synchronous inode update.
The IN_SIZEMOD and IN_IBLKDATA flags indicate changes to the
file size and block pointer fields in the inode. When these
fields have been changed, the fsync() and fsyncdata() system
calls must write the inode to ensure their semantics that the
file is on stable store.
The IN_SIZEMOD and IN_IBLKDATA flags cannot be cleared until
a synchronous write of the inode is done. If they are cleared
on an asynchronous write, then the inode may not yet have been
written to the disk when an fsync() or fsyncdata() call is done.
Absent these flags, these calls would not know that they needed
to write the inode. Thus, these flags only can be cleared on
synchronous writes of the inode. Since the inode will be locked
for the duration of the I/O that writes it to disk, no fsync()
or fsyncdata() will be able to run before the on-disk inode
is complete.
Reviewed by: kib
MFC with: -r361785
Differential revision: https://reviews.freebsd.org/D25072
Add xref to all SIM devices we currently have (including a rough indication
which ones are likely to fail).
Update to include all the CAM options.
Fix a few igor nits while I'm here.
Some automation tries to detect if nvd or nda is in used, and the presence of
both confuses it. Provide a knob to turn off nvd alias creation
(kern.cam.nda.nvd_compat=0) for these situations. The default is the same:
create the nvd compat link.
Summary:
Radix on AIM, and all of Book-E (currently), can do direct addressing of
user space, instead of needing to map user addresses into kernel space.
Take advantage of this to optimize the copy(9) functions for this
behavior, and avoid effectively NOP translations.
Test Plan: Tested on powerpcspe, powerpc64/booke, powerpc64/AIM
Reviewed by: bdragon
Differential Revision: https://reviews.freebsd.org/D25129
Summary:
The point of this addition is to cache CPU behavior 'features', to avoid
having to recompute based on CPU, etc.
The first such use case is to avoid the unnecessary manipulation of the
SLBs (Segment Lookaside Buffers) when using the Radix pmap on POWER9.
Since we already get the PCPU pointer wherever we swap the SLB entries,
we can use a cached flag to check if it's necessary to perform the
operation anyway, and skip it when not.
Reviewed by: bdragon
Differential Revision: https://reviews.freebsd.org/D24908
Instead, just skip marking pages valid if the read fails. Future
attempts to access such pages will notice that they are not marked valid
and try to read them from disk again.
Reviewed by: kib, markj
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D25138
Without this patch, if a SIGHUP is handled while the process is executing
get_exportlist(), that SIGHUP is essentially ignored because the got_sighup
variable is reset to 0 after get_exportlist().
This results in the exports file(s) not being reloaded until another SIGHUP
signal is sent to mountd.
This patch fixes this by resetting got_sighup to zero before the
get_exportlist() call while SIGHUP is blocked.
It also defines a delay time of 250msec before doing another exports reload
if there are RPC request(s) to process. This prevents repeated exports reloads
from delaying handling of RPC requests significantly.
PR: 246597
Reported by: patrykkotlowski@gmail.com
Tested by: patrykkotlowski@gmail.com
Reviewed by: markj
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D25127
Comparing the object files produced by GNU as 2.17.50 and Clang IAS
shows many immaterial changes in strtab etc., and one material change
in .text:
1bac: 4c 8b 4f 18 mov 0x18(%rdi),%r9
1bb0: eb 0e jmp 1bc0 <Skein1024_block_loop>
- 1bb2: 66 66 2e 0f 1f 84 00 data16 nopw %cs:0x0(%rax,%rax,1)
- 1bb9: 00 00 00 00
- 1bbd: 0f 1f 00 nopl (%rax)
+ 1bb2: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1)
+ 1bb9: 00 00 00
+ 1bbc: 0f 1f 40 00 nopl 0x0(%rax)
0000000000001bc0 <Skein1024_block_loop>:
Skein1024_block_loop():
1bc0: 4c 8b 47 10 mov 0x10(%rdi),%r8
1bc4: 4c 03 85 c0 00 00 00 add 0xc0(%rbp),%r8
That is, GNU as and Clang's integrated assembler use different multi-
byte NOPs for alignment (GNU as emits an 11 byte NOP + a 3 byte NOP,
while Clang IAS emits a 10 byte NOP + a 4 byte NOP).
Dependency cleanup hacks are not required, because we do not create
.depend files from GNU as.
Reviewed by: allanjude, arichardson, cem, tsoome
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D8434
r359473 removed the page unbusy logic from sendfile_iodone() because when
vm_pager_get_pages_async() would return an error after failing to start
the async I/O (eg. because VOP_BMAP failed), sendfile_swapin() would also
unbusy the pages, and it was wrong to unbusy twice. However this breaks
the case where vm_pager_get_pages_async() succeeds in starting an async I/O
and the async I/O is what fails. In this case, sendfile_iodone() must
unbusy the pages, and because sendfile_iodone() doesn't know which case
it is in, sendfile_iodone() must always unbusy pages and relookup pages
which have been substituted with bogus_page, which in turn means that
sendfile_swapin() must never do unbusy or relookup for pages which have
been given to vm_pager_get_pages_async(), even if there is an error.
Reviewed by: kib, markj
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D25136
r360979 erroneously assumed that the lowest mapping in an address space
would be a file mapping, but of course this is not true in general.
Reported and tested by: Frederic Chardon <chardon.frederic@gmail.com>
MFC after: 3 days
Clang IAS does not support the --defsym argument, and
.ifndef SKEIN_USE_ASM
gets turned into
.ifndef 1792
by the preprocessor, which results in
error: expected identifier after '.ifdef'
.ifndef 1792
^
Use #ifdef instead, which still works with GNU as.
Reviewed by: cem
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D25154
vfs.zfs.boot.primary_pool is only set on BIOS boot, use vfs.root.mountfrom
instead and update all vdevs on pool.
Reviewed by: allanjude
Sponsored by: Netflix, Klara Inc.
Differential Revision: https://reviews.freebsd.org/D25103
For synthetic aliases (just pseudonyms inferred from metadata like GPT or
UFS labels, GPT UUIDs, etc), use the GEOM provider aliasing system to create
a symlink to the real device instead of creating an independent device.
This makes it more clear which labels and devices correspond, and we can
safely have multiple labels to a single device accessed at once.
The confusingly named geom_label on-disk construct continues to behave
identically to how it did before.
This requires teaching GEOM's provider aliasing about the possibility
that aliases might be added later in time, and GEOM's devfs interaction
layer not to worry about existing aliases during retaste.
Discussed with: imp
Relnotes: sure, if we don't end up reverting it
Differential Revision: https://reviews.freebsd.org/D24968
Turns out this isn't a required call. I didn't pick it up because my
uncommitted changes involve new updateslot methods for cards I'm working
on.
Dunce hat to: adrian
GNU as seems to allow macro arguments without the '\' but clang is more
strict in that regard.
This change makes the source code compatible with LLVM's but does not yet
change the build system or rename it to .S.
The new code assembles identically with GNU as 2.17.50.
Reviewed By: emaste
Differential Revision: https://reviews.freebsd.org/D25143
Since we had a .set reorder, the nop after the "jal" was being placed after
the delay slot, resulting in two nops.
While changing this code also guard the .set noreorder with .set push/pop
and use $zero as the cpsetup save register since we don't need to save $gp.
Reviewed By: jhb
Differential Revision: https://reviews.freebsd.org/D25025
Copying the approach chosen in r309412. This fixes building the libc tests
on a macOS host since the macOS /bin/dd binary does not support status=none.
As there only seem to be two uses, this commit changes the two Makefiles.
If this becomes more common, we could also add a wrapper bootstrap script
that ignores status= and forwards the remaining args to the real dd.
Another alternative would be to remove the status flag and pipe stderr to
/dev/null, but them we lose error messages.
Reviewed By: brooks
Differential Revision: https://reviews.freebsd.org/D24785
and not only the software cache of that register. Else
pci_channel_offline() won't detect that the PCI device is gone when
using the LinuxKPI.
MFC after: 1 week
Sponsored by: Mellanox Technologies
description of items residing in a so-called union. FreeBSD currently
only supports 4 such pop levels.
If the push level is not restored within the processing of the same
HID item, an invalid memory location may be used for subsequent HID
item processing.
Verify that the push level is always valid when processing HID items.
Reported by: Andy Nguyen (Google)
MFC after: 3 days
Sponsored by: Mellanox Technologies
This is hopefully a big no-op unless you're running some extra
patches to flip on A-MSDU options in a driver.
802.11n supports sending A-MSDU in A-MPDU. That lets you do things
like pack small frames into an A-MSDU and stuff /those/ into an A-MPDU.
It allows for much more efficient airtime because you're not
wasting time sending small frames - which is still a problem when
doing A-MPDU as there's still per-frame overhead and minimum A-MPDU
density requirements.
It, however, is optional for 802.11n. A lot of stuff doesn't advertise
it (but does it, just wait!); and I know that ath10k does it and my
ath(4) driver work supports it.
Now, 802.11ac makes A-MSDU in A-MPDU something that can happen more
frequently, because even though you can send very large A-MPDUs
(like 1 megabyte and larger) you still have the small frame problem.
So, 802.11ac NICs like ath10k and iwm will support A-MSDU in A-MPDU
out of the box if it's enabled - and you can negotiate it.
So, let's lay down the ground work to enable A-MSDU in A-MPDU.
This will allow hardware like iwn(4) and ath(4) which supports
software A-MSDU but hardware A-MPDU to be more efficient.
Drivers that support A-MSDU in A-MPDU will set TX/RX htcap flags.
Note this is separate from the software A-MSDU encap path; /that/
dictates whether net80211 is doing A-MSDU encapsulation or not.
These HTC flags control negotiation, NOT encapsulation.
Once this negotiation and driver bits are done, hardware like
rtwn(4), run(4), and others will be able to use A-MSDU even without
A-MPDU working; right now FF and A-MSDU aren't even attempted
if you're an 11n node. It's a small hold-over from the initial
A-MPDU work and I know how to fix it, but to flip it on properly
I need to be able to negotiate or ignore A-MSDU in A-MPDU.
Oh and the fun part - some 11ac APs I've tested will quite happily
decap A-MSDU in A-MPDU even though they don't negotiate it when
doing 802.11n. So hey, I know it works - I just want to properly
handle things. :-)
Tested:
* AR9380, STA/AP mode
The 11b/11g ERP and slot time update handling are two things which weren't
migrated into the per-VAP state when Sam did the initial VAP work.
That makes sense for a lot of setups where net80211 is driving radio state
and the radio only cares about the shared state.
However, as noted by a now deleted comment, the ERP and slot time updates
aren't EXACTLY correct/accurate - they only take into account the most
RECENTLY created VAP, and the state updates when one creates/destroys
VAPs isn't exactly great.
So:
* track the short slot logic per VAP;
* whenever the slot time configuration changes, just push it into a deferred
task queue update so drivers don't have to serialise it themselves;
* if a driver registers a per-VAP slot time handler then it'll just get the
per VAP one;
* .. if a driver registers a global one then the legacy behaviour is maintained -
a single slot time is calculated and pushed out.
Note that the calculated slot time is better than the existing logic - if ANY
of the VAPs require long slot then it's disabled for all VAPs rather than
whatever the last configured VAP did.
Now, this isn't entirely complete - the rest of ERP tracking around short/long
slot capable station tracking needs to be converted into per-VAP, as well
as the preamble/barker flags. Luckily those also can be done in a similar
fashion - keep per-VAP counters/flags and unify them before doing the driver
update. I'll defer that work until later.
All the existing drivers can keep doing what they're doing with the global
slot time flags as that is maintained. One driver (iwi) used the per-VAP
flags instead of the ic flags, so now that driver will work properly.
This unblocks some ath10k porting work as the firmware takes the slot time
configuration per-VAP rather than globally, and some firmware handles
STA+AP and STA+STA (on same/different channels) configurations where
the firmware will switch slot time as appropriate.
Tested:
* AR9380, STA/AP mode
* AR9880 (ath10k), STA mode
net80211 currently doesn't negotiate A-MSDU in A-MPDU during ADDBA.
I've added the field in net80211 and this commit:
* Prints out the ADDBA field value during ADDBA;
* Adds some comments around where I need to follow up with some
negotiation logic.
Right now we don't have a driver flag anywhere which controls
whether A-MSDU in A-MPDU is allowed. I know it works (I have it
manually turned on at home on a couple test APs, heh!) but
I can't flip it on until we can negotiate it.
Tested:
* AR9380, STA/AP mode, printing out ADDBA requests
These are from the linux iwlwifi driver ;the default use smaller
maximum AMPDUs (8k) and a much smaller density (none.) The latter
could cause stability issues.
Tested:
* Tested on Intel 6300, STA mode.
Differential Revision: https://reviews.freebsd.org/D25113
Now that I have A-MSDU and A-MPDU coexisting together, we need to actually
announce if (a) it's permitted and (b) figure out if we should use it
when transmitting.
This just adds the field; it doesn't yet include it in ADDBA exchanges.
My AMD Ryzen system has 4 AHCI controllers, each supporting 16 MSI vectors.
Since two of the controllers have only one SATA port, limit to single MSI
saves system 30 interrupt vectors for free.
It may be possible to also limit number of MSI vectors to 4 and 8 for the
other two controllers, but according to the AHCI specification after that
controllers may revert to only one vector, that would be a bigger loss to
risk.
MFC after: 2 weeks
requires that new data on growing files be accessible. Thus, the
the fsyncdata() system call must update the on-disk inode when the
size of the file has changed.
This commit adds another inode update flag, IN_SIZEMOD, that gets
set any time that the file size changes. If either the IN_IBLKDATA
or the IN_SIZEMOD flag is set when fdatasync() is called, the
associated inode is synchronously written to disk. We could have
overloaded the IN_IBLKDATA flag to also track size changes since
the only (current) use case for these flags are for fsyncdata(),
but it does seem useful for possible future uses to separately
track the file size changes and the inode block pointer changes.
Reviewed by: kib
MFC with: -r361785
Differential revision: https://reviews.freebsd.org/D25072
