to add actions that run when a TCP frame is sent or received on a TCP
session in the ESTABLISHED state. In the base tree, this functionality is
only used for the h_ertt module, which is used by the cc_cdg, cc_chd, cc_hd,
and cc_vegas congestion control modules.
Presently, we incur overhead to check for hooks each time a TCP frame is
sent or received on an ESTABLISHED TCP session.
This change adds a new compile-time option (TCP_HHOOK) to determine whether
to include the hhook(9) framework for TCP. To retain backwards
compatibility, I added the TCP_HHOOK option to every configuration file that
already defined "options INET". (Therefore, this patch introduces no
functional change. In order to see a functional difference, you need to
compile a custom kernel without the TCP_HHOOK option.) This change will
allow users to easily exclude this functionality from their kernel, should
they wish to do so.
Note that any users who use a custom kernel configuration and use one of the
congestion control modules listed above will need to add the TCP_HHOOK
option to their kernel configuration.
Reviewed by: rrs, lstewart, hiren (previous version), sjg (makefiles only)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D8185
userland. It supports userland interfaces to UEFI Runtime Services. This is
indended to the the MI portion of EFI RuntimeServices support.
Differential Revision: https://reviews.freebsd.org/D8128
Reviewed by: kib@, wblock@, Ganael Laplanche
Reduce the cost of TLB invalidation on x86 by using per-CPU completion flags
Reduce contention during TLB invalidation operations by using a per-CPU
completion flag, rather than a single atomically-updated variable.
On a Westmere system (2 sockets x 4 cores x 1 threads), dtrace measurements
show that smp_tlb_shootdown is about 50% faster with this patch; observations
with VTune show that the percentage of time spent in invlrng_single_page on an
interrupt (actually doing invalidation, rather than synchronization) increases
from 31% with the old mechanism to 71% with the new one. (Running a basic file
server workload.)
Submitted by: Anton Rang <rang at acm.org>
Reviewed by: cem (earlier version)
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D8041
like other PCI network drivers. The sys/ofed directory is now mainly
reserved for generic infiniband code, with exception of the mthca driver.
- Add new manual page, mlx4en(4), describing how to configure and load
mlx4en.
- All relevant driver C-files are now prefixed mlx4, mlx4_en and
mlx4_ib respectivly to avoid object filename collisions when compiling
the kernel. This also fixes an issue with proper dependency file
generation for the C-files in question.
- Device mlxen is now device mlx4en and depends on device mlx4, see
mlx4en(4). Only the network device name remains unchanged.
- The mlx4 and mlx4en modules are now built by default on i386 and
amd64 targets. Only building the mlx4ib module depends on
WITH_OFED=YES .
Sponsored by: Mellanox Technologies
(1) Print the default segment %ss before adresses relative to %bp.
This is too cluttered for me, but so is printing some other default
prefixes, and this is a reasonable reminder that %ss is quite
likely to be different from %ds in 16-bit mode.
db_disasm still handles prefixes poorly, by trying to discard
redundant ones. This loses information, and sometimes the result
is wrong or misleading.
Clean up nearby initializations and dead code.
(2) Fix decoding of operand and address size prefixes in 16-bit mode.
They reverse the default in all modes.
Obtained from: (1) is partly from r1.4 (2003/11/08) in DFlyBSD (?)
db_segsize().
Use db_segsize() to set the default operand/address size for
disassembling. Allow overriding this with the "alternate" display
format /I. The API of db_disasm() should be debooleanized to pass a
more general request (amd64 needs overrides to sizes of 16, 32, and
64, but this commit doesn't implement anything for amd64 since much
larger changes are needed to restore the amd64 disassmbler's support
for non-default sizes).
Fix db_print_loc_and_inst() to ask for the normal format and not the
alternate in normal operation.
This is most useful for vm86 mode, but also works for 16-bit protected
mode.
Use db_segsize() to avoid trying to print a garbage stack trace if %cs
is 16 bits. Print something like the stack trace termination message
for a trap boundary instead.
Document that the alternate format is now useful on i386.
early printfs and debugging of vm86 initialization and some other early
initialization in some cases.) Add an option debug.late_console (with
default 1=off) to move console and kdb initialization back where it was.
Do the same for amd64 although there is no vm86 there.
On my test system, debug.late_console=0 works for the syscons, sio and
uart console drivers on amd64 and i386, and for vt on i386 but not on
amd64.
The early printfs fixed by debug.late_console=0 are:
- on i386, the message about lost memory above 4G
- with -v in otherwise normal use, about 20 printfs for SMAP
- other debugging messages for memory sizing. Mostly under -v and
not printed in normal use.
Document in a comment how much earlier the initialization and early
printf()s can be. That is very early for the console. Not much more
than curthread is needed. kdb use obviously needs to be not so early,
since it needs IDT initialization and that is done relatively late
for convenience and historical reasons.
preserve the ABI and API for applications. It was removed in the port
to amd64, but was remained as garbage giving a micro-pessimization and
spurious single-step traps on i386.
pcb_psl was intended to be used just to do a context switch of PSL_I,
but this context switch was null in most or all versions, and
mis-switching of PSL_T was done instead.
Some history:
- in 386BSD-0.0, cpu_switch() ran at splhigh() and splhigh() did too
much interrupt disabling, so interrupts were hard-disabled across
cpu_switch() and too many other places
- in 386BSD-0.0-patchkit through FreeBSD-4 and FreeBSD-5 before
SMPng, splhigh() did soft interrupt masking, and cpu_switch() was
excessively cautious and did a cli at the start and a sti at the
end to hard-disable interrupts across the switch
- SMPng replaced the spl's and cli's by spinlocks (just sched_lock?),
so interrupts were hard-disabled across cpu_switch() and too many
other places again
- initial attempts to fix this intended to restore some soft
interrupt disabling, but to support variations in this cpu_switch()
used pushfl/popfl into pcb_psl to avoid hard-coding the assumption
that the initial and final states have PSL_I enabled. But the
version with soft interrupt disabling wasn't used for long, or was
never committed, (except I always used my different version of it
for UP) so the pushfl/popl and pcb_psl to hold them have been doing
less than nothing for about 14 years.
While here, avoid using the old variable 'code' and remove it
in trap(). ('code' was meant for holding things like %dr6,
but is too small to hold %dr6 on amd64 and was reduced to an
obfuscation of tf_err, with early truncation on amd64.)
Submitted by: Michael Butler (imb@...)
This is not very easy to do, since ddb didn't know when traps are
for single-stepping. It more or less assumed that traps are either
breakpoints or single-step, but even for x86 this became inadequate
with the release of the i386 in ~1986, and FreeBSD passes it other
trap types for NMIs and panics.
On x86, teach ddb when a trap is for single stepping using the %dr6
register. Unknown traps are now treated almost the same as breakpoints
instead of as the same as single-steps. Previously, the classification
of breakpoints was almost correct and everything else was unknown so
had to be treated as a single-step. Now the classification of single-
steps is precise, the classification of breakpoints is almost correct
(as before) and everything else is unknown and treated like a
breakpoint.
This fixes:
- breakpoints not set by ddb, including the main one in kdb_enter(),
were treated as single-steps and not stopped on when stepping
(except for the usual, simple case of a step with residual count 1).
As special cases, kdb_enter() didn't stop for fatal traps or panics
- similarly for "hardware breakpoints".
Use a new MD macro IS_SSTEP_TRAP(type, code) to code to classify
single-steps. This is excessively complicated for bug-for-bug and
backwards compatibilty. Design errors apparently started in Mach
in ~1990 or perhaps in the FreeBSD interface in ~1993. Common trap
types like single steps should have a unique MI code (like the TRAP*
codes for user SIGTRAP) so that debuggers don't need macros like
IS_SSTEP_TRAP() to decode them. But 'type' is actually an ambiguous
MD trap number, and code was always 0 (now it is (int)%dr6 on x86).
So it was impossible to determine the trap type from the args.
Global variables had to be used.
There is already a classification macro db_pc_is_single_step(), but
this just gets in the way. It is only used to recover from bugs in
IS_BREAKPOINT_TRAP(). On some arches, IS_BREAKPOINT_TRAP() just
duplicates the ambiguity in 'type' and misclassifies single-steps as
breakpoints. It defaults to 'false', which is the opposite of what is
needed for bug-for-bug compatibility.
When this is cleaned up, MI classification bits should be passed in
'code'. This could be done now for positive-logic bits, since 'code'
was always 0, but some negative logic is needed for compatibility so
a simple MI classificition is not usable yet.
After reading %dr6, clear the single-step bit in it so that the type
of the next debugger trap can be decoded. This is a little
ddb-specific. ddb doesn't understand the need to clear this bit and
doing it before calling kdb is easiest. gdb would need to reverse
this to support hardware breakpoints, but it just doesn't support
them now since gdbstub doesn't support %dr*.
Fix a bug involving %dr6: when emulating a single-step trap for vm86,
set the bit for it in %dr6. Userland debuggers need this. ddb now
needs this for vm86 bios calls. The bit gets copied to 'code' then
cleared again.
Fix related style bugs:
- when clearing bits for hardware breakpoints in %dr6, spell the mask
as ~0xf on both amd64 and i386 to get the correct number of bits
using sign extension and not need a comment about using the wrong
mask on amd64 (amd64 traps for invalid results but clearing the
reserved top bits didn't trap since they are 0).
- rewrite my old wrong comments about using %dr6 for ddb watchpoints.
The 'cpu' and 'cpu_class' variables were always set to the same value
on amd64 and are legacy holdovers from i386. Remove them entirely on
amd64.
Reviewed by: imp, kib (older version)
Differential Revision: https://reviews.freebsd.org/D7888
SEL_UPL and sometimes PSL_VM. This is just a style change on amd64,
but on i386 it fixes 1 unimportant place where the PSL_VM check was
missing and starts fixing 1 important place where the PSL_VM check
had a logic error.
Fix logic errors in treating vm86 bioscall mode as kernel mode. The
main place checked all the necessary flags, but put the necessary
parentheses for the PSL_VM and PCB_VM86CALL checks in the wrong
place. The broken case is only reached if a vm86 bioscall uses a
%cs which is nonzero mod 4, but that is unusual -- most bios calls
start with %cs = 0xc000 or 0xf000 and rarely change it. Another
place was missing the check for PCB_VM86CALL, but was only reachable
if there are bugs virtualizing PSL_I.
Add a macro TF_HAS_STACKREGS() and use this instead of converting
open-coded checks of SEL_UPL, etc. to TRAPF_USERMODE() when we only
care about whether the frame has stack registers. This fixes 3
places in my recent fix for register variables in vm86 mode where I
messed up the PSL_VM check and cleans up other places.
Move PMAP_TS_REFERENCED_MAX out of the various pmap implementations and
into vm/pmap.h, and describe what its purpose is. Eliminate the archaic
"XXX" comment about its value. I don't believe that its exact value, e.g.,
5 versus 6, matters.
Update the arm64 and riscv pmap implementations of pmap_ts_referenced()
to opportunistically update the page's dirty field.
On amd64, use the PDE value already cached in a local variable rather than
dereferencing a pointer again and again.
Reviewed by: kib, markj
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D7836
In vm86.c, fix 2 (rarely used) cases where the return code lost the
single-step indicator. While here, fix 2 misspellings of PSL_T as
PSL_TF (TF is the CPU manufacturer's spelling, but we use T).
In trap.c, turn T_PROTFLT and T_STKFLT into T_TRCTRAP if
vm86_emulate() asked for this (it does this when the instruction is
being traced and was successully emulated). In the kernel case, we
used to deliver the trap as SIGTRAP to the process, where it always
terminated the process; now we deliver the trap as T_TRCTRAP to kdb,
where it normally gives single-stepping. In the user case, the only
difference is that we now clear PSL_T and initialize ucode properly.
Reviewed by: kib
Idle page zeroing has been disabled by default on all architectures since
r170816 and has some bugs that make it seemingly unusable. Specifically,
the idle-priority pagezero thread exacerbates contention for the free page
lock, and yields the CPU without releasing it in non-preemptive kernels. The
pagezero thread also does not behave correctly when superpage reservations
are enabled: its target is a function of v_free_count, which includes
reserved-but-free pages, but it is only able to zero pages belonging to the
physical memory allocator.
Reviewed by: alc, imp, kib
Differential Revision: https://reviews.freebsd.org/D7714
dependent pmap_ts_referenced() so that it updates the page's dirty field
if a modified bit is found while counting reference bits. This
opportunistic update can be performed at low cost and can eliminate the
need for some future calls to pmap_is_modified() by the machine-
independent layer.
Reviewed by: kib, markj
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
Differential Revision: https://reviews.freebsd.org/D7722
for zeroing pages in idle where nontemporal writes are clearly best.
This is almost a no-op since zeroing in idle works does nothing good
and is off by default. Fix END() statement forgotten in previous
commit.
Align the loop in sse2_pagezero(). Since it writes to main memory,
the loop doesn't have to be very carefully written to keep up.
Unrolling it was considered useless or harmful and was not done on
i386, but that was too careless.
Timing for i386: the loop was not unrolled at all, and moved only 4
bytes/iteration. So on a 2GHz CPU, it needed to run at 2 cycles/
iteration to keep up with a memory speed of just 4GB/sec. But when
it crossed a 16-byte boundary, on old CPUs it ran at 3 cycles/
iteration so it gave a maximum speed of 2.67GB/sec and couldn't even
keep up with PC3200 memory. Fix the alignment so that it keep up with
4GB/sec memory, and unroll once to get nearer to 8GB/sec. Further
unrolling might be useless or harmful since it would prevent the loop
fitting in 16-bytes. My test system with an old CPU and old DDR1 only
needed 5+ GB/sec. My test system with a new CPU and DDR3 doesn't need
any changes to keep up ~16GB/sec.
Timing for amd64: with 8-byte accesses and newer faster CPUs it is
easy to reach 16GB/sec but not so easy to go much faster. The
alignment doesn't matter much if the CPU is not very old. The loop
was already unrolled 4 times, but needs 32 bytes and uses a fancy
method that doesn't work for 2-way unrolling in 16 bytes. Just
align it to 32-bytes.
Early use of vm86 depends on the PIC being reset to mask interrupts,
but r286667 moved PIC initialization to after where vm86 may be first
used.
Move the PIC initialization up to immdiately before vm86 initialization.
All invocations of diff that I tried display this move poorly so that it
looks like PIC and vm86 initialization was moved later.
r286667 was to move console initialization later. The diffs are again
unreadable -- they show a large move that doesn't seem to involve the
console. The PIC initialization stayed just below the console
initialization where it could still be debugged but no longer works.
Later console initialization breaks mainly debugging vm86 initialization
and memory sizing using ddb and printf(). There are several printf()s
in the memory sizing that now go nowhere since message buffer
initialization has always been too late. Memory sizing is done by loader
for most users, but the lost messages for this case are even more
interesting than for an auto-probe since they tell you what the loader
found.
vm86 uses the tss, but r273995 moved tss initialization to after where
it may be first used, just because tss_esp0 now depends on later
initializations and/or amd64 does it later.
vm86 is first used for memory sizing in cases where the loader can't
figure out the size or is not used. Its initialization is placed
immediately before memory sizing to support this, and the tss was
initialized a little earlier.
Move everything in the tss initialization except for tss_esp0 back to
almost where it was, immediately before vm86 initialization (the
combined move is from before dblflt_tss initialization to after). Add
only early initialization of tss_esp0, later reloading of the tss, and
comments. The initial tss_esp0 no longer has space for the pcb since
initially the size of the pcb is not known and no pcb is needed.
(Later changes broke debugging at this point, so the nonexistent pcb
cannot be used by debuggers, and at the time of 273995 when ddb was
almost able to debug this problem it didn't need the pcb.) The
iniitial tss_esp0 still has a magic 16 bytes reserved for vm86
although I think this is unused too.
In all of these source files, the userspace pointer size corresponds
with the kernelspace pointer size, meaning that casting directly works.
As I'm planning on making 32-bit execution on 64-bit systems work as
well, use TO_PTR() here as well, so that the changes between source
files remain minimal.
Move msix_disable_migration under #ifdef SMP since it doesn't make sense
for !SMP kernels.
PR: 212014
Reported by: Glyn Grinstead <glyn@grinstead.org>
MFC after: 3 days
Copy over amd64's cloudabi64_sysvec.c into i386 and tailor it to work.
Again, we use a system call convention similar to FreeBSD, except that
there is no support for indirect system calls (%eax == 0).
Where i386 differs from amd64 is that we have to store thread/process
entry arguments on the stack instead of using registers. We also have to
put an extra pointer on the stack for TLS (for GSBASE). Place that
pointer in the empty slot that is normally used to hold return
addresses. That seems to keep the code simple.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D7590
This driver only supports 10Mb Ethernet using PIO (the hardware supports
DMA, but the driver only does PIO). There are not any PCCard adapters
supported by this driver, only ISA cards. In addition, it does not use
bus_space but instead uses bcopy with volatile pointers triggering a
host of warnings. (if_ie.c is one of 3 files always built with
-Wno-error)
Relnotes: yes
This hardware is not present on any modern systems. The driver is quite
hackish (raw inb/outb instead of bus_space, and raw inb/outb to random
I/O ports to enable ACPI since it predated proper ACPI support).
Relnotes: yes
The wl(4) driver supports pre-802.11 PCCard wireless adapters that
are slower than 802.11b. They do not work with any of the 802.11
framework and the driver hasn't been reported to actually work in a
long time.
Relnotes: yes
The si(4) driver supported multiport serial adapters for ISA, EISA, and
PCI buses. This driver does not use bus_space, instead it depends on
direct use of the pointer returned by rman_get_virtual(). It is also
still locked by Giant and calls for patch testing to convert it to use
bus_space were unanswered.
Relnotes: yes
Fix PC_REGS() so that printing of instructions works in some useful
cases. ddb only understands a single flat address space, but this
macro allows mapping $cs:$eip into vm86's flat address space well
enough for the MI parts of ddb. This doesn't work for the MD parts
that do stack traces, and there are no similar macros for data addresses.
PC_REGS() has to use the trapframe pointer instead of the pcb for this.
For other CPUs, the trapframe pointer is not available except by tracing
back to it. But tracing back through vm86 trapframes is broken even
starting with one.
exception is caught in kernel mode. There are third-party modules
which trigger the issue, and since the problem causes usermode state
corruption at least, panic in production kernels as well.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
difference between files.
For pc98, put x86/mp_x86.c into the same place as used by i386 file list.
Fix typo in comment.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Any sensible workflow will include a revision control system from which
to restore the old files if required. In normal usage, developers just
have to clean up the mess.
Reviewed by: jhb
Sponsored by: DARPA, AFRL
Differential Revision: https://reviews.freebsd.org/D7353
Set vm_guest to a new enum value (VM_GUEST_KVM) when kvm is detected and use
vm_guest in conditionals testing for KVM.
Also, fix a conditional checking if we're running in a VM which caught only
the generic VM case, but not more specific VMs (KVM, VMWare, etc.). (Spotted
by: vangyzen).
Differential revision: https://reviews.freebsd.org/D7172
Sponsored by: Dell Inc.
Approved by: kib (mentor), vangyzen (mentor)
Reviewed by: alc
MFC after: 4 weeks
If the hypervisor version is smaller than 4.6.0. Xen commits 74fd00 and
70a3cb are required on the hypervisor side for this to be fixed, and those
are only included in 4.6.0, so stay on the safe side and disable MSI-X
interrupt migration on anything older than 4.6.0.
It should not cause major performance degradation unless a lot of MSI-X
interrupts are allocated.
Sponsored by: Citrix Systems R&D
MFC after: 3 days
Reviewed by: jhb
Differential revision: https://reviews.freebsd.org/D7148
If tf_trapno contains garbage which appears to be equal to T_NMI,
e.g. due to thread previously entered kernel due to NMI, doreti
sequence skips ast, and does so until a trap or hardware interrupt
occur.
The visible effects of the issue are quite confusing. First, signals
delivery is postponed in observable ways. In particular, the
guarantee that unblocked async signals queue is flushed before a
return from syscall, is broken. Second, if there are pending signals,
all interruptible sleeps of the stuck thread are aborted immediately.
Since modern CPUs are relatively fast and tickless kernel generates
low interrupt rate, the faulty condition might exist for long time (in
an application time scale).
In collaboration with: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
FreeBSD support NX bit on X86_64 processors out of the box, for i386 emulation
use READ_IMPLIES_EXEC flag, introduced in r302515.
While here move common part of mmap() and mprotect() code to the files in compat/linux
to reduce code dupcliation between Linuxulator's.
Reported by: Johannes Jost Meixner, Shawn Webb
MFC after: 1 week
XMFC with: r302515, r302516
In Linux if this flag is set, PROT_READ implies PROT_EXEC for mmap().
Linux/i386 set this flag automatically if the binary requires executable stack.
READ_IMPLIES_EXEC flag will be used in the next Linux mmap() commit.
mp_maxid or CPU_FOREACH() as appropriate. This fixes a number of places in
the kernel that assumed CPU IDs are dense in [0, mp_ncpus) and would try,
for example, to run tasks on CPUs that did not exist or to allocate too
few buffers on systems with sparse CPU IDs in which there are holes in the
range and mp_maxid > mp_ncpus. Such circumstances generally occur on
systems with SMT, but on which SMT is disabled. This patch restores system
operation at least on POWER8 systems configured in this way.
There are a number of other places in the kernel with potential problems
in these situations, but where sparse CPU IDs are not currently known
to occur, mostly in the ARM machine-dependent code. These will be fixed
in a follow-up commit after the stable/11 branch.
PR: kern/210106
Reviewed by: jhb
Approved by: re (glebius)
threads, to make it less confusing and using modern kernel terms.
Rename the functions to reflect current use of the functions, instead
of the historic KSE conventions:
cpu_set_fork_handler -> cpu_fork_kthread_handler (for kthreads)
cpu_set_upcall -> cpu_copy_thread (for forks)
cpu_set_upcall_kse -> cpu_set_upcall (for new threads creation)
Reviewed by: jhb (previous version)
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (hrs)
Differential revision: https://reviews.freebsd.org/D6731
