When eagerly switching PKRU in switch_fpu_finish() it checks that
current is not a kernel thread as kernel threads will never use PKRU.
It's possible that this_cpu_read_stable() on current_task
(ie. get_current()) is returning an old cached value. To resolve this
reference next_p directly rather than relying on current.
As written it's possible when switching from a kernel thread to a
userspace thread to observe a cached PF_KTHREAD flag and never restore
the PKRU. And as a result this issue only occurs when switching
from a kernel thread to a userspace thread, switching from a non kernel
thread works perfectly fine because all that is considered in that
situation are the flags from some other non kernel task and the next fpu
is passed in to switch_fpu_finish().
This behavior only exists between 5.2 and 5.13 when it was fixed by a
rewrite decoupling PKRU from xstate, in:
commit 954436989cc5 ("x86/fpu: Remove PKRU handling from switch_fpu_finish()")
Unfortunately backporting the fix from 5.13 is probably not realistic as
it's part of a 60+ patch series which rewrites most of the PKRU handling.
Fixes: 0cecca9d03 ("x86/fpu: Eager switch PKRU state")
Signed-off-by: Brian Geffon <bgeffon@google.com>
Signed-off-by: Willis Kung <williskung@google.com>
Tested-by: Willis Kung <williskung@google.com>
Cc: <stable@vger.kernel.org> # v5.4.x
Cc: <stable@vger.kernel.org> # v5.10.x
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit aee8c67a4faa40a8df4e79316dbfc92d123989c1 ]
When *RSTOR from user memory raises an exception, there is no way to
differentiate them. That's bad because it forces the slow path even when
the failure was not a fault. If the operation raised eg. #GP then going
through the slow path is pointless.
Use _ASM_EXTABLE_FAULT() which stores the trap number and let the exception
fixup return the negated trap number as error.
This allows to separate the fast path and let it handle faults directly and
avoid the slow path for all other exceptions.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20210623121457.601480369@linutronix.de
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit f9dfb5e390fab2df9f7944bb91e7705aba14cd26 upstream.
The XSAVE init code initializes all enabled and supported components with
XRSTOR(S) to init state. Then it XSAVEs the state of the components back
into init_fpstate which is used in several places to fill in the init state
of components.
This works correctly with XSAVE, but not with XSAVEOPT and XSAVES because
those use the init optimization and skip writing state of components which
are in init state. So init_fpstate.xsave still contains all zeroes after
this operation.
There are two ways to solve that:
1) Use XSAVE unconditionally, but that requires to reshuffle the buffer when
XSAVES is enabled because XSAVES uses compacted format.
2) Save the components which are known to have a non-zero init state by other
means.
Looking deeper, #2 is the right thing to do because all components the
kernel supports have all-zeroes init state except the legacy features (FP,
SSE). Those cannot be hard coded because the states are not identical on all
CPUs, but they can be saved with FXSAVE which avoids all conditionals.
Use FXSAVE to save the legacy FP/SSE components in init_fpstate along with
a BUILD_BUG_ON() which reminds developers to validate that a newly added
component has all zeroes init state. As a bonus remove the now unused
copy_xregs_to_kernel_booting() crutch.
The XSAVE and reshuffle method can still be implemented in the unlikely
case that components are added which have a non-zero init state and no
other means to save them. For now, FXSAVE is just simple and good enough.
[ bp: Fix a typo or two in the text. ]
Fixes: 6bad06b768 ("x86, xsave: Use xsaveopt in context-switch path when supported")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Borislav Petkov <bp@suse.de>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20210618143444.587311343@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 510b80a6a0f1a0d114c6e33bcea64747d127973c upstream.
When user space brings PKRU into init state, then the kernel handling is
broken:
T1 user space
xsave(state)
state.header.xfeatures &= ~XFEATURE_MASK_PKRU;
xrstor(state)
T1 -> kernel
schedule()
XSAVE(S) -> T1->xsave.header.xfeatures[PKRU] == 0
T1->flags |= TIF_NEED_FPU_LOAD;
wrpkru();
schedule()
...
pk = get_xsave_addr(&T1->fpu->state.xsave, XFEATURE_PKRU);
if (pk)
wrpkru(pk->pkru);
else
wrpkru(DEFAULT_PKRU);
Because the xfeatures bit is 0 and therefore the value in the xsave
storage is not valid, get_xsave_addr() returns NULL and switch_to()
writes the default PKRU. -> FAIL #1!
So that wrecks any copy_to/from_user() on the way back to user space
which hits memory which is protected by the default PKRU value.
Assumed that this does not fail (pure luck) then T1 goes back to user
space and because TIF_NEED_FPU_LOAD is set it ends up in
switch_fpu_return()
__fpregs_load_activate()
if (!fpregs_state_valid()) {
load_XSTATE_from_task();
}
But if nothing touched the FPU between T1 scheduling out and back in,
then the fpregs_state is still valid which means switch_fpu_return()
does nothing and just clears TIF_NEED_FPU_LOAD. Back to user space with
DEFAULT_PKRU loaded. -> FAIL #2!
The fix is simple: if get_xsave_addr() returns NULL then set the
PKRU value to 0 instead of the restrictive default PKRU value in
init_pkru_value.
[ bp: Massage in minor nitpicks from folks. ]
Fixes: 0cecca9d03 ("x86/fpu: Eager switch PKRU state")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Rik van Riel <riel@surriel.com>
Tested-by: Babu Moger <babu.moger@amd.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20210608144346.045616965@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 9bfecd05833918526cc7357d55e393393440c5fa upstream.
While digesting the XSAVE-related horrors which got introduced with
the supervisor/user split, the recent addition of ENQCMD-related
functionality got on the radar and turned out to be similarly broken.
update_pasid(), which is only required when X86_FEATURE_ENQCMD is
available, is invoked from two places:
1) From switch_to() for the incoming task
2) Via a SMP function call from the IOMMU/SMV code
#1 is half-ways correct as it hacks around the brokenness of get_xsave_addr()
by enforcing the state to be 'present', but all the conditionals in that
code are completely pointless for that.
Also the invocation is just useless overhead because at that point
it's guaranteed that TIF_NEED_FPU_LOAD is set on the incoming task
and all of this can be handled at return to user space.
#2 is broken beyond repair. The comment in the code claims that it is safe
to invoke this in an IPI, but that's just wishful thinking.
FPU state of a running task is protected by fregs_lock() which is
nothing else than a local_bh_disable(). As BH-disabled regions run
usually with interrupts enabled the IPI can hit a code section which
modifies FPU state and there is absolutely no guarantee that any of the
assumptions which are made for the IPI case is true.
Also the IPI is sent to all CPUs in mm_cpumask(mm), but the IPI is
invoked with a NULL pointer argument, so it can hit a completely
unrelated task and unconditionally force an update for nothing.
Worse, it can hit a kernel thread which operates on a user space
address space and set a random PASID for it.
The offending commit does not cleanly revert, but it's sufficient to
force disable X86_FEATURE_ENQCMD and to remove the broken update_pasid()
code to make this dysfunctional all over the place. Anything more
complex would require more surgery and none of the related functions
outside of the x86 core code are blatantly wrong, so removing those
would be overkill.
As nothing enables the PASID bit in the IA32_XSS MSR yet, which is
required to make this actually work, this cannot result in a regression
except for related out of tree train-wrecks, but they are broken already
today.
Fixes: 20f0afd1fb ("x86/mmu: Allocate/free a PASID")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/87mtsd6gr9.ffs@nanos.tec.linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit e45122893a9870813f9bd7b4add4f613e6f29008 upstream.
Currently, requesting kernel FPU access doesn't distinguish which parts of
the extended ("FPU") state are needed. This is nice for simplicity, but
there are a few cases in which it's suboptimal:
- The vast majority of in-kernel FPU users want XMM/YMM/ZMM state but do
not use legacy 387 state. These users want MXCSR initialized but don't
care about the FPU control word. Skipping FNINIT would save time.
(Empirically, FNINIT is several times slower than LDMXCSR.)
- Code that wants MMX doesn't want or need MXCSR initialized.
_mmx_memcpy(), for example, can run before CR4.OSFXSR gets set, and
initializing MXCSR will fail because LDMXCSR generates an #UD when the
aforementioned CR4 bit is not set.
- Any future in-kernel users of XFD (eXtended Feature Disable)-capable
dynamic states will need special handling.
Add a more specific API that allows callers to specify exactly what they
want.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Krzysztof Piotr Olędzki <ole@ans.pl>
Link: https://lkml.kernel.org/r/aff1cac8b8fc7ee900cf73e8f2369966621b053f.1611205691.git.luto@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Pull x86 SEV-ES support from Borislav Petkov:
"SEV-ES enhances the current guest memory encryption support called SEV
by also encrypting the guest register state, making the registers
inaccessible to the hypervisor by en-/decrypting them on world
switches. Thus, it adds additional protection to Linux guests against
exfiltration, control flow and rollback attacks.
With SEV-ES, the guest is in full control of what registers the
hypervisor can access. This is provided by a guest-host exchange
mechanism based on a new exception vector called VMM Communication
Exception (#VC), a new instruction called VMGEXIT and a shared
Guest-Host Communication Block which is a decrypted page shared
between the guest and the hypervisor.
Intercepts to the hypervisor become #VC exceptions in an SEV-ES guest
so in order for that exception mechanism to work, the early x86 init
code needed to be made able to handle exceptions, which, in itself,
brings a bunch of very nice cleanups and improvements to the early
boot code like an early page fault handler, allowing for on-demand
building of the identity mapping. With that, !KASLR configurations do
not use the EFI page table anymore but switch to a kernel-controlled
one.
The main part of this series adds the support for that new exchange
mechanism. The goal has been to keep this as much as possibly separate
from the core x86 code by concentrating the machinery in two
SEV-ES-specific files:
arch/x86/kernel/sev-es-shared.c
arch/x86/kernel/sev-es.c
Other interaction with core x86 code has been kept at minimum and
behind static keys to minimize the performance impact on !SEV-ES
setups.
Work by Joerg Roedel and Thomas Lendacky and others"
* tag 'x86_seves_for_v5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (73 commits)
x86/sev-es: Use GHCB accessor for setting the MMIO scratch buffer
x86/sev-es: Check required CPU features for SEV-ES
x86/efi: Add GHCB mappings when SEV-ES is active
x86/sev-es: Handle NMI State
x86/sev-es: Support CPU offline/online
x86/head/64: Don't call verify_cpu() on starting APs
x86/smpboot: Load TSS and getcpu GDT entry before loading IDT
x86/realmode: Setup AP jump table
x86/realmode: Add SEV-ES specific trampoline entry point
x86/vmware: Add VMware-specific handling for VMMCALL under SEV-ES
x86/kvm: Add KVM-specific VMMCALL handling under SEV-ES
x86/paravirt: Allow hypervisor-specific VMMCALL handling under SEV-ES
x86/sev-es: Handle #DB Events
x86/sev-es: Handle #AC Events
x86/sev-es: Handle VMMCALL Events
x86/sev-es: Handle MWAIT/MWAITX Events
x86/sev-es: Handle MONITOR/MONITORX Events
x86/sev-es: Handle INVD Events
x86/sev-es: Handle RDPMC Events
x86/sev-es: Handle RDTSC(P) Events
...
Pull x86 PASID updates from Borislav Petkov:
"Initial support for sharing virtual addresses between the CPU and
devices which doesn't need pinning of pages for DMA anymore.
Add support for the command submission to devices using new x86
instructions like ENQCMD{,S} and MOVDIR64B. In addition, add support
for process address space identifiers (PASIDs) which are referenced by
those command submission instructions along with the handling of the
PASID state on context switch as another extended state.
Work by Fenghua Yu, Ashok Raj, Yu-cheng Yu and Dave Jiang"
* tag 'x86_pasid_for_5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/asm: Add an enqcmds() wrapper for the ENQCMDS instruction
x86/asm: Carve out a generic movdir64b() helper for general usage
x86/mmu: Allocate/free a PASID
x86/cpufeatures: Mark ENQCMD as disabled when configured out
mm: Add a pasid member to struct mm_struct
x86/msr-index: Define an IA32_PASID MSR
x86/fpu/xstate: Add supervisor PASID state for ENQCMD
x86/cpufeatures: Enumerate ENQCMD and ENQCMDS instructions
Documentation/x86: Add documentation for SVA (Shared Virtual Addressing)
iommu/vt-d: Change flags type to unsigned int in binding mm
drm, iommu: Change type of pasid to u32
A PASID is allocated for an "mm" the first time any thread binds to an
SVA-capable device and is freed from the "mm" when the SVA is unbound
by the last thread. It's possible for the "mm" to have different PASID
values in different binding/unbinding SVA cycles.
The mm's PASID (non-zero for valid PASID or 0 for invalid PASID) is
propagated to a per-thread PASID MSR for all threads within the mm
through IPI, context switch, or inherited. This is done to ensure that a
running thread has the right PASID in the MSR matching the mm's PASID.
[ bp: s/SVM/SVA/g; massage. ]
Suggested-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Link: https://lkml.kernel.org/r/1600187413-163670-10-git-send-email-fenghua.yu@intel.com
The xgetbv() function is needed in the pre-decompression boot code,
but asm/fpu/internal.h can't be included there directly. Doing so
opens the door to include-hell due to various include-magic in
boot/compressed/misc.h.
Avoid that by moving xgetbv()/xsetbv() to a separate header file and
include it instead.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200907131613.12703-27-joro@8bytes.org
Pull ptrace regset updates from Al Viro:
"Internal regset API changes:
- regularize copy_regset_{to,from}_user() callers
- switch to saner calling conventions for ->get()
- kill user_regset_copyout()
The ->put() side of things will have to wait for the next cycle,
unfortunately.
The balance is about -1KLoC and replacements for ->get() instances are
a lot saner"
* 'work.regset' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (41 commits)
regset: kill user_regset_copyout{,_zero}()
regset(): kill ->get_size()
regset: kill ->get()
csky: switch to ->regset_get()
xtensa: switch to ->regset_get()
parisc: switch to ->regset_get()
nds32: switch to ->regset_get()
nios2: switch to ->regset_get()
hexagon: switch to ->regset_get()
h8300: switch to ->regset_get()
openrisc: switch to ->regset_get()
riscv: switch to ->regset_get()
c6x: switch to ->regset_get()
ia64: switch to ->regset_get()
arc: switch to ->regset_get()
arm: switch to ->regset_get()
sh: convert to ->regset_get()
arm64: switch to ->regset_get()
mips: switch to ->regset_get()
sparc: switch to ->regset_get()
...
All instances of ->get() in arch/x86 switched; that might or might
not be worth splitting up. Notes:
* for xstateregs_get() the amount we want to store is determined at
the boot time; see init_xstate_size() and update_regset_xstate_info() for
details. task->thread.fpu.state.xsave ends with a flexible array member and
the amount of data in it depends upon the FPU features supported/enabled.
* fpregs_get() writes slightly less than full ->thread.fpu.state.fsave
(the last word is not copied); we pass the full size of state.fsave and let
membuf_write() trim to the amount declared by regset - __regset_get() will
make sure that the space in buffer is no more than that.
* copy_xstate_to_user() and its helpers are gone now.
* fpregs_soft_get() was getting user_regset_copyout() arguments
wrong. Since "x86: x86 user_regset math_emu" back in 2008... I really
doubt that it's worth splitting out for -stable, though - you need
a 486SX box for that to trigger...
[Kevin's braino fix for copy_xstate_to_kernel() essentially duplicated here]
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
In the LBR call stack mode, LBR information is used to reconstruct a
call stack. To get the complete call stack, perf has to save/restore
all LBR registers during a context switch. Due to a large number of the
LBR registers, this process causes a high CPU overhead. To reduce the
CPU overhead during a context switch, use the XSAVES/XRSTORS
instructions.
Every XSAVE area must follow a canonical format: the legacy region, an
XSAVE header and the extended region. Although the LBR information is
only kept in the extended region, a space for the legacy region and
XSAVE header is still required. Add a new dedicated structure for LBR
XSAVES support.
Before enabling XSAVES support, the size of the LBR state has to be
sanity checked, because:
- the size of the software structure is calculated from the max number
of the LBR depth, which is enumerated by the CPUID leaf for Arch LBR.
The size of the LBR state is enumerated by the CPUID leaf for XSAVE
support of Arch LBR. If the values from the two CPUID leaves are not
consistent, it may trigger a buffer overflow. For example, a hypervisor
may unconsciously set inconsistent values for the two emulated CPUID.
- unlike other state components, the size of an LBR state depends on the
max number of LBRs, which may vary from generation to generation.
Expose the function xfeature_size() for the sanity check.
The LBR XSAVES support will be disabled if the size of the LBR state
enumerated by CPUID doesn't match with the size of the software
structure.
The XSAVE instruction requires 64-byte alignment for state buffers. A
new macro is added to reflect the alignment requirement. A 64-byte
aligned kmem_cache is created for architecture LBR.
Currently, the structure for each state component is maintained in
fpu/types.h. The structure for the new LBR state component should be
maintained in the same place. Move structure lbr_entry to fpu/types.h as
well for broader sharing.
Add dedicated lbr_save/lbr_restore functions for LBR XSAVES support,
which invokes the corresponding xstate helpers to XSAVES/XRSTORS LBR
information at the context switch when the call stack mode is enabled.
Since the XSAVES/XRSTORS instructions will be eventually invoked, the
dedicated functions is named with '_xsaves'/'_xrstors' postfix.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Link: https://lkml.kernel.org/r/1593780569-62993-23-git-send-email-kan.liang@linux.intel.com
The perf subsystem will only need to save/restore the LBR state.
However, the existing helpers save all supported supervisor states to a
kernel buffer, which will be unnecessary. Two helpers are introduced to
only save/restore requested dynamic supervisor states. The supervisor
features in XFEATURE_MASK_SUPERVISOR_SUPPORTED and
XFEATURE_MASK_SUPERVISOR_UNSUPPORTED mask cannot be saved/restored using
these helpers.
The helpers will be used in the following patch.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Link: https://lkml.kernel.org/r/1593780569-62993-22-git-send-email-kan.liang@linux.intel.com
Last Branch Records (LBR) registers are used to log taken branches and
other control flows. In perf with call stack mode, LBR information is
used to reconstruct a call stack. To get the complete call stack, perf
has to save/restore all LBR registers during a context switch. Due to
the large number of the LBR registers, e.g., the current platform has
96 LBR registers, this process causes a high CPU overhead. To reduce
the CPU overhead during a context switch, an LBR state component that
contains all the LBR related registers is introduced in hardware. All
LBR registers can be saved/restored together using one XSAVES/XRSTORS
instruction.
However, the kernel should not save/restore the LBR state component at
each context switch, like other state components, because of the
following unique features of LBR:
- The LBR state component only contains valuable information when LBR
is enabled in the perf subsystem, but for most of the time, LBR is
disabled.
- The size of the LBR state component is huge. For the current
platform, it's 808 bytes.
If the kernel saves/restores the LBR state at each context switch, for
most of the time, it is just a waste of space and cycles.
To efficiently support the LBR state component, it is desired to have:
- only context-switch the LBR when the LBR feature is enabled in perf.
- only allocate an LBR-specific XSAVE buffer on demand.
(Besides the LBR state, a legacy region and an XSAVE header have to be
included in the buffer as well. There is a total of (808+576) byte
overhead for the LBR-specific XSAVE buffer. The overhead only happens
when the perf is actively using LBRs. There is still a space-saving,
on average, when it replaces the constant 808 bytes of overhead for
every task, all the time on the systems that support architectural
LBR.)
- be able to use XSAVES/XRSTORS for accessing LBR at run time.
However, the IA32_XSS should not be adjusted at run time.
(The XCR0 | IA32_XSS are used to determine the requested-feature
bitmap (RFBM) of XSAVES.)
A solution, called dynamic supervisor feature, is introduced to address
this issue, which
- does not allocate a buffer in each task->fpu;
- does not save/restore a state component at each context switch;
- sets the bit corresponding to the dynamic supervisor feature in
IA32_XSS at boot time, and avoids setting it at run time.
- dynamically allocates a specific buffer for a state component
on demand, e.g. only allocates LBR-specific XSAVE buffer when LBR is
enabled in perf. (Note: The buffer has to include the LBR state
component, a legacy region and a XSAVE header space.)
(Implemented in a later patch)
- saves/restores a state component on demand, e.g. manually invokes
the XSAVES/XRSTORS instruction to save/restore the LBR state
to/from the buffer when perf is active and a call stack is required.
(Implemented in a later patch)
A new mask XFEATURE_MASK_DYNAMIC and a helper xfeatures_mask_dynamic()
are introduced to indicate the dynamic supervisor feature. For the
systems which support the Architecture LBR, LBR is the only dynamic
supervisor feature for now. For the previous systems, there is no
dynamic supervisor feature available.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Link: https://lkml.kernel.org/r/1593780569-62993-21-git-send-email-kan.liang@linux.intel.com
When saving xstate to a kernel/user XSAVE area with the XSAVE family of
instructions, the current code applies the 'full' instruction mask (-1),
which tries to XSAVE all possible features. This method relies on
hardware to trim 'all possible' down to what is enabled in the
hardware. The code works well for now. However, there will be a
problem, if some features are enabled in hardware, but are not suitable
to be saved into all kernel XSAVE buffers, like task->fpu, due to
performance consideration.
One such example is the Last Branch Records (LBR) state. The LBR state
only contains valuable information when LBR is explicitly enabled by
the perf subsystem, and the size of an LBR state is large (808 bytes
for now). To avoid both CPU overhead and space overhead at each context
switch, the LBR state should not be saved into task->fpu like other
state components. It should be saved/restored on demand when LBR is
enabled in the perf subsystem. Current copy_xregs_to_* will trigger a
buffer overflow for such cases.
Three sites use the '-1' instruction mask which must be updated.
Two are saving/restoring the xstate to/from a kernel-allocated XSAVE
buffer and can use 'xfeatures_mask_all', which will save/restore all of
the features present in a normal task FPU buffer.
The last one saves the register state directly to a user buffer. It
could
also use 'xfeatures_mask_all'. Just as it was with the '-1' argument,
any supervisor states in the mask will be filtered out by the hardware
and not saved to the buffer. But, to be more explicit about what is
expected to be saved, use xfeatures_mask_user() for the instruction
mask.
KVM includes the header file fpu/internal.h. To avoid 'undefined
xfeatures_mask_all' compiling issue, move copy_fpregs_to_fpstate() to
fpu/core.c and export it, because:
- The xfeatures_mask_all is indirectly used via copy_fpregs_to_fpstate()
by KVM. The function which is directly used by other modules should be
exported.
- The copy_fpregs_to_fpstate() is a function, while xfeatures_mask_all
is a variable for the "internal" FPU state. It's safer to export a
function than a variable, which may be implicitly changed by others.
- The copy_fpregs_to_fpstate() is a big function with many checks. The
removal of the inline keyword should not impact the performance.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Link: https://lkml.kernel.org/r/1593780569-62993-20-git-send-email-kan.liang@linux.intel.com
Previously, kernel floating point code would run with the MXCSR control
register value last set by userland code by the thread that was active
on the CPU core just before kernel call. This could affect calculation
results if rounding mode was changed, or a crash if a FPU/SIMD exception
was unmasked.
Restore MXCSR to the kernel's default value.
[ bp: Carve out from a bigger patch by Petteri, add feature check, add
FNINIT call too (amluto). ]
Signed-off-by: Petteri Aimonen <jpa@git.mail.kapsi.fi>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=207979
Link: https://lkml.kernel.org/r/20200624114646.28953-2-bp@alien8.de
The XSAVES instruction takes a mask and saves only the features specified
in that mask. The kernel normally specifies that all features be saved.
XSAVES also unconditionally uses the "compacted format" which means that
all specified features are saved next to each other in memory. If a
feature is removed from the mask, all the features after it will "move
up" into earlier locations in the buffer.
Introduce copy_supervisor_to_kernel(), which saves only supervisor states
and then moves those states into the standard location where they are
normally found.
Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200512145444.15483-9-yu-cheng.yu@intel.com
The function copy_kernel_to_xregs_err() uses XRSTOR which can work with
standard or compacted format without supervisor xstates. However, when
supervisor xstates are present, XRSTORS must be used. Fix it by using
XRSTORS when supervisor state handling is enabled.
I also considered if there were additional cases where XRSTOR might be
mistakenly called instead of XRSTORS. There are only three XRSTOR sites
in the kernel:
1. copy_kernel_to_xregs_booting(), already switches between XRSTOR and
XRSTORS based on X86_FEATURE_XSAVES.
2. copy_user_to_xregs(), which *needs* XRSTOR because it is copying from
userspace and must never copy supervisor state with XRSTORS.
3. copy_kernel_to_xregs_err() mistakenly used XRSTOR only. Fix it.
[ bp: Massage commit message. ]
Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Link: https://lkml.kernel.org/r/20200512145444.15483-8-yu-cheng.yu@intel.com
Currently, fpu__clear() clears all fpregs and xstates. Once XSAVES
supervisor states are introduced, supervisor settings (e.g. CET xstates)
must remain active for signals; It is necessary to have separate functions:
- Create fpu__clear_user_states(): clear only user settings for signals;
- Create fpu__clear_all(): clear both user and supervisor settings in
flush_thread().
Also modify copy_init_fpstate_to_fpregs() to take a mask from above two
functions.
Remove obvious side-comment in fpu__clear(), while at it.
[ bp: Make the second argument of fpu__clear() bool after requesting it
a bunch of times during review.
- Add a comment about copy_init_fpstate_to_fpregs() locking needs. ]
Co-developed-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Link: https://lkml.kernel.org/r/20200512145444.15483-6-yu-cheng.yu@intel.com
Before the introduction of XSAVES supervisor states, 'xfeatures_mask' is
used at various places to determine XSAVE buffer components and XCR0 bits.
It contains only user xstates. To support supervisor xstates, it is
necessary to separate user and supervisor xstates:
- First, change 'xfeatures_mask' to 'xfeatures_mask_all', which represents
the full set of bits that should ever be set in a kernel XSAVE buffer.
- Introduce xfeatures_mask_supervisor() and xfeatures_mask_user() to
extract relevant xfeatures from xfeatures_mask_all.
Co-developed-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Link: https://lkml.kernel.org/r/20200512145444.15483-4-yu-cheng.yu@intel.com
XCNTXT_MASK is 'all supported xfeatures' before introducing supervisor
xstates. Rename it to XFEATURE_MASK_USER_SUPPORTED to make clear that
these are user xstates.
Replace XFEATURE_MASK_SUPERVISOR with the following:
- XFEATURE_MASK_SUPERVISOR_SUPPORTED: Currently nothing. ENQCMD and
Control-flow Enforcement Technology (CET) will be introduced in separate
series.
- XFEATURE_MASK_SUPERVISOR_UNSUPPORTED: Currently only Processor Trace.
- XFEATURE_MASK_SUPERVISOR_ALL: the combination of above.
Co-developed-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Link: https://lkml.kernel.org/r/20200512145444.15483-3-yu-cheng.yu@intel.com
The state/owner of the FPU is saved to fpu_fpregs_owner_ctx by pointing
to the context that is currently loaded. It never changed during the
lifetime of a task - it remained stable/constant.
After deferred FPU registers loading until return to userland was
implemented, the content of fpu_fpregs_owner_ctx may change during
preemption and must not be cached.
This went unnoticed for some time and was now noticed, in particular
since gcc 9 is caching that load in copy_fpstate_to_sigframe() and
reusing it in the retry loop:
copy_fpstate_to_sigframe()
load fpu_fpregs_owner_ctx and save on stack
fpregs_lock()
copy_fpregs_to_sigframe() /* failed */
fpregs_unlock()
*** PREEMPTION, another uses FPU, changes fpu_fpregs_owner_ctx ***
fault_in_pages_writeable() /* succeed, retry */
fpregs_lock()
__fpregs_load_activate()
fpregs_state_valid() /* uses fpu_fpregs_owner_ctx from stack */
copy_fpregs_to_sigframe() /* succeeds, random FPU content */
This is a comparison of the assembly produced by gcc 9, without vs with this
patch:
| # arch/x86/kernel/fpu/signal.c:173: if (!access_ok(buf, size))
| cmpq %rdx, %rax # tmp183, _4
| jb .L190 #,
|-# arch/x86/include/asm/fpu/internal.h:512: return fpu == this_cpu_read_stable(fpu_fpregs_owner_ctx) && cpu == fpu->last_cpu;
|-#APP
|-# 512 "arch/x86/include/asm/fpu/internal.h" 1
|- movq %gs:fpu_fpregs_owner_ctx,%rax #, pfo_ret__
|-# 0 "" 2
|-#NO_APP
|- movq %rax, -88(%rbp) # pfo_ret__, %sfp
…
|-# arch/x86/include/asm/fpu/internal.h:512: return fpu == this_cpu_read_stable(fpu_fpregs_owner_ctx) && cpu == fpu->last_cpu;
|- movq -88(%rbp), %rcx # %sfp, pfo_ret__
|- cmpq %rcx, -64(%rbp) # pfo_ret__, %sfp
|+# arch/x86/include/asm/fpu/internal.h:512: return fpu == this_cpu_read(fpu_fpregs_owner_ctx) && cpu == fpu->last_cpu;
|+#APP
|+# 512 "arch/x86/include/asm/fpu/internal.h" 1
|+ movq %gs:fpu_fpregs_owner_ctx(%rip),%rax # fpu_fpregs_owner_ctx, pfo_ret__
|+# 0 "" 2
|+# arch/x86/include/asm/fpu/internal.h:512: return fpu == this_cpu_read(fpu_fpregs_owner_ctx) && cpu == fpu->last_cpu;
|+#NO_APP
|+ cmpq %rax, -64(%rbp) # pfo_ret__, %sfp
Use this_cpu_read() instead this_cpu_read_stable() to avoid caching of
fpu_fpregs_owner_ctx during preemption points.
The Fixes: tag points to the commit where deferred FPU loading was
added. Since this commit, the compiler is no longer allowed to move the
load of fpu_fpregs_owner_ctx somewhere else / outside of the locked
section. A task preemption will change its value and stale content will
be observed.
[ bp: Massage. ]
Debugged-by: Austin Clements <austin@google.com>
Debugged-by: David Chase <drchase@golang.org>
Debugged-by: Ian Lance Taylor <ian@airs.com>
Fixes: 5f409e20b7 ("x86/fpu: Defer FPU state load until return to userspace")
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Rik van Riel <riel@surriel.com>
Tested-by: Borislav Petkov <bp@suse.de>
Cc: Aubrey Li <aubrey.li@intel.com>
Cc: Austin Clements <austin@google.com>
Cc: Barret Rhoden <brho@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Chase <drchase@golang.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: ian@airs.com
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Josh Bleecher Snyder <josharian@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20191128085306.hxfa2o3knqtu4wfn@linutronix.de
Link: https://bugzilla.kernel.org/show_bug.cgi?id=205663
Pull x86 FPU state handling updates from Borislav Petkov:
"This contains work started by Rik van Riel and brought to fruition by
Sebastian Andrzej Siewior with the main goal to optimize when to load
FPU registers: only when returning to userspace and not on every
context switch (while the task remains in the kernel).
In addition, this optimization makes kernel_fpu_begin() cheaper by
requiring registers saving only on the first invocation and skipping
that in following ones.
What is more, this series cleans up and streamlines many aspects of
the already complex FPU code, hopefully making it more palatable for
future improvements and simplifications.
Finally, there's a __user annotations fix from Jann Horn"
* 'x86-fpu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (29 commits)
x86/fpu: Fault-in user stack if copy_fpstate_to_sigframe() fails
x86/pkeys: Add PKRU value to init_fpstate
x86/fpu: Restore regs in copy_fpstate_to_sigframe() in order to use the fastpath
x86/fpu: Add a fastpath to copy_fpstate_to_sigframe()
x86/fpu: Add a fastpath to __fpu__restore_sig()
x86/fpu: Defer FPU state load until return to userspace
x86/fpu: Merge the two code paths in __fpu__restore_sig()
x86/fpu: Restore from kernel memory on the 64-bit path too
x86/fpu: Inline copy_user_to_fpregs_zeroing()
x86/fpu: Update xstate's PKRU value on write_pkru()
x86/fpu: Prepare copy_fpstate_to_sigframe() for TIF_NEED_FPU_LOAD
x86/fpu: Always store the registers in copy_fpstate_to_sigframe()
x86/entry: Add TIF_NEED_FPU_LOAD
x86/fpu: Eager switch PKRU state
x86/pkeys: Don't check if PKRU is zero before writing it
x86/fpu: Only write PKRU if it is different from current
x86/pkeys: Provide *pkru() helpers
x86/fpu: Use a feature number instead of mask in two more helpers
x86/fpu: Make __raw_xsave_addr() use a feature number instead of mask
x86/fpu: Add an __fpregs_load_activate() internal helper
...
Defer loading of FPU state until return to userspace. This gives
the kernel the potential to skip loading FPU state for tasks that
stay in kernel mode, or for tasks that end up with repeated
invocations of kernel_fpu_begin() & kernel_fpu_end().
The fpregs_lock/unlock() section ensures that the registers remain
unchanged. Otherwise a context switch or a bottom half could save the
registers to its FPU context and the processor's FPU registers would
became random if modified at the same time.
KVM swaps the host/guest registers on entry/exit path. This flow has
been kept as is. First it ensures that the registers are loaded and then
saves the current (host) state before it loads the guest's registers. The
swap is done at the very end with disabled interrupts so it should not
change anymore before theg guest is entered. The read/save version seems
to be cheaper compared to memcpy() in a micro benchmark.
Each thread gets TIF_NEED_FPU_LOAD set as part of fork() / fpu__copy().
For kernel threads, this flag gets never cleared which avoids saving /
restoring the FPU state for kernel threads and during in-kernel usage of
the FPU registers.
[
bp: Correct and update commit message and fix checkpatch warnings.
s/register/registers/ where it is used in plural.
minor comment corrections.
remove unused trace_x86_fpu_activate_state() TP.
]
Signed-off-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aubrey Li <aubrey.li@intel.com>
Cc: Babu Moger <Babu.Moger@amd.com>
Cc: "Chang S. Bae" <chang.seok.bae@intel.com>
Cc: Dmitry Safonov <dima@arista.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: kvm ML <kvm@vger.kernel.org>
Cc: Nicolai Stange <nstange@suse.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Waiman Long <longman@redhat.com>
Cc: x86-ml <x86@kernel.org>
Cc: Yi Wang <wang.yi59@zte.com.cn>
Link: https://lkml.kernel.org/r/20190403164156.19645-24-bigeasy@linutronix.de
The 64-bit case (both 64-bit and 32-bit frames) loads the new state from
user memory.
However, doing this is not desired if the FPU state is going to be
restored on return to userland: it would be required to disable
preemption in order to avoid a context switch which would set
TIF_NEED_FPU_LOAD. If this happens before the restore operation then the
loaded registers would become volatile.
Furthermore, disabling preemption while accessing user memory requires
to disable the pagefault handler. An error during FXRSTOR would then
mean that either a page fault occurred (and it would have to be retried
with enabled page fault handler) or a #GP occurred because the xstate is
bogus (after all, the signal handler can modify it).
In order to avoid that mess, copy the FPU state from userland, validate
it and then load it. The copy_kernel_…() helpers are basically just
like the old helpers except that they operate on kernel memory and the
fault handler just sets the error value and the caller handles it.
copy_user_to_fpregs_zeroing() and its helpers remain and will be used
later for a fastpath optimisation.
[ bp: Clarify commit message. ]
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aubrey Li <aubrey.li@intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: kvm ML <kvm@vger.kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190403164156.19645-22-bigeasy@linutronix.de
While most of a task's FPU state is only needed in user space, the
protection keys need to be in place immediately after a context switch.
The reason is that any access to userspace memory while running in
kernel mode also needs to abide by the memory permissions specified in
the protection keys.
The "eager switch" is a preparation for loading the FPU state on return
to userland. Instead of decoupling PKRU state from xstate, update PKRU
within xstate on write operations by the kernel.
For user tasks the PKRU should be always read from the xsave area and it
should not change anything because the PKRU value was loaded as part of
FPU restore.
For kernel threads the default "init_pkru_value" will be written. Before
this commit, the kernel thread would end up with a random value which it
inherited from the previous user task.
[ bigeasy: save pkru to xstate, no cache, don't use __raw_xsave_addr() ]
[ bp: update commit message, sort headers properly in asm/fpu/xstate.h ]
Signed-off-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aubrey Li <aubrey.li@intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Juergen Gross <jgross@suse.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: kvm ML <kvm@vger.kernel.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190403164156.19645-16-bigeasy@linutronix.de
user_fpu_begin() sets fpu_fpregs_owner_ctx to task's fpu struct. This is
always the case since there is no lazy FPU anymore.
fpu_fpregs_owner_ctx is used during context switch to decide if it needs
to load the saved registers or if the currently loaded registers are
valid. It could be skipped during a
taskA -> kernel thread -> taskA
switch because the switch to the kernel thread would not alter the CPU's
sFPU tate.
Since this field is always updated during context switch and
never invalidated, setting it manually (in user context) makes no
difference. A kernel thread with kernel_fpu_begin() block could
set fpu_fpregs_owner_ctx to NULL but a kernel thread does not use
user_fpu_begin().
This is a leftover from the lazy-FPU time.
Remove user_fpu_begin(), it does not change fpu_fpregs_owner_ctx's
content.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aubrey Li <aubrey.li@intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: kvm ML <kvm@vger.kernel.org>
Cc: Nicolai Stange <nstange@suse.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190403164156.19645-9-bigeasy@linutronix.de
The struct fpu.initialized member is always set to one for user tasks
and zero for kernel tasks. This avoids saving/restoring the FPU
registers for kernel threads.
The ->initialized = 0 case for user tasks has been removed in previous
changes, for instance, by doing an explicit unconditional init at fork()
time for FPU-less systems which was otherwise delayed until the emulated
opcode.
The context switch code (switch_fpu_prepare() + switch_fpu_finish())
can't unconditionally save/restore registers for kernel threads. Not
only would it slow down the switch but also load a zeroed xcomp_bv for
XSAVES.
For kernel_fpu_begin() (+end) the situation is similar: EFI with runtime
services uses this before alternatives_patched is true. Which means that
this function is used too early and it wasn't the case before.
For those two cases, use current->mm to distinguish between user and
kernel thread. For kernel_fpu_begin() skip save/restore of the FPU
registers.
During the context switch into a kernel thread don't do anything. There
is no reason to save the FPU state of a kernel thread.
The reordering in __switch_to() is important because the current()
pointer needs to be valid before switch_fpu_finish() is invoked so ->mm
is seen of the new task instead the old one.
N.B.: fpu__save() doesn't need to check ->mm because it is called by
user tasks only.
[ bp: Massage. ]
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aubrey Li <aubrey.li@intel.com>
Cc: Babu Moger <Babu.Moger@amd.com>
Cc: "Chang S. Bae" <chang.seok.bae@intel.com>
Cc: Dmitry Safonov <dima@arista.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: kvm ML <kvm@vger.kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Nicolai Stange <nstange@suse.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190403164156.19645-8-bigeasy@linutronix.de
This is a preparation for the removal of the ->initialized member in the
fpu struct.
__fpu__restore_sig() is deactivating the FPU via fpu__drop() and then
setting manually ->initialized followed by fpu__restore(). The result is
that it is possible to manipulate fpu->state and the state of registers
won't be saved/restored on a context switch which would overwrite
fpu->state:
fpu__drop(fpu):
...
fpu->initialized = 0;
preempt_enable();
<--- context switch
Don't access the fpu->state while the content is read from user space
and examined/sanitized. Use a temporary kmalloc() buffer for the
preparation of the FPU registers and once the state is considered okay,
load it. Should something go wrong, return with an error and without
altering the original FPU registers.
The removal of fpu__initialize() is a nop because fpu->initialized is
already set for the user task.
[ bp: Massage a bit. ]
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Borislav Petkov <bp@suse.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: kvm ML <kvm@vger.kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190403164156.19645-2-bigeasy@linutronix.de
User space tools which do automated task placement need information
about AVX-512 usage of tasks, because AVX-512 usage could cause core
turbo frequency drop and impact the running task on the sibling CPU.
The XSAVE hardware structure has bits that indicate when valid state
is present in registers unique to AVX-512 use. Use these bits to
indicate when AVX-512 has been in use and add per-task AVX-512 state
timestamp tracking to context switch.
Well-written AVX-512 applications are expected to clear the AVX-512
state when not actively using AVX-512 registers, so the tracking
mechanism is imprecise and can theoretically miss AVX-512 usage during
context switch. But it has been measured to be precise enough to be
useful under real-world workloads like tensorflow and linpack.
If higher precision is required, suggest user space tools to use the
PMU-based mechanisms in combination.
Signed-off-by: Aubrey Li <aubrey.li@linux.intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: aubrey.li@intel.com
Link: http://lkml.kernel.org/r/20190117183822.31333-1-aubrey.li@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
