Pull in r211435 from upstream llvm trunk (by Benjamin Kramer):
Legalizer: Add support for splitting insert_subvectors.
We handle this by spilling the whole thing to the stack and doing the
insertion as a store.
PR19492. This happens in real code because the vectorizer creates
v2i128 when AVX is enabled.
This fixes a "fatal error: error in backend: Do not know how to split
the result of this operator!" message encountered during compilation of
the net-p2p/libtorrent-rasterbar port.
Reported by: Evgeniy <iron@mail.ua>
MFC r267705:
Add the llvm patch for r267704.
Don't denounce peripherals on system shutdown. Together with r267321
(MFCed to stable/10 in r267775), we're now back to the pre-r228483
level of default verbosity. This in turn again typically allows for
reading information that userland might have printed on the screen
before initiating a halt, but still permits to debug potential device
shutdown problems on system shutdown via CAM_DEBUG etc.
Reviewed by: mav
Sponsored by: Bally Wulff Games & Entertainment GmbH
Avoid the USB device disconnected and controller shutdown clutter on system
shutdown by putting the former under !rebooting and turning the latter into
debug messages.
Reviewed by: hps
Sponsored by: Bally Wulff Games & Entertainment GmbH
Honor jails for unbound SCTP sockets when selecting source addresses,
reporting IP-addresses to the peer during the handshake, adding
addresses to the host, reporting the addresses via the sysctl
interface (used by netstat, for example) and reporting the
addresses to the application via socket options.
This issue was reported by Bernd Walter.
Consider the total number of descriptors available (and not just those
that are ready to be reclaimed) when deciding whether to resume tx after
a stall.
Allocating new bucket for bucket zone, never take it from the zone itself,
since it will almost certanly fail. Take next bigger zone instead.
This situation should not happen with original bucket zones configuration:
"32 Bucket" zone uses "64 Bucket" and vice versa. But if "64 Bucket" zone
lock is congested, zone may grow its bucket size and start biting itself.
Implement simple direct-mapped cache for popular filesystem identifiers to
avoid congestion on global mountlist_mtx mutex in vfs_busyfs(), while
traversing through the list of mount points.
This change significantly improves NFS server scalability, since it had
to do this translation for every request, and the global lock becomes quite
congested.
This code is more optimized for relatively small number of mount points.
On systems with hundreds of active mount points this simple cache may have
many collisions. But the original traversal code in that case should also
behave much worse, so we are not loosing much.
patch: add dry-run alias for compatibility with other implementations.
Other implementations of patch(1), including GNU patch and "svn patch"
have a --dry-run option which does the same as our -C or --check
option.
Add a new alias to make our implementation more compatible.
patch: unsign the line length to avoid overflows.
Patch(1) uses a short int for the line length, which is usually
sufficient for regular diffs, but makes no effort to signal
when there is an overflow.
Change the line length to an unsigned short int to better use
the fact that a length is never negative. The change is loosely
inspired on a related change in DragonFly, but we avoid spending
more memory than necessary.
While here adjust the messages to be clearer on what is happening.
Use atomics to modify numvnodes variable.
This allows to mostly avoid lock usage in getnewvnode_[drop_]reserve(),
that reduces number of global vnode_free_list_mtx mutex acquisitions
from 4 to 2 per NFS request on ZFS, improving SMP scalability.
Split RPC pool threads into number of smaller semi-isolated groups.
Old design with unified thread pool was good from the point of thread
utilization. But single pool-wide mutex became huge congestion point
for systems with many CPUs. To reduce the congestion create several
thread groups within a pool (one group for every 6 CPUs and 12 threads),
each group with own mutex. Each connection during its registration is
assigned to one of the groups in round-robin fashion. File affinify
code may still move requests between the groups, but otherwise groups
are self-contained.
Introduce new per-thread lock to protect the list of requests.
This allows to slightly simplify svc_run_internal() code: if we processed
all the requests in a queue, then we know that new one will not appear.
For some UDP packets (for example with 200 byte payload) and IP options,
the IP header and the UDP header are not in the same mbuf.
Add code to in_delayed_cksum() to deal with this case.
MFC r265713:
Use KASSERTs as suggested by glebius@
Don't free an mbuf twice. This only happens in very rare error
cases where the peer sends illegal sequencing information in
DATA chunks for an existing association.
Handle an edge case of address management similar to TCP.
This needs to be reconsidered when the address handling
will be reimplemented.
The patch is from rrs@.
* Provide information in error causes in ASCII instead of
proprietary binary format.
* Add support for a diagnostic information error cause.
The code is sysctlable and the default is 0, which
means it is not sent.
This is joint work with rrs@.
Put the offset of the CRC32C in csum_data instead of 0.
The virtio driver needs the offset to be stored in csum_data,
like in the case for UDP and TCP.
The virtio problem was reported by
Niu Zhixiong <kaiaixi@gmail.com>, who helped in debugging
and testing the patch.
