marcel Moolenaar [Sun, 21 Feb 2010 23:41:59 +0000 (23:41 +0000)]
Normalize nop instructions: Only use 0 for the immediate operand.

14 years agoRemove pm_active from struct pmap as it serves no purpose.
Marcel Moolenaar [Sun, 21 Feb 2010 23:10:13 +0000 (23:10 +0000)]
Remove pm_active from struct pmap as it serves no purpose.

MFC after: 1 week

14 years agoAlign the start of the clean submap to a superpage boundary. Although
Alan Cox [Sun, 21 Feb 2010 22:23:13 +0000 (22:23 +0000)]
Align the start of the clean submap to a superpage boundary.  Although
no superpage mappings are created within the clean submap, aligning the
start of the clean submap helps to prevent interference with kmem_alloc()'s
use of superpages.

14 years agoAdd a simple fan management callout to the SMU driver. This is designed
Nathan Whitehorn [Sun, 21 Feb 2010 22:13:57 +0000 (22:13 +0000)]
Add a simple fan management callout to the SMU driver. This is designed
such that a fancier thermal management algorithm can be run from user
space, but the kernel will at least ensure your machine does not either
sound like a wind tunnel or catch fire.

14 years agoFix several mistakes in this file, in order to allow individual fan speeds
Nathan Whitehorn [Sun, 21 Feb 2010 21:03:37 +0000 (21:03 +0000)]
Fix several mistakes in this file, in order to allow individual fan speeds
to be read and set correctly.

14 years agoPrefer exit(0) from main over return(0).
Ulrich Spörlein [Sun, 21 Feb 2010 21:03:15 +0000 (21:03 +0000)]
Prefer exit(0) from main over return(0).

Although not explicitly mentioned in style(9), it allows for easier
grepping of exit points. This reverts part of r203926.

Requested by: des
Approved by: philip (mentor)

14 years agoChanging proto_socketpair.c compilation and linking order revealed
Pawel Jakub Dawidek [Sun, 21 Feb 2010 19:56:47 +0000 (19:56 +0000)]
Changing proto_socketpair.c compilation and linking order revealed
a problem - we should simply ignore proto_server() if address
doesn't start with socketpair://, and not abort.

14 years agoDecompose the most lousy named file in sys/kern; kern_subr.c.
Ed Schouten [Sun, 21 Feb 2010 19:53:33 +0000 (19:53 +0000)]
Decompose the most lousy named file in sys/kern; kern_subr.c.

Although this file has historically been used as a dumping ground for
random functions, nowadays it only contains functions related to copying
bits {from,to} userspace and hash table utility functions.

Behold, subr_uio.c and subr_hash.c.

14 years agoFix another fo-pa of mine... duplicate patches should
Randall Stewart [Sun, 21 Feb 2010 17:27:20 +0000 (17:27 +0000)]
Fix another fo-pa of mine... duplicate patches should
not be applied and randy needs coffee in the morning
when working to help keep things sorted out... obviously :-)

14 years agoAdds the missing mips gdb files that I
Randall Stewart [Sun, 21 Feb 2010 17:25:00 +0000 (17:25 +0000)]
Adds the missing mips gdb files that I
foobar'd on my last commit.. Sorry about
that JC.. my fault.

Obtained from: JC

14 years agoifconfig(8) expects interface fooX to be supported by the module if_foo,
Robert Watson [Sun, 21 Feb 2010 15:25:47 +0000 (15:25 +0000)]
ifconfig(8) expects interface fooX to be supported by the module if_foo,
and will try to load it if it's not present.  To better meet these
expectations, change the module name for the loopback interface from
'loop' to 'if_lo'.  The loopback interface is always compiled into the
base kernel, so there are no resulting changes in kld files, etc.

Discussed with: brooks (ages ago)
MFC after: 1 week

14 years agoWhile there, trim some trailing whitespace.
Ed Schouten [Sun, 21 Feb 2010 14:58:01 +0000 (14:58 +0000)]
While there, trim some trailing whitespace.

14 years ago- Remove GNU bc/dc bits. There has not been any regressions seen after
Gabor Kovesdan [Sun, 21 Feb 2010 14:28:03 +0000 (14:28 +0000)]
- Remove GNU bc/dc bits. There has not been any regressions seen after
  BSD bc/dc import. It has been tested with make universe that this code
  is not used by anything any more.

Approved by: delphij (mentor)

14 years agoAdd proper const keywords to sysctl(3) parameters.
Ed Schouten [Sun, 21 Feb 2010 13:57:02 +0000 (13:57 +0000)]
Add proper const keywords to sysctl(3) parameters.

The `name' and `newp' arguments can be marked const, because the buffers
they refer to are never changed. While there, perform some other
cleanups:

- Remove K&R from sysctl.c.
- Implement sysctlbyname() using sysctlnametomib() to prevent
  duplication of an undocumented kernel interface.
- Fix some whitespace nits.

It seems the prototypes are now in sync with NetBSD as well.

14 years agoDocument the interaction between /etc/devfs.conf and
Gavin Atkinson [Sun, 21 Feb 2010 10:29:45 +0000 (10:29 +0000)]
Document the interaction between /etc/devfs.conf and
/etc/defaults/devfs.conf

PR: docs/117308
Submitted by: Mel <mel.xyzzy rachie.is-a-geek.net> (partially)
MFC after: 1 week

14 years agoAdd a "-x" option to chown(8)/chgrp(1) similar to the same option in
Gavin Atkinson [Sun, 21 Feb 2010 10:14:06 +0000 (10:14 +0000)]
Add a "-x" option to chown(8)/chgrp(1) similar to the same option in
du(1), cp(1) etc, to prevent the crossing of mountpoints whilst using the
commands recursively.

PR: bin/130855
Submitted by: keramida
MFC after: 1 month

14 years agoSome machines can not only consist of CPUs running at different speeds
Marius Strobl [Sun, 21 Feb 2010 09:25:53 +0000 (09:25 +0000)]
Some machines can not only consist of CPUs running at different speeds
but also of different types, f.e. Sun Fire V890 can be equipped with a
mix of UltraSPARC IV and IV+ CPUs, requiring different MMU initialization
and different workarounds for model specific errata. Therefore move the
CPU implementation number from a global variable to the per-CPU data.
Functions which are called before the latter is available are passed the
implementation number as a parameter now.

This file was missed in r204152.

14 years agoSince create_pagetables() zeroes the page tables, pmap_bootstrap() needn't
Alan Cox [Sun, 21 Feb 2010 03:49:39 +0000 (03:49 +0000)]
Since create_pagetables() zeroes the page tables, pmap_bootstrap() needn't
zero *CMAP1.

14 years ago- fix bootstrap for variable KVA_PAGES
Kip Macy [Sun, 21 Feb 2010 01:13:34 +0000 (01:13 +0000)]
- fix bootstrap for variable KVA_PAGES
- remove unused CADDR1
- hold lock across page table update

MFC after: 3 days

14 years agodon't hold spin lock across free
Kip Macy [Sun, 21 Feb 2010 01:12:18 +0000 (01:12 +0000)]
don't hold spin lock across free

14 years ago- make printf conditional
Kip Macy [Sun, 21 Feb 2010 01:11:39 +0000 (01:11 +0000)]
- make printf conditional
- fix witness warnings by making configuration lock a mutex

14 years agoremove atkbd from default config to avoid pulling in real-mode bios emulation
Kip Macy [Sun, 21 Feb 2010 01:06:07 +0000 (01:06 +0000)]
remove atkbd from default config to avoid pulling in real-mode bios emulation

14 years agoAdd __FBSDID.
Pyun YongHyeon [Sun, 21 Feb 2010 00:07:45 +0000 (00:07 +0000)]
Add __FBSDID.

Reviewed by: sam

14 years agoIncrease PCIe maximuim read request size to 2048. Because re(4) uses
Pyun YongHyeon [Sun, 21 Feb 2010 00:00:55 +0000 (00:00 +0000)]
Increase PCIe maximuim read request size to 2048. Because re(4) uses
Tx DMA burst size 2048, I beleive PCIe maximum read request size
also should match to the value of Tx DMA burst size. With this
change I can get more than 800Mbps for TCP bulk transfers.
Previously I was not able to get more than 700Mbps. If I enable TSO
it now shows 927Mbps.

14 years agoStarting with UltraSPARC IV CPUs the CPU caches are described with different
Marius Strobl [Sat, 20 Feb 2010 23:42:24 +0000 (23:42 +0000)]
Starting with UltraSPARC IV CPUs the CPU caches are described with different
OFW properties.

14 years agoSome machines can not only consist of CPUs running at different speeds
Marius Strobl [Sat, 20 Feb 2010 23:24:19 +0000 (23:24 +0000)]
Some machines can not only consist of CPUs running at different speeds
but also of different types, f.e. Sun Fire V890 can be equipped with a
mix of UltraSPARC IV and IV+ CPUs, requiring different MMU initialization
and different workarounds for model specific errata. Therefore move the
CPU implementation number from a global variable to the per-CPU data.
Functions which are called before the latter is available are passed the
implementation number as a parameter now.

14 years agoAdd TSO support on VLAN. Controller requires VLAN hardware tagging
Pyun YongHyeon [Sat, 20 Feb 2010 23:21:06 +0000 (23:21 +0000)]
Add TSO support on VLAN. Controller requires VLAN hardware tagging
to make TSO work on VLAN. So if VLAN hardware tagging is disabled
explicitly clear TSO on VLAN. While I'm here remove duplicated
VLAN_CAPABILITIES call.

14 years agoAdd TSO support on VLAN in fconfig(8).
Pyun YongHyeon [Sat, 20 Feb 2010 23:01:09 +0000 (23:01 +0000)]
Add TSO support on VLAN in fconfig(8).

Reviewed by: thompsa

14 years agoAdd TSO support on VLANs. Intentionally separated IFCAP_VLAN_HWTSO
Pyun YongHyeon [Sat, 20 Feb 2010 22:47:20 +0000 (22:47 +0000)]
Add TSO support on VLANs. Intentionally separated IFCAP_VLAN_HWTSO
from IFCAP_VLAN_HWTAGGING. I think some hardwares may be able to
TSO over VLAN without VLAN hardware tagging.
Driver changes and userland support will follow.

Reviewed by: thompsa

14 years agoSet curvnet earlier so that it also covers calls to sodisconnect(), which
Bjoern A. Zeeb [Sat, 20 Feb 2010 22:29:28 +0000 (22:29 +0000)]
Set curvnet earlier so that it also covers calls to sodisconnect(), which
before were possibly panicing the system in ULP code in the VIMAGE case.

Submitted by: Igor (igor ispsystem.com)
MFC after: 5 days

14 years agoCorrect inversed programming of ethernet hardware address on
Pyun YongHyeon [Sat, 20 Feb 2010 22:24:24 +0000 (22:24 +0000)]
Correct inversed programming of ethernet hardware address on
big-endian architecture.

Submitted by: C. Jayachandran <c.jayachandran at gmail dot com> (initial version)

14 years agoStart to implement ifnet DDB support:
Bjoern A. Zeeb [Sat, 20 Feb 2010 22:09:48 +0000 (22:09 +0000)]
Start to implement ifnet DDB support:
- 'show ifnets' prints a list of ifnet *s per virtual network stack,
- 'show ifnet <struct ifnet *>' prints fields matching the given ifp.

We do not yet print the complete set of fields and might want to
factor this out to an extra if_debug.c file in case this grows
a lot[1]. We may also want to grow 'show ifnet <if_xname>' support[1].

Sponsored by: ISPsystem
Suggested by: rwatson [1]
Reviewed by: rwatson
MFC after: 5 days

14 years agoAdd support for BCM54K2 found in combination with Apple K2 GMAC.
Marius Strobl [Sat, 20 Feb 2010 22:01:24 +0000 (22:01 +0000)]
Add support for BCM54K2 found in combination with Apple K2 GMAC.

Submitted by:   Andreas Tobler
Obtained from:  OpenBSD
MFC after:      1 week

14 years agoUpon virtual network stack teardown properly release the TCP syncache
Bjoern A. Zeeb [Sat, 20 Feb 2010 21:45:04 +0000 (21:45 +0000)]
Upon virtual network stack teardown properly release the TCP syncache
resources.

Sponsored by: ISPsystem
Reviewed by: rwatson
MFC After: 5 days

14 years agoEnhance a panic string to contain more useful debugging information.
Bjoern A. Zeeb [Sat, 20 Feb 2010 21:43:36 +0000 (21:43 +0000)]
Enhance a panic string to contain more useful debugging information.

Sponsored by: ISPsystem
Reviewed by: rwatson
MFC after: 5 days

14 years agoFix handling of SHUTDOWN-ACK chunk in COOKIE_WAIT and COOKIE_ECHOED.
Michael Tuexen [Sat, 20 Feb 2010 20:30:40 +0000 (20:30 +0000)]
Fix handling of SHUTDOWN-ACK chunk in COOKIE_WAIT and COOKIE_ECHOED.

MFC after: 1 week

14 years agoSplit up ip_drain() into an outer lock and iterator part and
Bjoern A. Zeeb [Sat, 20 Feb 2010 19:59:52 +0000 (19:59 +0000)]
Split up ip_drain() into an outer lock and iterator part and
a "locked" version that will only handle a single network stack
instance. The latter is called directly from ip_destroy().

Hook up an ip_destroy() function to release resources from the
legacy IP network layer upon virtual network stack teardown.

Sponsored by: ISPsystem
Reviewed by: rwatson
MFC After: 5 days

14 years agoEliminate an unused declaration.
Alan Cox [Sat, 20 Feb 2010 19:30:04 +0000 (19:30 +0000)]
Eliminate an unused declaration.

14 years agoThese contain JC's patch to get gdb sort of working
Randall Stewart [Sat, 20 Feb 2010 17:29:27 +0000 (17:29 +0000)]
These contain JC's patch to get gdb sort of working
on mips. Its not fully done yet but its a start.

Obtained from: JC - c.jayachandran@gmail.com

M    gnu/usr.bin/gdb/kgdb/trgt_mips.c
M    gnu/usr.bin/gdb/arch/mips/init.c
M    gnu/usr.bin/gdb/arch/mips/Makefile
M    gnu/usr.bin/Makefile
M    contrib/gdb/gdb/mips-tdep.h

14 years agoFix for the rge driver for boards without rge6 and rge7.
Randall Stewart [Sat, 20 Feb 2010 17:24:33 +0000 (17:24 +0000)]
Fix for the rge driver for boards without rge6 and rge7.
- changes to avoid adding theses instances on specific chips
Obtained from: C. Jayachandran - JC - c.jayachandran@gmail.com

14 years agoChanges for pci and pci-e support
Randall Stewart [Sat, 20 Feb 2010 17:19:16 +0000 (17:19 +0000)]
Changes for pci and pci-e support
- add bus_space_rmi_pci.c for PCI bus space
- files.xlr update for changes in files
- pcibus.c merged into xlr_pci.c (they were small files with inter-dependencies)
- xlr_pci.c - lot of changes here with few fixes, formatting cleanup
Obtained from: C. Jayachandran (JC) - c.jayachandran@gmail.com

14 years agoOpps forgot to add this:
Randall Stewart [Sat, 20 Feb 2010 17:12:07 +0000 (17:12 +0000)]
Opps forgot to add this:
- add bus_space_rmi_pci.c for PCI bus space

Obtained from: C. Jayachandran -  <c.jayachandran@gmail.com>

14 years agoCleanups for sys/mips/rmi/bus_space_rmi.c
Randall Stewart [Sat, 20 Feb 2010 16:32:33 +0000 (16:32 +0000)]
Cleanups for sys/mips/rmi/bus_space_rmi.c
- remove pci related code from bus_space_rmi.c, we will have another
file for PCI bus space functions which will do byte-swapping.
- remove local SWAP implementation
- added TODO stub for unimplemented functions

Obtained from: C. Jayachandran - c.jayachandran@gmail.com

14 years agoSome fixes to the current RMI interrupt handling, changes in this patch are:
Randall Stewart [Sat, 20 Feb 2010 16:30:29 +0000 (16:30 +0000)]
Some fixes to the current RMI interrupt handling, changes in this patch are:
- (cleanup) remove rmi specific 'struct mips_intrhand' - this is no
longer needed since 'struct intr_event' have all the required hooks
- add xlr_cpu_establish_hardintr, which has args for pre/post ithread
and filter hooks, so that the PCI code can add the PCI controller
interrupt ack code here
- make 'cpu_establish_hardintr' use the above function.
- (fix) change type of eirr/eimr from register_t to uint64_t. These
have to be 64bit otherwise we cannot handle interrupts from 32.
- (fix) use eimr to mask eirr before checking interrupts, so that we
will not handle masked interrupts.

Obtained from:  C. Jayachandran - c.jayachandran@gmail.com

14 years agoEnable NETIF_OPEN_CLOSE_ONCE on PowerPC OFW. This fixes netbooting on
Nathan Whitehorn [Sat, 20 Feb 2010 16:28:37 +0000 (16:28 +0000)]
Enable NETIF_OPEN_CLOSE_ONCE on PowerPC OFW. This fixes netbooting on
PowerPC Book-S hardware, which had been broken for a very long time.

Submitted by: Andreas Tobler
MFC after: 1 week