and kern.cam.ctl.disable tunable; those were introduced as a workaround
to make it possible to boot GENERIC on low memory machines.
With ctl(4) being built as a module and automatically loaded by ctladm(8),
this makes CTL work out of the box.
Reviewed by: ken
Sponsored by: FreeBSD Foundation
- use clock_gettime(2) as the time base for the emulated ACPI timer instead
of directly using rdtsc().
- don't advertise the invariant TSC capability to the guest to discourage it
from using the TSC as its time base.
Discussed with: jhb@ (about making 'smp_tsc' a global)
Reported by: Dan Mack on freebsd-virtualization@
Obtained from: NetApp
Introduce counter(9) API, that implements fast and raceless counters,
provided (but not limited to) for gathering of statistical data.
See http://lists.freebsd.org/pipermail/freebsd-arch/2013-April/014204.html
for more details.
In collaboration with: kib
Reviewed by: luigi
Tested by: ae, ray
Sponsored by: Nginx, Inc.
most kernels before FreeBSD 9.0. Remove such modules and respective kernel
options: atadisk, ataraid, atapicd, atapifd, atapist, atapicam. Remove the
atacontrol utility and some man pages. Remove useless now options ATA_CAM.
No objections: current@, stable@
MFC after: never
decode. This is to accomodate hardware assist implementations that do not
provide the 'guest linear address' as part of nested page fault collateral.
Submitted by: Anish Gupta (akgupt3 at gmail dot com)
do not map the b_pages pages into buffer_map KVA. The use of the
unmapped buffers eliminate the need to perform TLB shootdown for
mapping on the buffer creation and reuse, greatly reducing the amount
of IPIs for shootdown on big-SMP machines and eliminating up to 25-30%
of the system time on i/o intensive workloads.
The unmapped buffer should be explicitely requested by the GB_UNMAPPED
flag by the consumer. For unmapped buffer, no KVA reservation is
performed at all. The consumer might request unmapped buffer which
does have a KVA reserve, to manually map it without recursing into
buffer cache and blocking, with the GB_KVAALLOC flag.
When the mapped buffer is requested and unmapped buffer already
exists, the cache performs an upgrade, possibly reusing the KVA
reservation.
Unmapped buffer is translated into unmapped bio in g_vfs_strategy().
Unmapped bio carry a pointer to the vm_page_t array, offset and length
instead of the data pointer. The provider which processes the bio
should explicitely specify a readiness to accept unmapped bio,
otherwise g_down geom thread performs the transient upgrade of the bio
request by mapping the pages into the new bio_transient_map KVA
submap.
The bio_transient_map submap claims up to 10% of the buffer map, and
the total buffer_map + bio_transient_map KVA usage stays the
same. Still, it could be manually tuned by kern.bio_transient_maxcnt
tunable, in the units of the transient mappings. Eventually, the
bio_transient_map could be removed after all geom classes and drivers
can accept unmapped i/o requests.
Unmapped support can be turned off by the vfs.unmapped_buf_allowed
tunable, disabling which makes the buffer (or cluster) creation
requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped
buffers are only enabled by default on the architectures where
pmap_copy_page() was implemented and tested.
In the rework, filesystem metadata is not the subject to maxbufspace
limit anymore. Since the metadata buffers are always mapped, the
buffers still have to fit into the buffer map, which provides a
reasonable (but practically unreachable) upper bound on it. The
non-metadata buffer allocations, both mapped and unmapped, is
accounted against maxbufspace, as before. Effectively, this means that
the maxbufspace is forced on mapped and unmapped buffers separately.
The pre-patch bufspace limiting code did not worked, because
buffer_map fragmentation does not allow the limit to be reached.
By Jeff Roberson request, the getnewbuf() function was split into
smaller single-purpose functions.
Sponsored by: The FreeBSD Foundation
Discussed with: jeff (previous version)
Tested by: pho, scottl (previous version), jhb, bf
MFC after: 2 weeks
Replace the per-object resident and cached pages splay tree with a
path-compressed multi-digit radix trie.
Along with this, switch also the x86-specific handling of idle page
tables to using the radix trie.
This change is supposed to do the following:
- Allowing the acquisition of read locking for lookup operations of the
resident/cached pages collections as the per-vm_page_t splay iterators
are now removed.
- Increase the scalability of the operations on the page collections.
The radix trie does rely on the consumers locking to ensure atomicity of
its operations. In order to avoid deadlocks the bisection nodes are
pre-allocated in the UMA zone. This can be done safely because the
algorithm needs at maximum one new node per insert which means the
maximum number of the desired nodes is the number of available physical
frames themselves. However, not all the times a new bisection node is
really needed.
The radix trie implements path-compression because UFS indirect blocks
can lead to several objects with a very sparse trie, increasing the number
of levels to usually scan. It also helps in the nodes pre-fetching by
introducing the single node per-insert property.
This code is not generalized (yet) because of the possible loss of
performance by having much of the sizes in play configurable.
However, efforts to make this code more general and then reusable in
further different consumers might be really done.
The only KPI change is the removal of the function vm_page_splay() which
is now reaped.
The only KBI change, instead, is the removal of the left/right iterators
from struct vm_page, which are now reaped.
Further technical notes broken into mealpieces can be retrieved from the
svn branch:
http://svn.freebsd.org/base/user/attilio/vmcontention/
Sponsored by: EMC / Isilon storage division
In collaboration with: alc, jeff
Tested by: flo, pho, jhb, davide
Tested by: ian (arm)
Tested by: andreast (powerpc)
This can be done by using the new macros VMM_STAT_INTEL() and VMM_STAT_AMD().
Statistic counters that are common across the two are defined using VMM_STAT().
Suggested by: Anish Gupta
Discussed with: grehan
Obtained from: NetApp
pages around, taking array of vm_page_t both for source and
destination. Starting offsets and total transfer size are specified.
The function implements optimal algorithm for copying using the
platform-specific optimizations. For instance, on the architectures
were the direct map is available, no transient mappings are created,
for i386 the per-cpu ephemeral page frame is used. The code was
typically borrowed from the pmap_copy_page() for the same
architecture.
Only i386/amd64, powerpc aim and arm/arm-v6 implementations were
tested at the time of commit. High-level code, not committed yet to
the tree, ensures that the use of the function is only allowed after
explicit enablement.
For sparc64, the existing code has known issues and a stab is added
instead, to allow the kernel linking.
Sponsored by: The FreeBSD Foundation
Tested by: pho (i386, amd64), scottl (amd64), ian (arm and arm-v6)
MFC after: 2 weeks
future further optimizations where the vm_object lock will be held
in read mode most of the time the page cache resident pool of pages
are accessed for reading purposes.
The change is mostly mechanical but few notes are reported:
* The KPI changes as follow:
- VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK()
- VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK()
- VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK()
- VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED()
(in order to avoid visibility of implementation details)
- The read-mode operations are added:
VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(),
VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED()
* The vm/vm_pager.h namespace pollution avoidance (forcing requiring
sys/mutex.h in consumers directly to cater its inlining functions
using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h
consumers now must include also sys/rwlock.h.
* zfs requires a quite convoluted fix to include FreeBSD rwlocks into
the compat layer because the name clash between FreeBSD and solaris
versions must be avoided.
At this purpose zfs redefines the vm_object locking functions
directly, isolating the FreeBSD components in specific compat stubs.
The KPI results heavilly broken by this commit. Thirdy part ports must
be updated accordingly (I can think off-hand of VirtualBox, for example).
Sponsored by: EMC / Isilon storage division
Reviewed by: jeff
Reviewed by: pjd (ZFS specific review)
Discussed with: alc
Tested by: pho
will prevent the kernel from linking if the device driver are included
without the virtio module. Remove pci and scbus for the same reason.
Also explain the relationship and necessity of the virtio and virtio_pci
modules. Currently in FreeBSD, we only support VirtIO PCI, but it could
be replaced with a different interface (like MMIO) and the device
(network, block, etc) will still function.
Requested by: luigi
Approved by: grehan (mentor)
MFC after: 3 days
tunable by default.
This will allow GENERIC configurations to boot on small memory boxes, but
not require end users who want to use CTL to recompile their kernel. They
can simply set kern.cam.ctl.disable=0 in loader.conf.
The eventual solution to the memory usage problem is to change the way
CTL allocates memory to be more configurable, but this should fix things
for small memory situations in the mean time.
UPDATING: Explain the change in the CTL configuration, and
how users can enable CTL if they would like to use
it.
sys/conf/options: Add a new option, CTL_DISABLE, that prevents CTL
from initializing.
ctl.c: If CTL_DISABLE is turned on, don't initialize.
i386/conf/GENERIC,
amd64/conf/GENERIC: Re-enable device ctl, and add the CTL_DISABLE
option.
Rename the pv_entry_t iterator from pv_list to pv_next.
Besides being more correct technically (as the name seems to suggest
this is a list while it is an iterator), it will also be needed by
vm_radix work to avoid a nameclash on macro expansions.
Sponsored by: EMC / Isilon storage division
Reviewed by: alc, jeff
Tested by: flo, pho, jhb, davide
It unfortunately steals a fair chunk of RAM at startup even if it's not
actively used, which prevents FreeBSD VMs of 128MB from successfully
booting and running.
When CPU becomes idle, cpu_idleclock() calculates time to the next timer
event in order to reprogram hw timer. Return that time in sbintime_t to
the caller and pass it to acpi_cpu_idle(), where it can be used as one
more factor (quite precise) to extimate furter sleep time and choose
optimal sleep state. This is a preparatory change for further callout
improvements will be committed in the next days.
The commmit is not targeted for MFC.
VM_OBJECT_LOCKED() macro is only used to implement a custom version
of lock assertions right now (which likely spread out thanks to
copy and paste).
Remove it and implement actual assertions.
Sponsored by: EMC / Isilon storage division
Reviewed by: alc
Tested by: pho
machine/signal.h and machine/ucontext.h into common x86 includes,
copying from amd64 and merging with i386.
Kernel-only compat definitions are kept in the i386/include/sigframe.h
and i386/include/signal.h, to reduce amd64 kernel namespace pollution.
The amd64 compat uses its own definitions so far.
The _MACHINE_ELF_WANT_32BIT definition is to allow the
sys/boot/userboot/userboot/elf32_freebsd.c to use i386 ELF definitions
on the amd64 compile host. The same hack could be usefully abused by
other code too.
Prior to this change pinning was implemented via an ioctl (VM_SET_PINNING)
that called 'sched_bind()' on behalf of the user thread.
The ULE implementation of 'sched_bind()' bumps up 'td_pinned' which in turn
runs afoul of the assertion '(td_pinned == 0)' in userret().
Using the cpuset affinity to implement pinning of the vcpu threads works with
both 4BSD and ULE schedulers and has the happy side-effect of getting rid
of a bunch of code in vmm.ko.
Discussed with: grehan
This eliminates the need to recompile the kernel when the default value
of NKPT is not big enough - for e.g. when loading large kernel modules
or memory disk images from the loader.
If NKPT is defined in the kernel configuration file then it overrides the
dynamic calculation.
Reviewed by: alc, kib
- change 'pics' from STAILQ to TAILQ
- ensure that Local APIC is always first in 'pics'
Reviewed by: jhb
Tested by: Sergey V. Dyatko <sergey.dyatko@gmail.com>,
KAHO Toshikazu <kaho@elam.kais.kyoto-u.ac.jp>
MFC after: 12 days
can only be located at the beginning or the end of the BAR.
If the MSI-table is located in the middle of a BAR then we will split the
BAR into two and create two mappings - one before the table and one after
the table - leaving a hole in place of the table so accesses to it can be
trapped and emulated.
Obtained from: NetApp
The maximum length of an environment variable puts a limitation on the
number of passthru devices that can be specified via a single variable.
The workaround is to allow user to specify passthru devices via multiple
environment variables instead of a single one.
Obtained from: NetApp
FreeBSD TCP-level socket options (only the first two are). Instead,
using a mapping function and fail unsupported options as we do for other
socket option levels.
MFC after: 2 weeks
that 'smp_started != 0'.
This is required because the VT-x initialization calls smp_rendezvous()
to set the CR4_VMXE bit on all the cpus.
With this change we can preload vmm.ko from the loader.
Reported by: alfred@, sbruno@
Obtained from: NetApp
'bhyve' was developed by grehan@ and myself at NetApp (thanks!).
Special thanks to Peter Snyder, Joe Caradonna and Michael Dexter for their
support and encouragement.
Obtained from: NetApp
CPUs exhibit bad behavior if this is done (Intel Errata AAJ3, hangs on
Pentium-M, and trashing of the local APIC registers on a VIA C7). The
local APIC is implicitly mapped UC already via MTRRs, so the clflush isn't
necessary anyway.
MFC after: 2 weeks
This should not matter much when running on bare metal but it makes the guest
more friendly when running inside a virtual machine.
Discussed with: jhb
Obtained from: NetApp
During the early days of bhyve it did not support instruction emulation
which necessitated the use of x2apic to access the local apic. This is no
longer the case and the dependency on x2apic has gone away.
The x2apic patches can be considered independently of bhyve and will be
merged into head via projects/x2apic.
Discussed with: grehan
the guest to execute real or unpaged protected mode code - bhyve relies on
this feature to execute the AP bootstrap code.
Get rid of the hack that allowed bhyve to support SMP guests on processors
that do not have the "unrestricted guest" capability. This hack was entirely
FreeBSD-specific and would not work with any other guest OS.
Instead, limit the number of vcpus to 1 when executing on processors without
"unrestricted guest" capability.
Suggested by: grehan
Obtained from: NetApp
x2apic mode on the guest.
The guest can decide whether or not it wants to use legacy mmio or x2apic
access to the APIC by writing to the MSR_APICBASE register.
Obtained from: NetApp
Provide a tunable 'machdep.x2apic_desired' to let the administrator override
the default behavior.
Provide a read-only sysctl 'machdep.x2apic' to let the administrator know
whether the kernel is using x2apic or legacy mmio to access local apic.
Tested with Parallels Desktop 8 and bhyve hypervisors.
Also tested running on bare metal Intel Xeon E5-2658.
Obtained from: NetApp
Discussed with: jhb, attilio, avg, grehan
guest floating point state without having to know the
size of floating-point state.
Unstaticize fpurestore to allow the hypervisor to
save/restore guest state using fpusave/fpurestore
on the allocated FPU state area.
Reviewed by: kib
Obtained from: NetApp/bhyve
MFC after: 1 week
hierarchy of the page table entries which map the specified address.
Reviewed by: alc (previous version)
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
by clang in the local APIC code.
0x81 is a read-modify-write instruction - the EPT check
that only allowed read or write and not both has been
relaxed to allow read and write.
Reviewed by: neel
Obtained from: NetApp
On a nested page table fault the hypervisor will:
- fetch the instruction using the guest %rip and %cr3
- decode the instruction in 'struct vie'
- emulate the instruction in host kernel context for local apic accesses
- any other type of mmio access is punted up to user-space (e.g. ioapic)
The decoded instruction is passed as collateral to the user-space process
that is handling the PAGING exit.
The emulation code is fleshed out to include more addressing modes (e.g. SIB)
and more types of operands (e.g. imm8). The source code is unified into a
single file (vmm_instruction_emul.c) that is compiled into vmm.ko as well
as /usr/sbin/bhyve.
Reviewed by: grehan
Obtained from: NetApp
In the case where the underlying host had disabled MSI-X via the
"hw.pci.enable_msix" tunable, the ppt_setup_msix() function would fail
and return an error without properly cleaning up. This in turn would
cause a page fault on the next boot of the guest.
Fix this by calling ppt_teardown_msix() in all the error return paths.
Obtained from: NetApp
sleep, and perform the page allocations with VM_ALLOC_SYSTEM
class. Previously, the allocation was also allowed to completely drain
the reserve of the free pages, being translated to VM_ALLOC_INTERRUPT
request class for vm_page_alloc() and similar functions.
Allow the caller of malloc* to request the 'deep drain' semantic by
providing M_USE_RESERVE flag, now translated to VM_ALLOC_INTERRUPT
class. Previously, it resulted in less aggressive VM_ALLOC_SYSTEM
allocation class.
Centralize the translation of the M_* malloc(9) flags in the single
inline function malloc2vm_flags().
Discussion started by: "Sears, Steven" <Steven.Sears@netapp.com>
Reviewed by: alc, mdf (previous version)
Tested by: pho (previous version)
MFC after: 2 weeks
hypervisor. Apparently, hypervisors failed to filter out 'Standard
Extended Features' report from CPUID, but deliver #gp when
corresponding bit in %cr4 is toggled.
This shall be reconsidered later, after hypervisors correct the bug.
Reported and tested by: joel
Reviewed by: avg
MFC after: 2 weeks
between inline asm statements that would in turn modify the flags
value set by the first asm, and used by the second.
Solve by making the common error block a string that can be pulled
into the first inline asm, and using symbolic labels for asm variables.
bhyve can now build/run fine when compiled with clang.
Reviewed by: neel
Obtained from: NetApp
%gs, when supported. Note that WRFSBASE and WRGSBASE are not very
useful on FreeBSD right now, because a return from the kernel mode to
userspace reloads the bases specified by the sysarch(2) syscall, most
likely.
Enable the Supervisor Mode Execution Prevention (SMEP) when
supported. Since the loader(8) performs hand-off to the kernel with
the page tables which contradict the SMEP, postpone enabling the SMEP
on BSP until pmap switched for the proper kernel tables.
Debugged with the help from: avg
Tested by: avg, Michael Moll <kvedulv@kvedulv.de>
MFC after: 1 month
introduced with the IvyBridge CPUs. Provide the definitions for new
bits in CR3 and CR4 registers.
Tested by: avg, Michael Moll <kvedulv@kvedulv.de>
MFC after: 2 weeks
to vmcs_getreg(). Without this conversion vmcs_getreg() will return EINVAL.
In particular this prevented injection of the breakpoint exception into the
guest via the "-B" option to /usr/sbin/bhyve which is hugely useful when
debugging guest hangs.
This was broken in r241921.
Pointy hat: me
Obtained from: NetApp
vm page allocators do. This fixes a panic when a virtio block
device is mounted as root, with the host system dying in
vm_page_dirty with invalid bits.
Reviewed by: neel
Obtained from: NetApp
guest does a vm exit.
This allows us to trap any fpu access in the host context while the fpu still
has "dirty" state belonging to the guest.
Reported by: "s vas" on freebsd-virtualization@
Obtained from: NetApp
host cpu to the scheduler until the guest is ready to run again.
This implies that the host cpu utilization will now closely mirror the actual
load imposed by the guest vcpu.
Also, the vcpu mutex now needs to be of type MTX_SPIN since we need to acquire
it inside a critical section.
Obtained from: NetApp
If an IPI was delivered to this cpu before interrupts were disabled
then return right away via vmx_setjmp() with a return value of VMX_RETURN_AST.
Obtained from: NetApp
AMD BKDG for CPU families 10h and later requires that the memory
mapped config is always read into or written from al/ax/eax register.
Discussed with: kib, alc
Reviewed by: kib (earlier version)
MFC after: 25 days
instruction loads/stores at its will.
The macro __compiler_membar() is currently supported for both gcc and
clang, but kernel compilation will fail otherwise.
Reviewed by: bde, kib
Discussed with: dim, theraven
MFC after: 2 weeks
r234247.
Use, instead, the static intializer introduced in r239923 for x86 and
sparc64 intr_cpus, unwinding the code to the initial version.
Reviewed by: marius
chunks. This breaks the assumption that the entire memory segment is
contiguously allocated in the host physical address space.
This also paves the way to satisfy the 4KB page allocations by requesting
free pages from the VM subsystem as opposed to hard-partitioning host memory
at boot time.
associated with guest physical memory is contiguous.
Add check to vm_gpa2hpa() that the range indicated by [gpa,gpa+len) is all
contained within a single 4KB page.
associated with guest physical memory is contiguous.
In this case vm_malloc() was using vm_gpa2hpa() to indirectly infer whether
or not the address range had already been allocated.
Replace this instead with an explicit API 'vm_gpa_available()' that returns
TRUE if a page is available for allocation in guest physical address space.
bits under #ifdef _KERNEL but leave definitions for various structures
defined by standards ($PIR table, SMAP entries, etc.) available to
userland.
- Consolidate duplicate SMBIOS table structure definitions in ipmi(4)
and smbios(4) in <machine/pc/bios.h> and make them available to
userland.
MFC after: 2 weeks
page table fault. Use this when fetching the instruction bytes from the guest
memory.
Also modify the lapic_mmio() API so that a decoded instruction is fed into it
instead of having it fetch the instruction bytes from the guest. This is
useful for hardware assists like SVM that provide the faulting instruction
as part of the vmexit.
AP needs to be activated by spinning up an execution context for it.
The local apic emulation is now completely done in the hypervisor and it will
detect writes to the ICR_LO register that try to bring up the AP. In response
to such writes it will return to userspace with an exit code of SPINUP_AP.
Reviewed by: grehan
pmap_unmapdev()'s own direct efforts to destroy the page table entries are
redundant, so eliminate them.
Don't set PTE_W on the page table entry in pmap_kenter{,_attr}() on MIPS.
Setting PTE_W on MIPS is inconsistent with the implementation of this
function on other architectures. Moreover, PTE_W should not be set, unless
the pmap's wired mapping count is incremented, which pmap_kenter{,_attr}()
doesn't do.
MFC after: 10 days
generator, found on IvyBridge and supposedly later CPUs, accessible
with RDRAND instruction.
From the Intel whitepapers and articles about Bull Mountain, it seems
that we do not need to perform post-processing of RDRAND results, like
AES-encryption of the data with random IV and keys, which was done for
Padlock. Intel claims that sanitization is performed in hardware.
Make both Padlock and Bull Mountain random generators support code
covered by kernel config options, for the benefit of people who prefer
minimal kernels. Also add the tunables to disable hardware generator
even if detected.
Reviewed by: markm, secteam (simon)
Tested by: bapt, Michael Moll <kvedulv@kvedulv.de>
MFC after: 3 weeks
comment describing them. Both the function names and the comment had grown
stale. Quite some time has passed since these pmap implementations last
used the page's hold count to track the number of valid mapping within a
page table page. Also, returning TRUE from pmap_unwire_ptp() rather than
_pmap_unwire_ptp() eliminates a few instructions from callers like
pmap_enter_quick_locked() where pmap_unwire_ptp()'s return value is used
directly by a conditional statement.
- Move mwlfw from {amd64,i386}/conf/NOTES to sys/conf/NOTES (mwl(4) is
already present in sys/conf/NOTES).
- Remove duplicate mwl(4) entries from {amd64,i386}/conf/NOTES.
- While here, add a description to the sfxge line in amd64/conf/NOTES.
reason for generated trap. The dump of basic signal information and 8
bytes of the faulting instruction are printed on the controlling
terminal of the process, if the machdep.uprintf_signal syscal is
enabled.
The print is the only practical way to debug traps from a.out
processes I am aware of. Because I have to reimplement it each time I
debug an issue with a.out support on amd64, commit the hack to main
tree.
MFC after: 1 week
in long mode which transfers control to 32bit code segment. Unbreak
the lcall $7,$0 implementation on amd64 by putting the 64bit user code
segment' selector into call gate, and execute the 64bit trampoline
which converts the return frame into 32bit format and switches back to
32bit mode for executing int $0x80 trampoline.
Note that all jumps over the hoops are performed in the user mode.
MFC after: 1 week
It is not listed in the boot sequence in the MP specification (1.4),
and it is explicitly ignored on modern CPUs. It was only ever required
when bootstrapping systems with external APICs (that is, SMP machines
with 486s), which FreeBSD has never supported (and never will).
While here, tidy some comments and remove some banal ones.
matches the algorithm in the MP specification (1.4). Previously we
were sending out the deassert INIT IPI immediately after the initial
INIT IPI was sent.
typical hypervisor does not implement access to the required MSR,
causing #GP on boot.
Reported and tested by: olgeni
PR: amd64/170388
MFC after: 3 days
PTE's PG_M and PG_RW bits but not the physical page frame. First,
only perform vm_page_dirty() on a managed vm_page when the PG_M bit is
being cleared. If the updated PTE continues to have PG_M set, then
there is no requirement to perform vm_page_dirty(). Second, flush the
mapping from the TLB when PG_M alone is cleared, not just when PG_M
and PG_RW are cleared. Otherwise, a stale TLB entry may stop PG_M
from being set again on the next store to the virtual page. However,
since the vm_page's dirty field already shows the physical page as
being dirty, no actual harm comes from the PG_M bit not being set.
Nonetheless, it is potentially confusing to someone expecting to see
the PTE change after a store to the virtual page.
stopped threads. Implementation assumes that the thread's FPU context
is spilled into the PCB due to stop. This is mostly true, except when
FPU state for the thread is not initialized. Then the requests operate
on the garbage state which is currently left in the PCB, causing
confusion.
The situation is indeed observed after a signal delivery and before
#NM fault on execution of any FPU instruction in the signal handler,
since sendsig(9) drops FPU state for current thread, clearing
PCB_FPUINITDONE. When inspecting context state for the signal handler,
debugger sees the FPU state of the main program context instead of the
clear state supposed to be provided to handler.
Fix this by forcing clean FPU state in PCB user FPU save area by
performing getfpuregs(9) before accessing user FPU save area in
ptrace_machdep.c.
Note: this change will be merged to i386 kernel as well, where it is
much more important, since e.g. gdb on i386 uses PT_I386_GETXMMREGS to
inspect FPU context on CPUs that support SSE. Amd64 version of gdb
uses PT_GETFPREGS to inspect both 64 and 32 bit processes, which does
not exhibit the bug.
Reported by: bde
MFC after: 1 week
understands FPU hardware enough to catch SIGFPE and unmask exceptions
in control word, then it may as well properly handle return from
SIGFPE without causing an infinite loop of #MF exceptions due to
faulting instruction restart, when needed.
Clearing exceptions causes information loss for handlers which do
understand FPU hardware, and struct siginfo si_code member cannot be
considered adequate replacement for en_sw content due to translation.
Supposed reason for clearing the exceptions, which is IRQ13 handling
oddities, were never applicable to amd64.
Note: this change will be merged to i386 kernel as well, since we do
not support IRQ13 delivery of #MF notifications for some time.
Requested by: bde
MFC after: 1 week
amd64. It is implemented as __pure2 inline with non-volatile asm read
from pcpu, which allows a compiler to cache its results.
Convert most PCPU_GET(pcb) and curthread->td_pcb accesses into curpcb.
Note that __curthread() uses magic value 0 as an offsetof(struct pcpu,
pc_curthread). It seems to be done this way due to machine/pcpu.h
needs to be processed before sys/pcpu.h, because machine/pcpu.h
contributes machine-depended fields to the struct pcpu definition. As
result, machine/pcpu.h cannot use struct pcpu yet.
The __curpcb() also uses a magic constant instead of offsetof(struct
pcpu, pc_curpcb) for the same reason. The constants are now defined as
symbols and CTASSERTs are added to ensure that future KBI changes do
not break the code.
Requested and reviewed by: bde
MFC after: 3 weeks
occurs using the SSE math processor. Update comments describing the
handling of the exception status bits in coprocessors control words.
Remove GET_FPU_CW and GET_FPU_SW macros which were used only once.
Prefer to use curpcb to access pcb_save over the longer path of
referencing pcb through the thread structure.
Based on the submission by: Ed Alley <wea llnl gov>
PR: amd64/169927
Reviewed by: bde
MFC after: 3 weeks
mostly meets the guidelines set by the Intel SDM:
1. We use XRSTOR and XSAVE from the same CPL using the same linear
address for the store area
2. Contrary to the recommendations, we cannot zero the FPU save area
for a new thread, since fork semantic requires the copy of the
previous state. This advice seemingly contradicts to the advice
from the item 6.
3. We do use XSAVEOPT in the context switch code only, and the area
for XSAVEOPT already always contains the data saved by XSAVE.
4. We do not modify the save area between XRSTOR, when the area is
loaded into FPU context, and XSAVE. We always spit the fpu context
into save area and start emulation when directly writing into FPU
context.
5. We do not use segmented addressing to access save area, or rather,
always address it using %ds basing.
6. XSAVEOPT can be only executed in the area which was previously
loaded with XRSTOR, since context switch code checks for FPU use by
outgoing thread before saving, and thread which stopped emulation
forcibly get context loaded with XRSTOR.
7. The PCB cannot be paged out while FPU emulation is turned off, since
stack of the executing thread is never swapped out.
The context switch code is patched to issue XSAVEOPT instead of XSAVE
if supported. This approach eliminates one conditional in the context
switch code, which would be needed otherwise.
For user-visible machine context to have proper data, fpugetregs()
checks for unsaved extension blocks and manually copies pristine FPU
state into them, according to the description provided by CPUID leaf
0xd.
MFC after: 1 month
on x86 and use that to implement stop_emulating() in the fpu/npx code.
Reimplement start_emulating() in the non-XEN case by using load_cr0() and
rcr0() instead of the 'lmsw' and 'smsw' instructions. Intel explicitly
discourages the use of 'lmsw' and 'smsw' on 80386 and later processors in
the description of these instructions in Volume 2 of the ADM.
Reviewed by: kib
MFC after: 1 month
- Add generic support for opcodes that are escape bytes used for
multi-byte opcodes (such as the 0x0f prefix). Use this to replace
the hard-coded 0x0f special case and add support for three-byte
opcodes that use the 0x0f38 prefix.
- Decode all Intel VMX instructions. invept and invvpid in particular are
three-byte opcodes that use the 0x0f38 escape prefix.
- Rework how the special 'SDEP' size flag works such that the default
instruction name (i_name) is the instruction when the data size
prefix (0x66) is not specified, and the alternate name in i_extra is
used when the prefix is included.
- Add a new 'ADEP' size flag similar to 'SDEP' except that it chooses
between i_name and i_extra based on the address size prefix (0x67).
Use this to fix the decoding for jrcxz vs jecxz which is determined
by the address size prefix, not the operand size prefix. Also, jcxz
is not possible in 64-bit mode, but jrcxz is the default instruction
for that opcode.
- Add support for handling instructions that have a mandatory 'rep'
prefix (this means not outputting the 'repe ' prefix until determining
if it is used as part of an opcode). Make 'pause' less of a special
case this way.
- Decode 'cmpxchg16b' and 'cdqe' which are variants of other instructions
but with a REX.W prefix.
MFC after: 1 month
functions that manage PV entries. Specifically, remove the PV entry from
the containing PV list only after the corresponding PTE is destroyed.
Update the pmap's wired mapping count in pmap_enter() before the PV list
lock is acquired.
natively rather than hand-assembled versions. For xgetbv/xsetbv, add a
wrapper API to deal with xcr* registers: rxcr() and load_xcr().
Reviewed by: kib
MFC after: 1 month
at the point that it calls get_pv_entry(). Thus, pmap_enter()'s PV list
lock pointer must be passed to get_pv_entry() for those rare occasions
when get_pv_entry() calls reclaim_pv_chunk().
Update some related comments.
pmap_remove(). The execution of these functions is no longer serialized
by the pvh global lock.
Make some stylistic changes to the affected code for the sake of
consistency with related code elsewhere in the pmap.
to add PV list locking to pmap_pv_demote_pde(), it is necessary to change
the way that pmap_pv_demote_pde() allocates PV entries. Specifically,
once pmap_pv_demote_pde() begins modifying the PV lists, it can't allocate
any new PV chunks, because that could require the PV list lock to be
dropped. So, all necessary PV chunks must be allocated in advance. To my
surprise, this new approach is a few percent faster than the old one.
usermode, using shared page. The structures and functions have vdso
prefix, to indicate the intended location of the code in some future.
The versioned per-algorithm data is exported in the format of struct
vdso_timehands, which mostly repeats the content of in-kernel struct
timehands. Usermode reading of the structure can be lockless.
Compatibility export for 32bit processes on 64bit host is also
provided. Kernel also provides usermode with indication about
currently used timecounter, so that libc can fall back to syscall if
configured timecounter is unknown to usermode code.
The shared data updates are initiated both from the tc_windup(), where
a fast task is queued to do the update, and from sysctl handlers which
change timecounter. A manual override switch
kern.timecounter.fast_gettime allows to turn off the mechanism.
Only x86 architectures export the real algorithm data, and there, only
for tsc timecounter. HPET counters page could be exported as well, but
I prefer to not further glue the kernel and libc ABI there until
proper vdso-based solution is developed.
Minimal stubs neccessary for non-x86 architectures to still compile
are provided.
Discussed with: bde
Reviewed by: jhb
Tested by: flo
MFC after: 1 month
no longer necessary for free_pv_entry() to be serialized by the pvh global
lock.
Retire pmap_insert_entry() and pmap_remove_entry(). Once upon a time,
these functions were called from multiple places within the pmap. Now,
each has only one caller.
pmap_enter_quick(). These functions are no longer serialized by the pvh
global lock.
There is no need to release the PV list lock before calling free_pv_chunk()
in pmap_remove_pages().
- Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs.
These are available as t3_tom and t4_tom modules that augment cxgb(4)
and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as
usual with or without these extra features.
- iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the
works and will follow soon.
Build-tested with make universe.
30s overview
============
What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the
capabilities of an interface:
# ifconfig -m | grep TOE
Enable/disable TCP offload on an interface (just like any other ifnet
capability):
# ifconfig cxgbe0 toe
# ifconfig cxgbe0 -toe
Which connections are offloaded? Look for toe4 and/or toe6 in the
output of netstat and sockstat:
# netstat -np tcp | grep toe
# sockstat -46c | grep toe
Reviewed by: bz, gnn
Sponsored by: Chelsio communications.
MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
performing the return to usermode using full return path. This
consolidates the handling of exceptional situations in less number of
places, and is less code as well.
Reviewed by: jhb
MFC after: 1 week
layer, but it is read directly by the MI VM layer. This change introduces
pmap_page_is_write_mapped() in order to completely encapsulate all direct
access to PGA_WRITEABLE in the pmap layer.
Aesthetics aside, I am making this change because amd64 will likely begin
using an alternative method to track write mappings, and having
pmap_page_is_write_mapped() in place allows me to make such a change
without further modification to the MI VM layer.
As an added bonus, tidy up some nearby comments concerning page flags.
Reviewed by: kib
MFC after: 6 weeks
