mckusick [Thu, 7 Aug 2014 00:32:23 +0000 (00:32 +0000)]
MFC of r269303:
When restoring a UFS dump onto a ZFS filesystem, an assertion in
restore was failing because ZFS was reporting a blocksize that was
not a multiple of 1024. Replace restore's failed assertion with
code that writes restored files in a blocksize that works for
restore (a multiple of 1024) despite being non-optimal for ZFS.
marius [Tue, 5 Aug 2014 16:30:13 +0000 (16:30 +0000)]
MFC: r260457
The changes in r233781 (MFCed to stable/9 in r235515) attempted to make
logging during a machine check exception more readable. In practice they
prevented all logging during a machine check exception on at least some
systems. Specifically, when an uncorrected ECC error is detected in a DIMM
on a Nehalem/Westmere class machine, all CPUs receive a machine check
exception, but only CPUs on the same package as the memory controller for
the erroring DIMM log an error. The CPUs on the other package would complete
the scan of their machine check banks and panic before the first set of CPUs
could log an error. The end result was a clearer display during the panic
(no interleaved messages), but a crashdump without any useful info about
the error that occurred.
To handle this case, make all CPUs spin in the machine check handler
once they have completed their scan of their machine check banks until
at least one machine check error is logged. I tried using a DELAY()
instead so that the CPUs would not potentially hang forever, but that
was not reliable in testing.
While here, don't clear MCIP from MSR_MCG_STATUS before invoking panic.
Only clear it if the machine check handler does not panic and returns
to the interrupted thread.
Intel desktop Haswell CPUs may report benign corrected parity errors (see
HSD131 erratum in [1]) at a considerable rate. So filter these (default),
unless logging is enabled. Unfortunately, there really is no better way to
reasonably implement suppressing these errors than to just skipping them
in mca_log(). Given that they are reported for bank 0, they'd need to be
masked in MSR_MC0_CTL. However, P6 family processors require that register
to be set to either all 0s or all 1s, disabling way more than the one error
in question when using all 0s there. Alternatively, it could be masked for
the corresponding CMCI, but that still wouldn't keep the periodic scanner
from detecting these spurious errors. Apart from that, register contents of
MSR_MC0_CTL{,2} don't seem to be publicly documented, neither in the Intel
Architectures Developer's Manual nor in the Haswell datasheets.
Note that while HSD131 actually is only about C0-stepping as of revision
014 of the Intel desktop 4th generation processor family specification
update, these corrected errors also have been observed with D0-stepping
aka "Haswell Refresh".
markj [Tue, 5 Aug 2014 01:53:14 +0000 (01:53 +0000)]
MFC r267759, r267761
r267759:
Fix a couple of bugs on amd64 when fetching probe arguments beyond the
first five for probes entered through a UD fault (i.e. FBT probes).
Specifically, handle the fact that dtrace_invop_callsite must be
16 byte-aligned and thus may not immediately follow the call to
dtrace_invop() in dtrace_invop_start(). Also fetch register arguments and
the stack pointer through a struct trapframe instead of a struct reg.
r267761:
Fix some bugs when fetching probe arguments in i386. Firstly ensure that
the 4 byte-aligned dtrace_invop_callsite can be found and that it
immediately follows the call to dtrace_invop(). Secondly, fix some pointer
arithmetic to account for differences between struct i386_frame and illumos'
struct frame. Finally, ensure that dtrace_getarg() isn't inlined. It works
by following a fixed number of frame pointers to the probe site, so inlining
breaks it.
markj [Tue, 5 Aug 2014 00:25:19 +0000 (00:25 +0000)]
MFC r267706:
Allow creation of SDT probes from a module in which no providers are
defined. This ensures that the sdt:zfs:: probes appear despite the fact
the sdt provider is defined in the kernel rather than in zfs.ko.
markj [Mon, 4 Aug 2014 21:41:01 +0000 (21:41 +0000)]
MFC r256822:
When fetching function arguments out of a frame on amd64, explicitly select
the register based on the argument index rather than relying on the fields
in struct reg to be in the right order. This assumption is incorrect on
FreeBSD and generally led to bogus argument values for the sixth argument
of PID and USDT probes; the first five are passed directly to dtrace_probe()
via the fasttrap trap handler and so were correctly handled.
markj [Mon, 4 Aug 2014 15:36:23 +0000 (15:36 +0000)]
MFC r256571:
Add a function, memstr, which can be used to convert a buffer of
null-separated strings to a single string. This can be used to print the
full arguments of a process using execsnoop (from the DTrace toolkit) or
with the following one-liner:
Note that this relies on the process arguments being cached via the struct
proc, which means that it will not work for argvs longer than
kern.ps_arg_cache_limit. However, the following rather non-portable
script can be used to extract any argv at exec time:
truckman [Fri, 1 Aug 2014 15:08:47 +0000 (15:08 +0000)]
MFC r268780
Nuke the never-used RF_TIMESHARE feature, reducing the complexity of the
code. The consensus on arch@ is that this feature might have been useful
in the distant past, but is now just unnecessary bloat.
The int_rman_activate_resource() and int_rman_deactivate_resource()
functions become trivial, so manually inline them.
The special deferred handling of RF_ACTIVE is no longer needed in
reserve_resource_bound(), so eliminate the associated code at the
end of the function.
These changes reduce the object file size by more than 500 bytes on i386.
Update the rman.9 man page to reflect the removal of the RF_TIMESHARE
feature.
MFC r269106:
Add a 'raw' parameter to the 'modinfo' subcommand. This is handy when
trying to figure out why a QSFP+/SFP+ connector or cable wasn't
identified correctly by cxgbe(4). Its output looks like this:
MFC r264434:
DTrace's pid provider works by inserting breakpoint instructions at probe
sites and installing a hook at the kernel's trap handler. The fasttrap code
will emulate the overwritten instruction in some common cases, but otherwise
copies it out into some scratch space in the traced process' address space
and ensures that it's executed after returning from the trap.
In Solaris and illumos, this (per-thread) scratch space comes from some
reserved space in TLS, accessible via the fs segment register. This
approach is somewhat unappealing on FreeBSD since it would require some
modifications to rtld and jemalloc (for static TLS) to ensure that TLS is
executable, and would thus introduce dependencies on their implementation
details. I think it would also be impossible to safely trace static binaries
compiled without these modifications.
This change implements the functionality in a different way, by having
fasttrap map pages into the target process' address space on demand. Each
page is divided into 64-byte chunks for use by individual threads, and
fasttrap's process descriptor struct has been extended to keep track of
any scratch space allocated for the corresponding process.
With this change it's possible to trace all libc functions in a program,
e.g. with
MFC r263329:
Only invoke fasttrap hooks for traps from user mode, and ensure that they're
called with interrupts enabled. Calling fasttrap_pid_probe() with interrupts
disabled can lead to deadlock if fasttrap writes to the process' address
space.
MFC r262669:
When our linker merges .SUNW_dof sections from multiple files, it simply
concatenates the DOF tables into one section. Previously, the USDT init
code in drti.o would only look at the first table in the DOF section; with
this change, it iterates over all the tables, passing each DOF table to
the kernel.
marius [Tue, 29 Jul 2014 13:08:46 +0000 (13:08 +0000)]
MFC: r269050
- Copying and zeroing pages via temporary mappings involves updating the
corresponding page tables followed by accesses to the pages in question.
This sequence is subject to the situation exactly described in the "AMD64
Architecture Programmer's Manual Volume 2: System Programming" rev. 3.23,
"7.3.1 Special Coherency Considerations" [1, p. 171 f.]. Therefore, issuing
the INVLPG right after modifying the PTE bits is crucial.
For pmap_copy_page(), this has been broken in r124956 and later on carried
over to pmap_copy_pages() derived from the former, while all other places
in the i386 PMAP code use the correct order of instructions in this regard.
Fixing the latter breakage solves the problem of data corruption seen with
unmapped I/O enabled when running at least bare metal on AMD R-268D APUs.
However, this might also fix similar corruption reported for virtualized
environments.
- In pmap_copy_pages(), correctly set the cache bits on the source page being
copied. This change is thought to be a NOP for the real world, though. [2]
r268640:
Allow multi-byte reads in the private CHELSIO_T4_GET_I2C ioctl. The
firmware allows up to 48B to be read this way but the driver limits
itself to 8B at a time to remain compatible with old cxgbetool
binaries.
MFC r268420:
Remove IO_SYNC flag when writing extended file attributes on ZFS.
While it is possible to create and write file, modify its permissions, etc.
without ever doing sync, it looks odd that it is required for setting
extended file attributes on ZFS. UFS does not do sync there too.
Samba uses those extended attributes to store some its data, and doing it
synchronously by many times reduces file creation performance for systems
without SLOG device.
MFC 267883:
Expand r261243 even further and ignore any I/O port resources assigned to
PCI root bridges except for the one known-valid case on x86 where bridges
claim the I/O port registers used for PCI config space access.
MFC r268642:
libc/gen: small updates to code originating at OpenBSD
arc4random.c
- CVS rev. 1.22
Change arc4random_uniform() to calculate ``2**32 % upper_bound'' as
``-upper_bound % upper_bound''. Simplifies the code and makes it the
same on both ILP32 and LP64 architectures, and also slightly faster on
LP64 architectures by using a 32-bit remainder instead of a 64-bit
remainder.
- CVS rev. 1.23
Spacing
readpassphrase.c
-CVS rev. v 1.24
most obvious unsigned char casts for ctype
Use NULL instead of 0 (Patch by Sascha Wildner <saw at online.de> for Dragonfly)
Remove unnecessary semicolons (Patch by Sascha Wildner <saw at online.de> for Dragonfly)
Add support for arbitrary http requests [1]
Support EAGAIN in fetch_writev
Submitted by: Alex Hornung <alex at alexhornung.com> [1]
Reviewed by: des
MFC r268735:
Improve support for Intel Lynx Point USB 3.0 controllers by using the
USB 2.0 port mask in addition to the USB 3.0 port mask. The hardware
does not always accept when writing -1U to the port switching
registers.
dim [Sat, 19 Jul 2014 18:33:09 +0000 (18:33 +0000)]
MFC r268774:
After r261991, clang warnings about unused functions in the kernel were
completely silenced. Make sure these warnings appear again, so there is
some incentive to fix them, but do not error out the whole kernel build
for them.
Merge r256868,257276-257277,257515,257913 from head. These are fixes
required to make Xen buildable w/o INET. This is the same as r259541
in stable/10 by glebius.
>r263749 | imp | 2014-03-25 16:08:31 -0600 (Tue, 25 Mar 2014) | 18 lines
>Rather than require a makeoptions DEBUG to get debug correct,
>add it in kern.mk, but only if we're using clang. While this
>option is supported by both clang and gcc, in the future there
>may be changes to clang which change the defaults that require
>a tweak to build our kernel such that other tools in our tree
>will work. Set a good example by forcing -gdwarf-2 only for
>clang builds, and only if the user hasn't specified another
>dwarf level already. Update UPDATING to reflect the changed
>state of affairs. This also keeps us from having to update
>all the ARM kernels to add this, and also keeps us from
>in the future having to update all the MIPS kernels and is
>one less place the user will have to know to do something
>special for clang and one less thing developers will need
>to do when moving an architecture to clang.
Note: Due to the funkadellic way we used to do merges, I've made a
direct commit of the relevant text to UPDATING with no merge info
being recorded for it so I could commit it at the same time as the
other files.
tty.c 1.31
handle EINTR in the termios operations
Allow a single process to control multiple ttys (for pthreads using _REENTRANT)
using multiple EditLine objects.
pass lint on _LP64.
Don't depend on side effects inside an assert
MFC r268582:
Fix performance problems with AXGE network adapter in RX direction:
- Remove 4 extra bytes from the ethernet payload.
- The maximum RX buffer was incorrectly set. Increase it to 64K for
now, until the exact limit is understood.
- Enable hardware checksumming again.
- Make hardware data structure packed.
Improve markup, change references to nonexistent vt_vga(4), remove some
language redundancy, and move the examples so sections are in the
standard order.
MFC: r268008
There might be a potential race condition for the NFSv4 client
when a newly created file has another open done on it that
update the open mode. This patch moves the code that updates
the open mode up into the block where the mutex is held to
ensure this cannot happen. No bug caused by this potential
race has been observed, but this fix is a safety belt to ensure
it cannot happen.
Posix strptime() requires support for %t and %n, which were added
to the illumos port. Curiously we were skipping white spaces by
default in most other cases making %t meaningless.
We now skip spaces in the case of the %e specifier as strftime(3)
explicitly adds a space for the single digit case.
hiren [Thu, 10 Jul 2014 21:02:58 +0000 (21:02 +0000)]
MFC r256920
The TCP delayed ACK logic isn't aware of LRO passing up large aggregated
segments thinking it received only one segment. This causes it to enable
the delay the ACK for 100ms to wait for another segment which may never
come because all the data was received already.
Doing delayed ACK for LRO segments is bogus for two reasons: a) it pushes
us further away from acking every other packet; b) it introduces additional
delay in responding to the sender. The latter is especially bad because it
is in the nature of LRO to aggregated all segments of a burst with no more
coming until an ACK is sent back.
Change the delayed ACK logic to detect LRO segments by being larger than
the MSS for this connection and issuing an immediate ACK for them to keep
the ACK clock ticking without interruption.
MFC r266490, r266738, r267955, and r268209:
- Improve performance by fixing incorrect Rx/Tx handling
- Rename definition of AXGE_* to reflect reality
- Add new USB IDs
- Add proper rangechecks in "axge_rx_frame()" function and
fix receive loop header parsing.
- Disable hardware checksumming until it is properly tested.
Initialize r_flags the same way in all cases using a sanitized copy of
flags that has several bits cleared. The RF_WANTED and RF_FIRSTSHARE
bits are invalid in this context, and we want to defer setting RF_ACTIVE
in r_flags until later. This should make rman_get_flags() return
the correct answer in all cases.
Add a KASSERT() to catch callers which incorrectly pass the RF_WANTED
or RF_FIRSTSHARE flags.
Do a strict equality check on the share type bits of flags. In
particular, do an equality check on RF_PREFETCHABLE. The previous
code would allow one type of mismatch of RF_PREFETCHABLE but disallow
the other type of mismatch. Also, ignore the the RF_ALIGNMENT_MASK
bits since alignment validity should be handled by the amask check.
This field contains an integer value, but previous code did a strange
bitwise comparison on it.
Leave the original value of flags unmolested as a minor debug aid.
Change the start+amask overflow check to a KASSERT() since it is just
meant to catch a highly unlikely programming error in the caller.
MFC r267680:
Fix a code typo that prevented mkdir from firing (unnoticed
usually because another part of the code succeeded in making
the same directory).