The seccomp speculation control operates on all tasks of a process, but
only the current task of a process can update the MSR immediately. For the
other threads the update is deferred to the next context switch.
This creates the following situation with Process A and B:
Process A task 2 and Process B task 1 are pinned on CPU1. Process A task 2
does not have the speculation control TIF bit set. Process B task 1 has the
speculation control TIF bit set.
CPU0 CPU1
MSR bit is set
ProcB.T1 schedules out
ProcA.T2 schedules in
MSR bit is cleared
ProcA.T1
seccomp_update()
set TIF bit on ProcA.T2
ProcB.T1 schedules in
MSR is not updated <-- FAIL
This happens because the context switch code tries to avoid the MSR update
if the speculation control TIF bits of the incoming and the outgoing task
are the same. In the worst case ProcB.T1 and ProcA.T2 are the only tasks
scheduling back and forth on CPU1, which keeps the MSR stale forever.
In theory this could be remedied by IPIs, but chasing the remote task which
could be migrated is complex and full of races.
The straight forward solution is to avoid the asychronous update of the TIF
bit and defer it to the next context switch. The speculation control state
is stored in task_struct::atomic_flags by the prctl and seccomp updates
already.
Add a new TIF_SPEC_FORCE_UPDATE bit and set this after updating the
atomic_flags. Check the bit on context switch and force a synchronous
update of the speculation control if set. Use the same mechanism for
updating the current task.
Reported-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Woodhouse <dwmw@amazon.co.uk>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Casey Schaufler <casey.schaufler@intel.com>
Cc: Asit Mallick <asit.k.mallick@intel.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Jon Masters <jcm@redhat.com>
Cc: Waiman Long <longman9394@gmail.com>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: Dave Stewart <david.c.stewart@intel.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1811272247140.1875@nanos.tec.linutronix.de
The upcoming support for the virtual SPEC_CTRL MSR on AMD needs to reuse
speculative_store_bypass_update() to avoid code duplication. Add an
argument for supplying a thread info (TIF) value and create a wrapper
speculative_store_bypass_update_current() which is used at the existing
call site.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Some AMD processors only support a non-architectural means of enabling
speculative store bypass disable (SSBD). To allow a simplified view of
this to a guest, an architectural definition has been created through a new
CPUID bit, 0x80000008_EBX[25], and a new MSR, 0xc001011f. With this, a
hypervisor can virtualize the existence of this definition and provide an
architectural method for using SSBD to a guest.
Add the new CPUID feature, the new MSR and update the existing SSBD
support to use this MSR when present.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Borislav Petkov <bp@suse.de>
The AMD64_LS_CFG MSR is a per core MSR on Family 17H CPUs. That means when
hyperthreading is enabled the SSBD bit toggle needs to take both cores into
account. Otherwise the following situation can happen:
CPU0 CPU1
disable SSB
disable SSB
enable SSB <- Enables it for the Core, i.e. for CPU0 as well
So after the SSB enable on CPU1 the task on CPU0 runs with SSB enabled
again.
On Intel the SSBD control is per core as well, but the synchronization
logic is implemented behind the per thread SPEC_CTRL MSR. It works like
this:
CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL
i.e. if one of the threads enables a mitigation then this affects both and
the mitigation is only disabled in the core when both threads disabled it.
Add the necessary synchronization logic for AMD family 17H. Unfortunately
that requires a spinlock to serialize the access to the MSR, but the locks
are only shared between siblings.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
The SSBD enumeration is similarly to the other bits magically shared
between Intel and AMD though the mechanisms are different.
Make X86_FEATURE_SSBD synthetic and set it depending on the vendor specific
features or family dependent setup.
Change the Intel bit to X86_FEATURE_SPEC_CTRL_SSBD to denote that SSBD is
controlled via MSR_SPEC_CTRL and fix up the usage sites.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Intel collateral will reference the SSB mitigation bit in IA32_SPEC_CTL[2]
as SSBD (Speculative Store Bypass Disable).
Hence changing it.
It is unclear yet what the MSR_IA32_ARCH_CAPABILITIES (0x10a) Bit(4) name
is going to be. Following the rename it would be SSBD_NO but that rolls out
to Speculative Store Bypass Disable No.
Also fixed the missing space in X86_FEATURE_AMD_SSBD.
[ tglx: Fixup x86_amd_rds_enable() and rds_tif_to_amd_ls_cfg() as well ]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The Speculative Store Bypass vulnerability can be mitigated with the
Reduced Data Speculation (RDS) feature. To allow finer grained control of
this eventually expensive mitigation a per task mitigation control is
required.
Add a new TIF_RDS flag and put it into the group of TIF flags which are
evaluated for mismatch in switch_to(). If these bits differ in the previous
and the next task, then the slow path function __switch_to_xtra() is
invoked. Implement the TIF_RDS dependent mitigation control in the slow
path.
If the prctl for controlling Speculative Store Bypass is disabled or no
task uses the prctl then there is no overhead in the switch_to() fast
path.
Update the KVM related speculation control functions to take TID_RDS into
account as well.
Based on a patch from Tim Chen. Completely rewritten.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Some issues have been reported with the for loop in stop_this_cpu() that
issues the 'wbinvd; hlt' sequence. Reverting this sequence to halt()
has been shown to resolve the issue.
However, the wbinvd is needed when running with SME. The reason for the
wbinvd is to prevent cache flush races between encrypted and non-encrypted
entries that have the same physical address. This can occur when
kexec'ing from memory encryption active to inactive or vice-versa. The
important thing is to not have outside of kernel text memory references
(such as stack usage), so the usage of the native_*() functions is needed
since these expand as inline asm sequences. So instead of reverting the
change, rework the sequence.
Move the wbinvd instruction outside of the for loop as native_wbinvd()
and make its execution conditional on X86_FEATURE_SME. In the for loop,
change the asm 'wbinvd; hlt' sequence back to a halt sequence but use
the native_halt() call.
Fixes: bba4ed011a ("x86/mm, kexec: Allow kexec to be used with SME")
Reported-by: Dave Young <dyoung@redhat.com>
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Dave Young <dyoung@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Yu Chen <yu.c.chen@intel.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: kexec@lists.infradead.org
Cc: ebiederm@redhat.com
Cc: Borislav Petkov <bp@alien8.de>
Cc: Rui Zhang <rui.zhang@intel.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20180117234141.21184.44067.stgit@tlendack-t1.amdoffice.net
Pull x86 page table isolation fixes from Thomas Gleixner:
"A couple of urgent fixes for PTI:
- Fix a PTE mismatch between user and kernel visible mapping of the
cpu entry area (differs vs. the GLB bit) and causes a TLB mismatch
MCE on older AMD K8 machines
- Fix the misplaced CR3 switch in the SYSCALL compat entry code which
causes access to unmapped kernel memory resulting in double faults.
- Fix the section mismatch of the cpu_tss_rw percpu storage caused by
using a different mechanism for declaration and definition.
- Two fixes for dumpstack which help to decode entry stack issues
better
- Enable PTI by default in Kconfig. We should have done that earlier,
but it slipped through the cracks.
- Exclude AMD from the PTI enforcement. Not necessarily a fix, but if
AMD is so confident that they are not affected, then we should not
burden users with the overhead"
* 'x86-pti-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/process: Define cpu_tss_rw in same section as declaration
x86/pti: Switch to kernel CR3 at early in entry_SYSCALL_compat()
x86/dumpstack: Print registers for first stack frame
x86/dumpstack: Fix partial register dumps
x86/pti: Make sure the user/kernel PTEs match
x86/cpu, x86/pti: Do not enable PTI on AMD processors
x86/pti: Enable PTI by default
Pull x86 syscall entry code changes for PTI from Ingo Molnar:
"The main changes here are Andy Lutomirski's changes to switch the
x86-64 entry code to use the 'per CPU entry trampoline stack'. This,
besides helping fix KASLR leaks (the pending Page Table Isolation
(PTI) work), also robustifies the x86 entry code"
* 'WIP.x86-pti.entry-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (26 commits)
x86/cpufeatures: Make CPU bugs sticky
x86/paravirt: Provide a way to check for hypervisors
x86/paravirt: Dont patch flush_tlb_single
x86/entry/64: Make cpu_entry_area.tss read-only
x86/entry: Clean up the SYSENTER_stack code
x86/entry/64: Remove the SYSENTER stack canary
x86/entry/64: Move the IST stacks into struct cpu_entry_area
x86/entry/64: Create a per-CPU SYSCALL entry trampoline
x86/entry/64: Return to userspace from the trampoline stack
x86/entry/64: Use a per-CPU trampoline stack for IDT entries
x86/espfix/64: Stop assuming that pt_regs is on the entry stack
x86/entry/64: Separate cpu_current_top_of_stack from TSS.sp0
x86/entry: Remap the TSS into the CPU entry area
x86/entry: Move SYSENTER_stack to the beginning of struct tss_struct
x86/dumpstack: Handle stack overflow on all stacks
x86/entry: Fix assumptions that the HW TSS is at the beginning of cpu_tss
x86/kasan/64: Teach KASAN about the cpu_entry_area
x86/mm/fixmap: Generalize the GDT fixmap mechanism, introduce struct cpu_entry_area
x86/entry/gdt: Put per-CPU GDT remaps in ascending order
x86/dumpstack: Add get_stack_info() support for the SYSENTER stack
...
CR4 modifications are implemented as RMW operations which update a shadow
variable and write the result to CR4. The RMW operation is protected by
preemption disable, but there is no enforcement or debugging mechanism.
CR4 modifications happen also in interrupt context via
__native_flush_tlb_global(). This implementation does not affect a
interrupted thread context CR4 operation, because the CR4 toggle restores
the original content and does not modify the shadow variable.
So the current situation seems to be safe, but a recent patch tried to add
an actual RMW operation in interrupt context, which will cause subtle
corruptions.
To prevent that and make the CR4 handling future proof:
- Add a lockdep assertion to __cr4_set() which will catch interrupt
enabled invocations
- Disable interrupts in the cr4 manipulator inlines
- Rename cr4_toggle_bits() to cr4_toggle_bits_irqsoff(). This is called
from __switch_to_xtra() where interrupts are already disabled and
performance matters.
All other call sites are not performance critical, so the extra overhead of
an additional local_irq_save/restore() pair is not a problem. If new call
sites care about performance then the necessary _irqsoff() variants can be
added.
[ tglx: Condensed the patch by moving the irq protection inside the
manipulator functions. Updated changelog ]
Signed-off-by: Nadav Amit <namit@vmware.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Luck <tony.luck@intel.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: nadav.amit@gmail.com
Cc: linux-edac@vger.kernel.org
Link: https://lkml.kernel.org/r/20171125032907.2241-3-namit@vmware.com
Thomas Gleixner
Tim Chen
Joerg Roedel
Tom Lendacky
Konrad Rzeszutek Wilk
Linus Torvalds
Nick Desaulniers
Andy Lutomirski
Ingo Molnar
Pravin Shedge
Nadav Amit