dtrace_gethrtime() may be called outside of probe context, and in
particular, from the DTRACEIOC_BUFSNAP handler.
Disable interrupts rather than using sched_pin() to help ensure that
we don't call any external functions when in probe context.
PR: 218452
MFC after: 1 week
of low physical memory:
Update the comment about leaving the permanent mapping in place. This
also improves the wording of the comment. PTD 0 is still left alone
because it is fairly important that it was unmapped earlier, and the
comment now describes the unmapping of the other low PTDs that the code
actually does.
Reviewed by: kib
it by a transient double mapping for the one instruction in ACPI wakeup
where it is needed (and for many surrounding instructions in ACPI resume).
Invalidate the TLB as soon as convenient after undoing the transient
mapping. ACPI resume already has the strict ordering needed for this.
This fixes the non-trapping of null pointers and other garbage pointers
below NBPDR (except transiently). NBPDR is quite large (4MB, or 2MB for
PAE).
This fixes spurious traps at the first instruction in VM86 bioscalls.
The traps are for transiently missing read permission in the first
VM86 page (physical page 0) which was just written to at KERNBASE in
the kernel. The mechanism is unknown (it is not simply PG_G).
locore uses a similar but larger transient double mapping and needs
it for 2 instructions instead of 1. Unmap the first PDE in it after
the 2 instructions to detect most garbage pointers while bootstrapping.
pmap_bootstrap() finishes the unmapping.
Remove the avoidance of the double mapping for a recently fixed special
case. ACPI resume could use this avoidance (made non-special) to avoid
any problems with the transient double mapping, but no such problems
are known.
Update comments in locore. Many were for old versions of FreeBSD which
tried to map low memory r/o except for special cases, or might have
allowed access to low memory via physical offsets. Now all kernel
maps are r/w, and removal of of the double map disallows use of physical
offsets again.
when KERNLOAD is smaller than NBPDR (not the default) and PG_G is
enabled (the default if the CPU supports it). This case has relatively
minor problems with coherency of the permanent double mapping, but the
fix in r167869 to improve coherency creates page tables with 3 different
errors so never worked.
The permanent double mapping is fundamentally broken and will be removed
soon. It fundamentally breaks trapping for null pointers and requires
complications to avoid cache coherency bugs. It is currently used for
only a single instruction in ACPI resume,
Many fixes VM86 and/or ACPI and/or the double map were attempted near
r1200000. r167869 attempted to fix cache coherency bugs in an unusual
case, but the bugs were unreachable because older errors in page tables
caused a crash first.
This commit just makes r167869 work as intended. Part 1 of these fixes
fixed the other errors, but also stopped mapping the PDE for KERNBASE
as a large page, so double mapping of this PDE only causes the same
problems as when KERNLOAD is the default. Except for the problem of
trapping null pointers, r167869 could be used to fix these problems,
but it is inactive in usual cases. The only known other problem is
that incoherent permissions for page 0 cause spurious traps in VM86
BIOS calls.
Reviewed by: kib
When building the command to execute for compression, newsyslog was modifying
the generic arguments array instead of its own copy.
Meaning on the second file to compress with the same arguments, the command line
was not the one expected.
Fix it by creating one copy of the arguments per execution and modifying that
copy.
While here, print the command line executed in verbose mode.
Reported by: many
when KERNLOAD is not a multiple of NBPDR (not the default) and PSE is
enabled (the default if the CPU supports it). Addresses in PDEs must
be a multiple of NBPDR in the PSE case, but were not so in the crashing
case.
KERNLOAD defaults to NBPDR. NBPDR is 4 MB for !PAE and 2 MB for PAE.
The default can be changed by editing i386/include/vmparam.h or using
makeoptions. It can be changed to less than NBPDR to save real and
virtual memory at a small cost in time, or to more than NBPDR to waste
real and virtual memory. It must be larger than 1 MB and a multiple of
PAGE_SIZE. When it is less than NBPDR, it is necessarily not a multiple
of NBPDR. This case has much larger bugs which will be fixed in part 2.
The fix is to only use PSE for physical addresses above <KERNLOAD
rounded _up_ to an NBPDR boundary>. When the rounding is non-null,
this leaves part of the kernel not using large pages. Rounding down
would avoid this pessimization, but would break setting of PAT bits
on i/o pages if it goes below 1MB. Since rounding down always goes
below 1MB when KERNLOAD < NBPDR and the KERNLOAD > NBPDR case is not
useful, never round down.
Fix related style bugs (e.g., wrong literal values for NBPDR in comments).
Reviewed by: kib
HELP_FILES is a loader only thing, so move it to loader.mk. Only
generate the help file if HELP_FILES is defined. Adjust Makefiles to
new convention. Fix a few cases where ${.CURDIR}/ was missing
resulting in missing bits from the help files.
Sponsored by: Netflix
A comment in bcm_bsc_fill_tx_fifo() even lists sc_totlen > 0 as a
precondition for calling the routine. I apparently forgot to make the
code do what my comment said.
Otherwise a poorly timed lowmem event may attempt to acquire a destroyed
lock. Unregister the handler before destroying the ARC reclaim thread.
Reported by: gjb
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D13480
We may need to initialize multiple inode blocks before writing a given
inode. makefs(8) was only initializing a single block at a time, so
certain inode allocation patterns could lead to a situation where it
wrote an inode to an uninitialized block. That inode might be clobbered
by a later initialization, resulting in a filesystem image containing
directory entries that point to a seemingly unused inode.
Reviewed by: imp
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D13505
This change copies the existing amd64_cloudabi64.c to amd64_cloudabi32.c
and reimplements the functions for fetching system call arguments and
return values to use the same scheme as used by the vDSO that is used
when running cloudabi32 executables.
As arguments are automatically padded to 64-bit words by the vDSO in
userspace, we can copy the arguments directly into the array used by
truss(8) internally.
Reviewed by: jhb
Differential Revision: https://reviews.freebsd.org/D13516
In order to let truss(8) support tracing of 32-bit CloudABI
applications, we need to add a new ABI type to libsysdecode. We can
reuse the existing errno mapping table. Also link in the cloudabi32
system call table to translate system call names.
While there, remove all of the architecture ifdefs. There are not
needed, as the CloudABI data types and system call tables build fine on
any architecture. Building this unconditionally will make it easier to
do tracing for different compat modes, emulation, etc.
Reviewed by: jhb
Differential Revision: https://reviews.freebsd.org/D13516
vxlan_ftable entries are sorted in ascending order, due to wrong arguments
order it is possible to stop search before existing element will be found.
Then new element will be allocated in vxlan_ftable_update_locked() and can
be inserted in the list second time or trigger MPASS() assertion with
enabled INVARIANTS.
PR: 224371
MFC after: 1 week
Cherry-pick lld r315658 by Rui Ueyama:
This is not a mechanical transformation. Even though I believe this
patch is correct, I'm not 100% sure if lld with this patch behaves
exactly the same way as before on all edge cases. At least all tests
still pass.
I'm submitting this patch because it took almost a day to understand
this function, and I don't want to lose it.
This fixes jemalloc assertion failures observed at startup with i386
binaries and an lld-linked libc.so.
Reviewed by: dim
Obtained from: LLVM r315658
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D13503
weren't needed, and their existance interfered with things in subtle
ways. One of these subtle ways was that malloc could be different
based on what files were included when (even within the same .c file,
it turns out). Move to a single malloc implementation as well by
adding the calls to setheap() to gptboot.c and zfsboot.c. Once upon a
time, these boot loaders strove to not use libstand. However, with the
proliferation of features, that striving is too hard for too little
gain and lead to stupid mistakes.
This fixes the GELI-enabled (but not even using) boot environment. The
geli routines were calling libstand malloc but zfsboot.c and gptboot.c
were using the mini libstand malloc, so this failed when we tried to
probe for GELI partitions. Subtle changes in build order when moving
to self-contained stand build in r326593 toggled what it used from one
type to another due to odd nesting of the zfs implementation code that
differed subtly between zfsloader and zfsboot.
Sponsored by: Netflix
interactive console rather than the video port. qemu has issues with X
on my mac at the moment and this is the easiest path forward.
Sponsored by: Netflix
