Before my ZIL space optimization few years ago 128KB writes were logged
as two 64KB+ records in two 128KB log blocks. After that change it became
~124KB+/4KB+ in two 128KB log blocks to free space in the second block
for another record. Unfortunately in case of 128KB only writes, when space
in the second block remained unused, that change increased write latency by
imbalancing checksum computation time between parallel threads.
This change introduces new 68KB log block size, used for both writes below
67KB and 128KB-sharp writes. Writes of 68-127KB are still using one 128KB
block to not increase processing overhead. Writes above 131KB are still
using full 128KB blocks, since possible saving there is small. Mixed loads
will likely also fall back to previous 128KB, since code uses maximum of
the last 10 requested block sizes.
On a simple 128KB write test with queue depth of 1 this change demonstrates
~15-20% performance improvement.
MFC after: 2 weeks
Sponsored by: iXsystems, Inc.
This adds 8 and 16 bit versions of the cmpset and fcmpset functions. Macros
are used to generate all the flavors from the same set of instructions; the
macro expansion handles the couple minor differences between each size
variation (generating ldrexb/ldrexh/ldrex for 8/16/32, etc).
In addition to handling new sizes, the instruction sequences used for cmpset
and fcmpset are rewritten to be a bit shorter/faster, and the new sequence
will not return false when *dst==*old but the store-exclusive fails because
of concurrent writers. Instead, it just loops like ldrex/strex sequences
normally do until it gets a non-conflicted store. The manpage allows LL/SC
architectures to bogusly return false, but there's no reason to actually do
so, at least on arm.
Reviewed by: cognet
work. More precisely, it doesn't set ATTR_AP(ATTR_AP_USER) in the page
table entry, so any attempt to read from the mapped page by user space
generates a page fault. This problem has gone unnoticed because the page
fault handler, vm_fault(), will ultimately call pmap_enter(), which
replaces the non-working page table entry with one that has
ATTR_AP(ATTR_AP_USER) set.
This change reduces the number of page faults during a "buildworld" by
about 19.4%.
Reviewed by: andrew, markj
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D21841
Currently only suspend requests are acknowledged by writing an empty
string back to the xenstore control node, but poweroff or reboot
requests are not acknowledged and FreeBSD simply proceeds to perform
the desired action.
Fix this by acknowledging all requests, and remove the suspend specific
ack done in the handler.
Sponsored by: Citrix Systems R&D
MFC after: 3 days
As of r347221 the iflib legacy interrupt mode setup assumes that drivers
perform both receive and transmit processing from the interrupt handler.
This assumption is invalid in the vmxnet3 driver, so introduce the
IFLIB_SINGLE_IRQ_RX_ONLY flag to make iflib avoid tx processing in the
interrupt handler.
PR: 239118
Reported and tested by: Juraj Lutter <otis@sk.freebsd.org>
Obtained from: marius
Reviewed by: gallatin
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D21831
pv list lock is the main bottleneck during poudriere -j 104 and
pmap_remove_pages is the most impactful consumer. It frees chunks with the lock
held even though it plays no role in correctness. Moreover chunks are often
freed in groups, sample counts during buildkernel (0-sized frees removed):
value ------------- Distribution ------------- count
0 | 0
1 | 8
2 |@@@@@@@ 19329
4 |@@@@@@@@@@@@@@@@@@@@@@ 58517
8 | 1085
16 | 71
32 |@@@@@@@@@@ 24919
64 | 899
128 | 7
256 | 2
512 | 0
Thus:
1. batch freeing
2. move it past unlocking pv list
Reviewed by: alc (previous version), markj (previous version), kib
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D21832
including the TCP header in the first IP packet.
Enforce this in tcp_output(). In addition make sure that at least
one byte payload fits in the TCP segement to allow making progress.
Without this check, a kernel with INVARIANTS will panic.
This issue was found by running an instance of syzkaller.
Reviewed by: jtl@
MFC after: 3 days
Sponsored by: Netflix, Inc.
Differential Revision: https://reviews.freebsd.org/D21665
This API is still young enough that I would expect no one to be dependant on
this yet... Swap the ordering while it's young to match Linux values to
potentially ease implementation of linuxolator syscall, being able to reuse
existing constants.
r347183 bumped GEOM classes to SI_ORDER_SECOND to resolve a race between
them and the initialization of devsoftc.mtx in devinit, but missed this
dependency on g_flashmap that may now lose the race against GEOM
classes/g_init.
There's a great comment that describes the situation that has also been
updated with the new ordering of GEOM classes.
Reported by: bdragon
MFC after: 4 days
This driver is for the usb phy present on rockchip SoC.
It only support RK3399 and host mode for now.
The driver expose the usb clock needed by the usb controller.
On rockchip board it seems that the value in the DTS
are not enough for reseting the chip, I don't know if
the value are really incorrect or if DELAY is not precise
enough or if the rockchip gpio driver have some "lag" of some
kind or not.
For now just add more delay.
Module resets where not implemented when rockchip clocks were commited.
Implement them.
Since all resets registers are contiguous a driver only need to give
the start offset and the number of resets. This avoid to have to declare
every resets.
PLL_MIPI is the last important PLL that we missed.
Add support for it.
Since it's one of the possible parent for TCON0 also add this clock
now that we can.
While here add some info about what video related clocks should be
enabled at boot and with what frequency.
The REPRODUCIBLE_BUILD option is actually managed in two separate
files. src.opts.mk governs the setting for world builds and
kern.opts.mk governs it for kernel builds. r350550 only changed the
default for world builds.
Reported by: emaste
Reviewed by: emaste
Differential Revision: https://reviews.freebsd.org/D21444
initialisation, which is important when the host is dealing with a
SYN flood.
This affects the computation of the initial TCP sequence number for
the client side.
This has been discussed with secteam@.
Reviewed by: gallatin@
Sponsored by: Netflix, Inc.
Differential Revision: https://reviews.freebsd.org/D21616
for RACK specific socket options.
These issues were found by a syzkaller instance.
Reviewed by: rrs@
Sponsored by: Netflix, Inc.
Differential Revision: https://reviews.freebsd.org/D21825
and reinserting it back with an updated key.
This is one of dependencies for the upcoming stats(3) code.
Reviewed by: cem
Obtained from: Netflix
MFC after: 2 weeks
Sponsored by: Klara Inc, Netflix
Differential Revision: https://reviews.freebsd.org/D21786
Clang9/LLD9 appears to get quite confused with the instruction stream used
to obtain the tmpstack pointer, almost as though it thinks this is a C
function, so tries to optimize it. Since the AIM64 method doesn't use the
TOC to obtain the tmpstack, just follow that model, and lld won't get
confused.
Reported by: bdragon
MFC after: 2 weeks
On 64-bit platforms uintptr_t makes the copy twice as large as it should be.
This code isn't actually used in FreeBSD, since it's for guest mode only,
not hypervisor mode, but fixing it for completeness sake.
Reported by: bdragon (clang9 build)
- Ensure that the end of the mapping passed to vm_page_wire() is
page-aligned. vm_page_wire() expects this.
- Wire pages before reading data into them.
- Apply protections specified in the segment descriptor using
vm_map_protect() once relocation processing is done.
- On amd64, ensure that we load KLDs above KERNBASE, since they
are compiled with the "kernel" memory model by default.
Reviewed by: kib
MFC after: 2 weeks
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21756
We must also check for large mappings. pmap_page_is_mapped() is
mostly used in assertions, so the problem was not very noticeable.
Reviewed by: alc
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D21824
Software Kernel TLS needs to allocate a new destination crypto
buffer when encrypting data from the page cache, so as to avoid
overwriting shared clear-text file data with encrypted data
specific to a single socket. When the data is anonymous, eg, not
tied to a file, then we can encrypt in place and avoid allocating
a new page. This fixes a bug where the existing code always
assumes the data is private, and never encrypts in place. This
results in unneeded page allocations and potentially more memory
bandwidth consumption when doing socket writes.
When the code was written at Netflix, ktls_encrypt() looked at
private sendfile flags to determine if the pages being encrypted
where part of the page cache (coming from sendfile) or
anonymous (coming from sosend). This was broken internally at
Netflix when the sendfile flags were made private, and the
M_WRITABLE() check was added. Unfortunately, M_WRITABLE() will
always be false for M_NOMAP mbufs, since one cannot just mtod()
them.
This change introduces a new flags field to the mbuf_ext_pgs
struct by stealing a byte from the tls hdr. Note that the current
header is still 2 bytes larger than the largest header we
support: AES-CBC with explicit IV. We set MBUF_PEXT_FLAG_ANON
when creating an unmapped mbuf in m_uiotombuf_nomap() (which is
the path that socket writes take), and we check for that flag in
ktls_encrypt() when looking for anon pages.
Reviewed by: jhb
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21796
TLS 1.3 requires a few changes because 1.3 pretends to be 1.2
with a record type of application data. The "real" record type is
then included at the end of the user-supplied plaintext
data. This required adding a field to the mbuf_ext_pgs struct to
save the record type, and passing the real record type to the
sw_encrypt() ktls backend functions.
Reviewed by: jhb, hselasky
Sponsored by: Netflix
Differential Revision: D21801
The current mechanism is bogus in several ways:
- the limit is a percentage of total entries added, which means negative
entries get evicted all the time even if there are plenty of resources
- evicting code is almost not concurrent, which makes it unable to
remove entries fast enough when doing something as simple as -j 104
buildworld
- there is no support for performing mass removal if necessary
Vast majority of negative entries never get any hits. Only evicting
them when the filesystem demands it results in a significant growth of
the namecache with almost no improvement in the hit ratio.
Sample result about afer 90 minutes of poudriere -j 104:
current no evict % of the original
numneg 219737 2013157 916
numneghits 266711906 263544562 98 [1]
[1] this may look funny but there is a certain dose of variation to the
build
The number was chosen as something which mostly eliminates spurious
evictions during lighter workloads but still keeps the total at bay.
Sponsored by: The FreeBSD Foundation
