linux-2.6-microblaze.git
3 years agopowerpc/powernv/sriov: De-indent setup and teardown
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:11 +0000 (16:57 +1000)]
powerpc/powernv/sriov: De-indent setup and teardown

Remove the IODA2 PHB checks. We already assume IODA2 in several places so
there's not much point in wrapping most of the setup and teardown process
in an if block.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-12-oohall@gmail.com
3 years agopowerpc/powernv/sriov: Drop iov->pe_num_map[]
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:10 +0000 (16:57 +1000)]
powerpc/powernv/sriov: Drop iov->pe_num_map[]

Currently the iov->pe_num_map[] does one of two things depending on
whether single PE mode is being used or not. When it is, this contains an
array which maps a vf_index to the corresponding PE number. When single PE
mode is not being used this contains a scalar which is the base PE for the
set of enabled VFs (for for VFn is base + n).

The array was necessary because when calling pnv_ioda_alloc_pe() there is
no guarantee that the allocated PEs would be contigious. We can now
allocate contigious blocks of PEs so this is no longer an issue. This
allows us to drop the if (single_mode) {} .. else {} block scattered
through the SR-IOV code which is a nice clean up.

This also fixes a bug in pnv_pci_sriov_disable() which is the non-atomic
bitmap_clear() to manipulate the PE allocation map. Other users of the map
assume it will be accessed with atomic ops.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-11-oohall@gmail.com
3 years agopowerpc/powernv/pci: Refactor pnv_ioda_alloc_pe()
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:09 +0000 (16:57 +1000)]
powerpc/powernv/pci: Refactor pnv_ioda_alloc_pe()

Rework the PE allocation logic to allow allocating blocks of PEs rather
than individually. We'll use this to allocate contigious blocks of PEs for
the SR-IOVs.

This patch also adds code to pnv_ioda_alloc_pe() and pnv_ioda_reserve_pe() to
use the existing, but unused, phb->pe_alloc_mutex. Currently these functions
use atomic bit ops to release a currently allocated PE number. However,
the pnv_ioda_alloc_pe() wants to have exclusive access to the bit map while
scanning for hole large enough to accomodate the allocation size.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-10-oohall@gmail.com
3 years agopowerpc/powernv/sriov: Factor out M64 BAR setup
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:08 +0000 (16:57 +1000)]
powerpc/powernv/sriov: Factor out M64 BAR setup

The sequence required to use the single PE BAR mode is kinda janky and
requires a little explanation. The API was designed with P7-IOC style
windows where the setup process is something like:

1. Configure the window start / end address
2. Enable the window
3. Map the segments of each window to the PE

For Single PE BARs the process is:

1. Set the PE for segment zero on a disabled window
2. Set the range
3. Enable the window

Move the OPAL calls into their own helper functions where the quirks can be
contained.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-9-oohall@gmail.com
3 years agopowerpc/powernv/sriov: Simplify used window tracking
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:07 +0000 (16:57 +1000)]
powerpc/powernv/sriov: Simplify used window tracking

No need for the multi-dimensional arrays, just use a bitmap.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-8-oohall@gmail.com
3 years agopowerpc/powernv/sriov: Rename truncate_iov
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:06 +0000 (16:57 +1000)]
powerpc/powernv/sriov: Rename truncate_iov

This prevents SR-IOV being used by making the SR-IOV BAR resources
unallocatable. Rename it to reflect what it actually does.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-7-oohall@gmail.com
3 years agopowerpc/powernv/sriov: Explain how SR-IOV works on PowerNV
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:05 +0000 (16:57 +1000)]
powerpc/powernv/sriov: Explain how SR-IOV works on PowerNV

SR-IOV support on PowerNV is a byzantine maze of hooks. I have no idea
how anyone is supposed to know how it works except through a lot of
stuffering. Write up some docs about the overall story to help out
the next sucker^Wperson who needs to tinker with it.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-6-oohall@gmail.com
3 years agopowerpc/powernv/sriov: Move SR-IOV into a separate file
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:04 +0000 (16:57 +1000)]
powerpc/powernv/sriov: Move SR-IOV into a separate file

pci-ioda.c is getting a bit unwieldly due to the amount of stuff jammed in
there. The SR-IOV support can be extracted easily enough and is mostly
standalone, so move it into a separate file.

This patch also moves the PowerNV SR-IOV specific fields from pci_dn and
moves them into a platform specific structure. I'm not sure how they ended
up in there in the first place, but leaking platform specifics into common
code has proven to be a terrible idea so far so lets stop doing that.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-5-oohall@gmail.com
3 years agopowerpc/powernv/pci: Initialise M64 for IODA1 as a 1-1 window
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:03 +0000 (16:57 +1000)]
powerpc/powernv/pci: Initialise M64 for IODA1 as a 1-1 window

We pre-configure the m64 window for IODA1 as a 1-1 segment-PE mapping,
similar to PHB3. Currently the actual mapping of segments occurs in
pnv_ioda_pick_m64_pe(), but we can move it into pnv_ioda1_init_m64() and
drop the IODA1 specific code paths in the PE setup / teardown.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-4-oohall@gmail.com
3 years agopowerpc/powernv/pci: Add explicit tracking of the DMA setup state
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:02 +0000 (16:57 +1000)]
powerpc/powernv/pci: Add explicit tracking of the DMA setup state

There's an optimisation in the PE setup which skips performing DMA
setup for a PE if we only have bridges in a PE. The assumption being
that only "real" devices will DMA to system memory, which is probably
fair. However, if we start off with only bridge devices in a PE then
add a non-bridge device the new device won't be able to use DMA because
we never configured it.

Fix this (admittedly pretty weird) edge case by tracking whether we've done
the DMA setup for the PE or not. If a non-bridge device is added to the PE
(via rescan or hotplug, or whatever) we can set up DMA on demand.

This also means the only remaining user of the old "DMA Weight" code is
the IODA1 DMA setup code that it was originally added for, which is good.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-3-oohall@gmail.com
3 years agopowerpc/powernv/pci: Always tear down DMA windows on PE release
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:01 +0000 (16:57 +1000)]
powerpc/powernv/pci: Always tear down DMA windows on PE release

Currently we have these two functions:

pnv_pci_ioda2_release_dma_pe(), and
pnv_pci_ioda2_release_pe_dma()

The first is used when tearing down VF PEs and the other is used for normal
devices. There's very little difference between the two though. The latter
(non-VF) will skip a call to pnv_pci_ioda2_unset_window() unless
CONFIG_IOMMU_API=y is set. There's no real point in doing this so fold the
two together.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-2-oohall@gmail.com
3 years agopowerpc/powernv/pci: Add pci_bus_to_pnvhb() helper
Oliver O'Halloran [Wed, 22 Jul 2020 06:57:00 +0000 (16:57 +1000)]
powerpc/powernv/pci: Add pci_bus_to_pnvhb() helper

Add a helper to go from a pci_bus structure to the pnv_phb that hosts
that bus. There's a lot of instances of the following pattern:

struct pci_controller *hose = pci_bus_to_host(pdev->bus);
struct pnv_phb *phb = hose->private_data;

Without any other uses of the pci_controller inside the function. This
is hard to read since it requires you to memorise the contents of the
private data fields and kind of error prone since it involves blindly
assigning a void pointer. Add a helper to make it more concise and
explicit.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722065715.1432738-1-oohall@gmail.com
3 years agopowerpc/eeh: Move PE tree setup into the platform
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:31 +0000 (18:12 +1000)]
powerpc/eeh: Move PE tree setup into the platform

The EEH core has a concept of a "PE tree" to support PowerNV. The PE tree
follows the PCI bus structures because a reset asserted on an upstream
bridge will be propagated to the downstream bridges. On pseries there's a
1-1 correspondence between what the guest sees are a PHB and a PE so the
"tree" is really just a single node.

Current the EEH core is reponsible for setting up this PE tree which it
does by traversing the pci_dn tree. The structure of the pci_dn tree
matches the bus tree on PowerNV which leads to the PE tree being "correct"
this setup method doesn't make a whole lot of sense and it's actively
confusing for the pseries case where it doesn't really do anything.

We want to remove the dependence on pci_dn anyway so this patch move
choosing where to insert a new PE into the platform code rather than
being part of the generic EEH code. For PowerNV this simplifies the
tree building logic and removes the use of pci_dn. For pseries we
keep the existing logic. I'm not really convinced it does anything
due to the 1-1 PE-to-PHB correspondence so every device under that
PHB should be in the same PE, but I'd rather not remove it entirely
until we've had a chance to look at it more deeply.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-14-oohall@gmail.com
3 years agopowerpc/eeh: Drop pdn use in eeh_pe_tree_insert()
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:30 +0000 (18:12 +1000)]
powerpc/eeh: Drop pdn use in eeh_pe_tree_insert()

This is mostly just to make the subsequent diffs less noisy. No functional
changes.

One thing that needs calling out is the removal of the "config_addr"
variable and replacing it with edev->bdfn. The contents of edev->bdfn are
the same, however it's worth pointing out that what RTAS calls a
"config_addr" isn't the same as the bdfn. The config_addr is supposed to
be: <bus><devfn><reg> with each field being an 8 bit number. Various parts
of the EEH code use BDFN and "config_addr" as interchangeable quantities
even though they aren't really.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-13-oohall@gmail.com
3 years agopowerpc/eeh: Rename eeh_{add_to|remove_from}_parent_pe()
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:29 +0000 (18:12 +1000)]
powerpc/eeh: Rename eeh_{add_to|remove_from}_parent_pe()

The naming of eeh_{add_to|remove_from}_parent_pe() doesn't really reflect
what they actually do. If the PE referred to be edev->pe_config_addr
already exists under that PHB then the edev is added to that PE. However,
if the PE doesn't exist the a new one is created for the edev.

The bulk of the implementation of eeh_add_to_parent_pe() covers that
second case. Similarly, most of eeh_remove_from_parent_pe() is
determining when it's safe to delete a PE.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-12-oohall@gmail.com
3 years agopowerpc/eeh: Remove class code field from edev
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:28 +0000 (18:12 +1000)]
powerpc/eeh: Remove class code field from edev

The edev->class_code field is never referenced anywhere except for the
platform specific probe functions. The same information is available in
the pci_dev for PowerNV and in the pci_dn on pseries so we can remove
the field.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-11-oohall@gmail.com
3 years agopowerpc/eeh: Remove spurious use of pci_dn in eeh_dump_dev_log
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:27 +0000 (18:12 +1000)]
powerpc/eeh: Remove spurious use of pci_dn in eeh_dump_dev_log

Retrieve the domain, bus, device, and function numbers from the edev.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-10-oohall@gmail.com
3 years agopowerpc/eeh: Pass eeh_dev to eeh_ops->{read|write}_config()
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:26 +0000 (18:12 +1000)]
powerpc/eeh: Pass eeh_dev to eeh_ops->{read|write}_config()

Mechanical conversion of the eeh_ops interfaces to use eeh_dev to reference
a specific device rather than pci_dn. No functional changes.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-9-oohall@gmail.com
3 years agopowerpc/eeh: Pass eeh_dev to eeh_ops->resume_notify()
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:25 +0000 (18:12 +1000)]
powerpc/eeh: Pass eeh_dev to eeh_ops->resume_notify()

Mechanical conversion of the eeh_ops interfaces to use eeh_dev to reference
a specific device rather than pci_dn. No functional changes.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-8-oohall@gmail.com
3 years agopowerpc/eeh: Pass eeh_dev to eeh_ops->restore_config()
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:24 +0000 (18:12 +1000)]
powerpc/eeh: Pass eeh_dev to eeh_ops->restore_config()

Mechanical conversion of the eeh_ops interfaces to use eeh_dev to reference
a specific device rather than pci_dn. No functional changes.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-7-oohall@gmail.com
3 years agopowerpc/eeh: Remove VF config space restoration
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:23 +0000 (18:12 +1000)]
powerpc/eeh: Remove VF config space restoration

There's a bunch of strange things about this code. First up is that none of
the fields being written to are functional for a VF. The SR-IOV
specification lists then as "Reserved, but OS should preserve" so writing
new values to them doesn't do anything and is clearly wrong from a
correctness perspective.

However, since VFs are designed to be managed by the OS there is an
argument to be made that we should be saving and restoring some parts of
config space. We already sort of do that by saving the first 64 bytes of
config space in the eeh_dev (see eeh_dev->config_space[]). This is
inadequate since it doesn't even consider saving and restoring the PCI
capability structures. However, this is a problem with EEH in general and
that needs to be fixed for non-VF devices too.

There's no real reason to keep around this around so delete it.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-6-oohall@gmail.com
3 years agopowerpc/eeh: Kill off eeh_ops->get_pe_addr()
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:22 +0000 (18:12 +1000)]
powerpc/eeh: Kill off eeh_ops->get_pe_addr()

This is used in precisely one place which is in pseries specific platform
code.  There's no need to have the callback in eeh_ops since the platform
chooses the EEH PE addresses anyway. The PowerNV implementation has always
been a stub too so remove it.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-5-oohall@gmail.com
3 years agopowerpc/pseries: Stop using pdn->pe_number
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:21 +0000 (18:12 +1000)]
powerpc/pseries: Stop using pdn->pe_number

The pci_dn->pe_number field is mainly used to track the IODA PE number of a
device on PowerNV. At some point it grew a user in the pseries SR-IOV
support which muddies the waters a bit, so remove it.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-4-oohall@gmail.com
3 years agopowerpc/eeh: Move vf_index out of pci_dn and into eeh_dev
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:20 +0000 (18:12 +1000)]
powerpc/eeh: Move vf_index out of pci_dn and into eeh_dev

Drivers that do not support the PCI error handling callbacks are handled by
tearing down the device and re-probing them. If the device being removed is
a virtual function then we need to know the VF index so it can be removed
using the pci_iov_{add|remove}_virtfn() API.

Currently this is handled by looking up the pci_dn, and using the vf_index
that was stashed there when the pci_dn for the VF was created in
pcibios_sriov_enable(). We would like to eliminate the use of pci_dn
outside of pseries though so we need to provide the generic EEH code with
some other way to find the vf_index.

The easiest thing to do here is move the vf_index field out of pci_dn and
into eeh_dev.  Currently pci_dn and eeh_dev are allocated and initialized
together so this is a fairly minimal change in preparation for splitting
pci_dn and eeh_dev in the future.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-3-oohall@gmail.com
3 years agopowerpc/eeh: Remove eeh_dev.c
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:19 +0000 (18:12 +1000)]
powerpc/eeh: Remove eeh_dev.c

The only thing in this file is eeh_dev_init() which is allocates and
initialises an eeh_dev based on a pci_dn. This is only ever called from
pci_dn.c so move it into there and remove the file.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-2-oohall@gmail.com
3 years agopowerpc/eeh: Remove eeh_dev_phb_init_dynamic()
Oliver O'Halloran [Sat, 25 Jul 2020 08:12:18 +0000 (18:12 +1000)]
powerpc/eeh: Remove eeh_dev_phb_init_dynamic()

This function is a one line wrapper around eeh_phb_pe_create() and despite
the name it doesn't create any eeh_dev structures. Replace it with direct
calls to eeh_phb_pe_create() since that does what it says on the tin
and removes a layer of indirection.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200725081231.39076-1-oohall@gmail.com
3 years agopowerpc/watchpoint: Remove 512 byte boundary
Ravi Bangoria [Thu, 23 Jul 2020 09:08:13 +0000 (14:38 +0530)]
powerpc/watchpoint: Remove 512 byte boundary

Power10 has removed 512 bytes boundary from match criteria i.e. the watch
range can cross 512 bytes boundary.

Note: ISA 3.1 Book III 9.4 match criteria includes 512 byte limit but that
is a documentation mistake and hopefully will be fixed in the next version
of ISA. Though, ISA 3.1 change log mentions about removal of 512B boundary:

  Multiple DEAW:
  Added a second Data Address Watchpoint. [H]DAR is
  set to the first byte of overlap. 512B boundary is
  removed.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-11-ravi.bangoria@linux.ibm.com
3 years agopowerpc/watchpoint: Return available watchpoints dynamically
Ravi Bangoria [Thu, 23 Jul 2020 09:08:12 +0000 (14:38 +0530)]
powerpc/watchpoint: Return available watchpoints dynamically

So far Book3S Powerpc supported only one watchpoint. Power10 is
introducing 2nd DAWR. Enable 2nd DAWR support for Power10.
Availability of 2nd DAWR will depend on CPU_FTR_DAWR1.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-10-ravi.bangoria@linux.ibm.com
3 years agopowerpc/watchpoint: Guest support for 2nd DAWR hcall
Ravi Bangoria [Thu, 23 Jul 2020 09:08:11 +0000 (14:38 +0530)]
powerpc/watchpoint: Guest support for 2nd DAWR hcall

2nd DAWR can be set/unset using H_SET_MODE hcall with resource value 5.
Enable powervm guest support with that. This has no effect on kvm guest
because kvm will return error if guest does hcall with resource value 5.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-9-ravi.bangoria@linux.ibm.com
3 years agopowerpc/watchpoint: Rename current H_SET_MODE DAWR macro
Ravi Bangoria [Thu, 23 Jul 2020 09:08:10 +0000 (14:38 +0530)]
powerpc/watchpoint: Rename current H_SET_MODE DAWR macro

Current H_SET_MODE hcall macro name for setting/resetting DAWR0 is
H_SET_MODE_RESOURCE_SET_DAWR. Add suffix 0 to macro name as well.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Reviewed-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-8-ravi.bangoria@linux.ibm.com
3 years agopowerpc/watchpoint: Set CPU_FTR_DAWR1 based on pa-features bit
Ravi Bangoria [Thu, 23 Jul 2020 09:08:09 +0000 (14:38 +0530)]
powerpc/watchpoint: Set CPU_FTR_DAWR1 based on pa-features bit

As per the PAPR, bit 0 of byte 64 in pa-features property indicates
availability of 2nd DAWR registers. i.e. If this bit is set, 2nd
DAWR is present, otherwise not. Host generally uses "cpu-features",
which masks "pa-features". But "cpu-features" are still not used for
guests and thus this change is mostly applicable for guests only.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Tested-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-7-ravi.bangoria@linux.ibm.com
3 years agopowerpc/dt_cpu_ftrs: Add feature for 2nd DAWR
Ravi Bangoria [Thu, 23 Jul 2020 09:08:08 +0000 (14:38 +0530)]
powerpc/dt_cpu_ftrs: Add feature for 2nd DAWR

Add new device-tree feature for 2nd DAWR. If this feature is present,
2nd DAWR is supported, otherwise not.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-6-ravi.bangoria@linux.ibm.com
3 years agopowerpc/watchpoint: Enable watchpoint functionality on power10 guest
Ravi Bangoria [Thu, 23 Jul 2020 09:08:07 +0000 (14:38 +0530)]
powerpc/watchpoint: Enable watchpoint functionality on power10 guest

CPU_FTR_DAWR is by default enabled for host via CPU_FTRS_DT_CPU_BASE
(controlled by CONFIG_PPC_DT_CPU_FTRS). But cpu-features device-tree
node is not PAPR compatible and thus not yet used by kvm or pHyp
guests. Enable watchpoint functionality on power10 guest (both kvm
and powervm) by adding CPU_FTR_DAWR to CPU_FTRS_POWER10. Note that
this change does not enable 2nd DAWR support.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Tested-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-5-ravi.bangoria@linux.ibm.com
3 years agopowerpc/watchpoint: Fix DAWR exception for CACHEOP
Ravi Bangoria [Thu, 23 Jul 2020 09:08:06 +0000 (14:38 +0530)]
powerpc/watchpoint: Fix DAWR exception for CACHEOP

'ea' returned by analyse_instr() needs to be aligned down to cache
block size for CACHEOP instructions. analyse_instr() does not set
size for CACHEOP, thus size also needs to be calculated manually.

Fixes: 27985b2a640e ("powerpc/watchpoint: Don't ignore extraneous exceptions blindly")
Fixes: 74c6881019b7 ("powerpc/watchpoint: Prepare handler to handle more than one watchpoint")
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-4-ravi.bangoria@linux.ibm.com
3 years agopowerpc/watchpoint: Fix DAWR exception constraint
Ravi Bangoria [Thu, 23 Jul 2020 09:08:05 +0000 (14:38 +0530)]
powerpc/watchpoint: Fix DAWR exception constraint

Pedro Miraglia Franco de Carvalho noticed that on p8/p9, DAR value is
inconsistent with different type of load/store. Like for byte,word
etc. load/stores, DAR is set to the address of the first byte of
overlap between watch range and real access. But for quadword load/
store it's sometime set to the address of the first byte of real
access whereas sometime set to the address of the first byte of
overlap. This issue has been fixed in p10. In p10(ISA 3.1), DAR is
always set to the address of the first byte of overlap. Commit 27985b2a640e
("powerpc/watchpoint: Don't ignore extraneous exceptions blindly")
wrongly assumes that DAR is set to the address of the first byte of
overlap for all load/stores on p8/p9 as well. Fix that. With the fix,
we now rely on 'ea' provided by analyse_instr(). If analyse_instr()
fails, generate event unconditionally on p8/p9, and on p10 generate
event only if DAR is within a DAWR range.

Note: 8xx is not affected.

Fixes: 27985b2a640e ("powerpc/watchpoint: Don't ignore extraneous exceptions blindly")
Fixes: 74c6881019b7 ("powerpc/watchpoint: Prepare handler to handle more than one watchpoint")
Reported-by: Pedro Miraglia Franco de Carvalho <pedromfc@br.ibm.com>
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-3-ravi.bangoria@linux.ibm.com
3 years agopowerpc/watchpoint: Fix 512 byte boundary limit
Ravi Bangoria [Thu, 23 Jul 2020 09:08:04 +0000 (14:38 +0530)]
powerpc/watchpoint: Fix 512 byte boundary limit

Milton Miller reported that we are aligning start and end address to
wrong size SZ_512M. It should be SZ_512. Fix that.

While doing this change I also found a case where ALIGN() comparison
fails. Within a given aligned range, ALIGN() of two addresses does not
match when start address is pointing to the first byte and end address
is pointing to any other byte except the first one. But that's not true
for ALIGN_DOWN(). ALIGN_DOWN() of any two addresses within that range
will always point to the first byte. So use ALIGN_DOWN() instead of
ALIGN().

Fixes: e68ef121c1f4 ("powerpc/watchpoint: Use builtin ALIGN*() macros")
Reported-by: Milton Miller <miltonm@us.ibm.com>
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Tested-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200723090813.303838-2-ravi.bangoria@linux.ibm.com
3 years agopowerpc/book3s64/pkey: Disable pkey on POWER6 and before
Aneesh Kumar K.V [Sun, 26 Jul 2020 13:25:17 +0000 (18:55 +0530)]
powerpc/book3s64/pkey: Disable pkey on POWER6 and before

POWER6 only supports AMR update via privileged mode (MSR[PR] = 0,
SPRN_AMR=29) The PR=1 (userspace) alias for that SPR (SPRN_AMR=13) was
only supported from POWER7. Since we don't allow userspace modifying
of AMR value we should disable pkey support on P6 and before.

The hypervisor will still report pkey support via
"ibm,processor-storage-keys". Hence also check for P7 CPU_FTR bit to
decide on pkey support.

Fixes: f491fe3fb41e ("powerpc/book3s64/pkeys: Simplify the key initialization")
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200726132517.399076-1-aneesh.kumar@linux.ibm.com
3 years agopowerpc/sstep: Fix incorrect CONFIG symbol in scv handling
Michael Ellerman [Fri, 24 Jul 2020 10:42:54 +0000 (20:42 +1000)]
powerpc/sstep: Fix incorrect CONFIG symbol in scv handling

When I "fixed" the ppc64e build in Nick's recent patch, I typoed the
CONFIG symbol, resulting in one that doesn't exist. Fix it to use the
correct symbol.

Reported-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Fixes: 7fa95f9adaee ("powerpc/64s: system call support for scv/rfscv instructions")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200724131609.1640533-1-mpe@ellerman.id.au
3 years agopowerpc/test_emulate_sstep: Fix build error
Michael Ellerman [Thu, 23 Jul 2020 23:02:26 +0000 (09:02 +1000)]
powerpc/test_emulate_sstep: Fix build error

ppc64_book3e_allmodconfig fails with:

  arch/powerpc/lib/test_emulate_step.c: In function 'test_pld':
  arch/powerpc/lib/test_emulate_step.c:113:7: error: implicit declaration of function 'cpu_has_feature'
    113 |  if (!cpu_has_feature(CPU_FTR_ARCH_31)) {
        |       ^~~~~~~~~~~~~~~

Add an include of cpu_has_feature.h to fix it.

Fixes: b6b54b42722a ("powerpc/sstep: Add tests for prefixed integer load/stores")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200724004109.1461709-1-mpe@ellerman.id.au
3 years agoMerge branch 'scv' support into next
Michael Ellerman [Thu, 23 Jul 2020 07:43:44 +0000 (17:43 +1000)]
Merge branch 'scv' support into next

From Nick's cover letter:

Linux powerpc new system call instruction and ABI

System Call Vectored (scv) ABI
==============================

The scv instruction is introduced with POWER9 / ISA3, it comes with an
rfscv counter-part. The benefit of these instructions is
performance (trading slower SRR0/1 with faster LR/CTR registers, and
entering the kernel with MSR[EE] and MSR[RI] left enabled, which can
reduce MSR updates. The scv instruction has 128 levels (not enough to
cover the Linux system call space).

Assignment and advertisement
----------------------------
The proposal is to assign scv levels conservatively, and advertise
them with HWCAP feature bits as we add support for more.

Linux has not enabled FSCR[SCV] yet, so executing the scv instruction
will cause the kernel to log a "SCV facility unavilable" message, and
deliver a SIGILL with ILL_ILLOPC to the process. Linux has defined a
HWCAP2 bit PPC_FEATURE2_SCV for SCV support, but does not set it.

This change allocates the zero level ('scv 0'), advertised with
PPC_FEATURE2_SCV, which will be used to provide normal Linux system
calls (equivalent to 'sc').

Attempting to execute scv with other levels will cause a SIGILL to be
delivered the same as before, but will not log a "SCV facility
unavailable" message (because the processor facility is enabled).

Calling convention
------------------
The proposal is for scv 0 to provide the standard Linux system call
ABI with the following differences from sc convention[1]:

- LR is to be volatile across scv calls. This is necessary because the
  scv instruction clobbers LR. From previous discussion, this should
  be possible to deal with in GCC clobbers and CFI.

- cr1 and cr5-cr7 are volatile. This matches the C ABI and would allow
  the kernel system call exit to avoid restoring the volatile cr
  registers (although we probably still would anyway to avoid
  information leaks).

- Error handling: The consensus among kernel, glibc, and musl is to
  move to using negative return values in r3 rather than CR0[SO]=1 to
  indicate error, which matches most other architectures, and is
  closer to a function call.

Notes
-----
- r0,r4-r8 are documented as volatile in the ABI, but the kernel patch
  as submitted currently preserves them. This is to leave room for
  deciding which way to go with these. Some small benefit was found by
  preserving them[1] but I'm not convinced it's worth deviating from
  the C function call ABI just for this. Release code should follow
  the ABI.

Previous discussions:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-April/208691.html
https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-April/209268.html

[1] https://github.com/torvalds/linux/blob/master/Documentation/powerpc/syscall64-abi.rst
[2] https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-April/209263.html

3 years agoselftests/powerpc: Add test of memcmp at end of page
Michael Ellerman [Wed, 22 Jul 2020 05:53:15 +0000 (15:53 +1000)]
selftests/powerpc: Add test of memcmp at end of page

Update our memcmp selftest, to test the case where we're comparing up
to the end of a page and the subsequent page is not mapped. We have to
make sure we don't read off the end of the page and cause a fault.

We had a bug there in the past, fixed in commit
d9470757398a ("powerpc/64: Fix memcmp reading past the end of src/dest").

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200722055315.962391-1-mpe@ellerman.id.au
3 years agopowerpc/powernv/idle: Exclude mfspr on HID1, 4, 5 on P9 and above
Pratik Rajesh Sampat [Tue, 21 Jul 2020 15:37:08 +0000 (21:07 +0530)]
powerpc/powernv/idle: Exclude mfspr on HID1, 4, 5 on P9 and above

POWER9 onwards the support for the registers HID1, HID4, HID5 has been
receded.
Although mfspr on the above registers worked in Power9, In Power10
simulator is unrecognized. Moving their assignment under the
check for machines lower than Power9

Signed-off-by: Pratik Rajesh Sampat <psampat@linux.ibm.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200721153708.89057-4-psampat@linux.ibm.com
3 years agopowerpc/powernv/idle: Rename pnv_first_spr_loss_level variable
Pratik Rajesh Sampat [Tue, 21 Jul 2020 15:37:07 +0000 (21:07 +0530)]
powerpc/powernv/idle: Rename pnv_first_spr_loss_level variable

Replace the variable name from using "pnv_first_spr_loss_level" to
"deep_spr_loss_state".

pnv_first_spr_loss_level is supposed to be the earliest state that
has OPAL_PM_LOSE_FULL_CONTEXT set, in other places the kernel uses the
"deep" states as terminology. Hence renaming the variable to be coherent
to its semantics.

Signed-off-by: Pratik Rajesh Sampat <psampat@linux.ibm.com>
Acked-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200721153708.89057-3-psampat@linux.ibm.com
3 years agopowerpc/powernv/idle: Replace CPU feature check with PVR check
Pratik Rajesh Sampat [Tue, 21 Jul 2020 15:37:06 +0000 (21:07 +0530)]
powerpc/powernv/idle: Replace CPU feature check with PVR check

The POWER9 idle driver contains implementation-specific details that
means it is not suitable to run on any processor that implements ISA
v3.0 (e.g., POWER10), so only init the driver when running on a
POWER9.

Signed-off-by: Pratik Rajesh Sampat <psampat@linux.ibm.com>
[mpe: Use updated change log from Nick]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200721153708.89057-2-psampat@linux.ibm.com
3 years agopowerpc/mm/hash64: Remove comment that is no longer valid
Santosh Sivaraj [Tue, 21 Jul 2020 09:19:15 +0000 (14:49 +0530)]
powerpc/mm/hash64: Remove comment that is no longer valid

hash_low_64.S was removed in commit a43c0eb8364c ("powerpc/mm: Convert
4k insert from asm to C") and flush_hash_page() is no longer called
from any assembly routine.

Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
[mpe: Tweak comment wording]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200721091915.205006-1-santosh@fossix.org
3 years agoKVM: PPC: Fix typo on H_DISABLE_AND_GET hcall
Leonardo Bras [Tue, 7 Jul 2020 00:48:12 +0000 (21:48 -0300)]
KVM: PPC: Fix typo on H_DISABLE_AND_GET hcall

On PAPR+ the hcall() on 0x1B0 is called H_DISABLE_AND_GET, but got
defined as H_DISABLE_AND_GETC instead.

This define was introduced with a typo in commit <b13a96cfb055>
("[PATCH] powerpc: Extends HCALL interface for InfiniBand usage"), and was
later used without having the typo noticed.

Signed-off-by: Leonardo Bras <leobras.c@gmail.com>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200707004812.190765-1-leobras.c@gmail.com
3 years agopowerpc/spufs: Fix the type of ret in spufs_arch_write_note
Christoph Hellwig [Wed, 10 Jun 2020 08:55:54 +0000 (10:55 +0200)]
powerpc/spufs: Fix the type of ret in spufs_arch_write_note

Both the ->dump method and snprintf return an int.  So switch to an
int and properly handle errors from ->dump.

Fixes: 5456ffdee666 ("powerpc/spufs: simplify spufs core dumping")
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200610085554.5647-1-hch@lst.de
3 years agopowerpc/powernv: Machine check handler for POWER10
Nicholas Piggin [Thu, 2 Jul 2020 23:33:43 +0000 (09:33 +1000)]
powerpc/powernv: Machine check handler for POWER10

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200702233343.1128026-1-npiggin@gmail.com
3 years agopowerpc/pseries: PCIE PHB reset
Wen Xiong [Mon, 13 Jul 2020 14:39:33 +0000 (09:39 -0500)]
powerpc/pseries: PCIE PHB reset

Several device drivers hit EEH(Extended Error handling) when
triggering kdump on Pseries PowerVM. This patch implemented a reset of
the PHBs in pci general code when triggering kdump. PHB reset stop all
PCI transactions from normal kernel. We have tested the patch in
several enviroments:
  - direct slot adapters
  - adapters under the switch
  - a VF adapter in PowerVM
  - a VF adapter/adapter in KVM guest.

Signed-off-by: Wen Xiong <wenxiong@linux.vnet.ibm.com>
[mpe: Fix broken whitespace, subject & SOB formatting]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594651173-32166-1-git-send-email-wenxiong@linux.vnet.ibm.com
3 years agopowerpc: Select ARCH_HAS_MEMBARRIER_SYNC_CORE
Nicholas Piggin [Thu, 16 Jul 2020 01:35:22 +0000 (11:35 +1000)]
powerpc: Select ARCH_HAS_MEMBARRIER_SYNC_CORE

powerpc return from interrupt and return from system call sequences
are context synchronising.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200716013522.338318-1-npiggin@gmail.com
3 years agopowerpc/64: Fix an out of date comment about MMIO ordering
Palmer Dabbelt [Thu, 16 Jul 2020 19:38:20 +0000 (12:38 -0700)]
powerpc/64: Fix an out of date comment about MMIO ordering

This primitive has been renamed, but because it was spelled incorrectly in the
first place it must have escaped the fixup patch.  As far as I can tell this
logic is still correct: smp_mb__after_spinlock() uses the default smp_mb()
implementation, which is "sync" rather than "hwsync" but those are the same
(though I'm not that familiar with PowerPC).

Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200716193820.1141936-1-palmer@dabbelt.com
3 years agopowerpc/test_emulate_step: Move extern declaration to sstep.h
Balamuruhan S [Fri, 26 Jun 2020 09:51:58 +0000 (15:21 +0530)]
powerpc/test_emulate_step: Move extern declaration to sstep.h

fix checkpatch.pl warnings by moving extern declaration from source
file to headerfile.

Signed-off-by: Balamuruhan S <bala24@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200626095158.1031507-5-bala24@linux.ibm.com
3 years agopowerpc/sstep: Introduce macros to retrieve Prefix instruction operands
Balamuruhan S [Fri, 26 Jun 2020 09:51:57 +0000 (15:21 +0530)]
powerpc/sstep: Introduce macros to retrieve Prefix instruction operands

retrieve prefix instruction operands RA and pc relative bit R values
using macros and adopt it in sstep.c and test_emulate_step.c.

Signed-off-by: Balamuruhan S <bala24@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200626095158.1031507-4-bala24@linux.ibm.com
3 years agopowerpc/test_emulate_step: Add negative tests for prefixed addi
Balamuruhan S [Fri, 26 Jun 2020 09:51:56 +0000 (15:21 +0530)]
powerpc/test_emulate_step: Add negative tests for prefixed addi

testcases for `paddi` instruction to cover the negative case,
if R is equal to 1 and RA is not equal to 0, the instruction
form is invalid.

Signed-off-by: Balamuruhan S <bala24@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200626095158.1031507-3-bala24@linux.ibm.com
3 years agopowerpc/test_emulate_step: Enhancement to test negative scenarios
Balamuruhan S [Fri, 26 Jun 2020 09:51:55 +0000 (15:21 +0530)]
powerpc/test_emulate_step: Enhancement to test negative scenarios

add provision to declare test is a negative scenario, verify
whether emulation fails and avoid executing it.

Signed-off-by: Balamuruhan S <bala24@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200626095158.1031507-2-bala24@linux.ibm.com
3 years agopowerpc/xmon: Improve dumping prefixed instructions
Jordan Niethe [Tue, 2 Jun 2020 05:27:26 +0000 (15:27 +1000)]
powerpc/xmon: Improve dumping prefixed instructions

Currently prefixed instructions are dumped as two separate word
instructions. Use mread_instr() so that prefixed instructions are read
as such and update the incrementor in the loop to take this into
account.

'dump_func' is print_insn_powerpc() which comes from ppc-dis.c which is
taken from binutils. When this is updated prefixed instructions will be
disassembled.

Currently dumping prefixed instructions looks like this:
0:mon> di c000000000094168
c000000000094168  0x06000000    .long 0x6000000
c00000000009416c  0x392a0003    addi    r9,r10,3
c000000000094170  0x913f0028    stw     r9,40(r31)
c000000000094174  0xe93f002a    lwa     r9,40(r31)
c000000000094178  0x7d234b78    mr      r3,r9
c00000000009417c  0x383f0040    addi    r1,r31,64
c000000000094180  0xebe1fff8    ld      r31,-8(r1)
c000000000094184  0x4e800020    blr
c000000000094188  0x60000000    nop
 ...
c000000000094190  0x3c4c0121    addis   r2,r12,289
c000000000094194  0x38429670    addi    r2,r2,-27024
c000000000094198  0x7c0802a6    mflr    r0
c00000000009419c  0x60000000    nop
c0000000000941a0  0xe9240100    ld      r9,256(r4)
c0000000000941a4  0x39400001    li      r10,1

After this it looks like:
0:mon> di c000000000094168
c000000000094168  0x06000000 0x392a0003 .long 0x392a000306000000
c000000000094170  0x913f0028    stw     r9,40(r31)
c000000000094174  0xe93f002a    lwa     r9,40(r31)
c000000000094178  0x7d234b78    mr      r3,r9
c00000000009417c  0x383f0040    addi    r1,r31,64
c000000000094180  0xebe1fff8    ld      r31,-8(r1)
c000000000094184  0x4e800020    blr
c000000000094188  0x60000000    nop
 ...
c000000000094190  0x3c4c0121    addis   r2,r12,289
c000000000094194  0x38429570    addi    r2,r2,-27280
c000000000094198  0x7c0802a6    mflr    r0
c00000000009419c  0x60000000    nop
c0000000000941a0  0xe9240100    ld      r9,256(r4)
c0000000000941a4  0x39400001    li      r10,1
c0000000000941a8  0x3d02000b    addis   r8,r2,11

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200602052728.18227-2-jniethe5@gmail.com
3 years agopowerpc: Add a ppc_inst_as_str() helper
Jordan Niethe [Tue, 2 Jun 2020 05:27:25 +0000 (15:27 +1000)]
powerpc: Add a ppc_inst_as_str() helper

There are quite a few places where instructions are printed, this is
done using a '%x' format specifier. With the introduction of prefixed
instructions, this does not work well. Currently in these places,
ppc_inst_val() is used for the value for %x so only the first word of
prefixed instructions are printed.

When the instructions are word instructions, only a single word should
be printed. For prefixed instructions both the prefix and suffix should
be printed. To accommodate both of these situations, instead of a '%x'
specifier use '%s' and introduce a helper, __ppc_inst_as_str() which
returns a char *. The char * __ppc_inst_as_str() returns is buffer that
is passed to it by the caller.

It is cumbersome to require every caller of __ppc_inst_as_str() to now
declare a buffer. To make it more convenient to use __ppc_inst_as_str(),
wrap it in a macro that uses a compound statement to allocate a buffer
on the caller's stack before calling it.

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Acked-by: Segher Boessenkool <segher@kernel.crashing.org>
[mpe: Drop 0x prefix to match most existings uses, especially xmon]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200602052728.18227-1-jniethe5@gmail.com
3 years agopowerpc/sstep: Add tests for Prefixed Add Immediate
Jordan Niethe [Mon, 25 May 2020 02:59:23 +0000 (12:59 +1000)]
powerpc/sstep: Add tests for Prefixed Add Immediate

Use the existing support for testing compute type instructions to test
Prefixed Add Immediate (paddi). The R bit of the paddi instruction
controls whether current instruction address is used. Add test cases
for when R=1 and for R=0. paddi has a 34 bit immediate field formed by
concatenating si0 and si1. Add tests for the extreme values of this
field.

Skip the paddi tests if ISA v3.1 is unsupported.

Some of these test cases were added by Balamuruhan S.

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
[mpe: Fix conflicts with ppc-opcode.h changes, squash in .balign]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200525025923.19843-5-jniethe5@gmail.com
3 years agopowerpc/sstep: Let compute tests specify a required cpu feature
Jordan Niethe [Mon, 25 May 2020 02:59:22 +0000 (12:59 +1000)]
powerpc/sstep: Let compute tests specify a required cpu feature

An a array of struct compute_test's are used to declare tests for
compute instructions. Add a cpu_feature field to struct compute_test as
an optional way to specify a cpu feature that must be present. If not
present then skip the test.

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200525025923.19843-4-jniethe5@gmail.com
3 years agopowerpc/sstep: Set NIP in instruction emulation tests
Jordan Niethe [Mon, 25 May 2020 02:59:21 +0000 (12:59 +1000)]
powerpc/sstep: Set NIP in instruction emulation tests

The tests for emulation of compute instructions execute and
emulate an instruction and then compare the results to verify the
emulation. In ISA v3.1 there are instructions that operate relative to
the NIP. Therefore set the NIP in the regs used for the emulated
instruction to the location of the executed instruction so they will
give the same result.

This is a rework of a patch by Balamuruhan S.

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200525025923.19843-3-jniethe5@gmail.com
3 years agopowerpc/sstep: Add tests for prefixed floating-point load/stores
Jordan Niethe [Mon, 25 May 2020 02:59:20 +0000 (12:59 +1000)]
powerpc/sstep: Add tests for prefixed floating-point load/stores

Add tests for the prefixed versions of the floating-point load/stores
that are currently tested. This includes the following instructions:
  * Prefixed Load Floating-Point Single (plfs)
  * Prefixed Load Floating-Point Double (plfd)
  * Prefixed Store Floating-Point Single (pstfs)
  * Prefixed Store Floating-Point Double (pstfd)

Skip the new tests if ISA v3.10 is unsupported.

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
[mpe: Fix conflicts with ppc-opcode.h changes]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200525025923.19843-2-jniethe5@gmail.com
3 years agopowerpc/sstep: Add tests for prefixed integer load/stores
Jordan Niethe [Mon, 25 May 2020 02:59:19 +0000 (12:59 +1000)]
powerpc/sstep: Add tests for prefixed integer load/stores

Add tests for the prefixed versions of the integer load/stores that
are currently tested. This includes the following instructions:
  * Prefixed Load Doubleword (pld)
  * Prefixed Load Word and Zero (plwz)
  * Prefixed Store Doubleword (pstd)

Skip the new tests if ISA v3.1 is unsupported.

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
[mpe: Fix conflicts with ppc-opcode.h changes]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200525025923.19843-1-jniethe5@gmail.com
3 years agopowerpc/64s: system call support for scv/rfscv instructions
Nicholas Piggin [Thu, 11 Jun 2020 08:12:03 +0000 (18:12 +1000)]
powerpc/64s: system call support for scv/rfscv instructions

Add support for the scv instruction on POWER9 and later CPUs.

For now this implements the zeroth scv vector 'scv 0', as identical to
'sc' system calls, with the exception that LR is not preserved, nor
are volatile CR registers, and error is not indicated with CR0[SO],
but by returning a negative errno.

rfscv is implemented to return from scv type system calls. It can not
be used to return from sc system calls because those are defined to
preserve LR.

getpid syscall throughput on POWER9 is improved by 26% (428 to 318
cycles), largely due to reducing mtmsr and mtspr.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Fix ppc64e build]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200611081203.995112-3-npiggin@gmail.com
3 years agopowerpc/64s/exception: treat NIA below __end_interrupts as soft-masked
Nicholas Piggin [Thu, 11 Jun 2020 08:12:02 +0000 (18:12 +1000)]
powerpc/64s/exception: treat NIA below __end_interrupts as soft-masked

The scv instruction causes an interrupt which can enter the kernel with
MSR[EE]=1, thus allowing interrupts to hit at any time. These must not
be taken as normal interrupts, because they come from MSR[PR]=0 context,
and yet the kernel stack is not yet set up and r13 is not set to the
PACA).

Treat this as a soft-masked interrupt regardless of the soft masked
state. This does not affect behaviour yet, because currently all
interrupts are taken with MSR[EE]=0.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200611081203.995112-2-npiggin@gmail.com
3 years agopowerpc/perf: BHRB control to disable BHRB logic when not used
Athira Rajeev [Fri, 17 Jul 2020 14:38:23 +0000 (10:38 -0400)]
powerpc/perf: BHRB control to disable BHRB logic when not used

PowerISA v3.1 has few updates for the Branch History Rolling
Buffer(BHRB).

BHRB disable is controlled via Monitor Mode Control Register A (MMCRA)
bit, namely "BHRB Recording Disable (BHRBRD)". This field controls
whether BHRB entries are written when BHRB recording is enabled by
other bits. This patch implements support for this BHRB disable bit.
By setting 0b1 to this bit will disable the BHRB and by setting 0b0 to
this bit will have BHRB enabled. This addresses backward
compatibility (for older OS), since this bit will be cleared and
hardware will be writing to BHRB by default.

This patch addresses changes to set MMCRA (BHRBRD) at boot for
power10 (there by the core will run faster) and enable this feature
only on runtime ie, on explicit need from user. Also save/restore
MMCRA in the restore path of state-loss idle state to make sure we
keep BHRB disabled if it was not enabled on request at runtime.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-12-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/perf: Add Power10 BHRB filter support for PERF_SAMPLE_BRANCH_IND_CALL/COND
Athira Rajeev [Fri, 17 Jul 2020 14:38:22 +0000 (10:38 -0400)]
powerpc/perf: Add Power10 BHRB filter support for PERF_SAMPLE_BRANCH_IND_CALL/COND

PowerISA v3.1 introduce filtering support for
PERF_SAMPLE_BRANCH_IND_CALL/COND. The patch adds BHRB filter
support for "ind_call" and "cond" in power10_bhrb_filter_map().

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-11-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/perf: Ignore the BHRB kernel address filtering for P10
Athira Rajeev [Fri, 17 Jul 2020 14:38:21 +0000 (10:38 -0400)]
powerpc/perf: Ignore the BHRB kernel address filtering for P10

Commit bb19af816025 ("powerpc/perf: Prevent kernel address leak to
userspace via BHRB buffer") added a check in bhrb_read() to filter
the kernel address from BHRB buffer. This patch modified it to avoid
that check for PowerISA v3.1 based processors, since PowerISA v3.1
allows only MSR[PR]=1 address to be written to BHRB buffer.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-10-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/perf: power10 Performance Monitoring support
Athira Rajeev [Fri, 17 Jul 2020 14:38:20 +0000 (10:38 -0400)]
powerpc/perf: power10 Performance Monitoring support

Base enablement patch to register performance monitoring hardware
support for power10. Patch introduce the raw event encoding format,
defines the supported list of events, config fields for the event
attributes and their corresponding bit values which are exported via
sysfs.

Patch also enhances the support function in isa207_common.c to include
power10 pmu hardware.

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-9-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/perf: Add Power10 PMU feature to DT CPU features
Madhavan Srinivasan [Fri, 17 Jul 2020 14:38:19 +0000 (10:38 -0400)]
powerpc/perf: Add Power10 PMU feature to DT CPU features

Add Power10 feature function to DT CPU features, along with a Power10
specific init() to initialize PMU SPRs, sets the oprofile_cpu_type and
cpu_features. This will enable performance monitoring unit (PMU) for
Power10 in CPU features with "performance-monitor-power10".

For Power ISA v3.1, BHRB disable is controlled via Monitor Mode
Control Register A (MMCRA) bit, namely "BHRB Recording
Disable (BHRBRD)". This patch initializes MMCRA BHRBRD to disable BHRB
feature at boot for Power10.

Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
[mpe: Move MMCRA_BHRB_DISABLE as noted by jpn, drop CPU setup changes]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-8-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/xmon: Add PowerISA v3.1 PMU SPRs
Madhavan Srinivasan [Fri, 17 Jul 2020 14:38:18 +0000 (10:38 -0400)]
powerpc/xmon: Add PowerISA v3.1 PMU SPRs

PowerISA v3.1 added three new perfromance
monitoring unit (PMU) speical purpose register (SPR).
They are Monitor Mode Control Register 3 (MMCR3),
Sampled Instruction Event Register 2 (SIER2),
Sampled Instruction Event Register 3 (SIER3).

Patch here adds a new dump function dump_310_sprs
to print these SPR values.

Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-7-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agoKVM: PPC: Book3S HV: Save/restore new PMU registers
Athira Rajeev [Fri, 17 Jul 2020 14:38:17 +0000 (10:38 -0400)]
KVM: PPC: Book3S HV: Save/restore new PMU registers

Power ISA v3.1 has added new performance monitoring unit (PMU) special
purpose registers (SPRs). They are:

Monitor Mode Control Register 3 (MMCR3)
Sampled Instruction Event Register A (SIER2)
Sampled Instruction Event Register B (SIER3)

Add support to save/restore these new SPRs while entering/exiting
guest. Also include changes to support KVM_REG_PPC_MMCR3/SIER2/SIER3.
Add new SPRs to KVM API documentation.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-6-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/perf: Add support for ISA3.1 PMU SPRs
Madhavan Srinivasan [Fri, 17 Jul 2020 14:38:16 +0000 (10:38 -0400)]
powerpc/perf: Add support for ISA3.1 PMU SPRs

PowerISA v3.1 includes new performance monitoring unit(PMU)
special purpose registers (SPRs). They are

Monitor Mode Control Register 3 (MMCR3)
Sampled Instruction Event Register 2 (SIER2)
Sampled Instruction Event Register 3 (SIER3)

MMCR3 is added for further sampling related configuration
control. SIER2/SIER3 are added to provide additional
information about the sampled instruction.

Patch adds new PPMU flag called "PPMU_ARCH_31" to support handling of
these new SPRs, updates the struct thread_struct to include these new
SPRs, include MMCR3 in struct mmcr_regs. This is needed to support
programming of MMCR3 SPR during event_enable/disable. Patch also adds
the sysfs support for the MMCR3 SPR along with SPRN_ macros for these
new pmu SPRs.

Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
[mpe: Rename to PPMU_ARCH_31 as noted by jpn]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-5-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/perf: Update Power PMU cache_events to u64 type
Athira Rajeev [Fri, 17 Jul 2020 14:38:15 +0000 (10:38 -0400)]
powerpc/perf: Update Power PMU cache_events to u64 type

Events of type PERF_TYPE_HW_CACHE was described for Power PMU
as: int (*cache_events)[type][op][result];

where type, op, result values unpacked from the event attribute config
value is used to generate the raw event code at runtime.

So far the event code values which used to create these cache-related
events were within 32 bit and `int` type worked. In power10,
some of the event codes are of 64-bit value and hence update the
Power PMU cache_events to `u64` type in `power_pmu` struct.
Also propagate this change to existing all PMU driver code paths
which are using ppmu->cache_events.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-4-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agoKVM: PPC: Book3S HV: Cleanup updates for kvm vcpu MMCR
Athira Rajeev [Fri, 17 Jul 2020 14:38:14 +0000 (10:38 -0400)]
KVM: PPC: Book3S HV: Cleanup updates for kvm vcpu MMCR

Currently `kvm_vcpu_arch` stores all Monitor Mode Control registers
in a flat array in order: mmcr0, mmcr1, mmcra, mmcr2, mmcrs
Split this to give mmcra and mmcrs its own entries in vcpu and
use a flat array for mmcr0 to mmcr2. This patch implements this
cleanup to make code easier to read.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
[mpe: Fix MMCRA/MMCR2 uapi breakage as noted by paulus]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-3-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/perf: Update cpu_hw_event to use `struct` for storing MMCR registers
Athira Rajeev [Fri, 17 Jul 2020 14:38:13 +0000 (10:38 -0400)]
powerpc/perf: Update cpu_hw_event to use `struct` for storing MMCR registers

core-book3s currently uses array to store the MMCR registers as part
of per-cpu `cpu_hw_events`. This patch does a clean up to use `struct`
to store mmcr regs instead of array. This will make code easier to read
and reduces chance of any subtle bug that may come in the future, say
when new registers are added. Patch updates all relevant code that was
using MMCR array ( cpuhw->mmcr[x]) to use newly introduced `struct`.
This includes the PMU driver code for supported platforms (power5
to power9) and ISA macros for counter support functions.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1594996707-3727-2-git-send-email-atrajeev@linux.vnet.ibm.com
3 years agopowerpc/perf: Fix missing is_sier_aviable() during build
Madhavan Srinivasan [Sun, 14 Jun 2020 08:36:04 +0000 (14:06 +0530)]
powerpc/perf: Fix missing is_sier_aviable() during build

Compilation error:
  arch/powerpc/perf/perf_regs.c:80:undefined reference to `.is_sier_available'

Currently is_sier_available() is part of core-book3s.c, which is added
to build based on CONFIG_PPC_PERF_CTRS.

A config with CONFIG_PERF_EVENTS and without CONFIG_PPC_PERF_CTRS will
have a build break because of missing is_sier_available().

In practice it only breaks when CONFIG_FSL_EMB_PERF_EVENT=n because
that also guards the usage of is_sier_available(). That only happens
with CONFIG_PPC_BOOK3E_64=y and CONFIG_FSL_SOC_BOOKE=n.

Patch adds is_sier_available() in asm/perf_event.h to fix the build
break for configs missing CONFIG_PPC_PERF_CTRS.

Fixes: 333804dc3b7a ("powerpc/perf: Update perf_regs structure to include SIER")
Reported-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
[mpe: Add detail about CONFIG_FSL_SOC_BOOKE]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200614083604.302611-1-maddy@linux.ibm.com
3 years agopowerpc/64s/hash: Disable subpage_prot syscall by default
Nicholas Piggin [Fri, 3 Jul 2020 01:19:58 +0000 (11:19 +1000)]
powerpc/64s/hash: Disable subpage_prot syscall by default

The subpage_prot syscall was added for specialised system software
(Lx86) that has been discontinued for about 7 years, and is not thought
to be used elsewhere, so disable it by default.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200703011958.1166620-4-npiggin@gmail.com
3 years agopowerpc/64s: Remove PROT_SAO support
Nicholas Piggin [Fri, 3 Jul 2020 01:19:57 +0000 (11:19 +1000)]
powerpc/64s: Remove PROT_SAO support

ISA v3.1 does not support the SAO storage control attribute required to
implement PROT_SAO. PROT_SAO was used by specialised system software
(Lx86) that has been discontinued for about 7 years, and is not thought
to be used elsewhere, so removal should not cause problems.

We rather remove it than keep support for older processors, because
live migrating guest partitions to newer processors may not be possible
if SAO is in use (or worse allowed with silent races).

- PROT_SAO stays in the uapi header so code using it would still build.
- arch_validate_prot() is removed, the generic version rejects PROT_SAO
  so applications would get a failure at mmap() time.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Drop KVM change for the time being]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200703011958.1166620-3-npiggin@gmail.com
3 years agopowerpc: Remove stale calc_vm_prot_bits() comment
Nicholas Piggin [Fri, 3 Jul 2020 01:19:56 +0000 (11:19 +1000)]
powerpc: Remove stale calc_vm_prot_bits() comment

This comment is wrong, we wouldn't use calc_vm_prot_bits() here
because we are being called by calc_vm_prot_bits() to modify its
behaviour.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200703011958.1166620-2-npiggin@gmail.com
3 years agopowerpc: Remove unneeded inline functions
YueHaibing [Fri, 17 Jul 2020 11:27:14 +0000 (19:27 +0800)]
powerpc: Remove unneeded inline functions

Both of those functions are only called from 64-bit only code, so the
stubs should not be needed at all.

Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200717112714.19304-1-yuehaibing@huawei.com
3 years agomacintosh/therm_adt746x: Replace HTTP links with HTTPS ones
Alexander A. Klimov [Fri, 17 Jul 2020 18:29:40 +0000 (20:29 +0200)]
macintosh/therm_adt746x: Replace HTTP links with HTTPS ones

Rationale:
Reduces attack surface on kernel devs opening the links for MITM
as HTTPS traffic is much harder to manipulate.

Deterministic algorithm:
For each file:
  If not .svg:
    For each line:
      If doesn't contain `\bxmlns\b`:
        For each link, `\bhttp://[^# \t\r\n]*(?:\w|/)`:
  If neither `\bgnu\.org/license`, nor `\bmozilla\.org/MPL\b`:
            If both the HTTP and HTTPS versions
            return 200 OK and serve the same content:
              Replace HTTP with HTTPS.

Signed-off-by: Alexander A. Klimov <grandmaster@al2klimov.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200717182940.75484-1-grandmaster@al2klimov.de
3 years agomacintosh/adb: Replace HTTP links with HTTPS ones
Alexander A. Klimov [Fri, 17 Jul 2020 18:35:22 +0000 (20:35 +0200)]
macintosh/adb: Replace HTTP links with HTTPS ones

Rationale:
Reduces attack surface on kernel devs opening the links for MITM
as HTTPS traffic is much harder to manipulate.

Deterministic algorithm:
For each file:
  If not .svg:
    For each line:
      If doesn't contain `\bxmlns\b`:
        For each link, `\bhttp://[^# \t\r\n]*(?:\w|/)`:
  If neither `\bgnu\.org/license`, nor `\bmozilla\.org/MPL\b`:
            If both the HTTP and HTTPS versions
            return 200 OK and serve the same content:
              Replace HTTP with HTTPS.

Signed-off-by: Alexander A. Klimov <grandmaster@al2klimov.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200717183522.77605-1-grandmaster@al2klimov.de
3 years agopowerpc: Replace HTTP links with HTTPS ones
Alexander A. Klimov [Sat, 18 Jul 2020 10:39:58 +0000 (12:39 +0200)]
powerpc: Replace HTTP links with HTTPS ones

Rationale:
Reduces attack surface on kernel devs opening the links for MITM
as HTTPS traffic is much harder to manipulate.

Deterministic algorithm:
For each file:
  If not .svg:
    For each line:
      If doesn't contain `\bxmlns\b`:
        For each link, `\bhttp://[^# \t\r\n]*(?:\w|/)`:
  If neither `\bgnu\.org/license`, nor `\bmozilla\.org/MPL\b`:
            If both the HTTP and HTTPS versions
            return 200 OK and serve the same content:
              Replace HTTP with HTTPS.

Signed-off-by: Alexander A. Klimov <grandmaster@al2klimov.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200718103958.5455-1-grandmaster@al2klimov.de
3 years agopowerpc/spufs: Rework fcheck() usage
Michael Ellerman [Fri, 8 May 2020 13:06:33 +0000 (23:06 +1000)]
powerpc/spufs: Rework fcheck() usage

Currently the spu coredump code triggers an RCU warning:

  =============================
  WARNING: suspicious RCU usage
  5.7.0-rc3-01755-g7cd49f0b7ec7 #1 Not tainted
  -----------------------------
  include/linux/fdtable.h:95 suspicious rcu_dereference_check() usage!

  other info that might help us debug this:

  rcu_scheduler_active = 2, debug_locks = 1
  1 lock held by spu-coredump/1343:
   #0: c0000007fa22f430 (sb_writers#2){.+.+}-{0:0}, at: .do_coredump+0x1010/0x13c8

  stack backtrace:
  CPU: 0 PID: 1343 Comm: spu-coredump Not tainted 5.7.0-rc3-01755-g7cd49f0b7ec7 #1
  Call Trace:
    .dump_stack+0xec/0x15c (unreliable)
    .lockdep_rcu_suspicious+0x120/0x144
    .coredump_next_context+0x148/0x158
    .spufs_coredump_extra_notes_size+0x54/0x190
    .elf_coredump_extra_notes_size+0x34/0x50
    .elf_core_dump+0xe48/0x19d0
    .do_coredump+0xe50/0x13c8
    .get_signal+0x864/0xd88
    .do_notify_resume+0x158/0x3c8
    .interrupt_exit_user_prepare+0x19c/0x208
    interrupt_return+0x14/0x1c0

This comes from fcheck_files() via fcheck().

It's pretty clearly documented that fcheck() must be wrapped with
rcu_read_lock(), adding that fixes the RCU warning.

hch points out that once we've released the RCU read lock the file may
be closed and freed, which would leave us with a pointer to a freed
spu_context.

To avoid that, take a reference to the spu_context while we hold the
RCU read lock, and drop that reference later once we're done with the
context.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200508130633.2532759-1-mpe@ellerman.id.au
3 years agoselftests/powerpc: Run per_event_excludes test on Power8 or later
Michael Ellerman [Thu, 16 Jul 2020 12:21:42 +0000 (22:21 +1000)]
selftests/powerpc: Run per_event_excludes test on Power8 or later

The per_event_excludes test wants to run on Power8 or later. But
currently it checks that AT_BASE_PLATFORM *equals* power8, which means
it only runs on Power8.

Fix it to check for the ISA 2.07 feature, which will be set on Power8
and later CPUs.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200716122142.3776261-1-mpe@ellerman.id.au
3 years agopowerpc/book3s64/pkeys: Remove is_pkey_enabled()
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:46 +0000 (08:59 +0530)]
powerpc/book3s64/pkeys: Remove is_pkey_enabled()

There is only one caller to this function and the function is wrongly
named. Avoid further confusion w.r.t name and open code this at the
only call site. Also remove read_uamor(). There are no users for
the same after this.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-24-aneesh.kumar@linux.ibm.com
3 years agoselftests/powerpc: ptrace-pkey: Don't update expected UAMOR value
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:45 +0000 (08:59 +0530)]
selftests/powerpc: ptrace-pkey: Don't update expected UAMOR value

With commit 4a4a5e5d2aad ("powerpc/pkeys: key allocation/deallocation
must not change pkey registers") we are not updating UAMOR on key
allocation. So don't update the expected uamor value in the test.

Fixes: 4a4a5e5d2aad ("powerpc/pkeys: key allocation/deallocation must not change pkey registers")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-23-aneesh.kumar@linux.ibm.com
3 years agoselftests/powerpc: ptrace-pkey: Update the test to mark an invalid pkey correctly
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:44 +0000 (08:59 +0530)]
selftests/powerpc: ptrace-pkey: Update the test to mark an invalid pkey correctly

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-22-aneesh.kumar@linux.ibm.com
3 years agoselftests/powerpc: ptrace-pkey: Rename variables to make it easier to follow code
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:43 +0000 (08:59 +0530)]
selftests/powerpc: ptrace-pkey: Rename variables to make it easier to follow code

Rename variable to indicate that they are invalid values which we will
use to test ptrace update of pkeys.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-21-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/kuap: Move UAMOR setup to key init function
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:42 +0000 (08:59 +0530)]
powerpc/book3s64/kuap: Move UAMOR setup to key init function

UAMOR values are not application-specific. The kernel initializes
its value based on different reserved keys. Remove the thread-specific
UAMOR value and don't switch the UAMOR on context switch.

Move UAMOR initialization to key initialization code and remove
thread_struct.uamor because it is not used anymore.

Before commit: 4a4a5e5d2aad ("powerpc/pkeys: key allocation/deallocation must not change pkey registers")
we used to update uamor based on key allocation and free.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-20-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/keys/kuap: Reset AMR/IAMR values on kexec
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:41 +0000 (08:59 +0530)]
powerpc/book3s64/keys/kuap: Reset AMR/IAMR values on kexec

As we kexec across kernels that use AMR/IAMR for different purposes
we need to ensure that new kernels get kexec'd with a reset value
of AMR/IAMR. For ex: the new kernel can use key 0 for kernel mapping and the old
AMR value prevents access to key 0.

This patch also removes reset if IAMR and AMOR in kexec_sequence. Reset of AMOR
is not needed and the IAMR reset is partial (it doesn't do the reset
on secondary cpus) and is redundant with this patch.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-19-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/keys: Print information during boot.
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:40 +0000 (08:59 +0530)]
powerpc/book3s64/keys: Print information during boot.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-18-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/pkeys: Use MMU_FTR_PKEY instead of pkey_disabled static key
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:39 +0000 (08:59 +0530)]
powerpc/book3s64/pkeys: Use MMU_FTR_PKEY instead of pkey_disabled static key

Instead of pkey_disabled static key use mmu feature MMU_FTR_PKEY.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-17-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/pkeys: Use pkey_execute_disable_supported
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:38 +0000 (08:59 +0530)]
powerpc/book3s64/pkeys: Use pkey_execute_disable_supported

Use pkey_execute_disable_supported to check for execute key support instead
of pkey_disabled.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-16-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/kuep: Add MMU_FTR_KUEP
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:37 +0000 (08:59 +0530)]
powerpc/book3s64/kuep: Add MMU_FTR_KUEP

This will be used to enable/disable Kernel Userspace Execution
Prevention (KUEP).

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-15-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/pkeys: Add MMU_FTR_PKEY
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:36 +0000 (08:59 +0530)]
powerpc/book3s64/pkeys: Add MMU_FTR_PKEY

Parse storage keys related device tree entry in early_init_devtree
and enable MMU feature MMU_FTR_PKEY if pkeys are supported.

MMU feature is used instead of CPU feature because this enables us
to group MMU_FTR_KUAP and MMU_FTR_PKEY in asm feature fixup code.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-14-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/pkeys: Mark all the pkeys above max pkey as reserved
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:35 +0000 (08:59 +0530)]
powerpc/book3s64/pkeys: Mark all the pkeys above max pkey as reserved

The hypervisor can return less than max allowed pkey (for ex: 31) instead
of 32. We should mark all the pkeys above max allowed as reserved so
that we avoid the allocation of the wrong pkey(for ex: key 31 in the above
case) by userspace.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-13-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/pkeys: Make initial_allocation_mask static
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:34 +0000 (08:59 +0530)]
powerpc/book3s64/pkeys: Make initial_allocation_mask static

initial_allocation_mask is not used outside this file.
Also mark reserved_allocation_mask and initial_allocation_mask __ro_after_init;

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-12-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/pkeys: Convert pkey_total to num_pkey
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:33 +0000 (08:59 +0530)]
powerpc/book3s64/pkeys: Convert pkey_total to num_pkey

num_pkey now represents max number of keys supported such that we return
to userspace 0 - num_pkey - 1.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-11-aneesh.kumar@linux.ibm.com
3 years agopowerpc/book3s64/pkeys: Simplify pkey disable branch
Aneesh Kumar K.V [Thu, 9 Jul 2020 03:29:32 +0000 (08:59 +0530)]
powerpc/book3s64/pkeys: Simplify pkey disable branch

Make the default value FALSE (pkey enabled) and set to TRUE when we
find the total number of keys supported to be zero.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200709032946.881753-10-aneesh.kumar@linux.ibm.com