When estimating working set size, measure only allocation batches, not free
batches. Allocation and free patterns can be very different. For example,
ZFS on vm_lowmem event can free to UMA few gigabytes of memory in one call,
but it does not mean it will request the same amount back that fast too, in
fact it won't.
Update working set size on every reclamation call, shrinking caches faster
under pressure. Lack of this caused repeating vm_lowmem events squeezing
more and more memory out of real consumers only to make it stuck in UMA
caches. I saw ZFS drop ARC size in half before previous algorithm after
periodic WSS update decided to reclaim UMA caches.
Introduce voluntary reclamation of UMA caches not used for a long time. For
each zdom track longterm minimal cache size watermark, freeing some unused
items every UMA_TIMEOUT after first 15 minutes without cache misses. Freed
memory can get better use by other consumers. For example, ZFS won't grow
its ARC unless it see free memory, since it does not know it is not really
used. And even if memory is not really needed, periodic free during
inactivity periods should reduce its fragmentation.
Reviewed by: markj, jeff (previous version)
MFC after: 2 weeks
Sponsored by: iXsystems, Inc.
Differential Revision: https://reviews.freebsd.org/D29790
When processing INIT and INIT-ACK information, also during
COOKIE processing, delete the current association, when it
would end up in an inconsistent state.
MFC after: 3 days
PR#255523 reported that a file copy for a file with a large hole
to EOF on ZFS ran slowly over NFSv4.2.
The problem was that vn_generic_copy_file_range() would
loop around reading the hole's data and then see it is all
0s. It was coded this way since UFS always allocates a data
block near the end of the file, such that a hole to EOF never exists.
This patch modifies vn_generic_copy_file_range() to check for a
ENXIO returned from VOP_IOCTL(..FIOSEEKDATA..) and handle that
case as a hole to EOF. asomers@ confirms that it works for his
ZFS test case.
PR: 255523
Tested by: asomers
Reviewed by: asomers
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D30076
Not all interrupt controllers enable IPIs by default as the Arm
GIC specs make it an implementation defined option. As at least two
hypervisors have also previously masked the IPIs on boot.
As we already enable these IPIs on the non-boot CPUs it is expected
this is a safe operation.
Differential Revision: https://reviews.freebsd.org/D26975
This will allow us to allocate an unmapped memory resource, then
later map it with a specific memory attribute.
This is also needed for virtio with the modern PCI attachment.
Reviewed by: kib (via D29723)
Sponsored by: Innovate UK
Differential Revision: https://reviews.freebsd.org/D29694
It is defined as a uint64_t in the UEFI spec. As it's not used as a
pointer by the kernel follow this and define it as the same in the
kernel.
Reviewed by: kib, manu, imp
Sponsored by: Innovate UK
Differential Revision: https://reviews.freebsd.org/D29759
On arm64 we currently use a non-posted write for device memory, however
we should move to use posted writes. This is expected to work on most
hardware, however we will need to support a non-posted option for some
broken hardware.
Reviewed by: imp, manu, bcr (manpage)
Differential Revision: https://reviews.freebsd.org/D29722
Summary:
Since PCPU can live in a GPR for a while longer, let it, rather than
re-getting it in yet another register. MFSPR is an expensive operation,
12 clock latency on POWER9, so the fewer operations we need, the better.
Since the check is tightly coupled to the fetch, by reducing the number
of fetch+check, we reduce the stalls, and improve the performance
marginally. Buildworld was measured at a ~5-7% improvement on a single
run.
Reviewed By: nwhitehorn
Differential Revision: https://reviews.freebsd.org/D30003
It has to be zeroed before committing it to device.
We do that by allocating it with M_ZERO, but there was no
memory barrier or cache flush to ensure its sees it zeroed.
This fixes MSIX on LS1028A SoC.
Submitted by: Kornel Duleba <mindal@semihalf.com>
Reviewed by: andrew
Obtained from: Semihalf
Sponsored by: Alstom Group
Differential Revision: https://reviews.freebsd.org/D30033
When sending an IPI, if a previous IPI is still pending delivery,
native_lapic_ipi_vectored() waits for the previous IPI to be sent.
We've seen a few inexplicable panics with the current timeout of 50 ms.
Increase the timeout to 1 second and make it tunable.
No hardware specification mentions a timeout in this case; I checked
the Intel SDM, Intel MP spec, and Intel x2APIC spec. Linux and illumos
wait forever. In Linux, see __default_send_IPI_shortcut() in
arch/x86/kernel/apic/ipi.c. In illumos, see apic_send_ipi() in
usr/src/uts/i86pc/io/pcplusmp/apic_common.c. However, misbehaving hardware
could hang the system if we wait forever.
Reviewed by: mav kib
MFC after: 1 week
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D29942
Its use is for cases where some filler is needed for cmd, or we need an
indication that there were no cmd supplied, and so on.
Reviewed by: jhb
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D29935
Filter on fifos is real filter for the object, and not a filesystem
events filter like EVFILT_VNODE.
Reported by: markj using syzkaller
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
A testing on the real hardware uncovered an issue, and since I do not have
access to the machine, disable until the bug can be fixed.
Reported by: "Pieper, Jeffrey E" <jeffrey.e.pieper@intel.com>
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
When setting up trampoline mapping for LA57 switcher, it is possible
that TLB still has some random mapping at that address.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Drivers can specify padding of received frames with iri_pad field.
This can be used to enforce ip alignment by hardware.
Iflib ignored that padding when processing small frames,
which rendered this feature inoperable.
I found it while writing a driver for a NIC that can ip align
received packets. Note that this doesn't change behavior of existing
drivers as they all set iri_pad to 0.
Submitted by: Kornel Duleba <mindal@semihalf.com>
Reviewed by: gallatin
Obtained from: Semihalf
Sponsored by: Alstom Group
Differential Revision: https://reviews.freebsd.org/D30009
If we reassemble a packet we modify the IP header (to set the length and
remove the fragment offset information), but we failed to update the
checksum. On certain setups (mostly where we did not re-fragment again
afterwards) this could lead to us sending out packets with incorrect
checksums.
PR: 255432
MFC after: 1 week
Sponsored by: Rubicon Communications, LLC ("Netgate")
Differential Revision: https://reviews.freebsd.org/D30026
The original implementation only supports getting the address from legacy
BIOS (by searching for the SMBIOS_SIG pattern in a fixed address space).
Try to get the SMBIOS table from EFI through efirt (EFI Runtime Services)
firstly. Continue to search in the legacy BIOS if a NULL address is
returned from EFI.
By this way the ipmi function supports both legacy BIOS and UEFI systems.
Reviewed by: dab, vangyzen
MFC after: 1 week
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D30007
There is no need to panic in if_transmit if the checksums requested are
inconsistent with the frame being transmitted. This typically indicates
that the kernel and driver were built with different INET/INET6 options,
or there is some other kernel bug. The driver should just throw away
the requests that it doesn't understand and move on.
MFC after: 1 week
Sponsored by: Chelsio Communications
This is just clerical work to ease bug triage and may be used to set
expectations around the ability for anyone in the community to perform
testing and development on older parts.
Approved by: erj
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D29876
pipe_poll() may add the calling thread to the selinfo lists of both ends
of a pipe. It is ok to do this for the local end, since we know we hold
a reference on the file and so the local end is not closed. It is not
ok to do this for the remote end, which may already be closed and have
called seldrain(). In this scenario, when the polling thread wakes up,
it may end up referencing a freed selinfo.
Guard the selrecord() call appropriately.
Reviewed by: kib
Reported by: syzkaller+KASAN
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D30016
Teach poll(2) to support Linux-style POLLRDHUP events for sockets, if
requested. Triggered when the remote peer shuts down writing or closes
its end.
Reviewed by: kib
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D29757
Provide wrapper for the rnh_walktree_from() rib callback.
As currently `struct rib_head` is considered internal to the
routing subsystem, this wrapper is necessary to maintain isolation
from the external code.
Differential Revision: https://reviews.freebsd.org/D29971
MFC after: 1 week
Add suspend/resume callbacks to the driver and a live reset built around
them. This commit covers the basic NIC and future commits will expand
this functionality to other stateful parts of the chip. Suspend and
resume operate on the chip (the t?nex nexus device) and affect all its
ports. It is not possible to suspend/resume or reset individual ports.
All these operations can be performed on a running NIC. A reset will
look like a link bounce to the networking stack.
Here are some ways to exercise this functionality:
/* Manual suspend and resume. */
# devctl suspend t6nex0
# devctl resume t6nex0
/* Manual reset. */
# devctl reset t6nex0
/* Manual reset with driver sysctl. */
# sysctl dev.t6nex.0.reset=1
/* Automatic adapter reset on any fatal error. */
# hw.cxgbe.reset_on_fatal_err=1
Suspend disables the adapter (DMA, interrupts, and the port PHYs) and
marks the hardware as unavailable to the driver. All ifnets associated
with the adapter are still visible to the kernel but operations that
require hardware interaction will fail with ENXIO. All ifnets report
link-down while the adapter is suspended.
Resume will reattach to the card, reconfigure it as before, and recreate
the queues servicing the existing ifnets. The ifnets are able to send
and receive traffic as soon as the link comes back up.
Reset is roughly the same as a suspend and a resume with at least one of
these events in between: D0->D3Hot->D0, FLR, PCIe link retrain.
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Commit aad780464f added a function called nfscl_delegreturnvp()
to return delegations during the NFS VOP_RECLAIM().
The function erroneously assumed that nm_clp would
be non-NULL. It will be NULL for NFSV4.0 mounts until
a regular file is opened. It will also be NULL during
vflush() in nfs_unmount() for a forced dismount.
This patch adds a check for clp == NULL to fix this.
Also, since it makes no sense to call nfscl_delegreturnvp()
during a forced dismount, the patch adds a check for that
case and does not do the call during forced dismounts.
PR: 255436
Reported by: ish@amail.plala.or.jp
MFC after: 2 weeks
It was reported that a NFSv4.1 Linux client mount against
a FreeBSD12 server was hung, with the TCP connection in
CLOSE_WAIT state on the server.
When a NFSv4.1/4.2 mount is done and the back channel is
bound to the TCP connection, the soclose() is delayed until
a new TCP connection is bound to the back channel, due to
a reference count being held on the SVCXPRT structure in
the krpc for the socket. Without the soclose() call, the socket
will remain in CLOSE_WAIT and this somehow caused the Linux
client to hang.
This patch adds calls to soshutdown(.., SHUT_WR) that
are performed when the server side krpc sees that the
socket is no longer usable. Since this can be done
before the back channel is bound to a new TCP connection,
it allows the TCP connection to proceed to CLOSED state.
PR: 254590
Reported by: jbreitman@tildenparkcapital.com
Reviewed by: tuexen
Comments by: kevans
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D29526
* Fix 82574 Link Status Changes, carrying the OTHER mask bit around as
needed.
* Move igb-class LSC re-arming out of FAST back into the handler.
* Clarify spurious/other interrupt re-arms in FAST.
In MSI-X mode, 82574 and igb-class devices use an interrupt filter to
handle Link Status Changes. We want to do LSC re-arms in the handler
to take advantage of autoclear (EIAC) single shot behavior.
82574 uses 'Other' in ICR and IMS for LSC interrupt types when in MSI-X
mode, so we need to set and re-arm the 'Other' bit during attach and
after ICR reads in the FAST handler if not an LSC or after handling on
LSC due to autoclearing.
This work was primarily done to address the referenced PR, but inspired
some clarification and improvement for igb-class devices once the
intentions of previous bug fix attempts became clearer.
PR: 211219
Reported by: Alexey <aserp3@gmail.com>
Tested by: kbowling (I210 lagg), markj (I210)
Approved by: markj
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D29943
Currently, most of the rib(9) KPI does not use rnh pointers, using
fibnum and family parameters to determine the rib pointer instead.
This works well except for the case when we initialize new rib pointers
during fib growth.
In that case, there is no mapping between fib/family and the new rib,
as an entirely new rib pointer array is populated.
Address this by delaying fib algo initialization till after switching
to the new pointer array and updating the number of fibs.
Set datapath pointer to the dummy function, so the potential callers
won't crash the kernel in the brief moment when the rib exists, but
no fib algo is attached.
This change allows to avoid creating duplicates of existing rib functions,
with altered signature.
Differential Revision: https://reviews.freebsd.org/D29969
MFC after: 1 week
During early qemu development, the /soc node was marked as compatible
with "riscv-virtio-soc" instead of "simple-bus".
This was changed in qemu 53f54508dae6 in Sep 2018, and predates the
baseline required qemu version (5.0) for riscv by a wide margin.
The generic simplebus code handles attachment in all cases nowadays.
Sponsored by: Tag1 Consulting, Inc.
Reviewed by: jrtc27, mhorne
Differential Revision: https://reviews.freebsd.org/D30011
A lot more generic cam related things are done in mmc_sim so this simplify
the driver a lot.
Differential Revision: https://reviews.freebsd.org/D27487
Reviewed by: kibab
A lot more generic cam related things are done in mmc_sim so this simplify
the driver a lot.
Differential Revision: https://reviews.freebsd.org/D27486
Reviewed by: imp
