This reverts commits 0a4e6be9ca
and 80f7fdb1c7.
The task migration notifier was originally introduced in order to support
the pvclock vsyscall with non-synchronized TSC, but KVM only supports it
with synchronized TSC. Hence, on KVM the race condition is only needed
due to a bad implementation on the host side, and even then it's so rare
that it's mostly theoretical.
As far as KVM is concerned it's possible to fix the host, avoiding the
additional complexity in the vDSO and the (re)introduction of the task
migration notifier.
Xen, on the other hand, hasn't yet implemented vsyscall support at
all, so we do not care about its plans for non-synchronized TSC.
Reported-by: Peter Zijlstra <peterz@infradead.org>
Suggested-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Pull x86 vdso changes from Ingo Molnar:
"Misc vDSO updates"
* 'x86-vdso-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/vdso: Remove x32 intermediates during 'make clean'
x86/vdso: Teach 'make clean' to remove generated vdso-image-*.c files
x86/vdso32/syscall.S: Do not load __USER32_DS to %ss
x86/vdso: Fix the x86 vdso2c tool includes
ARM/ARM64: fixes for live migration, irqfd and ioeventfd support (enabling
vhost, too), page aging
s390: interrupt handling rework, allowing to inject all local interrupts
via new ioctl and to get/set the full local irq state for migration
and introspection. New ioctls to access memory by virtual address,
and to get/set the guest storage keys. SIMD support.
MIPS: FPU and MIPS SIMD Architecture (MSA) support. Includes some patches
from Ralf Baechle's MIPS tree.
x86: bugfixes (notably for pvclock, the others are small) and cleanups.
Another small latency improvement for the TSC deadline timer.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
iQEcBAABAgAGBQJVJ9vmAAoJEL/70l94x66DoMEH/R3rh8IMf4jTiWRkcqohOMPX
k1+NaSY/lCKayaSgggJ2hcQenMbQoXEOdslvaA/H0oC+VfJGK+lmU6E63eMyyhjQ
Y+Px6L85NENIzDzaVu/TIWWuhil5PvIRr3VO8cvntExRoCjuekTUmNdOgCvN2ObW
wswN2qRdPIeEj2kkulbnye+9IV4G0Ne9bvsmUdOdfSSdi6ZcV43JcvrpOZT++mKj
RrKB+3gTMZYGJXMMLBwMkdl8mK1ozriD+q0mbomT04LUyGlPwYLl4pVRDBqyksD7
KsSSybaK2E4i5R80WEljgDMkNqrCgNfg6VZe4n9Y+CfAAOToNnkMJaFEi+yuqbs=
=yu2b
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull KVM updates from Paolo Bonzini:
"First batch of KVM changes for 4.1
The most interesting bit here is irqfd/ioeventfd support for ARM and
ARM64.
Summary:
ARM/ARM64:
fixes for live migration, irqfd and ioeventfd support (enabling
vhost, too), page aging
s390:
interrupt handling rework, allowing to inject all local interrupts
via new ioctl and to get/set the full local irq state for migration
and introspection. New ioctls to access memory by virtual address,
and to get/set the guest storage keys. SIMD support.
MIPS:
FPU and MIPS SIMD Architecture (MSA) support. Includes some
patches from Ralf Baechle's MIPS tree.
x86:
bugfixes (notably for pvclock, the others are small) and cleanups.
Another small latency improvement for the TSC deadline timer"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (146 commits)
KVM: use slowpath for cross page cached accesses
kvm: mmu: lazy collapse small sptes into large sptes
KVM: x86: Clear CR2 on VCPU reset
KVM: x86: DR0-DR3 are not clear on reset
KVM: x86: BSP in MSR_IA32_APICBASE is writable
KVM: x86: simplify kvm_apic_map
KVM: x86: avoid logical_map when it is invalid
KVM: x86: fix mixed APIC mode broadcast
KVM: x86: use MDA for interrupt matching
kvm/ppc/mpic: drop unused IRQ_testbit
KVM: nVMX: remove unnecessary double caching of MAXPHYADDR
KVM: nVMX: checks for address bits beyond MAXPHYADDR on VM-entry
KVM: x86: cache maxphyaddr CPUID leaf in struct kvm_vcpu
KVM: vmx: pass error code with internal error #2
x86: vdso: fix pvclock races with task migration
KVM: remove kvm_read_hva and kvm_read_hva_atomic
KVM: x86: optimize delivery of TSC deadline timer interrupt
KVM: x86: extract blocking logic from __vcpu_run
kvm: x86: fix x86 eflags fixed bit
KVM: s390: migrate vcpu interrupt state
...
If we were migrated right after __getcpu, but before reading the
migration_count, we wouldn't notice that we read TSC of a different
VCPU, nor that KVM's bug made pvti invalid, as only migration_count
on source VCPU is increased.
Change vdso instead of updating migration_count on destination.
Cc: stable@vger.kernel.org
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Fixes: 0a4e6be9ca ("x86: kvm: Revert "remove sched notifier for cross-cpu migrations"")
Message-Id: <1428000263-11892-1-git-send-email-rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The existing clean-files rule was missing vdsox32.so and
vdsox32.so.dbg. We should really rename the intermediates to
allow a single rule to get them all.
Also-reported-by: Magnus Damm <damm+renesas@opensource.se>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Link: http://lkml.kernel.org/r/7fa2ad4a63bc6f52e214125900d54165ef06cc10.1427482099.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
After 'make clean' the following files were left in arch/x86/vdso/:
vdso-image-32-int80.c
vdso-image-32-syscall.c
vdso-image-32-sysenter.c
These file are generated during the build process and are present
in .gitignore, so remove them.
Signed-off-by: Andrey Skvortsov <andrej.skvortzov@gmail.com>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Link: http://lkml.kernel.org/r/f85bb7ef6f8c6f6aa4bf422348018c84321454f8.1427482099.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This vDSO code only gets used by 64-bit kernels, not 32-bit ones.
On 64-bit kernels, the data segment is the same for 32-bit and
64-bit userspace, and the SYSRET instruction loads %ss with its
selector.
So there's no need to repeat it by hand. Segment loads are somewhat
expensive: tens of cycles.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
[ Removed unnecessary comment. ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Drewry <wad@chromium.org>
Link: http://lkml.kernel.org/r/63da6d778f69fd0f1345d9287f6764d58be519fa.1427482099.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The following point:
2. per-CPU pvclock time info is updated if the
underlying CPU changes.
Is not true anymore since "KVM: x86: update pvclock area conditionally,
on cpu migration".
Add task migration notification back.
Problem noticed by Andy Lutomirski.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
CC: stable@kernel.org # 3.11+
On gcc5 the kernel does not link:
ld: .eh_frame_hdr table[4] FDE at 0000000000000648 overlaps table[5] FDE at 0000000000000670.
Because prior GCC versions always emitted NOPs on ALIGN directives, but
gcc5 started omitting them.
.LSTARTFDEDLSI1 says:
/* HACK: The dwarf2 unwind routines will subtract 1 from the
return address to get an address in the middle of the
presumed call instruction. Since we didn't get here via
a call, we need to include the nop before the real start
to make up for it. */
.long .LSTART_sigreturn-1-. /* PC-relative start address */
But commit 69d0627a7f ("x86 vDSO: reorder vdso32 code") from 2.6.25
replaced .org __kernel_vsyscall+32,0x90 by ALIGN right before
__kernel_sigreturn.
Of course, ALIGN need not generate any NOP in there. Esp. gcc5 collapses
vclock_gettime.o and int80.o together with no generated NOPs as "ALIGN".
So fix this by adding to that point at least a single NOP and make the
function ALIGN possibly with more NOPs then.
Kudos for reporting and diagnosing should go to Richard.
Reported-by: Richard Biener <rguenther@suse.de>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Cc: <stable@vger.kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1425543211-12542-1-git-send-email-jslaby@suse.cz
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch adds arch specific code for kernel address sanitizer.
16TB of virtual addressed used for shadow memory. It's located in range
[ffffec0000000000 - fffffc0000000000] between vmemmap and %esp fixup
stacks.
At early stage we map whole shadow region with zero page. Latter, after
pages mapped to direct mapping address range we unmap zero pages from
corresponding shadow (see kasan_map_shadow()) and allocate and map a real
shadow memory reusing vmemmap_populate() function.
Also replace __pa with __pa_nodebug before shadow initialized. __pa with
CONFIG_DEBUG_VIRTUAL=y make external function call (__phys_addr)
__phys_addr is instrumented, so __asan_load could be called before shadow
area initialized.
Signed-off-by: Andrey Ryabinin <a.ryabinin@samsung.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Konstantin Serebryany <kcc@google.com>
Cc: Dmitry Chernenkov <dmitryc@google.com>
Signed-off-by: Andrey Konovalov <adech.fo@gmail.com>
Cc: Yuri Gribov <tetra2005@gmail.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Jim Davis <jim.epost@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The theory behind vdso randomization is that it's mapped at a random
offset above the top of the stack. To avoid wasting a page of
memory for an extra page table, the vdso isn't supposed to extend
past the lowest PMD into which it can fit. Other than that, the
address should be a uniformly distributed address that meets all of
the alignment requirements.
The current algorithm is buggy: the vdso has about a 50% probability
of being at the very end of a PMD. The current algorithm also has a
decent chance of failing outright due to incorrect handling of the
case where the top of the stack is near the top of its PMD.
This fixes the implementation. The paxtest estimate of vdso
"randomisation" improves from 11 bits to 18 bits. (Disclaimer: I
don't know what the paxtest code is actually calculating.)
It's worth noting that this algorithm is inherently biased: the vdso
is more likely to end up near the end of its PMD than near the
beginning. Ideally we would either nix the PMD sharing requirement
or jointly randomize the vdso and the stack to reduce the bias.
In the mean time, this is a considerable improvement with basically
no risk of compatibility issues, since the allowed outputs of the
algorithm are unchanged.
As an easy test, doing this:
for i in `seq 10000`
do grep -P vdso /proc/self/maps |cut -d- -f1
done |sort |uniq -d
used to produce lots of output (1445 lines on my most recent run).
A tiny subset looks like this:
7fffdfffe000
7fffe01fe000
7fffe05fe000
7fffe07fe000
7fffe09fe000
7fffe0bfe000
7fffe0dfe000
Note the suspicious fe000 endings. With the fix, I get a much more
palatable 76 repeated addresses.
Reviewed-by: Kees Cook <keescook@chromium.org>
Cc: stable@vger.kernel.org
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
gcc-4.4.4:
arch/x86/vdso/vma.c: In function 'vgetcpu_cpu_init':
arch/x86/vdso/vma.c:247: error: unknown field 'limit0' specified in initializer
arch/x86/vdso/vma.c:247: warning: missing braces around initializer
arch/x86/vdso/vma.c:247: warning: (near initialization for '(anonymous).<anonymous>')
arch/x86/vdso/vma.c:248: error: unknown field 'limit' specified in initializer
arch/x86/vdso/vma.c:248: warning: excess elements in struct initializer
arch/x86/vdso/vma.c:248: warning: (near initialization for '(anonymous)')
....
I couldn't find any way of tricking it into accepting an initializer
format :(
Reported-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Fixes: 258801563b ("x86/vdso: Change the PER_CPU segment to use struct desc_struct")
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
IMO users ought not to be able to use 16-bit segments without
using modify_ldt. Fortunately, it's impossible to break
espfix64 by loading the PER_CPU segment into SS because it's
PER_CPU is marked read-only and SS cannot contain an RO segment,
but marking PER_CPU as 32-bit is less fragile.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/179f490d659307873eefd09206bebd417e2ab5ad.1411494540.git.luto@amacapital.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The first userspace attempt to read or write the PER_CPU segment
will write the accessed bit to the GDT. This is visible to
userspace using the LAR instruction, and it also pointlessly
dirties a cache line.
Set the segment's accessed bit at boot to prevent userspace
access to segments from having side effects.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/ac63814ca4c637a08ec2fd0360d67ca67560a9ee.1411494540.git.luto@amacapital.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Stephen Rothwell's compiler did something amazing: it unrolled a
loop, discovered that one iteration of that loop contained an
always-true test, and emitted a warning that will IMO only serve
to convince people to disable the warning.
That bogus warning caused me to wonder what prompted such an
absurdity from his compiler, and I discovered that the code in
question was, in fact, completely wrong -- I was looking things
up in the wrong array.
This affects 3.16 as well, but the only effect is to screw up
the error checking a bit. vdso2c's output is unaffected.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/53d96ad5.80ywqrbs33ZBCQej%25akpm@linux-foundation.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The core mm code will provide a default gate area based on
FIXADDR_USER_START and FIXADDR_USER_END if
!defined(__HAVE_ARCH_GATE_AREA) && defined(AT_SYSINFO_EHDR).
This default is only useful for ia64. arm64, ppc, s390, sh, tile, 64-bit
UML, and x86_32 have their own code just to disable it. arm, 32-bit UML,
and x86_64 have gate areas, but they have their own implementations.
This gets rid of the default and moves the code into ia64.
This should save some code on architectures without a gate area: it's now
possible to inline the gate_area functions in the default case.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Acked-by: Nathan Lynch <nathan_lynch@mentor.com>
Acked-by: H. Peter Anvin <hpa@linux.intel.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> [in principle]
Acked-by: Richard Weinberger <richard@nod.at> [for um]
Acked-by: Will Deacon <will.deacon@arm.com> [for arm64]
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Nathan Lynch <Nathan_Lynch@mentor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The VVAR area can, obviously, be read; that is kind of the point.
AFAIK this has no effect whatsoever unless x86 suddenly turns into a
nommu architecture. Nonetheless, not setting it is suspicious.
Reported-by: Nathan Lynch <Nathan_Lynch@mentor.com>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/e4c8bf4bc2725bda22c4a4b7d0c82adcd8f8d9b8.1406330779.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Now that we can tolerate extra things dangling off the end of the
vdso image, we can strip the vdso the old fashioned way rather than
using an overcomplicated custom stripping algorithm.
This is a partial reversion of:
6f121e5 x86, vdso: Reimplement vdso.so preparation in build-time C
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/50e01ed6dcc0575d20afd782f9fe98d5ee3e2d8a.1405040914.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Putting the vvar area after the vdso text is rather complicated: it
only works of the total length of the vdso text mapping is known at
vdso link time, and the linker doesn't allow symbol addresses to
depend on the sizes of non-allocatable data after the PT_LOAD
segment.
Moving the vvar area before the vdso text will allow is to safely
map non-allocatable data after the vdso text, which is a nice
simplification.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/156c78c0d93144ff1055a66493783b9e56813983.1405040914.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Relying on static functions used just once to get inlined (and
subsequently have dead code paths eliminated) is wrong: Compilers are
free to decide whether they do this, regardless of optimization level.
With this not happening for vdso_addr() (observed with gcc 4.1.x), an
unresolved reference to align_vdso_addr() causes the build to fail.
[ hpa: vdso_addr() is never actually used on x86-32, as calculate_addr
in map_vdso() is always false. It ought to be possible to clean
this up further, but this fixes the immediate problem. ]
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Link: http://lkml.kernel.org/r/53B5863B02000078000204D5@mail.emea.novell.com
Acked-by: Andy Lutomirski <luto@amacapital.net>
Tested-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Tested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Certain ld versions (observed with 2.20.0) put an empty .rela.dyn
section into shared object files, breaking the assumption on the number
of sections to be copied to the final output. Simply discard any empty
SHT_REL and SHT_RELA sections to address this.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Link: http://lkml.kernel.org/r/53B5861E02000078000204D1@mail.emea.novell.com
Acked-by: Andy Lutomirski <luto@amacapital.net>
Tested-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Tested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
vdso2c was checking for various types of relocations to detect when
the vdso had undefined symbols or was otherwise dependent on
relocation at load time. Undefined symbols in the vdso would fail if
accessed at runtime, and certain implementation errors (e.g. branch
profiling or incorrect symbol visibilities) could result in data
access through the GOT that requires relocations. This could be
as simple as:
extern char foo;
return foo;
Without some kind of visibility control, the compiler would assume
that foo could be interposed at load time and would generate a
relocation.
x86-64 and x32 (as opposed to i386) use explicit-addent (RELA) instead
of implicit-addent (REL) relocations for data access, and vdso2c
forgot to detect those.
Whether these bad relocations would actually fail at runtime depends
on what the linker sticks in the unrelocated references. Nonetheless,
these relocations have no business existing in the vDSO and should be
fixed rather than silently ignored.
This error could trigger on some configurations due to branch
profiling. The previous patch fixed that.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/74ef0c00b4d2a3b573e00a4113874e62f772e348.1403642755.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
DISABLE_BRANCH_PROFILING turns off branch profiling (i.e. a
redefinition of 'if'). Branch profiling depends on a bunch of
kernel-internal symbols and generates extra output sections, none of
which are useful or functional in the vDSO.
It's currently turned off for vclock_gettime.c, but vgetcpu.c also
triggers branch profiling, so just turn it off in the makefile.
This fixes the build on some configurations: the vdso could contain
undefined symbols, and the fake section table overflowed due to
ftrace's added sections.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/bf1ec29e03b2bbc081f6dcaefa64db1c3a83fb21.1403642755.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
With this change, doing 'make vdso_install' and telling gdb:
set debug-file-directory /lib/modules/KVER/vdso
will enable vdso debugging with symbols. This is useful for
testing, but kernel RPM builds will probably want to manually delete
these symlinks or otherwise do something sensible when they strip
the vdso/*.so files.
If ld does not support --build-id, then the symlinks will not be
created.
Note that kernel packagers that use vdso_install may need to adjust
their packaging scripts to accomdate this change. For example,
Fedora's scripts create build-id symlinks themselves in a different
location, so the spec should probably be updated to remove the
symlinks created by make vdso_install.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/a424b189ce3ced85fe1e82d032a20e765e0fe0d3.1403291930.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
.data doesn't need to be separate from .rodata: they're both readonly.
.altinstructions and .altinstr_replacement aren't needed by anything
except vdso2c; strip them from the final image.
While we're at it, rather than aligning the actual executable text,
just shove some unused-at-runtime data in between real data and
text.
My vdso image is still above 4k, but I'm disinclined to try to
trim it harder for 3.16. For future trimming, I suspect that these
sections could be moved to later in the file and dropped from
the in-memory image:
.gnu.version and .gnu.version_d (this may lose versions in gdb)
.eh_frame (should be harmless)
.eh_frame_hdr (I'm not really sure)
.hash (AFAIK nothing needs this section header)
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/2e96d0c49016ea6d026a614ae645e93edd325961.1403129369.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Fully stripping the vDSO has other unfortunate side effects:
- binutils is unable to find ELF notes without a SHT_NOTE section.
- Even elfutils has trouble: it can find ELF notes without a section
table at all, but if a section table is present, it won't look for
PT_NOTE.
- gdb wants section names to match between stripped DSOs and their
symbols; otherwise it will corrupt symbol addresses.
We're also breaking the rules: section 0 is supposed to be SHT_NULL.
Fix these problems by building a better fake section table. While
we're at it, we might as well let buggy Go versions keep working well
by giving the SHT_DYNSYM entry the correct size.
This is a bit unfortunate: it adds quite a bit of size to the vdso
image.
If/when binutils improves and the improved versions become widespread,
it would be worth considering dropping most of this.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/0e546a5eeaafdf1840e6ee654a55c1e727c26663.1403129369.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
"make vdso_install" installs unstripped versions of the vdso objects
for the benefit of the debugger. This was broken by checkin:
6f121e548f x86, vdso: Reimplement vdso.so preparation in build-time C
The filenames are different now, so update the Makefile to cope.
This still installs the 64-bit vdso as vdso64.so. We believe this
will be okay, as the only known user is a patched gdb which is known
to use build-ids, but if it turns out to be a problem we may have to
add a link.
Inspired by a patch from Sam Ravnborg.
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Reported-by: Josh Boyer <jwboyer@fedoraproject.org>
Tested-by: Josh Boyer <jwboyer@fedoraproject.org>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/b10299edd8ba98d17e07dafcd895b8ecf4d99eff.1402586707.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The Go runtime has a buggy vDSO parser that currently segfaults.
This writes an empty SHT_DYNSYM entry that causes Go's runtime to
malfunction by thinking that the vDSO is empty rather than
malfunctioning by running off the end and segfaulting.
This affects x86-64 only as far as we know, so we do not need this for
the i386 and x32 vdsos.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/d10618176c4bd39b457a5e85c497295c90cab1bc.1402620737.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
One final use of the macros from <endian.h> which are not available on
older system. In this case we had one sole case of *writing* a
littleendian number, but the number is SHN_UNDEF which is the constant
zero, so rather than dealing with the general case of littleendian
puts here, just document that the constant is zero and be done with
it.
Reported-and-Tested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/20140610135051.c3c34165f73d67d218b62bd9@linux-foundation.org
There are no standard functions for littleendian data (unlike
bigendian data.) Thus, use <tools/le_byteshift.h> to access
littleendian data members. Those are fairly inefficient, but it
doesn't matter for this purpose (and can be optimized later.) This
avoids portability problems.
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Tested-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/20140606140017.afb7f91142f66cb3dd13c186@linux-foundation.org
Make it a little clearer what the littleendian access macros in
vdso2c.[ch] actually do. This way they can probably also be moved to
a central location (e.g. tools/include) for the benefit of other host
tools.
We should avoid implementation namespace symbols when writing code
that is compiling for the compiler host, so avoid names starting with
double underscore or underscore-capital.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/2cf258df123cb24bad63c274c8563c050547d99d.1401464755.git.luto@amacapital.net
Using arch_vma_name to give special mappings a name is awkward. x86
currently implements it by comparing the start address of the vma to
the expected address of the vdso. This requires tracking the start
address of special mappings and is probably buggy if a special vma
is split or moved.
Improve _install_special_mapping to just name the vma directly. Use
it to give the x86 vvar area a name, which should make CRIU's life
easier.
As a side effect, the vvar area will show up in core dumps. This
could be considered weird and is fixable.
[hpa: I say we accept this as-is but be prepared to deal with knocking
out the vvars from core dumps if this becomes a problem.]
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/276b39b6b645fb11e345457b503f17b83c2c6fd0.1400538962.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
The oops can be triggered in qemu using -no-hpet (but not nohpet) by
reading a couple of pages past the end of the vdso text. This
should send SIGBUS instead of OOPSing.
The bug was introduced by:
commit 7a59ed415f
Author: Stefani Seibold <stefani@seibold.net>
Date: Mon Mar 17 23:22:09 2014 +0100
x86, vdso: Add 32 bit VDSO time support for 32 bit kernel
which is new in 3.15.
This will be fixed separately in 3.15, but that patch will not apply
to tip/x86/vdso. This is the equivalent fix for tip/x86/vdso and,
presumably, 3.16.
Cc: Stefani Seibold <stefani@seibold.net>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/c8b0a9a0b8d011a8b273cbb2de88d37190ed2751.1400538962.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
This unifies the vdso mapping code and teaches it how to map special
pages at addresses corresponding to symbols in the vdso image. The
new code is used for all vdso variants, but so far only the 32-bit
variants use the new vvar page position.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/b6d7858ad7b5ac3fd3c29cab6d6d769bc45d195e.1399317206.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Currently, vdso.so files are prepared and analyzed by a combination
of objcopy, nm, some linker script tricks, and some simple ELF
parsers in the kernel. Replace all of that with plain C code that
runs at build time.
All five vdso images now generate .c files that are compiled and
linked in to the kernel image.
This should cause only one userspace-visible change: the loaded vDSO
images are stripped more heavily than they used to be. Everything
outside the loadable segment is dropped. In particular, this causes
the section table and section name strings to be missing. This
should be fine: real dynamic loaders don't load or inspect these
tables anyway. The result is roughly equivalent to eu-strip's
--strip-sections option.
The purpose of this change is to enable the vvar and hpet mappings
to be moved to the page following the vDSO load segment. Currently,
it is possible for the section table to extend into the page after
the load segment, so, if we map it, it risks overlapping the vvar or
hpet page. This happens whenever the load segment is just under a
multiple of PAGE_SIZE.
The only real subtlety here is that the old code had a C file with
inline assembler that did 'call VDSO32_vsyscall' and a linker script
that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most
likely worked by accident: the linker script entry defines a symbol
associated with an address as opposed to an alias for the real
dynamic symbol __kernel_vsyscall. That caused ld to relocate the
reference at link time instead of leaving an interposable dynamic
relocation. Since the VDSO32_vsyscall hack is no longer needed, I
now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it
work. vdso2c will generate an error and abort the build if the
resulting image contains any dynamic relocations, so we won't
silently generate bad vdso images.
(Dynamic relocations are a problem because nothing will even attempt
to relocate the vdso.)
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
This code is used during CPU setup, and it isn't strictly speaking
related to the 32-bit vdso. It's easier to understand how this
works when the code is closer to its callers.
This also lets syscall32_cpu_init be static, which might save some
trivial amount of kernel text.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/4e466987204e232d7b55a53ff6b9739f12237461.1399317206.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Paolo Bonzini