freebsd.git
9 years agoUpdate mandoc to cvs snaphot from 20150302
bapt [Mon, 2 Mar 2015 16:58:57 +0000 (16:58 +0000)]
Update mandoc to cvs snaphot from 20150302

Use the new unsupp warning level to detect the unsupported manpages in man(1)

9 years agoRemove WIP code
bapt [Mon, 2 Mar 2015 16:48:00 +0000 (16:48 +0000)]
Remove WIP code

9 years agoImport CVS snapshot of mandoc as of 20150302
bapt [Mon, 2 Mar 2015 16:45:41 +0000 (16:45 +0000)]
Import CVS snapshot of mandoc as of 20150302

9 years agoInstall old texinfo pages in the form of mdoc(7) pages
bapt [Mon, 2 Mar 2015 11:49:01 +0000 (11:49 +0000)]
Install old texinfo pages in the form of mdoc(7) pages

9 years agoConvert texinfo to mdoc(7) using texi2mdoc
bapt [Mon, 2 Mar 2015 11:48:00 +0000 (11:48 +0000)]
Convert texinfo to mdoc(7) using texi2mdoc

9 years agoAdd support for the AP135 2.0 reference platform.
adrian [Mon, 2 Mar 2015 02:27:25 +0000 (02:27 +0000)]
Add support for the AP135 2.0 reference platform.

This is a QCA9558 SoC (2ghz 3x3) with an atheros 11ac PCIe 5GHz 3x3
NIC and an AR8327 gigabit ethernet switch.

TODO:

* The AR8327 gigabit switch support bugfixes are forthcoming.
* 11ac support and 11ac NIC support

9 years agoBring over the initial QCA955x SoC support framework.
adrian [Mon, 2 Mar 2015 02:24:46 +0000 (02:24 +0000)]
Bring over the initial QCA955x SoC support framework.

This is enough to bring up the basic SoC support.

What works thus far:

* The mips74k core, pll setup, and UART (or else well, stuff would
  be really difficult..)
* both USB 2.0 EHCI controllers
* on-board 2GHz 3x3 wifi (the other variant has 2GHz/5GHz wifi on-chip);
* arge0 - not yet sure why arge1 isn't firing off interrupts and thus
  handling traffic, but I will soon figure it out and fix it here.

Tested:

* AP135 reference design, QCA9558 SoC, pretending to be an 11n
  2GHz AP.

TODO:

* There's an interrupt mux hooking up devices to IP2 and IP3 - but it's
  not a read-and-clear or write-to-clear register.  So, trying to use it
  naively like I have been ends up with massive interrupt storms.
  For now the things that share those interrupts can just take them as
  shared interrupts and try to play nice.

* There's two PCIe root complexes /and/ one of them can actually be
  a PCIe device endpoint.  Yes, you heard right.  I have to teach the
  AR724x PCIe bridge code to handle multiple instances with multiple
  memory/irq regions, and then there'll be RC support, but EP support
  isn't on my TODO list.

* I'm not sure why arge1 isn't up and running.  I'll go figure that
  out soon and fix it here.

Thankyou to Qualcomm Atheros for providing me with hardware and
an abundance of documentation about these things.

9 years agoLay some groundwork for having this stuff hang off of AHB rather than
adrian [Mon, 2 Mar 2015 02:14:44 +0000 (02:14 +0000)]
Lay some groundwork for having this stuff hang off of AHB rather than
the CPU nexus.

* Add ahb as a possible bus attachment
* Lay a comment down to remind me or whoever else ends up trying
  to debug why the EEPROM isn't mapped in as to what's going on.

9 years ago[QCA955x] make the USB EHCI interrupts shareable.
adrian [Mon, 2 Mar 2015 02:08:43 +0000 (02:08 +0000)]
[QCA955x] make the USB EHCI interrupts shareable.

There's two EHCI controllers in the QCA955x SoCs - they have different
interrupts available via various demux registers, but they both tie to
IP3.

So for now, allow them to be sharable so they can hang off of IP3.

9 years agoAdd initial QCA955x support to if_arge.c.
adrian [Mon, 2 Mar 2015 01:53:47 +0000 (01:53 +0000)]
Add initial QCA955x support to if_arge.c.

Tested:

* AP135 development board, QCA9558 SoC.

9 years agoAdd a MII mode for SGMII.
adrian [Mon, 2 Mar 2015 01:23:59 +0000 (01:23 +0000)]
Add a MII mode for SGMII.

This appears on the AR934x and later chips, although it's not
something that's programmed via the arge0/arge1 register space.
It's just cosmetic.

9 years agosh: Fix more compiler warnings.
jilles [Sun, 1 Mar 2015 22:32:23 +0000 (22:32 +0000)]
sh: Fix more compiler warnings.

9 years agoMissed local diff.
nwhitehorn [Sun, 1 Mar 2015 21:47:38 +0000 (21:47 +0000)]
Missed local diff.

9 years agosh: Fix compiler warnings related to duplicate or missing declarations.
jilles [Sun, 1 Mar 2015 21:46:55 +0000 (21:46 +0000)]
sh: Fix compiler warnings related to duplicate or missing declarations.

9 years agonandfs_meta_bread() calls bread() which can set bp to NULL in some
imp [Sun, 1 Mar 2015 21:41:37 +0000 (21:41 +0000)]
nandfs_meta_bread() calls bread() which can set bp to NULL in some
error cases. Calling brelse() with a NULL pointer is not allowed,
so only call brelse() when the bp is non-NULL.

Reported by: Maxime Villard (reported as uninitialized variable)

9 years agoDon't leak 'used' in a few error cases.
imp [Sun, 1 Mar 2015 21:41:35 +0000 (21:41 +0000)]
Don't leak 'used' in a few error cases.

Reported by: Maxime Villard

9 years agoUnlock the main lock before returning rather than after to eliminate
imp [Sun, 1 Mar 2015 21:41:33 +0000 (21:41 +0000)]
Unlock the main lock before returning rather than after to eliminate
dead code that shouldn't have been dead.

Reported by: Maxime Villard

9 years agoInitialize NX stack capabilities and direct map status in pmap like on AIM.
nwhitehorn [Sun, 1 Mar 2015 21:23:23 +0000 (21:23 +0000)]
Initialize NX stack capabilities and direct map status in pmap like on AIM.

9 years agoMerge r278429 from ppc64:
nwhitehorn [Sun, 1 Mar 2015 21:20:18 +0000 (21:20 +0000)]
Merge r278429 from ppc64:

Fix an extremely subtle concurrency bug triggered by running on 32-thread
POWER8 systems. During thread switch, there was a very small window when
the stack pointer was set to the stack pointer of the outgoing thread, but
after the lock on that thread had already been released.

If, during that window, the outgoing thread were rescheduled on another CPU
and begin execution and an exception were taken on the original CPU, the
trap handler and the outgoing thread would simultaneously execute on the same
stack, causing memory corruption. Fix this by making sure to release the
old thread only after cpu_switch() is done with its stack.

MFC after: 2 weeks

9 years agoRecord the dependency to x86bios in vga_pci
dumbbell [Sun, 1 Mar 2015 20:54:29 +0000 (20:54 +0000)]
Record the dependency to x86bios in vga_pci

This fixes the build of XEN and XBOX kernels on i386, which was broken
in r279487.

While here, do not build vga_pci_repost() on PC98.

Reported by: bz@

9 years agoGive compilers a stronger hint to inline the functions
kargl [Sun, 1 Mar 2015 20:32:47 +0000 (20:32 +0000)]
Give compilers a stronger hint to inline the functions
pzero[f], qzero[f], pone[f], and qone[f].  While here
fix the function declarations in accordance with style(9).

9 years agoAR8327: Disable energy-efficient ethernet support in the PHYs.
adrian [Sun, 1 Mar 2015 20:32:35 +0000 (20:32 +0000)]
AR8327: Disable energy-efficient ethernet support in the PHYs.

I noticed that openwrt/linux does this, citing "instability", so
until they figure out why I'm going to disable it here as well.

Tested:

* QCA AP135 - QCA955x SoC + AR8327 switch.

9 years agoWhen j0() and j1() were converted to j0f() and j1f(), the threshold
kargl [Sun, 1 Mar 2015 20:26:03 +0000 (20:26 +0000)]
When j0() and j1() were converted to j0f() and j1f(), the threshold
values for the different invervals were not converted correctly.
Adjust the threshold values to values, which should agree with the
comments.

Reported by: cognet (j1f only)
Discussed with: pfg, bde
Reviewed by: bde

9 years agoBump the port mask on the AR8327 ethernet switch from 0x3f to 0x7f.
adrian [Sun, 1 Mar 2015 20:22:28 +0000 (20:22 +0000)]
Bump the port mask on the AR8327 ethernet switch from 0x3f to 0x7f.

So, it turns out that the AR8327 has 7 ports internally:

* GMAC0 / external (CPU) MAC0
* GMAC1 / port1 -> GMAC5 / port5: external switch port PHYs
* GMAC6 / external (CPU) MAC1

Now, depending upon how things are wired up, the second CPU port (MAC1)
can be wired to either the switch (port6), or through port5's PHY, bypassing
the GMAC+switch entirely.  Ie, it can pretend to be a boring PHY, saving
system designers from having to include a separate PHY for a "WAN" port.

Here's the rub - the AP135 board (QCA955x SoC) hooks up arge0 to
the second CPU port on the AR8327, but it's hooked up as RGMII.
So, in order to hook it up to the rest of the switch, it isn't configured
as a separate PHY - OpenWRT has it setup as connected via RGMII to
GMAC6 and (I'm guessing) it's set to be a WAN port by configuring up
port-based VLANs or something.

Thus, with a port mask of 0x3f, GMAC6 was never allowed to receive traffic
from any other port.  It could transmit fine, but not receive anything.

So, now it works enough for me to continue doing board bootstrapping.
Note, this isn't enough to make the QCA955x + AR8327 work - there's
a bunch of uncommitted work to both the platform SoC (interrupt handling,
ethernet, etc) and the ethernet switch (register access space, setup, etc)
that needs to happen.  However, this particular change is also relevant to
other SoCs, like the AR934x and AR7161, both of which can be glued to
this switch.

Tested:

* AP135 development board

TODO:

* Figure out whether I can somehow abuse another port mode to have this
  be a pass-through PHY, or whether I should just create some more boot
  time hints to explicitly set up port-based isolation so this works
  in a more useful way by default.

9 years agoMake the "automounted" flag work for FUSE filesystems.
trasz [Sun, 1 Mar 2015 18:26:26 +0000 (18:26 +0000)]
Make the "automounted" flag work for FUSE filesystems.

PR: 192852
Submitted by: taku at tackymt.homeip.net (earlier version)
MFC after: 1 month
Sponsored by: The FreeBSD Foundation

9 years agovt(4): Add support to "downgrade" from eg. vt_fb to vt_vga
dumbbell [Sun, 1 Mar 2015 12:54:22 +0000 (12:54 +0000)]
vt(4): Add support to "downgrade" from eg. vt_fb to vt_vga

The main purpose of this feature is to be able to unload a KMS driver.

When going back from the current vt(4) backend to the previous backend,
the previous backend is reinitialized with the special VDF_DOWNGRADE
flag set. Then the current driver is terminated with the new "vd_fini"
callback.

In the case of vt_fb and vt_vga, this allows the former to pass the
vgapci device vt_fb used to vt_vga so the device can be rePOSTed.

Differential Revision: https://reviews.freebsd.org/D687

9 years agovgapci: New vga_pci_repost() function
dumbbell [Sun, 1 Mar 2015 12:47:36 +0000 (12:47 +0000)]
vgapci: New vga_pci_repost() function

This can be used to restore the VGA mode after a KMS driver is unloaded.

Differential Revision: https://reviews.freebsd.org/D687

9 years agoFix the dtrace ARM atomic compare-and-set functions. These functions are
andrew [Sun, 1 Mar 2015 10:04:14 +0000 (10:04 +0000)]
Fix the dtrace ARM atomic compare-and-set functions. These functions are
expected to return the data in the memory location pointed at by target
after the operation. The FreeBSD atomic functions previously used return
either 0 or 1 to indicate if the comparison succeeded or not respectively.

With this change these functions only support ARMv6 and later are supported
by these functions.

Sponsored by: ABT Systems Ltd

9 years agoAdd very initial QCA955x awareness to the GPIO code.
adrian [Sun, 1 Mar 2015 07:00:34 +0000 (07:00 +0000)]
Add very initial QCA955x awareness to the GPIO code.

There's a lot more to come - the QCA955x has a bunch more GPIO MUX
configuration, reminiscent of what the ARM chips let you do - but
it'll have to come later.

9 years agoFlesh out some more QCA955x ethernet PLL setup.
adrian [Sun, 1 Mar 2015 06:59:32 +0000 (06:59 +0000)]
Flesh out some more QCA955x ethernet PLL setup.

9 years agoAdd Ethernet PLL values for the QCA955x.
adrian [Sun, 1 Mar 2015 06:54:59 +0000 (06:54 +0000)]
Add Ethernet PLL values for the QCA955x.

These are the same as the AR934x.

Obtained from: Linux openwrt

9 years agoMake QCA955X_GMAC_REG_ETH_CFG defined like most other registers like this.
adrian [Sun, 1 Mar 2015 06:52:23 +0000 (06:52 +0000)]
Make QCA955X_GMAC_REG_ETH_CFG defined like most other registers like this.

9 years agoAdd QCA955x support to the EHCI setup path.
adrian [Sun, 1 Mar 2015 06:05:01 +0000 (06:05 +0000)]
Add QCA955x support to the EHCI setup path.

Tested:

* QCA AP135 development board, USB rootfs.

9 years agoUse RW_NEW rather than calling bzero().
alc [Sun, 1 Mar 2015 05:18:02 +0000 (05:18 +0000)]
Use RW_NEW rather than calling bzero().

9 years agoCorrect a typo.
rstone [Sun, 1 Mar 2015 02:45:46 +0000 (02:45 +0000)]
Correct a typo.

X-MFC-With: r279458

9 years agoFree up the IPI slot used by IPI_STOP_HARD.
neel [Sun, 1 Mar 2015 02:31:27 +0000 (02:31 +0000)]
Free up the IPI slot used by IPI_STOP_HARD.

Change the numeric value of IPI_STOP_HARD so it doesn't occupy a valid IPI
slot. This can be done because IPI_STOP_HARD is actually delivered via NMI.

Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D1983

9 years agoTeach pciconf how to dump out SR-IOV capability
rstone [Sun, 1 Mar 2015 00:59:35 +0000 (00:59 +0000)]
Teach pciconf how to dump out SR-IOV capability

Differential Revision: https://reviews.freebsd.org/D1639
Reviewed by: jhb
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoValidate the schema that the PF driver passed to us
rstone [Sun, 1 Mar 2015 00:59:28 +0000 (00:59 +0000)]
Validate the schema that the PF driver passed to us

Differential Revision: https://reviews.freebsd.org/D90
Reviewed by: emaste
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoDocument the interface for defining a configuration schema
rstone [Sun, 1 Mar 2015 00:59:21 +0000 (00:59 +0000)]
Document the interface for defining a configuration schema

Differential Revision: https://reviews.freebsd.org/D89
Reviewed by: wblock, emaste, allanjude
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd an rc.d script to invoke iovctl(8) during boot
rstone [Sun, 1 Mar 2015 00:58:23 +0000 (00:58 +0000)]
Add an rc.d script to invoke iovctl(8) during boot

Differential Revision: https://reviews.freebsd.org/D88
Reviewed by: wblock, emaste, allanjude
MFC after: 1 month
Relnotes: yes
Sponsored by: Sandvine Inc.

9 years agoAdd main() for iovctl and hook iovctl into build
rstone [Sun, 1 Mar 2015 00:52:41 +0000 (00:52 +0000)]
Add main() for iovctl and hook iovctl into build

Differential Revision: https://reviews.freebsd.org/D87
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd functions for parsing the iovctl config file
rstone [Sun, 1 Mar 2015 00:52:34 +0000 (00:52 +0000)]
Add functions for parsing the iovctl config file

Add two functions for parsing the iovctl config file.  The config
file is parsed using libucl[1], which accepts most YAML files and
a superset of JSON.  The first function is an ad-hoc parser that
searches the file for the PF.DEVICE configuration value.  We need
to know that value in order to fetch the schema from the kernel,
and we need the schema in order to be able to fully parse the file.

The second function parses the config file and validates it
against a schema.  This function will exit with an error message
if any validation error occurs.  If it succeeds, the configuration
is returned as an nvlist suitable for passing to the kernel.

[1] https://github.com/vstakhov/libucl

Differential Revision: https://reviews.freebsd.org/D86
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd iovctl functions for validating config
rstone [Sun, 1 Mar 2015 00:52:28 +0000 (00:52 +0000)]
Add iovctl functions for validating config

Add an function to iovctl that validates the configuration against
a schema.  This function is able to assume that the parser has
done most of the validation already and it's only responsible for
applying default VF values specified in the config file, confirming
that all required parameters have been set and that no invalid VF
numbers have been specified.

Differential Revision: https://reviews.freebsd.org/D85
Reviewed by: bcr
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd manpage documenting iovctl config file format.
rstone [Sun, 1 Mar 2015 00:52:21 +0000 (00:52 +0000)]
Add manpage documenting iovctl config file format.

Differential Revision: https://reviews.freebsd.org/D84
Reviewed by: emaste, bcr, wblock
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd a manpage for iovctl(8)
rstone [Sun, 1 Mar 2015 00:52:15 +0000 (00:52 +0000)]
Add a manpage for iovctl(8)

Differential Revision: https://reviews.freebsd.org/D83
Reviewed by: wblock
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoRevert r279454. The new directory didn't get added to svn properly.
rstone [Sun, 1 Mar 2015 00:44:15 +0000 (00:44 +0000)]
Revert r279454.  The new directory didn't get added to svn properly.

Pointy hat to: rstone

9 years agoAdd main() for iovctl and hook iovctl into build
rstone [Sun, 1 Mar 2015 00:41:17 +0000 (00:41 +0000)]
Add main() for iovctl and hook iovctl into build

Differential Revision: https://reviews.freebsd.org/D87
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoPass SR-IOV configuration to kernel using an nvlist
rstone [Sun, 1 Mar 2015 00:40:57 +0000 (00:40 +0000)]
Pass SR-IOV configuration to kernel using an nvlist

Pass all SR-IOV configuration to the kernel using an nvlist.  The
main benefit that this offers is flexibility.  It allows a driver
to accept any number of parameters of any type supported by the
SR-IOV configuration infrastructure with having to make any
changes outside of the driver.

It also offers the user very fine-grained control over the
configuration of the VFs -- if they want, they can have different
configuration applied to every VF.

Differential Revision: https://reviews.freebsd.org/D82
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd function to validate the consistency of SR-IOV config
rstone [Sun, 1 Mar 2015 00:40:51 +0000 (00:40 +0000)]
Add function to validate the consistency of SR-IOV config

Add a function that validates that the user-provided SR-IOV
configuration is valid.  This includes basic checks that the
structure of the configuration is correct (e.g. all required
configuration nodes are present) as well as validating against
a configuration schema.

The schema validation consists of:
 - Ensuring that all required config parameters are present.
 - If the schema defines a default value for a parameter,
   adding the default value if the parameter is not set.
 - Ensuring that no parameters are specified in the config
   that are not defined in the schema.
 - Ensuring that have the correct type defined in the schema.
 - Ensuring that no configuration nodes are present for devices
   that do not exist.  For example, if 2 VFs are configured,
   then we validate that a node called VF-5 does not exist.

Differential Revision: https://reviews.freebsd.org/D81
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd infrastructure for exporting config schema from PF drivers
rstone [Sun, 1 Mar 2015 00:40:42 +0000 (00:40 +0000)]
Add infrastructure for exporting config schema from PF drivers

Differential Revision: https://reviews.freebsd.org/D80
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd interface to destroy SR-IOV VFs
rstone [Sun, 1 Mar 2015 00:40:34 +0000 (00:40 +0000)]
Add interface to destroy SR-IOV VFs

Differential Revision: https://reviews.freebsd.org/D79
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAllocate PCI I/O memory spaces for VFs
rstone [Sun, 1 Mar 2015 00:40:26 +0000 (00:40 +0000)]
Allocate PCI I/O memory spaces for VFs

When creating VFs, we must size each SR-IOV BAR on the PF and
allocate a configuous I/O memory window large enough for every VF.
However, the window only needs to be aligned to a boundary equal
to the size of the window for a single VF.

When a VF attempts to allocate an I/O memory resource, we must
intercept the request in the pci driver and pass it off to the
SR-IOV code, which will allocate the correct window from the
pre-allocated memory space for the PF.

Inform the pci driver about the size and address of the BARs on
the VF when the VF is created.  This is required by pciconf -b and
bhyve.

Differential Revision: https://reviews.freebsd.org/D78
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoEmulate the Device ID and Vendor ID registers for VFs
rstone [Sun, 1 Mar 2015 00:40:19 +0000 (00:40 +0000)]
Emulate the Device ID and Vendor ID registers for VFs

The SR-IOV standard requires VFs to read all-ones when the VID
and DID registers are read.  The VMM (hypervisor) is required to
emulate them instead.  Make pci_read_config() do this emulation.

Change pci_user.c to use pci_read_config() to read config space
registers instead of going directly to the pcib so that the
emulated VID/DID registers work correctly on VFs.  This is
required both for pciconf and bhyve PCI passthrough.

Differential Revision: https://reviews.freebsd.org/D77
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoImplement interface to create SR-IOV Virtual Functions
rstone [Sun, 1 Mar 2015 00:40:09 +0000 (00:40 +0000)]
Implement interface to create SR-IOV Virtual Functions

Implement the interace to create SR-IOV Virtual Functions (VFs).
When a driver registers that they support SR-IOV by calling
pci_setup_iov(), the SR-IOV code creates a new node in /dev/iov
for that device.  An ioctl can be invoked on that device to
create VFs and have the driver initialize them.

At this point, allocating memory I/O windows (BARs) is not
supported.

Differential Revision: https://reviews.freebsd.org/D76
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd manpages for SR-IOV enable/disable driver interface
rstone [Sun, 1 Mar 2015 00:40:01 +0000 (00:40 +0000)]
Add manpages for SR-IOV enable/disable driver interface

Differential Revision: https://reviews.freebsd.org/D75
Reviewed by: wblock
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoDocument pci_iov_attach/detach in pci.9
rstone [Sun, 1 Mar 2015 00:39:55 +0000 (00:39 +0000)]
Document pci_iov_attach/detach in pci.9

Differential Revision: https://reviews.freebsd.org/D74
Reviewed by: bcr, wblock, emaste
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAllow passthrough devices to be hinted.
rstone [Sun, 1 Mar 2015 00:39:48 +0000 (00:39 +0000)]
Allow passthrough devices to be hinted.

Allow the ppt driver to attach to devices that were hinted to be
passthrough devices by the PCI code creating them with a driver
name of "ppt".

Add a tunable that allows the IOMMU to be forced to be used.  With
SR-IOV passthrough devices the VFs may be created after vmm.ko is
loaded.  The current code will not initialize the IOMMU in that
case, meaning that the passthrough devices can't actually be used.

Differential Revision: https://reviews.freebsd.org/D73
Reviewed by: neel
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoAdd some pcib methods to get ARI-related information
rstone [Sun, 1 Mar 2015 00:39:40 +0000 (00:39 +0000)]
Add some pcib methods to get ARI-related information

Differential Revision: https://reviews.freebsd.org/D72
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoRefactor PCI resource allocation
rstone [Sun, 1 Mar 2015 00:39:33 +0000 (00:39 +0000)]
Refactor PCI resource allocation

Refactor PCI resource allocation code to allow a request for a
memory-mapped I/O window that is a multiple of a requested size.
This is needed by the SR-IOV code because the VF BARs are all
allocated contiguously.  We can't just allocate a resource that is
a multiple of a single VF BAR because the size of an allocation
implies its alignment requirement.

Differential Revision: https://reviews.freebsd.org/D71
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoRefactor PCI device creation
rstone [Sun, 1 Mar 2015 00:39:26 +0000 (00:39 +0000)]
Refactor PCI device creation

Refactor creation of PCI devices into helper methods that can be
used by the VF creation code.

Differential Revision: https://reviews.freebsd.org/D67
Reviewed by: jhb
MFC after:  1 month
Sponsored by: Sandvine Inc.

9 years agoFix build of nv_tests.cc
rstone [Sun, 1 Mar 2015 00:37:23 +0000 (00:37 +0000)]
Fix build of nv_tests.cc

nv_tests.cc managed to get two copies of several functions due to me
applying a patch in an unclean working tree.  My kingdom for an
"svn clean" command.

MFC after: 1 month
X-MFC-With: r279424

9 years agoMove libnv into the kernel and hook it into the kernel build
rstone [Sun, 1 Mar 2015 00:34:27 +0000 (00:34 +0000)]
Move libnv into the kernel and hook it into the kernel build

Differential Revision: https://reviews.freebsd.org/D1883
Reviewed by: jfv
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd macros to make code compile in kernel
rstone [Sun, 1 Mar 2015 00:22:53 +0000 (00:22 +0000)]
Add macros to make code compile in kernel

Make it possible to compile libnv in the kernel.  Mostly this
involves wrapping functions that have a different signature in
the kernel and in userland (e.g. malloc()) in a macro that will
conditionally expand to the right API depending on whether the
code is being compiled for the kernel or not.

I have also #ifdef'ed out all of file descriptor-handling code,
as well as the unsafe varargs functions.

Differential Revision: https://reviews.freebsd.org/D1882
Reviewed by: jfv
MFC after: 1 month
Sponsored by: Sandvine Inc

9 years agoAllow Illumos code to co-exist with nv(9)
rstone [Sun, 1 Mar 2015 00:22:45 +0000 (00:22 +0000)]
Allow Illumos code to co-exist with nv(9)

Differential Revision: https://reviews.freebsd.org/D1881
Reviewed by: jfv, will
Suggested by: pjd
MFC after: 1 month
Sponsored by: Sandvine Inc

9 years agoPrevent creation of an invalid nvlist
rstone [Sun, 1 Mar 2015 00:22:38 +0000 (00:22 +0000)]
Prevent creation of an invalid nvlist

If an nvlist is set as a child of another nvlist with
nvlist_move_nvlist then fail the operation and set the parent
nvlist to the error state.

Differential Revision: https://reviews.freebsd.org/D1880
Reviewers: jfv
MFC after: 1 month
Sponsored by: Sandvine Inc

9 years agoDon't allocate memory for operations that do not insert
rstone [Sun, 1 Mar 2015 00:22:31 +0000 (00:22 +0000)]
Don't allocate memory for operations that do not insert

Almost every operation performed on an nvlist was allocating a
new string to hold the key name.  The nvlist_exists* family of
functions would always return false if they failed to allocate
the string.  The rest of the functions would outright abort().
Fix the non-varargs variants of the functions to perform the
requested operations directly and the varargs versions to
allocate the string and call into the non-varargs versions.
The varargs versions are still broken and really can't be fixed,
so we might consider axing them entirely.  However, now the non-
varargs functions are always safe to call.

Differential Revision: https://reviews.freebsd.org/D1879
Reviewed by: pjd, jfv
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd function to force an nvlist into the error state
rstone [Sun, 1 Mar 2015 00:22:23 +0000 (00:22 +0000)]
Add function to force an nvlist into the error state

Add an nvlist_set_error() function that can be used to force an
nvlist into the error state.  This is useful both for writing
tests and for writing APIs that use nvlists internally.

Differential Revision: https://reviews.freebsd.org/D1878
Reviewed by: pjd, jfv
MFC After: 1 month
Sponsored by: Sandvine Inc.

9 years agoImplement asprintf in libkern
rstone [Sun, 1 Mar 2015 00:22:16 +0000 (00:22 +0000)]
Implement asprintf in libkern

Differential Revision: https://reviews.freebsd.org/D1877
Reviewed by: pjd, jfv
MFC After: 1 month
Sponsored by: Sandvine Inc.

9 years agoExtend the unit test to fix the bug caught in r277925
rstone [Sun, 1 Mar 2015 00:22:09 +0000 (00:22 +0000)]
Extend the unit test to fix the bug caught in r277925

Differential Revision: https://reviews.freebsd.org/D1888
MFC After: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd tests for dnvlist_take_*
rstone [Sun, 1 Mar 2015 00:22:03 +0000 (00:22 +0000)]
Add tests for dnvlist_take_*

Differential Revision: https://reviews.freebsd.org/D1876
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd tests for dnv_get_*
rstone [Sun, 1 Mar 2015 00:21:56 +0000 (00:21 +0000)]
Add tests for dnv_get_*

Differential Revision: https://reviews.freebsd.org/D1875
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd tests for nvlist_free* functions
rstone [Sun, 1 Mar 2015 00:21:50 +0000 (00:21 +0000)]
Add tests for nvlist_free* functions

Differential Revision: https://reviews.freebsd.org/D1874
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd tests for nvlist_take_*
rstone [Sun, 1 Mar 2015 00:21:43 +0000 (00:21 +0000)]
Add tests for nvlist_take_*

Differential Revision: https://reviews.freebsd.org/D1873
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd test cases for nvlist_move_*
rstone [Sun, 1 Mar 2015 00:21:37 +0000 (00:21 +0000)]
Add test cases for nvlist_move_*

Differential Revision: https://reviews.freebsd.org/D1872
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd tests for nvlist_pack/unpack
rstone [Sun, 1 Mar 2015 00:21:30 +0000 (00:21 +0000)]
Add tests for nvlist_pack/unpack

Differential Revision: https://reviews.freebsd.org/D1871
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd tests for nvlist_clone
rstone [Sun, 1 Mar 2015 00:21:24 +0000 (00:21 +0000)]
Add tests for nvlist_clone

Differential Revision: https://reviews.freebsd.org/D1870
Reviewed by: pjd, jfv
MFC after: 1 month
Sponsored by: Sandvine Inc

9 years agoTests of basic nvlist add functions
rstone [Sun, 1 Mar 2015 00:20:57 +0000 (00:20 +0000)]
Tests of basic nvlist add functions

Differential Revision: https://reviews.freebsd.org/D1869
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoRevert r279422. My "apply patch and commit" script wasn't adding
rstone [Sun, 1 Mar 2015 00:05:45 +0000 (00:05 +0000)]
Revert r279422.  My "apply patch and commit" script wasn't adding
new files properly.

Pointy hat to: rstone

9 years agoTests of basic nvlist add functions
rstone [Sun, 1 Mar 2015 00:01:44 +0000 (00:01 +0000)]
Tests of basic nvlist add functions

Differential Revision: https://reviews.freebsd.org/D1869
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoMake libnv headers includable from C++
rstone [Sun, 1 Mar 2015 00:01:36 +0000 (00:01 +0000)]
Make libnv headers includable from C++

Differential Revision: https://reviews.freebsd.org/D1868
Reviewed by: jfv, pjd
MFC after: 1 month
Sponsored by: Sandvine Inc.

9 years agoAdd another register definition for the AR8327.
adrian [Sat, 28 Feb 2015 23:59:29 +0000 (23:59 +0000)]
Add another register definition for the AR8327.

Obtained from: OpenWRT

9 years agoRemove the old DTrace test suite makefile - it was somewhat primitive and
markj [Sat, 28 Feb 2015 23:35:29 +0000 (23:35 +0000)]
Remove the old DTrace test suite makefile - it was somewhat primitive and
mostly unmaintained, and it has been superseded by the infrastructure added
in r279418.

Reviewed by: ngie
Sponsored by: EMC / Isilon Storage Divison

9 years agoAdd infrastructure to integrate the DTrace test suite with Kyua.
markj [Sat, 28 Feb 2015 23:30:06 +0000 (23:30 +0000)]
Add infrastructure to integrate the DTrace test suite with Kyua.

For each test category, we generate a script containing ATF test cases for
the tests under that category. Each test case simply runs dtest.pl (the
upstream test harness) with the corresponding test files. The exclude.sh
script is used to record info about tests which should be skipped or are
expected to fail; it is used to generate atf_skip and atf_expect_fail calls.
The genmakefiles.sh script can be used to regenerate the test makefiles when
new tests are brought it from upstream.

The test suite is currently not connected to the build as there is a small
number of lingering test issues which still need to be worked out. In the
meantime however, the test suite can be easily built and installed
manually from cddl/usr.sbin/dtrace/tests.

Reviewed by: ngie
Sponsored by: EMC / Isilon Storage Division

9 years agoUse the -shared option to create a shared library.
markj [Sat, 28 Feb 2015 23:14:35 +0000 (23:14 +0000)]
Use the -shared option to create a shared library.

MFC after: 1 week

9 years agoRemove a leading tab that causes a make(1) error when running the test.
markj [Sat, 28 Feb 2015 23:11:18 +0000 (23:11 +0000)]
Remove a leading tab that causes a make(1) error when running the test.

MFC after: 1 week

9 years agoOnly kill sleep processes that were forked from the test script.
markj [Sat, 28 Feb 2015 23:10:26 +0000 (23:10 +0000)]
Only kill sleep processes that were forked from the test script.

MFC after: 1 week

9 years agoStop hard-coding an incorrect path to rm(1).
markj [Sat, 28 Feb 2015 23:09:01 +0000 (23:09 +0000)]
Stop hard-coding an incorrect path to rm(1).

MFC after: 1 week

9 years agoAdd a missing include of an options header.
rstone [Sat, 28 Feb 2015 22:37:48 +0000 (22:37 +0000)]
Add a missing include of an options header.

watchdog.c does an #ifdef DDB but does not #include "opt_ddb.h".
Fixing this turned up a missing include file.

MFC after: 1 week
X-MFC-With: r261495, r279410

9 years agoDo not leak 'copy' buffer if bmap_truncate_indirect fails.
kan [Sat, 28 Feb 2015 22:24:45 +0000 (22:24 +0000)]
Do not leak 'copy' buffer if bmap_truncate_indirect fails.

Reported by: Brainy Code Scanner, by Maxime Villard.
MFC after: 2 weeks

9 years agoUnbreak 'make depend' with sfxge by removing debugging code activated in the
ngie [Sat, 28 Feb 2015 21:50:04 +0000 (21:50 +0000)]
Unbreak 'make depend' with sfxge by removing debugging code activated in the
INET || INET6 case

X-MFC with: r279398
Pointyhat to: arybchik

9 years agoCorrect the use of an unitialized variable in sendfind_getobj()
rstone [Sat, 28 Feb 2015 21:49:59 +0000 (21:49 +0000)]
Correct the use of an unitialized variable in sendfind_getobj()

When sendfile_getobj() is called on a DTYPE_SHM file, it never
initializes error, which is eventually returned to the caller.

Differential Revision: https://reviews.freebsd.org/D1989
Reviewed by: kib
Reported by:  Brainy Code Scanner, by Maxime Villard.

9 years agoAdd ofw_gpiobus_parse_gpios(), a new public function, to parse the gpios
loos [Sat, 28 Feb 2015 21:01:01 +0000 (21:01 +0000)]
Add ofw_gpiobus_parse_gpios(), a new public function, to parse the gpios
property for devices that doesn't descend directly from gpiobus.

The parser supports multiple pins, different GPIO controllers and can use
arbitrary names for the property (to match the many linux variants:
cd-gpios, power-gpios, wp-gpios, etc.).

Pass the driver name on ofw_gpiobus_add_fdt_child().  Update gpioled to
match.

An usage example of ofw_gpiobus_parse_gpios() will follow soon.

9 years agoSupposed fix for some SandyBridge mobile CPUs hang on AP startup when
kib [Sat, 28 Feb 2015 20:37:38 +0000 (20:37 +0000)]
Supposed fix for some SandyBridge mobile CPUs hang on AP startup when
x2APIC mode is detected and enabled.  Current theory is that switching
the APIC mode while an IPI is in flight might be the issue.

Postpone switching to x2APIC mode until we are guaranteed that all
starting IPIs are already send and aknowledged.  Use aps_ready signal
as an indication that the BSP is done with us.

Tested by: adrian
Sponsored by: The FreeBSD Foundation
MFC after: 2 months

9 years agoAvoid closing unallocated socket in case socreate fails.
kan [Sat, 28 Feb 2015 20:30:29 +0000 (20:30 +0000)]
Avoid closing unallocated socket in case socreate fails.

Found by: Brainy Code Scanner
Reported by: Maxime Villard <max@M00nBSD.net>
MFC after: 2 weeks

9 years agoAvoid lookup of CODESET aliases using uninitialized path
kan [Sat, 28 Feb 2015 20:30:25 +0000 (20:30 +0000)]
Avoid lookup of CODESET aliases using uninitialized path

We do not use iconv.alias file, so avoid using the vestiges
of the code that do.

Differential Revision: https://reviews.freebsd.org/D1729
Reviewed by: emaste
MFC after: 2 weeks

9 years agoRename and move gpiobus_alloc_ivars() and gpiobus_free_ivars() so they can
loos [Sat, 28 Feb 2015 20:02:41 +0000 (20:02 +0000)]
Rename and move gpiobus_alloc_ivars() and gpiobus_free_ivars() so they can
be used on non FDT systems.

This prevents access to uninitialized memory on drivers that try to access
pin flags on non FDT systems.

9 years agoSome fixes for fdescfs lookup code.
kib [Sat, 28 Feb 2015 19:57:22 +0000 (19:57 +0000)]
Some fixes for fdescfs lookup code.

Do not ever return doomed vnode from lookup.  This could happen, if
not checked, since dvp is relocked in the 'looking up ourselves' case.

In the other case, since dvp is relocked, mount point might go away
while fdesc_allocvp() is called.  Prevent the situation by doing
vfs_busy() before unlocking dvp.  Reuse the vn_vget_ino_gen() helper.

Reported and tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks

9 years agoEliminate a variable that became unused when VFS_LOCK_GIANT() was
alc [Sat, 28 Feb 2015 19:11:37 +0000 (19:11 +0000)]
Eliminate a variable that became unused when VFS_LOCK_GIANT() was
eliminated.

MFC after: 3 days

9 years agoAdd a driver for the Maxim DS3231 a low-cost, extremely accurate (+-2PPM)
loos [Sat, 28 Feb 2015 19:02:44 +0000 (19:02 +0000)]
Add a driver for the Maxim DS3231 a low-cost, extremely accurate (+-2PPM)
I2C real-time clock (RTC).

The DS3231 has an integrated temperature-compensated crystal oscillator
(TXCO) and crystal.

DS3231 has a temperature sensor, an independent 32kHz output (which can be
turned on and off by the driver) and another output that can be used as
interrupt for alarms or as a second square-wave output, which frequency and
operation mode can be set by driver sysctl(8) knobs.

Differential Revision: https://reviews.freebsd.org/D1016
Reviewed by: ian, rpaulo
Tested on: Raspberry pi model B

9 years agosfxge: compile out LRO if kernel is compiled without IPv4 and IPv6
arybchik [Sat, 28 Feb 2015 19:01:43 +0000 (19:01 +0000)]
sfxge: compile out LRO if kernel is compiled without IPv4 and IPv6

Sponsored by:   Solarflare Communications, Inc.
Approved by:    gnn (mentor)