For the dynamic I/O scheduler, make the TRIM stuff also count against
read bias so we do reads in preference to TRIMs. This helps a lot when
many trims are delivered at once from the upper layers as they tend to
delay READs due to priority inversion in the code today.
The non iosched case will be fixed when the trim comibing changes
needed for nvme come in later this year.
ian [Thu, 26 Jul 2018 20:03:11 +0000 (20:03 +0000)]
Re-apply r336625 which was reverted with r336638, now that the underlying
pw_scan(3) has been fixed in a way that doesn't perturb other callers of
it or the getpwnam(3) family.
Make pw(8) showuser work the same with or without -R <path> for non-root
users. Without -R, pw(8) uses getpwnam(3), which will open master.passwd
for the root user or passwd for non-root users. With -R <path> pw(8) was
always opening <path>/master.passwd, which would fail for a non-root user,
then falsely claim the userid you're trying to show doesn't exist.
Now for a non-root user it opens <path>/passwd, and populates the fields in
the returned struct passwd which aren't present in that file with well-known
canonical values, which duplicates the behavior of getpwnam(3). The net
effect is that the showuser output is identical whether using -R or not.
ian [Thu, 26 Jul 2018 18:34:38 +0000 (18:34 +0000)]
Make pw_scan(3) more compatible with getpwent(3) et. al. when processing
data from /etc/passwd rather than /etc/master.passwd.
The libc getpwent(3) and related functions automatically read master.passwd
when run by root, or passwd when run by a non-root user. When run by non-
root, getpwent() copes with the missing data by setting the corresponding
fields in the passwd struct to known values (zeroes for numbers, or a
pointer to an empty string for literals). When libutil's pw_scan(3) was
used to parse a line without the root-accessible data, it was leaving
garbage in the corresponding fields.
These changes rename the static pw_init() function used by getpwent() and
friends to __pw_initpwd(), and move it into pw_scan.c so that common init
code can be shared between libc and libutil. pw_scan(3) now calls
__pw_initpwd() before __pw_scan(), just like the getpwent() family does, so
that reading an arbitrary passwd file in either format and parsing it with
pw_scan(3) returns the same results as getpwent(3) would.
This also adds a new pw_initpwd(3) function to libutil, so that code which
creates passwd structs from scratch in some manner that doesn't involve
pw_scan() can initialize the struct to the values expected by lots of
existing code, which doesn't expect to encounter NULL pointers or garbage
values in some fields.
In the V5 ABI the flags are EF_ARM_ABI_FLOAT_HARD and
EF_ARM_ABI_FLOAT_SOFT. The flags have the same values as the legacy GCC
flags EF_ARM_VFP_FLOAT and EF_ARM_SOFT_FLOAT respectively.
The legacy names are kept for compatibility.
Reported by: Peter Smith (Linaro)
Reviewed by: imp
Sponsored by: The FreeBSD Foundation
SPE ABI uses the soft-float ABI, which splits doubles into two words. As such,
fabs(3) cannot work on a double directly. It's too costly to convert the
argument pair into a single double to use efdabs, so clear the top bit of the
high word, which is the sign bit.
Although the ffs (and later msdosfs) implementation in makefs is
independent of the one in kernel, it makes sense to keep differences to
a minimum in order to ease comparison and porting changes across.
Submitted by: Siva Mahadevan
Sponsored by: The FreeBSD Foundation
cxgbe(4): Consider rateunit before ratemode when displaying information
about a traffic class. This matches the order in which the firmware
evaluates unit and mode internally.
cxgbe(4): Better defaults for all cl-rl rate limiters.
Start in "class" instead of "flow" mode. This eliminates the need to
specify an MTU, which is not available that early anyway. It also
allows the user to manually configure ch-rl rate limiting after attach.
This used to fail because ch-rl isn't supported if cl-rl "flow" mode is
configured.
Set all traffic classes to 1Gbps during initialization. The goal is to
start off with _any_ valid configuration and 1Gbps works even for
gigabit cards.
Ignore Device Paths in the Boot Info that don't have Media path
nodes. These show up in default entries on SuperMicro motherboards and
elsewhere. Before, we couldn't find a block device associated with the
device path and return BAD_CHOICE which was an instant
failure. However, a VendHw node isn't specifc, so when we don't find a
media path, return NOT_SPECIFIC so that the rest of the algorithms
work.
Raise a proper SIGTRAP / TRAP_TRACE signal for a PT_STEP step on arm.
Previously, a step by PT_STEP resulted in no signal being raised to
the debugger so that a step was silently completed with the program
continuing to execute after the step. Fix by raising a SIGTRAP
signal with TRAP_TRACE as the signal code.
To simplify the error handling cases (if ptrace_clear_single_step()
fails, etc.) move the handling of PTRACE_BREAKPOINT into the
gdb_trapper() function. If ptrace_clear_single_step() fails,
gdb_trapper() won't claim the fault, and the default case of
SIGILL / ILL_OPC will be used.
cxgbetool(8): Require and validate only those inputs that are applicable
to the type of rate limiter being configured. For example, the class
WRR scheduler doesn't need any kbps limits (it just needs the weights
for each class), the channel scheduler doesn't need anything except the
aggregate kbps to limit the channel to, and so on.
MFC after: 3 days
Sponsored by: Chelsio Communications
dim [Wed, 25 Jul 2018 17:14:05 +0000 (17:14 +0000)]
Add a few forgotten files to ObsoleteFiles.inc:
* libcasper.so.0, which started in /usr/lib, then moved to /lib, but
was later replaced by libcasper.so.1
* 32-bit versions of static casper libraries
* 32-bit versions of static stand libraries
* 32-bit versions of static ifc(onfig) libraries
It works excellent, but KDB disassembler and DTrace FBT provider for
RISC-V do lack support for it. They currently handle 4-byte instructions
only, while C-compressed ISA extension introduces 2-byte instructions
freely mixing them together.
The original intention was 4 columns but with a usable a result. In
practice this was not the case. Increase the number of columns to 5
until humanize_number learns alternative ways of presenting the number.
Remove support for QLNX_RCV_IN_TASKQ - i.e., Rx only in TaskQ.
Added support for LLDP passthru
Upgrade ECORE to version 8.33.5.0
Upgrade STORMFW to version 8.33.7.0
Added support for SRIOV
All supported FreeBSD build host versions have backtrace.h, so we can
just eliminate that test. For futimes() we can test the compiler's
built-in __FreeBSD__ major version rather than relying on including
osreldate.h. This should reduce the frequency with which Clang gets
rebuilt when building world.
Reviewed by: dim
Sponsored by: The FreeBSD Foundation
This is an OFW initrd module that would load the initrd from device tree
parameters and give the to the md driver.
With this patch, it is possible to pass a rootfs image through kexec in PowerNV
mode (powerpc64). In order to user it, you should set the MD_ROOT_MEM option in
your kernel configuration.
andrew [Tue, 24 Jul 2018 16:31:16 +0000 (16:31 +0000)]
As with DPCPU create VNET_DEFINE_STATIC for when a variable needs to be
declaired static. This will allow us to change the definition on arm64
as it has the same issues described in r336349.
In C remquol() and thus also in C remainderl(), don't clobber the sign bit
of NaNs before possible returning a NaN.
The remquo*() and remainder*() functions should now give bitwise identical
results across arches and implementations, and bitwise consistent results
(with lower precisions having truncated mantissas) across precisions. x86
already had consistency across amd64 and i386 and precisions by using the
i387 consistently and normally not using the C versions. Inconsistencies
for C reqmquol() were first detected on sparc64.
Remove double second clearing of the sign bit and extra blank lines.
Fix the conversion to use nan_mix() in r336362. fmod*(x, y),
remainder*(x, y) and remquo*(x, y, quo) were broken for y = 0 by changing
multiplication by y to addition of y. (When y is 0, the result should be
NaN but became 1 for finite x.)
Use a new macro nan_mix_op() to give more control over the mixing, and
expand comments.
Recent re-testing missed finding this bug since I only tested the macro
version on amd64 and i386 and these arches don't use the C versions (they
use either asm versions or builtins).
Deprecate jedec_ts(4) and point users to jedec_dimm(4) instead
jedec_dimm(4) is a superset of the functionality of jedec_ts(4). Mark
jedec_ts(4) as removed in FreeBSD 12, and include a pointer to the migration
instructions in the jedec_dimm(4) manpage, in both the jedec_ts(4) code and
the jedec_ts(4) manpage. Add a note to the jedec_dimm(4) manpage about the
fact that it is a superset of jedec_ts(4).
This change will be MFCed to stable/11 and stable/10; the followup change
to actually remove jedec_ts(4) from -HEAD will not.
Finalize the boot manager protocol support for next-stage boot
loading.
If we are booting in a conforming UEFI Boot Manager Environment, then
use the BootCurrent variable to find the BootXXXX we're using. Once we
find that, then if it contains more than one EFI_DEVICE_PATH in its
what to boot section, try to use the last one as the kernel to
load. This will also set the default root partition as well. If
there's only one path, or if there's an error along the way, assume
that nothing specific was specified and revert to the old
algorithm. If something was specified, but not found, then fail the
boot. Otherwise you that, specific thing. On FreeBSD, this can be set
using efibootmgr -l <loader> -k <kernel>. We try a few variations of
kernel to cope with the fact that UEFI comes from a DOS world where
paths might be upper case and/or contain back-slashes.
Note: In an ideal world, we'd work out where we are in chain loading
by looking at the passed-in image handle and doing name
matching. However, that's unreliable since at least boot1.efi booted
images don't have that, hence the assumption that loader.efi needs to
load the last thing on the list, if possible.
The reason we fail for something specific is so that we can fully
participate in the UEFI Boot Manager Protocol and fail over to the
next item in the list of BootOrder choices when something goes wrong
at this stage.
This implements was was talked about in freebsd-arch@ last year
https://docs.freebsd.org/cgi/getmsg.cgi?fetch=3576+0+archive/2017/freebsd-arch/20171022.freebsd-arch
and documented in full (after changed resulting from the discussion) in
https://docs.google.com/document/d/1aK9IqF-60JPEbUeSAUAkYjF2W_8EnmczFs6RqCT90Jg/edit#
although one or two minor details may have been modified in this
implementation to make it work, and the ZFS MEDIA PATH extension isn't
implemented. This does not yet move things to ESP:\efi\freebsd\loader.efi.
Lookup a block device by it's device path. We use a 'loose' lookup
whereby we scan forward to the first Media Path portion of the device
path, then look at all our handles for one whose first Media Path
matches. This will also work if the device path pointed to has a
following file path (or paths) as that's ignored. It assumes that
there's only one media path node that describes the entire device,
which is true as of the latest UEFI spec (2.7 Errata A) as far as I've
been able to determine.
Returns true if the first node pointed to by devpath1 is identical to
the first node pointed to by devpath2, with care taken to not read
past the end of the valid parts of either devpath1 or
devpath2. Otherwise, returns false.
Store the number of handles we get back in efipart_nhandles rather
than the number of bytes. Don't divide by the element size every time
we have to iterate. Eliminate now-unused variables.
Takes a generic device path as its input. Scans through it to find the
first media_path node in it and returns a pointer to it. If none is
found, NULL is returned.
Compilers may define multiple variants of architecture-specific macros
(for example, both __x86_64 and __x86_64__). Add a note that the macros
documented in arch.7 are the preferred ones for FreeBSD.
Temporarily decompress a copy of a crash dump compressed with either
gzip or zstd and run various tools against the decompressed copy while
generating the crash information. The uncompressed copy is deleted when
the script exits.
Note that crashinfo is enabled by default, so this will attempt to
decompress the most recent compressed crash dump after a crash that
generates a compressed crash dump. Users who wish to only do offline
analysis of compressed crash dumps can disable crashinfo in rc.conf.
change interrupt event's list of handlers from TAILQ to CK_SLIST
The primary reason for this commit is to separate mechanical and nearly
mechanical code changes from an upcoming fix for unsafe teardown of
shared interrupt handlers that have only filters (see D15905).
The technical rationale is that SLIST is sufficient. The only operation
that gets worse performance -- O(n) instead of O(1) is a removal of a
handler, but it is not a critical operation and the list is expected to
be rather short.
Additionally, it is easier to reason about SLIST when considering the
concurrent lock-free access to the list from the interrupt context and
the interrupt thread.
CK_SLIST is used because the upcoming change depends on the memory order
provided by CK_SLIST insert and the fact that CL_SLIST remove does not
trash the linkage in a removed element.
While here, I also fixed a couple of whitespace issues, made code under
ifdef notyet compilable, added a lock assertion to ithread_update() and
made intr_event_execute_handlers() static as it had no external callers.
Fix the attempt to see if we're overriding the console in the command
line args. I had thought console would be NULL, but it's efi. Set it
to efi (as a clue) before we initialize the console, then test it to
see if it changed on the command line to do the automatic
override. This gets my serial console back.
Older zfs boot blocks don't support symlinks. install the link to
zfsloader as a hard link. While newer ones do, the whole point of the
link was to transition to the new world order smoothly. A hard link is
less flexible, but it works and will result in fewer bumps. Adjust
UPDATING entry to match.
ian [Sun, 22 Jul 2018 23:41:40 +0000 (23:41 +0000)]
Make pw(8) showuser work the same with or without -R <path> for non-root
users. Without -R, pw(8) uses getpwnam(3), which will open master.passwd
for the root user or passwd for non-root users. With -R <path> pw(8) was
always opening <path>/master.passwd, which would fail for a non-root user,
then falsely claim the userid you're trying to show doesn't exist.
Now for a non-root user it opens <path>/passwd and zeroes out the 3 fields
that aren't available in the passwd file, which duplicates the behavior of
getpwnam(3). The net effect is that the showuser output is identical
whether using -R or not.
The change made in r336593 assumes that the build is happening in a
svn checkout resulting in misleading debug output. Check that we're
actually working in an svn checkout before proceeding to call svn.
ian [Sun, 22 Jul 2018 22:34:20 +0000 (22:34 +0000)]
Set the pw_class field to NULL when scanning the non-master passwd file.
This avoids a null pointer deref in pw_dup(), which assumes that all
pointers are either NULL or valid.