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: NTom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>

AMD is proposing a VIRT_SPEC_CTRL MSR to handle the Speculative Store
Bypass Disable via MSR_AMD64_LS_CFG so that guests do not have to care
about the bit position of the SSBD bit and thus facilitate migration.
Also, the sibling coordination on Family 17H CPUs can only be done on
the host.

Extend x86_spec_ctrl_set_guest() and x86_spec_ctrl_restore_host() with an
extra argument for the VIRT_SPEC_CTRL MSR.

Hand in 0 from VMX and in SVM add a new virt_spec_ctrl member to the CPU
data structure which is going to be used in later patches for the actual
implementation.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>

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: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>

Add a ZEN feature bit so family-dependent static_cpu_has() optimizations
can be built for ZEN.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NKonrad 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: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>

The availability of the SPEC_CTRL MSR is enumerated by a CPUID bit on
Intel and implied by IBRS or STIBP support on AMD. That's just confusing
and in case an AMD CPU has IBRS not supported because the underlying
problem has been fixed but has another bit valid in the SPEC_CTRL MSR,
the thing falls apart.

Add a synthetic feature bit X86_FEATURE_MSR_SPEC_CTRL to denote the
availability on both Intel and AMD.

While at it replace the boot_cpu_has() checks with static_cpu_has() where
possible. This prevents late microcode loading from exposing SPEC_CTRL, but
late loading is already very limited as it does not reevaluate the
mitigation options and other bits and pieces. Having static_cpu_has() is
the simplest and least fragile solution.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>

Intel and AMD have different CPUID bits hence for those use synthetic bits
which get set on the respective vendor's in init_speculation_control(). So
that debacles like what the commit message of

  c65732e4 ("x86/cpu: Restore CPUID_8000_0008_EBX reload")

talks about don't happen anymore.
Signed-off-by: NBorislav Petkov <bp@suse.de>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Tested-by: NJörg Otte <jrg.otte@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Link: https://lkml.kernel.org/r/20180504161815.GG9257@pd.tnic

svm_vcpu_run() invokes x86_spec_ctrl_restore_host() after VMEXIT, but
before the host GS is restored. x86_spec_ctrl_restore_host() uses 'current'
to determine the host SSBD state of the thread. 'current' is GS based, but
host GS is not yet restored and the access causes a triple fault.

Move the call after the host GS restore.

Fixes: 885f82bf x86/process: Allow runtime control of Speculative Store Bypass
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: NPaolo Bonzini <pbonzini@redhat.com>

Cast val and (val >> 32) to (u32), so that they fit in a
general-purpose register in both 32-bit and 64-bit code.

[ tglx: Made it u32 instead of uintptr_t ]

Fixes: c65732e4 ("x86/cpu: Restore CPUID_8000_0008_EBX reload")
Signed-off-by: NJim Mattson <jmattson@google.com>
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Acked-by: NLinus Torvalds <torvalds@linux-foundation.org>

Fixes: 7bb4d366 ("x86/bugs: Make cpu_show_common() static")
Fixes: 24f7fc83 ("x86/bugs: Provide boot parameters for the spec_store_bypass_disable mitigation")
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

cpu_show_common() is not used outside of arch/x86/kernel/cpu/bugs.c, so
make it static.
Signed-off-by: NJiri Kosina <jkosina@suse.cz>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

__ssb_select_mitigation() returns one of the members of enum ssb_mitigation,
not ssb_mitigation_cmd; fix the prototype to reflect that.

Fixes: 24f7fc83 ("x86/bugs: Provide boot parameters for the spec_store_bypass_disable mitigation")
Signed-off-by: NJiri Kosina <jkosina@suse.cz>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

Fix some typos, improve formulations, end sentences with a fullstop.
Signed-off-by: NBorislav Petkov <bp@suse.de>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

The style for the 'status' file is CamelCase or this. _.

Fixes: fae1fa0f ("proc: Provide details on speculation flaw mitigations")
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

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: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

Unless explicitly opted out of, anything running under seccomp will have
SSB mitigations enabled. Choosing the "prctl" mode will disable this.

[ tglx: Adjusted it to the new arch_seccomp_spec_mitigate() mechanism ]
Signed-off-by: NKees Cook <keescook@chromium.org>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

The migitation control is simpler to implement in architecture code as it
avoids the extra function call to check the mode. Aside of that having an
explicit seccomp enabled mode in the architecture mitigations would require
even more workarounds.

Move it into architecture code and provide a weak function in the seccomp
code. Remove the 'which' argument as this allows the architecture to decide
which mitigations are relevant for seccomp.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

If a seccomp user is not interested in Speculative Store Bypass mitigation
by default, it can set the new SECCOMP_FILTER_FLAG_SPEC_ALLOW flag when
adding filters.
Signed-off-by: NKees Cook <keescook@chromium.org>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

Use PR_SPEC_FORCE_DISABLE in seccomp() because seccomp does not allow to
widen restrictions.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

For certain use cases it is desired to enforce mitigations so they cannot
be undone afterwards. That's important for loader stubs which want to
prevent a child from disabling the mitigation again. Will also be used for
seccomp(). The extra state preserving of the prctl state for SSB is a
preparatory step for EBPF dymanic speculation control.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

There's no reason for these to be changed after boot.
Signed-off-by: NKees Cook <keescook@chromium.org>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

When speculation flaw mitigations are opt-in (via prctl), using seccomp
will automatically opt-in to these protections, since using seccomp
indicates at least some level of sandboxing is desired.
Signed-off-by: NKees Cook <keescook@chromium.org>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

As done with seccomp and no_new_privs, also show speculation flaw
mitigation state in /proc/$pid/status.
Signed-off-by: NKees Cook <keescook@chromium.org>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

Adjust arch_prctl_get/set_spec_ctrl() to operate on tasks other than
current.

This is needed both for /proc/$pid/status queries and for seccomp (since
thread-syncing can trigger seccomp in non-current threads).
Signed-off-by: NKees Cook <keescook@chromium.org>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>

Add prctl based control for Speculative Store Bypass mitigation and make it
the default mitigation for Intel and AMD.

Andi Kleen provided the following rationale (slightly redacted):

 There are multiple levels of impact of Speculative Store Bypass:

 1) JITed sandbox.
    It cannot invoke system calls, but can do PRIME+PROBE and may have call
    interfaces to other code

 2) Native code process.
    No protection inside the process at this level.

 3) Kernel.

 4) Between processes. 

 The prctl tries to protect against case (1) doing attacks.

 If the untrusted code can do random system calls then control is already
 lost in a much worse way. So there needs to be system call protection in
 some way (using a JIT not allowing them or seccomp). Or rather if the
 process can subvert its environment somehow to do the prctl it can already
 execute arbitrary code, which is much worse than SSB.

 To put it differently, the point of the prctl is to not allow JITed code
 to read data it shouldn't read from its JITed sandbox. If it already has
 escaped its sandbox then it can already read everything it wants in its
 address space, and do much worse.

 The ability to control Speculative Store Bypass allows to enable the
 protection selectively without affecting overall system performance.

Based on an initial patch from Tim Chen. Completely rewritten.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>

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: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>

Add two new prctls to control aspects of speculation related vulnerabilites
and their mitigations to provide finer grained control over performance
impacting mitigations.

PR_GET_SPECULATION_CTRL returns the state of the speculation misfeature
which is selected with arg2 of prctl(2). The return value uses bit 0-2 with
the following meaning:

Bit  Define           Description
0    PR_SPEC_PRCTL    Mitigation can be controlled per task by
                      PR_SET_SPECULATION_CTRL
1    PR_SPEC_ENABLE   The speculation feature is enabled, mitigation is
                      disabled
2    PR_SPEC_DISABLE  The speculation feature is disabled, mitigation is
                      enabled

If all bits are 0 the CPU is not affected by the speculation misfeature.

If PR_SPEC_PRCTL is set, then the per task control of the mitigation is
available. If not set, prctl(PR_SET_SPECULATION_CTRL) for the speculation
misfeature will fail.

PR_SET_SPECULATION_CTRL allows to control the speculation misfeature, which
is selected by arg2 of prctl(2) per task. arg3 is used to hand in the
control value, i.e. either PR_SPEC_ENABLE or PR_SPEC_DISABLE.

The common return values are:

EINVAL  prctl is not implemented by the architecture or the unused prctl()
        arguments are not 0
ENODEV  arg2 is selecting a not supported speculation misfeature

PR_SET_SPECULATION_CTRL has these additional return values:

ERANGE  arg3 is incorrect, i.e. it's not either PR_SPEC_ENABLE or PR_SPEC_DISABLE
ENXIO   prctl control of the selected speculation misfeature is disabled

The first supported controlable speculation misfeature is
PR_SPEC_STORE_BYPASS. Add the define so this can be shared between
architectures.

Based on an initial patch from Tim Chen and mostly rewritten.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>

Having everything in nospec-branch.h creates a hell of dependencies when
adding the prctl based switching mechanism. Move everything which is not
required in nospec-branch.h to spec-ctrl.h and fix up the includes in the
relevant files.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

Expose the CPUID.7.EDX[31] bit to the guest, and also guard against various
combinations of SPEC_CTRL MSR values.

The handling of the MSR (to take into account the host value of SPEC_CTRL
Bit(2)) is taken care of in patch:

  KVM/SVM/VMX/x86/spectre_v2: Support the combination of guest and host IBRS
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

AMD does not need the Speculative Store Bypass mitigation to be enabled.

The parameters for this are already available and can be done via MSR
C001_1020. Each family uses a different bit in that MSR for this.

[ tglx: Expose the bit mask via a variable and move the actual MSR fiddling
  	into the bugs code as that's the right thing to do and also required
	to prepare for dynamic enable/disable ]
Suggested-by: NBorislav Petkov <bp@suse.de>
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

Intel and AMD SPEC_CTRL (0x48) MSR semantics may differ in the
future (or in fact use different MSRs for the same functionality).

As such a run-time mechanism is required to whitelist the appropriate MSR
values.

[ tglx: Made the variable __ro_after_init ]
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

Intel CPUs expose methods to:

 - Detect whether RDS capability is available via CPUID.7.0.EDX[31],

 - The SPEC_CTRL MSR(0x48), bit 2 set to enable RDS.

 - MSR_IA32_ARCH_CAPABILITIES, Bit(4) no need to enable RRS.

With that in mind if spec_store_bypass_disable=[auto,on] is selected set at
boot-time the SPEC_CTRL MSR to enable RDS if the platform requires it.

Note that this does not fix the KVM case where the SPEC_CTRL is exposed to
guests which can muck with it, see patch titled :
 KVM/SVM/VMX/x86/spectre_v2: Support the combination of guest and host IBRS.

And for the firmware (IBRS to be set), see patch titled:
 x86/spectre_v2: Read SPEC_CTRL MSR during boot and re-use reserved bits

[ tglx: Distangled it from the intel implementation and kept the call order ]
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

Contemporary high performance processors use a common industry-wide
optimization known as "Speculative Store Bypass" in which loads from
addresses to which a recent store has occurred may (speculatively) see an
older value. Intel refers to this feature as "Memory Disambiguation" which
is part of their "Smart Memory Access" capability.

Memory Disambiguation can expose a cache side-channel attack against such
speculatively read values. An attacker can create exploit code that allows
them to read memory outside of a sandbox environment (for example,
malicious JavaScript in a web page), or to perform more complex attacks
against code running within the same privilege level, e.g. via the stack.

As a first step to mitigate against such attacks, provide two boot command
line control knobs:

 nospec_store_bypass_disable
 spec_store_bypass_disable=[off,auto,on]

By default affected x86 processors will power on with Speculative
Store Bypass enabled. Hence the provided kernel parameters are written
from the point of view of whether to enable a mitigation or not.
The parameters are as follows:

 - auto - Kernel detects whether your CPU model contains an implementation
	  of Speculative Store Bypass and picks the most appropriate
	  mitigation.

 - on   - disable Speculative Store Bypass
 - off  - enable Speculative Store Bypass

[ tglx: Reordered the checks so that the whole evaluation is not done
  	when the CPU does not support RDS ]
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

Add the CPU feature bit CPUID.7.0.EDX[31] which indicates whether the CPU
supports Reduced Data Speculation.

[ tglx: Split it out from a later patch ]
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

Add the sysfs file for the new vulerability. It does not do much except
show the words 'Vulnerable' for recent x86 cores.

Intel cores prior to family 6 are known not to be vulnerable, and so are
some Atoms and some Xeon Phi.

It assumes that older Cyrix, Centaur, etc. cores are immune.
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

A guest may modify the SPEC_CTRL MSR from the value used by the
kernel. Since the kernel doesn't use IBRS, this means a value of zero is
what is needed in the host.

But the 336996-Speculative-Execution-Side-Channel-Mitigations.pdf refers to
the other bits as reserved so the kernel should respect the boot time
SPEC_CTRL value and use that.

This allows to deal with future extensions to the SPEC_CTRL interface if
any at all.

Note: This uses wrmsrl() instead of native_wrmsl(). I does not make any
difference as paravirt will over-write the callq *0xfff.. with the wrmsrl
assembler code.
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

The 336996-Speculative-Execution-Side-Channel-Mitigations.pdf refers to all
the other bits as reserved. The Intel SDM glossary defines reserved as
implementation specific - aka unknown.

As such at bootup this must be taken it into account and proper masking for
the bits in use applied.

A copy of this document is available at
https://bugzilla.kernel.org/show_bug.cgi?id=199511

[ tglx: Made x86_spec_ctrl_base __ro_after_init ]
Suggested-by: NJon Masters <jcm@redhat.com>
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

Those SysFS functions have a similar preamble, as such make common
code to handle them.
Suggested-by: NBorislav Petkov <bp@suse.de>
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

Combine the various logic which goes through all those
x86_cpu_id matching structures in one function.
Suggested-by: NBorislav Petkov <bp@suse.de>
Signed-off-by: NKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NBorislav Petkov <bp@suse.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>

The macro is not type safe and I did look for why that "g" constraint for
the asm doesn't work: it's because the asm is more fundamentally wrong.

It does

        movl %[val], %%eax

but "val" isn't a 32-bit value, so then gcc will pass it in a register, 
and generate code like

        movl %rsi, %eax

and gas will complain about a nonsensical 'mov' instruction (it's moving a 
64-bit register to a 32-bit one).

Passing it through memory will just hide the real bug - gcc still thinks 
the memory location is 64-bit, but the "movl" will only load the first 32 
bits and it all happens to work because x86 is little-endian.

Convert it to a type safe inline function with a little trick which hands
the feature into the ALTERNATIVE macro.
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Reviewed-by: NIngo Molnar <mingo@kernel.org>