openwrt/staging/blogic.git
6 years agoarm64: Kconfig: Reword UNMAP_KERNEL_AT_EL0 kconfig entry
Will Deacon [Tue, 14 Nov 2017 16:19:39 +0000 (16:19 +0000)]
arm64: Kconfig: Reword UNMAP_KERNEL_AT_EL0 kconfig entry

Although CONFIG_UNMAP_KERNEL_AT_EL0 does make KASLR more robust, it's
actually more useful as a mitigation against speculation attacks that
can leak arbitrary kernel data to userspace through speculation.

Reword the Kconfig help message to reflect this, and make the option
depend on EXPERT so that it is on by default for the majority of users.

Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
6 years agoarm64: use RET instruction for exiting the trampoline
Will Deacon [Tue, 14 Nov 2017 16:15:59 +0000 (16:15 +0000)]
arm64: use RET instruction for exiting the trampoline

Speculation attacks against the entry trampoline can potentially resteer
the speculative instruction stream through the indirect branch and into
arbitrary gadgets within the kernel.

This patch defends against these attacks by forcing a misprediction
through the return stack: a dummy BL instruction loads an entry into
the stack, so that the predicted program flow of the subsequent RET
instruction is to a branch-to-self instruction which is finally resolved
as a branch to the kernel vectors with speculation suppressed.

Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
6 years agoarm64: v8.4: Support for new floating point multiplication instructions
Dongjiu Geng [Wed, 13 Dec 2017 10:13:56 +0000 (18:13 +0800)]
arm64: v8.4: Support for new floating point multiplication instructions

ARM v8.4 extensions add new neon instructions for performing a
multiplication of each FP16 element of one vector with the corresponding
FP16 element of a second vector, and to add or subtract this without an
intermediate rounding to the corresponding FP32 element in a third vector.

This patch detects this feature and let the userspace know about it via a
HWCAP bit and MRS emulation.

Cc: Dave Martin <Dave.Martin@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Dongjiu Geng <gengdongjiu@huawei.com>
Reviewed-by: Dave Martin <Dave.Martin@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
6 years agoarm64: asid: Do not replace active_asids if already 0
Catalin Marinas [Wed, 27 Dec 2017 15:12:56 +0000 (15:12 +0000)]
arm64: asid: Do not replace active_asids if already 0

Under some uncommon timing conditions, a generation check and
xchg(active_asids, A1) in check_and_switch_context() on P1 can race with
an ASID roll-over on P2. If P2 has not seen the update to
active_asids[P1], it can re-allocate A1 to a new task T2 on P2. P1 ends
up waiting on the spinlock since the xchg() returned 0 while P2 can go
through a second ASID roll-over with (T2,A1,G2) active on P2. This
roll-over copies active_asids[P1] == A1,G1 into reserved_asids[P1] and
active_asids[P2] == A1,G2 into reserved_asids[P2]. A subsequent
scheduling of T1 on P1 and T2 on P2 would match reserved_asids and get
their generation bumped to G3:

P1 P2
--                                      --
TTBR0.BADDR = T0
TTBR0.ASID = A0
asid_generation = G1
check_and_switch_context(T1,A1,G1)
  generation match
check_and_switch_context(T2,A0,G0)
            new_context()
    ASID roll-over
    asid_generation = G2
    flush_context()
      active_asids[P1] = 0
      asid_map[A1] = 0
      reserved_asids[P1] = A0,G0
  xchg(active_asids, A1)
    active_asids[P1] = A1,G1
    xchg returns 0
  spin_lock_irqsave()
    allocated ASID (T2,A1,G2)
    asid_map[A1] = 1
  active_asids[P2] = A1,G2
...
check_and_switch_context(T3,A0,G0)
  new_context()
    ASID roll-over
    asid_generation = G3
    flush_context()
      active_asids[P1] = 0
      asid_map[A1] = 1
      reserved_asids[P1] = A1,G1
      reserved_asids[P2] = A1,G2
    allocated ASID (T3,A2,G3)
    asid_map[A2] = 1
  active_asids[P2] = A2,G3
  new_context()
    check_update_reserved_asid(A1,G1)
      matches reserved_asid[P1]
      reserved_asid[P1] = A1,G3
  updated T1 ASID to (T1,A1,G3)
check_and_switch_context(T2,A1,G2)
  new_context()
    check_and_switch_context(A1,G2)
      matches reserved_asids[P2]
      reserved_asids[P2] = A1,G3
  updated T2 ASID to (T2,A1,G3)

At this point, we have two tasks, T1 and T2 both using ASID A1 with the
latest generation G3. Any of them is allowed to be scheduled on the
other CPU leading to two different tasks with the same ASID on the same
CPU.

This patch changes the xchg to cmpxchg so that the active_asids is only
updated if non-zero to avoid a race with an ASID roll-over on a
different CPU.

The ASID allocation algorithm has been formally verified using the TLA+
model checker (see
https://git.kernel.org/pub/scm/linux/kernel/git/cmarinas/kernel-tla.git/tree/asidalloc.tla
for the spec).

Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
6 years agoarm64: make label allocation style consistent in tishift
Jason A. Donenfeld [Tue, 7 Nov 2017 02:24:04 +0000 (11:24 +0900)]
arm64: make label allocation style consistent in tishift

This is entirely cosmetic, but somehow it was missed when sending
differing versions of this patch. This just makes the file a bit more
uniform.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
6 years agoARM64 / cpuidle: Use new cpuidle macro for entering retention state
Prashanth Prakash [Wed, 15 Nov 2017 17:11:50 +0000 (10:11 -0700)]
ARM64 / cpuidle: Use new cpuidle macro for entering retention state

CPU_PM_CPU_IDLE_ENTER_RETENTION skips calling cpu_pm_enter() and
cpu_pm_exit(). By not calling cpu_pm functions in idle entry/exit
paths we can reduce the latency involved in entering and exiting
the low power idle state.

On ARM64 based Qualcomm server platform we measured below overhead
for calling cpu_pm_enter and cpu_pm_exit for retention states.

workload: stress --hdd #CPUs --hdd-bytes 32M  -t 30
Average overhead of cpu_pm_enter - 1.2us
Average overhead of cpu_pm_exit  - 3.1us

Acked-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
6 years agocpuidle: Add new macro to enter a retention idle state
Prashanth Prakash [Wed, 15 Nov 2017 17:11:49 +0000 (10:11 -0700)]
cpuidle: Add new macro to enter a retention idle state

If a CPU is entering a low power idle state where it doesn't lose any
context, then there is no need to call cpu_pm_enter()/cpu_pm_exit().
Add a new macro(CPU_PM_CPU_IDLE_ENTER_RETENTION) to be used by cpuidle
drivers when they are entering retention state. By not calling
cpu_pm_enter and cpu_pm_exit we reduce the latency involved in
entering and exiting the retention idle states.

CPU_PM_CPU_IDLE_ENTER_RETENTION assumes that no state is lost and
hence CPU PM notifiers will not be called. We may need a broader
change if we need to support partial retention states effeciently.

On ARM64 based Qualcomm Server Platform we measured below overhead for
for calling cpu_pm_enter and cpu_pm_exit for retention states.

workload: stress --hdd #CPUs --hdd-bytes 32M  -t 30
        Average overhead of cpu_pm_enter - 1.2us
        Average overhead of cpu_pm_exit  - 3.1us

Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoMerge branch 'for-next/52-bit-pa' into for-next/core
Catalin Marinas [Fri, 22 Dec 2017 17:40:58 +0000 (17:40 +0000)]
Merge branch 'for-next/52-bit-pa' into for-next/core

* for-next/52-bit-pa:
  arm64: enable 52-bit physical address support
  arm64: allow ID map to be extended to 52 bits
  arm64: handle 52-bit physical addresses in page table entries
  arm64: don't open code page table entry creation
  arm64: head.S: handle 52-bit PAs in PTEs in early page table setup
  arm64: handle 52-bit addresses in TTBR
  arm64: limit PA size to supported range
  arm64: add kconfig symbol to configure physical address size

7 years agoarm64: enable 52-bit physical address support
Kristina Martsenko [Wed, 13 Dec 2017 17:07:25 +0000 (17:07 +0000)]
arm64: enable 52-bit physical address support

Now that 52-bit physical address support is in place, add the kconfig
symbol to enable it. As described in ARMv8.2, the larger addresses are
only supported with the 64k granule. Also ensure that PAN is configured
(or TTBR0 PAN is not), as explained in an earlier patch in this series.

Tested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Bob Picco <bob.picco@oracle.com>
Reviewed-by: Bob Picco <bob.picco@oracle.com>
Signed-off-by: Kristina Martsenko <kristina.martsenko@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoarm64: allow ID map to be extended to 52 bits
Kristina Martsenko [Wed, 13 Dec 2017 17:07:24 +0000 (17:07 +0000)]
arm64: allow ID map to be extended to 52 bits

Currently, when using VA_BITS < 48, if the ID map text happens to be
placed in physical memory above VA_BITS, we increase the VA size (up to
48) and create a new table level, in order to map in the ID map text.
This is okay because the system always supports 48 bits of VA.

This patch extends the code such that if the system supports 52 bits of
VA, and the ID map text is placed that high up, then we increase the VA
size accordingly, up to 52.

One difference from the current implementation is that so far the
condition of VA_BITS < 48 has meant that the top level table is always
"full", with the maximum number of entries, and an extra table level is
always needed. Now, when VA_BITS = 48 (and using 64k pages), the top
level table is not full, and we simply need to increase the number of
entries in it, instead of creating a new table level.

Tested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Bob Picco <bob.picco@oracle.com>
Reviewed-by: Bob Picco <bob.picco@oracle.com>
Signed-off-by: Kristina Martsenko <kristina.martsenko@arm.com>
[catalin.marinas@arm.com: reduce arguments to __create_hyp_mappings()]
[catalin.marinas@arm.com: reworked/renamed __cpu_uses_extended_idmap_level()]
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoarm64: handle 52-bit physical addresses in page table entries
Kristina Martsenko [Wed, 13 Dec 2017 17:07:21 +0000 (17:07 +0000)]
arm64: handle 52-bit physical addresses in page table entries

The top 4 bits of a 52-bit physical address are positioned at bits
12..15 of a page table entry. Introduce macros to convert between a
physical address and its placement in a table entry, and change all
macros/functions that access PTEs to use them.

Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Tested-by: Bob Picco <bob.picco@oracle.com>
Reviewed-by: Bob Picco <bob.picco@oracle.com>
Signed-off-by: Kristina Martsenko <kristina.martsenko@arm.com>
[catalin.marinas@arm.com: some long lines wrapped]
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoarm64: don't open code page table entry creation
Kristina Martsenko [Wed, 13 Dec 2017 17:07:20 +0000 (17:07 +0000)]
arm64: don't open code page table entry creation

Instead of open coding the generation of page table entries, use the
macros/functions that exist for this - pfn_p*d and p*d_populate. Most
code in the kernel already uses these macros, this patch tries to fix
up the few places that don't. This is useful for the next patch in this
series, which needs to change the page table entry logic, and it's
better to have that logic in one place.

The KVM extended ID map is special, since we're creating a level above
CONFIG_PGTABLE_LEVELS and the required function isn't available. Leave
it as is and add a comment to explain it. (The normal kernel ID map code
doesn't need this change because its page tables are created in assembly
(__create_page_tables)).

Tested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Bob Picco <bob.picco@oracle.com>
Reviewed-by: Bob Picco <bob.picco@oracle.com>
Signed-off-by: Kristina Martsenko <kristina.martsenko@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoarm64: head.S: handle 52-bit PAs in PTEs in early page table setup
Kristina Martsenko [Wed, 13 Dec 2017 17:07:19 +0000 (17:07 +0000)]
arm64: head.S: handle 52-bit PAs in PTEs in early page table setup

The top 4 bits of a 52-bit physical address are positioned at bits
12..15 in page table entries. Introduce a macro to move the bits there,
and change the early ID map and swapper table setup code to use it.

Tested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Bob Picco <bob.picco@oracle.com>
Reviewed-by: Bob Picco <bob.picco@oracle.com>
Signed-off-by: Kristina Martsenko <kristina.martsenko@arm.com>
[catalin.marinas@arm.com: additional comments for clarification]
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoarm64: handle 52-bit addresses in TTBR
Kristina Martsenko [Wed, 13 Dec 2017 17:07:18 +0000 (17:07 +0000)]
arm64: handle 52-bit addresses in TTBR

The top 4 bits of a 52-bit physical address are positioned at bits 2..5
in the TTBR registers. Introduce a couple of macros to move the bits
there, and change all TTBR writers to use them.

Leave TTBR0 PAN code unchanged, to avoid complicating it. A system with
52-bit PA will have PAN anyway (because it's ARMv8.1 or later), and a
system without 52-bit PA can only use up to 48-bit PAs. A later patch in
this series will add a kconfig dependency to ensure PAN is configured.

In addition, when using 52-bit PA there is a special alignment
requirement on the top-level table. We don't currently have any VA_BITS
configuration that would violate the requirement, but one could be added
in the future, so add a compile-time BUG_ON to check for it.

Tested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Bob Picco <bob.picco@oracle.com>
Reviewed-by: Bob Picco <bob.picco@oracle.com>
Signed-off-by: Kristina Martsenko <kristina.martsenko@arm.com>
[catalin.marinas@arm.com: added TTBR_BADD_MASK_52 comment]
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoarm64: limit PA size to supported range
Kristina Martsenko [Wed, 13 Dec 2017 17:07:17 +0000 (17:07 +0000)]
arm64: limit PA size to supported range

We currently copy the physical address size from
ID_AA64MMFR0_EL1.PARange directly into TCR.(I)PS. This will not work for
4k and 16k granule kernels on systems that support 52-bit physical
addresses, since 52-bit addresses are only permitted with the 64k
granule.

To fix this, fall back to 48 bits when configuring the PA size when the
kernel does not support 52-bit PAs. When it does, fall back to 52, to
avoid similar problems in the future if the PA size is ever increased
above 52.

Tested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Bob Picco <bob.picco@oracle.com>
Reviewed-by: Bob Picco <bob.picco@oracle.com>
Signed-off-by: Kristina Martsenko <kristina.martsenko@arm.com>
[catalin.marinas@arm.com: tcr_set_pa_size macro renamed to tcr_compute_pa_size]
[catalin.marinas@arm.com: comments added to tcr_compute_pa_size]
[catalin.marinas@arm.com: definitions added for TCR_*PS_SHIFT]
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoarm64: add kconfig symbol to configure physical address size
Kristina Martsenko [Wed, 13 Dec 2017 17:07:16 +0000 (17:07 +0000)]
arm64: add kconfig symbol to configure physical address size

ARMv8.2 introduces support for 52-bit physical addresses. To prepare for
supporting this, add a new kconfig symbol to configure the physical
address space size. The symbols will be used in subsequent patches.
Currently the only choice is 48, a later patch will add the option of 52
once the required code is in place.

Tested-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Tested-by: Bob Picco <bob.picco@oracle.com>
Reviewed-by: Bob Picco <bob.picco@oracle.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Kristina Martsenko <kristina.martsenko@arm.com>
[catalin.marinas@arm.com: folded minor patches into this one]
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoMerge branch 'kpti' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Catalin Marinas [Mon, 11 Dec 2017 16:10:30 +0000 (16:10 +0000)]
Merge branch 'kpti' of git://git./linux/kernel/git/arm64/linux

Support for unmapping the kernel when running in userspace (aka
"KAISER").

* 'kpti' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: kaslr: Put kernel vectors address in separate data page
  arm64: mm: Introduce TTBR_ASID_MASK for getting at the ASID in the TTBR
  perf: arm_spe: Fail device probe when arm64_kernel_unmapped_at_el0()
  arm64: Kconfig: Add CONFIG_UNMAP_KERNEL_AT_EL0
  arm64: entry: Add fake CPU feature for unmapping the kernel at EL0
  arm64: tls: Avoid unconditional zeroing of tpidrro_el0 for native tasks
  arm64: erratum: Work around Falkor erratum #E1003 in trampoline code
  arm64: entry: Hook up entry trampoline to exception vectors
  arm64: entry: Explicitly pass exception level to kernel_ventry macro
  arm64: mm: Map entry trampoline into trampoline and kernel page tables
  arm64: entry: Add exception trampoline page for exceptions from EL0
  arm64: mm: Invalidate both kernel and user ASIDs when performing TLBI
  arm64: mm: Add arm64_kernel_unmapped_at_el0 helper
  arm64: mm: Allocate ASIDs in pairs
  arm64: mm: Fix and re-enable ARM64_SW_TTBR0_PAN
  arm64: mm: Rename post_ttbr0_update_workaround
  arm64: mm: Remove pre_ttbr0_update_workaround for Falkor erratum #E1003
  arm64: mm: Move ASID from TTBR0 to TTBR1
  arm64: mm: Temporarily disable ARM64_SW_TTBR0_PAN
  arm64: mm: Use non-global mappings for kernel space

7 years agoarm64: kaslr: Put kernel vectors address in separate data page
Will Deacon [Wed, 6 Dec 2017 11:24:02 +0000 (11:24 +0000)]
arm64: kaslr: Put kernel vectors address in separate data page

The literal pool entry for identifying the vectors base is the only piece
of information in the trampoline page that identifies the true location
of the kernel.

This patch moves it into a page-aligned region of the .rodata section
and maps this adjacent to the trampoline text via an additional fixmap
entry, which protects against any accidental leakage of the trampoline
contents.

Suggested-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Introduce TTBR_ASID_MASK for getting at the ASID in the TTBR
Will Deacon [Fri, 1 Dec 2017 17:33:48 +0000 (17:33 +0000)]
arm64: mm: Introduce TTBR_ASID_MASK for getting at the ASID in the TTBR

There are now a handful of open-coded masks to extract the ASID from a
TTBR value, so introduce a TTBR_ASID_MASK and use that instead.

Suggested-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoperf: arm_spe: Fail device probe when arm64_kernel_unmapped_at_el0()
Will Deacon [Mon, 27 Nov 2017 15:49:53 +0000 (15:49 +0000)]
perf: arm_spe: Fail device probe when arm64_kernel_unmapped_at_el0()

When running with the kernel unmapped whilst at EL0, the virtually-addressed
SPE buffer is also unmapped, which can lead to buffer faults if userspace
profiling is enabled and potentially also when writing back kernel samples
unless an expensive drain operation is performed on exception return.

For now, fail the SPE driver probe when arm64_kernel_unmapped_at_el0().

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: Kconfig: Add CONFIG_UNMAP_KERNEL_AT_EL0
Will Deacon [Tue, 14 Nov 2017 14:41:01 +0000 (14:41 +0000)]
arm64: Kconfig: Add CONFIG_UNMAP_KERNEL_AT_EL0

Add a Kconfig entry to control use of the entry trampoline, which allows
us to unmap the kernel whilst running in userspace and improve the
robustness of KASLR.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: entry: Add fake CPU feature for unmapping the kernel at EL0
Will Deacon [Tue, 14 Nov 2017 14:38:19 +0000 (14:38 +0000)]
arm64: entry: Add fake CPU feature for unmapping the kernel at EL0

Allow explicit disabling of the entry trampoline on the kernel command
line (kpti=off) by adding a fake CPU feature (ARM64_UNMAP_KERNEL_AT_EL0)
that can be used to toggle the alternative sequences in our entry code and
avoid use of the trampoline altogether if desired. This also allows us to
make use of a static key in arm64_kernel_unmapped_at_el0().

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: tls: Avoid unconditional zeroing of tpidrro_el0 for native tasks
Will Deacon [Tue, 14 Nov 2017 14:33:28 +0000 (14:33 +0000)]
arm64: tls: Avoid unconditional zeroing of tpidrro_el0 for native tasks

When unmapping the kernel at EL0, we use tpidrro_el0 as a scratch register
during exception entry from native tasks and subsequently zero it in
the kernel_ventry macro. We can therefore avoid zeroing tpidrro_el0
in the context-switch path for native tasks using the entry trampoline.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: erratum: Work around Falkor erratum #E1003 in trampoline code
Will Deacon [Tue, 14 Nov 2017 14:29:19 +0000 (14:29 +0000)]
arm64: erratum: Work around Falkor erratum #E1003 in trampoline code

We rely on an atomic swizzling of TTBR1 when transitioning from the entry
trampoline to the kernel proper on an exception. We can't rely on this
atomicity in the face of Falkor erratum #E1003, so on affected cores we
can issue a TLB invalidation to invalidate the walk cache prior to
jumping into the kernel. There is still the possibility of a TLB conflict
here due to conflicting walk cache entries prior to the invalidation, but
this doesn't appear to be the case on these CPUs in practice.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: entry: Hook up entry trampoline to exception vectors
Will Deacon [Tue, 14 Nov 2017 14:24:29 +0000 (14:24 +0000)]
arm64: entry: Hook up entry trampoline to exception vectors

Hook up the entry trampoline to our exception vectors so that all
exceptions from and returns to EL0 go via the trampoline, which swizzles
the vector base register accordingly. Transitioning to and from the
kernel clobbers x30, so we use tpidrro_el0 and far_el1 as scratch
registers for native tasks.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: entry: Explicitly pass exception level to kernel_ventry macro
Will Deacon [Tue, 14 Nov 2017 14:20:21 +0000 (14:20 +0000)]
arm64: entry: Explicitly pass exception level to kernel_ventry macro

We will need to treat exceptions from EL0 differently in kernel_ventry,
so rework the macro to take the exception level as an argument and
construct the branch target using that.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Map entry trampoline into trampoline and kernel page tables
Will Deacon [Tue, 14 Nov 2017 14:14:17 +0000 (14:14 +0000)]
arm64: mm: Map entry trampoline into trampoline and kernel page tables

The exception entry trampoline needs to be mapped at the same virtual
address in both the trampoline page table (which maps nothing else)
and also the kernel page table, so that we can swizzle TTBR1_EL1 on
exceptions from and return to EL0.

This patch maps the trampoline at a fixed virtual address in the fixmap
area of the kernel virtual address space, which allows the kernel proper
to be randomized with respect to the trampoline when KASLR is enabled.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: entry: Add exception trampoline page for exceptions from EL0
Will Deacon [Tue, 14 Nov 2017 14:07:40 +0000 (14:07 +0000)]
arm64: entry: Add exception trampoline page for exceptions from EL0

To allow unmapping of the kernel whilst running at EL0, we need to
point the exception vectors at an entry trampoline that can map/unmap
the kernel on entry/exit respectively.

This patch adds the trampoline page, although it is not yet plugged
into the vector table and is therefore unused.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Invalidate both kernel and user ASIDs when performing TLBI
Will Deacon [Thu, 10 Aug 2017 13:13:33 +0000 (14:13 +0100)]
arm64: mm: Invalidate both kernel and user ASIDs when performing TLBI

Since an mm has both a kernel and a user ASID, we need to ensure that
broadcast TLB maintenance targets both address spaces so that things
like CoW continue to work with the uaccess primitives in the kernel.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Add arm64_kernel_unmapped_at_el0 helper
Will Deacon [Tue, 14 Nov 2017 13:58:08 +0000 (13:58 +0000)]
arm64: mm: Add arm64_kernel_unmapped_at_el0 helper

In order for code such as TLB invalidation to operate efficiently when
the decision to map the kernel at EL0 is determined at runtime, this
patch introduces a helper function, arm64_kernel_unmapped_at_el0, to
determine whether or not the kernel is mapped whilst running in userspace.

Currently, this just reports the value of CONFIG_UNMAP_KERNEL_AT_EL0,
but will later be hooked up to a fake CPU capability using a static key.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Allocate ASIDs in pairs
Will Deacon [Thu, 10 Aug 2017 13:10:28 +0000 (14:10 +0100)]
arm64: mm: Allocate ASIDs in pairs

In preparation for separate kernel/user ASIDs, allocate them in pairs
for each mm_struct. The bottom bit distinguishes the two: if it is set,
then the ASID will map only userspace.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Fix and re-enable ARM64_SW_TTBR0_PAN
Will Deacon [Thu, 10 Aug 2017 12:58:16 +0000 (13:58 +0100)]
arm64: mm: Fix and re-enable ARM64_SW_TTBR0_PAN

With the ASID now installed in TTBR1, we can re-enable ARM64_SW_TTBR0_PAN
by ensuring that we switch to a reserved ASID of zero when disabling
user access and restore the active user ASID on the uaccess enable path.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Rename post_ttbr0_update_workaround
Will Deacon [Thu, 10 Aug 2017 12:34:30 +0000 (13:34 +0100)]
arm64: mm: Rename post_ttbr0_update_workaround

The post_ttbr0_update_workaround hook applies to any change to TTBRx_EL1.
Since we're using TTBR1 for the ASID, rename the hook to make it clearer
as to what it's doing.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Remove pre_ttbr0_update_workaround for Falkor erratum #E1003
Will Deacon [Thu, 10 Aug 2017 12:29:06 +0000 (13:29 +0100)]
arm64: mm: Remove pre_ttbr0_update_workaround for Falkor erratum #E1003

The pre_ttbr0_update_workaround hook is called prior to context-switching
TTBR0 because Falkor erratum E1003 can cause TLB allocation with the wrong
ASID if both the ASID and the base address of the TTBR are updated at
the same time.

With the ASID sitting safely in TTBR1, we no longer update things
atomically, so we can remove the pre_ttbr0_update_workaround macro as
it's no longer required. The erratum infrastructure and documentation
is left around for #E1003, as it will be required by the entry
trampoline code in a future patch.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Move ASID from TTBR0 to TTBR1
Will Deacon [Thu, 10 Aug 2017 12:19:09 +0000 (13:19 +0100)]
arm64: mm: Move ASID from TTBR0 to TTBR1

In preparation for mapping kernelspace and userspace with different
ASIDs, move the ASID to TTBR1 and update switch_mm to context-switch
TTBR0 via an invalid mapping (the zero page).

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Temporarily disable ARM64_SW_TTBR0_PAN
Will Deacon [Thu, 10 Aug 2017 12:04:48 +0000 (13:04 +0100)]
arm64: mm: Temporarily disable ARM64_SW_TTBR0_PAN

We're about to rework the way ASIDs are allocated, switch_mm is
implemented and low-level kernel entry/exit is handled, so keep the
ARM64_SW_TTBR0_PAN code out of the way whilst we do the heavy lifting.

It will be re-enabled in a subsequent patch.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoarm64: mm: Use non-global mappings for kernel space
Will Deacon [Thu, 10 Aug 2017 11:56:18 +0000 (12:56 +0100)]
arm64: mm: Use non-global mappings for kernel space

In preparation for unmapping the kernel whilst running in userspace,
make the kernel mappings non-global so we can avoid expensive TLB
invalidation on kernel exit to userspace.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Tested-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
7 years agoLinux 4.15-rc3
Linus Torvalds [Mon, 11 Dec 2017 01:56:26 +0000 (17:56 -0800)]
Linux 4.15-rc3

7 years agohpfs: don't bother with the i_version counter or f_version
Jeff Layton [Thu, 23 Nov 2017 19:34:40 +0000 (20:34 +0100)]
hpfs: don't bother with the i_version counter or f_version

HPFS does not set SB_I_VERSION and does not use the i_version counter
internally.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Mikulas Patocka <mikulas@twibright.com>
Reviewed-by: Mikulas Patocka <mikulas@twibright.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agofutex: futex_wake_op, fix sign_extend32 sign bits
Jiri Slaby [Thu, 30 Nov 2017 14:35:44 +0000 (15:35 +0100)]
futex: futex_wake_op, fix sign_extend32 sign bits

sign_extend32 counts the sign bit parameter from 0, not from 1.  So we
have to use "11" for 12th bit, not "12".

This mistake means we have not allowed negative op and cmp args since
commit 30d6e0a4190d ("futex: Remove duplicated code and fix undefined
behaviour") till now.

Fixes: 30d6e0a4190d ("futex: Remove duplicated code and fix undefined behaviour")
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Darren Hart <dvhart@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoMerge tag 'for-4.15-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
Linus Torvalds [Sun, 10 Dec 2017 16:30:04 +0000 (08:30 -0800)]
Merge tag 'for-4.15-rc3-tag' of git://git./linux/kernel/git/kdave/linux

Pull btrfs fixes from David Sterba:
 "This contains a few fixes (error handling, quota leak, FUA vs
  nobarrier mount option).

  There's one one worth mentioning separately - an off-by-one fix that
  leads to overwriting first byte of an adjacent page with 0, out of
  bounds of the memory allocated by an ioctl. This is under a privileged
  part of the ioctl, can be triggerd in some subvolume layouts"

* tag 'for-4.15-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: Fix possible off-by-one in btrfs_search_path_in_tree
  Btrfs: disable FUA if mounted with nobarrier
  btrfs: fix missing error return in btrfs_drop_snapshot
  btrfs: handle errors while updating refcounts in update_ref_for_cow
  btrfs: Fix quota reservation leak on preallocated files

7 years agoMerge tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Sun, 10 Dec 2017 16:26:59 +0000 (08:26 -0800)]
Merge tag 'armsoc-fixes' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Olof Johansson:

 - A revert of all SCPI changes from the 4.15 merge window. They had
   regressions on the Amlogic platforms, and the submaintainer isn't
   around to fix these bugs due to vacation, etc. So we agreed to revert
   and revisit in next release cycle.

 - A series fixing a number of bugs for ARM CCN interconnect, around
   module unload, smp_processor_id() in preemptable context, and fixing
   some memory allocation failure checks.

 - A handful of devicetree fixes for different platforms, fixing
   warnings and errors that were previously ignored by the compiler.

 - The usual set of mostly minor fixes for different platforms.

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (42 commits)
  ARM64: dts: meson-gx: fix UART pclk clock name
  ARM: omap2: hide omap3_save_secure_ram on non-OMAP3 builds
  arm: dts: nspire: Add missing #phy-cells to usb-nop-xceiv
  ARM: dts: Fix dm814x missing phy-cells property
  ARM: dts: Fix elm interrupt compiler warning
  bus: arm-ccn: fix module unloading Error: Removing state 147 which has instances left.
  bus: arm-cci: Fix use of smp_processor_id() in preemptible context
  bus: arm-ccn: Fix use of smp_processor_id() in preemptible context
  bus: arm-ccn: Simplify code
  bus: arm-ccn: Check memory allocation failure
  bus: arm-ccn: constify attribute_group structures.
  firmware: arm_scpi: Revert updates made during v4.15 merge window
  arm: dts: marvell: Add missing #phy-cells to usb-nop-xceiv
  arm64: dts: sort vendor subdirectories in Makefile alphabetically
  meson-gx-socinfo: Fix package id parsing
  ARM: meson: fix spelling mistake: "Couln't" -> "Couldn't"
  ARM: dts: meson: fix the memory region of the GPIO interrupt controller
  ARM: dts: meson: correct the sort order for the the gpio_intc node
  MAINTAINERS: exclude other Socionext SoC DT files from ARM/UNIPHIER entry
  arm64: dts: uniphier: remove unnecessary interrupt-parent
  ...

7 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sun, 10 Dec 2017 16:24:16 +0000 (08:24 -0800)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fixes from Radim Krčmář:
 "ARM:
   - A number of issues in the vgic discovered using SMATCH
   - A bit one-off calculation in out stage base address mask (32-bit
     and 64-bit)
   - Fixes to single-step debugging instructions that trap for other
     reasons such as MMMIO aborts
   - Printing unavailable hyp mode as error
   - Potential spinlock deadlock in the vgic
   - Avoid calling vgic vcpu free more than once
   - Broken bit calculation for big endian systems

 s390:
   - SPDX tags
   - Fence storage key accesses from problem state
   - Make sure that irq_state.flags is not used in the future

  x86:
   - Intercept port 0x80 accesses to prevent host instability (CVE)
   - Use userspace FPU context for guest FPU (mainly an optimization
     that fixes a double use of kernel FPU)
   - Do not leak one page per module load
   - Flush APIC page address cache from MMU invalidation notifiers"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (28 commits)
  KVM: x86: fix APIC page invalidation
  KVM: s390: Fix skey emulation permission check
  KVM: s390: mark irq_state.flags as non-usable
  KVM: s390: Remove redundant license text
  KVM: s390: add SPDX identifiers to the remaining files
  KVM: VMX: fix page leak in hardware_setup()
  KVM: VMX: remove I/O port 0x80 bypass on Intel hosts
  x86,kvm: remove KVM emulator get_fpu / put_fpu
  x86,kvm: move qemu/guest FPU switching out to vcpu_run
  KVM: arm/arm64: Fix broken GICH_ELRSR big endian conversion
  KVM: arm/arm64: kvm_arch_destroy_vm cleanups
  KVM: arm/arm64: Fix spinlock acquisition in vgic_set_owner
  kvm: arm: don't treat unavailable HYP mode as an error
  KVM: arm/arm64: Avoid attempting to load timer vgic state without a vgic
  kvm: arm64: handle single-step of hyp emulated mmio instructions
  kvm: arm64: handle single-step during SError exceptions
  kvm: arm64: handle single-step of userspace mmio instructions
  kvm: arm64: handle single-stepping trapped instructions
  KVM: arm/arm64: debug: Introduce helper for single-step
  arm: KVM: Fix VTTBR_BADDR_MASK BUG_ON off-by-one
  ...

7 years agoMerge branch 'fixes' into for-next
Olof Johansson [Sun, 10 Dec 2017 04:23:58 +0000 (20:23 -0800)]
Merge branch 'fixes' into for-next

* fixes:
  ARM64: dts: meson-gx: fix UART pclk clock name
  ARM: dts: Fix dm814x missing phy-cells property
  ARM: dts: Fix elm interrupt compiler warning
  bus: arm-ccn: fix module unloading Error: Removing state 147 which has instances left.
  bus: arm-cci: Fix use of smp_processor_id() in preemptible context
  bus: arm-ccn: Fix use of smp_processor_id() in preemptible context
  bus: arm-ccn: Simplify code
  bus: arm-ccn: Check memory allocation failure
  bus: arm-ccn: constify attribute_group structures.
  meson-gx-socinfo: Fix package id parsing
  ARM: meson: fix spelling mistake: "Couln't" -> "Couldn't"
  ARM: dts: meson: fix the memory region of the GPIO interrupt controller
  ARM: dts: meson: correct the sort order for the the gpio_intc node

7 years agoMerge tag 'amlogic-fixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git/khilman...
Olof Johansson [Sun, 10 Dec 2017 04:23:29 +0000 (20:23 -0800)]
Merge tag 'amlogic-fixes-1' of git://git./linux/kernel/git/khilman/linux-amlogic into fixes

Amlogic fixes for v4.15-rc
- GPIO interrupt fixes
- socinfo fix for GX series
- fix typo

* tag 'amlogic-fixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git/khilman/linux-amlogic:
  ARM64: dts: meson-gx: fix UART pclk clock name
  meson-gx-socinfo: Fix package id parsing
  ARM: meson: fix spelling mistake: "Couln't" -> "Couldn't"
  ARM: dts: meson: fix the memory region of the GPIO interrupt controller
  ARM: dts: meson: correct the sort order for the the gpio_intc node

Signed-off-by: Olof Johansson <olof@lixom.net>
7 years agoMerge tag 'ccn/fixes-for-4.15' of git://git.linaro.org/people/pawel.moll/linux into...
Olof Johansson [Sun, 10 Dec 2017 04:22:46 +0000 (20:22 -0800)]
Merge tag 'ccn/fixes-for-4.15' of git://git.linaro.org/people/pawel.moll/linux into fixes

bus: ARM CCN and CCI PMU driver fixes

This is a bunch of fixes CCN and (guest starring this time) CCI drivers.

* Check for potential of failed allocation for the driver name string
* Manage CPU ID properly at allocation (both CCN and CCI)
* Fix module unload warnings related to objects release order
* Small improvements like using allocating printfs and proper
  attributes constification

The one fixing potential issues have been cc-ed to stable.

* tag 'ccn/fixes-for-4.15' of git://git.linaro.org/people/pawel.moll/linux:
  bus: arm-ccn: fix module unloading Error: Removing state 147 which has instances left.
  bus: arm-cci: Fix use of smp_processor_id() in preemptible context
  bus: arm-ccn: Fix use of smp_processor_id() in preemptible context
  bus: arm-ccn: Simplify code
  bus: arm-ccn: Check memory allocation failure
  bus: arm-ccn: constify attribute_group structures.

Signed-off-by: Olof Johansson <olof@lixom.net>
7 years agoMerge tag 'omap-for-v4.15/fixes-dt-warnings' of git://git.kernel.org/pub/scm/linux...
Olof Johansson [Sun, 10 Dec 2017 04:22:01 +0000 (20:22 -0800)]
Merge tag 'omap-for-v4.15/fixes-dt-warnings' of git://git./linux/kernel/git/tmlind/linux-omap into fixes

Two fixes for dts compiler warnings

These recently started showing up with better dtc checks being
introduced.

* tag 'omap-for-v4.15/fixes-dt-warnings' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap:
  ARM: dts: Fix dm814x missing phy-cells property
  ARM: dts: Fix elm interrupt compiler warning

Signed-off-by: Olof Johansson <olof@lixom.net>
7 years agoMerge tag 'keys-fixes-20171208' of git://git.kernel.org/pub/scm/linux/kernel/git...
James Morris [Sat, 9 Dec 2017 03:39:48 +0000 (14:39 +1100)]
Merge tag 'keys-fixes-20171208' of git://git./linux/kernel/git/dhowells/linux-fs into keys-for-linus

Assorted fixes for keyrings, ASN.1, X.509 and PKCS#7.

7 years agokmemcheck: rip it out for real
Michal Hocko [Wed, 6 Dec 2017 10:27:57 +0000 (11:27 +0100)]
kmemcheck: rip it out for real

Commit 4675ff05de2d ("kmemcheck: rip it out") has removed the code but
for some reason SPDX header stayed in place.  This looks like a rebase
mistake in the mmotm tree or the merge mistake.  Let's drop those
leftovers as well.

Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Fri, 8 Dec 2017 21:32:44 +0000 (13:32 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) CAN fixes from Martin Kelly (cancel URBs properly in all the CAN usb
    drivers).

 2) Revert returning -EEXIST from __dev_alloc_name() as this propagates
    to userspace and broke some apps. From Johannes Berg.

 3) Fix conn memory leaks and crashes in TIPC, from Jon Malloc and Cong
    Wang.

 4) Gianfar MAC can't do EEE so don't advertise it by default, from
    Claudiu Manoil.

 5) Relax strict netlink attribute validation, but emit a warning. From
    David Ahern.

 6) Fix regression in checksum offload of thunderx driver, from Florian
    Westphal.

 7) Fix UAPI bpf issues on s390, from Hendrik Brueckner.

 8) New card support in iwlwifi, from Ihab Zhaika.

 9) BBR congestion control bug fixes from Neal Cardwell.

10) Fix port stats in nfp driver, from Pieter Jansen van Vuuren.

11) Fix leaks in qualcomm rmnet, from Subash Abhinov Kasiviswanathan.

12) Fix DMA API handling in sh_eth driver, from Thomas Petazzoni.

13) Fix spurious netpoll warnings in bnxt_en, from Calvin Owens.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (67 commits)
  net: mvpp2: fix the RSS table entry offset
  tcp: evaluate packet losses upon RTT change
  tcp: fix off-by-one bug in RACK
  tcp: always evaluate losses in RACK upon undo
  tcp: correctly test congestion state in RACK
  bnxt_en: Fix sources of spurious netpoll warnings
  tcp_bbr: reset long-term bandwidth sampling on loss recovery undo
  tcp_bbr: reset full pipe detection on loss recovery undo
  tcp_bbr: record "full bw reached" decision in new full_bw_reached bit
  sfc: pass valid pointers from efx_enqueue_unwind
  gianfar: Disable EEE autoneg by default
  tcp: invalidate rate samples during SACK reneging
  can: peak/pcie_fd: fix potential bug in restarting tx queue
  can: usb_8dev: cancel urb on -EPIPE and -EPROTO
  can: kvaser_usb: cancel urb on -EPIPE and -EPROTO
  can: esd_usb2: cancel urb on -EPIPE and -EPROTO
  can: ems_usb: cancel urb on -EPIPE and -EPROTO
  can: mcba_usb: cancel urb on -EPROTO
  usbnet: fix alignment for frames with no ethernet header
  tcp: use current time in tcp_rcv_space_adjust()
  ...

7 years agoMerge tag 'media/v4.15-2' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
Linus Torvalds [Fri, 8 Dec 2017 21:18:47 +0000 (13:18 -0800)]
Merge tag 'media/v4.15-2' of git://git./linux/kernel/git/mchehab/linux-media

Pull media fixes from Mauro Carvalho Chehab:

 "A series of fixes for the media subsytem:

   - The largest amount of fixes in this series is with regards to
     comments that aren't kernel-doc, but start with "/**".

     A new check added for 4.15 makes it to produce a *huge* amount of
     new warnings (I'm compiling here with W=1). Most of the patches in
     this series fix those.

     No code changes - just comment changes at the source files

   - rc: some fixed in order to better handle RC repetition codes

   - v4l-async: use the v4l2_dev from the root notifier when matching
     sub-devices

   - v4l2-fwnode: Check subdev count after checking port

   - ov 13858 and et8ek8: compilation fix with randconfigs

   - usbtv: a trivial new USB ID addition

   - dibusb-common: don't do DMA on stack on firmware load

   - imx274: Fix error handling, add MAINTAINERS entry

   - sir_ir: detect presence of port"

* tag 'media/v4.15-2' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media: (50 commits)
  media: imx274: Fix error handling, add MAINTAINERS entry
  media: v4l: async: use the v4l2_dev from the root notifier when matching sub-devices
  media: v4l2-fwnode: Check subdev count after checking port
  media: et8ek8: select V4L2_FWNODE
  media: ov13858: Select V4L2_FWNODE
  media: rc: partial revert of "media: rc: per-protocol repeat period"
  media: dvb: i2c transfers over usb cannot be done from stack
  media: dvb-frontends: complete kernel-doc markups
  media: docs: add documentation for frontend attach info
  media: dvb_frontends: fix kernel-doc macros
  media: drivers: remove "/**" from non-kernel-doc comments
  media: lm3560: add a missing kernel-doc parameter
  media: rcar_jpu: fix two kernel-doc markups
  media: vsp1: add a missing kernel-doc parameter
  media: soc_camera: fix a kernel-doc markup
  media: mt2063: fix some kernel-doc warnings
  media: radio-wl1273: fix a parameter name at kernel-doc macro
  media: s3c-camif: add missing description at s3c_camif_find_format()
  media: mtk-vpu: add description for wdt fields at struct mtk_vpu
  media: vdec: fix some kernel-doc warnings
  ...

7 years agoMerge tag 'drm-fixes-for-v4.15-rc3' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Fri, 8 Dec 2017 21:11:57 +0000 (13:11 -0800)]
Merge tag 'drm-fixes-for-v4.15-rc3' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "This pull is a bit larger than I'd like but a large bunch of it is
  license fixes, AMD wanted to fix the licenses for a bunch of files
  that were missing them,

 Otherwise a bunch of TTM regression fix since the hugepage support,
 some i915 and gvt fixes, a core connector free in a safe context fix,
 and one bridge fix"

* tag 'drm-fixes-for-v4.15-rc3' of git://people.freedesktop.org/~airlied/linux: (26 commits)
  drm/bridge: analogix dp: Fix runtime PM state in get_modes() callback
  Revert "drm/i915: Display WA #1133 WaFbcSkipSegments:cnl, glk"
  drm/vc4: Fix false positive WARN() backtrace on refcount_inc() usage
  drm/i915: Call i915_gem_init_userptr() before taking struct_mutex
  drm/exynos: remove unnecessary function declaration
  drm/exynos: remove unnecessary descrptions
  drm/exynos: gem: Drop NONCONTIG flag for buffers allocated without IOMMU
  drm/exynos: Fix dma-buf import
  drm/ttm: swap consecutive allocated pooled pages v4
  drm: safely free connectors from connector_iter
  drm/i915/gvt: set max priority for gvt context
  drm/i915/gvt: Don't mark vgpu context as inactive when preempted
  drm/i915/gvt: Limit read hw reg to active vgpu
  drm/i915/gvt: Export intel_gvt_render_mmio_to_ring_id()
  drm/i915/gvt: Emulate PCI expansion ROM base address register
  drm/ttm: swap consecutive allocated cached pages v3
  drm/ttm: roundup the shrink request to prevent skip huge pool
  drm/ttm: add page order support in ttm_pages_put
  drm/ttm: add set_pages_wb for handling page order more than zero
  drm/ttm: add page order in page pool
  ...

7 years agoMerge tag 'md/4.15-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md
Linus Torvalds [Fri, 8 Dec 2017 21:03:02 +0000 (13:03 -0800)]
Merge tag 'md/4.15-rc2' of git://git./linux/kernel/git/shli/md

Pull md fixes from Shaohua Li:
 "Some MD fixes.

  The notable one is a raid5-cache deadlock bug with dm-raid, others are
  not significant"

* tag 'md/4.15-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md:
  md/raid1/10: add missed blk plug
  md: limit mdstat resync progress to max_sectors
  md/r5cache: move mddev_lock() out of r5c_journal_mode_set()
  md/raid5: correct degraded calculation in raid5_error

7 years agoMerge tag 'devicetree-fixes-for-4.15-part2' of git://git.kernel.org/pub/scm/linux...
Linus Torvalds [Fri, 8 Dec 2017 21:00:51 +0000 (13:00 -0800)]
Merge tag 'devicetree-fixes-for-4.15-part2' of git://git./linux/kernel/git/robh/linux

Pull DeviceTree fixes from Rob Herring:
 "Another set of DT fixes:

   - Fixes from overlay code rework. A trifecta of fixes to the locking,
     an out of bounds access, and a memory leak in of_overlay_apply()

   - Clean-up at25 eeprom binding document

   - Remove leading '0x' in unit-addresses from binding docs"

* tag 'devicetree-fixes-for-4.15-part2' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  of: overlay: Make node skipping in init_overlay_changeset() clearer
  of: overlay: Fix out-of-bounds write in init_overlay_changeset()
  of: overlay: Fix (un)locking in of_overlay_apply()
  of: overlay: Fix memory leak in of_overlay_apply() error path
  dt-bindings: eeprom: at25: Document device-specific compatible values
  dt-bindings: eeprom: at25: Grammar s/are can/can/
  dt-bindings: Remove leading 0x from bindings notation
  of: overlay: Remove else after goto
  of: Spelling s/changset/changeset/
  of: unittest: Remove bogus overlay mutex release from overlay_data_add()

7 years agoMerge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
Linus Torvalds [Fri, 8 Dec 2017 20:58:51 +0000 (12:58 -0800)]
Merge tag 'for_linus' of git://git./linux/kernel/git/mst/vhost

Pull virtio bugfixes from Michael Tsirkin:
 "A couple of minor bugfixes"

* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
  virtio_net: fix return value check in receive_mergeable()
  virtio_mmio: add cleanup for virtio_mmio_remove
  virtio_mmio: add cleanup for virtio_mmio_probe

7 years agoMerge tag 'for-linus-4.15-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 8 Dec 2017 20:53:43 +0000 (12:53 -0800)]
Merge tag 'for-linus-4.15-rc3-tag' of git://git./linux/kernel/git/xen/tip

Pull xen fixes from Juergen Gross:
 "Just two small fixes for the new pvcalls frontend driver"

* tag 'for-linus-4.15-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen/pvcalls: Fix a check in pvcalls_front_remove()
  xen/pvcalls: check for xenbus_read() errors

7 years agoMerge tag 'powerpc-4.15-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Fri, 8 Dec 2017 20:52:09 +0000 (12:52 -0800)]
Merge tag 'powerpc-4.15-4' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:

 "One notable fix for kexec on Power9, where we were not clearing MMU
  PID properly which sometimes leads to hangs. Finally debugged to a
  root cause by Nick.

  A revert of a patch which tried to rework our panic handling to get
  more output on the console, but inadvertently broke reporting the
  panic to the hypervisor, which apparently people care about.

  Then a fix for an oops in the PMU code, and finally some s/%p/%px/ in
  xmon.

  Thanks to: David Gibson, Nicholas Piggin, Ravi Bangoria"

* tag 'powerpc-4.15-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/xmon: Don't print hashed pointers in xmon
  powerpc/64s: Initialize ISAv3 MMU registers before setting partition table
  Revert "powerpc: Do not call ppc_md.panic in fadump panic notifier"
  powerpc/perf: Fix oops when grouping different pmu events

7 years agoMerge tag 'linux-can-fixes-for-4.15-20171208' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Fri, 8 Dec 2017 19:53:54 +0000 (14:53 -0500)]
Merge tag 'linux-can-fixes-for-4.15-20171208' of git://git./linux/kernel/git/mkl/linux-can

Marc Kleine-Budde says:

====================
pull-request: can 2017-12-08

this is a pull request of 6 patches for net/master.

Martin Kelly provides 5 patches for various USB based CAN drivers, that
properly cancel the URBs on adapter unplug, so that the driver doesn't
end up in an endless loop. Stephane Grosjean provides a patch to restart
the tx queue if zero length packages are transmitted.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'wireless-drivers-for-davem-2017-12-08' of git://git.kernel.org/pub/scm...
David S. Miller [Fri, 8 Dec 2017 19:48:49 +0000 (14:48 -0500)]
Merge tag 'wireless-drivers-for-davem-2017-12-08' of git://git./linux/kernel/git/kvalo/wireless-drivers

Kalle Valo says:

====================
wireless-drivers fixes for 4.15

Second set of fixes for 4.15. This time a lot of iwlwifi patches and
two brcmfmac patches. Most important here are the MIC and IVC fixes
for iwlwifi to unbreak 9000 series.

iwlwifi

* fix rate-scaling to not start lowest possible rate

* fix the TX queue hang detection for AP/GO modes

* fix the TX queue hang timeout in monitor interfaces

* fix packet injection

* remove a wrong error message when dumping PCI registers

* fix race condition with RF-kill

* tell mac80211 when the MIC has been stripped (9000 series)

* tell mac80211 when the IVC has been stripped (9000 series)

* add 2 new PCI IDs, one for 9000 and one for 22000

* fix a queue hang due during a P2P Remain-on-Channel operation

brcmfmac

* fix a race which sometimes caused a crash during sdio unbind

* fix a kernel-doc related build error
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mvpp2: fix the RSS table entry offset
Antoine Tenart [Fri, 8 Dec 2017 09:24:20 +0000 (10:24 +0100)]
net: mvpp2: fix the RSS table entry offset

The macro used to access or set an RSS table entry was using an offset
of 8, while it should use an offset of 0. This lead to wrongly configure
the RSS table, not accessing the right entries.

Fixes: 1d7d15d79fb4 ("net: mvpp2: initialize the RSS tables")
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'tcp-RACK-loss-recovery-bug-fixes'
David S. Miller [Fri, 8 Dec 2017 19:14:12 +0000 (14:14 -0500)]
Merge branch 'tcp-RACK-loss-recovery-bug-fixes'

Yuchung Cheng says:

====================
tcp: RACK loss recovery bug fixes

This patch set has four minor bug fixes in TCP RACK loss recovery.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: evaluate packet losses upon RTT change
Yuchung Cheng [Thu, 7 Dec 2017 19:33:33 +0000 (11:33 -0800)]
tcp: evaluate packet losses upon RTT change

RACK skips an ACK unless it advances the most recently delivered
TX timestamp (rack.mstamp). Since RACK also uses the most recent
RTT to decide if a packet is lost, RACK should still run the
loss detection whenever the most recent RTT changes. For example,
an ACK that does not advance the timestamp but triggers the cwnd
undo due to reordering, would then use the most recent (higher)
RTT measurement to detect further losses.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: fix off-by-one bug in RACK
Yuchung Cheng [Thu, 7 Dec 2017 19:33:32 +0000 (11:33 -0800)]
tcp: fix off-by-one bug in RACK

RACK should mark a packet lost when remaining wait time is zero.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: always evaluate losses in RACK upon undo
Yuchung Cheng [Thu, 7 Dec 2017 19:33:31 +0000 (11:33 -0800)]
tcp: always evaluate losses in RACK upon undo

When sender detects spurious retransmission, all packets
marked lost are remarked to be in-flight. However some may
be considered lost based on its timestamps in RACK. This patch
forces RACK to re-evaluate, which may be skipped previously if
the ACK does not advance RACK timestamp.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: correctly test congestion state in RACK
Yuchung Cheng [Thu, 7 Dec 2017 19:33:30 +0000 (11:33 -0800)]
tcp: correctly test congestion state in RACK

RACK does not test the loss recovery state correctly to compute
the reordering window. It assumes if lost_out is zero then TCP is
not in loss recovery. But it can be zero during recovery before
calling tcp_rack_detect_loss(): when an ACK acknowledges all
packets marked lost before receiving this ACK, but has not yet
to discover new ones by tcp_rack_detect_loss(). The fix is to
simply test the congestion state directly.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Fix sources of spurious netpoll warnings
Calvin Owens [Fri, 8 Dec 2017 17:05:26 +0000 (09:05 -0800)]
bnxt_en: Fix sources of spurious netpoll warnings

After applying 2270bc5da3497945 ("bnxt_en: Fix netpoll handling") and
903649e718f80da2 ("bnxt_en: Improve -ENOMEM logic in NAPI poll loop."),
we still see the following WARN fire:

  ------------[ cut here ]------------
  WARNING: CPU: 0 PID: 1875170 at net/core/netpoll.c:165 netpoll_poll_dev+0x15a/0x160
  bnxt_poll+0x0/0xd0 exceeded budget in poll
  <snip>
  Call Trace:
   [<ffffffff814be5cd>] dump_stack+0x4d/0x70
   [<ffffffff8107e013>] __warn+0xd3/0xf0
   [<ffffffff8107e07f>] warn_slowpath_fmt+0x4f/0x60
   [<ffffffff8179519a>] netpoll_poll_dev+0x15a/0x160
   [<ffffffff81795f38>] netpoll_send_skb_on_dev+0x168/0x250
   [<ffffffff817962fc>] netpoll_send_udp+0x2dc/0x440
   [<ffffffff815fa9be>] write_ext_msg+0x20e/0x250
   [<ffffffff810c8125>] call_console_drivers.constprop.23+0xa5/0x110
   [<ffffffff810c9549>] console_unlock+0x339/0x5b0
   [<ffffffff810c9a88>] vprintk_emit+0x2c8/0x450
   [<ffffffff810c9d5f>] vprintk_default+0x1f/0x30
   [<ffffffff81173df5>] printk+0x48/0x50
   [<ffffffffa0197713>] edac_raw_mc_handle_error+0x563/0x5c0 [edac_core]
   [<ffffffffa0197b9b>] edac_mc_handle_error+0x42b/0x6e0 [edac_core]
   [<ffffffffa01c3a60>] sbridge_mce_output_error+0x410/0x10d0 [sb_edac]
   [<ffffffffa01c47cc>] sbridge_check_error+0xac/0x130 [sb_edac]
   [<ffffffffa0197f3c>] edac_mc_workq_function+0x3c/0x90 [edac_core]
   [<ffffffff81095f8b>] process_one_work+0x19b/0x480
   [<ffffffff810967ca>] worker_thread+0x6a/0x520
   [<ffffffff8109c7c4>] kthread+0xe4/0x100
   [<ffffffff81884c52>] ret_from_fork+0x22/0x40

This happens because we increment rx_pkts on -ENOMEM and -EIO, resulting
in rx_pkts > 0. Fix this by only bumping rx_pkts if we were actually
given a non-zero budget.

Signed-off-by: Calvin Owens <calvinowens@fb.com>
Acked-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoARM64: dts: meson-gx: fix UART pclk clock name
Neil Armstrong [Mon, 4 Dec 2017 09:04:53 +0000 (10:04 +0100)]
ARM64: dts: meson-gx: fix UART pclk clock name

The clock-names for pclk was wrongly set to "core", but the bindings
specifies "pclk".
This was not cathed until the legacy non-documented bindings were removed.

Reported-by: Andreas Färber <afaerber@suse.de>
Fixes: f72d6f6037b7 ("ARM64: dts: meson-gx: use stable UART bindings with correct gate clock")
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Kevin Hilman <khilman@baylibre.com>
7 years agoMerge branch 'tcp-bbr-sampling-fixes'
David S. Miller [Fri, 8 Dec 2017 18:27:44 +0000 (13:27 -0500)]
Merge branch 'tcp-bbr-sampling-fixes'

Neal Cardwell says:

====================
TCP BBR sampling fixes for loss recovery undo

This patch series has a few minor bug fixes for cases where spurious
loss recoveries can trick BBR estimators into estimating that the
available bandwidth is much lower than the true available bandwidth.
In both cases the fix here is to just reset the estimator upon loss
recovery undo.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp_bbr: reset long-term bandwidth sampling on loss recovery undo
Neal Cardwell [Thu, 7 Dec 2017 17:43:32 +0000 (12:43 -0500)]
tcp_bbr: reset long-term bandwidth sampling on loss recovery undo

Fix BBR so that upon notification of a loss recovery undo BBR resets
long-term bandwidth sampling.

Under high reordering, reordering events can be interpreted as loss.
If the reordering and spurious loss estimates are high enough, this
can cause BBR to spuriously estimate that we are seeing loss rates
high enough to trigger long-term bandwidth estimation. To avoid that
problem, this commit resets long-term bandwidth sampling on loss
recovery undo events.

Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp_bbr: reset full pipe detection on loss recovery undo
Neal Cardwell [Thu, 7 Dec 2017 17:43:31 +0000 (12:43 -0500)]
tcp_bbr: reset full pipe detection on loss recovery undo

Fix BBR so that upon notification of a loss recovery undo BBR resets
the full pipe detection (STARTUP exit) state machine.

Under high reordering, reordering events can be interpreted as loss.
If the reordering and spurious loss estimates are high enough, this
could previously cause BBR to spuriously estimate that the pipe is
full.

Since spurious loss recovery means that our overall sending will have
slowed down spuriously, this commit gives a flow more time to probe
robustly for bandwidth and decide the pipe is really full.

Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp_bbr: record "full bw reached" decision in new full_bw_reached bit
Neal Cardwell [Thu, 7 Dec 2017 17:43:30 +0000 (12:43 -0500)]
tcp_bbr: record "full bw reached" decision in new full_bw_reached bit

This commit records the "full bw reached" decision in a new
full_bw_reached bit. This is a pure refactor that does not change the
current behavior, but enables subsequent fixes and improvements.

In particular, this enables simple and clean fixes because the full_bw
and full_bw_cnt can be unconditionally zeroed without worrying about
forgetting that we estimated we filled the pipe in Startup. And it
enables future improvements because multiple code paths can be used
for estimating that we filled the pipe in Startup; any new code paths
only need to set this bit when they think the pipe is full.

Note that this fix intentionally reduces the width of the full_bw_cnt
counter, since we have never used the most significant bit.

Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosfc: pass valid pointers from efx_enqueue_unwind
Bert Kenward [Thu, 7 Dec 2017 17:18:58 +0000 (17:18 +0000)]
sfc: pass valid pointers from efx_enqueue_unwind

The bytes_compl and pkts_compl pointers passed to efx_dequeue_buffers
cannot be NULL. Add a paranoid warning to check this condition and fix
the one case where they were NULL.

efx_enqueue_unwind() is called very rarely, during error handling.
Without this fix it would fail with a NULL pointer dereference in
efx_dequeue_buffer, with efx_enqueue_skb in the call stack.

Fixes: e9117e5099ea ("sfc: Firmware-Assisted TSO version 2")
Reported-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Tested-by: Jarod Wilson <jarod@redhat.com>
Acked-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agogianfar: Disable EEE autoneg by default
Claudiu Manoil [Thu, 7 Dec 2017 16:44:23 +0000 (18:44 +0200)]
gianfar: Disable EEE autoneg by default

This controller does not support EEE, but it may connect to a PHY
which supports EEE and advertises EEE by default, while its link
partner also advertises EEE. If this happens, the PHY enters low
power mode when the traffic rate is low and causes packet loss.
This patch disables EEE advertisement by default for any PHY that
gianfar connects to, to prevent the above unwanted outcome.

Signed-off-by: Shaohui Xie <Shaohui.Xie@nxp.com>
Tested-by: Yangbo Lu <Yangbo.lu@nxp.com>
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Fri, 8 Dec 2017 18:10:17 +0000 (10:10 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/s390/linux

Pull s390 fixes from Martin Schwidefsky:

 - three more patches in regard to the SPDX license tags. The missing
   tags for the files in arch/s390/kvm will be merged via the KVM tree.
   With that all s390 related files should have their SPDX tags.

 - a patch to get rid of 'struct timespec' in the DASD driver.

 - bug fixes

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390: fix compat system call table
  s390/mm: fix off-by-one bug in 5-level page table handling
  s390: Remove redudant license text
  s390: add a few more SPDX identifiers
  s390/dasd: prevent prefix I/O error
  s390: always save and restore all registers on context switch
  s390/dasd: remove 'struct timespec' usage
  s390/qdio: restrict target-full handling to IQDIO
  s390/qdio: consider ERROR buffers for inbound-full condition
  s390/virtio: add BSD license to virtio-ccw

7 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Fri, 8 Dec 2017 18:08:23 +0000 (10:08 -0800)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fixes from Will Deacon:
 "Fix some more FP register fallout from the SVE patches and also some
  problems with the PGD tracking in our software PAN emulation code,
  after we received a crash report from a 3.18 kernel running a
  backport.

  Summary:

   - fix SW PAN pgd shadowing for kernel threads, EFI and exiting user
     tasks

   - fix FP register leak when a task_struct is re-allocated

   - fix potential use-after-free in FP state tracking used by KVM"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64/sve: Avoid dereference of dead task_struct in KVM guest entry
  arm64: SW PAN: Update saved ttbr0 value on enter_lazy_tlb
  arm64: SW PAN: Point saved ttbr0 at the zero page when switching to init_mm
  arm64: fpsimd: Abstract out binding of task's fpsimd context to the cpu.
  arm64: fpsimd: Prevent registers leaking from dead tasks

7 years agoMerge tag 'acpi-4.15-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Fri, 8 Dec 2017 18:05:53 +0000 (10:05 -0800)]
Merge tag 'acpi-4.15-rc3' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI fix from Rafael Wysocki:
 "This fixes an out of bounds warning from KASAN in the ACPI CPPC
  driver"

* tag 'acpi-4.15-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI / CPPC: Fix KASAN global out of bounds warning

7 years agoMerge tag 'pm-4.15-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Fri, 8 Dec 2017 17:58:10 +0000 (09:58 -0800)]
Merge tag 'pm-4.15-rc3' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fix from Rafael Wysocki:
 "This fixes an issue in the device runtime PM framework that prevents
  customer devices from resuming if runtime PM is disabled for one or
  more of their supplier devices (as reflected by device links between
  those devices)"

* tag 'pm-4.15-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  PM / runtime: Fix handling of suppliers with disabled runtime PM

7 years agoof: overlay: Make node skipping in init_overlay_changeset() clearer
Geert Uytterhoeven [Fri, 8 Dec 2017 13:13:03 +0000 (14:13 +0100)]
of: overlay: Make node skipping in init_overlay_changeset() clearer

Make it more clear that nodes without "__overlay__" subnodes are
skipped, by reverting the logic and using continue.
This also reduces indentation level.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Rob Herring <robh@kernel.org>
7 years agoof: overlay: Fix out-of-bounds write in init_overlay_changeset()
Geert Uytterhoeven [Fri, 8 Dec 2017 13:13:02 +0000 (14:13 +0100)]
of: overlay: Fix out-of-bounds write in init_overlay_changeset()

If an overlay has no "__symbols__" node, but it has nodes without
"__overlay__" subnodes at the end (e.g. a "__fixups__" node), after
filling in all fragments for nodes with "__overlay__" subnodes,
"fragment = &fragments[cnt]" will point beyond the end of the allocated
array.

Hence writing to "fragment->overlay" will overwrite unallocated memory,
which may lead to a crash later.

Fix this by deferring both the assignment to "fragment" and the
offending write afterwards until we know for sure the node has an
"__overlay__" subnode, and thus a valid entry in "fragments[]".

Fixes: 61b4de4e0b384f4a ("of: overlay: minor restructuring")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Rob Herring <robh@kernel.org>
7 years agoX.509: fix comparisons of ->pkey_algo
Eric Biggers [Fri, 8 Dec 2017 15:13:29 +0000 (15:13 +0000)]
X.509: fix comparisons of ->pkey_algo

->pkey_algo used to be an enum, but was changed to a string by commit
4e8ae72a75aa ("X.509: Make algo identifiers text instead of enum").  But
two comparisons were not updated.  Fix them to use strcmp().

This bug broke signature verification in certain configurations,
depending on whether the string constants were deduplicated or not.

Fixes: 4e8ae72a75aa ("X.509: Make algo identifiers text instead of enum")
Cc: <stable@vger.kernel.org> # v4.6+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agoKEYS: reject NULL restriction string when type is specified
Eric Biggers [Fri, 8 Dec 2017 15:13:29 +0000 (15:13 +0000)]
KEYS: reject NULL restriction string when type is specified

keyctl_restrict_keyring() allows through a NULL restriction when the
"type" is non-NULL, which causes a NULL pointer dereference in
asymmetric_lookup_restriction() when it calls strcmp() on the
restriction string.

But no key types actually use a "NULL restriction" to mean anything, so
update keyctl_restrict_keyring() to reject it with EINVAL.

Reported-by: syzbot <syzkaller@googlegroups.com>
Fixes: 97d3aa0f3134 ("KEYS: Add a lookup_restriction function for the asymmetric key type")
Cc: <stable@vger.kernel.org> # v4.12+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agosecurity: keys: remove redundant assignment to key_ref
Colin Ian King [Fri, 8 Dec 2017 15:13:29 +0000 (15:13 +0000)]
security: keys: remove redundant assignment to key_ref

Variable key_ref is being assigned a value that is never read;
key_ref is being re-assigned a few statements later.  Hence this
assignment is redundant and can be removed.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
7 years agoX.509: use crypto_shash_digest()
Eric Biggers [Fri, 8 Dec 2017 15:13:29 +0000 (15:13 +0000)]
X.509: use crypto_shash_digest()

Use crypto_shash_digest() instead of crypto_shash_init() followed by
crypto_shash_finup().  (For simplicity only; they are equivalent.)

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agoKEYS: be careful with error codes in public_key_verify_signature()
Eric Biggers [Fri, 8 Dec 2017 15:13:29 +0000 (15:13 +0000)]
KEYS: be careful with error codes in public_key_verify_signature()

In public_key_verify_signature(), if akcipher_request_alloc() fails, we
return -ENOMEM.  But that error code was set 25 lines above, and by
accident someone could easily insert new code in between that assigns to
'ret', which would introduce a signature verification bypass.  Make the
code clearer by moving the -ENOMEM down to where it is used.

Additionally, the callers of public_key_verify_signature() only consider
a negative return value to be an error.  This means that if any positive
return value is accidentally introduced deeper in the call stack (e.g.
'return EBADMSG' instead of 'return -EBADMSG' somewhere in RSA),
signature verification will be bypassed.  Make things more robust by
having public_key_verify_signature() warn about positive errors and
translate them into -EINVAL.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agopkcs7: use crypto_shash_digest()
Eric Biggers [Fri, 8 Dec 2017 15:13:28 +0000 (15:13 +0000)]
pkcs7: use crypto_shash_digest()

Use crypto_shash_digest() instead of crypto_shash_init() followed by
crypto_shash_finup().  (For simplicity only; they are equivalent.)

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agopkcs7: fix check for self-signed certificate
Eric Biggers [Fri, 8 Dec 2017 15:13:28 +0000 (15:13 +0000)]
pkcs7: fix check for self-signed certificate

pkcs7_validate_trust_one() used 'x509->next == x509' to identify a
self-signed certificate.  That's wrong; ->next is simply the link in the
linked list of certificates in the PKCS#7 message.  It should be
checking ->signer instead.  Fix it.

Fortunately this didn't actually matter because when we re-visited
'x509' on the next iteration via 'x509->signer', it was already seen and
not verified, so we returned -ENOKEY anyway.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
7 years agopkcs7: return correct error code if pkcs7_check_authattrs() fails
Eric Biggers [Fri, 8 Dec 2017 15:13:28 +0000 (15:13 +0000)]
pkcs7: return correct error code if pkcs7_check_authattrs() fails

If pkcs7_check_authattrs() returns an error code, we should pass that
error code on, rather than using ENOMEM.

Fixes: 99db44350672 ("PKCS#7: Appropriately restrict authenticated attributes and content type")
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
7 years ago509: fix printing uninitialized stack memory when OID is empty
Eric Biggers [Fri, 8 Dec 2017 15:13:28 +0000 (15:13 +0000)]
509: fix printing uninitialized stack memory when OID is empty

Callers of sprint_oid() do not check its return value before printing
the result.  In the case where the OID is zero-length, -EBADMSG was
being returned without anything being written to the buffer, resulting
in uninitialized stack memory being printed.  Fix this by writing
"(bad)" to the buffer in the cases where -EBADMSG is returned.

Fixes: 4f73175d0375 ("X.509: Add utility functions to render OIDs as strings")
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agoX.509: fix buffer overflow detection in sprint_oid()
Eric Biggers [Fri, 8 Dec 2017 15:13:28 +0000 (15:13 +0000)]
X.509: fix buffer overflow detection in sprint_oid()

In sprint_oid(), if the input buffer were to be more than 1 byte too
small for the first snprintf(), 'bufsize' would underflow, causing a
buffer overflow when printing the remainder of the OID.

Fortunately this cannot actually happen currently, because no users pass
in a buffer that can be too small for the first snprintf().

Regardless, fix it by checking the snprintf() return value correctly.

For consistency also tweak the second snprintf() check to look the same.

Fixes: 4f73175d0375 ("X.509: Add utility functions to render OIDs as strings")
Cc: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
7 years agoX.509: reject invalid BIT STRING for subjectPublicKey
Eric Biggers [Fri, 8 Dec 2017 15:13:27 +0000 (15:13 +0000)]
X.509: reject invalid BIT STRING for subjectPublicKey

Adding a specially crafted X.509 certificate whose subjectPublicKey
ASN.1 value is zero-length caused x509_extract_key_data() to set the
public key size to SIZE_MAX, as it subtracted the nonexistent BIT STRING
metadata byte.  Then, x509_cert_parse() called kmemdup() with that bogus
size, triggering the WARN_ON_ONCE() in kmalloc_slab().

This appears to be harmless, but it still must be fixed since WARNs are
never supposed to be user-triggerable.

Fix it by updating x509_cert_parse() to validate that the value has a
BIT STRING metadata byte, and that the byte is 0 which indicates that
the number of bits in the bitstring is a multiple of 8.

It would be nice to handle the metadata byte in asn1_ber_decoder()
instead.  But that would be tricky because in the general case a BIT
STRING could be implicitly tagged, and/or could legitimately have a
length that is not a whole number of bytes.

Here was the WARN (cleaned up slightly):

    WARNING: CPU: 1 PID: 202 at mm/slab_common.c:971 kmalloc_slab+0x5d/0x70 mm/slab_common.c:971
    Modules linked in:
    CPU: 1 PID: 202 Comm: keyctl Tainted: G    B            4.14.0-09238-g1d3b78bbc6e9 #26
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-20171110_100015-anatol 04/01/2014
    task: ffff880033014180 task.stack: ffff8800305c8000
    Call Trace:
     __do_kmalloc mm/slab.c:3706 [inline]
     __kmalloc_track_caller+0x22/0x2e0 mm/slab.c:3726
     kmemdup+0x17/0x40 mm/util.c:118
     kmemdup include/linux/string.h:414 [inline]
     x509_cert_parse+0x2cb/0x620 crypto/asymmetric_keys/x509_cert_parser.c:106
     x509_key_preparse+0x61/0x750 crypto/asymmetric_keys/x509_public_key.c:174
     asymmetric_key_preparse+0xa4/0x150 crypto/asymmetric_keys/asymmetric_type.c:388
     key_create_or_update+0x4d4/0x10a0 security/keys/key.c:850
     SYSC_add_key security/keys/keyctl.c:122 [inline]
     SyS_add_key+0xe8/0x290 security/keys/keyctl.c:62
     entry_SYSCALL_64_fastpath+0x1f/0x96

Fixes: 42d5ec27f873 ("X.509: Add an ASN.1 decoder")
Cc: <stable@vger.kernel.org> # v3.7+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
7 years agoASN.1: check for error from ASN1_OP_END__ACT actions
Eric Biggers [Fri, 8 Dec 2017 15:13:27 +0000 (15:13 +0000)]
ASN.1: check for error from ASN1_OP_END__ACT actions

asn1_ber_decoder() was ignoring errors from actions associated with the
opcodes ASN1_OP_END_SEQ_ACT, ASN1_OP_END_SET_ACT,
ASN1_OP_END_SEQ_OF_ACT, and ASN1_OP_END_SET_OF_ACT.  In practice, this
meant the pkcs7_note_signed_info() action (since that was the only user
of those opcodes).  Fix it by checking for the error, just like the
decoder does for actions associated with the other opcodes.

This bug allowed users to leak slab memory by repeatedly trying to add a
specially crafted "pkcs7_test" key (requires CONFIG_PKCS7_TEST_KEY).

In theory, this bug could also be used to bypass module signature
verification, by providing a PKCS#7 message that is misparsed such that
a signature's ->authattrs do not contain its ->msgdigest.  But it
doesn't seem practical in normal cases, due to restrictions on the
format of the ->authattrs.

Fixes: 42d5ec27f873 ("X.509: Add an ASN.1 decoder")
Cc: <stable@vger.kernel.org> # v3.7+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
7 years agoASN.1: fix out-of-bounds read when parsing indefinite length item
Eric Biggers [Fri, 8 Dec 2017 15:13:27 +0000 (15:13 +0000)]
ASN.1: fix out-of-bounds read when parsing indefinite length item

In asn1_ber_decoder(), indefinitely-sized ASN.1 items were being passed
to the action functions before their lengths had been computed, using
the bogus length of 0x80 (ASN1_INDEFINITE_LENGTH).  This resulted in
reading data past the end of the input buffer, when given a specially
crafted message.

Fix it by rearranging the code so that the indefinite length is resolved
before the action is called.

This bug was originally found by fuzzing the X.509 parser in userspace
using libFuzzer from the LLVM project.

KASAN report (cleaned up slightly):

    BUG: KASAN: slab-out-of-bounds in memcpy ./include/linux/string.h:341 [inline]
    BUG: KASAN: slab-out-of-bounds in x509_fabricate_name.constprop.1+0x1a4/0x940 crypto/asymmetric_keys/x509_cert_parser.c:366
    Read of size 128 at addr ffff880035dd9eaf by task keyctl/195

    CPU: 1 PID: 195 Comm: keyctl Not tainted 4.14.0-09238-g1d3b78bbc6e9 #26
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-20171110_100015-anatol 04/01/2014
    Call Trace:
     __dump_stack lib/dump_stack.c:17 [inline]
     dump_stack+0xd1/0x175 lib/dump_stack.c:53
     print_address_description+0x78/0x260 mm/kasan/report.c:252
     kasan_report_error mm/kasan/report.c:351 [inline]
     kasan_report+0x23f/0x350 mm/kasan/report.c:409
     memcpy+0x1f/0x50 mm/kasan/kasan.c:302
     memcpy ./include/linux/string.h:341 [inline]
     x509_fabricate_name.constprop.1+0x1a4/0x940 crypto/asymmetric_keys/x509_cert_parser.c:366
     asn1_ber_decoder+0xb4a/0x1fd0 lib/asn1_decoder.c:447
     x509_cert_parse+0x1c7/0x620 crypto/asymmetric_keys/x509_cert_parser.c:89
     x509_key_preparse+0x61/0x750 crypto/asymmetric_keys/x509_public_key.c:174
     asymmetric_key_preparse+0xa4/0x150 crypto/asymmetric_keys/asymmetric_type.c:388
     key_create_or_update+0x4d4/0x10a0 security/keys/key.c:850
     SYSC_add_key security/keys/keyctl.c:122 [inline]
     SyS_add_key+0xe8/0x290 security/keys/keyctl.c:62
     entry_SYSCALL_64_fastpath+0x1f/0x96

    Allocated by task 195:
     __do_kmalloc_node mm/slab.c:3675 [inline]
     __kmalloc_node+0x47/0x60 mm/slab.c:3682
     kvmalloc ./include/linux/mm.h:540 [inline]
     SYSC_add_key security/keys/keyctl.c:104 [inline]
     SyS_add_key+0x19e/0x290 security/keys/keyctl.c:62
     entry_SYSCALL_64_fastpath+0x1f/0x96

Fixes: 42d5ec27f873 ("X.509: Add an ASN.1 decoder")
Reported-by: Alexander Potapenko <glider@google.com>
Cc: <stable@vger.kernel.org> # v3.7+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agoKEYS: add missing permission check for request_key() destination
Eric Biggers [Fri, 8 Dec 2017 15:13:27 +0000 (15:13 +0000)]
KEYS: add missing permission check for request_key() destination

When the request_key() syscall is not passed a destination keyring, it
links the requested key (if constructed) into the "default" request-key
keyring.  This should require Write permission to the keyring.  However,
there is actually no permission check.

This can be abused to add keys to any keyring to which only Search
permission is granted.  This is because Search permission allows joining
the keyring.  keyctl_set_reqkey_keyring(KEY_REQKEY_DEFL_SESSION_KEYRING)
then will set the default request-key keyring to the session keyring.
Then, request_key() can be used to add keys to the keyring.

Both negatively and positively instantiated keys can be added using this
method.  Adding negative keys is trivial.  Adding a positive key is a
bit trickier.  It requires that either /sbin/request-key positively
instantiates the key, or that another thread adds the key to the process
keyring at just the right time, such that request_key() misses it
initially but then finds it in construct_alloc_key().

Fix this bug by checking for Write permission to the keyring in
construct_get_dest_keyring() when the default keyring is being used.

We don't do the permission check for non-default keyrings because that
was already done by the earlier call to lookup_user_key().  Also,
request_key_and_link() is currently passed a 'struct key *' rather than
a key_ref_t, so the "possessed" bit is unavailable.

We also don't do the permission check for the "requestor keyring", to
continue to support the use case described by commit 8bbf4976b59f
("KEYS: Alter use of key instantiation link-to-keyring argument") where
/sbin/request-key recursively calls request_key() to add keys to the
original requestor's destination keyring.  (I don't know of any users
who actually do that, though...)

Fixes: 3e30148c3d52 ("[PATCH] Keys: Make request-key create an authorisation key")
Cc: <stable@vger.kernel.org> # v2.6.13+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agoKEYS: remove unnecessary get/put of explicit dest_keyring
Eric Biggers [Fri, 8 Dec 2017 15:13:27 +0000 (15:13 +0000)]
KEYS: remove unnecessary get/put of explicit dest_keyring

In request_key_and_link(), in the case where the dest_keyring was
explicitly specified, there is no need to get another reference to
dest_keyring before calling key_link(), then drop it afterwards.  This
is because by definition, we already have a reference to dest_keyring.

This change is useful because we'll be making
construct_get_dest_keyring() able to return an error code, and we don't
want to have to handle that error here for no reason.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
7 years agotcp: invalidate rate samples during SACK reneging
Yousuk Seung [Thu, 7 Dec 2017 21:41:34 +0000 (13:41 -0800)]
tcp: invalidate rate samples during SACK reneging

Mark tcp_sock during a SACK reneging event and invalidate rate samples
while marked. Such rate samples may overestimate bw by including packets
that were SACKed before reneging.

< ack 6001 win 10000 sack 7001:38001
< ack 7001 win 0 sack 8001:38001 // Reneg detected
> seq 7001:8001 // RTO, SACK cleared.
< ack 38001 win 10000

In above example the rate sample taken after the last ack will count
7001-38001 as delivered while the actual delivery rate likely could
be much lower i.e. 7001-8001.

This patch adds a new field tcp_sock.sack_reneg and marks it when we
declare SACK reneging and entering TCP_CA_Loss, and unmarks it after
the last rate sample was taken before moving back to TCP_CA_Open. This
patch also invalidates rate samples taken while tcp_sock.is_sack_reneg
is set.

Fixes: b9f64820fb22 ("tcp: track data delivery rate for a TCP connection")
Signed-off-by: Yousuk Seung <ysseung@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocan: peak/pcie_fd: fix potential bug in restarting tx queue
Stephane Grosjean [Thu, 7 Dec 2017 15:13:43 +0000 (16:13 +0100)]
can: peak/pcie_fd: fix potential bug in restarting tx queue

Don't rely on can_get_echo_skb() return value to wake the network tx
queue up: can_get_echo_skb() returns 0 if the echo array slot was not
occupied, but also when the DLC of the released echo frame was 0.

Signed-off-by: Stephane Grosjean <s.grosjean@peak-system.com>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: usb_8dev: cancel urb on -EPIPE and -EPROTO
Martin Kelly [Tue, 5 Dec 2017 19:15:50 +0000 (11:15 -0800)]
can: usb_8dev: cancel urb on -EPIPE and -EPROTO

In mcba_usb, we have observed that when you unplug the device, the driver will
endlessly resubmit failing URBs, which can cause CPU stalls. This issue
is fixed in mcba_usb by catching the codes seen on device disconnect
(-EPIPE and -EPROTO).

This driver also resubmits in the case of -EPIPE and -EPROTO, so fix it
in the same way.

Signed-off-by: Martin Kelly <mkelly@xevo.com>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: kvaser_usb: cancel urb on -EPIPE and -EPROTO
Martin Kelly [Tue, 5 Dec 2017 19:15:49 +0000 (11:15 -0800)]
can: kvaser_usb: cancel urb on -EPIPE and -EPROTO

In mcba_usb, we have observed that when you unplug the device, the driver will
endlessly resubmit failing URBs, which can cause CPU stalls. This issue
is fixed in mcba_usb by catching the codes seen on device disconnect
(-EPIPE and -EPROTO).

This driver also resubmits in the case of -EPIPE and -EPROTO, so fix it
in the same way.

Signed-off-by: Martin Kelly <mkelly@xevo.com>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: esd_usb2: cancel urb on -EPIPE and -EPROTO
Martin Kelly [Tue, 5 Dec 2017 19:15:48 +0000 (11:15 -0800)]
can: esd_usb2: cancel urb on -EPIPE and -EPROTO

In mcba_usb, we have observed that when you unplug the device, the driver will
endlessly resubmit failing URBs, which can cause CPU stalls. This issue
is fixed in mcba_usb by catching the codes seen on device disconnect
(-EPIPE and -EPROTO).

This driver also resubmits in the case of -EPIPE and -EPROTO, so fix it
in the same way.

Signed-off-by: Martin Kelly <mkelly@xevo.com>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: ems_usb: cancel urb on -EPIPE and -EPROTO
Martin Kelly [Tue, 5 Dec 2017 19:15:47 +0000 (11:15 -0800)]
can: ems_usb: cancel urb on -EPIPE and -EPROTO

In mcba_usb, we have observed that when you unplug the device, the driver will
endlessly resubmit failing URBs, which can cause CPU stalls. This issue
is fixed in mcba_usb by catching the codes seen on device disconnect
(-EPIPE and -EPROTO).

This driver also resubmits in the case of -EPIPE and -EPROTO, so fix it
in the same way.

Signed-off-by: Martin Kelly <mkelly@xevo.com>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>