kernel - Minor adjustment to set_flags cast
* filedesc flags are a char. Avoid potential future issues by typing
it and cast the int argument from fsetfdflags() to clean the
implementation up a bit. Leave the argument as an int for now.
sys/vfs/hammer2: Remove hardlink related obsolete comments
These "real inode" comments are from old hardlink design
(e.g. e12ae3a59f03333cd0ce1dc645dd4719c32592b1 in 2015).
Most have been removed, so remove remaining ones.
Also remove "isopen" comments for hammer2_inode_unlink_finisher().
isopen was removed in bb16050ede5e81262312893fc8aaa54959cf993b
in 2022.
sys/vfs/hammer2: Add missing ctime updates
This fixes many errors in pjdfstest.
The inode in question is already "modified" for either mtime or nlinks,
so no extra chain flushing only to update ctime.
<sys/param.h>: Supplement 500706 with RTM_VERSION and _CMSG_ALIGN changes
There were RTM_VERSION bump (from 6 to 7) and _CMSG_ALIGN() changes
(alignment raised from 4 to 8) prior to the revision 500706, but those
changes themselves didn't bump the __DragonFly_version. So supplement
the 500706 revision notes with such changes.
See also the commits of those changes:
* Commit b3aa44a1fb12417a3c45a78a0620e126ae6b624d
Roy Marples <roy at marples.name>
Wed, 11 Sep 2019 19:25:11 +0000 (20:25 +0100)
build: Adjust _CMSG_ALIGN() for portability
* Commit 43a373152df2d405c9940983e584e6a25e76632d
Roy Marples <roy at marples.name>
Thu, 12 Sep 2019 08:13:30 +0000 (09:13 +0100)
route(4): Add ifam_pid and ifam_addrflags to ifa_msghdr
Debug - Fixing fdpassing.c test program
* Use proper alignment macros. This test program must be really
old, from a time when sizeof(long) used to be 4 but now is 8.
usr.sbin/makefs: Sync with sys/vfs/hammer2
Sync with following sys/vfs/hammer2 commits.
$ git log --pretty="%h %s" 55066b18244fe9350bb96f93d65807d82b5cd98b.. -- sys/vfs/hammer2/
4f646c6893 sys/vfs/hammer2: Use HAMMER2_LOOKUP_ALWAYS in xop_readdir
6d68a182a9 sys/vfs/hammer2: Add missing error handling in hammer2_remount()
34ad347f7f sys/vfs/hammer2: Fix "Properly set ronly flag for PFS"
9a5c949af7 sys/vfs/hammer2: Properly set ronly flag for PFS
0c9702fdbe sys/vfs/hammer2: Remove duplicate spmp->pfs_hmps[0] initialization
8860585752 sys/vfs/hammer2: Ignore unused return value from hammer2_xop_feed()
bd41e746e6 sys/vfs/hammer2: Add SEEK_HOLE / SEEK_DATA support
sys/vfs/hammer2: Use HAMMER2_LOOKUP_ALWAYS in xop_readdir
hammer2_xop_readdir() should explicitly use HAMMER2_LOOKUP_ALWAYS
for chain lookup, as hammer2_vop_readdir() can access chain data
for (a long enough) dirent name. Note that chain data is loaded
with or without ALWAYS in this case.
drm/linux - More closely match linux wait-for-completion
Linux wait-for-completion functions decrement `done` counter if a
completion did not signaled with complete_all. Do the same on DragonFly.
Proposed-By: dillon
drm/radeon - Fix panic when launching Xorg server on Radeon Xpress 1100/1150 GPU
I2C adapter lock_ops were not initialized on certain radeon hardware.
Add explicit call to i2c_add_adapter which initializes lock_ops.
Fixes #3374.
Reported-by: hjckr
drm - Update to Linux 4.20.17
core:
* Implement basic DMA-BUF support
amdgpu:
* Add support for Picasso, Raven2, and Vega20
i915:
* Add support up to Ice Lake
* Fix hardware acceleration
sys/vfs/hammer2: Fix "Properly set ronly flag for PFS"
hammer2_remount() is update once only for hmp->ronly,
so pmp->ronly needed to be updated first independently,
otherwise causes problems with mount update after boot.
sys/vfs/hammer2: Properly set ronly flag for PFS
Redo reverted b18a59622759a0617ee06c73b282dfc317ae70bf.
Once ->ronly becomes 0, it never becomes non-zero again.
(Update to ro still works as MNT_RDONLY is set in sys_mount().)
sys/vfs/hammer2: Ignore unused return value from hammer2_xop_feed()
hammer2_xop_feed callers that never use the return value
(except for these two) simply ignore it.
sys/vfs/hammer2: Add SEEK_HOLE / SEEK_DATA support
Similar to ext2 and ufs using VOP_BMAP based vn_bmap_seekhole().
- Add xop_bmap which corresponds to .vop_bmap.
- Add BUF_CMD_SEEK to prevent xop_bmap from being used for other
purposes e.g. cluster read / write. Note that vop_bmap_args::a_cmd
is only used in here and HAMMER1 (supports BUF_CMD_READ only,
and SEEK_XXX is unsupported).
atkbd: Fix first keystroke force reset
Some i8042 falsely return KBD_ACK for ECHO command which cause the
keyboard echo test to fail.
Without passing echo test, the keyboard is considered as unconfigured.
Although this is incorrect behavior, we still consider it correct to
prevent the forced reset (a step to configure a keyboard) of the entire
keyboard when the first key interrupt is reached.
This should fix keyboard probing at least for Framework Laptop [1].
Taken-from: FreeBSD [2] (minus typos)
[1]: https://reviews.freebsd.org/D50498
[2]: https://github.com/freebsd/freebsd-src/commit/cba5d7ab32fd0b4e9c4312cc967baf137377f812
atkbd: remove unused code that measures inb() duration
We are not using the retry count as FreeBSD does.
Instead, we are using TOTALDELAY / CHECKTIMEOUT, which measures the
actual execution time after each loop iteration and breaks out of the
loop when the total time (70ms or 200ms in most cases) has expired.