We have a special segment descriptor entry in the GDT, whose sole purpose is to
encode the CPU and node numbers in its limit (size) field. There are user-space
instructions that allow the reading of the limit field, which gives us a really
fast way to read the CPU and node IDs from the vDSO for example.
But the naming of related functionality does not make this clear, at all:
VDSO_CPU_SIZE
VDSO_CPU_MASK
__CPU_NUMBER_SEG
GDT_ENTRY_CPU_NUMBER
vdso_encode_cpu_node
vdso_read_cpu_node
There's a number of problems:
- The 'VDSO_CPU_SIZE' doesn't really make it clear that these are number
of bits, nor does it make it clear which 'CPU' this refers to, i.e.
that this is about a GDT entry whose limit encodes the CPU and node number.
- Furthermore, the 'CPU_NUMBER' naming is actively misleading as well,
because the segment limit encodes not just the CPU number but the
node ID as well ...
So use a better nomenclature all around: name everything related to this trick
as 'CPUNODE', to make it clear that this is something special, and add
_BITS to make it clear that these are number of bits, and propagate this to
every affected name:
VDSO_CPU_SIZE => VDSO_CPUNODE_BITS
VDSO_CPU_MASK => VDSO_CPUNODE_MASK
__CPU_NUMBER_SEG => __CPUNODE_SEG
GDT_ENTRY_CPU_NUMBER => GDT_ENTRY_CPUNODE
vdso_encode_cpu_node => vdso_encode_cpunode
vdso_read_cpu_node => vdso_read_cpunode
This, beyond being less confusing, also makes it easier to grep for all related
functionality:
$ git grep -i cpunode arch/x86
Also, while at it, fix "return is not a function" style sloppiness in vdso_encode_cpunode().
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chang S. Bae <chang.seok.bae@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Markus T Metzger <markus.t.metzger@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Shankar <ravi.v.shankar@intel.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/1537312139-5580-2-git-send-email-chang.seok.bae@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
vgetcyc() is full of barriers, so fetching values out of the vvar
page before vgetcyc() for use after vgetcyc() results in poor code
generation. Put vgetcyc() first to avoid this problem.
Also, pull the tv_sec division into the loop and put all the ts
writes together. The old code wrote ts->tv_sec on each iteration
before the syscall fallback check and then added in the offset
afterwards, which forced the compiler to pointlessly copy base->sec
to ts->tv_sec on each iteration. The new version seems to generate
sensible code.
Saves several cycles. With this patch applied, the result is faster
than before the clock_gettime() rewrite.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/3c05644d010b72216aa286a6d20b5078d5fae5cd.1538762487.git.luto@kernel.org
The code flow for the vclocks is convoluted as it requires the vclocks
which can be invalidated separately from the vsyscall_gtod_data sequence to
store the fact in a separate variable. That's inefficient.
Restructure the code so the vclock readout returns cycles and the
conversion to nanoseconds is handled at the call site.
If the clock gets invalidated or vclock is already VCLOCK_NONE, return
U64_MAX as the cycle value, which is invalid for all clocks and leave the
sequence loop immediately in that case by calling the fallback function
directly.
This allows to remove the gettimeofday fallback as it now uses the
clock_gettime() fallback and does the nanoseconds to microseconds
conversion in the same way as it does when the vclock is functional. It
does not make a difference whether the division by 1000 happens in the
kernel fallback or in userspace.
Generates way better code and gains a few cycles back.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Matt Rickard <matt@softrans.com.au>
Cc: Stephen Boyd <sboyd@kernel.org>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Florian Weimer <fweimer@redhat.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: devel@linuxdriverproject.org
Cc: virtualization@lists.linux-foundation.org
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Juergen Gross <jgross@suse.com>
Link: https://lkml.kernel.org/r/20180917130707.657928937@linutronix.de
It's desired to support more clocks in the VDSO, e.g. CLOCK_TAI. This
results either in indirect calls due to the larger switch case, which then
requires retpolines or when the compiler is forced to avoid jump tables it
results in even more conditionals.
To avoid both variants which are bad for performance the high resolution
functions and the coarse grained functions will be collapsed into one for
each. That requires to store the clock specific base time in an array.
Introcude struct vgtod_ts for storage and convert the data store, the
update function and the individual clock functions over to use it.
The new storage does not longer use gtod_long_t for seconds depending on 32
or 64 bit compile because this needs to be the full 64bit value even for
32bit when a Y2038 function is added. No point in keeping the distinction
alive in the internal representation.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Matt Rickard <matt@softrans.com.au>
Cc: Stephen Boyd <sboyd@kernel.org>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Florian Weimer <fweimer@redhat.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: devel@linuxdriverproject.org
Cc: virtualization@lists.linux-foundation.org
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Juergen Gross <jgross@suse.com>
Link: https://lkml.kernel.org/r/20180917130707.324679401@linutronix.de
The syscall fallbacks in the vDSO have incorrect asm constraints.
They are not marked as writing to their outputs -- instead, they are
marked as clobbering "memory", which is useless. In particular, gcc
is smart enough to know that the timespec parameter hasn't escaped,
so a memory clobber doesn't clobber it. And passing a pointer as an
asm *input* does not tell gcc that the pointed-to value is changed.
Add in the fact that the asm instructions weren't volatile, and gcc
was free to omit them entirely unless their sole output (the return
value) is used. Which it is (phew!), but that stops happening with
some upcoming patches.
As a trivial example, the following code:
void test_fallback(struct timespec *ts)
{
vdso_fallback_gettime(CLOCK_MONOTONIC, ts);
}
compiles to:
00000000000000c0 <test_fallback>:
c0: c3 retq
To add insult to injury, the RCX and R11 clobbers on 64-bit
builds were missing.
The "memory" clobber is also unnecessary -- no ordering with respect to
other memory operations is needed, but that's going to be fixed in a
separate not-for-stable patch.
Fixes: 2aae950b21 ("x86_64: Add vDSO for x86-64 with gettimeofday/clock_gettime/getcpu")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/2c0231690551989d2fafa60ed0e7b5cc8b403908.1538422295.git.luto@kernel.org
The SYSCALL64 trampoline has a couple of nice properties:
- The usual sequence of SWAPGS followed by two GS-relative accesses to
set up RSP is somewhat slow because the GS-relative accesses need
to wait for SWAPGS to finish. The trampoline approach allows
RIP-relative accesses to set up RSP, which avoids the stall.
- The trampoline avoids any percpu access before CR3 is set up,
which means that no percpu memory needs to be mapped in the user
page tables. This prevents using Meltdown to read any percpu memory
outside the cpu_entry_area and prevents using timing leaks
to directly locate the percpu areas.
The downsides of using a trampoline may outweigh the upsides, however.
It adds an extra non-contiguous I$ cache line to system calls, and it
forces an indirect jump to transfer control back to the normal kernel
text after CR3 is set up. The latter is because x86 lacks a 64-bit
direct jump instruction that could jump from the trampoline to the entry
text. With retpolines enabled, the indirect jump is extremely slow.
Change the code to map the percpu TSS into the user page tables to allow
the non-trampoline SYSCALL64 path to work under PTI. This does not add a
new direct information leak, since the TSS is readable by Meltdown from the
cpu_entry_area alias regardless. It does allow a timing attack to locate
the percpu area, but KASLR is more or less a lost cause against local
attack on CPUs vulnerable to Meltdown regardless. As far as I'm concerned,
on current hardware, KASLR is only useful to mitigate remote attacks that
try to attack the kernel without first gaining RCE against a vulnerable
user process.
On Skylake, with CONFIG_RETPOLINE=y and KPTI on, this reduces syscall
overhead from ~237ns to ~228ns.
There is a possible alternative approach: Move the trampoline within 2G of
the entry text and make a separate copy for each CPU. This would allow a
direct jump to rejoin the normal entry path. There are pro's and con's for
this approach:
+ It avoids a pipeline stall
- It executes from an extra page and read from another extra page during
the syscall. The latter is because it needs to use a relative
addressing mode to find sp1 -- it's the same *cacheline*, but accessed
using an alias, so it's an extra TLB entry.
- Slightly more memory. This would be one page per CPU for a simple
implementation and 64-ish bytes per CPU or one page per node for a more
complex implementation.
- More code complexity.
The current approach is chosen for simplicity and because the alternative
does not provide a significant benefit, which makes it worth.
[ tglx: Added the alternative discussion to the changelog ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Borislav Petkov <bp@suse.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/8c7c6e483612c3e4e10ca89495dc160b1aa66878.1536015544.git.luto@kernel.org
The STACKLEAK feature (initially developed by PaX Team) has the following
benefits:
1. Reduces the information that can be revealed through kernel stack leak
bugs. The idea of erasing the thread stack at the end of syscalls is
similar to CONFIG_PAGE_POISONING and memzero_explicit() in kernel
crypto, which all comply with FDP_RIP.2 (Full Residual Information
Protection) of the Common Criteria standard.
2. Blocks some uninitialized stack variable attacks (e.g. CVE-2017-17712,
CVE-2010-2963). That kind of bugs should be killed by improving C
compilers in future, which might take a long time.
This commit introduces the code filling the used part of the kernel
stack with a poison value before returning to userspace. Full
STACKLEAK feature also contains the gcc plugin which comes in a
separate commit.
The STACKLEAK feature is ported from grsecurity/PaX. More information at:
https://grsecurity.net/https://pax.grsecurity.net/
This code is modified from Brad Spengler/PaX Team's code in the last
public patch of grsecurity/PaX based on our understanding of the code.
Changes or omissions from the original code are ours and don't reflect
the original grsecurity/PaX code.
Performance impact:
Hardware: Intel Core i7-4770, 16 GB RAM
Test #1: building the Linux kernel on a single core
0.91% slowdown
Test #2: hackbench -s 4096 -l 2000 -g 15 -f 25 -P
4.2% slowdown
So the STACKLEAK description in Kconfig includes: "The tradeoff is the
performance impact: on a single CPU system kernel compilation sees a 1%
slowdown, other systems and workloads may vary and you are advised to
test this feature on your expected workload before deploying it".
Signed-off-by: Alexander Popov <alex.popov@linux.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
Pull Kbuild updates from Masahiro Yamada:
- verify depmod is installed before modules_install
- support build salt in case build ids must be unique between builds
- allow users to specify additional host compiler flags via HOST*FLAGS,
and rename internal variables to KBUILD_HOST*FLAGS
- update buildtar script to drop vax support, add arm64 support
- update builddeb script for better debarch support
- document the pit-fall of if_changed usage
- fix parallel build of UML with O= option
- make 'samples' target depend on headers_install to fix build errors
- remove deprecated host-progs variable
- add a new coccinelle script for refcount_t vs atomic_t check
- improve double-test coccinelle script
- misc cleanups and fixes
* tag 'kbuild-v4.19' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild: (41 commits)
coccicheck: return proper error code on fail
Coccinelle: doubletest: reduce side effect false positives
kbuild: remove deprecated host-progs variable
kbuild: make samples really depend on headers_install
um: clean up archheaders recipe
kbuild: add %asm-generic to no-dot-config-targets
um: fix parallel building with O= option
scripts: Add Python 3 support to tracing/draw_functrace.py
builddeb: Add automatic support for sh{3,4}{,eb} architectures
builddeb: Add automatic support for riscv* architectures
builddeb: Add automatic support for m68k architecture
builddeb: Add automatic support for or1k architecture
builddeb: Add automatic support for sparc64 architecture
builddeb: Add automatic support for mips{,64}r6{,el} architectures
builddeb: Add automatic support for mips64el architecture
builddeb: Add automatic support for ppc64 and powerpcspe architectures
builddeb: Introduce functions to simplify kconfig tests in set_debarch
builddeb: Drop check for 32-bit s390
builddeb: Change architecture detection fallback to use dpkg-architecture
builddeb: Skip architecture detection when KBUILD_DEBARCH is set
...
Pull x86 PTI updates from Thomas Gleixner:
"The Speck brigade sadly provides yet another large set of patches
destroying the perfomance which we carefully built and preserved
- PTI support for 32bit PAE. The missing counter part to the 64bit
PTI code implemented by Joerg.
- A set of fixes for the Global Bit mechanics for non PCID CPUs which
were setting the Global Bit too widely and therefore possibly
exposing interesting memory needlessly.
- Protection against userspace-userspace SpectreRSB
- Support for the upcoming Enhanced IBRS mode, which is preferred
over IBRS. Unfortunately we dont know the performance impact of
this, but it's expected to be less horrible than the IBRS
hammering.
- Cleanups and simplifications"
* 'x86/pti' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (60 commits)
x86/mm/pti: Move user W+X check into pti_finalize()
x86/relocs: Add __end_rodata_aligned to S_REL
x86/mm/pti: Clone kernel-image on PTE level for 32 bit
x86/mm/pti: Don't clear permissions in pti_clone_pmd()
x86/mm/pti: Fix 32 bit PCID check
x86/mm/init: Remove freed kernel image areas from alias mapping
x86/mm/init: Add helper for freeing kernel image pages
x86/mm/init: Pass unconverted symbol addresses to free_init_pages()
mm: Allow non-direct-map arguments to free_reserved_area()
x86/mm/pti: Clear Global bit more aggressively
x86/speculation: Support Enhanced IBRS on future CPUs
x86/speculation: Protect against userspace-userspace spectreRSB
x86/kexec: Allocate 8k PGDs for PTI
Revert "perf/core: Make sure the ring-buffer is mapped in all page-tables"
x86/mm: Remove in_nmi() warning from vmalloc_fault()
x86/entry/32: Check for VM86 mode in slow-path check
perf/core: Make sure the ring-buffer is mapped in all page-tables
x86/pti: Check the return value of pti_user_pagetable_walk_pmd()
x86/pti: Check the return value of pti_user_pagetable_walk_p4d()
x86/entry/32: Add debug code to check entry/exit CR3
...
Pull x86 vdso update from Thomas Gleixner:
"Use LD to link the VDSO libs instead of indirecting trough CC which
causes build failures with Clang"
* 'x86-vdso-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86: vdso: Use $LD instead of $CC to link
Pull x86 asm updates from Thomas Gleixner:
"The lowlevel and ASM code updates for x86:
- Make stack trace unwinding more reliable
- ASM instruction updates for better code generation
- Various cleanups"
* 'x86-asm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/entry/64: Add two more instruction suffixes
x86/asm/64: Use 32-bit XOR to zero registers
x86/build/vdso: Simplify 'cmd_vdso2c'
x86/build/vdso: Remove unused vdso-syms.lds
x86/stacktrace: Enable HAVE_RELIABLE_STACKTRACE for the ORC unwinder
x86/unwind/orc: Detect the end of the stack
x86/stacktrace: Do not fail for ORC with regs on stack
x86/stacktrace: Clarify the reliable success paths
x86/stacktrace: Remove STACKTRACE_DUMP_ONCE
x86/stacktrace: Do not unwind after user regs
x86/asm: Use CC_SET/CC_OUT in percpu_cmpxchg8b_double() to micro-optimize code generation
The vdso{32,64}.so can fail to link with CC=clang when clang tries to find
a suitable GCC toolchain to link these libraries with.
/usr/bin/ld: arch/x86/entry/vdso/vclock_gettime.o:
access beyond end of merged section (782)
This happens because the host environment leaked into the cross compiler
environment due to the way clang searches for suitable GCC toolchains.
Clang is a retargetable compiler, and each invocation of it must provide
--target=<something> --gcc-toolchain=<something> to allow it to find the
correct binutils for cross compilation. These flags had been added to
KBUILD_CFLAGS, but the vdso code uses CC and not KBUILD_CFLAGS (for various
reasons) which breaks clang's ability to find the correct linker when cross
compiling.
Most of the time this goes unnoticed because the host linker is new enough
to work anyway, or is incompatible and skipped, but this cannot be reliably
assumed.
This change alters the vdso makefile to just use LD directly, which
bypasses clang and thus the searching problem. The makefile will just use
${CROSS_COMPILE}ld instead, which is always what we want. This matches the
method used to link vmlinux.
This drops references to DISABLE_LTO; this option doesn't seem to be set
anywhere, and not knowing what its possible values are, it's not clear how
to convert it from CC to LD flag.
Signed-off-by: Alistair Strachan <astrachan@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: kernel-team@android.com
Cc: joel@joelfernandes.org
Cc: Andi Kleen <andi.kleen@intel.com>
Link: https://lkml.kernel.org/r/20180803173931.117515-1-astrachan@google.com
error_entry and error_exit communicate the user vs. kernel status of
the frame using %ebx. This is unnecessary -- the information is in
regs->cs. Just use regs->cs.
This makes error_entry simpler and makes error_exit more robust.
It also fixes a nasty bug. Before all the Spectre nonsense, the
xen_failsafe_callback entry point returned like this:
ALLOC_PT_GPREGS_ON_STACK
SAVE_C_REGS
SAVE_EXTRA_REGS
ENCODE_FRAME_POINTER
jmp error_exit
And it did not go through error_entry. This was bogus: RBX
contained garbage, and error_exit expected a flag in RBX.
Fortunately, it generally contained *nonzero* garbage, so the
correct code path was used. As part of the Spectre fixes, code was
added to clear RBX to mitigate certain speculation attacks. Now,
depending on kernel configuration, RBX got zeroed and, when running
some Wine workloads, the kernel crashes. This was introduced by:
commit 3ac6d8c787 ("x86/entry/64: Clear registers for exceptions/interrupts, to reduce speculation attack surface")
With this patch applied, RBX is no longer needed as a flag, and the
problem goes away.
I suspect that malicious userspace could use this bug to crash the
kernel even without the offending patch applied, though.
[ Historical note: I wrote this patch as a cleanup before I was aware
of the bug it fixed. ]
[ Note to stable maintainers: this should probably get applied to all
kernels. If you're nervous about that, a more conservative fix to
add xorl %ebx,%ebx; incl %ebx before the jump to error_exit should
also fix the problem. ]
Reported-and-tested-by: M. Vefa Bicakci <m.v.b@runbox.com>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Cc: xen-devel@lists.xenproject.org
Fixes: 3ac6d8c787 ("x86/entry/64: Clear registers for exceptions/interrupts, to reduce speculation attack surface")
Link: http://lkml.kernel.org/r/b5010a090d3586b2d6e06c7ad3ec5542d1241c45.1532282627.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The vDSO needs to have a unique build id in a similar manner
to the kernel and modules. Use the build salt macro.
Acked-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Laura Abbott <labbott@redhat.com>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
