membar_producer is supposed to be a store-store barrier.
Also, in the code that FreeBSD has ported from illumos membar_producer
is used only with regular stores to regular memory (with respect to
caching).
We do not have an MI primitive for the store-store barrier, so
atomic_thread_fence_rel is the closest we have as it provides
(load | store) -> store barrier.
Previously, membar_producer was an empty function call on all 32-bit
arm-s, 32-bit powerpc, riscv and all mips variants. I think that it was
inadequate.
On other platforms, such as amd64, arm64, i386, powerpc64, sparc64,
membar_producer was implemented using stronger primitives than required
for a store-store barrier with respect to regular memory access.
For example, it used sfence on amd64 and lock-ed nop in i386 (despite TSO).
On powerpc64 we now use recommended lwsync instead of eieio.
On sparc64 FreeBSD uses TSO mode.
On arm64/aarch64 we now use dmb sy instead of dmb ish. Not sure if this
is an improvement, actually.
After this change we can drop opensolaris_atomic.S for aarch64, amd64,
powerpc64 and sparc64 as all required atomic operations have either
direct or light-weight mapping to FreeBSD native atomic operations.
Discussed with: kib
MFC after: 4 weeks
starting at the max. domain, and then work down. Then existing FreeBSD
drivers will attach. Interrupt routing from the VMD MSI-X to the NVME
drive is not well known, so any interrupt is sent to all children that
register.
VROC used Intel meta data so graid(8) works with it. However, graid(8)
supports RAID 0,1,10 for read and write. I have some early code to
support writes with RAID 5. Note that RAID 5 can have life issues
with SSDs since it can cause write amplification from updating the parity
data.
Hot plug support needs a change to skip the following check to work:
if (pcib_request_feature(dev, PCI_FEATURE_HP) != 0) {
in sys/dev/pci/pci_pci.c.
Looked at by: imp, rpokala, bcr
Differential Revision: https://reviews.freebsd.org/D21383
The cursor boundary tag is statically allocated in the vmem instead of
from the vmem_bt_zone. Explicitly remove it from the vmem's segment
list in vmem_destroy before freeing all the segments from the vmem.
Reviewed by: markj
MFC after: 1 week
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D21953
This ensures the clip task won't race with t4_destroy_clip_table.
While here, make some mutex destroys unconditional since attach always
initializes them.
Reviewed by: np
MFC after: 1 week
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D21952
Cast the pointers to (uintptr_t) before assigning to type
uint64_t. This eliminates an error from gcc when we cast the pointer
to a larger integer type.
Ensure the epoch_call() function is not called more than one time
before the callback has been executed, by always checking the
RS_FUNERAL_SCHD flag before invoking epoch_call().
The "rs_number_dead" is balanced again after r353353.
Discussed with: rrs@
Sponsored by: Mellanox Technologies
the epoch section towards return statement. Since entering epoch
is cheap, it is easier to cover the whole function with epoch,
rather than try to properly maintain its state.
If the bootloader enabled DMA we need to fully reset the DMA controller
otherwise we might have some stale data in it that provoke weird
behavior.
MFC after: 1 week
atomic_cas_32 is implemented using atomic_fcmpset_32 on all platforms.
Ditto for atomic_cas_64 and atomic_fcmpset_64 on platforms that have it.
The only exception is sparc64 that provides MD atomic_cas_32 and
atomic_cas_64.
This is slightly inefficient as fcmpset reports whether the operation
updated the target and that information is not needed for cas.
Nevertheless, there is less code to maintain and to add for new platforms.
Also, the operations are done inline now as opposed to function calls before.
atomic_add_64_nv is implemented using atomic_fetchadd_64 on platforms
that provide it.
casptr, cas32, atomic_or_8, atomic_or_8_nv are completely removed as they
have no users.
atomic_mtx that is used to emulate 64-bit atomics on platforms that lack
them is defined only on those platforms.
As a result, platform specific opensolaris_atomic.S files have lost most of
their code. The only exception is i386 where the compat+contrib code
provides 64-bit atomics for userland use. That code assumes availability of
cmpxchg8b instruction. FreeBSD does not have that assumption for i386
userland and does not provide 64-bit atomics. Hopefully, this can and will
be fixed.
MFC after: 3 weeks
called in syscall context, so it must enter epoch itself. This
changeset originates from early version of the patch, and somehow
slipped to the final version.
Reported by: pho
As unp_internalize() processes the input control messages, it builds
an output mbuf chain containing the internalized representations of
those messages. In one special case, that of an empty SCM_RIGHTS
message, the message is simply discarded. However, the loop which
appends mbufs to the output chain assumed that each iteration would
produce an mbuf, resulting in a null pointer dereference if an empty
SCM_RIGHTS message was followed by a non-empty message.
Fix this by advancing the output mbuf chain tail pointer only if an
internalized control message was produced.
Reported by: syzbot+1b5cced0f7fad26ae382@syzkaller.appspotmail.com
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
This adds a TOE hook to allocate a KTLS session. It also recognizes
TLS mbufs in the socket buffer and sends those to the NIC using a TLS
work request to encrypt the record before segmenting it.
TOE TLS support must be enabled via the dev.t6nex.<N>.tls sysctl in
addition to enabling KTLS.
Reviewed by: np, gallatin
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D21891
This adds the glue to allocate TLS sessions and invokes it from
the TLS enable socket option handler. This also adds some counters
for active TOE sessions.
The TOE KTLS mode is returned by getsockopt(TLSTX_TLS_MODE) when
TOE KTLS is in use on a socket, but cannot be set via setsockopt().
To simplify various checks, a TLS session now includes an explicit
'mode' member set to the value returned by TLSTX_TLS_MODE. Various
places that used to check 'sw_encrypt' against NULL to determine
software vs ifnet (NIC) TLS now check 'mode' instead.
Reviewed by: np, gallatin
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D21891
ABI already guarantees the direction is forward. Note this does not take care
of i386-specific cld's.
Reviewed by: kib
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D21906
The PCI block in the adapter requires this field to be set to a valid
queue ID. It is not clear why it did not fail on all machines, but
the effect was that crypto operations reading input data via DMA
failed with an internal PCI read error on machines with 128G or more
of RAM.
Reported by: gallatin
Reviewed by: np
MFC after: 3 days
Sponsored by: Chelsio Communications
As noted by the commit message, callouts are now persistant
and should not be in the auto-zero section of the RQ's and SQ's.
This fixes an assert when using the TX completion event
factor feature with mlx5en(4).
Found by: gallatin@
MFC after: 3 days
Sponsored by: Mellanox Technologies
protected by IF_ADDR_LOCK(), which was a mutex, so that two simultaneous
if_setlladdr() can't execute. Later it was switched to IF_ADDR_RLOCK(),
likely by a mistake. Later it was switched to NET_EPOCH_ENTER(). Then I
incorrectly added NET_EPOCH_ASSERT() here.
In reality ifp->if_addr never goes away and never changes its length. So,
doing bcopy() in it is always "safe", meaning it won't dereference a wrong
pointer or write into someone's else memory. Of course doing two bcopy() in
parallel would result in a mess of two addresses, but net epoch doesn't
protect against that, neither IF_ADDR_RLOCK() did.
So for now, just remove the assertion and leave for later a proper fix.
Reported by: markj
During a CoW fault, we must check for both 4KB and 2MB mappings before
clearing PGA_WRITEABLE on the old mapping's page. Previously we were
only checking for 4KB mappings. This was missed in r344106.
MFC after: 3 days
Sponsored by: The FreeBSD Foundation
This matches the state prior to r353149 and fixes crashes with DRM
modules.
Reported and tested by: cy, garga, Krasznai Andras
Fixes: r353149 ("amd64 pmap: implement per-superpage locks")
Sponsored by: The FreeBSD Foundation
pmap_remove_l3() may remove the last mapping of a page, in which case
it must clear PGA_WRITEABLE.
Reported by: Jenkins, via lwhsu
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
As long as we support ZFS on 32-bit platforms we should do this for all
64-bit variables that are modified in a lockless fashion using atomic
operations. Otherwise, there is a risk of a reading a torn value.
Here is a rationale for why I am doing this in dmu_object_alloc_impl:
- it's very recent code
- the code deals with object IDs and a number of objects in a file
system can overflow 32 bits
- incorrect allocation of an object ID may result in hard to debug
problems
- fixing all plain reads of 64-bit atomic variables is not a trivial
undertaking to do in one shot, so I chose to do it incrementally
MFC after: 3 weeks
X-MFC after: r353301, r353176
Make sure the vnet_shutdown field is not set until after all
VNET_SYSUNINIT()'s in the SI_SUB_VNET_DONE subsystem have been
executed. Especially the vnet_if_return() functions requires that
if_move() is still operational.
Reported by: lwhsu@
MFC after: 1 week
Sponsored by: Mellanox Technologies
At the moment i386 does not provide 64-bit atomic operations in
userland. Exposing some atomic_*_64 defines can cause unnecessary
confusion.
Discussed with: kib
MFC after: 2 weeks
In case the implementation ever changes from using a chain of next pointers,
then changing the macro definition will be necessary, but changing all the
files that iterate over vm_map entries will not.
Drop a counter in vm_object.c that would have an effect only if the
vm_map entry count was wrong.
Discussed with: alc
Reviewed by: markj
Tested by: pho (earlier version)
Differential Revision: https://reviews.freebsd.org/D21882
|
This adds two implementations for each atomic_fcmpset_ and atomic_cmpset_
short and char functions, selectable at compile time for the target
architecture. By default, it uses a generic shift-and-mask to perform atomic
updates to sub-components of 32-bit words from <sys/_atomic_subword.h>.
However, if ISA_206_ATOMICS is defined it uses the ll/sc instructions for
halfword and bytes, introduced in PowerISA 2.06. These instructions are
supported by all IBM processors from POWER7 on, as well as the Freescale/NXP
e6500 core. Although the e5500 and e500mc both implement PowerISA 2.06 they
do not implement these instructions.
As part of this, clean up the atomic_(f)cmpset_acq and _rel wrappers, by
using macros to reduce code duplication.
ISA_206_ATOMICS requires clang or newer binutils (2.20 or later).
Differential Revision: https://reviews.freebsd.org/D21682
Acquire the inp lock before checking whether the socket is already bound,
and around updates to the inp_vflag field.
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D21867
About 11 minutes of poudriere -s -j 104 and probing on return value of
trylocks reveals that over 10% of attempts fail, which in turn means
there are more atomics performed than necessary.
Trylocking was there to try preventing migration, but it's not very likely
to happen if the lock is uncontested.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D21925
When epoch(9) was introduced to network stack, it was basically
dropped in place of existing locking, which was mutexes and
rwlocks. For the sake of performance mutex covered areas were
as small as possible, so became epoch covered areas.
However, epoch doesn't introduce any contention, it just delays
memory reclaim. So, there is no point to minimise epoch covered
areas in sense of performance. Meanwhile entering/exiting epoch
also has non-zero CPU usage, so doing this less often is a win.
Not the least is also code maintainability. In the new paradigm
we can assume that at any stage of processing a packet, we are
inside network epoch. This makes coding both input and output
path way easier.
On output path we already enter epoch quite early - in the
ip_output(), in the ip6_output().
This patch does the same for the input path. All ISR processing,
network related callouts, other ways of packet injection to the
network stack shall be performed in net_epoch. Any leaf function
that walks network configuration now asserts epoch.
Tricky part is configuration code paths - ioctls, sysctls. They
also call into leaf functions, so some need to be changed.
This patch would introduce more epoch recursions (see EPOCH_TRACE)
than we had before. They will be cleaned up separately, as several
of them aren't trivial. Note, that unlike a lock recursion the
epoch recursion is safe and just wastes a bit of resources.
Reviewed by: gallatin, hselasky, cy, adrian, kristof
Differential Revision: https://reviews.freebsd.org/D19111
The new sysctl was not added to the siftr.4 man page at the time.
This updates the man page, and removes one left over trailing whitespace.
Submitted by: Richard Scheffenegger
Reviewed by: bcr@
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D21619
This provides a framework to define a template describing
a set of "variables of interest" and the intended way for
the framework to maintain them (for example the maximum, sum,
t-digest, or a combination thereof). Afterwards the user
code feeds in the raw data, and the framework maintains
these variables inside a user-provided, opaque stats blobs.
The framework also provides a way to selectively extract the
stats from the blobs. The stats(3) framework can be used in
both userspace and the kernel.
See the stats(3) manual page for details.
This will be used by the upcoming TCP statistics gathering code,
https://reviews.freebsd.org/D20655.
The stats(3) framework is disabled by default for now, except
in the NOTES kernel (for QA); it is expected to be enabled
in amd64 GENERIC after a cool down period.
Reviewed by: sef (earlier version)
Obtained from: Netflix
Relnotes: yes
Sponsored by: Klara Inc, Netflix
Differential Revision: https://reviews.freebsd.org/D20477
Remove the now obsolete vnet_state field. This greatly simplifies the
detection of VNET shutdown and avoids code duplication.
Discussed with: bz@
MFC after: 1 week
Sponsored by: Mellanox Technologies
