dtrace_trap() consumes page and protection faults triggered by code running
in DTrace probe context. Such faults occur with interrupts disabled and are
detected using a per-CPU flag. Regular faults cause dtrace_trap() to be
called with interrupts enabled, and nothing was ensuring that the flag was
read from the correct CPU. This may result in dtrace_trap() consuming
unrelated page and protection faults when DTrace is enabled, causing the
fault handler to return without actually having handled the fault.
Diagnosed by: Ryan Libby <rlibby@gmail.com>
MFC after: 3 days
Sponsored by: Dell EMC Isilon
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
Currently this argument is a pointer into the stack which is used by FBT
to fetch the first five probe arguments. On all non-x86 architectures it's
simply the trapframe address, so this change has no functional impact. On
amd64 it's a pointer into the trapframe such that stack[1 .. 5] gives the
first five argument registers, which are deliberately grouped together in
the amd64 trapframe definition.
A trapframe argument simplifies the invop handlers on !x86 and makes the
x86 FBT invop handler easier to understand. Moreover, it allows for invop
handlers that may want to modify the register set of the interrupted thread.
This allows the hrtimer to be used earlier during boot. This is required
for boot-time DTrace: anonymous enablings are created during
SI_SUB_DTRACE_ANON, which runs before APs are started. In particular,
the DTrace deadman timer requires that the hrtimer be functional.
MFC after: 2 weeks
It's redundant at the moment since it can be obtained from the trapframe
on the architectures where DTrace is supported, but this won't be the case
with ARM.
the upstream implementation and helps ensure that a trap induced by tracing
fbt::trap:entry is handled without recursively generating another trap.
This makes it possible to run most (but not all) of the DTrace tests under
common/safety/ without triggering a kernel panic.
Submitted by: Anton Rang <anton.rang@isilon.com> (original version)
Phabric: D95
The skew calculation here is exactly backwards. We were able to repro
it on a multi-package ESX server running a FreeBSD VM, where the TSCs
can be pretty evil.
MFC after: 1 week
Submitted by: Jeff Ford <jeffrey.ford2@isilon.com>
Reviewed by: avg, gnn
cpuset_t objects.
That is going to offer the underlying support for a simple bump of
MAXCPU and then support for number of cpus > 32 (as it is today).
Right now, cpumask_t is an int, 32 bits on all our supported architecture.
cpumask_t on the other side is implemented as an array of longs, and
easilly extendible by definition.
The architectures touched by this commit are the following:
- amd64
- i386
- pc98
- arm
- ia64
- XEN
while the others are still missing.
Userland is believed to be fully converted with the changes contained
here.
Some technical notes:
- This commit may be considered an ABI nop for all the architectures
different from amd64 and ia64 (and sparc64 in the future)
- per-cpu members, which are now converted to cpuset_t, needs to be
accessed avoiding migration, because the size of cpuset_t should be
considered unknown
- size of cpuset_t objects is different from kernel and userland (this is
primirally done in order to leave some more space in userland to cope
with KBI extensions). If you need to access kernel cpuset_t from the
userland please refer to example in this patch on how to do that
correctly (kgdb may be a good source, for example).
- Support for other architectures is going to be added soon
- Only MAXCPU for amd64 is bumped now
The patch has been tested by sbruno and Nicholas Esborn on opteron
4 x 12 pack CPUs. More testing on big SMP is expected to came soon.
pluknet tested the patch with his 8-ways on both amd64 and i386.
Tested by: pluknet, sbruno, gianni, Nicholas Esborn
Reviewed by: jeff, jhb, sbruno
safer for i386 because it can be easily over 4 GHz now. More worse, it can
be easily changed by user with 'machdep.tsc_freq' tunable (directly) or
cpufreq(4) (indirectly). Note it is intentionally not used in performance
critical paths to avoid performance regression (but we should, in theory).
Alternatively, we may add "virtual TSC" with lower frequency if maximum
frequency overflows 32 bits (and ignore possible incoherency as we do now).
Currently dtrace_gethrtime uses formula similar to the following for
converting TSC ticks to nanoseconds:
rdtsc() * 10^9 / tsc_freq
The dividend overflows 64-bit type and wraps-around every 2^64/10^9 =
18446744073 ticks which is just a few seconds on modern machines.
Now we instead use precalculated scaling factor of
10^9*2^N/tsc_freq < 2^32 and perform TSC value multiplication separately
for each 32-bit half. This allows to avoid overflow of the dividend
described above.
The idea is taken from OpenSolaris.
This has an added feature of always scaling TSC with invariant value
regardless of TSC frequency changes. Thus the timestamps will not be
accurate if TSC actually changes, but they are always proportional to
TSC ticks and thus monotonic. This should be much better than current
formula which produces wildly different non-monotonic results on when
tsc_freq changes.
Also drop write-only 'cp' variable from amd64 dtrace_gethrtime_init()
to make it identical to the i386 twin.
PR: kern/127441
Tested by: Thomas Backman <serenity@exscape.org>
Reviewed by: jhb
Discussed with: current@, bde, gnn
Silence from: jb
Approved by: re (gnn)
MFC after: 1 week
