Move smp_num_siblings and cpu_llc_id to cpu/common.c so that they're
always present as symbols and not only in the CONFIG_SMP case. Then,
other code using them doesn't need ugly ifdeffery anymore. Get rid of
some ifdeffery.
The introduction of generic_max_swapfile_size and arch-specific versions has
broken linking on x86 with CONFIG_SWAP=n due to undefined reference to
'generic_max_swapfile_size'. Fix it by compiling the x86-specific
max_swapfile_size() only with CONFIG_SWAP=y.
Commit 0cc3cd21657b ("cpu/hotplug: Boot HT siblings at least once")
breaks non-SMP builds.
[ I suspect the 'bool' fields should just be made to be bitfields and be
exposed regardless of configuration, but that's a separate cleanup
that I'll leave to the owners of this file for later. - Linus ]
Fixes: 0cc3cd21657b ("cpu/hotplug: Boot HT siblings at least once") Cc: Dave Hansen <dave.hansen@intel.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Tony Luck <tony.luck@intel.com> Signed-off-by: Abel Vesa <abelvesa@linux.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The function has an inline "return false;" definition with CONFIG_SMP=n
but the "real" definition is also visible leading to "redefinition of
‘apic_id_is_primary_thread’" compiler error.
[ ... and some older changes in the 4.17.y backport too ...] Signed-off-by: David Woodhouse <dwmw@amazon.co.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This will be used by 'perf trace' to show these strings when beautifying
the prctl syscall args. At some point we'll be able to say something
like:
'perf trace --all-cpus -e prctl(option=*SPEC*)'
To filter by arg by name.
This silences this warning when building tools/perf:
Warning: Kernel ABI header at 'tools/include/uapi/linux/prctl.h' differs from latest version at 'include/uapi/linux/prctl.h'
Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Wang Nan <wangnan0@huawei.com> Link: https://lkml.kernel.org/n/tip-zztsptwhc264r8wg44tqh5gp@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: David Woodhouse <dwmw@amazon.co.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The mmio tracer sets io mapping PTEs and PMDs to non present when enabled
without inverting the address bits, which makes the PTE entry vulnerable
for L1TF.
Make it use the right low level macros to actually invert the address bits
to protect against L1TF.
In principle this could be avoided because MMIO tracing is not likely to be
enabled on production machines, but the fix is straigt forward and for
consistency sake it's better to get rid of the open coded PTE manipulation.
set_memory_np() is used to mark kernel mappings not present, but it has
it's own open coded mechanism which does not have the L1TF protection of
inverting the address bits.
Replace the open coded PTE manipulation with the L1TF protecting low level
PTE routines.
Some cases in THP like:
- MADV_FREE
- mprotect
- split
mark the PMD non present for temporarily to prevent races. The window for
an L1TF attack in these contexts is very small, but it wants to be fixed
for correctness sake.
Use the proper low level functions for pmd/pud_mknotpresent() to address
this.
For kernel mappings PAGE_PROTNONE is not necessarily set for a non present
mapping, but the inversion logic explicitely checks for !PRESENT and
PROT_NONE.
Remove the PROT_NONE check and make the inversion unconditional for all not
present mappings.
Josh reported that the late SMT evaluation in cpu_smt_state_init() sets
cpu_smt_control to CPU_SMT_NOT_SUPPORTED in case that 'nosmt' was supplied
on the kernel command line as it cannot differentiate between SMT disabled
by BIOS and SMT soft disable via 'nosmt'. That wreckages the state and
makes the sysfs interface unusable.
Rework this so that during bringup of the non boot CPUs the availability of
SMT is determined in cpu_smt_allowed(). If a newly booted CPU is not a
'primary' thread then set the local cpu_smt_available marker and evaluate
this explicitely right after the initial SMP bringup has finished.
SMT evaulation on x86 is a trainwreck as the firmware has all the
information _before_ booting the kernel, but there is no interface to query
it.
Fixes: 73d5e2b47264 ("cpu/hotplug: detect SMT disabled by BIOS") Reported-by: Josh Poimboeuf <jpoimboe@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When nested virtualization is in use, VMENTER operations from the nested
hypervisor into the nested guest will always be processed by the bare metal
hypervisor, and KVM's "conditional cache flushes" mode in particular does a
flush on nested vmentry. Therefore, include the "skip L1D flush on
vmentry" bit in KVM's suggested ARCH_CAPABILITIES setting.
Add the relevant Documentation.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Bit 3 of ARCH_CAPABILITIES tells a hypervisor that L1D flush on vmentry is
not needed. Add a new value to enum vmx_l1d_flush_state, which is used
either if there is no L1TF bug at all, or if bit 3 is set in ARCH_CAPABILITIES.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The last missing piece to having vmx_l1d_flush() take interrupts after
VMEXIT into account is to set the kvm_cpu_l1tf_flush_l1d per-cpu flag on
irq entry.
Issue calls to kvm_set_cpu_l1tf_flush_l1d() from entering_irq(),
ipi_entering_ack_irq(), smp_reschedule_interrupt() and
uv_bau_message_interrupt().
Suggested-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Nicolai Stange <nstange@suse.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This causes compilation errors because of the header guards becoming
effective in the second inclusion: symbols/macros that had been defined
before wouldn't be available to intermediate headers in the #include chain
anymore.
A possible workaround would be to move the definition of irq_cpustat_t
into its own header and include that from both, asm/hardirq.h and
asm/apic.h.
However, this wouldn't solve the real problem, namely asm/harirq.h
unnecessarily pulling in all the linux/irq.h cruft: nothing in
asm/hardirq.h itself requires it. Also, note that there are some other
archs, like e.g. arm64, which don't have that #include in their
asm/hardirq.h.
Remove the linux/irq.h #include from x86' asm/hardirq.h.
Fix resulting compilation errors by adding appropriate #includes to *.c
files as needed.
Note that some of these *.c files could be cleaned up a bit wrt. to their
set of #includes, but that should better be done from separate patches, if
at all.
Part of the L1TF mitigation for vmx includes flushing the L1D cache upon
VMENTRY.
L1D flushes are costly and two modes of operations are provided to users:
"always" and the more selective "conditional" mode.
If operating in the latter, the cache would get flushed only if a host side
code path considered unconfined had been traversed. "Unconfined" in this
context means that it might have pulled in sensitive data like user data
or kernel crypto keys.
The need for L1D flushes is tracked by means of the per-vcpu flag
l1tf_flush_l1d. KVM exit handlers considered unconfined set it. A
vmx_l1d_flush() subsequently invoked before the next VMENTER will conduct a
L1d flush based on its value and reset that flag again.
Currently, interrupts delivered "normally" while in root operation between
VMEXIT and VMENTER are not taken into account. Part of the reason is that
these don't leave any traces and thus, the vmx code is unable to tell if
any such has happened.
As proposed by Paolo Bonzini, prepare for tracking all interrupts by
introducing a new per-cpu flag, "kvm_cpu_l1tf_flush_l1d". It will be in
strong analogy to the per-vcpu ->l1tf_flush_l1d.
A later patch will make interrupt handlers set it.
For the sake of cache locality, group kvm_cpu_l1tf_flush_l1d into x86'
per-cpu irq_cpustat_t as suggested by Peter Zijlstra.
Provide the helpers kvm_set_cpu_l1tf_flush_l1d(),
kvm_clear_cpu_l1tf_flush_l1d() and kvm_get_cpu_l1tf_flush_l1d(). Make them
trivial resp. non-existent for !CONFIG_KVM_INTEL as appropriate.
Let vmx_l1d_flush() handle kvm_cpu_l1tf_flush_l1d in the same way as
l1tf_flush_l1d.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com> Suggested-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Nicolai Stange <nstange@suse.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
An upcoming patch will extend KVM's L1TF mitigation in conditional mode
to also cover interrupts after VMEXITs. For tracking those, stores to a
new per-cpu flag from interrupt handlers will become necessary.
In order to improve cache locality, this new flag will be added to x86's
irq_cpustat_t.
Make some space available there by shrinking the ->softirq_pending bitfield
from 32 to 16 bits: the number of bits actually used is only NR_SOFTIRQS,
i.e. 10.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Nicolai Stange <nstange@suse.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The vmx_l1d_flush_always static key is only ever evaluated if
vmx_l1d_should_flush is enabled. In that case however, there are only two
L1d flushing modes possible: "always" and "conditional".
The "conditional" mode's implementation tends to require more sophisticated
logic than the "always" mode.
Avoid inverted logic by replacing the 'vmx_l1d_flush_always' static key
with a 'vmx_l1d_flush_cond' one.
If SMT is disabled in BIOS, the CPU code doesn't properly detect it.
The /sys/devices/system/cpu/smt/control file shows 'on', and the 'l1tf'
vulnerabilities file shows SMT as vulnerable.
Fix it by forcing 'cpu_smt_control' to CPU_SMT_NOT_SUPPORTED in such a
case. Unfortunately the detection can only be done after bringing all
the CPUs online, so we have to overwrite any previous writes to the
variable.
Reported-by: Joe Mario <jmario@redhat.com> Tested-by: Jiri Kosina <jkosina@suse.cz> Fixes: f048c399e0f7 ("x86/topology: Provide topology_smt_supported()") Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The slow path in vmx_l1d_flush() reads from vmx_l1d_flush_pages in order
to evict the L1d cache.
However, these pages are never cleared and, in theory, their data could be
leaked.
More importantly, KSM could merge a nested hypervisor's vmx_l1d_flush_pages
to fewer than 1 << L1D_CACHE_ORDER host physical pages and this would break
the L1d flushing algorithm: L1D on x86_64 is tagged by physical addresses.
Fix this by initializing the individual vmx_l1d_flush_pages with a
different pattern each.
Rename the "empty_zp" asm constraint identifier in vmx_l1d_flush() to
"flush_pages" to reflect this change.
pfn_modify_allowed() and arch_has_pfn_modify_check() are outside of the
!__ASSEMBLY__ section in include/asm-generic/pgtable.h, which confuses
assembler on archs that don't have __HAVE_ARCH_PFN_MODIFY_ALLOWED (e.g.
ia64) and breaks build:
include/asm-generic/pgtable.h: Assembler messages:
include/asm-generic/pgtable.h:538: Error: Unknown opcode `static inline bool pfn_modify_allowed(unsigned long pfn,pgprot_t prot)'
include/asm-generic/pgtable.h:540: Error: Unknown opcode `return true'
include/asm-generic/pgtable.h:543: Error: Unknown opcode `static inline bool arch_has_pfn_modify_check(void)'
include/asm-generic/pgtable.h:545: Error: Unknown opcode `return false'
arch/ia64/kernel/entry.S:69: Error: `mov' does not fit into bundle
Move those two static inlines into the !__ASSEMBLY__ section so that they
don't confuse the asm build pass.
Fixes: 42e4089c7890 ("x86/speculation/l1tf: Disallow non privileged high MMIO PROT_NONE mappings") Signed-off-by: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Introduce the 'l1tf=' kernel command line option to allow for boot-time
switching of mitigation that is used on processors affected by L1TF.
The possible values are:
full
Provides all available mitigations for the L1TF vulnerability. Disables
SMT and enables all mitigations in the hypervisors. SMT control via
/sys/devices/system/cpu/smt/control is still possible after boot.
Hypervisors will issue a warning when the first VM is started in
a potentially insecure configuration, i.e. SMT enabled or L1D flush
disabled.
full,force
Same as 'full', but disables SMT control. Implies the 'nosmt=force'
command line option. sysfs control of SMT and the hypervisor flush
control is disabled.
flush
Leaves SMT enabled and enables the conditional hypervisor mitigation.
Hypervisors will issue a warning when the first VM is started in a
potentially insecure configuration, i.e. SMT enabled or L1D flush
disabled.
flush,nosmt
Disables SMT and enables the conditional hypervisor mitigation. SMT
control via /sys/devices/system/cpu/smt/control is still possible
after boot. If SMT is reenabled or flushing disabled at runtime
hypervisors will issue a warning.
flush,nowarn
Same as 'flush', but hypervisors will not warn when
a VM is started in a potentially insecure configuration.
off
Disables hypervisor mitigations and doesn't emit any warnings.
Default is 'flush'.
Let KVM adhere to these semantics, which means:
- 'lt1f=full,force' : Performe L1D flushes. No runtime control
possible.
- 'l1tf=full'
- 'l1tf-flush'
- 'l1tf=flush,nosmt' : Perform L1D flushes and warn on VM start if
SMT has been runtime enabled or L1D flushing
has been run-time enabled
- 'l1tf=flush,nowarn' : Perform L1D flushes and no warnings are emitted.
- 'l1tf=off' : L1D flushes are not performed and no warnings
are emitted.
KVM can always override the L1D flushing behavior using its 'vmentry_l1d_flush'
module parameter except when lt1f=full,force is set.
This makes KVM's private 'nosmt' option redundant, and as it is a bit
non-systematic anyway (this is something to control globally, not on
hypervisor level), remove that option.
Add the missing Documentation entry for the l1tf vulnerability sysfs file
while at it.
All mitigation modes can be switched at run time with a static key now:
- Use sysfs_streq() instead of strcmp() to handle the trailing new line
from sysfs writes correctly.
- Make the static key management handle multiple invocations properly.
- Set the module parameter file to RW
In preparation of allowing run time control for L1D flushing, move the
setup code to the module parameter handler.
In case of pre module init parsing, just store the value and let vmx_init()
do the actual setup after running kvm_init() so that enable_ept is having
the correct state.
During run-time invoke it directly from the parameter setter to prepare for
run-time control.
If Extended Page Tables (EPT) are disabled or not supported, no L1D
flushing is required. The setup function can just avoid setting up the L1D
flush for the EPT=n case.
Invoke it after the hardware setup has be done and enable_ept has the
correct state and expose the EPT disabled state in the mitigation status as
well.
The VMX module parameter to control the L1D flush should become
writeable.
The MSR list is set up at VM init per guest VCPU, but the run time
switching is based on a static key which is global. Toggling the MSR list
at run time might be feasible, but for now drop this optimization and use
the regular MSR write to make run-time switching possible.
The default mitigation is the conditional flush anyway, so for extra
paranoid setups this will add some small overhead, but the extra code
executed is in the noise compared to the flush itself.
Aside of that the EPT disabled case is not handled correctly at the moment
and the MSR list magic is in the way for fixing that as well.
If it's really providing a significant advantage, then this needs to be
revisited after the code is correct and the control is writable.
Writing 'off' to /sys/devices/system/cpu/smt/control offlines all SMT
siblings. Writing 'on' merily enables the abilify to online them, but does
not online them automatically.
Make 'on' more useful by onlining all offline siblings.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The IA32_FLUSH_CMD MSR needs only to be written on VMENTER. Extend
add_atomic_switch_msr() with an entry_only parameter to allow storing the
MSR only in the guest (ENTRY) MSR array.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There is no semantic change but this change allows an unbalanced amount of
MSRs to be loaded on VMEXIT and VMENTER, i.e. the number of MSRs to save or
restore on VMEXIT or VMENTER may be different.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
336996-Speculative-Execution-Side-Channel-Mitigations.pdf defines a new MSR
(IA32_FLUSH_CMD aka 0x10B) which has similar write-only semantics to other
MSRs defined in the document.
The semantics of this MSR is to allow "finer granularity invalidation of
caching structures than existing mechanisms like WBINVD. It will writeback
and invalidate the L1 data cache, including all cachelines brought in by
preceding instructions, without invalidating all caches (eg. L2 or
LLC). Some processors may also invalidate the first level level instruction
cache on a L1D_FLUSH command. The L1 data and instruction caches may be
shared across the logical processors of a core."
Use it instead of the loop based L1 flush algorithm.
A copy of this document is available at
https://bugzilla.kernel.org/show_bug.cgi?id=199511
[ tglx: Avoid allocating pages when the MSR is available ]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To mitigate the L1 Terminal Fault vulnerability it's required to flush L1D
on VMENTER to prevent rogue guests from snooping host memory.
CPUs will have a new control MSR via a microcode update to flush L1D with a
single MSR write, but in the absence of microcode a fallback to a software
based flush algorithm is required.
Add a software flush loop which is based on code from Intel.
[ tglx: Split out from combo patch ]
[ bpetkov: Polish the asm code ]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add a mitigation mode parameter "vmentry_l1d_flush" for CVE-2018-3620, aka
L1 terminal fault. The valid arguments are:
- "always" L1D cache flush on every VMENTER.
- "cond" Conditional L1D cache flush, explained below
- "never" Disable the L1D cache flush mitigation
"cond" is trying to avoid L1D cache flushes on VMENTER if the code executed
between VMEXIT and VMENTER is considered safe, i.e. is not bringing any
interesting information into L1D which might exploited.
[ tglx: Split out from a larger patch ]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the L1TF CPU bug is present we allow the KVM module to be loaded as the
major of users that use Linux and KVM have trusted guests and do not want a
broken setup.
Cloud vendors are the ones that are uncomfortable with CVE 2018-3620 and as
such they are the ones that should set nosmt to one.
Setting 'nosmt' means that the system administrator also needs to disable
SMT (Hyper-threading) in the BIOS, or via the 'nosmt' command line
parameter, or via the /sys/devices/system/cpu/smt/control. See commit 05736e4ac13c ("cpu/hotplug: Provide knobs to control SMT").
Other mitigations are to use task affinity, cpu sets, interrupt binding,
etc - anything to make sure that _only_ the same guests vCPUs are running
on sibling threads.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Due to the way Machine Check Exceptions work on X86 hyperthreads it's
required to boot up _all_ logical cores at least once in order to set the
CR4.MCE bit.
So instead of ignoring the sibling threads right away, let them boot up
once so they can configure themselves. After they came out of the initial
boot stage check whether its a "secondary" sibling and cancel the operation
which puts the CPU back into offline state.
Reported-by: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Tony Luck <tony.luck@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Dave Hansen reported, that it's outright dangerous to keep SMT siblings
disabled completely so they are stuck in the BIOS and wait for SIPI.
The reason is that Machine Check Exceptions are broadcasted to siblings and
the soft disabled sibling has CR4.MCE = 0. If a MCE is delivered to a
logical core with CR4.MCE = 0, it asserts IERR#, which shuts down or
reboots the machine. The MCE chapter in the SDM contains the following
blurb:
Because the logical processors within a physical package are tightly
coupled with respect to shared hardware resources, both logical
processors are notified of machine check errors that occur within a
given physical processor. If machine-check exceptions are enabled when
a fatal error is reported, all the logical processors within a physical
package are dispatched to the machine-check exception handler. If
machine-check exceptions are disabled, the logical processors enter the
shutdown state and assert the IERR# signal. When enabling machine-check
exceptions, the MCE flag in control register CR4 should be set for each
logical processor.
Reverting the commit which ignores siblings at enumeration time solves only
half of the problem. The core cpuhotplug logic needs to be adjusted as
well.
This thoughtful engineered mechanism also turns the boot process on all
Intel HT enabled systems into a MCE lottery. MCE is enabled on the boot CPU
before the secondary CPUs are brought up. Depending on the number of
physical cores the window in which this situation can happen is smaller or
larger. On a HSW-EX it's about 750ms:
MCE is enabled on the boot CPU:
[ 0.244017] mce: CPU supports 22 MCE banks
The corresponding sibling #72 boots:
[ 1.008005] .... node #0, CPUs: #72
That means if an MCE hits on physical core 0 (logical CPUs 0 and 72)
between these two points the machine is going to shutdown. At least it's a
known safe state.
It's obvious that the early boot can be hit by an MCE as well and then runs
into the same situation because MCEs are not yet enabled on the boot CPU.
But after enabling them on the boot CPU, it does not make any sense to
prevent the kernel from recovering.
Adjust the nosmt kernel parameter documentation as well.
Reverts: 2207def700f9 ("x86/apic: Ignore secondary threads if nosmt=force") Reported-by: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Tony Luck <tony.luck@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Jan has noticed that pte_pfn and co. resp. pfn_pte are incorrect for
CONFIG_PAE because phys_addr_t is wider than unsigned long and so the
pte_val reps. shift left would get truncated. Fix this up by using proper
types.
Fixes: 6b28baca9b1f ("x86/speculation/l1tf: Protect PROT_NONE PTEs against speculation") Reported-by: Jan Beulich <JBeulich@suse.com> Signed-off-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The PAE 3-level paging code currently doesn't mitigate L1TF by flipping the
offset bits, and uses the high PTE word, thus bits 32-36 for type, 37-63 for
offset. The lower word is zeroed, thus systems with less than 4GB memory are
safe. With 4GB to 128GB the swap type selects the memory locations vulnerable
to L1TF; with even more memory, also the swap offfset influences the address.
This might be a problem with 32bit PAE guests running on large 64bit hosts.
By continuing to keep the whole swap entry in either high or low 32bit word of
PTE we would limit the swap size too much. Thus this patch uses the whole PAE
PTE with the same layout as the 64bit version does. The macros just become a
bit tricky since they assume the arch-dependent swp_entry_t to be 32bit.
Signed-off-by: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The TOPOEXT reenablement is a workaround for broken BIOSen which didn't
enable the CPUID bit. amd_get_topology_early(), however, relies on
that bit being set so that it can read out the CPUID leaf and set
smp_num_siblings properly.
Move the reenablement up to early_init_amd(). While at it, simplify
amd_get_topology_early().
nosmt on the kernel command line merely prevents the onlining of the
secondary SMT siblings.
nosmt=force makes the APIC detection code ignore the secondary SMT siblings
completely, so they even do not show up as possible CPUs. That reduces the
amount of memory allocations for per cpu variables and saves other
resources from being allocated too large.
This is not fully equivalent to disabling SMT in the BIOS because the low
level SMT enabling in the BIOS can result in partitioning of resources
between the siblings, which is not undone by just ignoring them. Some CPUs
can use the full resources when their sibling is not onlined, but this is
depending on the CPU family and model and it's not well documented whether
this applies to all partitioned resources. That means depending on the
workload disabling SMT in the BIOS might result in better performance.
Linus analysis of the Intel manual:
The intel optimization manual is not very clear on what the partitioning
rules are.
I find:
"In general, the buffers for staging instructions between major pipe
stages are partitioned. These buffers include µop queues after the
execution trace cache, the queues after the register rename stage, the
reorder buffer which stages instructions for retirement, and the load
and store buffers.
In the case of load and store buffers, partitioning also provided an
easier implementation to maintain memory ordering for each logical
processor and detect memory ordering violations"
but some of that partitioning may be relaxed if the HT thread is "not
active":
"In Intel microarchitecture code name Sandy Bridge, the micro-op queue
is statically partitioned to provide 28 entries for each logical
processor, irrespective of software executing in single thread or
multiple threads. If one logical processor is not active in Intel
microarchitecture code name Ivy Bridge, then a single thread executing
on that processor core can use the 56 entries in the micro-op queue"
but I do not know what "not active" means, and how dynamic it is. Some of
that partitioning may be entirely static and depend on the early BIOS
disabling of HT, and even if we park the cores, the resources will just be
wasted.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To support force disabling of SMT it's required to know the number of
thread siblings early. amd_get_topology() cannot be called before the APIC
driver is selected, so split out the part which initializes
smp_num_siblings and invoke it from amd_early_init().
Old code used to check whether CPUID ext max level is >= 0x80000008 because
that last leaf contains the number of cores of the physical CPU. The three
functions called there now do not depend on that leaf anymore so the check
can go.
Make use of the new early detection function to initialize smp_num_siblings
on the boot cpu before the MP-Table or ACPI/MADT scan happens. That's
required for force disabling SMT.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To support force disabling of SMT it's required to know the number of
thread siblings early. detect_extended_topology() cannot be called before
the APIC driver is selected, so split out the part which initializes
smp_num_siblings.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To support force disabling of SMT it's required to know the number of
thread siblings early. detect_ht() cannot be called before the APIC driver
is selected, so split out the part which initializes smp_num_siblings.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Provide a command line and a sysfs knob to control SMT.
The command line options are:
'nosmt': Enumerate secondary threads, but do not online them
'nosmt=force': Ignore secondary threads completely during enumeration
via MP table and ACPI/MADT.
The sysfs control file has the following states (read/write):
'on': SMT is enabled. Secondary threads can be freely onlined
'off': SMT is disabled. Secondary threads, even if enumerated
cannot be onlined
'forceoff': SMT is permanentely disabled. Writes to the control
file are rejected.
'notsupported': SMT is not supported by the CPU
The command line option 'nosmt' sets the sysfs control to 'off'. This
can be changed to 'on' to reenable SMT during runtime.
The command line option 'nosmt=force' sets the sysfs control to
'forceoff'. This cannot be changed during runtime.
When SMT is 'on' and the control file is changed to 'off' then all online
secondary threads are offlined and attempts to online a secondary thread
later on are rejected.
When SMT is 'off' and the control file is changed to 'on' then secondary
threads can be onlined again. The 'off' -> 'on' transition does not
automatically online the secondary threads.
When the control file is set to 'forceoff', the behaviour is the same as
setting it to 'off', but the operation is irreversible and later writes to
the control file are rejected.
When the control status is 'notsupported' then writes to the control file
are rejected.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The asymmetry caused a warning to trigger if the bootup was stopped in state
CPUHP_AP_ONLINE_IDLE. The warning no longer triggers as kthread_park() can
now be invoked on already or still parked threads. But there is still no
reason to have this be asymmetric.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the CPU is supporting SMT then the primary thread can be found by
checking the lower APIC ID bits for zero. smp_num_siblings is used to build
the mask for the APIC ID bits which need to be taken into account.
This uses the MPTABLE or ACPI/MADT supplied APIC ID, which can be different
than the initial APIC ID in CPUID. But according to AMD the lower bits have
to be consistent. Intel gave a tentative confirmation as well.
Preparatory patch to support disabling SMT at boot/runtime.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The static key sched_smt_present is only updated at boot time when SMT
siblings have been detected. Booting with maxcpus=1 and bringing the
siblings online after boot rebuilds the scheduling domains correctly but
does not update the static key, so the SMT code is not enabled.
Let the key be updated in the scheduler CPU hotplug code to fix this.
Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For the L1TF workaround its necessary to limit the swap file size to below
MAX_PA/2, so that the higher bits of the swap offset inverted never point
to valid memory.
Add a mechanism for the architecture to override the swap file size check
in swapfile.c and add a x86 specific max swapfile check function that
enforces that limit.
The check is only enabled if the CPU is vulnerable to L1TF.
In VMs with 42bit MAX_PA the typical limit is 2TB now, on a native system
with 46bit PA it is 32TB. The limit is only per individual swap file, so
it's always possible to exceed these limits with multiple swap files or
partitions.
Signed-off-by: Andi Kleen <ak@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For L1TF PROT_NONE mappings are protected by inverting the PFN in the page
table entry. This sets the high bits in the CPU's address space, thus
making sure to point to not point an unmapped entry to valid cached memory.
Some server system BIOSes put the MMIO mappings high up in the physical
address space. If such an high mapping was mapped to unprivileged users
they could attack low memory by setting such a mapping to PROT_NONE. This
could happen through a special device driver which is not access
protected. Normal /dev/mem is of course access protected.
To avoid this forbid PROT_NONE mappings or mprotect for high MMIO mappings.
Valid page mappings are allowed because the system is then unsafe anyways.
It's not expected that users commonly use PROT_NONE on MMIO. But to
minimize any impact this is only enforced if the mapping actually refers to
a high MMIO address (defined as the MAX_PA-1 bit being set), and also skip
the check for root.
For mmaps this is straight forward and can be handled in vm_insert_pfn and
in remap_pfn_range().
For mprotect it's a bit trickier. At the point where the actual PTEs are
accessed a lot of state has been changed and it would be difficult to undo
on an error. Since this is a uncommon case use a separate early page talk
walk pass for MMIO PROT_NONE mappings that checks for this condition
early. For non MMIO and non PROT_NONE there are no changes.
Signed-off-by: Andi Kleen <ak@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
L1TF core kernel workarounds are cheap and normally always enabled, However
they still should be reported in sysfs if the system is vulnerable or
mitigated. Add the necessary CPU feature/bug bits.
- Extend the existing checks for Meltdowns to determine if the system is
vulnerable. All CPUs which are not vulnerable to Meltdown are also not
vulnerable to L1TF
- Check for 32bit non PAE and emit a warning as there is no practical way
for mitigation due to the limited physical address bits
- If the system has more than MAX_PA/2 physical memory the invert page
workarounds don't protect the system against the L1TF attack anymore,
because an inverted physical address will also point to valid
memory. Print a warning in this case and report that the system is
vulnerable.
Add a function which returns the PFN limit for the L1TF mitigation, which
will be used in follow up patches for sanity and range checks.
[ tglx: Renamed the CPU feature bit to L1TF_PTEINV ]
Signed-off-by: Andi Kleen <ak@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The L1TF workaround doesn't make any attempt to mitigate speculate accesses
to the first physical page for zeroed PTEs. Normally it only contains some
data from the early real mode BIOS.
It's not entirely clear that the first page is reserved in all
configurations, so add an extra reservation call to make sure it is really
reserved. In most configurations (e.g. with the standard reservations)
it's likely a nop.
Signed-off-by: Andi Kleen <ak@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When PTEs are set to PROT_NONE the kernel just clears the Present bit and
preserves the PFN, which creates attack surface for L1TF speculation
speculation attacks.
This is important inside guests, because L1TF speculation bypasses physical
page remapping. While the host has its own migitations preventing leaking
data from other VMs into the guest, this would still risk leaking the wrong
page inside the current guest.
This uses the same technique as Linus' swap entry patch: while an entry is
is in PROTNONE state invert the complete PFN part part of it. This ensures
that the the highest bit will point to non existing memory.
The invert is done by pte/pmd_modify and pfn/pmd/pud_pte for PROTNONE and
pte/pmd/pud_pfn undo it.
This assume that no code path touches the PFN part of a PTE directly
without using these primitives.
This doesn't handle the case that MMIO is on the top of the CPU physical
memory. If such an MMIO region was exposed by an unpriviledged driver for
mmap it would be possible to attack some real memory. However this
situation is all rather unlikely.
For 32bit non PAE the inversion is not done because there are really not
enough bits to protect anything.
Q: Why does the guest need to be protected when the HyperVisor already has
L1TF mitigations?
A: Here's an example:
Physical pages 1 2 get mapped into a guest as
GPA 1 -> PA 2
GPA 2 -> PA 1
through EPT.
The L1TF speculation ignores the EPT remapping.
Now the guest kernel maps GPA 1 to process A and GPA 2 to process B, and
they belong to different users and should be isolated.
A sets the GPA 1 PA 2 PTE to PROT_NONE to bypass the EPT remapping and
gets read access to the underlying physical page. Which in this case
points to PA 2, so it can read process B's data, if it happened to be in
L1, so isolation inside the guest is broken.
There's nothing the hypervisor can do about this. This mitigation has to
be done in the guest itself.
[ tglx: Massaged changelog ]
Signed-off-by: Andi Kleen <ak@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
With L1 terminal fault the CPU speculates into unmapped PTEs, and resulting
side effects allow to read the memory the PTE is pointing too, if its
values are still in the L1 cache.
For swapped out pages Linux uses unmapped PTEs and stores a swap entry into
them.
To protect against L1TF it must be ensured that the swap entry is not
pointing to valid memory, which requires setting higher bits (between bit
36 and bit 45) that are inside the CPUs physical address space, but outside
any real memory.
To do this invert the offset to make sure the higher bits are always set,
as long as the swap file is not too big.
Note there is no workaround for 32bit !PAE, or on systems which have more
than MAX_PA/2 worth of memory. The later case is very unlikely to happen on
real systems.
[AK: updated description and minor tweaks by. Split out from the original
patch ]
If pages are swapped out, the swap entry is stored in the corresponding
PTE, which has the Present bit cleared. CPUs vulnerable to L1TF speculate
on PTE entries which have the present bit set and would treat the swap
entry as phsyical address (PFN). To mitigate that the upper bits of the PTE
must be set so the PTE points to non existent memory.
The swap entry stores the type and the offset of a swapped out page in the
PTE. type is stored in bit 9-13 and offset in bit 14-63. The hardware
ignores the bits beyond the phsyical address space limit, so to make the
mitigation effective its required to start 'offset' at the lowest possible
bit so that even large swap offsets do not reach into the physical address
space limit bits.
Move offset to bit 9-58 and type to bit 59-63 which are the bits that
hardware generally doesn't care about.
That, in turn, means that if you on desktop chip with only 40 bits of
physical addressing, now that the offset starts at bit 9, there needs to be
30 bits of offset actually *in use* until bit 39 ends up being set, which
means when inverted it will again point into existing memory.
So that's 4 terabyte of swap space (because the offset is counted in pages,
so 30 bits of offset is 42 bits of actual coverage). With bigger physical
addressing, that obviously grows further, until the limit of the offset is
hit (at 50 bits of offset - 62 bits of actual swap file coverage).
This is a preparatory change for the actual swap entry inversion to protect
against L1TF.
[ AK: Updated description and minor tweaks. Split into two parts ]
[ tglx: Massaged changelog ]
L1 Terminal Fault (L1TF) is a speculation related vulnerability. The CPU
speculates on PTE entries which do not have the PRESENT bit set, if the
content of the resulting physical address is available in the L1D cache.
The OS side mitigation makes sure that a !PRESENT PTE entry points to a
physical address outside the actually existing and cachable memory
space. This is achieved by inverting the upper bits of the PTE. Due to the
address space limitations this only works for 64bit and 32bit PAE kernels,
but not for 32bit non PAE.
This mitigation applies to both host and guest kernels, but in case of a
64bit host (hypervisor) and a 32bit PAE guest, inverting the upper bits of
the PAE address space (44bit) is not enough if the host has more than 43
bits of populated memory address space, because the speculation treats the
PTE content as a physical host address bypassing EPT.
The host (hypervisor) protects itself against the guest by flushing L1D as
needed, but pages inside the guest are not protected against attacks from
other processes inside the same guest.
For the guest the inverted PTE mask has to match the host to provide the
full protection for all pages the host could possibly map into the
guest. The hosts populated address space is not known to the guest, so the
mask must cover the possible maximal host address space, i.e. 52 bit.
On 32bit PAE the maximum PTE mask is currently set to 44 bit because that
is the limit imposed by 32bit unsigned long PFNs in the VMs. This limits
the mask to be below what the host could possible use for physical pages.
The L1TF PROT_NONE protection code uses the PTE masks to determine which
bits to invert to make sure the higher bits are set for unmapped entries to
prevent L1TF speculation attacks against EPT inside guests.
In order to invert all bits that could be used by the host, increase
__PHYSICAL_PAGE_SHIFT to 52 to match 64bit.
The real limit for a 32bit PAE kernel is still 44 bits because all Linux
PTEs are created from unsigned long PFNs, so they cannot be higher than 44
bits on a 32bit kernel. So these extra PFN bits should be never set. The
only users of this macro are using it to look at PTEs, so it's safe.
[ tglx: Massaged changelog ]
Signed-off-by: Andi Kleen <ak@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The article "Spectre Returns! Speculation Attacks using the Return Stack
Buffer" [1] describes two new (sub-)variants of spectrev2-like attacks,
making use solely of the RSB contents even on CPUs that don't fallback to
BTB on RSB underflow (Skylake+).
Mitigate userspace-userspace attacks by always unconditionally filling RSB on
context switch when the generic spectrev2 mitigation has been enabled.
[1] https://arxiv.org/pdf/1807.07940.pdf
Signed-off-by: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: Borislav Petkov <bp@suse.de> Cc: David Woodhouse <dwmw@amazon.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/nycvar.YFH.7.76.1807261308190.997@cbobk.fhfr.pm Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Nadav reported that on guests we're failing to rewrite the indirect
calls to CALLEE_SAVE paravirt functions. In particular the
pv_queued_spin_unlock() call is left unpatched and that is all over the
place. This obviously wrecks Spectre-v2 mitigation (for paravirt
guests) which relies on not actually having indirect calls around.
The reason is an incorrect clobber test in paravirt_patch_call(); this
function rewrites an indirect call with a direct call to the _SAME_
function, there is no possible way the clobbers can be different
because of this.
Therefore remove this clobber check. Also put WARNs on the other patch
failure case (not enough room for the instruction) which I've not seen
trigger in my (limited) testing.
Three live kernel image disassemblies for lock_sock_nested (as a small
function that illustrates the problem nicely). PRE is the current
situation for guests, POST is with this patch applied and NATIVE is with
or without the patch for !guests.
Use the correct IRQ line for the MSI controller in the PCIe host
controller. Apparently a different IRQ line is used compared to other
i.MX6 variants. Without this change MSI IRQs aren't properly propagated
to the upstream interrupt controller.
__legitimize_mnt() has two problems - one is that in case of success
the check of mount_lock is not ordered wrt preceding increment of
refcount, making it possible to have successful __legitimize_mnt()
on one CPU just before the otherwise final mntpu() on another,
with __legitimize_mnt() not seeing mntput() taking the lock and
mntput() not seeing the increment done by __legitimize_mnt().
Solved by a pair of barriers.
Another is that failure of __legitimize_mnt() on the second
read_seqretry() leaves us with reference that'll need to be
dropped by caller; however, if that races with final mntput()
we can end up with caller dropping rcu_read_lock() and doing
mntput() to release that reference - with the first mntput()
having freed the damn thing just as rcu_read_lock() had been
dropped. Solution: in "do mntput() yourself" failure case
grab mount_lock, check if MNT_DOOMED has been set by racing
final mntput() that has missed our increment and if it has -
undo the increment and treat that as "failure, caller doesn't
need to drop anything" case.
It's not easy to hit - the final mntput() has to come right
after the first read_seqretry() in __legitimize_mnt() *and*
manage to miss the increment done by __legitimize_mnt() before
the second read_seqretry() in there. The things that are almost
impossible to hit on bare hardware are not impossible on SMP
KVM, though...
mntput_no_expire() does the calculation of total refcount under mount_lock;
unfortunately, the decrement (as well as all increments) are done outside
of it, leading to false positives in the "are we dropping the last reference"
test. Consider the following situation:
* mnt is a lazy-umounted mount, kept alive by two opened files. One
of those files gets closed. Total refcount of mnt is 2. On CPU 42
mntput(mnt) (called from __fput()) drops one reference, decrementing component
* After it has looked at component #0, the process on CPU 0 does
mntget(), incrementing component #0, gets preempted and gets to run again -
on CPU 69. There it does mntput(), which drops the reference (component #69)
and proceeds to spin on mount_lock.
* On CPU 42 our first mntput() finishes counting. It observes the
decrement of component #69, but not the increment of component #0. As the
result, the total it gets is not 1 as it should've been - it's 0. At which
point we decide that vfsmount needs to be killed and proceed to free it and
shut the filesystem down. However, there's still another opened file
on that filesystem, with reference to (now freed) vfsmount, etc. and we are
screwed.
It's not a wide race, but it can be reproduced with artificial slowdown of
the mnt_get_count() loop, and it should be easier to hit on SMP KVM setups.
Fix consists of moving the refcount decrement under mount_lock; the tricky
part is that we want (and can) keep the fast case (i.e. mount that still
has non-NULL ->mnt_ns) entirely out of mount_lock. All places that zero
mnt->mnt_ns are dropping some reference to mnt and they call synchronize_rcu()
before that mntput(). IOW, if mntput() observes (under rcu_read_lock())
a non-NULL ->mnt_ns, it is guaranteed that there is another reference yet to
be dropped.
RCU pathwalk relies upon the assumption that anything that changes
->d_inode of a dentry will invalidate its ->d_seq. That's almost
true - the one exception is that the final dput() of already unhashed
dentry does *not* touch ->d_seq at all. Unhashing does, though,
so for anything we'd found by RCU dcache lookup we are fine.
Unfortunately, we can *start* with an unhashed dentry or jump into
it.
We could try and be careful in the (few) places where that could
happen. Or we could just make the final dput() invalidate the damn
thing, unhashed or not. The latter is much simpler and easier to
backport, so let's do it that way.
Reported-by: "Dae R. Jeong" <threeearcat@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since mountpoint crossing can happen without leaving lazy mode,
root dentries do need the same protection against having their
memory freed without RCU delay as everything else in the tree.
It's partially hidden by RCU delay between detaching from the
mount tree and dropping the vfsmount reference, but the starting
point of pathwalk can be on an already detached mount, in which
case umount-caused RCU delay has already passed by the time the
lazy pathwalk grabs rcu_read_lock(). If the starting point
happens to be at the root of that vfsmount *and* that vfsmount
covers the entire filesystem, we get trouble.
This is purely a preparatory patch for upcoming changes during the 4.19
merge window.
We have a function called "boot_cpu_state_init()" that isn't really
about the bootup cpu state: that is done much earlier by the similarly
named "boot_cpu_init()" (note lack of "state" in name).
This function initializes some hotplug CPU state, and needs to run after
the percpu data has been properly initialized. It even has a comment to
that effect.
Except it _doesn't_ actually run after the percpu data has been properly
initialized. On x86 it happens to do that, but on at least arm and
arm64, the percpu base pointers are initialized by the arch-specific
'smp_prepare_boot_cpu()' hook, which ran _after_ boot_cpu_state_init().
This had some unexpected results, and in particular we have a patch
pending for the merge window that did the obvious cleanup of using
'this_cpu_write()' in the cpu hotplug init code:
which is obviously the right thing to do. Except because of the
ordering issue, it actually failed miserably and unexpectedly on arm64.
So this just fixes the ordering, and changes the name of the function to
be 'boot_cpu_hotplug_init()' to make it obvious that it's about cpu
hotplug state, because the core CPU state was supposed to have already
been done earlier.
Marked for stable, since the (not yet merged) patch that will show this
problem is marked for stable.
Surround scsi_execute() calls with scsi_autopm_get_device() and
scsi_autopm_put_device(). Note: removing sr_mutex protection from the
scsi_cd_get() and scsi_cd_put() calls is safe because the purpose of
sr_mutex is to serialize cdrom_*() calls.
This patch avoids that complaints similar to the following appear in the
kernel log if runtime power management is enabled:
INFO: task systemd-udevd:650 blocked for more than 120 seconds.
Not tainted 4.18.0-rc7-dbg+ #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
systemd-udevd D28176 650 513 0x00000104
Call Trace:
__schedule+0x444/0xfe0
schedule+0x4e/0xe0
schedule_preempt_disabled+0x18/0x30
__mutex_lock+0x41c/0xc70
mutex_lock_nested+0x1b/0x20
__blkdev_get+0x106/0x970
blkdev_get+0x22c/0x5a0
blkdev_open+0xe9/0x100
do_dentry_open.isra.19+0x33e/0x570
vfs_open+0x7c/0xd0
path_openat+0x6e3/0x1120
do_filp_open+0x11c/0x1c0
do_sys_open+0x208/0x2d0
__x64_sys_openat+0x59/0x70
do_syscall_64+0x77/0x230
entry_SYSCALL_64_after_hwframe+0x49/0xbe
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Cc: Maurizio Lombardi <mlombard@redhat.com> Cc: Johannes Thumshirn <jthumshirn@suse.de> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: <stable@vger.kernel.org> Tested-by: Johannes Thumshirn <jthumshirn@suse.de> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
While working on bpf_tcp_sendmsg() code, I noticed that when a
sk->sk_err is set we error out with err = sk->sk_err. However
this is problematic since sk->sk_err is a positive error value
and therefore we will neither go into sk_stream_error() nor will
we report an error back to user space. I had this case with EPIPE
and user space was thinking sendmsg() succeeded since EPIPE is
a positive value, thinking we submitted 32 bytes. Fix it by
negating the sk->sk_err value.
Fixes: 4f738adba30a ("bpf: create tcp_bpf_ulp allowing BPF to monitor socket TX/RX data") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In bpf_tcp_sendmsg() the sk_alloc_sg() may fail. In the case of
ENOMEM, it may also mean that we've partially filled the scatterlist
entries with pages. Later jumping to sk_stream_wait_memory()
we could further fail with an error for several reasons, however
we miss to call free_start_sg() if the local sk_msg_buff was used.
Fixes: 4f738adba30a ("bpf: create tcp_bpf_ulp allowing BPF to monitor socket TX/RX data") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
skb_shinfo() can change when calling __pskb_pull_tail(): Don't cache
its return value.
Cc: stable@vger.kernel.org Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Wei Liu <wei.liu2@citrix.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If zram supports writeback feature, it's no longer a
BD_CAP_SYNCHRONOUS_IO device beause zram does asynchronous IO operations
for incompressible pages.
Do not pretend to be synchronous IO device. It makes the system very
sluggish due to waiting for IO completion from upper layers.
Furthermore, it causes a user-after-free problem because swap thinks the
opearion is done when the IO functions returns so it can free the page
(e.g., lock_page_or_retry and goto out_release in do_swap_page) but in
fact, IO is asynchronous so the driver could access a just freed page
afterward.
This patch fixes the problem.
BUG: Bad page state in process qemu-system-x86 pfn:3dfab21
page:ffffdfb137eac840 count:0 mapcount:0 mapping:0000000000000000 index:0x1
flags: 0x17fffc000000008(uptodate)
raw: 017fffc000000008dead000000000100dead0000000002000000000000000000
raw: 0000000000000001000000000000000000000000ffffffff0000000000000000
page dumped because: PAGE_FLAGS_CHECK_AT_PREP flag set
bad because of flags: 0x8(uptodate)
CPU: 4 PID: 1039 Comm: qemu-system-x86 Tainted: G B 4.18.0-rc5+ #1
Hardware name: Supermicro Super Server/X10SRL-F, BIOS 2.0b 05/02/2017
Call Trace:
dump_stack+0x5c/0x7b
bad_page+0xba/0x120
get_page_from_freelist+0x1016/0x1250
__alloc_pages_nodemask+0xfa/0x250
alloc_pages_vma+0x7c/0x1c0
do_swap_page+0x347/0x920
__handle_mm_fault+0x7b4/0x1110
handle_mm_fault+0xfc/0x1f0
__get_user_pages+0x12f/0x690
get_user_pages_unlocked+0x148/0x1f0
__gfn_to_pfn_memslot+0xff/0x3c0 [kvm]
try_async_pf+0x87/0x230 [kvm]
tdp_page_fault+0x132/0x290 [kvm]
kvm_mmu_page_fault+0x74/0x570 [kvm]
kvm_arch_vcpu_ioctl_run+0x9b3/0x1990 [kvm]
kvm_vcpu_ioctl+0x388/0x5d0 [kvm]
do_vfs_ioctl+0xa2/0x630
ksys_ioctl+0x70/0x80
__x64_sys_ioctl+0x16/0x20
do_syscall_64+0x55/0x100
entry_SYSCALL_64_after_hwframe+0x44/0xa9
This warning happened in the push_dl_task(), because
__add_running_bw()->cpufreq_update_util() is getting the rq_clock of
the later_rq before its update, which takes place at activate_task().
The fix then is to update the rq_clock before calling add_running_bw().
To avoid double rq_clock_update() call, we set ENQUEUE_NOCLOCK flag to
activate_task().
Reported-by: Daniel Casini <daniel.casini@santannapisa.it> Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Juri Lelli <juri.lelli@redhat.com> Cc: Clark Williams <williams@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Luca Abeni <luca.abeni@santannapisa.it> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Tommaso Cucinotta <tommaso.cucinotta@santannapisa.it> Fixes: e0367b12674b sched/deadline: Move CPU frequency selection triggering points Link: http://lkml.kernel.org/r/ca31d073a4788acf0684a8b255f14fea775ccf20.1532077269.git.bristot@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9fb8d5dc4b64 ("stop_machine, Disable preemption when waking two stopper threads")
does not fully address the race condition that can occur
as follows:
On one CPU, call it CPU 3, thread 1 invokes
cpu_stop_queue_two_works(2, 3,...), and the execution is such
that thread 1 queues the works for migration/2 and migration/3,
and is preempted after releasing the locks for migration/2 and
migration/3, but before waking the threads.
Then, On CPU 2, a kworker, call it thread 2, is running,
and it invokes cpu_stop_queue_two_works(1, 2,...), such that
thread 2 queues the works for migration/1 and migration/2.
Meanwhile, on CPU 3, thread 1 resumes execution, and wakes
migration/2 and migration/3. This means that when CPU 2
releases the locks for migration/1 and migration/2, but before
it wakes those threads, it can be preempted by migration/2.
If thread 2 is preempted by migration/2, then migration/2 will
execute the first work item successfully, since migration/3
was woken up by CPU 3, but when it goes to execute the second
work item, it disables preemption, calls multi_cpu_stop(),
and thus, CPU 2 will wait forever for migration/1, which should
have been woken up by thread 2. However migration/1 cannot be
woken up by thread 2, since it is a kworker, so it is affine to
CPU 2, but CPU 2 is running migration/2 with preemption
disabled, so thread 2 will never run.
Disable preemption after queueing works for stopper threads
to ensure that the operation of queueing the works and waking
the stopper threads is atomic.
Co-Developed-by: Prasad Sodagudi <psodagud@codeaurora.org> Co-Developed-by: Pavankumar Kondeti <pkondeti@codeaurora.org> Signed-off-by: Isaac J. Manjarres <isaacm@codeaurora.org> Signed-off-by: Prasad Sodagudi <psodagud@codeaurora.org> Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bigeasy@linutronix.de Cc: gregkh@linuxfoundation.org Cc: matt@codeblueprint.co.uk Fixes: 9fb8d5dc4b64 ("stop_machine, Disable preemption when waking two stopper threads") Link: http://lkml.kernel.org/r/1531856129-9871-1-git-send-email-isaacm@codeaurora.org Signed-off-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Way back in 4.9, we committed 4cd13c21b207 ("softirq: Let ksoftirqd do
its job"), and ever since we've had small nagging issues with it. For
example, we've had:
1ff688209e2e ("watchdog: core: make sure the watchdog_worker is not deferred") 8d5755b3f77b ("watchdog: softdog: fire watchdog even if softirqs do not get to run") 217f69743681 ("net: busy-poll: allow preemption in sk_busy_loop()")
all of which worked around some of the effects of that commit.
The DVB people have also complained that the commit causes excessive USB
URB latencies, which seems to be due to the USB code using tasklets to
schedule USB traffic. This seems to be an issue mainly when already
living on the edge, but waiting for ksoftirqd to handle it really does
seem to cause excessive latencies.
Now Hanna Hawa reports that this issue isn't just limited to USB URB and
DVB, but also causes timeout problems for the Marvell SoC team:
"I'm facing kernel panic issue while running raid 5 on sata disks
connected to Macchiatobin (Marvell community board with Armada-8040
SoC with 4 ARMv8 cores of CA72) Raid 5 built with Marvell DMA engine
and async_tx mechanism (ASYNC_TX_DMA [=y]); the DMA driver (mv_xor_v2)
uses a tasklet to clean the done descriptors from the queue"
The latency problem causes a panic:
mv_xor_v2 f0400000.xor: dma_sync_wait: timeout!
Kernel panic - not syncing: async_tx_quiesce: DMA error waiting for transaction
We've discussed simply just reverting the original commit entirely, and
also much more involved solutions (with per-softirq threads etc). This
patch is intentionally stupid and fairly limited, because the issue
still remains, and the other solutions either got sidetracked or had
other issues.
We should probably also consider the timer softirqs to be synchronous
and not be delayed to ksoftirqd (since they were the issue with the
earlier watchdog problems), but that should be done as a separate patch.
This does only the tasklet cases.
Reported-and-tested-by: Hanna Hawa <hannah@marvell.com> Reported-and-tested-by: Josef Griebichler <griebichler.josef@gmx.at> Reported-by: Mauro Carvalho Chehab <mchehab@s-opensource.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Eric Dumazet <edumazet@google.com> Cc: Ingo Molnar <mingo@kernel.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>