KMSAN instrumentation requires thread-local storage to track
initialization state for function parameters and return values. This
buffer is accessed as part of each function prologue. It is provided by
the KMSAN runtime, which looks up a pointer in the current thread's
structure.
When KMSAN is configured, init_secondary() is instrumented, but this
means that GS.base must be initialized first, otherwise the runtime
cannot safely access curthread. Work around this by loading GS.base
before calling init_secondary(), so that the runtime can at least check
curthread == NULL and return a pointer to some dummy storage. Note that
init_secondary() still must reload GS.base after calling lgdt(), which
loads a selector into %gs, which in turn clears the base register.
Reviewed by: kib
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31336
There is no reason to initialize it to anything else, and this matches
initialization of the BSP. No functional change intended.
Reviewed by: kib
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31336
gcc failed as it didn't inlined the builtins and generates calls to
the libgcc, ld can't find libgcc as cross-toolchain libgcc is not installed.
To avoid this add internal vDSO ffs functions without optimized builtins.
Reported by: jhb
MFC after: 2 weeks
The timer is not used on ARM.
Submitted by: Elliott Mitchell <ehem+freebsd@m5p.com>
Reviewed by: royger
Differential Revision: https://reviews.freebsd.org/D29041
Stop using temporal page table with 1:1 mapping of low 1G populated over
the whole VA. Use 1:1 mapping of low 4G temporarily installed in the
normal kernel page table.
The features are:
- now there is one less step for startup asm to perform
- the startup code still needs to be at lower 1G because CPU starts in
real mode. But everything else can be located anywhere in low 4G
because it is accessed by non-paged 32bit protected mode. Note that
kernel page table root page is at low 4G, as well as the kernel itself.
- the page table pages can be allocated by normal allocator, there is
no need to carve them from the phys_avail segments at very early time.
The allocation of the page for startup code still requires some magic.
Pages are freed after APs are ignited.
- la57 startup for APs is less tricky, we directly load the final page
table and do not need to tweak the paging mode.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D31121
To determine whether to use alternate signal stack or not,
we need to use the native signal number, not the one translated
with bsd_to_linux_signal().
In practical terms, this fixes golang.
Reviewed By: dchagin
Fixes: 135dd0cab51
Sponsored By: EPSRC
Differential Revision: https://reviews.freebsd.org/D31298
When a cmpset for removing the PG_RW bit in pmap_promote_pde() fails,
there is no need to repeat the alignment, PG_A, and PG_V tests just to
reload the PTE's value. The only bit that we need be concerned with at
this point is PG_M. Use fcmpset instead.
MFC after: 1 week
Old expression happens to provide the correct answer, but assumes that
kernel is loaded at physical address zero, with 2M gap. Do not use
kernphys to calculate KVA of kernel text start, just explicitly write
out KERNBASE and the hole size.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D31121
The call to pmap_allow_2m_x_page() in pmap_enter_object() is redundant.
Specifically, even without the call to pmap_allow_2m_x_page() in
pmap_enter_object(), pmap_allow_2m_x_page() is eventually called by
pmap_enter_pde(), so the outcome will be the same. Essentially,
calling pmap_allow_2m_x_page() in pmap_enter_object() amounts to
"optimizing" for the unexpected case.
Reviewed by: kib
MFC after: 1 week
Initial patch from submitter was adapted by me to prevent unconditional
FUTEX_REQUEUE use.
PR: 255947
Submitted by: Philippe Michaud-Boudreault
Differential Revision: https://reviews.freebsd.org/D30332
The vDSO initialisation order should be as follows:
- native abi init via exec_sysvec_init();
- vDSO symbols queued to the linux_vdso_syms list;
- linux_vdso_install();
- linux_exec_sysvec_init();
As the exec_sysvec_init() called with SI_ORDER_ANY (last) at SI_SUB_EXEC
order, move linux_vdso_install() and linux_exec_sysvec_init() to the
SI_SUB_EXEC+1 order.
Reviewed by: trasz
Differential Revision: https://reviews.freebsd.org/D30902
MFC after 2 weeks
In order to reduce diff between arches constify vdso install/deinstall
functions like arm64.
Reviewed by: emaste
Differential revision: https://reviews.freebsd.org/D30901
MFC after: 2 weeks
The vDSO (virtual dynamic shared object) is a small shared library that the
kernel maps R/O into the address space of all Linux processes on image
activation. The vDSO is a fully formed ELF image, shared by all processes
with the same ABI, has no process private data.
The primary purpose of the vDSO:
- non-executable stack, signal trampolines not copied to the stack;
- signal trampolines unwind, mandatory for the NPTL;
- to avoid contex-switch overhead frequently used system calls can be
implemented in the vDSO: for now gettimeofday, clock_gettime.
The first two have been implemented, so add the implementation of system
calls.
System calls implemenation based on a native timekeeping code with some
limitations:
- ifunc can't be used, as vDSO r/o mapped to the process VA and rtld
can't relocate symbols;
- reading HPET memory is not implemented for now (TODO).
In case on any error vDSO system calls fallback to the kernel system
calls. For unimplemented vDSO system calls added prototypes which call
corresponding kernel system call.
Tested by: trasz (arm64)
Differential revision: https://reviews.freebsd.org/D30900
MFC after: 2 weeks
Temporary add stubs to the Linux emulation layer which calls the existing hook.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D30911
MFC after: 2 weeks
In preparation for vDSO code revision get rid of incomplete vDSO methods
from locore, but leave .note.Linux section commented out.
.note.Linux section is used by glibc rtld to get the kernel version, that
saves one system call call. I'll try to implement it later, if figure out
how to use it with jails.
MFC after: 2 weeks
The underlying types for both are the same so arguably this doesn't
really matter, but using the wrong type is still confusing and
technically incorrect.
The syscall number is stored in the same register as the syscall return
on amd64 (and possibly other architectures) and so it is impossible to
recover in the signal handler after the call has returned. This small
tweak delivers it in the `si_value` field of the signal, which is
sufficient to catch capability violations and emulate them with a call
to a more-privileged process in the signal handler.
This reapplies 3a522ba1bc852c3d4660a4fa32e4a94999d09a47 with a fix for
the static assertion failure on i386.
Approved by: markj (mentor)
Reviewed by: kib, bcr (manpages)
Differential Revision: https://reviews.freebsd.org/D29185
pmap_copy() is used to speculatively create mappings, so those mappings
should not have their access bit preset.
Reviewed by: kib, markj
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D31162
Reduce the live ranges for three variables so that they do not span the
call to PHYS_TO_VM_PAGE(). This enables the compiler to generate
slightly smaller machine code.
Reviewed by: kib, markj
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D31161
The ACPI parsing code around rid range was wrong on assuming there is
only one pair of start/end device id range. Besides, ivhd_dev_parse()
never work as supposed. The start/end rid info was always zero.
Restructure the code to build dynamic-sized tables for each IOMMU softc
holding device entries. The device entries are enumerated to find a
suitable IOMMU unit. Operations on devices not governed (e.g. the IOMMU
unit itself) are no-op from now on. There are also a minor fix on wrong
%b formatting string usage.
Tested on my EPYC 7282.
Sponsored by: The FreeBSD Foundation
Reviewed by: grehan
Differential Revision: https://reviews.freebsd.org/D30827
This controller supports 2.5G/1G/100MB/10MB speeds, and allows
tx/rx checksum offload, TSO, LRO, and multi-queue operation.
The driver was derived from code contributed by Intel, and modified
by Netgate to fit into the iflib framework.
Thanks to Mike Karels for testing and feedback on the driver.
Reviewed by: bcr (manpages), kbowling, scottl, erj
MFC after: 1 month
Relnotes: yes
Sponsored by: Rubicon Communications, LLC ("Netgate")
Differential Revision: https://reviews.freebsd.org/D30668
Remove deugging stuff, since it's arguably not needed in a minimal
setup. Also vlan, tuntap and gif since they can be loaded.
imp didn't include the part of the patch that removed xen guest support.
Xen guest is relatively small and has no way of being loaded.
Reviewed by: imp
PR: 229564
MFC After: 3 days
The syscall number is stored in the same register as the syscall return
on amd64 (and possibly other architectures) and so it is impossible to
recover in the signal handler after the call has returned. This small
tweak delivers it in the `si_value` field of the signal, which is
sufficient to catch capability violations and emulate them with a call
to a more-privileged process in the signal handler.
Approved by: markj (mentor)
Reviewed by: kib, bcr (manpages)
Differential Revision: https://reviews.freebsd.org/D29185
Otherwise KASAN may generate false positives if the trapframe was
written into a poisoned region of the stack.
Reported by: pho
Sponsored by: The FreeBSD Foundation
Use sysentvec hooks to only call umtx_thread_exit/umtx_exec, which handle
robust mutexes, for native FreeBSD ABI. Similarly, there is no sense
in calling sigfastblock_clear() for non-native ABIs.
Requested by: dchagin
Reviewed by: dchagin, markj (previous version)
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D30987
In a few places, on a failed compare-and-set, both the amd64 pmap and
the arm64 pmap repeat tests on bits that won't change state while the
pmap is locked. Eliminate some of these unnecessary tests.
Reviewed by: andrew, kib, markj
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D31014
Implement dumping core for Linux binaries on amd64, for both
32- and 64-bit executables. Some bits are still missing.
This is based on a prototype by chuck@.
Reviewed By: kib
Sponsored By: EPSRC
Differential Revision: https://reviews.freebsd.org/D30019
Eliminate some unnecessary unlocking and relocking when we have to retry
the operation to avoid deadlock. (All of the other pmap functions that
iterate over a PV list already implemented retries without these same
unlocking and relocking operations.)
Reviewed by: kib, markj
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D30951
This adds `sv_elf_core_osabi`, `sv_elf_core_abi_vendor`,
and `sv_elf_core_prepare_notes` fields to `struct sysentvec`,
and modifies imgact_elf.c to make use of them instead
of hardcoding FreeBSD-specific values. It also updates all
of the ABI definitions to preserve current behaviour.
This makes it possible to implement non-native ELF coredump
support without unnecessary code duplication. It will be used
for Linux coredumps.
Reviewed By: kib
Sponsored By: EPSRC
Differential Revision: https://reviews.freebsd.org/D30921
Hide the vDSO defines to the linux32_sysvec as they are not intended to
be used outside of it. Fix LINUX32_PS_STRINGS, use the size of
struct linux32_ps_strings instead of a numeric constant.
MFC after: 2 weeks
Assuming we can't run on i486, i586 class cpu, retire linux_kplatform var
and use hardcoded 'machine' value in linux_newuname().
I have added linux_kplatform for consistency with linux_platform which is
placed in to vdso to avoid excess copyout it on stack for AT_PLATFORM at
exec time.
This is the first stage of Linuxulator's vdso revision.
Reviewed by: trasz, imp
Differential Revision: https://reviews.freebsd.org/D30774
MFC after: 2 weeks
Stop confusing people, retire COMPAT_LINUX and COMPAT_LINUX32 kernel
build options. Since we have 32 and 64 bit Linux emulators, we can't build both
emulators together into the kernel. I don't think it matters, Linux emulation
depends on loadable modules (via rc).
Cut LINPROCFS and LINSYSFS for consistency.
PR: 215061
Reviewed by: bcr (manpages), trasz
Differential Revision: https://reviews.freebsd.org/D30751
MFC after: 2 weeks
