fssconfig(8): Clarify some details.
- Note up front that fssdev can be mounted with mount(8).
- Emphasize that fssconfig -c takes a mount point, not an arbitrary
directory.
- Break the prose paragraph of options for snapstore into a bulleted
list.
- Clarify that you needn't use fssconfig(8) at all with dump(8).
- Move dump -X/-x option up front to make it more obvious.
Pull up following revision(s) (requested by joe in ticket #1344):
usr.sbin/npf/npfctl/npf_build.c: revision 1.64,1.67
improve table load error in npf
an EEXIST should clearly state already defined and a load failure should
also clearly be stated. do not class all error returns under EEXIST
since npf_table_insert can return other errors aside EEXIST.
provide more context into npf reload failures with const tables.
PR lib/60740
Pull up following revision(s) (requested by joe in ticket #494):
usr.sbin/npf/npfctl/npf_build.c: revision 1.64,1.67
improve table load error in npf
an EEXIST should clearly state already defined and a load failure should
also clearly be stated. do not class all error returns under EEXIST
since npf_table_insert can return other errors aside EEXIST.
provide more context into npf reload failures with const tables.
PR lib/60740
Pull up following revision(s) (requested by tsutsui in ticket #493):
sys/kern/vfs_bio.c: revision 1.309
vfs_bio: retry fresh buffer allocation if recycling fails
getnewbuf() uses buf_lotsfree() to choose whether to
allocate a fresh buffer before trying BQ_AGE and BQ_LRU.
If fresh allocation is skipped and no buffer can actually be
recycled, it sleeps on needbuffer_cv even when the buffer cache
is still below its high water mark.
In that situation, a thread which could release a buffer may
itself need another buffer, resulting in a deadlock.
To avoid this, if recycling finds no usable buffer and bufmem is
still below bufmem_hiwater, make one more PR_NOWAIT fresh allocation
attempt before sleeping. See PR kern/60584 for details.
Pull up following revision(s) (requested by tsutsui in ticket #492):
sys/arch/atari/atari/atari_init.c: revision 1.121
sys/arch/atari/atari/atari_init.c: revision 1.122
atari: Fix fatal 68030 MMU TC initialization bug
ATARITT kernels from NetBSD 11.0 and -current fail to boot on my
Atari TT030.
The inline asm that loads TC specifies only `"a" (&tc)` as an input,
so the compiler may eliminate the initialization of tc as a dead store.
Then the generated code loads an uninitialized value from the stack
into TC.
Add an `"m" (tc)` input operand to the inline asm so that the compiler
knows that the asm reads tc from memory.
[26 lines not shown]
Add various Intel models (Lakefield, Granite Rapids, Arrow Lake, Panther Lake,
WildCat, Clearwater Forest, Sierra Forest, Lunar Lake).
Fix Yonah name (missing comma between Pentium and Core Duo).
Add few more AMD K8 models (Lima, Sherman, Windsor FX, Tyler, Richmond).
Fix overly aggressive rejection of valid servers.
Patch taken from https://bugs.ntp.org/3877 . This patch is likely to be
included in the next release.
Fixes PR bin/60831
Rewrite Thumb-2 NEON into ARM form before dispatching the trap.
Now neon_handler() recognize it and turns the FPU on.
Fixes execution of Thumb-2 userland on ARMv7 NEON kernel, which
previously was SIGILLed on the first SIMD code after switching the
FPU off.
Use lfence to serialize rdtsc for VIA Nano and up, same as Linux do.
Unlike Linux I didn't add a 64-bit guard on assumption that it should work
the same way and the 64-bit version will likely be the preferred one by users.
planning to pull up to netbsd-11, netbsd-10.
PR kern/68010 - sys/conf/osrelease.sh - add comments
Add some comments to osrelease.sh to explain its purpose,
keep away gawkers, etc...
While here, fix a couple of sh quoting botches.
NFCI.
No pullups.
Include model 7 in CPUID_TO_MODEL() extended model check.
Accommodates Zhaoxin CPUs using extended model.
Update the comment accordingly and fix one typo in it.
evbarm/INTEGRATOR_CP: establish interrupt handler for PL181 controller
PR/60818 plmmc_ifpga.c does not establish interrupt handler for PL181 controller
Capture the handle returned from ifpga_intr_establish - someone might
write the code to call ifpga_intr_disestablish.
fssconfig(8): some minor improvements
Try to clarify snapstore a bit more with more explicit language.
The proverbial desperate sysadmin at 3am doesn't need an exegetic
exercise.
fssconfig(8): some minor improvements
Avoid understcore in fss_dev, as it doesn't mesh well with the
underlining used for .Ar in plain text.
Make [-c] case visually match the other "action" options -l, and -u.
risc-v: remove initialisation of ci_shootdowncpus as it's unused.
RISC-V (currently) uses SBI services to invalidate TLB entries so
ci_shootdowncpus is unused. Remove its initialisation, but retain
the field to avoid changing struct cpuinfo layout, and it may be
used in the future.
Handle sync, interlace in GTF modes, explicity init name in EDID and GTF.
Avoid stack garbage passed in those uninitialized fields - fixes
wsdisplay drivers that actually read these fields.
Fixes and features for PPC440 core, PPC460EX chip, Sam460ex board.
- Add ebc - PPC440 family External Bus Controller driver.
- Add samfpga - Driver for Sam460ex on-board FPGA (provides 80 GPIOs,
soft power off, 3x led(4), etc. Lives on the ebc.
- Enable PPC460EX on-chip GPIO via gpio_opb.
- Implement a bunch of workarounds for 440/460 hardware bugs.
- Enable pm2fb in SAM460EX config (necessary fixes to pm2fb pending
commit).
- Disable OHCI in SAM460EX config - due to hardware bug operating
two AHB masters at the same time may cause lockups. And it sure does,
there's no known software workaround.
- Don't allow passing PSL_FP from MSR userland bits (userland can cause
panic).
- Fix ghost interrupt deliveries where asserted line re-entered the
exception.
- Ensure interrupt level-mask flag is set by default.
- Tag mmaped DMA address as cacheable unless memory was explicitly
allocated as uncached (only for 440, OEA probably could use it too).
[11 lines not shown]
