Sascha Wildner [Mon, 31 Aug 2015 18:01:31 +0000 (20:01 +0200)]
Add a acpicall(8) utility for debugging and tweaking purposes.
It is based on ports' sysutils/acpi_call (from Maxim Ignatenko) with
a few changes by me:
* Rename acpi_call -> acpicall.
* Ioctl handling is in the main acpi.ko module.
* To enable it, the debug.acpi.allow_method_calls tunable needs to
be set.
* In acpi_call, the mandatory -p option was used to pass the method's
namespace path. I removed the option and made the path acpicall(8)'s
argument.
* Wrote a manual page and cleaned up a bit.
The separate acpiio_mcall.h file was added because ACPIIO_DO_MCALL's
argument struct uses ACPICA types, so it needs acpi.h which acpiio.h
(a public header used by some ports) so far didn't need. So to avoid
any hassle, I put the ACPIIO_DO_MCALL ioctl into a separate header
(it's only used by acpicall(8) anyway).
The changes to kdump(1) and truss(1) are to include and build with
acpi.h.
Tested-by: tollens
Matthew Dillon [Mon, 31 Aug 2015 02:55:53 +0000 (19:55 -0700)]
hammer2 - bug fixes
* When a dirty inode is reclaimed meta-data changes stored in the inode
structure could sometimes get lost. These changes must be synchronized
to the chains in the underlying cluster nodes.
Repurpose the unlinkq into a more general 'sideq' that handles both
deletion-on-reclaim and dirty-meta-data-on-reclaim.
Add a flag to the inode to represent when it is on the sideq, preventing
situations where it might be added twice.
* Enable the free->allocated transition in the bulkfree code. Add
additional statistics and an underflow/overflow check for
hammer2_bmap_data->avail. Also adjust the volume free space
in both directions. Do not update any live field if 'nofree' is
set for the chain (~1GB granularity representing possible
frontend/backend races).
Sepherosa Ziehau [Sun, 30 Aug 2015 14:39:47 +0000 (22:39 +0800)]
uipc: Hold per-unp token for all unp usrreq methods
Matthew Dillon [Sun, 30 Aug 2015 21:17:35 +0000 (14:17 -0700)]
hammer2 - Refactor bulkfree
* Change the bulkfree scan from breadth-first to depth-first. This
improves disk performance significantly (~2x) and is also needed for the
duplicate-detection feature.
* Create an 8-way set-associative hash table similar to what the live
dedup code uses. Record the data_off for elements we have processed
and detect if a duplicate is encountered so we do not have to re-process
the duplicate subtree.
Also prioritize the table based on the aggregate bottom-up inode count
to reduce the chance that a top-level duplicate (aka snapshot) will get
kicked out of the hash table.
* Clean up the hammer2_chain_scan() API, making it more bref-centric
which allows us to avoid instantiating chain structures for leaf
entities. This significantly improves performance and increases
flexibility.
* Manual page adjustments for kern.maxvnodes settings suggestions.
John Marino [Sun, 30 Aug 2015 18:59:42 +0000 (20:59 +0200)]
iswdigit(3), iswxdigit(3): Don't limit to U007F
We inherited a valid range of 00 - 127 for checking wide characters for
type digit or xdigit.
This is a mistake. The relationship between digits (iswdigit) and
numbers (iswnumber, iswalnum) is that digits are a subset of numbers.
Digits are also a subset of xdigits (iswxdigit). Digits are limited
to radix-10. Numeric typess, on the other hand, include factions,
Roman Numerals, circled, etc. It should cover the entire unicode
(UTF-8) range.
Unfortunately, DragonFly doesn't support the "number" type yet. The
iswnumber and iswalnum just use the "digit" definition. Note that
before this commit, they used full range for "digit" type while the
iswdigit() and iswdigits() themselves were limited.
John Marino [Sun, 30 Aug 2015 15:25:58 +0000 (17:25 +0200)]
UTF8 locales: Refine Latin supplement more
The multiplication and division sign were missing, and the control
characters were not outlined. Also set superscript 1,2,3 as digits.
There are not showing up with iswdigit() function so that requires
further investigation (iswdigit does work for '0','1',...'9' however)
John Marino [Sun, 30 Aug 2015 13:39:42 +0000 (15:39 +0200)]
UTF8 locales: Include inverted exclamation mark too
I was off by one character when I defined the first range on the previous
commit. It starts with an inverted exclamation mark, not the cent sign.
John Marino [Sun, 30 Aug 2015 13:12:41 +0000 (15:12 +0200)]
UTF8 locales: Complete implemenation of Latin-1 Supplement
The Latin-1 Supplement block of UTF-8 (U0080-U00FF) was not fully
implemented. Specifically it was missing U00A1 (inverted exclamation)
through U00BF (inverted question mark). Some popular characters this
affected was cent sign, pound sign, Yen sign, broken bar, copyright
symbol and superscripts. On international keyboards, AltGR + number
key wouldn't output correctly. This addition to the manual ctype input
definitions (and subsequent regenerations) will fix these issues.
Reported by: profmakx, ivadasz
Diagnostics: YRabbit
Matthew Dillon [Sun, 30 Aug 2015 03:02:14 +0000 (20:02 -0700)]
hammer2 - live dedup, cleanup
* First attempt at a live dedup. The H2 strategy code now caches
{data_off, crc} info to track recently accessed data blocks. The
cache is checked in the strategy_write code after device-level
block encoding. If we get a cache hit, the disk block is compared
against the write data and reused if it matches.
* This 'live' dedup should catch most typical 'cp' or 'cpdup' style
commands. There will also be a bulk dedup capable of catching
everything.
* Note that 'df' output might be a bit confusing because the 'Used'
field represents the topology and does not take into account dedups.
'Avail' is calculated from the actual freemap. To make things look
right the total disk size is adjusted upward so it matches
Used+Avail. This mechanism will likely change.
Here is an example with one copy of /usr/src and 13 copies of /usr/src.
The first copy eats around 872MB, and a 'du' will show each copy eating
about the same. But because of dedup each subsequent copy actually only
eats around 160MB as you can see from the 'Avail' field:
test40# df -h /mnt
Filesystem Size Used Avail Capacity
/dev/serno/WD-WX51A82J2299.s1f@LOCAL 99G 934M 99G 1%
Filesystem Size Used Avail Capacity
/dev/serno/WD-WX51A82J2299.s1f@LOCAL 106G 8.5G 97G 8%
* Rename hammer2_bulkscan.c to hammer2_bulkfree.c since that is
basically all it does.
* Move the synchronization code to its own file, hammer2_synchro.c.
(note: This code is currently in rip-up mode and will not operate
properly).
Sascha Wildner [Sat, 29 Aug 2015 11:32:04 +0000 (13:32 +0200)]
kernel: Use NULL for pointers in DRIVER_MODULE* calls.
Sascha Wildner [Sat, 29 Aug 2015 10:44:53 +0000 (12:44 +0200)]
kernel: Add gnu/ to SUBDIR for MODULES_WITH_WORLD.
I forgot to do that in
b993bb87ded5c733e1f0213efbef1be2660b2536.
Matthew Dillon [Fri, 28 Aug 2015 20:49:47 +0000 (13:49 -0700)]
hammer2 - bulkfree work, rip-up cluster sync.
* bulkfree no longer attempts to flush. Instead it deals with races against
live by refusing to free blocks in L1 freemap chains that have been modified
since the last sync. This is a temporary workaround.
* No longer propagate modify_tid during a flush. modify_tid is now used
as a localized but cluster-aware TID (whereas mirror_tid is only localized
to a cluster node).
* Start work on adding an update_tid to the blockref. This will ultimately
be used by the cluster synchronization code instead of modify_tid.
* Adjust the DESIGN document for the new synchronization concept.
Sascha Wildner [Fri, 28 Aug 2015 13:55:24 +0000 (15:55 +0200)]
kernel: Adjust sys/conf/files for recent drm changes.
Sascha Wildner [Fri, 28 Aug 2015 13:49:48 +0000 (15:49 +0200)]
kernel/cxm: Add pcidevs.h to the Makefile (forgotten in dcb4b80).
Matthew Dillon [Fri, 28 Aug 2015 06:15:17 +0000 (23:15 -0700)]
hammer2 - stabilization, sequencing
* Change the way XOPs are dispatched. Instead of dispatching a XOP
to specific service threads in the xgrp we now queue the XOP to a
per-cluster-node xopq and allow the service threads to compete for
work.
* Implement XOP dependency tracking. Cluster nodes can complete execution
of a XOP in any order and the frontend may complete early on-quorum while
the XOP is still executing on other cluster nodes. In this situation
the inode lock may be released or cycled and another, dependent XOP may
be issued, potentially executing out of order on the cluster nodes that
have not yet finished (or possibly even started) the previous XOP.
With the new queueing mechanism we also implement dependency tracking
using the xop->ip1, ip2, and ip3 fields. Dependent XOPs on (slow) cluster
nodes will be held-back and ordering of dependent XOPS will be enforced.
This fixes one of several cluster-related issues that will need to be
addressed in order to free the frontend of slow backend cluster nodes.
* The strategy XOP now uses a per-XOP lock instead of an xgrp lock
(xgrp is no longer used in the xop structure and XOPs are now
distributed to potentially different xgrps for each cluster node).
* Normalize some structural field names.
* bulkfree now takes a snapshot of vchain, using a separate (temporary)
chain structure which isolates the entire topology scan from the
live filesystem.
This required minor adjustments to base/count handling for
HAMMER2_BREF_TYPE_VOLUME and HAMMER2_BREF_TYPE_FREEMAP.
* Debugging adjustments.
Matthew Dillon [Thu, 27 Aug 2015 06:31:43 +0000 (23:31 -0700)]
hammer2 - hide some debugging kprintfs
* Hide debugging kprintfs related to flushes so we do not default to
feeding console messages back to syslogd.
Sascha Wildner [Thu, 27 Aug 2015 17:09:21 +0000 (19:09 +0200)]
<regex.h>: REG_ENOSYS was removed in Issue 7 but up to 6 it's needed.
Sascha Wildner [Thu, 27 Aug 2015 16:46:08 +0000 (18:46 +0200)]
Remove references to colldef(1) and mklocale(1) in manpages & comments.
François Tigeot [Thu, 27 Aug 2015 18:53:22 +0000 (20:53 +0200)]
drm/i915: Update to Linux 3.17
* With an Haswell stability fix from Linux 3.19
* Display-Port code not completely synced with Linux 3.17, this was
breaking things as noticed by Rimvydas Jasinskas
* Broadwell and Cherryview support improvements
* Runtime power-management improvements
* Panel Self-Refresh (PSR) now enabled by default on Haswell and Broadwell,
leading to further power savings
* Interrupt handling improvements
* backlight brightness locking fixes by Yellow Rabbit
* As usual, all sort of little fixes everywhere
Sascha Wildner [Thu, 27 Aug 2015 16:34:55 +0000 (18:34 +0200)]
Remove colldef(1) manpage too via 'make upgrade'.
François Tigeot [Thu, 27 Aug 2015 06:58:34 +0000 (08:58 +0200)]
drm: Add linux/notifier.h
Obtained-from: FreeBSD/OFED
Matthew Dillon [Thu, 27 Aug 2015 05:45:37 +0000 (22:45 -0700)]
hammer2 - cleanup
* Cleanup some printfs.
Matthew Dillon [Thu, 27 Aug 2015 05:45:21 +0000 (22:45 -0700)]
mount_hammer2 - Allow serno/ relative paths
* Allow paths to omit the /dev.
Matthew Dillon [Thu, 27 Aug 2015 05:21:21 +0000 (22:21 -0700)]
kernel - Add root mount support for hammer2
* Don't try to interpret the device string in mountroot for
hammer2 either (it was already not being interpreted for hammer).
Matthew Dillon [Thu, 27 Aug 2015 05:21:00 +0000 (22:21 -0700)]
hammer2 - Add root mount support.
* Add root mount support to the hammer2 vfs.
Matthew Dillon [Thu, 27 Aug 2015 03:49:18 +0000 (20:49 -0700)]
hammer2 - Require a @label in the device path
* If the user forgets to specify an @label in a hammer2 mount, complain
in a way that does not leave the user confused.
* newfs_hammer2 no longer creates a "BOOT" and "ROOT" PFS by default.
It still always creates the "LOCAL" PFS which we may use later to hold
hammer2 configuration data.
Now hammer2 creates one additional label by default, when no -L option is
specified, based on the partition from the device path:
'a' - newfs_hammer2 will create a "BOOT" PFS
'd' - newfs_hammer2 will create a "ROOT" PFS
* - newfs_hammer2 will create a "DATA" PFS
* When -L is specified to indicate label(s) to create, newfs_hammer2 no longer
creates any default labels other than "LOCAL" (which must always be
created).
* Adjust manual page.
Matthew Dillon [Thu, 27 Aug 2015 03:45:55 +0000 (20:45 -0700)]
hammer2 - Fix hammer2 probe in boot code
* During boot2 the BIOS dskread might not be able to handle a 64KB read
request. Break requests up into 16KB pieces in the hammer2 read code,
allowing DragonFly to boot from a HAMMER2 boot partition.
* Automatically dive the "BOOT" label under the superoot.
Matthew Dillon [Thu, 27 Aug 2015 03:44:26 +0000 (20:44 -0700)]
hammer2 - Add WANT_HAMMER2 to buildworld
* Setting WANT_HAMMER=YES in /etc/make.conf will build the hammer2 pieces.
Matthew Dillon [Thu, 27 Aug 2015 03:37:04 +0000 (20:37 -0700)]
hammer2 - Fix unlink/reclaim panic, add macro for "BOOT" key
* Fix confusion where an inode would erronously be placed on the unlinkq
even though it was not moved to the hidden directory.
* Add HAMMER2_BOOT_KEY for use by libstand.
Matthew Dillon [Thu, 27 Aug 2015 03:32:55 +0000 (20:32 -0700)]
kernel - allow '@' key at mountroot> prompt
* Allow the @ key to generate an '@' at the mountroot<> prompt
so HAMMER2 label specifications can be typed in.
Sepherosa Ziehau [Wed, 26 Aug 2015 12:32:03 +0000 (20:32 +0800)]
uipc: Fix lockless unp_conn accessing and uipc_detach() race.
THREAD1 THREAD2
uipc_send(unp) uipc_detach(unp2)
{ {
lock(unp); unp_free(unp2)
unp2 = unp->unp_conn; {
: /* unp2 ref is 0 */
unp_reference(unp2); unp_detach(unp2); (***)
/* unp2 ref is 1 */ }
: }
unp_free(unp2) :
{ :
/* unp2 ref is 0 */ :
unp_detach(unp2); (***) :
} :
unlock(unp); :
} :
Two calls of unp_detach() on unp2!
To fix this race, we drop all connections before calling unp_free()
on uipc_detach() and uipc_abort() path.
Sepherosa Ziehau [Wed, 26 Aug 2015 12:31:37 +0000 (20:31 +0800)]
uipc: Assert global token is held in unp_connect_pair
Sascha Wildner [Wed, 26 Aug 2015 16:32:01 +0000 (18:32 +0200)]
regex.3: Add LIBRARY section.
Sepherosa Ziehau [Tue, 25 Aug 2015 15:12:42 +0000 (23:12 +0800)]
uipc: Return error from unp_connect_pair() if one of unps is not attached
Sepherosa Ziehau [Tue, 25 Aug 2015 14:06:57 +0000 (22:06 +0800)]
uipc: Don't abuse unp_conn for unconnect DGRAM unix sockets.
Use unp_find_lockref() to located the target unp instead.
Sepherosa Ziehau [Tue, 25 Aug 2015 14:05:14 +0000 (22:05 +0800)]
uipc: Fix various races on unp_connect() path.
And factor out unp_find_lockref(), which will be used to avoid
abusing unpcb.unp_conn on uipc_send() for unconnected DGRAM
unix sockets.
Sascha Wildner [Tue, 25 Aug 2015 17:38:46 +0000 (19:38 +0200)]
kernel: Sync sys/conf/files with recent drm(4) and radeonkms(4) updates.
This fixes building drm into the kernel.
François Tigeot [Tue, 25 Aug 2015 11:45:44 +0000 (13:45 +0200)]
drm: Make the hlist API compatible with Linux 3.17
* Replace hlist_add_after_rcu by hlist_add_behind_rcu
* hlist_for_each_entry() now only uses 3 arguments
François Tigeot [Tue, 25 Aug 2015 05:32:42 +0000 (07:32 +0200)]
kernel/linux: Add hweight64()
François Tigeot [Tue, 25 Aug 2015 05:15:27 +0000 (07:15 +0200)]
kernel: Add bitcount64 to sys/systm.h
Obtained-from: FreeBSD
Sepherosa Ziehau [Mon, 24 Aug 2015 14:02:20 +0000 (22:02 +0800)]
uipc: Use flag to indicate that the unpcb was detached/aborted.
Since unpcb detach (so->so_pcb set to NULL) could be delayed by
extra references, we use a flag to indicate that the unpcb was
detached/aborted. And hold per-unpcb token in uipc_detach() and
uipc_abort(), since this flag is also checked in other uipc_userreq
methods, e.g. uipc_send().
This commit makes uipc_abort() reentrant.
Sepherosa Ziehau [Mon, 24 Aug 2015 13:52:18 +0000 (21:52 +0800)]
uipc: Use atomic operation for unp_flags
Sascha Wildner [Mon, 24 Aug 2015 18:15:15 +0000 (20:15 +0200)]
sort.1: Clean up mdoc.
While here, set WITHOUT_NLS in the Makefile.
Sascha Wildner [Mon, 24 Aug 2015 18:01:18 +0000 (20:01 +0200)]
tcpdrop.8: Remove unneeded .Nm arguments.
Sascha Wildner [Mon, 24 Aug 2015 16:59:03 +0000 (18:59 +0200)]
Sort SEE ALSO in various manual pages.
Sascha Wildner [Mon, 24 Aug 2015 16:58:30 +0000 (18:58 +0200)]
mdoc.local: Add 4.0 version so it can be used in manpages (procctl(2)).
This is one of the rare cases where a new feature appeared "first" in
a release (i.e. was MFC'd).
Sascha Wildner [Mon, 24 Aug 2015 16:56:46 +0000 (18:56 +0200)]
Add MLINKS for 3 kernel functions documented on the led(4) manpage.
Sascha Wildner [Mon, 24 Aug 2015 16:51:39 +0000 (18:51 +0200)]
libusbhid: Add missing MLINK for hid_get_report_id().
Sascha Wildner [Mon, 24 Aug 2015 16:48:41 +0000 (18:48 +0200)]
Add MLINKS for recently added tree(3) macros.
Sascha Wildner [Mon, 24 Aug 2015 16:30:00 +0000 (18:30 +0200)]
mbintowcr.3: Further mdoc cleanup.
Sascha Wildner [Mon, 24 Aug 2015 16:16:34 +0000 (18:16 +0200)]
mbintowcr.3: Mention missing functions.
Sascha Wildner [Mon, 24 Aug 2015 15:57:37 +0000 (17:57 +0200)]
<xlocale/_wchar.h>: Fix wrong prototype.
Imre Vadász [Sun, 23 Aug 2015 21:09:43 +0000 (23:09 +0200)]
drm/radeon: Update temperature sensor support after update to Linux 3.17
We can apparently just use the rdev->pm.dpm.thermal.min_temp and
rdev->pm.dpm.thermal.min_temp values as WARN and CRIT threshholds
respectively.
Sepherosa Ziehau [Sun, 23 Aug 2015 14:44:49 +0000 (22:44 +0800)]
uipc: Factor out unp_getsocktoken()/unp_reltoken()
Streamline the code a little bit. This also paves the way to make
uipc_abort() reentrant.
Sascha Wildner [Sun, 23 Aug 2015 23:24:51 +0000 (01:24 +0200)]
libc: Add missing futimens(2) manual page link.
François Tigeot [Sun, 23 Aug 2015 20:11:16 +0000 (22:11 +0200)]
drm: Move DMI definitions to their usual Linux headers
François Tigeot [Sun, 23 Aug 2015 19:52:29 +0000 (21:52 +0200)]
drm/linux: Improve the workqueue subsystem
Adding some APIs used by the drm/i915 code from Linux 3.17
Tomohiro Kusumi [Sun, 23 Aug 2015 11:53:50 +0000 (20:53 +0900)]
sys/vfs/hammer: Don't fail to downgrade cursor
7ef2d7b should have placed end: before
hammer_cursor_downgrade(cursor);.
If hammer volume-del starts to reblock because of non-empty
volume, and first reblocks data, but skips reblock for btree
node as the node was not in that volume (data and btree node
stored in different volume which is usually a rare case),
then volume-del returns from hammer_reblock_helper() without
donwgrading upgraded lock.
This could happen only on volume-del, but not hammer reblock
command since hammer reblock command never hits goto end;.
Tomohiro Kusumi [Sun, 23 Aug 2015 11:12:49 +0000 (20:12 +0900)]
sys/vfs/hammer: Fix wrong comment
It doesn't return EDEADLK.
Tomohiro Kusumi [Sun, 23 Aug 2015 10:53:33 +0000 (19:53 +0900)]
sys/vfs/hammer: Fix possible panic with vfs.hammer.debug_general=0x4000
bc996e6 needed to check NULL-elm case for node reblock
when (vfs.hammer.debug_general & 0x4000) != 0.
The above sysctl has 0 by default, so nothing will happen
as long as keeping a default value when reblock starts.
Tomohiro Kusumi [Sun, 23 Aug 2015 05:49:33 +0000 (14:49 +0900)]
sys/vfs/hammer: Minor cleanups
Nothing changes, but this probably has better readability
(and also better for grep) as it doesn't make code looks
like it's about inode. nnode and nnode->ondisk must exist
at this point.
Also sync a comment between internal and leaf.
Also add a comment on internal and leaf reblock.
François Tigeot [Sat, 22 Aug 2015 21:26:26 +0000 (23:26 +0200)]
drm/radeon: Wrong authorship in previous commit
The big drm/radeon update to Linux 3.17 was really made by Rimvydas
Jasinskas but pushed by myself.
I must have changed the authorship unknowingly while rebasing a
bugfix and pushed the commit without noticing :-(
François Tigeot [Sat, 22 Aug 2015 16:20:38 +0000 (18:20 +0200)]
drm/radeon: Update to Linux 3.17 (v2)
Catch up with recent updates in i915kms driver.
While there perform some cleaning in drm and drm/ttm too.
This smallish(1.2MB) diff between Linux v3.12..v3.17 brings
few improvements for radeon cards support on DragonFly BSD.
Driver is VERY experimental even if FF 720p "Hey Mamma"
looks much more pleasing to the eyes on my Xeon box.
Me being Fortran developer(who wants to run OpenCL on BSDs too)
was literally throwing diffs at the fan to see what sticks on.
After cleaning up the mess a bit, these features seems to behave:
kms-syscons switching on R7 240 and 5770 JUNIPER;
current DPorts Xorg;
Xorg-next(1.17.2) + Mesa-next(glamor for radeonsi);
2D tiling now works out of the box on OLAND;
glxgears/cubemap/cuberender mesa demos;
firefox youtube videos(still some blinking on overlays);
mpv -vo vdpau --hwdec=vdpau bsd-0088.mp4 (--vo=opengl too);
UVD (decoding limited to 1080p frame streams);
fragging in OpenArena maxed-out(even ttys[0-8] to quickly check mail);
hdmi audio;
DPM (enabled by default);
gpu temperature monitoring through hw.sensors framework;
automatic firmware loading.
TODO:
update drm/ttm, any help would be really appreciated;
ww_mutex, yep still good old "homegrown" locks;
vma, newer shrinker api;
dma_buf/prime;
iic/i2c rework;
kldunload radeonkms;
better gart sizes support;
newer firmware format support;
fix failing ring sync test to ring-5(UVD);
check how driver supports X2 and IGP cards;
better OpenCL through Clover;
many more.
Thanks goes to:
ftigeot@efnet for opportunity and i915 work
YRabbit@efnet for extensive testing of hdmi A/V on TURKS
ivadasz@efnet for temperature sensors support
mneumann@efnet for testing on HAWAII
AMD for investing time and effort in opensource drivers
NetBSD and Linux gpu developers
Special thanks to:
everyone behind freebsd-ports-graphics/opencl (keep pushing the mngt!)
and that Sweet special girl out there.
François Tigeot [Sat, 22 Aug 2015 18:21:31 +0000 (20:21 +0200)]
drm: Reintroduce drm_kms_free()
* The Linux version of kfree() is a macro with one argument.
* The real kfree() function takes two arguments.
* kfree() cannot be simply called via a function pointer from
Linux code.
* Re-add the special purpose drm_kms_free() wrapper function
to handle this situation. It was wrongly removed during one of the
drm_crtc.c syncs with Linux.
Reported-by: ivadasz
zrj [Sat, 15 Aug 2015 07:08:52 +0000 (10:08 +0300)]
drm/radeon: Add pcie_get_readrq(), pcie_set_readrq() functions
pcie_set_readrq() needs something more clever(clear and set).
Extracted-from: drm/radeon/evergreen.c
Tomohiro Kusumi [Sat, 22 Aug 2015 07:59:11 +0000 (16:59 +0900)]
sys/vfs/hammer: Add hammer_move_node()
Procedures to move internal and leaf node are basically the same.
Also sync some comments between internal and leaf.
Tomohiro Kusumi [Sat, 22 Aug 2015 04:01:08 +0000 (13:01 +0900)]
sbin/hammer: Print total bytes of reblocked nodes
just like it shows both "data elements" and "data bytes"
in the next two lines.
"btree bytes" is kind of obvious since btree nodes are
all 4KB, and there isn't a field for "btree bytes" in
struct hammer_ioc_reblock as well, however this isn't
obvious for users so it should print it.
Tomohiro Kusumi [Sat, 22 Aug 2015 03:52:34 +0000 (12:52 +0900)]
sys/vfs/hammer: Change obsolete hammer_ioc_reblock fields to reserved
record_count and record_moves fields in
hammer_ioc_reblock are not used and not necessary.
These fields were used when there was a distinction
between data and meta reblock (see
bf686dbe). Meta
reblock is now a part of data reblock, so these two
are no longer used.
Tomohiro Kusumi [Fri, 21 Aug 2015 12:13:40 +0000 (21:13 +0900)]
sys/vfs/hammer: Don't try to reblock everything on volume-del
Add volume# option to reblock ioctl and use that on hammer
volume-del instead of trying to reblock every data/node that
the reblocker has faced.
This is only used internally by hammer volume-del ioctl when
it finds the volume-to-remove is not empty and needs to get
reblocked. hammer reblock command does not provide an option
to specify volume# since it shouldn't be necessary for users
in general.
This makes huge difference as it cuts extra and unnecessary
i/o from reblocking, especially when the volume to remove
only has small amount data in it. The following examples show
much of the time spent on reblocking on volume-del comes from
unnecessary i/o. Example1 spent most of the time moving data
in /dev/da1 which wasn't necessary.
Example 1 - before this commit
# newfs_hammer -L TEST /dev/da1 /dev/da2 /dev/da3 >/dev/null
# mount_hammer /dev/da1:/dev/da2:/dev/da3 /HAMMER
# dd if=/dev/zero of=/HAMMER/out bs=1000000 count=120000 >/dev/null 2>&1
# sync; sync; sync; sync
# df -Th /HAMMER
Filesystem Type Size Used Avail Capacity Mounted on
TEST hammer 334G 112G 222G 34% /HAMMER
# hammer -f /dev/da1:/dev/da2:/dev/da3 blockmap | grep -A2 layer1
layer1
4000000000000000 @
2000000000800000 blocks-free 0
4000000000000000 zone=4 app=8388608 free=0
4000000000800000 zone=4 app=8388608 free=0
--
layer1
4010000000000000 @
2010000000000000 blocks-free 14145
4010000000000000 zone=4 app=8388608 free=0
4010000000800000 zone=10 app=8388608 free=0
--
layer1
4020000000000000 @
2020000000000000 blocks-free 14268
4020000000000000 zone=4 app=8388608 free=0
4020000000800000 zone=0 app=0 free=8388608
/* /dev/da1 is 100% used
/dev/da2 is partially used
/dev/da3 is 0% used */
# hammer volume-blkdevs /HAMMER
/dev/da1:/dev/da2:/dev/da3
# time hammer volume-del /dev/da2 /HAMMER
real 14m29.119s
user 0m0.000s
sys 7m14.388s
# hammer volume-blkdevs /HAMMER
/dev/da1:/dev/da3
Example 2 - with this commit
# newfs_hammer -L TEST /dev/da1 /dev/da2 /dev/da3 >/dev/null
# mount_hammer /dev/da1:/dev/da2:/dev/da3 /HAMMER
# dd if=/dev/zero of=/HAMMER/out bs=1000000 count=120000 >/dev/null 2>&1
# sync; sync; sync; sync
# df -Th /HAMMER
Filesystem Type Size Used Avail Capacity Mounted on
TEST hammer 334G 112G 222G 34% /HAMMER
# hammer -f /dev/da1:/dev/da2:/dev/da3 blockmap | grep -A2 layer1
layer1
4000000000000000 @
2000000000800000 blocks-free 0
4000000000000000 zone=4 app=8388608 free=0
4000000000800000 zone=4 app=8388608 free=0
--
layer1
4010000000000000 @
2010000000000000 blocks-free 14145
4010000000000000 zone=4 app=8388608 free=0
4010000000800000 zone=10 app=8388608 free=0
--
layer1
4020000000000000 @
2020000000000000 blocks-free 14268
4020000000000000 zone=4 app=8388608 free=0
4020000000800000 zone=0 app=0 free=8388608
/* /dev/da1 is 100% used
/dev/da2 is partially used
/dev/da3 is 0% used */
# hammer volume-blkdevs /HAMMER
/dev/da1:/dev/da2:/dev/da3
# time hammer volume-del /dev/da2 /HAMMER
real 0m8.139s
user 0m0.000s
sys 0m4.177s
# hammer volume-blkdevs /HAMMER
/dev/da1:/dev/da3
Tomohiro Kusumi [Fri, 21 Aug 2015 12:13:05 +0000 (21:13 +0900)]
sys/vfs/hammer: Remove unnecessary error check
error is always ==0 at this point.
Tomohiro Kusumi [Fri, 21 Aug 2015 12:12:06 +0000 (21:12 +0900)]
sys/vfs/hammer: Don't repeat layer1 blockmap check of volume-to-remove
hammer's blockmap allocator can simply skip the whole volume
without going through unnecessary goto;s when the allocator's
current zone2 offset is in the volume-to-remove.
This is similar to 0871ec5 except that this commit does not
need to check layer1->phys_offset because that volume is what
hammer volume-del is wanting to remove.
Tomohiro Kusumi [Thu, 20 Aug 2015 14:46:56 +0000 (23:46 +0900)]
sys/vfs/hammer: Adjust hammer_debug_general
Change hammer_debug_general in hammer_blockmap_getfree()
to 0x4000 which is the same value as those in
sys/vfs/hammer/hammer_reblock.c.
This is better since hammer_blockmap_getfree() is made
for reblock and only used by reblock (0x0800 picks up
other debug kprintfs that aren't related to reblock).
Also note that hammer_debug_general values are randomly
picked based on use cases rather than fixed value for
each file/etc.
Add some comments on the conditionals that are used to
start data/node reblock since this part is not too easy
to understand.
Tomohiro Kusumi [Thu, 20 Aug 2015 10:55:25 +0000 (19:55 +0900)]
sbin/hammer: Move two blockmap functions to hammer/blockmap.c
from hammer/ondisk.c.
See sys/vfs/hammer/hammer_blockmap.c for details.
alloc_blockmap() should be in hammer/blockmap.c than
hammer/ondisk.c considering alloc_blockmap() is the
userspace version of hammer_blockmap_alloc() in
sys/vfs/hammer/hammer_blockmap.c which basically does
the same thing. If blockmap_lookup() is located in
hammer/blockmap.c then this one should be too.
alloc_bigblock() is the userspace only function that is
only used by newfs_hammer when newfs_hammer initially
allocates the whole 8MB chunks, however since the idea
of alloc_bigblock() is the same as alloc_blockmap()
(difference in allocation size), it makes better sense
to have this in hammer/blockmap.c as well.
(There are other blockmap related functions but those
are more of formatting initial ondisk image based on
blockmap info than blockmap handling itself, so those
should stay in hammer/ondisk.c)
Tomohiro Kusumi [Thu, 20 Aug 2015 10:22:42 +0000 (19:22 +0900)]
sbin/hammer: Cleanup pwrite(2) wrapper
Just make writehammerbuf() return 0 or -1 instead of void,
just like readhammerbuf() returns 0 or -1.
Tomohiro Kusumi [Thu, 20 Aug 2015 10:22:06 +0000 (19:22 +0900)]
sbin/hammer: Cleanup hammer show
Also change "=%d" to "=B%d" on error to clearly distinct
from non-error case "=%d%%".
'B' usually represents error (bad) in hammer show.
Tomohiro Kusumi [Wed, 19 Aug 2015 16:13:48 +0000 (01:13 +0900)]
sbin/newfs_hammer: Explicitly tell which volume is root volume
hammer volume-del says "Cannot remove root-volume" if one
tries to remove the root volume, however no hammer commands
actually tell which one is the root volume.
This commit makes newfs_hammer printf the root volume name,
followed by the existing undo/etc info of the root volume.
Also refer to the root volume in manpages.
(It may be helpful if other hammer commands like info or
volume-list show the root volume name just like newfs_hammer,
however it's not too trivial to do that. One reason is because
hammer seems to have been designed that way so users don't
need to care about any detail of underlying fs volumes.
This commit is trivial since newfs_hammer code knows which
one is the root volume)
Tomohiro Kusumi [Wed, 19 Aug 2015 14:58:05 +0000 (23:58 +0900)]
sys/vfs/hammer: Minor fix on volume format on volume-add
It's better to just copy volume relative offset from
the existing root volume than having the magic number
"HAMMER_BUFSIZE * 16" in both kernel and userspace.
Tomohiro Kusumi [Wed, 19 Aug 2015 12:31:40 +0000 (21:31 +0900)]
sbin/hammer: Merge check_volume()
There are two check_volume() that basically do the same.
The one in sbin/hammer/cmd_volume.c is copied from newfs_hammer.
Merge them into one.
Tomohiro Kusumi [Wed, 19 Aug 2015 12:03:13 +0000 (21:03 +0900)]
sbin/newfs_hammer: Fix minor memory leak
vol->fd is supposed to be already opened by setup_volume()
by the time check_volume(vol) is called.
Tomohiro Kusumi [Tue, 18 Aug 2015 11:32:28 +0000 (20:32 +0900)]
sys/vfs/hammer: Remove obsolete code and comments
from very early stage of hammer development.
Tomohiro Kusumi [Wed, 19 Aug 2015 10:05:09 +0000 (19:05 +0900)]
sys/vfs/hammer: Fix wrong function name in kprintf
Use the right name.
Allocating tid isn't always start of transaction.
Tomohiro Kusumi [Tue, 18 Aug 2015 13:08:05 +0000 (22:08 +0900)]
sys/vfs/hammer: Add hammer_flush_dirty()
Tomohiro Kusumi [Tue, 18 Aug 2015 11:25:04 +0000 (20:25 +0900)]
sys/vfs/hammer: Use HAMMER_OBJID_ROOT for root inode#
instead of hardcoded 1.
Tomohiro Kusumi [Tue, 18 Aug 2015 11:08:21 +0000 (20:08 +0900)]
sys/vfs/ufs: Fix typo in variable name
Tomohiro Kusumi [Tue, 18 Aug 2015 10:11:04 +0000 (19:11 +0900)]
sys/vfs/hammer: Check ronly when overwriting volume header
Check readonly mount before overwriting volume header
with non-NULL data on load/unload.
(hammer volume-add|del ioctls which are the only cases
with non-NULL data do have readonly mount check before
load/unload, so there isn't any issue without ronly==0
but the check should be done here too)
Tomohiro Kusumi [Tue, 18 Aug 2015 10:06:59 +0000 (19:06 +0900)]
sys/vfs/hammer: Use either HAMMER: or hammer:
Should be either of the following.
kprintf("HAMMER: ...\n");
kprintf("hammer: ...\n");
The existing code mostly uses "HAMMER:" so use "HAMMER:".
Assuming no one or no userspace triggers anything via
"hammer:" string in dmesg.
zrj [Thu, 13 Aug 2015 06:37:44 +0000 (09:37 +0300)]
drm/linux: Add pci_bus compat structure
zrj [Thu, 13 Aug 2015 06:26:52 +0000 (09:26 +0300)]
drm/linux: Expand pdev structure
Will be used to reduce direct dependency on drm functions in radeon.
Also will help a lot when updating drm itself.
François Tigeot [Sat, 22 Aug 2015 15:28:46 +0000 (17:28 +0200)]
drm: Really remove the old implementation of list_sort()
François Tigeot [Sat, 22 Aug 2015 13:53:03 +0000 (15:53 +0200)]
drm: Use the NetBSD implementation of list_sort()
* The NetBSD import is more similar to the Linux version and doesn't
do dynamic memory allocation
* Remove the old incomplete FreeBSD implementation
* It wasn't even using M_DRM with its kmalloc() calls, potentially
leaking memory
zrj [Mon, 17 Aug 2015 12:49:08 +0000 (15:49 +0300)]
drm/linux: Add list_sort() implementation
Taken-from: NetBSD
Sepherosa Ziehau [Sat, 22 Aug 2015 14:05:18 +0000 (22:05 +0800)]
uipc: Wrap unp test with UNP_ISATTACHED()
Sepherosa Ziehau [Sat, 22 Aug 2015 13:19:38 +0000 (21:19 +0800)]
socket: Allow keeping a reference on the new socket in sonewconn_faddr
It will be used to fix unix socket races.
zrj [Sat, 15 Aug 2015 14:26:17 +0000 (17:26 +0300)]
drm: Implement gcd64.h math function
Adapted-from: lib/libc/stdlib/getopt_long.c
zrj [Sun, 16 Aug 2015 09:32:55 +0000 (12:32 +0300)]
drm/linux: Implement simple capable() priv helper
zrj [Mon, 17 Aug 2015 11:35:11 +0000 (14:35 +0300)]
drm/linux: Add define for jiffies_64 compat
Likely is wrong, but better than nothing
zrj [Mon, 17 Aug 2015 12:15:54 +0000 (15:15 +0300)]
drm/linux: Implement atomic64_add/atomic64_sub macros
Sascha Wildner [Sat, 22 Aug 2015 08:36:18 +0000 (10:36 +0200)]
mbintowcr.3: Remove unneeded .Pp
zrj [Sun, 16 Aug 2015 14:03:48 +0000 (17:03 +0300)]
mkregtable: Sync script to Linux v3.14
This fixes sscanf handling vulnerability
Taken-from: Linux drm/radeon/mkregtable.c
zrj [Mon, 3 Aug 2015 08:01:17 +0000 (11:01 +0300)]
drm: const'ify ioctls table (v2)
Because, there is no reason for it not to be const.
We never modify the contents of drm_ioctls, so make it const.