commit b94b737331 upstream.
Instead of blacklisting all model 79 CPUs when attempting a late
microcode loading, limit that only to CPUs with microcode revisions <
0x0b000021 because only on those late loading may cause a system hang.
For such processors either:
a) a BIOS update which might contain a newer microcode revision
or
b) the early microcode loading method
should be considered.
Processors with revisions 0x0b000021 or higher will not experience such
hangs.
For more details, see erratum BDF90 in document #334165 (Intel Xeon
Processor E7-8800/4800 v4 Product Family Specification Update) from
September 2017.
[ bp: Heavily massage commit message and pr_* statements. ]
Fixes: 723f2828a9 ("x86/microcode/intel: Disable late loading on model 79")
Signed-off-by: Jia Zhang <qianyue.zj@alibaba-inc.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Tony Luck <tony.luck@intel.com>
Cc: x86-ml <x86@kernel.org>
Link: http://lkml.kernel.org/r/1514772287-92959-1-git-send-email-qianyue.zj@alibaba-inc.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 0cb5b30698 upstream.
Guest GPR values are live in the hardware GPRs at VM-exit. Do not
leave any guest values in hardware GPRs after the guest GPR values are
saved to the vcpu_vmx structure.
This is a partial mitigation for CVE 2017-5715 and CVE 2017-5753.
Specifically, it defeats the Project Zero PoC for CVE 2017-5715.
Suggested-by: Eric Northup <digitaleric@google.com>
Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Eric Northup <digitaleric@google.com>
Reviewed-by: Benjamin Serebrin <serebrin@google.com>
Reviewed-by: Andrew Honig <ahonig@google.com>
[Paolo: Add AMD bits, Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit be07a6a118 upstream.
Fix a commit 72b22bbad1 ("MIPS: Don't assume 64-bit FP registers for
FP regset") public API regression, then activated by commit 1db1af84d6
("MIPS: Basic MSA context switching support"), that caused the FCSR
register not to be read or written for CONFIG_CPU_HAS_MSA kernel
configurations (regardless of actual presence or absence of the MSA
feature in a given processor) with ptrace(2) PTRACE_GETREGSET and
PTRACE_SETREGSET requests nor recorded in core dumps.
This is because with !CONFIG_CPU_HAS_MSA configurations the whole of
`elf_fpregset_t' array is bulk-copied as it is, which includes the FCSR
in one half of the last, 33rd slot, whereas with CONFIG_CPU_HAS_MSA
configurations array elements are copied individually, and then only the
leading 32 FGR slots while the remaining slot is ignored.
Correct the code then such that only FGR slots are copied in the
respective !MSA and MSA helpers an then the FCSR slot is handled
separately in common code. Use `ptrace_setfcr31' to update the FCSR
too, so that the read-only mask is respected.
Retrieving a correct value of FCSR is important in debugging not only
for the human to be able to get the right interpretation of the
situation, but for correct operation of GDB as well. This is because
the condition code bits in FSCR are used by GDB to determine the
location to place a breakpoint at when single-stepping through an FPU
branch instruction. If such a breakpoint is placed incorrectly (i.e.
with the condition reversed), then it will be missed, likely causing the
debuggee to run away from the control of GDB and consequently breaking
the process of investigation.
Fortunately GDB continues using the older PTRACE_GETFPREGS ptrace(2)
request which is unaffected, so the regression only really hits with
post-mortem debug sessions using a core dump file, in which case
execution, and consequently single-stepping through branches is not
possible. Of course core files created by buggy kernels out there will
have the value of FCSR recorded clobbered, but such core files cannot be
corrected and the person using them simply will have to be aware that
the value of FCSR retrieved is not reliable.
Which also means we can likely get away without defining a replacement
API which would ensure a correct value of FSCR to be retrieved, or none
at all.
This is based on previous work by Alex Smith, extensively rewritten.
Signed-off-by: Alex Smith <alex@alex-smith.me.uk>
Signed-off-by: James Hogan <james.hogan@mips.com>
Signed-off-by: Maciej W. Rozycki <macro@mips.com>
Fixes: 72b22bbad1 ("MIPS: Don't assume 64-bit FP registers for FP regset")
Cc: Paul Burton <Paul.Burton@mips.com>
Cc: Dave Martin <Dave.Martin@arm.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/17928/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 80b3ffce01 upstream.
Update commit d614fd58a2 ("mips/ptrace: Preserve previous registers
for short regset write") bug and consistently consume all data supplied
to `fpr_set_msa' with the ptrace(2) PTRACE_SETREGSET request, such that
a zero data buffer counter is returned where insufficient data has been
given to fill a whole number of FP general registers.
In reality this is not going to happen, as the caller is supposed to
only supply data covering a whole number of registers and it is verified
in `ptrace_regset' and again asserted in `fpr_set', however structuring
code such that the presence of trailing partial FP general register data
causes `fpr_set_msa' to return with a non-zero data buffer counter makes
it appear that this trailing data will be used if there are subsequent
writes made to FP registers, which is going to be the case with the FCSR
once the missing write to that register has been fixed.
Fixes: d614fd58a2 ("mips/ptrace: Preserve previous registers for short regset write")
Signed-off-by: Maciej W. Rozycki <macro@mips.com>
Cc: James Hogan <james.hogan@mips.com>
Cc: Paul Burton <Paul.Burton@mips.com>
Cc: Alex Smith <alex@alex-smith.me.uk>
Cc: Dave Martin <Dave.Martin@arm.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/17927/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit a03fe72572 upstream.
In preparation to fix a commit 72b22bbad1 ("MIPS: Don't assume 64-bit
FP registers for FP regset") FCSR access regression factor out
NT_PRFPREG regset access helpers for the non-MSA and the MSA variants
respectively, to avoid having to deal with excessive indentation in the
actual fix.
No functional change, however use `target->thread.fpu.fpr[0]' rather
than `target->thread.fpu.fpr[i]' for FGR holding type size determination
as there's no `i' variable to refer to anymore, and for the factored out
`i' variable declaration use `unsigned int' rather than `unsigned' as
its type, following the common style.
Signed-off-by: Maciej W. Rozycki <macro@mips.com>
Fixes: 72b22bbad1 ("MIPS: Don't assume 64-bit FP registers for FP regset")
Cc: James Hogan <james.hogan@mips.com>
Cc: Paul Burton <Paul.Burton@mips.com>
Cc: Alex Smith <alex@alex-smith.me.uk>
Cc: Dave Martin <Dave.Martin@arm.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/17925/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit b67336eee3 upstream.
Fix an API loophole introduced with commit 9791554b45 ("MIPS,prctl:
add PR_[GS]ET_FP_MODE prctl options for MIPS"), where the caller of
prctl(2) is incorrectly allowed to make a change to CP0.Status.FR or
CP0.Config5.FRE register bits even if CONFIG_MIPS_O32_FP64_SUPPORT has
not been enabled, despite that an executable requesting the mode
requested via ELF file annotation would not be allowed to run in the
first place, or for n64 and n64 ABI tasks which do not have non-default
modes defined at all. Add suitable checks to `mips_set_process_fp_mode'
and bail out if an invalid mode change has been requested for the ABI in
effect, even if the FPU hardware or emulation would otherwise allow it.
Always succeed however without taking any further action if the mode
requested is the same as one already in effect, regardless of whether
any mode change, should it be requested, would actually be allowed for
the task concerned.
Signed-off-by: Maciej W. Rozycki <macro@mips.com>
Fixes: 9791554b45 ("MIPS,prctl: add PR_[GS]ET_FP_MODE prctl options for MIPS")
Reviewed-by: Paul Burton <paul.burton@mips.com>
Cc: James Hogan <james.hogan@mips.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/17800/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit e39d200fa5 upstream.
Reported by syzkaller:
BUG: KASAN: stack-out-of-bounds in write_mmio+0x11e/0x270 [kvm]
Read of size 8 at addr ffff8803259df7f8 by task syz-executor/32298
CPU: 6 PID: 32298 Comm: syz-executor Tainted: G OE 4.15.0-rc2+ #18
Hardware name: LENOVO ThinkCentre M8500t-N000/SHARKBAY, BIOS FBKTC1AUS 02/16/2016
Call Trace:
dump_stack+0xab/0xe1
print_address_description+0x6b/0x290
kasan_report+0x28a/0x370
write_mmio+0x11e/0x270 [kvm]
emulator_read_write_onepage+0x311/0x600 [kvm]
emulator_read_write+0xef/0x240 [kvm]
emulator_fix_hypercall+0x105/0x150 [kvm]
em_hypercall+0x2b/0x80 [kvm]
x86_emulate_insn+0x2b1/0x1640 [kvm]
x86_emulate_instruction+0x39a/0xb90 [kvm]
handle_exception+0x1b4/0x4d0 [kvm_intel]
vcpu_enter_guest+0x15a0/0x2640 [kvm]
kvm_arch_vcpu_ioctl_run+0x549/0x7d0 [kvm]
kvm_vcpu_ioctl+0x479/0x880 [kvm]
do_vfs_ioctl+0x142/0x9a0
SyS_ioctl+0x74/0x80
entry_SYSCALL_64_fastpath+0x23/0x9a
The path of patched vmmcall will patch 3 bytes opcode 0F 01 C1(vmcall)
to the guest memory, however, write_mmio tracepoint always prints 8 bytes
through *(u64 *)val since kvm splits the mmio access into 8 bytes. This
leaks 5 bytes from the kernel stack (CVE-2017-17741). This patch fixes
it by just accessing the bytes which we operate on.
Before patch:
syz-executor-5567 [007] .... 51370.561696: kvm_mmio: mmio write len 3 gpa 0x10 val 0x1ffff10077c1010f
After patch:
syz-executor-13416 [002] .... 51302.299573: kvm_mmio: mmio write len 3 gpa 0x10 val 0xc1010f
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Marc Zyngier <marc.zyngier@arm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This needs to happen early in kaiser_pagetable_walk(), before the
hierarchy is established so that _PAGE_USER permission can be really
set.
A proper fix would be to teach kaiser_pagetable_walk() to update those
permissions but the vsyscall page is the only exception here so ...
Signed-off-by: Borislav Petkov <bp@suse.de>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 1e5476815f upstream.
The recent changes for PTI touch cpu_tlbstate from various tlb_flush
inlines. cpu_tlbstate is exported as GPL symbol, so this causes a
regression when building out of tree drivers for certain graphics cards.
Aside of that the export was wrong since it was introduced as it should
have been EXPORT_PER_CPU_SYMBOL_GPL().
Use the correct PER_CPU export and drop the _GPL to restore the previous
state which allows users to utilize the cards they payed for.
As always I'm really thrilled to make this kind of change to support the
#friends (or however the hot hashtag of today is spelled) from that closet
sauce graphics corp.
Fixes: 1e02ce4ccc ("x86: Store a per-cpu shadow copy of CR4")
Fixes: 6fd166aae7 ("x86/mm: Use/Fix PCID to optimize user/kernel switches")
Reported-by: Kees Cook <keescook@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Thomas Backlund <tmb@mageia.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 310d82784f upstream.
Add qemu idle sleep support when running under qemu with SeaBIOS PDC
firmware.
Like the power architecture we use the "or" assembler instructions,
which translate to nops on real hardware, to indicate that qemu shall
idle sleep.
Signed-off-by: Helge Deller <deller@gmx.de>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 88776c0e70 upstream.
Qemu for PARISC reported on a 32bit SMP parisc kernel strange failures
about "Not-handled unaligned insn 0x0e8011d6 and 0x0c2011c9."
Those opcodes evaluate to the ldcw() assembly instruction which requires
(on 32bit) an alignment of 16 bytes to ensure atomicity.
As it turns out, qemu is correct and in our assembly code in entry.S and
pacache.S we don't pay attention to the required alignment.
This patch fixes the problem by aligning the lock offset in assembly
code in the same manner as we do in our C-code.
Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 79435ac78d upstream.
This used to setup the LP_COUNT register automatically, but now has been
removed.
There was an earlier fix 3c7c7a2fc8 which fixed instance in delay.h but
somehow missed this one as gcc change had not made its way into
production toolchains and was not pedantic as it is now !
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Kaiser cannot be used on paravirtualized MMUs (namely reading and writing CR3).
This does not work with KAISER as the CR3 switch from and to user space PGD
would require to map the whole XEN_PV machinery into both.
More importantly, enabling KAISER on Xen PV doesn't make too much sense, as PV
guests use distinct %cr3 values for kernel and user already.
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Let kaiser_flush_tlb_on_return_to_user() do the X86_FEATURE_PCID
check, instead of each caller doing it inline first: nobody needs
to optimize for the noPCID case, it's clearer this way, and better
suits later changes. Replace those no-op X86_CR3_PCID_KERN_FLUSH lines
by a BUILD_BUG_ON() in load_new_mm_cr3(), in case something changes.
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
I found asm/tlbflush.h too twisty, and think it safer not to avoid
__native_flush_tlb_global_irq_disabled() in the kaiser_enabled case,
but instead let it handle kaiser_enabled along with cr3: it can just
use __native_flush_tlb() for that, no harm in re-disabling preemption.
(This is not the same change as Kirill and Dave have suggested for
upstream, flipping PGE in cr4: that's neat, but needs a cpu_has_pge
check; cr3 is enough for kaiser, and thought to be cheaper than cr4.)
Also delete the X86_FEATURE_INVPCID invpcid_flush_all_nonglobals()
preference from __native_flush_tlb(): unlike the invpcid_flush_all()
preference in __native_flush_tlb_global(), it's not seen in upstream
4.14, and was recently reported to be surprisingly slow.
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
I have not observed a might_sleep() warning from setup_fixmap_gdt()'s
use of kaiser_add_mapping() in our tree (why not?), but like upstream
we have not provided a way for that to pass is_atomic true down to
kaiser_pagetable_walk(), and at startup it's far from a likely source
of trouble: so just delete the walk's is_atomic arg and might_sleep().
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now that we're playing the ALTERNATIVE game, use that more efficient
method: instead of user-mapping an extra page, and reading an extra
cacheline each time for x86_cr3_pcid_noflush.
Neel has found that __stringify(bts $X86_CR3_PCID_NOFLUSH_BIT, %rax)
is a working substitute for the "bts $63, %rax" in these ALTERNATIVEs;
but the one line with $63 in looks clearer, so let's stick with that.
Worried about what happens with an ALTERNATIVE between the jump and
jump label in another ALTERNATIVE? I was, but have checked the
combinations in SWITCH_KERNEL_CR3_NO_STACK at entry_SYSCALL_64,
and it does a good job.
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
AMD (and possibly other vendors) are not affected by the leak
KAISER is protecting against.
Keep the "nopti" for traditional reasons and add pti=<on|off|auto>
like upstream.
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Concentrate it in arch/x86/mm/kaiser.c and use the upstream string "nopti".
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Added "nokaiser" boot option: an early param like "noinvpcid".
Most places now check int kaiser_enabled (#defined 0 when not
CONFIG_KAISER) instead of #ifdef CONFIG_KAISER; but entry_64.S
and entry_64_compat.S are using the ALTERNATIVE technique, which
patches in the preferred instructions at runtime. That technique
is tied to x86 cpu features, so X86_FEATURE_KAISER is fabricated.
Prior to "nokaiser", Kaiser #defined _PAGE_GLOBAL 0: revert that,
but be careful with both _PAGE_GLOBAL and CR4.PGE: setting them when
nokaiser like when !CONFIG_KAISER, but not setting either when kaiser -
neither matters on its own, but it's hard to be sure that _PAGE_GLOBAL
won't get set in some obscure corner, or something add PGE into CR4.
By omitting _PAGE_GLOBAL from __supported_pte_mask when kaiser_enabled,
all page table setup which uses pte_pfn() masks it out of the ptes.
It's slightly shameful that the same declaration versus definition of
kaiser_enabled appears in not one, not two, but in three header files
(asm/kaiser.h, asm/pgtable.h, asm/tlbflush.h). I felt safer that way,
than with #including any of those in any of the others; and did not
feel it worth an asm/kaiser_enabled.h - kernel/cpu/common.c includes
them all, so we shall hear about it if they get out of synch.
Cleanups while in the area: removed the silly #ifdef CONFIG_KAISER
from kaiser.c; removed the unused native_get_normal_pgd(); removed
the spurious reg clutter from SWITCH_*_CR3 macro stubs; corrected some
comments. But more interestingly, set CR4.PSE in secondary_startup_64:
the manual is clear that it does not matter whether it's 0 or 1 when
4-level-pts are enabled, but I was distracted to find cr4 different on
BSP and auxiliaries - BSP alone was adding PSE, in probe_page_size_mask().
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
An error from kaiser_add_mapping() here is not at all likely, but
Eric Biggers rightly points out that __free_ldt_struct() relies on
new_ldt->size being initialized: move that up.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When removing the bogus comment from kaiser_remove_mapping(),
I really ought to have checked the extent of its bogosity: as
Neel points out, there is nothing to stop unmap_pud_range_nofree()
from continuing beyond the end of a pud (and starting in the wrong
position on the next).
Fix kaiser_remove_mapping() to constrain the extent and advance pgd
pointer correctly: use pgd_addr_end() macro as used throughout base
mm (but don't assume page-rounded start and size in this case).
But this bug was very unlikely to trigger in this backport: since
any buddy allocation is contained within a single pud extent, and
we are not using vmapped stacks (and are only mapping one page of
stack anyway): the only way to hit this bug here would be when
freeing a large modified ldt.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Neel Natu points out that paranoid_entry() was wrong to assume that
an entry that did not need swapgs would not need SWITCH_KERNEL_CR3:
paranoid_entry (used for debug breakpoint, int3, double fault or MCE;
though I think it's only the MCE case that is cause for concern here)
can break in at an awkward time, between cr3 switch and swapgs, but
its handling always needs kernel gs and kernel cr3.
Easy to fix in itself, but paranoid_entry() also needs to convey to
paranoid_exit() (and my reading of macro idtentry says paranoid_entry
and paranoid_exit are always paired) how to restore the prior state.
The swapgs state is already conveyed by %ebx (0 or 1), so extend that
also to convey when SWITCH_USER_CR3 will be needed (2 or 3).
(Yes, I'd much prefer that 0 meant no swapgs, whereas it's the other
way round: and a convention shared with error_entry() and error_exit(),
which I don't want to touch. Perhaps I should have inverted the bit
for switch cr3 too, but did not.)
paranoid_exit() would be straightforward, except for TRACE_IRQS: it
did TRACE_IRQS_IRETQ when doing swapgs, but TRACE_IRQS_IRETQ_DEBUG
when not: which is it supposed to use when SWITCH_USER_CR3 is split
apart from that? As best as I can determine, commit 5963e317b1
("ftrace/x86: Do not change stacks in DEBUG when calling lockdep")
missed the swapgs case, and should have used TRACE_IRQS_IRETQ_DEBUG
there too (the discrepancy has nothing to do with the liberal use
of _NO_STACK and _UNSAFE_STACK hereabouts: TRACE_IRQS_OFF_DEBUG has
just been used in all cases); discrepancy lovingly preserved across
several paranoid_exit() cleanups, but I'm now removing it.
Neel further indicates that to use SWITCH_USER_CR3_NO_STACK there in
paranoid_exit() is now not only unnecessary but unsafe: might corrupt
syscall entry's unsafe_stack_register_backup of %rax. Just use
SWITCH_USER_CR3: and delete SWITCH_USER_CR3_NO_STACK altogether,
before we make the mistake of using it again.
hughd adds: this commit fixes an issue in the Kaiser-without-PCIDs
part of the series, and ought to be moved earlier, if you decided
to make a release of Kaiser-without-PCIDs.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Mostly this commit is just unshouting X86_CR3_PCID_KERN_VAR and
X86_CR3_PCID_USER_VAR: we usually name variables in lower-case.
But why does x86_cr3_pcid_noflush need to be __aligned(PAGE_SIZE)?
Ah, it's a leftover from when kaiser_add_user_map() once complained
about mapping the same page twice. Make it __read_mostly instead.
(I'm a little uneasy about all the unrelated data which shares its
page getting user-mapped too, but that was so before, and not a big
deal: though we call it user-mapped, it's not mapped with _PAGE_USER.)
And there is a little change around the two calls to do_nmi().
Previously they set the NOFLUSH bit (if PCID supported) when
forcing to kernel context before do_nmi(); now they also have the
NOFLUSH bit set (if PCID supported) when restoring context after:
nothing done in do_nmi() should require a TLB to be flushed here.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Why was 4 chosen for kernel PCID and 6 for user PCID?
No good reason in a backport where PCIDs are only used for Kaiser.
If we continue with those, then we shall need to add Andy Lutomirski's
4.13 commit 6c690ee103 ("x86/mm: Split read_cr3() into read_cr3_pa()
and __read_cr3()"), which deals with the problem of read_cr3() callers
finding stray bits in the cr3 that they expected to be page-aligned;
and for hibernation, his 4.14 commit f34902c5c6 ("x86/hibernate/64:
Mask off CR3's PCID bits in the saved CR3").
But if 0 is used for kernel PCID, then there's no need to add in those
commits - whenever the kernel looks, it sees 0 in the lower bits; and
0 for kernel seems an obvious choice.
And I naughtily propose 128 for user PCID. Because there's a place
in _SWITCH_TO_USER_CR3 where it takes note of the need for TLB FLUSH,
but needs to reset that to NOFLUSH for the next occasion. Currently
it does so with a "movb $(0x80)" into the high byte of the per-cpu
quadword, but that will cause a machine without PCID support to crash.
Now, if %al just happened to have 0x80 in it at that point, on a
machine with PCID support, but 0 on a machine without PCID support...
(That will go badly wrong once the pgd can be at a physical address
above 2^56, but even with 5-level paging, physical goes up to 2^52.)
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We have many machines (Westmere, Sandybridge, Ivybridge) supporting
PCID but not INVPCID: on these load_new_mm_cr3() simply crashed.
Flushing user context inside load_new_mm_cr3() without the use of
invpcid is difficult: momentarily switch from kernel to user context
and back to do so? I'm not sure whether that can be safely done at
all, and would risk polluting user context with kernel internals,
and kernel context with stale user externals.
Instead, follow the hint in the comment that was there: change
X86_CR3_PCID_USER_VAR to be a per-cpu variable, then load_new_mm_cr3()
can leave a note in it, for SWITCH_USER_CR3 on return to userspace to
flush user context TLB, instead of default X86_CR3_PCID_USER_NOFLUSH.
Which works well enough that there's no need to do it this way only
when invpcid is unsupported: it's a good alternative to invpcid here.
But there's a couple of inlines in asm/tlbflush.h that need to do the
same trick, so it's best to localize all this per-cpu business in
mm/kaiser.c: moving that part of the initialization from setup_pcid()
to kaiser_setup_pcid(); with kaiser_flush_tlb_on_return_to_user() the
function for noting an X86_CR3_PCID_USER_FLUSH. And let's keep a
KAISER_SHADOW_PGD_OFFSET in there, to avoid the extra OR on exit.
I did try to make the feature tests in asm/tlbflush.h more consistent
with each other: there seem to be far too many ways of performing such
tests, and I don't have a good grasp of their differences. At first
I converted them all to be static_cpu_has(): but that proved to be a
mistake, as the comment in __native_flush_tlb_single() hints; so then
I reversed and made them all this_cpu_has(). Probably all gratuitous
change, but that's the way it's working at present.
I am slightly bothered by the way non-per-cpu X86_CR3_PCID_KERN_VAR
gets re-initialized by each cpu (before and after these changes):
no problem when (as usual) all cpus on a machine have the same
features, but in principle incorrect. However, my experiment
to per-cpu-ify that one did not end well...
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Merged performance improvements to Kaiser, using distinct kernel
and user Process Context Identifiers to minimize the TLB flushing.
[This work actually all from Dave Hansen 2017-08-30:
still omitting trackswitch mods, and KAISER_REAL_SWITCH deleted.]
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The kaiser update made an interesting choice, never to free any shadow
page tables. Contention on global spinlock was worrying, particularly
with it held across page table scans when freeing. Something had to be
done: I was going to add refcounting; but simply never to free them is
an appealing choice, minimizing contention without complicating the code
(the more a page table is found already, the less the spinlock is used).
But leaking pages in this way is also a worry: can we get away with it?
At the very least, we need a count to show how bad it actually gets:
in principle, one might end up wasting about 1/256 of memory that way
(1/512 for when direct-mapped pages have to be user-mapped, plus 1/512
for when they are user-mapped from the vmalloc area on another occasion
(but we don't have vmalloc'ed stacks, so only large ldts are vmalloc'ed).
Add per-cpu stat NR_KAISERTABLE: including 256 at startup for the
shared pgd entries, and 1 for each intermediate page table added
thereafter for user-mapping - but leave out the 1 per mm, for its
shadow pgd, because that distracts from the monotonic increase.
Shown in /proc/vmstat as nr_overhead (0 if kaiser not enabled).
In practice, it doesn't look so bad so far: more like 1/12000 after
nine hours of gtests below; and movable pageblock segregation should
tend to cluster the kaiser tables into a subset of the address space
(if not, they will be bad for compaction too). But production may
tell a different story: keep an eye on this number, and bring back
lighter freeing if it gets out of control (maybe a shrinker).
["nr_overhead" should of course say "nr_kaisertable", if it needs
to stay; but for the moment we are being coy, preferring that when
Joe Blow notices a new line in his /proc/vmstat, he does not get
too curious about what this "kaiser" stuff might be.]
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We fail to see what CONFIG_KAISER_REAL_SWITCH is for: it seems to be
left over from early development, and now just obscures tricky parts
of the code. Delete it before adding PCIDs, or nokaiser boot option.
(Or if there is some good reason to keep the option, then it needs
a help text - and a "depends on KAISER", so that all those without
KAISER are not asked the question. But we'd much rather delete it.)
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There's a 0x1000 in various places, which looks better with a name.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
While trying to get our gold link to work, four cleanups:
matched the gdt_page declaration to its definition;
in fiddling unsuccessfully with PERCPU_INPUT(), lined up backslashes;
lined up the backslashes according to convention in percpu-defs.h;
deleted the unused irq_stack_pointer addition to irq_stack_union.
Sad to report that aligning backslashes does not appear to help gold
align to 8192: but while these did not help, they are worth keeping.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Yes, unmap_pud_range_nofree()'s declaration ought to be in a
header file really, but I'm not sure we want to use it anyway:
so for now just declare it inside kaiser_remove_mapping().
And there doesn't seem to be such a thing as unmap_p4d_range(),
even in a 5-level paging tree.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kaiser_add_user_map() took no notice when kaiser_pagetable_walk() failed.
And avoid its might_sleep() when atomic (though atomic at present unused).
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Avoid perf crashes: place debug_store in the user-mapped per-cpu area
instead of allocating, and use page allocator plus kaiser_add_mapping()
to keep the BTS and PEBS buffers user-mapped (that is, present in the
user mapping, though visible only to kernel and hardware). The PEBS
fixup buffer does not need this treatment.
The need for a user-mapped struct debug_store showed up before doing
any conscious perf testing: in a couple of kernel paging oopses on
Westmere, implicating the debug_store offset of the per-cpu area.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
