Currently, requesting kernel FPU access doesn't distinguish which parts of
the extended ("FPU") state are needed. This is nice for simplicity, but
there are a few cases in which it's suboptimal:
- The vast majority of in-kernel FPU users want XMM/YMM/ZMM state but do
not use legacy 387 state. These users want MXCSR initialized but don't
care about the FPU control word. Skipping FNINIT would save time.
(Empirically, FNINIT is several times slower than LDMXCSR.)
- Code that wants MMX doesn't want or need MXCSR initialized.
_mmx_memcpy(), for example, can run before CR4.OSFXSR gets set, and
initializing MXCSR will fail because LDMXCSR generates an #UD when the
aforementioned CR4 bit is not set.
- Any future in-kernel users of XFD (eXtended Feature Disable)-capable
dynamic states will need special handling.
Add a more specific API that allows callers to specify exactly what they
want.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Krzysztof Piotr Olędzki <ole@ans.pl>
Link: https://lkml.kernel.org/r/aff1cac8b8fc7ee900cf73e8f2369966621b053f.1611205691.git.luto@kernel.org
membarrier()'s MEMBARRIER_CMD_PRIVATE_EXPEDITED_SYNC_CORE is documented as
syncing the core on all sibling threads but not necessarily the calling
thread. This behavior is fundamentally buggy and cannot be used safely.
Suppose a user program has two threads. Thread A is on CPU 0 and thread B
is on CPU 1. Thread A modifies some text and calls
membarrier(MEMBARRIER_CMD_PRIVATE_EXPEDITED_SYNC_CORE).
Then thread B executes the modified code. If, at any point after
membarrier() decides which CPUs to target, thread A could be preempted and
replaced by thread B on CPU 0. This could even happen on exit from the
membarrier() syscall. If this happens, thread B will end up running on CPU
0 without having synced.
In principle, this could be fixed by arranging for the scheduler to issue
sync_core_before_usermode() whenever switching between two threads in the
same mm if there is any possibility of a concurrent membarrier() call, but
this would have considerable overhead. Instead, make membarrier() sync the
calling CPU as well.
As an optimization, this avoids an extra smp_mb() in the default
barrier-only mode and an extra rseq preempt on the caller.
Fixes: 70216e18e5 ("membarrier: Provide core serializing command, *_SYNC_CORE")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Link: https://lore.kernel.org/r/250ded637696d490c69bef1877148db86066881c.1607058304.git.luto@kernel.org
membarrier() does not explicitly sync_core() remote CPUs; instead, it
relies on the assumption that an IPI will result in a core sync. On x86,
this may be true in practice, but it's not architecturally reliable. In
particular, the SDM and APM do not appear to guarantee that interrupt
delivery is serializing. While IRET does serialize, IPI return can
schedule, thereby switching to another task in the same mm that was
sleeping in a syscall. The new task could then SYSRET back to usermode
without ever executing IRET.
Make this more robust by explicitly calling sync_core_before_usermode()
on remote cores. (This also helps people who search the kernel tree for
instances of sync_core() and sync_core_before_usermode() -- one might be
surprised that the core membarrier code doesn't currently show up in a
such a search.)
Fixes: 70216e18e5 ("membarrier: Provide core serializing command, *_SYNC_CORE")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/776b448d5f7bd6b12690707f5ed67bcda7f1d427.1607058304.git.luto@kernel.org
sync_core_before_usermode() had an incorrect optimization. If the kernel
returns from an interrupt, it can get to usermode without IRET. It just has
to schedule to a different task in the same mm and do SYSRET. Fortunately,
there were no callers of sync_core_before_usermode() that could have had
in_irq() or in_nmi() equal to true, because it's only ever called from the
scheduler.
While at it, clarify a related comment.
Fixes: 70216e18e5 ("membarrier: Provide core serializing command, *_SYNC_CORE")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/5afc7632be1422f91eaf7611aaaa1b5b8580a086.1607058304.git.luto@kernel.org
On Xen PV, #DB doesn't use IST. It still needs to be correctly routed
depending on whether it came from user or kernel mode.
Get rid of DECLARE/DEFINE_IDTENTRY_XEN -- it was too hard to follow the
logic. Instead, route #DB and NMI through DECLARE/DEFINE_IDTENTRY_RAW on
Xen, and do the right thing for #DB. Also add more warnings to the
exc_debug* handlers to make this type of failure more obvious.
This fixes various forms of corruption that happen when usermode
triggers #DB on Xen PV.
Fixes: 4c0dcd8350 ("x86/entry: Implement user mode C entry points for #DB and #MCE")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/4163e733cce0b41658e252c6c6b3464f33fdff17.1593795633.git.luto@kernel.org
Debuggers expect that doing PTRACE_GETREGS, then poking at a tracee
and maybe letting it run for a while, then doing PTRACE_SETREGS will
put the tracee back where it was. In the specific case of a 32-bit
tracer and tracee, the PTRACE_GETREGS/SETREGS data structure doesn't
have fs_base or gs_base fields, so FSBASE and GSBASE fields are
never stored anywhere. Everything used to still work because
nonzero FS or GS would result full reloads of the segment registers
when the tracee resumes, and the bases associated with FS==0 or
GS==0 are irrelevant to 32-bit code.
Adding FSGSBASE support broke this: when FSGSBASE is enabled, FSBASE
and GSBASE are now restored independently of FS and GS for all tasks
when context-switched in. This means that, if a 32-bit tracer
restores a previous state using PTRACE_SETREGS but the tracee's
pre-restore and post-restore bases don't match, then the tracee is
resumed with the wrong base.
Fix it by explicitly loading the base when a 32-bit tracer pokes FS
or GS on a 64-bit kernel.
Also add a test case.
Fixes: 673903495c ("x86/process/64: Use FSBSBASE in switch_to() if available")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/229cc6a50ecbb701abd50fe4ddaf0eda888898cd.1593192140.git.luto@kernel.org
There are several copies of get_eflags() and set_eflags() and they all are
buggy. Consolidate them and fix them. The fixes are:
Add memory clobbers. These are probably unnecessary but they make sure
that the compiler doesn't move something past one of these calls when it
shouldn't.
Respect the redzone on x86_64. There has no failure been observed related
to this, but it's definitely a bug.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/982ce58ae8dea2f1e57093ee894760e35267e751.1593191971.git.luto@kernel.org
The SYSENTER frame setup was nonsense. It worked by accident because the
normal code into which the Xen asm jumped (entry_SYSENTER_32/compat) threw
away SP without touching the stack. entry_SYSENTER_compat was recently
modified such that it relied on having a valid stack pointer, so now the
Xen asm needs to invoke it with a valid stack.
Fix it up like SYSCALL: use the Xen-provided frame and skip the bare
metal prologue.
Fixes: 1c3e5d3f60 ("x86/entry: Make entry_64_compat.S objtool clean")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Link: https://lkml.kernel.org/r/947880c41ade688ff4836f665d0c9fcaa9bd1201.1593191971.git.luto@kernel.org
If the kernel erroneously allows WRGSBASE and user code writes a
negative value, paranoid_entry will get confused. Check for this by
writing a negative value to GSBASE and doing SYSENTER with TF set. A
successful run looks like:
[RUN] SYSENTER with TF, invalid state, and GSBASE < 0
[SKIP] Illegal instruction
A failed run causes a kernel hang, and I believe it's because we
double-fault and then get a never ending series of page faults and,
when we exhaust the double fault stack we double fault again,
starting the process over.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/f4f71efc91b9eae5e3dae21c9aee1c70cf5f370e.1590620529.git.luto@kernel.org
With the new FSGSBASE instructions, FS and GSABSE can be efficiently read
and writen in __switch_to(). Use that capability to preserve the full
state.
This will enable user code to do whatever it wants with the new
instructions without any kernel-induced gotchas. (There can still be
architectural gotchas: movl %gs,%eax; movl %eax,%gs may change GSBASE if
WRGSBASE was used, but users are expected to read the CPU manual before
doing things like that.)
This is a considerable speedup. It seems to save about 100 cycles
per context switch compared to the baseline 4.6-rc1 behavior on a
Skylake laptop. This is mostly due to avoiding the WRMSR operation.
[ chang: 5~10% performance improvements were seen with a context switch
benchmark that ran threads with different FS/GSBASE values (to the
baseline 4.16). Minor edit on the changelog. ]
[ tglx: Masaage changelog ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Chang S. Bae <chang.seok.bae@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Link: https://lkml.kernel.org/r/1557309753-24073-8-git-send-email-chang.seok.bae@intel.com
Link: https://lkml.kernel.org/r/20200528201402.1708239-6-sashal@kernel.org
BUG/WARN are cleverly optimized using UD2 to handle the BUG/WARN out of
line in an exception fixup.
But if BUG or WARN is issued in a funny RCU context, then the
idtentry_enter...() path might helpfully WARN that the RCU context is
invalid, which results in infinite recursion.
Split the BUG/WARN handling into an nmi_enter()/nmi_exit() path in
exc_invalid_op() to increase the chance to survive the experience.
[ tglx: Make the declaration match the implementation ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/f8fe40e0088749734b4435b554f73eee53dcf7a8.1591932307.git.luto@kernel.org
KVM overloads #PF to indicate two types of not-actually-page-fault
events. Right now, the KVM guest code intercepts them by modifying
the IDT and hooking the #PF vector. This makes the already fragile
fault code even harder to understand, and it also pollutes call
traces with async_page_fault and do_async_page_fault for normal page
faults.
Clean it up by moving the logic into do_page_fault() using a static
branch. This gets rid of the platform trap_init override mechanism
completely.
[ tglx: Fixed up 32bit, removed error code from the async functions and
massaged coding style ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Alexandre Chartre <alexandre.chartre@oracle.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200505134059.169270470@linutronix.de
/dev/random and getrandom() never warn. Split the meat of
urandom_read() into urandom_read_nowarn() and leave the warning code
in urandom_read().
This has no effect on kernel behavior, but it makes subsequent
patches more straightforward. It also makes the fact that
getrandom() never warns more obvious.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Link: https://lore.kernel.org/r/c87ab200588de746431d9f916501ef11e5242b13.1577088521.git.luto@kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
seg_segment_reg() should be unreachable with task == current.
Rather than confusingly trying to make it work, just explicitly
disable this case.
(regset->get is used for current in the coredump code, but the ->set
interface is only used for ptrace, and you can't ptrace yourself.)
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The old x86_32 doublefault_fn() was old and crufty, and it did not
even try to recover. do_double_fault() is much nicer. Rewrite the
32-bit double fault code to sanitize CPU state and call
do_double_fault(). This is mostly an exercise i386 archaeology.
With this patch applied, 32-bit double faults get a real stack trace,
just like 64-bit double faults.
[ mingo: merged the patch to a later kernel base. ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There are three problems with the current layout of the doublefault
stack and TSS. First, the TSS is only cacheline-aligned, which is
not enough -- if the hardware portion of the TSS (struct x86_hw_tss)
crosses a page boundary, horrible things happen [0]. Second, the
stack and TSS are global, so simultaneous double faults on different
CPUs will cause massive corruption. Third, the whole mechanism
won't work if user CR3 is loaded, resulting in a triple fault [1].
Let the doublefault stack and TSS share a page (which prevents the
TSS from spanning a page boundary), make it percpu, and move it into
cpu_entry_area. Teach the stack dump code about the doublefault
stack.
[0] Real hardware will read past the end of the page onto the next
*physical* page if a task switch happens. Virtual machines may
have any number of bugs, and I would consider it reasonable for
a VM to summarily kill the guest if it tries to task-switch to
a page-spanning TSS.
[1] Real hardware triple faults. At least some VMs seem to hang.
I'm not sure what's going on.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The 64-bit doublefault handler is much nicer than the 32-bit one.
As a first step toward unifying them, make the 64-bit handler
self-contained. This should have no effect no functional effect
except in the odd case of x86_64 with CONFIG_DOUBLEFAULT=n in which
case it will change the logging a bit.
This also gets rid of CONFIG_DOUBLEFAULT configurability on 64-bit
kernels. It didn't do anything useful -- CONFIG_DOUBLEFAULT=n
didn't actually disable doublefault handling on x86_64.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
UNWIND_ESPFIX_STACK needs to read the GDT, and the GDT mapping that
can be accessed via %fs is not mapped in the user pagetables. Use
SGDT to find the cpu_entry_area mapping and read the espfix offset
from that instead.
Reported-and-tested-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
If the kernel accidentally uses DS or ES while the user values are
loaded, it will work fine for sane userspace. In the interest of
simulating maximally insane userspace, make sigreturn_32 zero out DS
and ES for the nasty parts so that inadvertent use of these segments
will crash.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@kernel.org
For reasons that I haven't quite fully diagnosed, running
mov_ss_trap_32 on a 32-bit kernel results in an infinite loop in
userspace. This appears to be because the hacky SYSENTER test
doesn't segfault as desired; instead it corrupts the program state
such that it infinite loops.
Fix it by explicitly clearing EBP before doing SYSENTER. This will
give a more reliable segfault.
Fixes: 59c2a7226f ("x86/selftests: Add mov_to_ss test")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@kernel.org
Right now, we do some fancy parts of the exception entry path while SS
might have a nonzero base: we fill in regs->ss and regs->sp, and we
consider switching to the kernel stack. This results in regs->ss and
regs->sp referring to a non-flat stack and it may result in
overflowing the entry stack. The former issue means that we can try to
call iret_exc on a non-flat stack, which doesn't work.
Tested with selftests/x86/sigreturn_32.
Fixes: 45d7b25574 ("x86/entry/32: Enter the kernel via trampoline stack")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@kernel.org
This will allow us to get percpu access working before FIXUP_FRAME,
which will allow us to unwind ESPFIX earlier.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@kernel.org
When re-building the IRET frame we use %eax as an destination %esp,
make sure to then also match the segment for when there is a nonzero
SS base (ESPFIX).
[peterz: Changelog and minor edits]
Fixes: 3c88c692c2 ("x86/stackframe/32: Provide consistent pt_regs")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@kernel.org
The double fault TSS was missing GS setup, which is needed for stack
canaries to work.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@kernel.org
For unfortunate historical reasons, the x32 syscalls and the x86_64
syscalls are not all numbered the same. As an example, ioctl() is nr 16 on
x86_64 but 514 on x32.
This has potentially nasty consequences, since it means that there are two
valid RAX values to do ioctl(2) and two invalid RAX values. The valid
values are 16 (i.e. ioctl(2) using the x86_64 ABI) and (514 | 0x40000000)
(i.e. ioctl(2) using the x32 ABI).
The invalid values are 514 and (16 | 0x40000000). 514 will enter the
"COMPAT_SYSCALL_DEFINE3(ioctl, ...)" entry point with in_compat_syscall()
and in_x32_syscall() returning false, whereas (16 | 0x40000000) will enter
the native entry point with in_compat_syscall() and in_x32_syscall()
returning true. Both are bogus, and both will exercise code paths in the
kernel and in any running seccomp filters that really ought to be
unreachable.
Splitting out the x32 syscalls into their own tables, allows both bogus
invocations to return -ENOSYS. I've checked glibc, musl, and Bionic, and
all of them appear to call syscalls with their correct numbers, so this
change should have no effect on them.
There is an added benefit going forward: new syscalls that need special
handling on x32 can share the same number on x32 and x86_64. This means
that the special syscall range 512-547 can be treated as a legacy wart
instead of something that may need to be extended in the future.
Also add a selftest to verify the new behavior.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/208024256b764312598f014ebfb0a42472c19354.1562185330.git.luto@kernel.org
I'm working on some code that detects at build time if there's a
COMPAT_SYSCALL_DEFINE() that is not referenced in the x86 syscall tables.
It catches three offenders: rt_sigsuspend(), rt_sigprocmask(), and
sendfile64().
For rt_sigsuspend() and rt_sigprocmask(), the only potential difference
between the native and compat versions is that the compat version converts
the sigset_t, but, on little endian architectures, the conversion is a
no-op. This is why they both currently work on x86.
To make the code more consistent, and to make the upcoming patches work,
rewire x86 to use the compat vesions.
sendfile64() is more complicated, and will be addressed separately.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/51643ac3157b5921eae0e172a8a0b1d953e68ebb.1562185330.git.luto@kernel.org
Currently, it's an int. This is bizarre. Fortunately, the code using it
still works: ~__X32_SYSCALL_BIT is also int, so, if nr is unsigned long,
then C kindly sign-extends the ~__X32_SYSCALL_BIT part, and it actually
results in the desired value.
This is far more subtle than it deserves to be. Syscall numbers are, for
all practical purposes, unsigned long, so make __X32_SYSCALL_BIT be
unsigned long.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/99b0d83ad891c67105470a1a6b63243fd63a5061.1562185330.git.luto@kernel.org
The APIC, per spec, is fundamentally confused and thinks that interrupt
vectors 16-31 are valid. This makes no sense -- the CPU reserves vectors
0-31 for exceptions (faults, traps, etc). Obviously, no device should
actually produce an interrupt with vector 16-31, but robustness can be
improved by setting the APIC TPR class to 1, which will prevent delivery of
an interrupt with a vector below 32.
Note: This is *not* intended as a security measure against attackers who
control malicious hardware. Any PCI or similar hardware that can be
controlled by an attacker MUST be behind a functional IOMMU that remaps
interrupts. The purpose of this change is to reduce the chance that a
certain class of device malfunctions crashes the kernel in hard-to-debug
ways.
Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/dc04a9f8b234d7b0956a8d2560b8945bcd9c4bf7.1563117760.git.luto@kernel.org
paranoid_exit needs to restore CR3 before GSBASE. Doing it in the opposite
order crashes if the exception came from a context with user GSBASE and
user CR3 -- RESTORE_CR3 cannot resture user CR3 if run with user GSBASE.
This results in infinitely recursing exceptions if user code does SYSENTER
with TF set if both FSGSBASE and PTI are enabled.
The old code worked if user code just set TF without SYSENTER because #DB
from user mode is special cased in idtentry and paranoid_exit doesn't run.
Fix it by cleaning up the spaghetti code. All that paranoid_exit needs to
do is to disable IRQs, handle IRQ tracing, then restore CR3, and restore
GSBASE. Simply do those actions in that order.
Fixes: 708078f657 ("x86/entry/64: Handle FSGSBASE enabled paranoid entry/exit")
Reported-by: Vegard Nossum <vegard.nossum@oracle.com>
Signed-off-by: Chang S. Bae <chang.seok.bae@intel.com>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "H . Peter Anvin" <hpa@zytor.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ravi Shankar <ravi.v.shankar@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Link: https://lkml.kernel.org/r/59725ceb08977359489fbed979716949ad45f616.1562035429.git.luto@kernel.org
With vsyscall emulation on, a readable vsyscall page is still exposed that
contains syscall instructions that validly implement the vsyscalls.
This is required because certain dynamic binary instrumentation tools
attempt to read the call targets of call instructions in the instrumented
code. If the instrumented code uses vsyscalls, then the vsyscall page needs
to contain readable code.
Unfortunately, leaving readable memory at a deterministic address can be
used to help various ASLR bypasses, so some hardening value can be gained
by disallowing vsyscall reads.
Given how rarely the vsyscall page needs to be readable, add a mechanism to
make the vsyscall page be execute only.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Kees Cook <keescook@chromium.org>
Cc: Florian Weimer <fweimer@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Kernel Hardening <kernel-hardening@lists.openwall.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/d17655777c21bc09a7af1bbcf74e6f2b69a51152.1561610354.git.luto@kernel.org
With the new FSGSBASE instructions, FS and GSABSE can be efficiently read
and writen in __switch_to(). Use that capability to preserve the full
state.
This will enable user code to do whatever it wants with the new
instructions without any kernel-induced gotchas. (There can still be
architectural gotchas: movl %gs,%eax; movl %eax,%gs may change GSBASE if
WRGSBASE was used, but users are expected to read the CPU manual before
doing things like that.)
This is a considerable speedup. It seems to save about 100 cycles
per context switch compared to the baseline 4.6-rc1 behavior on a
Skylake laptop.
[ chang: 5~10% performance improvements were seen with a context switch
benchmark that ran threads with different FS/GSBASE values (to the
baseline 4.16). Minor edit on the changelog. ]
[ tglx: Masaage changelog ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Chang S. Bae <chang.seok.bae@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Cc: Ravi Shankar <ravi.v.shankar@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Link: https://lkml.kernel.org/r/1557309753-24073-8-git-send-email-chang.seok.bae@intel.com
GCC 5.5.0 sometimes cleverly hoists reads of the pvclock and/or hvclock
pages before the vclock mode checks. This creates a path through
vclock_gettime() in which no vclock is enabled at all (due to disabled
TSC on old CPUs, for example) but the pvclock or hvclock page
nevertheless read. This will segfault on bare metal.
This fixes commit 459e3a2153 ("gcc-9: properly declare the
{pv,hv}clock_page storage") in the sense that, before that commit, GCC
didn't seem to generate the offending code. There was nothing wrong
with that commit per se, and -stable maintainers should backport this to
all supported kernels regardless of whether the offending commit was
present, since the same crash could just as easily be triggered by the
phase of the moon.
On GCC 9.1.1, this doesn't seem to affect the generated code at all, so
I'm not too concerned about performance regressions from this fix.
Cc: stable@vger.kernel.org
Cc: x86@kernel.org
Cc: Borislav Petkov <bp@alien8.de>
Reported-by: Duncan Roe <duncan_roe@optusnet.com.au>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The IRQ stack lives in percpu space, so an IRQ handler that overflows it
will overwrite other data structures.
Use vmap() to remap the IRQ stack so that it will have the usual guard
pages that vmap()/vmalloc() allocations have. With this, the kernel will
panic immediately on an IRQ stack overflow.
[ tglx: Move the map code to a proper place and invoke it only when a CPU
is about to be brought online. No point in installing the map at
early boot for all possible CPUs. Fail the CPU bringup if the vmap()
fails as done for all other preparatory stages in CPU hotplug. ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Nicolai Stange <nstange@suse.de>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190414160146.363733568@linutronix.de
Andy Lutomirski