openwrt/staging/blogic.git
5 years agopowerpc/powernv: Fix build with IOMMU_API=n
Michael Ellerman [Fri, 13 Sep 2019 13:50:06 +0000 (23:50 +1000)]
powerpc/powernv: Fix build with IOMMU_API=n

The builds breaks when IOMMU_API=n, eg. skiroot_defconfig:

  arch/powerpc/platforms/powernv/npu-dma.c:96:28: error: 'get_gpu_pci_dev_and_pe' defined but not used
  arch/powerpc/platforms/powernv/npu-dma.c:126:13: error: 'pnv_npu_set_window' defined but not used

Fixes: b4d37a7b6934 ("powerpc/powernv: Remove unused pnv_npu_try_dma_set_bypass() function")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/eeh: Fix build with STACKTRACE=n
Michael Ellerman [Fri, 13 Sep 2019 13:32:13 +0000 (23:32 +1000)]
powerpc/eeh: Fix build with STACKTRACE=n

The build breaks when STACKTRACE=n, eg. skiroot_defconfig:

  arch/powerpc/kernel/eeh_event.c:124:23: error: implicit declaration of function 'stack_trace_save'

Fix it with some ifdefs for now.

Fixes: 25baf3d81614 ("powerpc/eeh: Defer printing stack trace")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/xive: Fix bogus error code returned by OPAL
Greg Kurz [Wed, 11 Sep 2019 15:52:18 +0000 (17:52 +0200)]
powerpc/xive: Fix bogus error code returned by OPAL

There's a bug in skiboot that causes the OPAL_XIVE_ALLOCATE_IRQ call
to return the 32-bit value 0xffffffff when OPAL has run out of IRQs.
Unfortunatelty, OPAL return values are signed 64-bit entities and
errors are supposed to be negative. If that happens, the linux code
confusingly treats 0xffffffff as a valid IRQ number and panics at some
point.

A fix was recently merged in skiboot:

e97391ae2bb5 ("xive: fix return value of opal_xive_allocate_irq()")

but we need a workaround anyway to support older skiboots already
in the field.

Internally convert 0xffffffff to OPAL_RESOURCE which is the usual error
returned upon resource exhaustion.

Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/156821713818.1985334.14123187368108582810.stgit@bahia.lan
5 years agopowerpc/pseries: correctly track irq state in default idle
Nathan Lynch [Tue, 10 Sep 2019 22:52:44 +0000 (17:52 -0500)]
powerpc/pseries: correctly track irq state in default idle

prep_irq_for_idle() is intended to be called before entering
H_CEDE (and it is used by the pseries cpuidle driver). However the
default pseries idle routine does not call it, leading to mismanaged
lazy irq state when the cpuidle driver isn't in use. Manifestations of
this include:

* Dropped IPIs in the time immediately after a cpu comes
  online (before it has installed the cpuidle handler), making the
  online operation block indefinitely waiting for the new cpu to
  respond.

* Hitting this WARN_ON in arch_local_irq_restore():
/*
 * We should already be hard disabled here. We had bugs
 * where that wasn't the case so let's dbl check it and
 * warn if we are wrong. Only do that when IRQ tracing
 * is enabled as mfmsr() can be costly.
 */
if (WARN_ON_ONCE(mfmsr() & MSR_EE))
__hard_irq_disable();

Call prep_irq_for_idle() from pseries_lpar_idle() and honor its
result.

Fixes: 363edbe2614a ("powerpc: Default arch idle could cede processor on pseries")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190910225244.25056-1-nathanl@linux.ibm.com
5 years agopowerpc/watchpoint: Disable watchpoint hit by larx/stcx instructions
Ravi Bangoria [Tue, 10 Sep 2019 13:15:13 +0000 (18:45 +0530)]
powerpc/watchpoint: Disable watchpoint hit by larx/stcx instructions

If watchpoint exception is generated by larx/stcx instructions, the
reservation created by larx gets lost while handling exception, and
thus stcx instruction always fails. Generally these instructions are
used in a while(1) loop, for example spinlocks. And because stcx
never succeeds, it loops forever and ultimately hangs the system.

Note that ptrace anyway works in one-shot mode and thus for ptrace
we don't change the behaviour. It's up to ptrace user to take care
of this.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Acked-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190910131513.30499-1-ravi.bangoria@linux.ibm.com
5 years agopowerpc/powernv: Add new opal message type
Vasant Hegde [Mon, 26 Aug 2019 06:57:01 +0000 (12:27 +0530)]
powerpc/powernv: Add new opal message type

We have OPAL_MSG_PRD message type to pass prd related messages from
OPAL to `opal-prd`. It can handle messages upto 64 bytes. We have a
requirement to send bigger than 64 bytes of data from OPAL to
`opal-prd`. Lets add new message type (OPAL_MSG_PRD2) to pass bigger
data.

Signed-off-by: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>
[mpe: Make the error string clear that it's the PRD2 event that failed]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190826065701.8853-2-hegdevasant@linux.vnet.ibm.com
5 years agopowerpc/powernv: Enhance opal message read interface
Vasant Hegde [Mon, 26 Aug 2019 06:57:00 +0000 (12:27 +0530)]
powerpc/powernv: Enhance opal message read interface

Use "opal-msg-size" device tree property to allocate memory for
"opal_msg".

Signed-off-by: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>
[mpe: s/uint32_t/u32/ and mark opal_msg_size as __ro_after_init]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190826065701.8853-1-hegdevasant@linux.vnet.ibm.com
5 years agopowerpc/powernv: Remove unused pnv_npu_try_dma_set_bypass() function
Christoph Hellwig [Tue, 3 Sep 2019 16:51:47 +0000 (18:51 +0200)]
powerpc/powernv: Remove unused pnv_npu_try_dma_set_bypass() function

Neither pnv_npu_try_dma_set_bypass() nor the pnv_npu_dma_set_32() and
pnv_npu_dma_set_bypass() helpers called by it are used anywhere in the
kernel tree, so remove them.

mpe: They're unused since 2d6ad41b2c21 ("powerpc/powernv: use the
generic iommu bypass code") removed the last usage.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903165147.11099-1-hch@lst.de
5 years agoseltests/powerpc: Add a selftest for memcpy_mcsafe
Santosh Sivaraj [Tue, 3 Sep 2019 21:43:59 +0000 (03:13 +0530)]
seltests/powerpc: Add a selftest for memcpy_mcsafe

Appropriate self tests for memcpy_mcsafe

Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903214359.23887-2-santosh@fossix.org
5 years agopowerpc/memcpy: Fix stack corruption for smaller sizes
Santosh Sivaraj [Tue, 3 Sep 2019 21:43:58 +0000 (03:13 +0530)]
powerpc/memcpy: Fix stack corruption for smaller sizes

For sizes lesser than 128 bytes, the code branches out early without saving
the stack frame, which when restored later drops frame of the caller.

Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903214359.23887-1-santosh@fossix.org
5 years agopowerpc: Add attributes for setjmp/longjmp
Segher Boessenkool [Wed, 4 Sep 2019 14:11:07 +0000 (14:11 +0000)]
powerpc: Add attributes for setjmp/longjmp

The setjmp function should be declared as "returns_twice", or bad
things can happen[1]. This does not actually change generated code in
my testing.

The longjmp function should be declared as "noreturn", so that the
compiler can optimise calls to it better. This makes the generated
code a little shorter.

1: https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#index-returns_005ftwice-function-attribute

Signed-off-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c02ce4a573f3bac907e2c70957a2d1275f910013.1567605586.git.segher@kernel.crashing.org
5 years agopowerpc: Remove empty comment
Jordan Niethe [Tue, 13 Aug 2019 05:12:12 +0000 (15:12 +1000)]
powerpc: Remove empty comment

Commit 2874c5fd2842 ("treewide: Replace GPLv2 boilerplate/reference with
SPDX - rule 152") left an empty comment in machdep.h, as the boilerplate
was the only text in the comment. Remove the empty comment.

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190813051212.6387-1-jniethe5@gmail.com
5 years agopowerpc/imc: Dont create debugfs files for cpu-less nodes
Madhavan Srinivasan [Tue, 27 Aug 2019 10:16:35 +0000 (15:46 +0530)]
powerpc/imc: Dont create debugfs files for cpu-less nodes

Commit <684d984038aa> ('powerpc/powernv: Add debugfs interface for
imc-mode and imc') added debugfs interface for the nest imc pmu
devices to support changing of different ucode modes. Primarily adding
this capability for debug. But when doing so, the code did not
consider the case of cpu-less nodes. So when reading the _cmd_ or
_mode_ file of a cpu-less node will create this crash.

  Faulting instruction address: 0xc0000000000d0d58
  Oops: Kernel access of bad area, sig: 11 [#1]
  ...
  CPU: 67 PID: 5301 Comm: cat Not tainted 5.2.0-rc6-next-20190627+ #19
  NIP:  c0000000000d0d58 LR: c00000000049aa18 CTR:c0000000000d0d50
  REGS: c00020194548f9e0 TRAP: 0300   Not tainted  (5.2.0-rc6-next-20190627+)
  MSR:  9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE>  CR:28022822  XER: 00000000
  CFAR: c00000000049aa14 DAR: 000000000003fc08 DSISR:40000000 IRQMASK: 0
  ...
  NIP imc_mem_get+0x8/0x20
  LR  simple_attr_read+0x118/0x170
  Call Trace:
    simple_attr_read+0x70/0x170 (unreliable)
    debugfs_attr_read+0x6c/0xb0
    __vfs_read+0x3c/0x70
     vfs_read+0xbc/0x1a0
    ksys_read+0x7c/0x140
    system_call+0x5c/0x70

Patch fixes the issue with a more robust check for vbase to NULL.

Before patch, ls output for the debugfs imc directory

  # ls /sys/kernel/debug/powerpc/imc/
  imc_cmd_0    imc_cmd_251  imc_cmd_253  imc_cmd_255  imc_mode_0    imc_mode_251  imc_mode_253  imc_mode_255
  imc_cmd_250  imc_cmd_252  imc_cmd_254  imc_cmd_8    imc_mode_250  imc_mode_252  imc_mode_254  imc_mode_8

After patch, ls output for the debugfs imc directory

  # ls /sys/kernel/debug/powerpc/imc/
  imc_cmd_0  imc_cmd_8  imc_mode_0  imc_mode_8

Actual bug here is that, we have two loops with potentially different
loop counts. That is, in imc_get_mem_addr_nest(), loop count is
obtained from the dt entries. But in case of export_imc_mode_and_cmd(),
loop was based on for_each_nid() count. Patch fixes the loop count in
latter based on the struct mem_info. Ideally it would be better to
have array size in struct imc_pmu.

Fixes: 684d984038aa ('powerpc/powernv: Add debugfs interface for imc-mode and imc')
Reported-by: Qian Cai <cai@lca.pw>
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190827101635.6942-1-maddy@linux.vnet.ibm.com
5 years agopowerpc/64s/radix: introduce options to disable use of the tlbie instruction
Nicholas Piggin [Mon, 2 Sep 2019 15:29:31 +0000 (01:29 +1000)]
powerpc/64s/radix: introduce options to disable use of the tlbie instruction

Introduce two options to control the use of the tlbie instruction. A
boot time option which completely disables the kernel using the
instruction, this is currently incompatible with HASH MMU, KVM, and
coherent accelerators.

And a debugfs option can be switched at runtime and avoids using tlbie
for invalidating CPU TLBs for normal process and kernel address
mappings. Coherent accelerators are still managed with tlbie, as will
KVM partition scope translations.

Cross-CPU TLB flushing is implemented with IPIs and tlbiel. This is a
basic implementation which does not attempt to make any optimisation
beyond the tlbie implementation.

This is useful for performance testing among other things. For example
in certain situations on large systems, using IPIs may be faster than
tlbie as they can be directed rather than broadcast. Later we may also
take advantage of the IPIs to do more interesting things such as trim
the mm cpumask more aggressively.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190902152931.17840-7-npiggin@gmail.com
5 years agopowerpc/64s: remove unnecessary translation cache flushes at boot
Nicholas Piggin [Mon, 2 Sep 2019 15:29:30 +0000 (01:29 +1000)]
powerpc/64s: remove unnecessary translation cache flushes at boot

The various translation structure invalidations performed in early boot
when the MMU is off are not required, because everything is invalidated
immediately before a CPU first enables its MMU (see early_init_mmu
and early_init_mmu_secondary).

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190902152931.17840-6-npiggin@gmail.com
5 years agopowerpc/64s/pseries: radix flush translations before MMU is enabled at boot
Nicholas Piggin [Mon, 2 Sep 2019 15:29:29 +0000 (01:29 +1000)]
powerpc/64s/pseries: radix flush translations before MMU is enabled at boot

Radix guests are responsible for managing their own translation caches,
so make them match bare metal radix and hash, and make each CPU flush
all its translations right before enabling its MMU.

Radix guests may not flush partition scope translations, so in
tlbiel_all, make these flushes conditional on CPU_FTR_HVMODE. Process
scope translations are the only type visible to the guest.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190902152931.17840-5-npiggin@gmail.com
5 years agopowerpc/64s: make mmu_partition_table_set_entry TLB flush optional
Nicholas Piggin [Mon, 2 Sep 2019 15:29:28 +0000 (01:29 +1000)]
powerpc/64s: make mmu_partition_table_set_entry TLB flush optional

No functional change.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190902152931.17840-4-npiggin@gmail.com
5 years agopowerpc/64s/radix: tidy up TLB flushing code
Nicholas Piggin [Mon, 2 Sep 2019 15:29:27 +0000 (01:29 +1000)]
powerpc/64s/radix: tidy up TLB flushing code

There should be no functional changes.

- Use calls to existing radix_tlb.c functions in flush_partition.

- Rename radix__flush_tlb_lpid to radix__flush_all_lpid and similar,
  because they flush everything, matching flush_all_mm rather than
  flush_tlb_mm for the lpid.

- Remove some unused radix_tlb.c flush primitives.

Signed-off: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190902152931.17840-3-npiggin@gmail.com
5 years agopowerpc/64s: remove register_process_table callback
Nicholas Piggin [Mon, 2 Sep 2019 15:29:26 +0000 (01:29 +1000)]
powerpc/64s: remove register_process_table callback

This callback is only required because the partition table init comes
before process table allocation on powernv (aka bare metal aka native).

Change the order to allocate the process table first, and remove the
callback.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190902152931.17840-2-npiggin@gmail.com
5 years agoselftests/powerpc: Add basic EEH selftest
Oliver O'Halloran [Tue, 3 Sep 2019 10:16:05 +0000 (20:16 +1000)]
selftests/powerpc: Add basic EEH selftest

Use the new eeh_dev_check and eeh_dev_break interfaces to test EEH
recovery.  Historically this has been done manually using platform specific
EEH error injection facilities (e.g. via RTAS). However, documentation on
how to use these facilities is haphazard at best and non-existent at worst
so it's hard to develop a cross-platform test.

The new debugfs interfaces allow the kernel to handle the platform specific
details so we can write a more generic set of sets. This patch adds the
most basic of recovery tests where:

a) Errors are injected and recovered from sequentially,
b) Errors are not injected into PCI-PCI bridges, such as PCIe switches.
c) Errors are only injected into device function zero.
d) No errors are injected into Virtual Functions.

a), b) and c) are largely due to limitations of Linux's EEH support.  EEH
recovery is serialised in the EEH recovery thread which forces a).
Similarly, multi-function PCI devices are almost always grouped into the
same PE so injecting an error on one function exercises the same code
paths. c) is because we currently more or less ignore PCI bridges during
recovery and assume that the recovered topology will be the same as the
original.

d) is due to the limits of the eeh_dev_break interface. With the current
implementation we can't inject an error into a specific VF without
potentially causing additional errors on other VFs. Due to the serialised
recovery process we might end up timing out waiting for another function to
recover before the function of interest is recovered. The platform specific
error injection facilities are finer-grained and allow this capability, but
doing that requires working out how to use those facilities first.

Basicly, it's better than nothing and it's a base to build on.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-15-oohall@gmail.com
5 years agopowerpc/eeh: Add a eeh_dev_break debugfs interface
Oliver O'Halloran [Tue, 3 Sep 2019 10:16:04 +0000 (20:16 +1000)]
powerpc/eeh: Add a eeh_dev_break debugfs interface

Add an interface to debugfs for generating an EEH event on a given device.
This works by disabling memory accesses to and from the device by setting
the PCI_COMMAND register (or the VF Memory Space Enable on the parent PF).

This is a somewhat portable alternative to using the platform specific
error injection mechanisms since those tend to be either hard to use, or
straight up broken. For pseries the interfaces also requires the use of
/dev/mem which is probably going to go away in a post-LOCKDOWN world
(and it's a horrific hack to begin with) so moving to a kernel-provided
interface makes sense and provides a sane, cross-platform interface for
userspace so we can write more generic testing scripts.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-14-oohall@gmail.com
5 years agopowerpc/eeh: Add debugfs interface to run an EEH check
Oliver O'Halloran [Tue, 3 Sep 2019 10:16:03 +0000 (20:16 +1000)]
powerpc/eeh: Add debugfs interface to run an EEH check

Detecting an frozen EEH PE usually occurs when an MMIO load returns a 0xFFs
response. When performing EEH testing using the EEH error injection feature
available on some platforms there is no simple way to kick-off the kernel's
recovery process since any accesses from userspace (usually /dev/mem) will
bypass the MMIO helpers in the kernel which check if a 0xFF response is due
to an EEH freeze or not.

If a device contains a 0xFF byte in it's config space it's possible to
trigger the recovery process via config space read from userspace, but this
is not a reliable method. If a driver is bound to the device an in use it
will frequently trigger the MMIO check, but this is also inconsistent.

To solve these problems this patch adds a debugfs file called
"eeh_dev_check" which accepts a <domain>:<bus>:<dev>.<fn> string and runs
eeh_dev_check_failure() on it. This is the same check that's done when the
kernel gets a 0xFF result from an config or MMIO read with the added
benifit that it can be reliably triggered from userspace.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-13-oohall@gmail.com
5 years agopowerpc/eeh: Set attention indicator while recovering
Oliver O'Halloran [Tue, 3 Sep 2019 10:16:02 +0000 (20:16 +1000)]
powerpc/eeh: Set attention indicator while recovering

I am the RAS team. Hear me roar.

Roar.

On a more serious note, being able to locate failed devices can be helpful.
Set the attention indicator if the slot supports it once we've determined
the device is present and only clear it if the device is fully recovered.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-12-oohall@gmail.com
5 years agopci-hotplug/pnv_php: Add attention indicator support
Oliver O'Halloran [Tue, 3 Sep 2019 10:16:01 +0000 (20:16 +1000)]
pci-hotplug/pnv_php: Add attention indicator support

pnv_php is generally used with PCIe bridges which provide a native
interface for setting the attention and power indicator LEDs. Wire up
those interfaces even if firmware does not have support for them (yet...)

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-11-oohall@gmail.com
5 years agopci-hotplug/pnv_php: Add support for IODA3 Power9 PHBs
Oliver O'Halloran [Tue, 3 Sep 2019 10:16:00 +0000 (20:16 +1000)]
pci-hotplug/pnv_php: Add support for IODA3 Power9 PHBs

Currently we check that an IODA2 compatible PHB is upstream of this slot.
This is mainly to avoid pnv_php creating slots for the various "virtual
PHBs" that we create for NVLink. There's no real need for this restriction
so allow it on IODA3.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-10-oohall@gmail.com
5 years agopci-hotplug/pnv_php: Add a reset_slot() callback
Oliver O'Halloran [Tue, 3 Sep 2019 10:15:59 +0000 (20:15 +1000)]
pci-hotplug/pnv_php: Add a reset_slot() callback

When performing EEH recovery of devices in a hotplug slot we need to use
the slot driver's ->reset_slot() callback to prevent spurious hotplug
events due to spurious DLActive and PresDet change interrupts. Add a
reset_slot() callback to pnv_php so we can handle recovery of devices
in pnv_php managed slots.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-9-oohall@gmail.com
5 years agopowernv/eeh: Use generic code to handle hot resets
Oliver O'Halloran [Tue, 3 Sep 2019 10:15:58 +0000 (20:15 +1000)]
powernv/eeh: Use generic code to handle hot resets

When we reset PCI devices managed by a hotplug driver the reset may
generate spurious hotplug events that cause the PCI device we're resetting
to be torn down accidently. This is a problem for EEH (when the driver is
EEH aware) since we want to leave the OS PCI device state intact so that
the device can be re-set without losing any resources (network, disks,
etc) provided by the driver.

Generic PCI code provides the pci_bus_error_reset() function to handle
resetting a PCI Device (or bus) by using the reset method provided by the
hotplug slot driver. We can use this function if the EEH core has
requested a hot reset (common case) without tripping over the hotplug
driver.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-8-oohall@gmail.com
5 years agopowerpc/eeh: Remove stale CAPI comment
Oliver O'Halloran [Tue, 3 Sep 2019 10:15:57 +0000 (20:15 +1000)]
powerpc/eeh: Remove stale CAPI comment

Support for switching CAPI cards into and out of CAPI mode was removed a
while ago. Drop the comment since it's no longer relevant.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-7-oohall@gmail.com
5 years agopowerpc/eeh: Defer printing stack trace
Oliver O'Halloran [Tue, 3 Sep 2019 10:15:56 +0000 (20:15 +1000)]
powerpc/eeh: Defer printing stack trace

Currently we print a stack trace in the event handler to help with
debugging EEH issues. In the case of suprise hot-unplug this is unneeded,
so we want to prevent printing the stack trace unless we know it's due to
an actual device error. To accomplish this, we can save a stack trace at
the point of detection and only print it once the EEH recovery handler has
determined the freeze was due to an actual error.

Since the whole point of this is to prevent spurious EEH output we also
move a few prints out of the detection thread, or mark them as pr_debug
so anyone interested can get output from the eeh_check_dev_failure()
if they want.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-6-oohall@gmail.com
5 years agopowerpc/eeh: Check slot presence state in eeh_handle_normal_event()
Oliver O'Halloran [Tue, 3 Sep 2019 10:15:55 +0000 (20:15 +1000)]
powerpc/eeh: Check slot presence state in eeh_handle_normal_event()

When a device is surprise removed while undergoing IO we will probably
get an EEH PE freeze due to MMIO timeouts and other errors. When a freeze
is detected we send a recovery event to the EEH worker thread which will
notify drivers, and perform recovery as needed.

In the event of a hot-remove we don't want recovery to occur since there
isn't a device to recover. The recovery process is fairly long due to
the number of wait states (required by PCIe) which causes problems when
devices are removed and replaced (e.g. hot swapping of U.2 NVMe drives).

To determine if we need to skip the recovery process we can use the
get_adapter_state() operation of the hotplug_slot to determine if the
slot contains a device or not, and if the slot is empty we can skip
recovery entirely.

One thing to note is that the slot being EEH frozen does not prevent the
hotplug driver from working. We don't have the EEH recovery thread
remove any of the devices since it's assumed that the hotplug driver
will handle tearing down the slot state.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-5-oohall@gmail.com
5 years agopowerpc/eeh: Make permanently failed devices non-actionable
Oliver O'Halloran [Tue, 3 Sep 2019 10:15:54 +0000 (20:15 +1000)]
powerpc/eeh: Make permanently failed devices non-actionable

If a device is torn down by a hotplug slot driver it's marked as removed
and marked as permaantly failed. There's no point in trying to recover a
permernantly failed device so it should be considered un-actionable.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-4-oohall@gmail.com
5 years agopowerpc/eeh: Fix race when freeing PDNs
Oliver O'Halloran [Tue, 3 Sep 2019 10:15:53 +0000 (20:15 +1000)]
powerpc/eeh: Fix race when freeing PDNs

When hot-adding devices we rely on the hotplug driver to create pci_dn's
for the devices under the hotplug slot. Converse, when hot-removing the
driver will remove the pci_dn's that it created. This is a problem because
the pci_dev is still live until it's refcount drops to zero. This can
happen if the driver is slow to tear down it's internal state. Ideally, the
driver would not attempt to perform any config accesses to the device once
it's been marked as removed, but sometimes it happens. As a result, we
might attempt to access the pci_dn for a device that has been torn down and
the kernel may crash as a result.

To fix this, don't free the pci_dn unless the corresponding pci_dev has
been released.  If the pci_dev is still live, then we mark the pci_dn with
a flag that indicates the pci_dev's release function should free it.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-3-oohall@gmail.com
5 years agopowerpc/eeh: Clean up EEH PEs after recovery finishes
Oliver O'Halloran [Tue, 3 Sep 2019 10:15:52 +0000 (20:15 +1000)]
powerpc/eeh: Clean up EEH PEs after recovery finishes

When the last device in an eeh_pe is removed the eeh_pe structure itself
(and any empty parents) are freed since they are no longer needed. This
results in a crash when a hotplug driver is involved since the following
may occur:

1. Device is suprise removed.
2. Driver performs an MMIO, which fails and queues and eeh_event.
3. Hotplug driver receives a hotplug interrupt and removes any
   pci_devs that were under the slot.
4. pci_dev is torn down and the eeh_pe is freed.
5. The EEH event handler thread processes the eeh_event and crashes
   since the eeh_pe pointer in the eeh_event structure is no
   longer valid.

Crashing is generally considered poor form. Instead of doing that use
the fact PEs are marked as EEH_PE_INVALID to keep them around until the
end of the recovery cycle, at which point we can safely prune any empty
PEs.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190903101605.2890-2-oohall@gmail.com
5 years agopowerpc/64s/exception: reduce page fault unnecessary loads
Nicholas Piggin [Fri, 2 Aug 2019 10:57:01 +0000 (20:57 +1000)]
powerpc/64s/exception: reduce page fault unnecessary loads

This avoids 3 loads in the radix page fault case, 1 load in the
hash fault case, and 2 loads in the hash miss page fault case.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-37-npiggin@gmail.com
5 years agopowerpc/64s/exception: Remove pointless KVM handler name bifurcation
Nicholas Piggin [Fri, 2 Aug 2019 10:57:00 +0000 (20:57 +1000)]
powerpc/64s/exception: Remove pointless KVM handler name bifurcation

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-36-npiggin@gmail.com
5 years agopowerpc/64s/exception: program check handler do not branch into a macro
Nicholas Piggin [Fri, 2 Aug 2019 10:56:59 +0000 (20:56 +1000)]
powerpc/64s/exception: program check handler do not branch into a macro

It is clever, but the small code saving is not worth the spaghetti of
jumping to a label in an expanded macro, particularly when the label
is just a number rather than a descriptive name.

So expand the INT_COMMON macro twice, once for the stack and no stack
cases, and branch to those. The slight code size increase is worth
the improved clarity of branches for this non-performance critical
code.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-35-npiggin@gmail.com
5 years agopowerpc/64s/exception: move interrupt entry code above the common handler
Nicholas Piggin [Fri, 2 Aug 2019 10:56:58 +0000 (20:56 +1000)]
powerpc/64s/exception: move interrupt entry code above the common handler

This better reflects the order in which the code is executed.

No generated code change except BUG line number constants.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-34-npiggin@gmail.com
5 years agopowerpc/64s/exception: INT_COMMON add DAR, DSISR, reconcile options
Nicholas Piggin [Fri, 2 Aug 2019 10:56:57 +0000 (20:56 +1000)]
powerpc/64s/exception: INT_COMMON add DAR, DSISR, reconcile options

Move DAR and DSISR saving to pt_regs into INT_COMMON. Also add an
option to expand RECONCILE_IRQ_STATE.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-33-npiggin@gmail.com
5 years agopowerpc/64s/exception: Expand EXCEPTION_PROLOG_COMMON_1 and 2 into caller
Nicholas Piggin [Fri, 2 Aug 2019 10:56:56 +0000 (20:56 +1000)]
powerpc/64s/exception: Expand EXCEPTION_PROLOG_COMMON_1 and 2 into caller

No generated code change except BUG line number constants.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-32-npiggin@gmail.com
5 years agopowerpc/64s/exception: Expand EXCEPTION_COMMON macro into caller
Nicholas Piggin [Fri, 2 Aug 2019 10:56:55 +0000 (20:56 +1000)]
powerpc/64s/exception: Expand EXCEPTION_COMMON macro into caller

No generated code change except BUG line number constants.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-31-npiggin@gmail.com
5 years agopowerpc/64s/exception: Add INT_COMMON gas macro to generate common exception code
Nicholas Piggin [Fri, 2 Aug 2019 10:56:54 +0000 (20:56 +1000)]
powerpc/64s/exception: Add INT_COMMON gas macro to generate common exception code

No generated code change except BUG line number constants.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-30-npiggin@gmail.com
5 years agopowerpc/64s/exception: Merge EXCEPTION_PROLOG_COMMON_2/3
Nicholas Piggin [Fri, 2 Aug 2019 10:56:53 +0000 (20:56 +1000)]
powerpc/64s/exception: Merge EXCEPTION_PROLOG_COMMON_2/3

Merge EXCEPTION_PROLOG_COMMON_3 into EXCEPTION_PROLOG_COMMON_2.

No generated code change except BUG line number constants.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-29-npiggin@gmail.com
5 years agopowerpc/64s/exception: KVM_HANDLER reorder arguments to match other macros
Nicholas Piggin [Fri, 2 Aug 2019 10:56:52 +0000 (20:56 +1000)]
powerpc/64s/exception: KVM_HANDLER reorder arguments to match other macros

Also change argument name (n -> vec) to match others.

No generated code change.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-28-npiggin@gmail.com
5 years agopowerpc/64s/exception: Add INT_KVM_HANDLER gas macro
Nicholas Piggin [Fri, 2 Aug 2019 10:56:51 +0000 (20:56 +1000)]
powerpc/64s/exception: Add INT_KVM_HANDLER gas macro

Replace the 4 variants of cpp macros with one gas macro.

No generated code change except BUG line number constants.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-27-npiggin@gmail.com
5 years agopowerpc/64s/exception: INT_HANDLER support HDAR/HDSISR and use it in HDSI
Nicholas Piggin [Fri, 2 Aug 2019 10:56:50 +0000 (20:56 +1000)]
powerpc/64s/exception: INT_HANDLER support HDAR/HDSISR and use it in HDSI

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-26-npiggin@gmail.com
5 years agopowerpc/64s/exception: Add the virt variant of the denorm interrupt handler
Nicholas Piggin [Fri, 2 Aug 2019 10:56:49 +0000 (20:56 +1000)]
powerpc/64s/exception: Add the virt variant of the denorm interrupt handler

All other virt handlers have the prolog code in the virt vector rather
than branch to the real vector. Follow this pattern in the denorm virt
handler.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-25-npiggin@gmail.com
5 years agopowerpc/64s/exception: remove EXCEPTION_PROLOG_0/1, rename _2
Nicholas Piggin [Fri, 2 Aug 2019 10:56:48 +0000 (20:56 +1000)]
powerpc/64s/exception: remove EXCEPTION_PROLOG_0/1, rename _2

EXCEPTION_PROLOG_0 and _1 have only a single caller, so expand them
into it.

Rename EXCEPTION_PROLOG_2_REAL to INT_SAVE_SRR_AND_JUMP and
EXCEPTION_PROLOG_2_VIRT to INT_VIRT_SAVE_SRR_AND_JUMP, which are
more descriptive.

No generated code change except BUG line number constants.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-24-npiggin@gmail.com
5 years agopowerpc/64s/exceptions: Use keyword params to shorten arg lists
Michael Ellerman [Thu, 29 Aug 2019 13:36:08 +0000 (23:36 +1000)]
powerpc/64s/exceptions: Use keyword params to shorten arg lists

The argument lists for the INT_HANDLER macro are getting a bit
unwieldy. Use keyword parameters with default values to shorten them.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190830011426.16810-1-mpe@ellerman.id.au
5 years agopowerpc/64s/exception: Replace PROLOG macros and EXC helpers with a gas macro
Nicholas Piggin [Fri, 2 Aug 2019 10:56:47 +0000 (20:56 +1000)]
powerpc/64s/exception: Replace PROLOG macros and EXC helpers with a gas macro

This creates a single macro that generates the exception prolog code,
with variants specified by arguments, rather than assorted nested
macros for different variants.

The increasing length of macro argument list is not nice to read or
modify, but this is a temporary condition that will be improved in
later changes.

No generated code change except BUG line number constants and label
names.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-23-npiggin@gmail.com
5 years agopowerpc/64s/exception: remove 0xb00 handler
Nicholas Piggin [Fri, 2 Aug 2019 10:56:46 +0000 (20:56 +1000)]
powerpc/64s/exception: remove 0xb00 handler

This vector is not used by any supported processor, and has been
implemented as an unknown exception going back to 2.6. There is
nothing special about 0xb00, so remove it like other unused
vectors.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-22-npiggin@gmail.com
5 years agopowerpc/64s/exception: Fix performance monitor virt handler
Nicholas Piggin [Fri, 2 Aug 2019 10:56:45 +0000 (20:56 +1000)]
powerpc/64s/exception: Fix performance monitor virt handler

The perf virt handler uses EXCEPTION_PROLOG_2_REAL rather than _VIRT.
In practice this is okay because the _REAL variant is usable by virt
mode interrupts, but should be fixed (and is a performance win).

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-21-npiggin@gmail.com
5 years agopowerpc/64s/exception: Add EXC_HV_OR_STD, which selects HSRR if HVMODE
Nicholas Piggin [Fri, 2 Aug 2019 10:56:44 +0000 (20:56 +1000)]
powerpc/64s/exception: Add EXC_HV_OR_STD, which selects HSRR if HVMODE

Add EXC_HV_OR_STD and use it to consolidate the 0x500 external
interrupt.

Executed code is unchanged.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-20-npiggin@gmail.com
5 years agopowerpc/64s/exception: move head-64.h exception code to exception-64s.S
Nicholas Piggin [Fri, 2 Aug 2019 10:56:43 +0000 (20:56 +1000)]
powerpc/64s/exception: move head-64.h exception code to exception-64s.S

The head-64.h code should deal only with the head code sections
and offset calculations.

No generated code change except BUG line number constants.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-19-npiggin@gmail.com
5 years agopowerpc/64s/exception: Fix DAR load for handle_page_fault error case
Nicholas Piggin [Fri, 2 Aug 2019 10:56:42 +0000 (20:56 +1000)]
powerpc/64s/exception: Fix DAR load for handle_page_fault error case

This buglet goes back to before the 64/32 arch merge, but it does not
seem to have had practical consequences because bad_page_fault does
not use the 2nd argument, but rather regs->dar/nip.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-18-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check improve labels and comments
Nicholas Piggin [Fri, 2 Aug 2019 10:56:41 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check improve labels and comments

Short forward and backward branches can be given number labels,
but larger significant divergences in code path a more readable
if they're given descriptive names.

Also adjusts a comment to account for guest delivery.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-17-npiggin@gmail.com
5 years agopowerpc/64s/exception: untangle early machine check handler branch
Nicholas Piggin [Fri, 2 Aug 2019 10:56:40 +0000 (20:56 +1000)]
powerpc/64s/exception: untangle early machine check handler branch

machine_check_early_common now branches to machine_check_handle_early
which is its only caller.

Move interleaving code out of the way, and remove the branch.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-16-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check move unrecoverable handling out of line
Nicholas Piggin [Fri, 2 Aug 2019 10:56:39 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check move unrecoverable handling out of line

Similarly to the previous change, all callers of the unrecoverable
handler run relocated so can reach it with a direct branch. This makes
it easy to move out of line, which makes the "normal" path less
cluttered and easier to follow.

MSR[ME] manipulation still requires the rfi, so that is moved out of
line to its own function.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-15-npiggin@gmail.com
5 years agopowerpc/64s/exception: simplify machine check early path
Nicholas Piggin [Fri, 2 Aug 2019 10:56:38 +0000 (20:56 +1000)]
powerpc/64s/exception: simplify machine check early path

machine_check_handle_early_common can reach machine_check_handle_early
directly now that it runs at the relocated address, so just branch
directly.

The rfi sequence is required to enable MSR[ME] but that step is moved
into a helper function, making the code easier to follow.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-14-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check move tramp code
Nicholas Piggin [Fri, 2 Aug 2019 10:56:37 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check move tramp code

Following convention, move the tramp code (unrelocated) above the
common handlers (relocated).

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-13-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check restructure to reuse common macros
Nicholas Piggin [Fri, 2 Aug 2019 10:56:36 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check restructure to reuse common macros

Follow the pattern of sreset and HMI handlers more closely: use
EXCEPTION_PROLOG_COMMON_1 rather than open-coding it, and run the
handler at the relocated location.

This helps later simplification and code sharing.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-12-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check pseries should skip the late handler for kernel...
Nicholas Piggin [Fri, 2 Aug 2019 10:56:35 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check pseries should skip the late handler for kernel MCEs

The powernv machine check handler copes with taking a MCE from one of
three contexts, guest, kernel, and user. In each case the early
handler runs first on a special stack, then:

- The guest case branches to the KVM interrupt handler (via standard
  interrupt macros).
- The user case will run the "late" handler which is like a normal
  interrupt that runs in virtual mode and uses the regular kernel
  stack.
- The kernel case queues the event and schedules it for processing
  with irq work.

The last case is important, it must not enable virtual memory because
the MMU state may not be set up to deal with that (e.g., SLB might be
clear), it must not use the regular kernel stack for similar reasons
(e.g., might be in OPAL with OPAL stack in r1), and the kernel does
not expect anything to touch its stack if interrupts are disabled.

The pseries handler does not do this queueing, but instead it always
runs the late handler for host MCEs, which has some of the same
problems.

Now that pseries is using machine_check_events, change it to do the
same as powernv and queue events for kernel MCEs.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-11-npiggin@gmail.com
5 years agopowerpc/64s/pseries: machine check convert to use common event code
Nicholas Piggin [Fri, 2 Aug 2019 10:56:34 +0000 (20:56 +1000)]
powerpc/64s/pseries: machine check convert to use common event code

The common machine_check_event data structures and queues are mostly
platform independent, with powernv decoding SRR1/DSISR/etc., into
machine_check_event objects.

This patch converts pseries to use this infrastructure by decoding
fwnmi/rtas data into machine_check_event objects.

This allows queueing to be used by a subsequent change to delay the
virtual mode handling of machine checks that occur in kernel space
where it is unsafe to switch immediately to virtual mode, similarly
to powernv.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Fix implicit fallthrough warnings in mce_handle_error()]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-10-npiggin@gmail.com
5 years agopowerpc/64s/powernv: machine check dump SLB contents
Nicholas Piggin [Fri, 2 Aug 2019 10:56:33 +0000 (20:56 +1000)]
powerpc/64s/powernv: machine check dump SLB contents

Re-use the code introduced in pseries to save and dump the contents
of the SLB in the case of an SLB involved machine check exception.

This patch also avoids allocating the SLB save array on pseries radix.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-9-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check use correct cfar for late handler
Nicholas Piggin [Fri, 2 Aug 2019 10:56:32 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check use correct cfar for late handler

Bare metal machine checks run an "early" handler in real mode before
running the main handler which reports the event.

The main handler runs exactly as a normal interrupt handler, after the
"windup" which sets registers back as they were at interrupt entry.
CFAR does not get restored by the windup code, so that will be wrong
when the handler is run.

Restore the CFAR to the saved value before running the late handler.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-8-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check remove machine_check_pSeries_0 branch
Nicholas Piggin [Fri, 2 Aug 2019 10:56:31 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check remove machine_check_pSeries_0 branch

This label has only one caller, so unwind the branch and move it
inline. The location of the comment is adjusted to match similar
one in system reset.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-7-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check pseries should always run the early handler
Nicholas Piggin [Fri, 2 Aug 2019 10:56:30 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check pseries should always run the early handler

Now that pseries with fwnmi registered runs the early machine check
handler, there is no good reason to special case the non-fwnmi case
and skip the early handler. Reducing the code and number of paths is
a top priority for asm code, it's better to handle this in C where
possible (and the pseries early handler is a no-op if fwnmi is not
registered).

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-6-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check adjust RFI target
Nicholas Piggin [Fri, 2 Aug 2019 10:56:29 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check adjust RFI target

The host kernel delivery case for powernv does RFI_TO_USER_OR_KERNEL,
but should just use RFI_TO_KERNEL which makes it clear this is not a
user case.

This is not a bug because RFI_TO_USER_OR_KERNEL deals with kernel
returns just fine.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-5-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check fix KVM guest test
Nicholas Piggin [Fri, 2 Aug 2019 10:56:28 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check fix KVM guest test

The machine_check_handle_early hypervisor guest test is skipped if
!HVMODE or MSR[HV]=0, which is wrong for PR or nested hypervisors
that could be running a guest in this state.

Test HSTATE_IN_GUEST up front and use that to branch out to the KVM
handler, then MSR[PR] alone can test for this kernel's userspace.
This matches all other interrupt handling.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-4-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check remove bitrotted comment
Nicholas Piggin [Fri, 2 Aug 2019 10:56:27 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check remove bitrotted comment

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-3-npiggin@gmail.com
5 years agopowerpc/64s/exception: machine check fwnmi remove HV case
Nicholas Piggin [Fri, 2 Aug 2019 10:56:26 +0000 (20:56 +1000)]
powerpc/64s/exception: machine check fwnmi remove HV case

fwnmi does not trigger in HV mode, so remove always-true feature test.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802105709.27696-2-npiggin@gmail.com
5 years agopowerpc/configs: Enable secure guest support in pseries and ppc64 defconfigs
Ryan Grimm [Tue, 20 Aug 2019 02:13:26 +0000 (23:13 -0300)]
powerpc/configs: Enable secure guest support in pseries and ppc64 defconfigs

Enables running as a secure guest in platforms with an Ultravisor.

Signed-off-by: Ryan Grimm <grimm@linux.vnet.ibm.com>
Signed-off-by: Ram Pai <linuxram@us.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-17-bauerman@linux.ibm.com
5 years agopowerpc/pseries/svm: Force SWIOTLB for secure guests
Anshuman Khandual [Tue, 20 Aug 2019 02:13:24 +0000 (23:13 -0300)]
powerpc/pseries/svm: Force SWIOTLB for secure guests

SWIOTLB checks range of incoming CPU addresses to be bounced and sees if
the device can access it through its DMA window without requiring bouncing.
In such cases it just chooses to skip bouncing. But for cases like secure
guests on powerpc platform all addresses need to be bounced into the shared
pool of memory because the host cannot access it otherwise. Hence the need
to do the bouncing is not related to device's DMA window and use of bounce
buffers is forced by setting swiotlb_force.

Also, connect the shared memory conversion functions into the
ARCH_HAS_MEM_ENCRYPT hooks and call swiotlb_update_mem_attributes() to
convert SWIOTLB's memory pool to shared memory.

Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
[ bauerman: Use ARCH_HAS_MEM_ENCRYPT hooks to share swiotlb memory pool. ]
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-15-bauerman@linux.ibm.com
5 years agopowerpc/pseries/iommu: Don't use dma_iommu_ops on secure guests
Thiago Jung Bauermann [Tue, 20 Aug 2019 02:13:23 +0000 (23:13 -0300)]
powerpc/pseries/iommu: Don't use dma_iommu_ops on secure guests

Secure guest memory is inacessible to devices so regular DMA isn't
possible.

In that case set devices' dma_map_ops to NULL so that the generic
DMA code path will use SWIOTLB to bounce buffers for DMA.

Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-14-bauerman@linux.ibm.com
5 years agopowerpc/pseries/svm: Disable doorbells in SVM guests
Sukadev Bhattiprolu [Tue, 20 Aug 2019 02:13:22 +0000 (23:13 -0300)]
powerpc/pseries/svm: Disable doorbells in SVM guests

Normally, the HV emulates some instructions like MSGSNDP, MSGCLRP
from a KVM guest. To emulate the instructions, it must first read
the instruction from the guest's memory and decode its parameters.

However for a secure guest (aka SVM), the page containing the
instruction is in secure memory and the HV cannot access directly.
It would need the Ultravisor (UV) to facilitate accessing the
instruction and parameters but the UV currently does not have
the support for such accesses.

Until the UV has such support, disable doorbells in SVMs. This might
incur a performance hit but that is yet to be quantified.

With this patch applied (needed only in SVMs not needed for HV) we
are able to launch SVM guests with multi-core support. Eg:

qemu -smp sockets=2,cores=2,threads=2.

Fix suggested by Benjamin Herrenschmidt. Thanks to input from
Paul Mackerras, Ram Pai and Michael Anderson.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-13-bauerman@linux.ibm.com
5 years agopowerpc/pseries/svm: Export guest SVM status to user space via sysfs
Ryan Grimm [Tue, 20 Aug 2019 02:13:21 +0000 (23:13 -0300)]
powerpc/pseries/svm: Export guest SVM status to user space via sysfs

User space might want to know it's running in a secure VM.  It can't do
a mfmsr because mfmsr is a privileged instruction.

The solution here is to create a cpu attribute:

/sys/devices/system/cpu/svm

which will read 0 or 1 based on the S bit of the current CPU.

Signed-off-by: Ryan Grimm <grimm@linux.vnet.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-12-bauerman@linux.ibm.com
5 years agopowerpc/pseries/svm: Unshare all pages before kexecing a new kernel
Ram Pai [Tue, 20 Aug 2019 02:13:20 +0000 (23:13 -0300)]
powerpc/pseries/svm: Unshare all pages before kexecing a new kernel

A new kernel deserves a clean slate. Any pages shared with the hypervisor
is unshared before invoking the new kernel. However there are exceptions.
If the new kernel is invoked to dump the current kernel, or if there is a
explicit request to preserve the state of the current kernel, unsharing
of pages is skipped.

NOTE: While testing crashkernel, make sure at least 256M is reserved for
crashkernel. Otherwise SWIOTLB allocation will fail and crash kernel will
fail to boot.

Signed-off-by: Ram Pai <linuxram@us.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-11-bauerman@linux.ibm.com
5 years agopowerpc/pseries/svm: Use shared memory for Debug Trace Log (DTL)
Anshuman Khandual [Tue, 20 Aug 2019 02:13:19 +0000 (23:13 -0300)]
powerpc/pseries/svm: Use shared memory for Debug Trace Log (DTL)

Secure guests need to share the DTL buffers with the hypervisor. To that
end, use a kmem_cache constructor which converts the underlying buddy
allocated SLUB cache pages into shared memory.

Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-10-bauerman@linux.ibm.com
5 years agopowerpc/pseries/svm: Use shared memory for LPPACA structures
Anshuman Khandual [Tue, 20 Aug 2019 02:13:18 +0000 (23:13 -0300)]
powerpc/pseries/svm: Use shared memory for LPPACA structures

LPPACA structures need to be shared with the host. Hence they need to be in
shared memory. Instead of allocating individual chunks of memory for a
given structure from memblock, a contiguous chunk of memory is allocated
and then converted into shared memory. Subsequent allocation requests will
come from the contiguous chunk which will be always shared memory for all
structures.

While we are able to use a kmem_cache constructor for the Debug Trace Log,
LPPACAs are allocated very early in the boot process (before SLUB is
available) so we need to use a simpler scheme here.

Introduce helper is_svm_platform() which uses the S bit of the MSR to tell
whether we're running as a secure guest.

Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-9-bauerman@linux.ibm.com
5 years agopowerpc/pseries: Add and use LPPACA_SIZE constant
Thiago Jung Bauermann [Tue, 20 Aug 2019 02:13:17 +0000 (23:13 -0300)]
powerpc/pseries: Add and use LPPACA_SIZE constant

Helps document what the hard-coded number means.

Also take the opportunity to fix an #endif comment.

Suggested-by: Alexey Kardashevskiy <aik@linux.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-8-bauerman@linux.ibm.com
5 years agopowerpc: Introduce the MSR_S bit
Sukadev Bhattiprolu [Tue, 20 Aug 2019 02:13:16 +0000 (23:13 -0300)]
powerpc: Introduce the MSR_S bit

Protected Execution Facility (PEF) is an architectural change for
POWER 9 that enables Secure Virtual Machines (SVMs). When enabled,
PEF adds a new higher privileged mode, called Ultravisor mode, to
POWER architecture.

The hardware changes include the following:

  * There is a new bit in the MSR that determines whether the current
    process is running in secure mode, MSR(S) bit 41. MSR(S)=1, process
    is in secure mode, MSR(s)=0 process is in normal mode.

  * The MSR(S) bit can only be set by the Ultravisor.

  * HRFID cannot be used to set the MSR(S) bit. If the hypervisor needs
    to return to a SVM it must use an ultracall. It can determine if
    the VM it is returning to is secure.

  * The privilege of a process is now determined by three MSR bits,
    MSR(S, HV, PR). In each of the tables below the modes are listed
    from least privilege to highest privilege. The higher privilege
    modes can access all the resources of the lower privilege modes.

    **Secure Mode MSR Settings**

       +---+---+---+---------------+
       | S | HV| PR|Privilege      |
       +===+===+===+===============+
       | 1 | 0 | 1 | Problem       |
       +---+---+---+---------------+
       | 1 | 0 | 0 | Privileged(OS)|
       +---+---+---+---------------+
       | 1 | 1 | 0 | Ultravisor    |
       +---+---+---+---------------+
       | 1 | 1 | 1 | Reserved      |
       +---+---+---+---------------+

    **Normal Mode MSR Settings**

       +---+---+---+---------------+
       | S | HV| PR|Privilege      |
       +===+===+===+===============+
       | 0 | 0 | 1 | Problem       |
       +---+---+---+---------------+
       | 0 | 0 | 0 | Privileged(OS)|
       +---+---+---+---------------+
       | 0 | 1 | 0 | Hypervisor    |
       +---+---+---+---------------+
       | 0 | 1 | 1 | Problem (HV)  |
       +---+---+---+---------------+

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Signed-off-by: Ram Pai <linuxram@us.ibm.com>
[ cclaudio: Update the commit message ]
Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-7-bauerman@linux.ibm.com
5 years agopowerpc/pseries/svm: Add helpers for UV_SHARE_PAGE and UV_UNSHARE_PAGE
Ram Pai [Tue, 20 Aug 2019 02:13:15 +0000 (23:13 -0300)]
powerpc/pseries/svm: Add helpers for UV_SHARE_PAGE and UV_UNSHARE_PAGE

These functions are used when the guest wants to grant the hypervisor
access to certain pages.

Signed-off-by: Ram Pai <linuxram@us.ibm.com>
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-6-bauerman@linux.ibm.com
5 years agopowerpc/prom_init: Add the ESM call to prom_init
Ram Pai [Tue, 20 Aug 2019 02:13:14 +0000 (23:13 -0300)]
powerpc/prom_init: Add the ESM call to prom_init

Make the Enter-Secure-Mode (ESM) ultravisor call to switch the VM to secure
mode. Pass kernel base address and FDT address so that the Ultravisor is
able to verify the integrity of the VM using information from the ESM blob.

Add "svm=" command line option to turn on switching to secure mode.

Signed-off-by: Ram Pai <linuxram@us.ibm.com>
[ andmike: Generate an RTAS os-term hcall when the ESM ucall fails. ]
Signed-off-by: Michael Anderson <andmike@linux.ibm.com>
[ bauerman: Cleaned up the code a bit. ]
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-5-bauerman@linux.ibm.com
5 years agopowerpc: Add support for adding an ESM blob to the zImage wrapper
Benjamin Herrenschmidt [Tue, 20 Aug 2019 02:13:13 +0000 (23:13 -0300)]
powerpc: Add support for adding an ESM blob to the zImage wrapper

For secure VMs, the signing tool will create a ticket called the "ESM blob"
for the Enter Secure Mode ultravisor call with the signatures of the kernel
and initrd among other things.

This adds support to the wrapper script for adding that blob via the "-e"
option to the zImage.pseries.

It also adds code to the zImage wrapper itself to retrieve and if necessary
relocate the blob, and pass its address to Linux via the device-tree, to be
later consumed by prom_init.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[ bauerman: Minor adjustments to some comments. ]
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-4-bauerman@linux.ibm.com
5 years agopowerpc/pseries: Introduce option to build secure virtual machines
Thiago Jung Bauermann [Tue, 20 Aug 2019 02:13:12 +0000 (23:13 -0300)]
powerpc/pseries: Introduce option to build secure virtual machines

Introduce CONFIG_PPC_SVM to control support for secure guests and include
Ultravisor-related helpers when it is selected

Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820021326.6884-3-bauerman@linux.ibm.com
5 years agoMerge branch 'topic/ppc-kvm' into next
Michael Ellerman [Thu, 29 Aug 2019 23:52:57 +0000 (09:52 +1000)]
Merge branch 'topic/ppc-kvm' into next

Merge our ppc-kvm topic branch to bring in the Ultravisor support
patches.

5 years agoMerge branch 'topic/mem-encrypt' into next
Michael Ellerman [Thu, 29 Aug 2019 23:49:28 +0000 (09:49 +1000)]
Merge branch 'topic/mem-encrypt' into next

This branch has some cross-arch patches that are a prequisite for the
SVM work. They're in a topic branch in case any of the other arch
maintainers want to merge them to resolve conflicts.

5 years agopowerpc/powernv: Add ultravisor message log interface
Claudio Carvalho [Wed, 28 Aug 2019 13:05:21 +0000 (23:05 +1000)]
powerpc/powernv: Add ultravisor message log interface

The ultravisor (UV) provides an in-memory console which follows the
OPAL in-memory console structure.

This patch extends the OPAL msglog code to initialize the UV memory
console and provide the "/sys/firmware/ultravisor/msglog" interface
for userspace to view the UV message log.

Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Tested-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190828130521.26764-2-mpe@ellerman.id.au
5 years agopowerpc/powernv/opal-msglog: Refactor memcons code
Claudio Carvalho [Wed, 28 Aug 2019 13:05:20 +0000 (23:05 +1000)]
powerpc/powernv/opal-msglog: Refactor memcons code

This patch refactors the code in opal-msglog that operates on the OPAL
memory console in order to make it cleaner and also allow the reuse of
the new memcons_* functions.

Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Tested-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190828130521.26764-1-mpe@ellerman.id.au
5 years agopowerpc/kvm: Use UV_RETURN ucall to return to ultravisor
Sukadev Bhattiprolu [Thu, 22 Aug 2019 03:48:38 +0000 (00:48 -0300)]
powerpc/kvm: Use UV_RETURN ucall to return to ultravisor

When an SVM makes an hypercall or incurs some other exception, the
Ultravisor usually forwards (a.k.a. reflects) the exceptions to the
Hypervisor. After processing the exception, Hypervisor uses the
UV_RETURN ultracall to return control back to the SVM.

The expected register state on entry to this ultracall is:

* Non-volatile registers are restored to their original values.
* If returning from an hypercall, register R0 contains the return value
  (unlike other ultracalls) and, registers R4 through R12 contain any
  output values of the hypercall.
* R3 contains the ultracall number, i.e UV_RETURN.
* If returning with a synthesized interrupt, R2 contains the
  synthesized interrupt number.

Thanks to input from Paul Mackerras, Ram Pai and Mike Anderson.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190822034838.27876-8-cclaudio@linux.ibm.com
5 years agopowerpc/powernv: Access LDBAR only if ultravisor disabled
Claudio Carvalho [Thu, 22 Aug 2019 03:48:37 +0000 (00:48 -0300)]
powerpc/powernv: Access LDBAR only if ultravisor disabled

LDBAR is a per-thread SPR populated and used by the thread-imc pmu
driver to dump the data counter into memory. It contains memory along
with few other configuration bits. LDBAR is populated and enabled only
when any of the thread imc pmu events are monitored.

In ultravisor enabled systems, LDBAR becomes ultravisor privileged and
an attempt to write to it will cause a Hypervisor Emulation Assistance
interrupt.

In ultravisor enabled systems, the ultravisor is responsible to maintain
the LDBAR (e.g. save and restore it).

This restricts LDBAR access to only when ultravisor is disabled.

Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Reviewed-by: Ryan Grimm <grimm@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190822034838.27876-7-cclaudio@linux.ibm.com
5 years agopowerpc/mm: Write to PTCR only if ultravisor disabled
Claudio Carvalho [Thu, 22 Aug 2019 03:48:36 +0000 (00:48 -0300)]
powerpc/mm: Write to PTCR only if ultravisor disabled

In ultravisor enabled systems, PTCR becomes ultravisor privileged only
for writing and an attempt to write to it will cause a Hypervisor
Emulation Assitance interrupt.

This patch uses the set_ptcr_when_no_uv() function to restrict PTCR
writing to only when ultravisor is disabled.

Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190822034838.27876-6-cclaudio@linux.ibm.com
5 years agopowerpc/mm: Use UV_WRITE_PATE ucall to register a PATE
Michael Anderson [Thu, 22 Aug 2019 03:48:35 +0000 (00:48 -0300)]
powerpc/mm: Use UV_WRITE_PATE ucall to register a PATE

When Ultravisor (UV) is enabled, the partition table is stored in secure
memory and can only be accessed via the UV. The Hypervisor (HV) however
maintains a copy of the partition table in normal memory to allow Nest MMU
translations to occur (for normal VMs). The HV copy includes partition
table entries (PATE)s for secure VMs which would currently be unused
(Nest MMU translations cannot access secure memory) but they would be
needed as we add functionality.

This patch adds the UV_WRITE_PATE ucall which is used to update the PATE
for a VM (both normal and secure) when Ultravisor is enabled.

Signed-off-by: Michael Anderson <andmike@linux.ibm.com>
Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Signed-off-by: Ram Pai <linuxram@us.ibm.com>
[ cclaudio: Write the PATE in HV's table before doing that in UV's ]
Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Reviewed-by: Ryan Grimm <grimm@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190822034838.27876-5-cclaudio@linux.ibm.com
5 years agopowerpc/powernv: Introduce FW_FEATURE_ULTRAVISOR
Claudio Carvalho [Thu, 22 Aug 2019 03:48:34 +0000 (00:48 -0300)]
powerpc/powernv: Introduce FW_FEATURE_ULTRAVISOR

In PEF enabled systems, some of the resources which were previously
hypervisor privileged are now ultravisor privileged and controlled by
the ultravisor firmware.

This adds FW_FEATURE_ULTRAVISOR to indicate if PEF is enabled.

The host kernel can use FW_FEATURE_ULTRAVISOR, for instance, to skip
accessing resources (e.g. PTCR and LDBAR) in case PEF is enabled.

Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
[ andmike: Device node name to "ibm,ultravisor" ]
Signed-off-by: Michael Anderson <andmike@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190822034838.27876-4-cclaudio@linux.ibm.com
5 years agopowerpc/kernel: Add ucall_norets() ultravisor call handler
Claudio Carvalho [Thu, 22 Aug 2019 03:48:33 +0000 (00:48 -0300)]
powerpc/kernel: Add ucall_norets() ultravisor call handler

The ultracalls (ucalls for short) allow the Secure Virtual Machines
(SVM)s and hypervisor to request services from the ultravisor such as
accessing a register or memory region that can only be accessed when
running in ultravisor-privileged mode.

This patch adds the ucall_norets() ultravisor call handler.

The specific service needed from an ucall is specified in register
R3 (the first parameter to the ucall). Other parameters to the
ucall, if any, are specified in registers R4 through R12.

Return value of all ucalls is in register R3. Other output values
from the ucall, if any, are returned in registers R4 through R12.

Each ucall returns specific error codes, applicable in the context
of the ucall. However, like with the PowerPC Architecture Platform
Reference (PAPR), if no specific error code is defined for a particular
situation, then the ucall will fallback to an erroneous
parameter-position based code. i.e U_PARAMETER, U_P2, U_P3 etc depending
on the ucall parameter that may have caused the error.

Every host kernel (powernv) needs to be able to do ucalls in case it
ends up being run in a machine with ultravisor enabled. Otherwise, the
kernel may crash early in boot trying to access ultravisor resources,
for instance, trying to set the partition table entry 0. Secure guests
also need to be able to do ucalls and its kernel may not have
CONFIG_PPC_POWERNV=y. For that reason, the ucall.S file is placed under
arch/powerpc/kernel.

If ultravisor is not enabled, the ucalls will be redirected to the
hypervisor which must handle/fail the call.

Thanks to inputs from Ram Pai and Michael Anderson.

Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190822034838.27876-3-cclaudio@linux.ibm.com
5 years agoDocumentation/powerpc: Ultravisor API
Sukadev Bhattiprolu [Thu, 22 Aug 2019 03:48:32 +0000 (00:48 -0300)]
Documentation/powerpc: Ultravisor API

Protected Execution Facility (PEF) is an architectural change for
POWER 9 that enables Secure Virtual Machines (SVMs). When enabled,
PEF adds a new higher privileged mode, called Ultravisor mode, to POWER
architecture. Along with the new mode there is new firmware called the
Protected Execution Ultravisor (or Ultravisor for short).

POWER 9 DD2.3 chips (PVR=0x004e1203) or greater will be PEF-capable.

Attached documentation provides an overview of PEF and defines the API
for various interfaces that must be implemented in the Ultravisor
firmware as well as in the KVM Hypervisor.

Based on input from Mike Anderson, Thiago Bauermann, Claudio Carvalho,
Ben Herrenschmidt, Guerney Hunt, Paul Mackerras.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Signed-off-by: Ram Pai <linuxram@linux.ibm.com>
Signed-off-by: Guerney Hunt <gdhh@linux.ibm.com>
Reviewed-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Reviewed-by: Michael Anderson <andmike@linux.ibm.com>
Reviewed-by: Thiago Bauermann <bauerman@linux.ibm.com>
Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190822034838.27876-2-cclaudio@linux.ibm.com
5 years agoDocumentation/powerpc: Add ELF note documentation
Maxiwell S. Garcia [Thu, 29 Aug 2019 15:50:21 +0000 (12:50 -0300)]
Documentation/powerpc: Add ELF note documentation

The ELF note documentation describes the types and descriptors to be
used with the PowerPC namespace.

Signed-off-by: Maxiwell S. Garcia <maxiwell@linux.ibm.com>
Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829155021.2915-3-maxiwell@linux.ibm.com
5 years agopowerpc: Add PowerPC Capabilities ELF note
Claudio Carvalho [Thu, 29 Aug 2019 15:50:20 +0000 (12:50 -0300)]
powerpc: Add PowerPC Capabilities ELF note

Add the PowerPC name and the PPC_ELFNOTE_CAPABILITIES type in the
kernel binary ELF note. This type is a bitmap that can be used to
advertise kernel capabilities to userland.

This patch also defines PPCCAP_ULTRAVISOR_BIT as being the bit zero.

Suggested-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Claudio Carvalho <cclaudio@linux.ibm.com>
[ maxiwell: Define the 'PowerPC' type in the elfnote.h ]
Signed-off-by: Maxiwell S. Garcia <maxiwell@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829155021.2915-2-maxiwell@linux.ibm.com
5 years agopowerpc/powernv/ioda: Remove obsolete iommu_table_ops::exchange callbacks
Alexey Kardashevskiy [Thu, 29 Aug 2019 08:52:52 +0000 (18:52 +1000)]
powerpc/powernv/ioda: Remove obsolete iommu_table_ops::exchange callbacks

As now we have xchg_no_kill/tce_kill, these are not used anymore so
remove them.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829085252.72370-6-aik@ozlabs.ru
5 years agopowerpc/pseries/iommu: Switch to xchg_no_kill
Alexey Kardashevskiy [Thu, 29 Aug 2019 08:52:51 +0000 (18:52 +1000)]
powerpc/pseries/iommu: Switch to xchg_no_kill

This is the last implementation of iommu_table_ops::exchange() which
we are about to remove.

This implements xchg_no_kill() for pseries. Since it is paravirtual
platform, the hypervisor does TCE invalidations and we do not have
to deal with it here, hence no tce_kill() hook.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829085252.72370-5-aik@ozlabs.ru
5 years agovfio/spapr_tce: Invalidate multiple TCEs at once
Alexey Kardashevskiy [Thu, 29 Aug 2019 08:52:50 +0000 (18:52 +1000)]
vfio/spapr_tce: Invalidate multiple TCEs at once

Invalidating a TCE cache entry for each updated TCE is quite expensive.
This makes use of the new iommu_table_ops::xchg_no_kill()/tce_kill()
callbacks to bring down the time spent in mapping a huge guest DMA window.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829085252.72370-4-aik@ozlabs.ru