powerpc/603: Use SPRN_SDR1 to store the pgdir phys address

On the 603, SDR1 is not used.

In order to free SPRN_SPRG2, use SPRN_SDR1 to store the pgdir
phys addr.

But only some bits of SDR1 can be used (0xffff01ff).
As the pgdir is 4k aligned, rotate it by 4 bits to the left.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/7370574b49d8476878ce5480726197993cb76108.1606285014.git.christophe.leroy@csgroup.eu

powerpc/32s: Don't use SPRN_SPRG_PGDIR in hash_page

SPRN_SPRG_PGDIR is there mainly to speedup SW TLB miss handlers
for powerpc 603.

We need to free SPRN_SPRG2 to reduce the mess with CONFIG_VMAP_STACK.

In hash_page(), reading PGDIR from thread_struct will be in the noise
performance wise.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4adca19b7120cdf619956768ed09e74fc6a558f3.1606285014.git.christophe.leroy@csgroup.eu

powerpc/32s: Fix an FTR_SECTION_ELSE

An FTR_SECTION_ELSE is in the middle of
BEGIN_MMU_FTR_SECTION/ALT_MMU_FTR_SECTION_END_IFSET

Change it to MMU_FTR_SECTION_ELSE

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/61790f1a91692950a6bb5bb53d6d514d9bcdad74.1606285014.git.christophe.leroy@csgroup.eu

powerpc/32s: Don't hash_preload() kernel text

We now always map kernel text with BATs. Neither need to preload
hash with kernel text addresses nor ensure they are never evicted.

This is more or less a revert of commit ee4f2ea48674 ("[POWERPC] Fix
32-bit mm operations when not using BATs")

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0a0bab7fadd89aa829e33420fbc10d60c59040a7.1606285014.git.christophe.leroy@csgroup.eu

powerpc/32s: Always map kernel text and rodata with BATs

Since commit 2b279c0348af ("powerpc/32s: Allow mapping with BATs with
DEBUG_PAGEALLOC"), there is no real situation where mapping without
BATs is required.

In order to simplify memory handling, always map kernel text
and rodata with BATs even when "nobats" kernel parameter is set.

Also fix the 603 TLB miss exceptions that don't require anymore
kernel page table if DEBUG_PAGEALLOC.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/da51f7ec632825a4ce43290a904aad61648408c0.1606285013.git.christophe.leroy@csgroup.eu

ocxl: Add new kernel traces

Add specific kernel traces which provide information on mmu notifier and on
pages range.

Signed-off-by: Christophe Lombard <clombard@linux.vnet.ibm.com>
Acked-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201125155013.39955-6-clombard@linux.vnet.ibm.com

ocxl: Add mmu notifier

Add invalidate_range mmu notifier, when required (ATSD access of MMIO
registers is available), to initiate TLB invalidation commands.
For the time being, the ATSD0 set of registers is used by default.

The pasid and bdf values have to be configured in the Process Element
Entry.
The PEE must be set up to match the BDF/PASID of the AFU.

Signed-off-by: Christophe Lombard <clombard@linux.vnet.ibm.com>
Acked-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201125155013.39955-5-clombard@linux.vnet.ibm.com

ocxl: Update the Process Element Entry

To complete the MMIO based mechanism, the fields: PASID, bus, device and
function of the Process Element Entry have to be filled. (See
OpenCAPI Power Platform Architecture document)

                   Hypervisor Process Element Entry
Word
    0 1 .... 7  8  ...... 12  13 ..15  16.... 19  20 ........... 31
0                  OSL Configuration State (0:31)
1                  OSL Configuration State (32:63)
2               PASID                      |    Reserved
3       Bus   |   Device    |Function |        Reserved
4                             Reserved
5                             Reserved
6                               ....

Signed-off-by: Christophe Lombard <clombard@linux.vnet.ibm.com>
Acked-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201125155013.39955-4-clombard@linux.vnet.ibm.com

ocxl: Initiate a TLB invalidate command

When a TLB Invalidate is required for the Logical Partition, the following
sequence has to be performed:

1. Load MMIO ATSD AVA register with the necessary value, if required.
2. Write the MMIO ATSD launch register to initiate the TLB Invalidate
command.
3. Poll the MMIO ATSD status register to determine when the TLB Invalidate
   has been completed.

Signed-off-by: Christophe Lombard <clombard@linux.vnet.ibm.com>
Acked-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201125155013.39955-3-clombard@linux.vnet.ibm.com

ocxl: Assign a register set to a Logical Partition

Platform specific function to assign a register set to a Logical Partition.
The "ibm,mmio-atsd" property, provided by the firmware, contains the 16
base ATSD physical addresses (ATSD0 through ATSD15) of the set of MMIO
registers (XTS MMIO ATSDx LPARID/AVA/launch/status register).

For the time being, the ATSD0 set of registers is used by default.

Signed-off-by: Christophe Lombard <clombard@linux.vnet.ibm.com>
Acked-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201125155013.39955-2-clombard@linux.vnet.ibm.com

powerpc/perf: MMCR0 control for PMU registers under PMCC=00

PowerISA v3.1 introduces new control bit (PMCCEXT) for restricting
access to group B PMU registers in problem state when
MMCR0 PMCC=0b00. In problem state and when MMCR0 PMCC=0b00,
setting the Monitor Mode Control Register bit 54 (MMCR0 PMCCEXT),
will restrict read permission on Group B Performance Monitor
Registers (SIER, SIAR, SDAR and MMCR1). When this bit is set to zero,
group B registers will be readable. In other platforms (like power9),
the older behaviour is retained where group B PMU SPRs are readable.

Patch adds support for MMCR0 PMCCEXT bit in power10 by enabling
this bit during boot and during the PMU event enable/disable callback
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/1606409684-1589-8-git-send-email-atrajeev@linux.vnet.ibm.com

powerpc/perf: Fix to update cache events with l2l3 events in power10

Export l2l3 events (PM_L2_ST_MISS and PM_L2_ST) and LLC-prefetches
(PM_L3_PF_MISS_L3) via sysfs, and also add these to list of
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/1606409684-1589-7-git-send-email-atrajeev@linux.vnet.ibm.com

powerpc/perf: Fix to update generic event codes for power10

Fix the event code for events: branch-instructions (to PM_BR_FIN),
branch-misses (to PM_MPRED_BR_FIN) and cache-misses (to
PM_LD_DEMAND_MISS_L1_FIN) for power10 PMU. Update the
list of generic events with this modified event code.

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/1606409684-1589-6-git-send-email-atrajeev@linux.vnet.ibm.com

powerpc/perf: Add generic and cache event list for power10 DD1

There are event code updates for some of the generic events
and cache events for power10. Inorder to maintain the current
event codes work with DD1 also, create a new array of generic_events,
cache_events and pmu_attr_groups with suffix _dd1, example,
power10_events_attr_dd1. So that further updates to event codes
can be made in the original list, ie, power10_events_attr. Update the
power10 pmu init code to pick the dd1 list while registering
the power PMU, based on the pvr (Processor Version Register) value.

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/1606409684-1589-5-git-send-email-atrajeev@linux.vnet.ibm.com

powerpc/perf: Fix the PMU group constraints for threshold events in power10

The PMU group constraints mask for threshold events covers
all thresholding bits which includes threshold control value
(start/stop), select value as well as thresh_cmp value (MMCRA[9:18].
In power9, thresh_cmp bits were part of the event code. But in case
of power10, thresh_cmp bits are not part of event code due to
inclusion of MMCR3 bits. Hence thresh_cmp is not valid for
group constraints for power10.

Fix the PMU group constraints checking for threshold events in
power10 by using constraint mask and value for only threshold control
and select bits.

Fixes: a64e697cef23 ("powerpc/perf: power10 Performance Monitoring support")
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/1606409684-1589-4-git-send-email-atrajeev@linux.vnet.ibm.com

powerpc/perf: Update the PMU group constraints for l2l3 events in power10

In Power9, L2/L3 bus events are always available as a
"bank" of 4 events. To obtain the counts for any of the
l2/l3 bus events in a given bank, the user will have to
program PMC4 with corresponding l2/l3 bus event for that
bank.

Commit 59029136d750 ("powerpc/perf: Add constraints for power9 l2/l3 bus events")
enforced this rule in Power9. But this is not valid for
Power10, since in Power10 Monitor Mode Control Register2
(MMCR2) has bits to configure l2/l3 event bits. Hence remove
this PMC4 constraint check from power10.

Since the l2/l3 bits in MMCR2 are not per-pmc, patch handles
group constrints checks for l2/l3 bits in MMCR2.

Fixes: a64e697cef23 ("powerpc/perf: power10 Performance Monitoring support")
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/1606409684-1589-3-git-send-email-atrajeev@linux.vnet.ibm.com

powerpc/perf: Fix to update radix_scope_qual in power10

power10 uses bit 9 of the raw event code as RADIX_SCOPE_QUAL.
This bit is used for enabling the radix process events.
Patch fixes the PMU counter support functions to program bit
18 of MMCR1 ( Monitor Mode Control Register1 ) with the
RADIX_SCOPE_QUAL bit value. Since this field is not per-pmc,
add this to PMU group constraints to make sure events in a
group will have same bit value for this field. Use bit 21 as
constraint bit field for radix_scope_qual. Patch also updates
the power10 raw event encoding layout information, format field
and constraints bit layout to include the radix_scope_qual bit.

Fixes: a64e697cef23 ("powerpc/perf: power10 Performance Monitoring support")
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/1606409684-1589-2-git-send-email-atrajeev@linux.vnet.ibm.com

powerpc/book3s64/pkeys: Optimize KUAP and KUEP feature disabled case

If FTR_BOOK3S_KUAP is disabled, kernel will continue to run with the same AMR
value with which it was entered. Hence there is a high chance that
we can return without restoring the AMR value. This also helps the case
when applications are not using the pkey feature. In this case, different
applications will have the same AMR values and hence we can avoid restoring
AMR in this case too.

Also avoid isync() if not really needed.

Do the same for IAMR.

null-syscall benchmark results:

With smap/smep disabled:
Without patch:
957.95 ns    2778.17 cycles
With patch:
858.38 ns    2489.30 cycles

With smap/smep enabled:
Without patch:
1017.26 ns    2950.36 cycles
With patch:
1021.51 ns    2962.44 cycles

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/20201127044424.40686-23-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kup: Check max key supported before enabling kup

Don't enable KUEP/KUAP if we support less than or equal to 3 keys.

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/20201202043854.76406-1-aneesh.kumar@linux.ibm.com

powerpc/book3s64/hash/kuep: Enable KUEP on hash

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-21-aneesh.kumar@linux.ibm.com

powerpc/book3s64/hash/kuap: Enable kuap on hash

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-20-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuep: Use Key 3 to implement KUEP with hash translation.

Radix use IAMR Key 0 and hash translation use IAMR key 3.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-19-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuap: Use Key 3 to implement KUAP with hash translation.

Radix use AMR Key 0 and hash translation use AMR key 3.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-18-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuap: Improve error reporting with KUAP

With hash translation use DSISR_KEYFAULT to identify a wrong access.
With Radix we look at the AMR value and type of fault.

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/20201127044424.40686-17-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuap: Restrict access to userspace based on userspace AMR

If an application has configured address protection such that read/write is
denied using pkey even the kernel should receive a FAULT on accessing the same.

This patch use user AMR value stored in pt_regs.amr to achieve the same.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-16-aneesh.kumar@linux.ibm.com

powerpc/book3s64/pkeys: Don't update SPRN_AMR when in kernel mode.

Now that kernel correctly store/restore userspace AMR/IAMR values, avoid
manipulating AMR and IAMR from the kernel on behalf of userspace.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-15-aneesh.kumar@linux.ibm.com

powerpc/ptrace-view: Use pt_regs values instead of thread_struct based one.

We will remove thread.amr/iamr/uamor in a later 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/20201127044424.40686-14-aneesh.kumar@linux.ibm.com

powerpc/book3s64/pkeys: Reset userspace AMR correctly on exec

On fork, we inherit from the parent and on exec, we should switch to default_amr values.

Also, avoid changing the AMR register value within the kernel. The kernel now runs with
different AMR values.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-13-aneesh.kumar@linux.ibm.com

powerpc/book3s64/pkeys: Inherit correctly on fork.

Child thread.kuap value is inherited from the parent in copy_thread_tls. We still
need to make sure when the child returns from a fork in the kernel we start with the kernel
default AMR value.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-12-aneesh.kumar@linux.ibm.com

powerpc/book3s64/pkeys: Store/restore userspace AMR/IAMR correctly on entry and exit from kernel

This prepare kernel to operate with a different value than userspace AMR/IAMR.
For this, AMR/IAMR need to be saved and restored on entry and return from the
kernel.

With KUAP we modify kernel AMR when accessing user address from the kernel
via copy_to/from_user interfaces. We don't need to modify IAMR value in
similar fashion.

If MMU_FTR_PKEY is enabled we need to save AMR/IAMR in pt_regs on entering
kernel from userspace. If not we can assume that AMR/IAMR is not modified
from userspace.

We need to save AMR if we have MMU_FTR_BOOK3S_KUAP feature enabled and we are
interrupted within kernel. This is required so that if we get interrupted
within copy_to/from_user we continue with the right AMR value.

If we hae MMU_FTR_BOOK3S_KUEP enabled we need to restore IAMR on
return to userspace beause kernel will be running with a different
IAMR value.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-11-aneesh.kumar@linux.ibm.com

powerpc/exec: Set thread.regs early during exec

In later patches during exec, we would like to access default regs.amr to
control access to the user mapping. Having thread.regs set early makes the
code changes simpler.

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/20201127044424.40686-10-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuap: Use Key 3 for kernel mapping with hash translation

This patch updates kernel hash page table entries to use storage key 3
for its mapping. This implies all kernel access will now use key 3 to
control READ/WRITE. The patch also prevents the allocation of key 3 from
userspace and UAMOR value is updated such that userspace cannot modify key 3.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201127044424.40686-9-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuap: Rename MMU_FTR_RADIX_KUAP and MMU_FTR_KUEP

This is in preparation to adding support for kuap with hash translation.
In preparation for that rename/move kuap related functions to
non radix names. Also move the feature bit closer to MMU_FTR_KUEP.

MMU_FTR_KUEP is renamed to MMU_FTR_BOOK3S_KUEP to indicate the feature
is only relevant to BOOK3S_64

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/20201127044424.40686-8-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuep: Move KUEP related function outside radix

The next set of patches adds support for kuep with hash translation.
In preparation for that rename/move kuap related functions to
non radix names.

Also set MMU_FTR_KUEP and add the missing isync().

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/20201127044424.40686-7-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuap: Move KUAP related function outside radix

The next set of patches adds support for kuap with hash translation.
In preparation for that rename/move kuap related functions to
non radix names.

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/20201127044424.40686-6-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuap/kuep: Move uamor setup to pkey init

This patch consolidates UAMOR update across pkey, kuap and kuep features.
The boot cpu initialize UAMOR via pkey init and both radix/hash do the
secondary cpu UAMOR init in early_init_mmu_secondary.

We don't check for mmu_feature in radix secondary init because UAMOR
is a supported SPRN with all CPUs supporting radix translation.
The old code was not updating UAMOR if we had smap disabled and smep enabled.
This change handles that case.

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/20201127044424.40686-5-aneesh.kumar@linux.ibm.com

powerpc/book3s64/kuap/kuep: Add PPC_PKEY config on book3s64

The config CONFIG_PPC_PKEY is used to select the base support that is
required for PPC_MEM_KEYS, KUAP, and KUEP. Adding this dependency
reduces the code complexity(in terms of #ifdefs) and enables us to
move some of the initialization code to pkeys.c

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/20201127044424.40686-4-aneesh.kumar@linux.ibm.com

KVM: PPC: BOOK3S: PR: Ignore UAMOR SPR

With power7 and above we expect the cpu to support keys. The
number of keys are firmware controlled based on device tree.
PR KVM do not expose key details via device tree. Hence when running with PR KVM
we do run with MMU_FTR_KEY support disabled. But we can still
get updates on UAMOR. Hence ignore access to them and for mfstpr return
0 indicating no AMR/IAMR update is no allowed.

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/20201127044424.40686-3-aneesh.kumar@linux.ibm.com

powerpc: Add new macro to handle NESTED_IFCLR

This will be used by the following patches

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/20201127044424.40686-2-aneesh.kumar@linux.ibm.com

powerpc/64s: Tidy machine check SLB logging

Since ISA v3.0, SLB no longer uses the slb_cache, and stab_rr is no
longer correlated with SLB allocation. Move those to pre-3.0.

While here, improve some alignments and reduce whitespace.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201128070728.825934-9-npiggin@gmail.com

powerpc/64s: Remove "Host" from MCE logging

"Host" caused machine check is printed when the kernel sees a MCE
hit in this kernel or userspace, and "Guest" if it hit one of its
guests. This is confusing when a guest kernel handles a hypervisor-
delivered MCE, it also prints "Host".

Just remove "Host". "Guest" is adequate to make the distinction.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201128070728.825934-8-npiggin@gmail.com

powerpc/64s/pseries: Add ERAT specific machine check handler

Don't treat ERAT MCEs as SLB, don't save the SLB and use a specific
ERAT flush to recover it.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201128070728.825934-7-npiggin@gmail.com

powerpc/64s/powernv: Ratelimit harmless HMI error printing

Harmless HMI errors can be triggered by guests in some cases, and don't
contain much useful information anyway. Ratelimit these to avoid
flooding the console/logs.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Use dedicated ratelimit state, not printk_ratelimit()]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201128070728.825934-6-npiggin@gmail.com

KVM: PPC: Book3S HV: Ratelimit machine check messages coming from guests

A number of machine check exceptions are triggerable by the guest.
Ratelimit these to avoid a guest flooding the host console and logs.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Use dedicated ratelimit state, not printk_ratelimit()]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201128070728.825934-5-npiggin@gmail.com

KVM: PPC: Book3S HV: Don't attempt to recover machine checks for FWNMI enabled guests

Guests that can deal with machine checks would actually prefer the
hypervisor not to try recover for them. For example if SLB multi-hits
are recovered by the hypervisor by clearing the SLB then the guest
will not be able to log the contents and debug its programming error.

If guests don't register for FWNMI, they may not be so capable and so
the hypervisor will continue to recover for those.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201128070728.825934-4-npiggin@gmail.com

powerpc/64s/powernv: Allow KVM to handle guest machine check details

KVM has strategies to perform machine check recovery. If a MCE hits
in a guest, have the low level handler just decode and save the MCE
but not try to recover anything, so KVM can deal with it.

The host does not own SLBs and does not need to report the SLB state
in case of a multi-hit for example, or know about the virtual memory
map of the guest.

UE and memory poisoning of guest pages in the host is one thing that
is possibly not completely robust at the moment, but this too needs
to go via KVM (possibly via the guest and back out to host via hcall)
rather than being handled at a low level in the host handler.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201128070728.825934-3-npiggin@gmail.com

powerpc/ps3: make system bus's remove and shutdown callbacks return void

The driver core ignores the return value of struct device_driver::remove
because there is only little that can be done. For the shutdown callback
it's ps3_system_bus_shutdown() which ignores the return value.

To simplify the quest to make struct device_driver::remove return void,
let struct ps3_system_bus_driver::remove return void, too. All users
already unconditionally return 0, this commit makes it obvious that
returning an error code is a bad idea and ensures future users behave
accordingly.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201126165950.2554997-2-u.kleine-koenig@pengutronix.de

ALSA: ppc: drop if block with always false condition

The remove callback is only called for devices that were probed
successfully before. As the matching probe function cannot complete
without error if dev->match_id != PS3_MATCH_ID_SOUND, we don't have to
check this here.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201126165950.2554997-1-u.kleine-koenig@pengutronix.de

powerpc/paravirt: Use is_kvm_guest() in vcpu_is_preempted()

If its a shared LPAR but not a KVM guest, then see if the vCPU is
related to the calling vCPU. On PowerVM, only cores can be preempted.
So if one vCPU is a non-preempted state, we can decipher that all
other vCPUs sharing the same core are in non-preempted state.

Performance results:

  $ perf stat -r 5 -a perf bench sched pipe -l 10000000 (lesser time is better)

  powerpc/next
       35,107,951.20 msec cpu-clock                 #  255.898 CPUs utilized            ( +-  0.31% )
          23,655,348      context-switches          #    0.674 K/sec                    ( +-  3.72% )
              14,465      cpu-migrations            #    0.000 K/sec                    ( +-  5.37% )
              82,463      page-faults               #    0.002 K/sec                    ( +-  8.40% )
   1,127,182,328,206      cycles                    #    0.032 GHz                      ( +-  1.60% )  (66.67%)
      78,587,300,622      stalled-cycles-frontend   #    6.97% frontend cycles idle     ( +-  0.08% )  (50.01%)
     654,124,218,432      stalled-cycles-backend    #   58.03% backend cycles idle      ( +-  1.74% )  (50.01%)
     834,013,059,242      instructions              #    0.74  insn per cycle
                                                    #    0.78  stalled cycles per insn  ( +-  0.73% )  (66.67%)
     132,911,454,387      branches                  #    3.786 M/sec                    ( +-  0.59% )  (50.00%)
       2,890,882,143      branch-misses             #    2.18% of all branches          ( +-  0.46% )  (50.00%)

             137.195 +- 0.419 seconds time elapsed  ( +-  0.31% )

  powerpc/next + patchset
       29,981,702.64 msec cpu-clock                 #  255.881 CPUs utilized            ( +-  1.30% )
          40,162,456      context-switches          #    0.001 M/sec                    ( +-  0.01% )
               1,110      cpu-migrations            #    0.000 K/sec                    ( +-  5.20% )
              62,616      page-faults               #    0.002 K/sec                    ( +-  3.93% )
   1,430,030,626,037      cycles                    #    0.048 GHz                      ( +-  1.41% )  (66.67%)
      83,202,707,288      stalled-cycles-frontend   #    5.82% frontend cycles idle     ( +-  0.75% )  (50.01%)
     744,556,088,520      stalled-cycles-backend    #   52.07% backend cycles idle      ( +-  1.39% )  (50.01%)
     940,138,418,674      instructions              #    0.66  insn per cycle
                                                    #    0.79  stalled cycles per insn  ( +-  0.51% )  (66.67%)
     146,452,852,283      branches                  #    4.885 M/sec                    ( +-  0.80% )  (50.00%)
       3,237,743,996      branch-misses             #    2.21% of all branches          ( +-  1.18% )  (50.01%)

              117.17 +- 1.52 seconds time elapsed  ( +-  1.30% )

This is around 14.6% improvement in performance.

Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Waiman Long <longman@redhat.com>
[mpe: Fold in performance results from cover letter]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201202050456.164005-5-srikar@linux.vnet.ibm.com

powerpc: Reintroduce is_kvm_guest() as a fast-path check

Introduce a static branch that would be set during boot if the OS
happens to be a KVM guest. Subsequent checks to see if we are on KVM
will rely on this static branch. This static branch would be used in
vcpu_is_preempted() in a subsequent patch.

Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Waiman Long <longman@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201202050456.164005-4-srikar@linux.vnet.ibm.com

powerpc: Rename is_kvm_guest() to check_kvm_guest()

We want to reuse the is_kvm_guest() name in a subsequent patch but
with a new body. Hence rename is_kvm_guest() to check_kvm_guest(). No
additional changes.

Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Waiman Long <longman@redhat.com>
Signed-off-by: kernel test robot <lkp@intel.com> # int -> bool fix
[mpe: Fold in fix from lkp to use true/false not 0/1]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201202050456.164005-3-srikar@linux.vnet.ibm.com

powerpc: Refactor is_kvm_guest() declaration to new header

Only code/declaration movement, in anticipation of doing a KVM-aware
vcpu_is_preempted(). No additional changes.

Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Waiman Long <longman@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201202050456.164005-2-srikar@linux.vnet.ibm.com

powerpc: show registers when unwinding interrupt frames

It's often useful to know the register state for interrupts in
the stack frame. In the below example (with this patch applied),
the important information is the state of the page fault.

A blatant case like this probably rather should have the page
fault regs passed down to the warning, but quite often there are
less obvious cases where an interrupt shows up that might give
some more clues.

The downside is longer and more complex bug output.

  Bug: Write fault blocked by AMR!
  WARNING: CPU: 0 PID: 72 at arch/powerpc/include/asm/book3s/64/kup-radix.h:164 __do_page_fault+0x880/0xa90
  Modules linked in:
  CPU: 0 PID: 72 Comm: systemd-gpt-aut Not tainted
  NIP:  c00000000006e2f0 LR: c00000000006e2ec CTR: 0000000000000000
  REGS: c00000000a4f3420 TRAP: 0700
  MSR:  8000000000021033 <SF,ME,IR,DR,RI,LE>  CR: 28002840  XER: 20040000
  CFAR: c000000000128be0 IRQMASK: 3
  GPR00: c00000000006e2ec c00000000a4f36c0 c0000000014f0700 0000000000000020
  GPR04: 0000000000000001 c000000001290f50 0000000000000001 c000000001290f80
  GPR08: c000000001612b08 0000000000000000 0000000000000000 00000000ffffe0f7
  GPR12: 0000000048002840 c0000000016e0000 c00c000000021c80 c000000000fd6f60
  GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000
  GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004
  GPR24: 0000000002000000 0000000000000300 0000000002000000 c00000000a5b0c00
  GPR28: 0000000000000000 000000000a000000 00007fffb2a90038 c00000000a4f3820
  NIP [c00000000006e2f0] __do_page_fault+0x880/0xa90
  LR [c00000000006e2ec] __do_page_fault+0x87c/0xa90
  Call Trace:
  [c00000000a4f36c0] [c00000000006e2ec] __do_page_fault+0x87c/0xa90 (unreliable)
  [c00000000a4f3780] [c000000000e1c034] do_page_fault+0x34/0x90
  [c00000000a4f37b0] [c000000000008908] data_access_common_virt+0x158/0x1b0
  --- interrupt: 300 at __copy_tofrom_user_base+0x9c/0x5a4
  NIP:  c00000000009b028 LR: c000000000802978 CTR: 0000000000000800
  REGS: c00000000a4f3820 TRAP: 0300
  MSR:  800000000280b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE>  CR: 24004840  XER: 00000000
  CFAR: c00000000009aff4 DAR: 00007fffb2a90038 DSISR: 0a000000 IRQMASK: 0
  GPR00: 0000000000000000 c00000000a4f3ac0 c0000000014f0700 00007fffb2a90028
  GPR04: c000000008720010 0000000000010000 0000000000000000 0000000000000000
  GPR08: 0000000000000000 0000000000000000 0000000000000000 0000000000000001
  GPR12: 0000000000004000 c0000000016e0000 c00c000000021c80 c000000000fd6f60
  GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000
  GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004
  GPR24: c00000000a4f3c80 c000000008720000 0000000000010000 0000000000000000
  GPR28: 0000000000010000 0000000008720000 0000000000010000 c000000001515b98
  NIP [c00000000009b028] __copy_tofrom_user_base+0x9c/0x5a4
  LR [c000000000802978] copyout+0x68/0xc0
  --- interrupt: 300
  [c00000000a4f3af0] [c0000000008074b8] copy_page_to_iter+0x188/0x540
  [c00000000a4f3b50] [c00000000035c678] generic_file_buffered_read+0x358/0xd80
  [c00000000a4f3c40] [c0000000004c1e90] blkdev_read_iter+0x50/0x80
  [c00000000a4f3c60] [c00000000045733c] new_sync_read+0x12c/0x1c0
  [c00000000a4f3d00] [c00000000045a1f0] vfs_read+0x1d0/0x240
  [c00000000a4f3d50] [c00000000045a7f4] ksys_read+0x84/0x140
  [c00000000a4f3da0] [c000000000033a60] system_call_exception+0x100/0x280
  [c00000000a4f3e10] [c00000000000c508] system_call_common+0xf8/0x2f8
  Instruction dump:
  eae10078 3be0000b 4bfff890 60420000 792917e1 4182ff18 3c82ffab 3884a5e0
  3c62ffab 3863a6e8 480ba891 60000000 <0fe00000> 3be0000b 4bfff860 e93c0938

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201107023305.2384874-1-npiggin@gmail.com

powerpc/perf: Invoke per-CPU variable access with disabled interrupts

The power_pmu_event_init() callback access per-cpu variable
(cpu_hw_events) to check for event constraints and Branch Stack
(BHRB). Current usage is to disable preemption when accessing the
per-cpu variable, but this does not prevent timer callback from
interrupting event_init. Fix this by using 'local_irq_save/restore'
to make sure the code path is invoked with disabled interrupts.

This change is tested in mambo simulator to ensure that, if a timer
interrupt comes in during the per-cpu access in event_init, it will be
soft masked and replayed later. For testing purpose, introduced a
udelay() in power_pmu_event_init() to make sure a timer interrupt arrives
while in per-cpu variable access code between local_irq_save/resore.
As expected the timer interrupt was replayed later during local_irq_restore
called from power_pmu_event_init. This was confirmed by adding
breakpoint in mambo and checking the backtrace when timer_interrupt
was hit.

Reported-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
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/1606814880-1720-1-git-send-email-atrajeev@linux.vnet.ibm.com

selftests/powerpc: Fix uninitialized variable warning

Patch fixes uninitialized variable warning in bad_accesses test
which causes the selftests build to fail in older distibutions

bad_accesses.c: In function ‘bad_access’:
bad_accesses.c:52:9: error: ‘x’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
   printf("Bad - no SEGV! (%c)\n", x);
         ^
cc1: all warnings being treated as errors

Signed-off-by: Harish <harish@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201201092403.238182-1-harish@linux.ibm.com

selftests/powerpc: update .gitignore

I did an in-place build of the self-tests and found that it left
the tree dirty.

Add missed test binaries to .gitignore

Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201201144427.1228745-1-dja@axtens.net

powerpc/feature-fixups: use a semicolon rather than a comma

In a bunch of our security flushes, we use a comma rather than
a semicolon to 'terminate' an assignment. Nothing breaks, but
checkpatch picks it up if you copy it into another flush.

Switch to semicolons for ending statements.

Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201201144344.1228421-1-dja@axtens.net

powerpc/pseries: Define PCI bus speed for Gen4 and Gen5

Update bus speed definition for PCI Gen4 and 5.

Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201130152949.26467-1-fbarrat@linux.ibm.com

powerpc: Allow relative pointers in bug table entries

This enables GENERIC_BUG_RELATIVE_POINTERS on Power so that 32-bit
offsets are stored in the bug entries rather than 64-bit pointers.
While this doesn't save space for 32-bit machines, use it anyway so
there is only one code path.

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201201005203.15210-1-jniethe5@gmail.com

powerpc/xmon: Fix build failure for 8xx

With CONFIG_PPC_8xx and CONFIG_XMON set, kernel build fails with

  arch/powerpc/xmon/xmon.c:1379:12: error: 'find_free_data_bpt' defined
  but not used [-Werror=unused-function]

Fix it by enclosing find_free_data_bpt() inside #ifndef CONFIG_PPC_8xx.

Fixes: 30df74d67d48 ("powerpc/watchpoint/xmon: Support 2nd DAWR")
Reported-by: kernel test robot <lkp@intel.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/20201130034406.288047-1-ravi.bangoria@linux.ibm.com

powerpc: Use common STABS_DEBUG and DWARF_DEBUG and ELF_DETAILS macro

Use the common STABS_DEBUG and DWARF_DEBUG and ELF_DETAILS macro rule for
the linker script in an effort.

Signed-off-by: Youling Tang <tangyouling@loongson.cn>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1606460857-2723-1-git-send-email-tangyouling@loongson.cn

powerpc/64: Fix an EMIT_BUG_ENTRY in head_64.S

Commit 63ce271b5e37 ("powerpc/prom: convert PROM_BUG() to standard
trap") added an EMIT_BUG_ENTRY for the trap after the branch to
start_kernel(). The EMIT_BUG_ENTRY was for the address "0b", however the
trap was not labeled with "0". Hence the address used for bug is in
relative_toc() where the previous "0" label is. Label the trap as "0" so
the correct address is used.

Fixes: 63ce271b5e37 ("powerpc/prom: convert PROM_BUG() to standard trap")
Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20201130004404.30953-1-jniethe5@gmail.com

powerpc/vdso: Cleanup vdso.h

Rename the guard define to _ASM_POWERPC_VDSO_H

And remove useless #ifdef __KERNEL__

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/9902590d410cd1c2afa48b83b277faf0711f07b2.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove VDSO32_LBASE and VDSO64_LBASE

VDSO32_LBASE and VDSO64_LBASE are 0. Remove them to simplify code.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/6c4d6570d886bbe1cc471e8ca01602e4b4d9beb5.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove DBG()

DBG() is not used anymore. Remove it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e11a9b50e709f197bb3aa2ed1d80d2dee8714afc.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove vdso_ready

There is no way to get out of vdso_init() prematuraly anymore.

Remove vdso_ready as it will always be 1.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0e1e18c6329b848aa3edeeba76509b4d76182e7d.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove vdso_setup()

vdso_fixup_features() cannot fail anymore and that's
the only function called by vdso_setup().

vdso_setup() has become trivial and can be removed.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/11522eec6140f510a8c89c63cbb739277d097fdc.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove lib32_elfinfo and lib64_elfinfo

lib32_elfinfo and lib64_elfinfo are not used anymore, remove them.

Also remove vdso32_kbase and vdso64_kbase while removing the
last use.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/01ac65abf22f0428f8f764525a7d84459c54d806.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove symbol section information in struct lib32/64_elfinfo

The members related to the symbol section in struct lib32_elfinfo and
struct lib64_elfinfo are not used anymore, removed them.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b779e5b7cc0354e2f87fd407fe5b02f4a8a73825.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove unused text member in struct lib32/64_elfinfo

The text member in struct lib32_elfinfo and struct lib64_elfinfo
is not used, remove it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f53dcc9bb1946a7854d15b34d03d3d2e2003848c.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove vdso_patches[] and associated functions

vdso_patches[] is now empty, remove it and remove
all functions that depends on it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/27d75debd6e4ddeaffe1d66ffed1e7526684a004.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove runtime generated sigtramp offsets

Signal trampoline offsets are now generated at buildtime.

Runtime generated offsets are not used anymore, remove them.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/7c192d35a437151837cf4c48aeccb42380d6daac.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove __kernel_datapage_offset

__kernel_datapage_offset is not used anymore, remove it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ddb5c746bec4e1a026d7c85243213a1876ef844f.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove vdso32_pages and vdso64_pages

vdso32_pages and vdso64_pages are not used anymore.

Remove them.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/bce021f616cbaf39dfb5766cf7ef114adcb918d9.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Merge __kernel_sync_dicache_p5() into __kernel_sync_dicache()

__kernel_sync_dicache_p5() is an alternative to
__kernel_sync_dicache() when cpu has CPU_FTR_COHERENT_ICACHE

Remove this alternative function and merge
__kernel_sync_dicache_p5() into __kernel_sync_dicache() using
standard CPU feature fixup.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4c7dcc6544882761b2b0249d7a8ec2c3a8088cb5.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Use builtin symbols to locate fixup section

Add builtin symbols to locate fixup section and use them
instead of locating sections through elf headers at runtime.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/2954526981859ca1ccfcfc7a7c4263920e9ddfcb.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Retrieve sigtramp offsets at buildtime

This is copied from arm64.

Instead of using runtime generated signal trampoline offsets,
get offsets at buildtime.

If the said trampoline doesn't exist, build will fail. So no
need to check whether the trampoline exists or not in the VDSO.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f8bfd6812c3e3678b1cdb4d55a52f9eb022b40d3.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove unused \tmp param in __get_datapage()

The \tmp param is not used anymore, remove it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4b13f897dcccce8ae03c031a4598cf26b32e2f1c.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Simplify __get_datapage()

The VDSO datapage and the text pages are always located immediately
next to each other, so it can be hardcoded without an indirection
through __kernel_datapage_offset

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b08f5ef99d64cfc38f79b7ad5310d9b4d2479eeb.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Move vdso datapage up front

Move the vdso datapage in front of the VDSO area,
before vdso test.

This will allow to remove the __kernel_datapage_offset symbol
and simplify __get_datapage() in following patches.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b68c99b6e8ee0b1d99bfa4c7e34c359fc1bc1000.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Replace vdso_base by vdso

All other architectures but s390 use a void pointer named 'vdso'
to reference the VDSO mapping.

In a following patch, the VDSO data page will be put in front of
text, vdso_base will then not anymore point to VDSO text.

To avoid confusion between vdso_base and VDSO text, rename vdso_base
into vdso and make it a void __user *.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/8e6cefe474aa4ceba028abb729485cd46c140990.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Provide vdso_remap()

Provide vdso_remap() through _install_special_mapping() and
drop arch_remap().

This adds a test of the size and returns -EINVAL if the size
is not correct.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/373c66f768fa9cc8890f3b55462209a98c522326.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Move to _install_special_mapping() and remove arch_vma_name()

Copied from commit 2fea7f6c98f5 ("arm64: vdso: move to
_install_special_mapping and remove arch_vma_name").

Use the new _install_special_mapping() API added by
commit a62c34bd2a8a ("x86, mm: Improve _install_special_mapping
and fix x86 vdso naming") which obsolete install_special_mapping().

And remove arch_vma_name() as the name is handled by the new API.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: kernel test robot <lkp@intel.com>
[mpe: Squash fix to use PTR_ERR_OR_ZERO() from lkp]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e7e5dfe0f93234e31051f2a610b4b07f50b0082f.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Simplify arch_setup_additional_pages() exit

To simplify arch_setup_additional_pages() exit, rename
it __arch_setup_additional_pages() and create a caller
arch_setup_additional_pages() which does the locking.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/603c1d039d3f928ee95e547fcd2219fcf4c3b514.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Use VDSO size in arch_setup_additional_pages()

In arch_setup_additional_pages(), instead of using number of VDSO
pages and recalculate VDSO size, directly use the VDSO size.

As vdso_ready is set, vdso_pages can't be 0 so just remove the test.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4edfa548c3885a430b765335dc720105716e273f.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove unnecessary ifdefs in vdso_pagelist initialization

No need of all those #ifdefs around the pagelist initialisation,
use IS_ENABLED(), GCC will kick out unused static variables.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f9333432e329b1fcbbbf846cb1cd4a1c4127a60b.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Refactor 32 bits and 64 bits pages setup

The setup of VDSO pages is identical for 32 bits VDSO and
64 bits VDSO.

Refactor that setup.

And use &vdsoXX_start which is synonym of vdsoXX_kbase.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/269ffb54c37fc1d46128f77d7a39f88ef4a9957d.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove NULL termination element in vdso_pagelist

No need of a NULL last element in pagelists, install_special_mapping()
knows how long the list is.

Remove that element.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e58d95ab859e3cbc9bae3c9ce2959e17d2864f5d.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Remove get_page() in vdso_pagelist initialization

Partly copied from commit 16fb1a9bec61 ("arm64: vdso: clean up
vdso_pagelist initialization").

No need to get_page() the vdso text/data - these are part of the
kernel image.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/9d14540bd10832b6c9519d74fb5728fdc4974b36.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Rename syscall_map_32/64 to simplify vdso_setup_syscall_map()

Today vdso_data structure has:
- syscall_map_32[] and syscall_map_64[] on PPC64
- syscall_map_32[] on PPC32

On PPC32, syscall_map_32[] is populated using sys_call_table[].

On PPC64, syscall_map_64[] is populated using sys_call_table[]
and syscal_map_32[] is populated using compat_sys_call_table[].

To simplify vdso_setup_syscall_map(),
- On PPC32 rename syscall_map_32[] into syscall_map[],
- On PPC64 rename syscall_map_64[] into syscall_map[],
- On PPC64 rename syscall_map_32[] into compat_syscall_map[].

That way, syscall_map[] gets populated using sys_call_table[] and
compat_syscall_map[] gets population using compat_sys_call_table[].

Also define an empty compat_syscall_map[] on PPC32 to avoid ifdefs.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/472734be0d9991eee320a06824219a5b2663736b.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Add missing includes and clean vdso_setup_syscall_map()

Instead of including extern references locally in
vdso_setup_syscall_map(), add the missing headers.

sys_ni_syscall() being a function, cast its address to
an unsigned long instead of declaring it as a fake
unsigned long object.

At the same time, remove a comment which paraphrases the
function name.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b4afedce748ed2858299ceab5ae29b52109263ef.1601197618.git.christophe.leroy@csgroup.eu

powerpc/vdso: Stripped VDSO is not needed, don't build it

Since commit 24b659a13866 ("powerpc: Use unstripped VDSO image for
more accurate profiling data"), only the unstripped VDSO image
has been used.

Partially revert commit 8150caad0226 ("[POWERPC] powerpc vDSO: install
unstripped copies on disk") to avoid building the stripped version.

And the unstripped version in $(MODLIB)/vdso/ is not required
anymore as it is the one embedded in the kernel image.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5986ca25be44fe6e9790486304507f240077d8c4.1601197618.git.christophe.leroy@csgroup.eu

powerpc/signal32: Transform save_user_regs() and save_tm_user_regs() in 'unsafe' version

Change those two functions to be used within a user access block.

For that, change save_general_regs() to and unsafe_save_general_regs(),
then replace all user accesses by unsafe_ versions.

This series leads to a reduction from 2.55s to 1.73s of
the system CPU time with the following microbench app
on an mpc832x with KUAP (approx 32%)

Without KUAP, the difference is in the noise.

void sigusr1(int sig) { }

int main(int argc, char **argv)
{
int i = 100000;

signal(SIGUSR1, sigusr1);
for (;i--;)
    raise(SIGUSR1);
exit(0);
}

An additional 0.10s reduction is achieved by removing
CONFIG_PPC_FPU, as the mpc832x has no FPU.

A bit less spectacular on an 8xx as KUAP is less heavy, prior to
the series (with KUAP) it ran in 8.10 ms. Once applies the removal
of FPU regs handling, we get 7.05s. With the full series, we get 6.9s.
If artificially re-activating FPU regs handling with the full series,
we get 7.6s.

So for the 8xx, the removal of the FPU regs copy is what makes the
difference, but the rework of handle_signal also have a benefit.

Same as above, without KUAP the difference is in the noise.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Fixup typo in SPE handling]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c7b37b385ccf9666066452e58f018a86573f83e8.1597770847.git.christophe.leroy@csgroup.eu

powerpc/signal32: Isolate non-copy actions in save_user_regs() and save_tm_user_regs()

Reorder actions in save_user_regs() and save_tm_user_regs() to
regroup copies together in order to switch to user_access_begin()
logic in a later patch.

Move non-copy actions into new functions called
prepare_save_user_regs() and prepare_save_tm_user_regs().

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f6eac65781b4a57220477c8864bca2b57f29a5d5.1597770847.git.christophe.leroy@csgroup.eu

powerpc/signal: Create 'unsafe' versions of copy_[ck][fpr/vsx]_to_user()

For the non VSX version, that's trivial. Just use unsafe_copy_to_user()
instead of __copy_to_user().

For the VSX version, remove the intermediate step through a buffer and
use unsafe_put_user() directly. This generates a far smaller code which
is acceptable to inline, see below:

Standard VSX version:

0000000000000000 <.copy_fpr_to_user>:
   0: 7c 08 02 a6  mflr    r0
   4: fb e1 ff f8  std     r31,-8(r1)
   8: 39 00 00 20  li      r8,32
   c: 39 24 0b 80  addi    r9,r4,2944
  10: 7d 09 03 a6  mtctr   r8
  14: f8 01 00 10  std     r0,16(r1)
  18: f8 21 fe 71  stdu    r1,-400(r1)
  1c: 39 41 00 68  addi    r10,r1,104
  20: e9 09 00 00  ld      r8,0(r9)
  24: 39 4a 00 08  addi    r10,r10,8
  28: 39 29 00 10  addi    r9,r9,16
  2c: f9 0a 00 00  std     r8,0(r10)
  30: 42 00 ff f0  bdnz    20 <.copy_fpr_to_user+0x20>
  34: e9 24 0d 80  ld      r9,3456(r4)
  38: 3d 42 00 00  addis   r10,r2,0
3a: R_PPC64_TOC16_HA .toc
  3c: eb ea 00 00  ld      r31,0(r10)
3e: R_PPC64_TOC16_LO_DS .toc
  40: f9 21 01 70  std     r9,368(r1)
  44: e9 3f 00 00  ld      r9,0(r31)
  48: 81 29 00 20  lwz     r9,32(r9)
  4c: 2f 89 00 00  cmpwi   cr7,r9,0
  50: 40 9c 00 18  bge     cr7,68 <.copy_fpr_to_user+0x68>
  54: 4c 00 01 2c  isync
  58: 3d 20 40 00  lis     r9,16384
  5c: 79 29 07 c6  rldicr  r9,r9,32,31
  60: 7d 3d 03 a6  mtspr   29,r9
  64: 4c 00 01 2c  isync
  68: 38 a0 01 08  li      r5,264
  6c: 38 81 00 70  addi    r4,r1,112
  70: 48 00 00 01  bl      70 <.copy_fpr_to_user+0x70>
70: R_PPC64_REL24 .__copy_tofrom_user
  74: 60 00 00 00  nop
  78: e9 3f 00 00  ld      r9,0(r31)
  7c: 81 29 00 20  lwz     r9,32(r9)
  80: 2f 89 00 00  cmpwi   cr7,r9,0
  84: 40 9c 00 18  bge     cr7,9c <.copy_fpr_to_user+0x9c>
  88: 4c 00 01 2c  isync
  8c: 39 20 ff ff  li      r9,-1
  90: 79 29 00 44  rldicr  r9,r9,0,1
  94: 7d 3d 03 a6  mtspr   29,r9
  98: 4c 00 01 2c  isync
  9c: 38 21 01 90  addi    r1,r1,400
  a0: e8 01 00 10  ld      r0,16(r1)
  a4: eb e1 ff f8  ld      r31,-8(r1)
  a8: 7c 08 03 a6  mtlr    r0
  ac: 4e 80 00 20  blr

'unsafe' simulated VSX version (The ... are only nops) using
unsafe_copy_fpr_to_user() macro:

unsigned long copy_fpr_to_user(void __user *to,
       struct task_struct *task)
{
unsafe_copy_fpr_to_user(to, task, failed);
return 0;
failed:
return 1;
}

0000000000000000 <.copy_fpr_to_user>:
   0: 39 00 00 20  li      r8,32
   4: 39 44 0b 80  addi    r10,r4,2944
   8: 7d 09 03 a6  mtctr   r8
   c: 7c 69 1b 78  mr      r9,r3
...
  20: e9 0a 00 00  ld      r8,0(r10)
  24: f9 09 00 00  std     r8,0(r9)
  28: 39 4a 00 10  addi    r10,r10,16
  2c: 39 29 00 08  addi    r9,r9,8
  30: 42 00 ff f0  bdnz    20 <.copy_fpr_to_user+0x20>
  34: e9 24 0d 80  ld      r9,3456(r4)
  38: f9 23 01 00  std     r9,256(r3)
  3c: 38 60 00 00  li      r3,0
  40: 4e 80 00 20  blr
...
  50: 38 60 00 01  li      r3,1
  54: 4e 80 00 20  blr

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/29f6c4b8e7a5bbc61e6a8801b78bbf493f9f819e.1597770847.git.christophe.leroy@csgroup.eu

powerpc/signal32: Switch swap_context() to user_access_begin() logic

As this was the last user of put_sigset_t(), remove it as well.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c3ac4f2d134a3391bb51bdaa2d00e9a409aba9f8.1597770847.git.christophe.leroy@csgroup.eu

powerpc/signal32: Add and use unsafe_put_sigset_t()

put_sigset_t() calls copy_to_user() for copying two words.

This is terribly inefficient for copying two words.

By switching to unsafe_put_user(), we end up with something as
simple as:

 3cc:   81 3d 00 00     lwz     r9,0(r29)
 3d0:   91 26 00 b4     stw     r9,180(r6)
 3d4:   81 3d 00 04     lwz     r9,4(r29)
 3d8:   91 26 00 b8     stw     r9,184(r6)

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/06def97e87ac1c4ae8e3197e0982e1fab7b3c8ae.1597770847.git.christophe.leroy@csgroup.eu

signal: Add unsafe_put_compat_sigset()

Implement 'unsafe' version of put_compat_sigset()

For the bigendian, use unsafe_put_user() directly
to avoid intermediate copy through the stack.

For the littleendian, use a straight unsafe_copy_to_user().

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/537c7082ee309a0bb9c67a50c5d9dd929aedb82d.1597770847.git.christophe.leroy@csgroup.eu

powerpc/signal32: Remove ifdefery in middle of if/else

MSR_TM_ACTIVE() is always defined and returns always 0 when
CONFIG_PPC_TRANSACTIONAL_MEM is not selected, so the awful
ifdefery in the middle of an if/else can be removed.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f3c36d687e4228f58d5c207a4036aa9ddcc7420a.1597770847.git.christophe.leroy@csgroup.eu

powerpc/signal32: Switch handle_rt_signal32() to user_access_begin() logic

On the same way as handle_signal32(), replace all user
accesses with equivalent unsafe_ versions, and move the
trampoline code icache flush outside the user access block.

Functions that have no unsafe_ equivalent also remains outside
the access block.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/2974314226256f958e2984912b48883ef1754185.1597770847.git.christophe.leroy@csgroup.eu