powerpc: make stack walking KASAN-safe

Make our stack-walking code KASAN-safe by using __no_sanitize_address.
Generic code, arm64, s390 and x86 all make accesses unchecked for similar
sorts of reasons: when unwinding a stack, we might touch memory that KASAN
has marked as being out-of-bounds. In ppc64 KASAN development, I hit this
sometimes when checking for an exception frame - because we're checking
an arbitrary offset into the stack frame.

See commit 20955746320e ("s390/kasan: avoid false positives during stack
unwind"), commit bcaf669b4bdb ("arm64: disable kasan when accessing
frame->fp in unwind_frame"), commit 91e08ab0c851 ("x86/dumpstack:
Prevent KASAN false positive warnings") and commit 6e22c8366416
("tracing, kasan: Silence Kasan warning in check_stack of stack_tracer").

Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210614120907.1952321-1-dja@axtens.net

selftests/powerpc: EBB selftest for MMCR0 control for PMU SPRs in ISA v3.1

With the MMCR0 control bit (PMCCEXT) in ISA v3.1, read access to
group B registers is restricted when MMCR0 PMCC=0b00. In other
platforms (like power9), the older behaviour works where group B
PMU SPRs are readable.

Patch creates a selftest which verifies that the test takes a
SIGILL when attempting to read PMU registers via helper function
"dump_ebb_state" for ISA v3.1.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Tested-by: Nageswara R Sastry <rnsastry@linux.ibm.com <mailto:rnsastry@linux.ibm.com>>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1621950703-1532-3-git-send-email-atrajeev@linux.vnet.ibm.com

selftests/powerpc: Fix "no_handler" EBB selftest

The "no_handler_test" in ebb selftests attempts to read the PMU
registers twice via helper function "dump_ebb_state". First dump is
just before closing of event and the second invocation is done after
closing of the event. The original intention of second
dump_ebb_state was to dump the state of registers at the end of
the test when the counters are frozen. But this will be achieved
with the first call itself since sample period is set to low value
and PMU will be frozen by then. Hence patch removes the
dump which was done before closing of the event.

Reported-by: Shirisha Ganta <shirisha.ganta1@ibm.com>
Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Tested-by: Nageswara R Sastry <rnsastry@linux.ibm.com <mailto:rnsastry@linux.ibm.com>>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1621950703-1532-2-git-send-email-atrajeev@linux.vnet.ibm.com

powerpc: Move update_power8_hid0() into its only user

update_power8_hid0() is used only by powernv platform subcore.c

Move it there.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/37f41d74faa0c66f90b373e243e8b1ee37a1f6fa.1623219019.git.christophe.leroy@csgroup.eu

powerpc: Remove proc_trap()

proc_trap() has never been 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/827944ea12d470c2f862635f48b5ee6c1520351f.1623217909.git.christophe.leroy@csgroup.eu

powerpc/32: Remove __main()

Comment says that __main() is there to make GCC happy.

It's been there since the implementation of ppc arch in Linux 1.3.45.

ppc32 is the only architecture having that. Even ppc64 doesn't have it.

Seems like GCC is still happy without it.

Drop it for good.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d01028f8166b98584eec536b52f14c5e3f98ff6b.1623172922.git.christophe.leroy@csgroup.eu

powerpc/32s: Rename PTE_SIZE to PTE_T_SIZE

PTE_SIZE means PTE page table size in most placed, whereas
in hash_low.S in means size of one entry in the table.

Rename it PTE_T_SIZE, and define it directly in hash_low.S
instead of going through asm-offsets.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/83a008a9fd6cc3f2bbcb470f592555d260ed7a3d.1623063174.git.christophe.leroy@csgroup.eu

powerpc: Define swapper_pg_dir[] in C

Don't duplicate swapper_pg_dir[] in each platform's head.S

Define it in mm/pgtable.c

Define MAX_PTRS_PER_PGD because on book3s/64 PTRS_PER_PGD is
not a constant.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5e3f1b8a4695c33ccc80aa3870e016bef32b85e1.1623063174.git.christophe.leroy@csgroup.eu

powerpc: Define empty_zero_page[] in C

At the time being, empty_zero_page[] is defined in each
platform head.S.

Define it in mm/mem.c instead, and put it in BSS section instead
of the DATA section. Commit 5227cfa71f9e ("arm64: mm: place
empty_zero_page in bss") explains why it is interesting to have
it in BSS.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5838caffa269e0957c5a50cc85477876220298b0.1623063174.git.christophe.leroy@csgroup.eu

powerpc/selftests: Use gettid() instead of getppid() for null_syscall

gettid() is 10% lighter than getppid(), use it for null_syscall selftest.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0ad62673d3e063f848e7c99d719bb966efd433e8.1622809833.git.christophe.leroy@csgroup.eu

powerpc/nohash: Remove DEBUG_HARDER

DEBUG_HARDER is not user selectable.

Remove it together with related messages.

Also remove two pr_devel() messages that should
likely have been pr_hard().

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0f25109b0e12fdd1e6541dedbb2212cc53526a57.1622712515.git.christophe.leroy@csgroup.eu

powerpc/nohash: Remove DEBUG_CLAMP_LAST_CONTEXT

DEBUG_CLAMP_LAST_CONTEXT was there in the old days to reduce
number of contexts in order to ease debugging implementation
of context switching, but that's been quite stable during
years now.

As it is not user selectable, 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/da81837b452e8b9f1657b529b9c3050dc10b9770.1622712515.git.christophe.leroy@csgroup.eu

powerpc/nohash: Remove DEBUG_MAP_CONSISTENCY

mmu_context handling has been there for years, so we
would know if there was problems with maps.

DEBUG_MAP_CONSISTENCY is not user selectable, 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/6fe2b88956db53f8d6ee221525b2c5dc6aec82c6.1622712515.git.christophe.leroy@csgroup.eu

powerpc/nohash: Remove CONFIG_SMP #ifdefery in mmu_context.h

Everything can be done even when CONFIG_SMP is not selected.

Just use IS_ENABLED() where relevant and rely on GCC to
opt out unneeded code and variables when CONFIG_SMP is not set.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/cc13b87b0f750a538621876ecc24c22a07e7c8be.1622712515.git.christophe.leroy@csgroup.eu

powerpc/nohash: Convert set_context() to C

ppc8xx already has set_context() in C.

Other ones have it in assembly. The only thing it does is to
write the context id into SPRN_PID.

Do it in C.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/a5d0759064f3831c6b88af49ef5d3b05ba1c4dad.1622712515.git.christophe.leroy@csgroup.eu

powerpc/nohash: Refactor update of BDI2000 pointers in switch_mmu_context()

Instead of duplicating the update of BDI2000 pointers in
set_context(), do it directly from switch_mmu_context().

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4c54997edd3548fa54717915e7c6ebaf60f208c0.1622712515.git.christophe.leroy@csgroup.eu

powerpc/kuap: Force inlining of all first level KUAP helpers.

All KUAP helpers defined in asm/kup.h are single line functions
that should be inlined. But on book3s/32 build, we get many
instances of <prevent_write_to_user.constprop.0>.

Force inlining of those helpers.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/8479a862e165a57a855292d47e24c259a578f5a0.1622711627.git.christophe.leroy@csgroup.eu

powerpc/kuap: Remove to/from/size parameters of prevent_user_access()

prevent_user_access() doesn't use anymore to/from/size parameters.

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/b7113662fd2c26e4c33e9d705de324bd3860822e.1622708530.git.christophe.leroy@csgroup.eu

powerpc/kuap: Remove KUAP_CURRENT_XXX

book3s/32 was the only user of KUAP_CURRENT_XXX.

After rework of book3s/32 KUAP, it is 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/549214ecf6887d965645e664520d4886663c5ffb.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Activate KUAP and KUEP by default

Now that KUAP and KUEP have been significantly optimised and can be
disabled at boot time using 'nosmap' and 'nosmep' kernel parameters,
them can be active by default like in other powerpc platforms.

It is still possible to disable them completely in the configuration.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/86c7c74a3ba5312daea7e9658b096e2bcc6f4b64.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Rework Kernel Userspace Access Protection

On book3s/32, KUAP is provided by toggling Ks bit in segment registers.
One segment register addresses 256M of virtual memory.

At the time being, KUAP implements a complex logic to apply the
unlock/lock on the exact number of segments covering the user range
to access, with saving the boundaries of the range of segments in
a member of thread struct.

But most if not all user accesses are within a single segment.

Rework KUAP with a different approach:
- Open only one segment, the one corresponding to the starting
address of the range to be accessed.
- If a second segment is involved, it will generate a page fault. The
segment will then be open by the page fault handler.

The kuap member of thread struct will now contain:
- The start address of the current on going user access, that will be
used to know which segment to lock at the end of the user access.
- ~0 when no user access is open
- ~1 when additionnal segments are opened by a page fault.

Then, at lock time
- When only one segment is open, close it.
- When several segments are open, close all user segments.

Almost 100% of the time, only one segment will be involved.

In interrupts, inline the function that unlock/lock all segments,
because not inlining them implies a lot of register save/restore.

With the patch, writing value 128 in userspace in perf_copy_attr() is
done with 16 instructions:

    3890: 93 82 04 dc  stw     r28,1244(r2)
    3894: 7d 20 e5 26  mfsrin  r9,r28
    3898: 55 29 00 80  rlwinm  r9,r9,0,2,0
    389c: 7d 20 e1 e4  mtsrin  r9,r28
    38a0: 4c 00 01 2c  isync

    38a4: 39 20 00 80  li      r9,128
    38a8: 91 3c 00 00  stw     r9,0(r28)

    38ac: 81 42 04 dc  lwz     r10,1244(r2)
    38b0: 39 00 ff ff  li      r8,-1
    38b4: 91 02 04 dc  stw     r8,1244(r2)
    38b8: 2c 0a ff fe  cmpwi   r10,-2
    38bc: 41 82 00 88  beq     3944 <perf_copy_attr+0x36c>
    38c0: 7d 20 55 26  mfsrin  r9,r10
    38c4: 65 29 40 00  oris    r9,r9,16384
    38c8: 7d 20 51 e4  mtsrin  r9,r10
    38cc: 4c 00 01 2c  isync
...
    3944: 48 00 00 01  bl      3944 <perf_copy_attr+0x36c>
3944: R_PPC_REL24 kuap_lock_all_ool

Before the patch it was 118 instructions. In reality only 42 are
executed in most cases, but GCC is not able to see that a properly
aligned user access cannot involve more than one segment.

    5060: 39 1d 00 04  addi    r8,r29,4
    5064: 3d 20 b0 00  lis     r9,-20480
    5068: 7c 08 48 40  cmplw   r8,r9
    506c: 40 81 00 08  ble     5074 <perf_copy_attr+0x2cc>
    5070: 3d 00 b0 00  lis     r8,-20480
    5074: 39 28 ff ff  addi    r9,r8,-1
    5078: 57 aa 00 06  rlwinm  r10,r29,0,0,3
    507c: 55 29 27 3e  rlwinm  r9,r9,4,28,31
    5080: 39 29 00 01  addi    r9,r9,1
    5084: 7d 29 53 78  or      r9,r9,r10
    5088: 91 22 04 dc  stw     r9,1244(r2)
    508c: 7d 20 ed 26  mfsrin  r9,r29
    5090: 55 29 00 80  rlwinm  r9,r9,0,2,0
    5094: 7c 08 50 40  cmplw   r8,r10
    5098: 40 81 00 c0  ble     5158 <perf_copy_attr+0x3b0>
    509c: 7d 46 50 f8  not     r6,r10
    50a0: 7c c6 42 14  add     r6,r6,r8
    50a4: 54 c6 27 be  rlwinm  r6,r6,4,30,31
    50a8: 7d 20 51 e4  mtsrin  r9,r10
    50ac: 3c ea 10 00  addis   r7,r10,4096
    50b0: 39 29 01 11  addi    r9,r9,273
    50b4: 7f 88 38 40  cmplw   cr7,r8,r7
    50b8: 55 29 02 06  rlwinm  r9,r9,0,8,3
    50bc: 40 9d 00 9c  ble     cr7,5158 <perf_copy_attr+0x3b0>

    50c0: 2f 86 00 00  cmpwi   cr7,r6,0
    50c4: 41 9e 00 4c  beq     cr7,5110 <perf_copy_attr+0x368>
    50c8: 2f 86 00 01  cmpwi   cr7,r6,1
    50cc: 41 9e 00 2c  beq     cr7,50f8 <perf_copy_attr+0x350>
    50d0: 2f 86 00 02  cmpwi   cr7,r6,2
    50d4: 41 9e 00 14  beq     cr7,50e8 <perf_copy_attr+0x340>
    50d8: 7d 20 39 e4  mtsrin  r9,r7
    50dc: 39 29 01 11  addi    r9,r9,273
    50e0: 3c e7 10 00  addis   r7,r7,4096
    50e4: 55 29 02 06  rlwinm  r9,r9,0,8,3
    50e8: 7d 20 39 e4  mtsrin  r9,r7
    50ec: 39 29 01 11  addi    r9,r9,273
    50f0: 3c e7 10 00  addis   r7,r7,4096
    50f4: 55 29 02 06  rlwinm  r9,r9,0,8,3
    50f8: 7d 20 39 e4  mtsrin  r9,r7
    50fc: 3c e7 10 00  addis   r7,r7,4096
    5100: 39 29 01 11  addi    r9,r9,273
    5104: 7f 88 38 40  cmplw   cr7,r8,r7
    5108: 55 29 02 06  rlwinm  r9,r9,0,8,3
    510c: 40 9d 00 4c  ble     cr7,5158 <perf_copy_attr+0x3b0>
    5110: 7d 20 39 e4  mtsrin  r9,r7
    5114: 39 29 01 11  addi    r9,r9,273
    5118: 3c c7 10 00  addis   r6,r7,4096
    511c: 55 29 02 06  rlwinm  r9,r9,0,8,3
    5120: 7d 20 31 e4  mtsrin  r9,r6
    5124: 39 29 01 11  addi    r9,r9,273
    5128: 3c c6 10 00  addis   r6,r6,4096
    512c: 55 29 02 06  rlwinm  r9,r9,0,8,3
    5130: 7d 20 31 e4  mtsrin  r9,r6
    5134: 39 29 01 11  addi    r9,r9,273
    5138: 3c c7 30 00  addis   r6,r7,12288
    513c: 55 29 02 06  rlwinm  r9,r9,0,8,3
    5140: 7d 20 31 e4  mtsrin  r9,r6
    5144: 3c e7 40 00  addis   r7,r7,16384
    5148: 39 29 01 11  addi    r9,r9,273
    514c: 7f 88 38 40  cmplw   cr7,r8,r7
    5150: 55 29 02 06  rlwinm  r9,r9,0,8,3
    5154: 41 9d ff bc  bgt     cr7,5110 <perf_copy_attr+0x368>

    5158: 4c 00 01 2c  isync
    515c: 39 20 00 80  li      r9,128
    5160: 91 3d 00 00  stw     r9,0(r29)

    5164: 38 e0 00 00  li      r7,0
    5168: 90 e2 04 dc  stw     r7,1244(r2)
    516c: 7d 20 ed 26  mfsrin  r9,r29
    5170: 65 29 40 00  oris    r9,r9,16384
    5174: 40 81 00 c0  ble     5234 <perf_copy_attr+0x48c>
    5178: 7d 47 50 f8  not     r7,r10
    517c: 7c e7 42 14  add     r7,r7,r8
    5180: 54 e7 27 be  rlwinm  r7,r7,4,30,31
    5184: 7d 20 51 e4  mtsrin  r9,r10
    5188: 3d 4a 10 00  addis   r10,r10,4096
    518c: 39 29 01 11  addi    r9,r9,273
    5190: 7c 08 50 40  cmplw   r8,r10
    5194: 55 29 02 06  rlwinm  r9,r9,0,8,3
    5198: 40 81 00 9c  ble     5234 <perf_copy_attr+0x48c>

    519c: 2c 07 00 00  cmpwi   r7,0
    51a0: 41 82 00 4c  beq     51ec <perf_copy_attr+0x444>
    51a4: 2c 07 00 01  cmpwi   r7,1
    51a8: 41 82 00 2c  beq     51d4 <perf_copy_attr+0x42c>
    51ac: 2c 07 00 02  cmpwi   r7,2
    51b0: 41 82 00 14  beq     51c4 <perf_copy_attr+0x41c>
    51b4: 7d 20 51 e4  mtsrin  r9,r10
    51b8: 39 29 01 11  addi    r9,r9,273
    51bc: 3d 4a 10 00  addis   r10,r10,4096
    51c0: 55 29 02 06  rlwinm  r9,r9,0,8,3
    51c4: 7d 20 51 e4  mtsrin  r9,r10
    51c8: 39 29 01 11  addi    r9,r9,273
    51cc: 3d 4a 10 00  addis   r10,r10,4096
    51d0: 55 29 02 06  rlwinm  r9,r9,0,8,3
    51d4: 7d 20 51 e4  mtsrin  r9,r10
    51d8: 3d 4a 10 00  addis   r10,r10,4096
    51dc: 39 29 01 11  addi    r9,r9,273
    51e0: 7c 08 50 40  cmplw   r8,r10
    51e4: 55 29 02 06  rlwinm  r9,r9,0,8,3
    51e8: 40 81 00 4c  ble     5234 <perf_copy_attr+0x48c>
    51ec: 7d 20 51 e4  mtsrin  r9,r10
    51f0: 39 29 01 11  addi    r9,r9,273
    51f4: 3c ea 10 00  addis   r7,r10,4096
    51f8: 55 29 02 06  rlwinm  r9,r9,0,8,3
    51fc: 7d 20 39 e4  mtsrin  r9,r7
    5200: 39 29 01 11  addi    r9,r9,273
    5204: 3c e7 10 00  addis   r7,r7,4096
    5208: 55 29 02 06  rlwinm  r9,r9,0,8,3
    520c: 7d 20 39 e4  mtsrin  r9,r7
    5210: 39 29 01 11  addi    r9,r9,273
    5214: 3c ea 30 00  addis   r7,r10,12288
    5218: 55 29 02 06  rlwinm  r9,r9,0,8,3
    521c: 7d 20 39 e4  mtsrin  r9,r7
    5220: 3d 4a 40 00  addis   r10,r10,16384
    5224: 39 29 01 11  addi    r9,r9,273
    5228: 7c 08 50 40  cmplw   r8,r10
    522c: 55 29 02 06  rlwinm  r9,r9,0,8,3
    5230: 41 81 ff bc  bgt     51ec <perf_copy_attr+0x444>

    5234: 4c 00 01 2c  isync

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Export the ool handlers to fix build errors]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d9121f96a7c4302946839a0771f5d1daeeb6968c.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Allow disabling KUAP at boot time

PPC64 uses MMU features to enable/disable KUAP at boot time.
But feature fixups are applied way too early on PPC32.

Now that all KUAP related actions are in C following the
conversion of KUAP initial setup and context switch in C,
static branches can be used to enable/disable KUAP.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Export disable_kuap_key to fix build errors]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/cd79e8008455fba5395d099f9bb1305c039b931c.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Allow disabling KUEP at boot time

PPC64 uses MMU features to enable/disable KUEP at boot time.
But feature fixups are applied way too early on PPC32.

Now that all KUEP related actions are in C following the
conversion of KUEP initial setup and context switch in C,
static branches can be used to enable/disable KUEP.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/7745a2c3a08ec46302920a3f48d1cb9b5469dbbb.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Initialise KUAP and KUEP in C

In order to selectively activate KUAP and KUEP in a following patch,
perform KUAP and KUEP initialisation in C.

Unlike PPC64, PPC32 doesn't have an early_setup_secondary(),
so do it in start_secondary().

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/87be72023448dd4e476744ed279b8c04b8d08a1c.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Simplify calculation of segment register content

segment register has VSID on bits 8-31.
Bits 4-7 are reserved, there is no requirement to set them to 0.

VSIDs are calculated from VSID of SR0 by adding 0x111.

Even with highest possible VSID which would be 0xFFFFF0,
adding 16 times 0x111 results in 0x1001100.

So, the reserved bits are never overflowed, no need to clear
the reserved bits after each calculation.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ddc1cfd2ec8f3b2395c6a4d7f2b0c1aa1b1e64fb.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Convert switch_mmu_context() to C

switch_mmu_context() does things that can easily be done in C.

For updating user segments, we have update_user_segments().

As mentionned in commit b5efec00b671 ("powerpc/32s: Move KUEP
locking/unlocking in C"), update_user_segments() has the loop
unrolled which is a significant performance gain.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/05c0875ad8220c03452c3a334946e207c6ca04d6.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: move CTX_TO_VSID() into mmu-hash.h

In order to reuse it in switch_mmu_context(), this
patch moves CTX_TO_VSID() macro into asm/book3s/32/mmu-hash.h

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/26b36ef2939234a04b37baf6ffe50cba81f5d1b7.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Refactor update of user segment registers

KUEP implements the update of user segment registers.

Move it into mmu-hash.h in order to use it from other places.

And inline kuep_lock() and kuep_unlock(). Inlining kuep_lock() is
important for system_call_exception(), otherwise system_call_exception()
has to save into stack the system call parameters that are used just
after, and doing that takes more instructions than kuep_lock() itself.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/24591ca480d14a62ef910e38a5273d551262c4a2.1622708530.git.christophe.leroy@csgroup.eu

powerpc/32s: Move setup_{kuep/kuap}() into {kuep/kuap}.c

Avoids the #ifdef in mmu.c

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0b7a13d414837e58264edc336b89c2fe9f35f9bc.1622708530.git.christophe.leroy@csgroup.eu

powerpc/8xx: Allow disabling KUAP at boot time

PPC64 uses MMU features to enable/disable KUAP at boot time.
But feature fixups are applied way too early on PPC32.

But since commit c16728835eec ("powerpc/32: Manage KUAP in C"),
all KUAP is in C so it is now possible to use static branches.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/3dca510ce555335261a47c4799167da698f569c0.1622782111.git.christophe.leroy@csgroup.eu

powerpc/44x: Implement Kernel Userspace Exec Protection (KUEP)

Powerpc 44x has two bits for exec protection in TLBs: one
for user (UX) and one for superviser (SX).

Clear SX on user pages in TLB miss handlers to provide KUEP.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/169310e08152aa1d96c979770291d165ec6896ae.1622616032.git.christophe.leroy@csgroup.eu

powerpc: Remove CONFIG_PPC_MMU_NOHASH_32

Since commit Fixes: 555904d07eef ("powerpc/8xx: MM_SLICE is not needed anymore"),
CONFIG_PPC_MMU_NOHASH_32 has not been used.

Remove it.

Reported-by: Tom Rix <trix@redhat.com>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/bf1e074f6fb213a1c4cc4964370bdce4b648d647.1622706812.git.christophe.leroy@csgroup.eu

powerpc/optprobes: use PPC_RAW_ macros

Use PPC_RAW_ macros to simplify the code.

And use PPC_LO/PPC_HI instead of IMM_L/IMM_H which are for
internal use inside ppc-opcode.h

Those macros are self explanatory, comments can go 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/5a167b8ba4d33a5c09cd504f0c862e25ffe85459.1621516826.git.christophe.leroy@csgroup.eu

powerpc/optprobes: Compact code source a bit.

Now that lines can be up to 100 chars long, minimise the
amount of split lines to increase readability.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/8ebbd977ea8cf8d706d82458f2a21acd44562a99.1621516826.git.christophe.leroy@csgroup.eu

powerpc/optprobes: Minimise casts

nip is already an unsigned long, no cast needed.

op_callback_addr and emulate_step_addr are kprobe_opcode_t *.
There value is obtained with ppc_kallsyms_lookup_name() which
returns 'unsigned long', and there values are used create_branch()
which expects 'unsigned long'. So change them to 'unsigned long'
to avoid casting them back and forth.

can_optimize() used p->addr several times as 'unsigned long'.
Use a local 'unsigned long' variable and avoid casting multiple times.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e03192a6d4123242a275e71ce2ba0bb4d90700c1.1621516826.git.christophe.leroy@csgroup.eu

powerpc/inst: Refactor PPC32 and PPC64 versions

ppc_inst() ppc_inst_prefixed() ppc_inst_swab() can easily be made common
to both PPC32 and PPC64.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d54c63dcac6d190e1cc0d2fe3259d6e621928cdf.1621516826.git.christophe.leroy@csgroup.eu

powerpc: Don't use 'struct ppc_inst' to reference instruction location

'struct ppc_inst' is an internal representation of an instruction, but
in-memory instructions are and will remain a table of 'u32' forever.

Replace all 'struct ppc_inst *' used for locating an instruction in
memory by 'u32 *'. This removes a lot of undue casts to 'struct
ppc_inst *'.

It also helps locating ab-use of 'struct ppc_inst' dereference.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Fix ppc_inst_next(), use u32 instead of unsigned int]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/7062722b087228e42cbd896e39bfdf526d6a340a.1621516826.git.christophe.leroy@csgroup.eu

powerpc/lib/code-patching: Don't use struct 'ppc_inst' for runnable code in tests.

'struct ppc_inst' is meant to represent an instruction internally, it
is not meant to dereference code in memory.

For testing code patching, use patch_instruction() to properly
write into memory the code to be tested.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d8425fb42a4adebc35b7509f121817eeb02fac31.1621516826.git.christophe.leroy@csgroup.eu

powerpc/lib/code-patching: Make instr_is_branch_to_addr() static

instr_is_branch_to_addr() is only used in code-patching.c

Make it static.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5f6b9c8c83170ed310953eac2f5b14539bfc964a.1621516826.git.christophe.leroy@csgroup.eu

powerpc: Do not dereference code as 'struct ppc_inst' (uprobe, code-patching, feature-fixups)

'struct ppc_inst' is an internal structure to represent an instruction,
it is not directly the representation of that instruction in text code.
It is not meant to map and dereference code.

Dereferencing code directly through 'struct ppc_inst' has two main issues:
- On powerpc, structs are expected to be 8 bytes aligned while code is
spread every 4 byte.
- Should a non prefixed instruction lie at the end of the page and the
following page not be mapped, it would generate a page fault.

In-memory code must be accessed with ppc_inst_read().

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c9a1201dd0a66b4a0f91f0fb46d9385cbf030feb.1621516826.git.christophe.leroy@csgroup.eu

powerpc/inst: Avoid pointer dereferencing in ppc_inst_equal()

Avoid casting/dereferencing ppc_inst() as u64* , check each member
of the struct when relevant.

And remove the 0xff initialisation of the suffix for non
prefixed instruction. An instruction with 0xff as a suffix
might be invalid, but still is a prefixed instruction and
has to be considered as this.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d8b155e930b7a9708ca110e8ff0ace6713a7af75.1621516826.git.christophe.leroy@csgroup.eu

powerpc/inst: Improve readability of get_user_instr() and friends

Remove unneeded line splits.

And remove unneeded local variable initialisation.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/fb097fda78cc6852905ef00f8f7bf371b6cc66f7.1621516826.git.christophe.leroy@csgroup.eu

powerpc/inst: Reduce casts in get_user_instr()

Declare __gui_ptr as 'u32 *' instead of casting it at each use to
'unsigned int *' (which is an equivalent type).

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Use u32 * instead of unsigned int *]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/2c2123998e05535d08ba03a96ea1eea921d06a86.1621516826.git.christophe.leroy@csgroup.eu

powerpc/inst: Fix sparse detection on get_user_instr()

get_user_instr() lacks sparse detection for the __user tag.

This is because __gui_ptr is assigned with a cast.

Fix that by adding a __chk_user_ptr()

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0320e5b41a794fd456ab8c5993bbfadcf9e1d8b4.1621516826.git.christophe.leroy@csgroup.eu

powerpc: Replace PPC_INST_NOP by PPC_RAW_NOP()

On the road to removing all PPC_INST_xx defines in
asm/ppc-opcodes.h, change PPC_INST_NOP to PPC_RAW_NOP().

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ad46c195ca1b8572629ef07ba6bfe247585239a6.1621506159.git.christophe.leroy@csgroup.eu

powerpc/traps: Start using PPC_RAW_xx() macros

Start using PPC_RAW_xx() macros where relevant.

PPC_INST_SYNC is used to both represent the 'sync' instruction and
the family of synchronisation instructions. Keep it for the later,
maybe we'll change the name in the future to avoid confusion.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0945c155d6cb113431185fc1296ac127359fe29b.1621506159.git.christophe.leroy@csgroup.eu

powerpc/lib/feature-fixups: Use PPC_RAW_xxx() macros

Use PPC_RAW_xxx() macros instead of open coding assembly
opcodes.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Fix bad converison in do_stf_exit_barrier_fixups()]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e79cd8e111ca13bf8c61a384bac365aa7e207647.1621506159.git.christophe.leroy@csgroup.eu

powerpc/ebpf32: Use _Rx macros instead of __REG_Rx ones

To increase readability, use _Rx macros instead of __REG_Rx.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/eb7ec6297b5d16f141c5866da3975b418e47431b.1621506159.git.christophe.leroy@csgroup.eu

powerpc/ebpf64: Use PPC_RAW_MFLR()

Use PPC_RAW_MFLR() instead of open coding with PPC_INST_MFLR.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c1887623e91e8b4da36e669e4c74de86320a5092.1621506159.git.christophe.leroy@csgroup.eu

powerpc/ftrace: Use PPC_RAW_MFLR() and PPC_RAW_NOP()

Use PPC_RAW_MFLR() instead of open coding with PPC_INST_MFLR.

Same for PPC_INST_NOP.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/98fd4d717810b7c4032a1edf62dd6fe638e64329.1621506159.git.christophe.leroy@csgroup.eu

powerpc/security: Use PPC_RAW_BLR() and PPC_RAW_NOP()

On the road to remove all use of PPC_INST_xxx, replace
PPC_INST_BLR by PPC_RAW_BLR(). Same for PPC_INST_NOP.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c04f88d0e53d2122fbbe92226892a01ebc668b6a.1621506159.git.christophe.leroy@csgroup.eu

powerpc/modules: Use PPC_RAW_xx() macros

To improve readability, use PPC_RAW_xx() macros instead of
open coding. Those macros are self-explanatory so the comments
can go 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/99d9ee8849d3992beeadb310a665aae01c3abfb1.1621506159.git.christophe.leroy@csgroup.eu

powerpc/signal: Use PPC_RAW_xx() macros

To improve readability, use PPC_RAW_xx() macros instead of
open coding. Those macros are self-explanatory so the comments
can go 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/4ca2bfdca2f47a293d05f61eb3c4e487ee170f1f.1621506159.git.christophe.leroy@csgroup.eu

powerpc/lib/code-patching: Use PPC_RAW_() macros

Instead of open coding with PPC_INST_ defines, use
PPC_RAW_() macros. It improves readability.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/8c92f1d9e825ee47c6f88fe43ad42d2a8cc2ab4a.1621506159.git.christophe.leroy@csgroup.eu

powerpc/opcodes: Add shorter macros for registers for use with PPC_RAW_xx()

Today we have __REG_Rx macros . They are mainly meant for
internal use by macros __PPC_RA() and friends macros which
allows uses like __PPC_RA(R12).

When used with PPC_RAW_xx() macros, it gives a result which is
not very readable.

Add shorter macros _Rx in order to improve readability when
used with PPC_RAW_xx() macros.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ec34d92b7c2f810622261acfeeed4b0a0f4d01bd.1621506159.git.christophe.leroy@csgroup.eu

powerpc: Rework PPC_RAW_xxx() macros for prefixed instructions

At the time being, we have PPC_RAW_PLXVP() and PPC_RAW_PSTXVP() which
provide a 64 bits value, and then it gets split by open coding to
format it into a 'struct ppc_inst' instruction.

Instead, define a PPC_RAW_xxx_P() and a PPC_RAW_xxx_S() to be used
as is.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5d146b31b943e7ad674894421db4feef54804b9b.1621506159.git.christophe.leroy@csgroup.eu

powerpc: Don't handle ALTIVEC/SPE in ASM in _switch(). Do it in C.

_switch() saves and restores ALTIVEC and SPE status.
For altivec this is redundant with what __switch_to() does with
save_sprs() and restore_sprs() and giveup_all() before
calling _switch().

Add support for SPI in save_sprs() and restore_sprs() and
remove things from _switch().

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/8ab21fd93d6e0047aa71e6509e5e312f14b2991b.1620998075.git.christophe.leroy@csgroup.eu

powerpc: Force inlining of csum_add()

Commit 328e7e487a46 ("powerpc: force inlining of csum_partial() to
avoid multiple csum_partial() with GCC10") inlined csum_partial().

Now that csum_partial() is inlined, GCC outlines csum_add() when
called by csum_partial().

c064fb28 <csum_add>:
c064fb28: 7c 63 20 14  addc    r3,r3,r4
c064fb2c: 7c 63 01 94  addze   r3,r3
c064fb30: 4e 80 00 20  blr

c0665fb8 <csum_add>:
c0665fb8: 7c 63 20 14  addc    r3,r3,r4
c0665fbc: 7c 63 01 94  addze   r3,r3
c0665fc0: 4e 80 00 20  blr

c066719c: 7c 9a c0 2e  lwzx    r4,r26,r24
c06671a0: 38 60 00 00  li      r3,0
c06671a4: 7f 1a c2 14  add     r24,r26,r24
c06671a8: 4b ff ee 11  bl      c0665fb8 <csum_add>
c06671ac: 80 98 00 04  lwz     r4,4(r24)
c06671b0: 4b ff ee 09  bl      c0665fb8 <csum_add>
c06671b4: 80 98 00 08  lwz     r4,8(r24)
c06671b8: 4b ff ee 01  bl      c0665fb8 <csum_add>
c06671bc: a0 98 00 0c  lhz     r4,12(r24)
c06671c0: 4b ff ed f9  bl      c0665fb8 <csum_add>
c06671c4: 7c 63 18 f8  not     r3,r3
c06671c8: 81 3f 00 68  lwz     r9,104(r31)
c06671cc: 81 5f 00 a0  lwz     r10,160(r31)
c06671d0: 7d 29 18 14  addc    r9,r9,r3
c06671d4: 7d 29 01 94  addze   r9,r9
c06671d8: 91 3f 00 68  stw     r9,104(r31)
c06671dc: 7d 1a 50 50  subf    r8,r26,r10
c06671e0: 83 01 00 10  lwz     r24,16(r1)
c06671e4: 83 41 00 18  lwz     r26,24(r1)

The sum with 0 is useless, should have been skipped.
And there is even one completely unused instance of csum_add().

In file included from ./include/net/checksum.h:22,
                 from ./include/linux/skbuff.h:28,
                 from ./include/linux/icmp.h:16,
                 from net/ipv6/ip6_tunnel.c:23:
./arch/powerpc/include/asm/checksum.h: In function '__ip6_tnl_rcv':
./arch/powerpc/include/asm/checksum.h:94:22: warning: inlining failed in call to 'csum_add': call is unlikely and code size would grow [-Winline]
   94 | static inline __wsum csum_add(__wsum csum, __wsum addend)
      |                      ^~~~~~~~
./arch/powerpc/include/asm/checksum.h:172:31: note: called from here
  172 |                         sum = csum_add(sum, (__force __wsum)*(const u32 *)buff);
      |                               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
./arch/powerpc/include/asm/checksum.h:94:22: warning: inlining failed in call to 'csum_add': call is unlikely and code size would grow [-Winline]
   94 | static inline __wsum csum_add(__wsum csum, __wsum addend)
      |                      ^~~~~~~~
./arch/powerpc/include/asm/checksum.h:177:31: note: called from here
  177 |                         sum = csum_add(sum, (__force __wsum)
      |                               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                                             *(const u32 *)(buff + 4));
      |                                             ~~~~~~~~~~~~~~~~~~~~~~~~~
./arch/powerpc/include/asm/checksum.h:94:22: warning: inlining failed in call to 'csum_add': call is unlikely and code size would grow [-Winline]
   94 | static inline __wsum csum_add(__wsum csum, __wsum addend)
      |                      ^~~~~~~~
./arch/powerpc/include/asm/checksum.h:183:31: note: called from here
  183 |                         sum = csum_add(sum, (__force __wsum)
      |                               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                                             *(const u32 *)(buff + 8));
      |                                             ~~~~~~~~~~~~~~~~~~~~~~~~~
./arch/powerpc/include/asm/checksum.h:94:22: warning: inlining failed in call to 'csum_add': call is unlikely and code size would grow [-Winline]
   94 | static inline __wsum csum_add(__wsum csum, __wsum addend)
      |                      ^~~~~~~~
./arch/powerpc/include/asm/checksum.h:186:31: note: called from here
  186 |                         sum = csum_add(sum, (__force __wsum)
      |                               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                                             *(const u16 *)(buff + 12));
      |                                             ~~~~~~~~~~~~~~~~~~~~~~~~~~

Force inlining of csum_add().

     94c: 80 df 00 a0  lwz     r6,160(r31)
     950: 7d 28 50 2e  lwzx    r9,r8,r10
     954: 7d 48 52 14  add     r10,r8,r10
     958: 80 aa 00 04  lwz     r5,4(r10)
     95c: 80 ff 00 68  lwz     r7,104(r31)
     960: 7d 29 28 14  addc    r9,r9,r5
     964: 7d 29 01 94  addze   r9,r9
     968: 7d 08 30 50  subf    r8,r8,r6
     96c: 80 aa 00 08  lwz     r5,8(r10)
     970: a1 4a 00 0c  lhz     r10,12(r10)
     974: 7d 29 28 14  addc    r9,r9,r5
     978: 7d 29 01 94  addze   r9,r9
     97c: 7d 29 50 14  addc    r9,r9,r10
     980: 7d 29 01 94  addze   r9,r9
     984: 7d 29 48 f8  not     r9,r9
     988: 7c e7 48 14  addc    r7,r7,r9
     98c: 7c e7 01 94  addze   r7,r7
     990: 90 ff 00 68  stw     r7,104(r31)

In the non-inlined version, the first sum with 0 was performed.
Here it is skipped.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f7f4d4e364de6e473da874468b903da6e5d97adc.1620713272.git.christophe.leroy@csgroup.eu

Merge branch 'fixes' into next

Merge our fixes branch which has a number of important fixes, notably
the fix for initrd corruption, as well as the fixes for scv vs ptrace.

powerpc/tau: Remove superfluous parameter in alloc_workqueue() call

This avoids an (optional) compiler warning:

arch/powerpc/kernel/tau_6xx.c: In function 'TAU_init':
arch/powerpc/kernel/tau_6xx.c:204:30: error: too many arguments for format [-Werror=format-extra-args]
  tau_workq = alloc_workqueue("tau", WQ_UNBOUND, 1, 0);

Fixes: b1c6a0a10bfa ("powerpc/tau: Convert from timer to workqueue")
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Finn Thain <fthain@linux-m68k.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/a1456e8bbd33ef702e3ff6f14b1bf3919241c62b.1623398307.git.fthain@linux-m68k.org

powerpc: Fix initrd corruption with relative jump labels

Commit b0b3b2c78ec0 ("powerpc: Switch to relative jump labels") switched
us to using relative jump labels. That involves changing the code,
target and key members in struct jump_entry to be relative to the
address of the jump_entry, rather than absolute addresses.

We have two static inlines that create a struct jump_entry,
arch_static_branch() and arch_static_branch_jump(), as well as an asm
macro ARCH_STATIC_BRANCH, which is used by the pseries-only hypervisor
tracing code.

Unfortunately we missed updating the key to be a relative reference in
ARCH_STATIC_BRANCH.

That causes a pseries kernel to have a handful of jump_entry structs
with bad key values. Instead of being a relative reference they instead
hold the full address of the key.

However the code doesn't expect that, it still adds the key value to the
address of the jump_entry (see jump_entry_key()) expecting to get a
pointer to a key somewhere in kernel data.

The table of jump_entry structs sits in rodata, which comes after the
kernel text. In a typical build this will be somewhere around 15MB. The
address of the key will be somewhere in data, typically around 20MB.
Adding the two values together gets us a pointer somewhere around 45MB.

We then call static_key_set_entries() with that bad pointer and modify
some members of the struct static_key we think we are pointing at.

A pseries kernel is typically ~30MB in size, so writing to ~45MB won't
corrupt the kernel itself. However if we're booting with an initrd,
depending on the size and exact location of the initrd, we can corrupt
the initrd. Depending on how exactly we corrupt the initrd it can either
cause the system to not boot, or just corrupt one of the files in the
initrd.

The fix is simply to make the key value relative to the jump_entry
struct in the ARCH_STATIC_BRANCH macro.

Fixes: b0b3b2c78ec0 ("powerpc: Switch to relative jump labels")
Reported-by: Anastasia Kovaleva <a.kovaleva@yadro.com>
Reported-by: Roman Bolshakov <r.bolshakov@yadro.com>
Reported-by: Greg Kurz <groug@kaod.org>
Reported-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Tested-by: Daniel Axtens <dja@axtens.net>
Tested-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210614131440.312360-1-mpe@ellerman.id.au

powerpc/perf: Simplify Makefile

arch/powerpc/Kbuild decend into arch/powerpc/perf/ only when
CONFIG_PERF_EVENTS is selected, so there is not need to take
CONFIG_PERF_EVENTS into account in arch/powerpc/perf/Makefile.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Michal Suchánek <msuchanek@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d37f61afca55b5b33787b643890e061ae1c18f5f.1620396045.git.christophe.leroy@csgroup.eu

powerpc/prom_init: Move custom isspace() to its own namespace

If by some reason any of the headers will include ctype.h
we will have a name collision. Avoid this by moving isspace()
to the dedicate namespace.

First appearance of the code is in the commit cf68787b68a2
("powerpc/prom_init: Evaluate mem kernel parameter for early allocation").

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
[mpe: Reformat prom_isxdigit() now that we allow longer lines]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210510144925.58195-1-andriy.shevchenko@linux.intel.com

selftests/powerpc: Remove the repeated declaration

Function 'event_ebb_init' and 'event_leader_ebb_init' are declared
twice in the header file, so remove the repeated declaration.

Signed-off-by: Shaokun Zhang <zhangshaokun@hisilicon.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1622529385-5938-1-git-send-email-zhangshaokun@hisilicon.com

powerpc/spider-pci: Remove set but not used variable 'val'

Fixes gcc '-Wunused-but-set-variable' warning:
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format

arch/powerpc/platforms/cell/spider-pci.c: In function 'spiderpci_io_flush':
arch/powerpc/platforms/cell/spider-pci.c:28:6: warning:
variable ‘val’ set but not used [-Wunused-but-set-variable]

It never used since introduction.

Signed-off-by: Baokun Li <libaokun1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210601085319.140461-1-libaokun1@huawei.com

powerpc/spufs: Remove set but not used variable 'dummy'

Fixes gcc '-Wunused-but-set-variable' warning:
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format
# WARNING: Fixes tag on line 3 doesn't match correct format

arch/powerpc/platforms/cell/spufs/switch.c: In function 'check_ppu_mb_stat':
arch/powerpc/platforms/cell/spufs/switch.c:1660:6: warning:
variable ‘dummy’ set but not used [-Wunused-but-set-variable]

arch/powerpc/platforms/cell/spufs/switch.c: In function 'check_ppuint_mb_stat':
arch/powerpc/platforms/cell/spufs/switch.c:1675:6: warning:
variable ‘dummy’ set but not used [-Wunused-but-set-variable]

It never used since introduction.

Signed-off-by: Baokun Li <libaokun1@huawei.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210601085127.139598-1-libaokun1@huawei.com

powerpc/52xx: Add fallthrough in mpc52xx_wdt_ioctl()

With gcc 10.3, there is this compiler error:

  compiler.h:56:26: error: this statement may fall through
  mpc52xx_gpt.c:586:2: note: here
    586 |  case WDIOC_GETTIMEOUT:
        |  ^~~~

So add the fallthrough pseudo keyword.

Signed-off-by: Tom Rix <trix@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210601190200.2637776-1-trix@redhat.com

powerpc/signal64: Copy siginfo before changing regs->nip

In commit 96d7a4e06fab ("powerpc/signal64: Rewrite handle_rt_signal64()
to minimise uaccess switches") the 64-bit signal code was rearranged to
use user_write_access_begin/end().

As part of that change the call to copy_siginfo_to_user() was moved
later in the function, so that it could be done after the
user_write_access_end().

In particular it was moved after we modify regs->nip to point to the
signal trampoline. That means if copy_siginfo_to_user() fails we exit
handle_rt_signal64() with an error but with regs->nip modified, whereas
previously we would not modify regs->nip until the copy succeeded.

Returning an error from signal delivery but with regs->nip updated
leaves the process in a sort of half-delivered state. We do immediately
force a SEGV in signal_setup_done(), called from do_signal(), so the
process should never run in the half-delivered state.

However that SEGV is not delivered until we've gone around to
do_notify_resume() again, so it's possible some tracing could observe
the half-delivered state.

There are other cases where we fail signal delivery with regs partly
updated, eg. the write to newsp and SA_SIGINFO, but the latter at least
is very unlikely to fail as it reads back from the frame we just wrote
to.

Looking at other arches they seem to be more careful about leaving regs
unchanged until the copy operations have succeeded, and in general that
seems like good hygenie.

So although the current behaviour is not cleary buggy, it's also not
clearly correct. So move the call to copy_siginfo_to_user() up prior to
the modification of regs->nip, which is closer to the old behaviour, and
easier to reason about.

Fixes: 96d7a4e06fab ("powerpc/signal64: Rewrite handle_rt_signal64() to minimise uaccess switches")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210608134605.2783677-1-mpe@ellerman.id.au

powerpc/ps3: Add dma_mask to ps3_dma_region

Commit f959dcd6ddfd29235030e8026471ac1b022ad2b0 (dma-direct: Fix
potential NULL pointer dereference) added a null check on the
dma_mask pointer of the kernel's device structure.

Add a dma_mask variable to the ps3_dma_region structure and set
the device structure's dma_mask pointer to point to this new variable.

Fixes runtime errors like these:
# WARNING: Fixes tag on line 10 doesn't match correct format
# WARNING: Fixes tag on line 10 doesn't match correct format

  ps3_system_bus_match:349: dev=8.0(sb_01), drv=8.0(ps3flash): match
  WARNING: CPU: 0 PID: 1 at kernel/dma/mapping.c:151 .dma_map_page_attrs+0x34/0x1e0
  ps3flash sb_01: ps3stor_setup:193: map DMA region failed

Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/562d0c9ea0100a30c3b186bcc7adb34b0bbd2cd7.1622746428.git.geoff@infradead.org

powerpc/ps3: Warn on PS3 device errors

To aid debugging PS3 boot problems change the log level
of several PS3 device errors from pr_debug to pr_warn.

Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/eb5c1c10da0bbdeb27c8b069187b4f58e429e837.1622746428.git.geoff@infradead.org

powerpc/ps3: Add CONFIG_PS3_VERBOSE_RESULT option

To aid debugging, add a new PS3 kernel config option
PS3_VERBOSE_RESULT that, when enabled, will print more
verbose messages for the result of LV1 hypercalls.

Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0ce4b6969a08094a747bd382dbfd30b72ebc192d.1622746428.git.geoff@infradead.org

powerpc/ps3: Re-align DTB in image

Change the PS3 linker script to align the DTB at 8 bytes,
the same alignment as that of the of the 'generic' powerpc
linker script.

Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/245897ed65e402686a4b114ba618e935cb5c6506.1622822173.git.geoff@infradead.org

powerpc/ps3: Add firmware version to sysfs

Add a new sysfs entry /sys/firmware/ps3/fw-version that exports
the PS3's firmware version.

The firmware version is available through an LV1 hypercall, and we've
been printing it to the boot log, but haven't provided an easy way for
user utilities to get it.

Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/41509b2da647cd34b1331cc4756c8774b1e284eb.1622822173.git.geoff@infradead.org

powerpc/barrier: Avoid collision with clang's __lwsync macro

A change in clang 13 results in the __lwsync macro being defined as
__builtin_ppc_lwsync, which emits 'lwsync' or 'msync' depending on what
the target supports. This breaks the build because of -Werror in
arch/powerpc, along with thousands of warnings:

 In file included from arch/powerpc/kernel/pmc.c:12:
 In file included from include/linux/bug.h:5:
 In file included from arch/powerpc/include/asm/bug.h:109:
 In file included from include/asm-generic/bug.h:20:
 In file included from include/linux/kernel.h:12:
 In file included from include/linux/bitops.h:32:
 In file included from arch/powerpc/include/asm/bitops.h:62:
 arch/powerpc/include/asm/barrier.h:49:9: error: '__lwsync' macro redefined [-Werror,-Wmacro-redefined]
 #define __lwsync()      __asm__ __volatile__ (stringify_in_c(LWSYNC) : : :"memory")
        ^
 <built-in>:308:9: note: previous definition is here
 #define __lwsync __builtin_ppc_lwsync
        ^
 1 error generated.

Undefine this macro so that the runtime patching introduced by
commit 2d1b2027626d ("powerpc: Fixup lwsync at runtime") continues to
work properly with clang and the build no longer breaks.

Cc: stable@vger.kernel.org
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://github.com/ClangBuiltLinux/linux/issues/1386
Link: https://github.com/llvm/llvm-project/commit/62b5df7fe2b3fda1772befeda15598fbef96a614
Link: https://lore.kernel.org/r/20210528182752.1852002-1-nathan@kernel.org

powerpc/mem: Add back missing header to fix 'no previous prototype' error

Commit b26e8f27253a ("powerpc/mem: Move cache flushing functions into
mm/cacheflush.c") removed asm/sparsemem.h which is required when
CONFIG_MEMORY_HOTPLUG is selected to get the declaration of
create_section_mapping().

Add it back.

Fixes: b26e8f27253a ("powerpc/mem: Move cache flushing functions into mm/cacheflush.c")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/3e5b63bb3daab54a1eb9c20221c2e9528c4db9b3.1622883330.git.christophe.leroy@csgroup.eu

Revert "powerpc/kernel/iommu: Align size for IOMMU_PAGE_SIZE() to save TCEs"

This reverts commit 3c0468d4451eb6b4f6604370639f163f9637a479.

That commit was breaking alignment guarantees for the DMA address when
allocating coherent mappings, as described in
Documentation/core-api/dma-api-howto.rst

It was also noticed by Mellanox' driver:
[ 1515.763621] mlx5_core c002:01:00.0: mlx5_frag_buf_alloc_node:146:(pid 13402): unexpected map alignment: 0x0800000000c61000, page_shift=16
[ 1515.763635] mlx5_core c002:01:00.0: mlx5_cqwq_create:181:(pid
13402): mlx5_frag_buf_alloc_node() failed, -12

Fixes: 3c0468d4451e ("powerpc/kernel/iommu: Align size for IOMMU_PAGE_SIZE() to save TCEs")
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210526144540.117795-1-fbarrat@linux.ibm.com

KVM: PPC: Book3S HV: Save host FSCR in the P7/8 path

Similar to commit 25edcc50d76c ("KVM: PPC: Book3S HV: Save and restore
FSCR in the P9 path"), ensure the P7/8 path saves and restores the host
FSCR. The logic explained in that patch actually applies there to the
old path well: a context switch can be made before kvmppc_vcpu_run_hv
restores the host FSCR and returns.

Now both the p9 and the p7/8 paths now save and restore their FSCR, it
no longer needs to be restored at the end of kvmppc_vcpu_run_hv

Fixes: b005255e12a3 ("KVM: PPC: Book3S HV: Context-switch new POWER8 SPRs")
Cc: stable@vger.kernel.org # v3.14+
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Fabiano Rosas <farosas@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210526125851.3436735-1-npiggin@gmail.com

powerpc: Fix reverse map real-mode address lookup with huge vmalloc

real_vmalloc_addr() does not currently work for huge vmalloc, which is
what the reverse map can be allocated with for radix host, hash guest.

Extract the hugepage aware equivalent from eeh code into a helper, and
convert existing sites including this one to use it.

Fixes: 8abddd968a30 ("powerpc/64s/radix: Enable huge vmalloc mappings")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210526120005.3432222-1-npiggin@gmail.com

powerpc/kprobes: Fix validation of prefixed instructions across page boundary

When checking if the probed instruction is the suffix of a prefixed
instruction, we access the instruction at the previous word. If the
probed instruction is the very first word of a module, we can end up
trying to access an invalid page.

Fix this by skipping the check for all instructions at the beginning of
a page. Prefixed instructions cannot cross a 64-byte boundary and as
such, we don't expect to encounter a suffix as the very first word in a
page for kernel text. Even if there are prefixed instructions crossing
a page boundary (from a module, for instance), the instruction will be
illegal, so preventing probing on the suffix of such prefix instructions
isn't worthwhile.

Fixes: b4657f7650ba ("powerpc/kprobes: Don't allow breakpoints on suffixes")
Cc: stable@vger.kernel.org # v5.8+
Reported-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0df9a032a05576a2fa8e97d1b769af2ff0eafbd6.1621416666.git.naveen.n.rao@linux.vnet.ibm.com

tty: hvc: udbg_hvc: retry putc on -EAGAIN

hvterm_raw_put_chars() calls hvc_put_chars(), which may return -EAGAIN
when the underlying hcall returns a "busy" status, but udbg_hvc_putc()
doesn't handle this. When using xmon on a PowerVM guest, this can
result in incomplete or garbled output when printing relatively large
amounts of data quickly, such as when dumping the kernel log buffer.

Call again on -EAGAIN.

Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210514214422.3019105-1-nathanl@linux.ibm.com

powerpc/xmon: make dumping log buffer contents more reliable

Log buffer entries that are too long for dump_log_buf()'s small
local buffer are:

* silently discarded when a single-line entry is too long;
  kmsg_dump_get_line() returns true but sets &len to 0.
* silently truncated to the last fitting new line when a multi-line
  entry is too long, e.g. register dumps from __show_regs(); this
  seems undetectable via the kmsg_dump API.

xmon_printf()'s internal buffer is already 1KB; enlarge
dump_log_buf()'s own buffer to match and make it statically
allocated. Verified that this allows complete printing of register
dumps on ppc64le with both CONFIG_PRINTK_TIME=y and
CONFIG_PRINTK_CALLER=y.

Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210514162420.2911458-1-nathanl@linux.ibm.com

powerpc/kprobes: Replace ppc_optinsn by common optinsn

Commit 51c9c0843993 ("powerpc/kprobes: Implement Optprobes")
implemented a powerpc specific version of optinsn in order
to workaround the 32Mb limitation for direct branches.

Instead of implementing a dedicated powerpc version, use the
common optinsn and override the allocation and freeing functions.

This also indirectly remove the CLANG warning about
is_kprobe_ppc_optinsn_slot() not being use, and the powerpc will
now benefit from commit 5b485629ba0d ("kprobes, extable: Identify
kprobes trampolines as kernel text area")

Suggested-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ec5e85f9f9abcfecc959a03495f4a7858eb4d203.1620896780.git.christophe.leroy@csgroup.eu

kprobes: Allow architectures to override optinsn page allocation

Some architectures like powerpc require a non standard
allocation of optinsn page, because module pages are
too far from the kernel for direct branches.

Define weak alloc_optinsn_page() and free_optinsn_page(), that
fall back on alloc_insn_page() and free_insn_page() when not
overridden by the architecture.

Suggested-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/40a43d6df1fdf41ade36e9a46e60a4df774ca9f6.1620896780.git.christophe.leroy@csgroup.eu

powerpc: Kconfig: disable CONFIG_COMPAT for clang < 12

Until clang-12, clang would attempt to assemble 32b powerpc assembler in
64b emulation mode when using a 64b target triple with -m32, leading to
errors during the build of the compat VDSO. Simply disable all of
CONFIG_COMPAT; users should upgrade to the latest release of clang for
proper support.

Suggested-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://github.com/ClangBuiltLinux/linux/issues/1160
Link: https://github.com/llvm/llvm-project/commits/2288319733cd5f525bf7e24dece08bfcf9d0ff9e
Link: https://groups.google.com/g/clang-built-linux/c/ayNmi3HoNdY/m/XJAGj_G2AgAJ
Link: https://lore.kernel.org/r/20210518205858.2440344-1-ndesaulniers@google.com

powerpc/powernv/pci: fix header guard

While looking at -Wundef warnings, the #if CONFIG_EEH stood out as a
possible candidate to convert to #ifdef CONFIG_EEH.

It seems that based on Kconfig dependencies it's not possible to build
this file without CONFIG_EEH enabled, but based on upstream discussion,
it's not clear yet that CONFIG_EEH should be enabled by default.

For now, simply fix the -Wundef warning.

Suggested-by: Nathan Chancellor <nathan@kernel.org>
Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://github.com/ClangBuiltLinux/linux/issues/570
Link: https://lore.kernel.org/lkml/67f6cd269684c9aa8463ff4812c3b4605e6739c3.camel@perches.com/
Link: https://lore.kernel.org/lkml/CAOSf1CGoN5R0LUrU=Y=UWho1Z_9SLgCX8s3SbFJXwJXc5BYz4A@mail.gmail.com/
Link: https://lore.kernel.org/r/20210518204044.2390064-1-ndesaulniers@google.com

powerpc/sstep: Add tests for setb instruction

This adds selftests for setb instruction.

Signed-off-by: Sathvika Vasireddy <sathvika@linux.vnet.ibm.com>
Tested-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b05b61ccb5f10279d46fed490796f32ea2ccc270.1620727160.git.sathvika@linux.vnet.ibm.com

powerpc/sstep: Add emulation support for ‘setb’ instruction

This adds emulation support for the following instruction:
   * Set Boolean (setb)

Signed-off-by: Sathvika Vasireddy <sathvika@linux.vnet.ibm.com>
Tested-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/7b735b0c898da0db2af8628a64df2f5114596f22.1620727160.git.sathvika@linux.vnet.ibm.com

powerpc/Makefile: Add ppc32/ppc64_randconfig targets

Make it easier to generate a 32 or 64-bit specific randconfig.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Tested-by: Randy Dunlap <rdunlap@infradead.org>
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Requested-by: Randy Dunlap <rdunlap@infradead.org>
Link: https://lore.kernel.org/r/20210428132700.3426100-1-mpe@ellerman.id.au

powerpc/pseries: minor enhancements in dlpar_memory_remove_by_ic()

We don't need the 'lmbs_available' variable to count the valid LMBs and
to check if we have less than 'lmbs_to_remove'. We must ensure that the
entire LMB range must be removed, so we can error out immediately if any
LMB in the range is marked as reserved.

Add a couple of comments explaining the reasoning behind the differences
we have in this function in contrast to what it is done in its sister
function, dlpar_memory_remove_by_count().

Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210512202809.95363-5-danielhb413@gmail.com

powerpc/pseries: break early in dlpar_memory_remove_by_count() loops

After marking the LMBs as reserved depending on dlpar_remove_lmb() rc,
we evaluate whether we need to add the LMBs back or if we can release
the LMB DRCs. In both cases, a for_each_drmem_lmb() loop without a break
condition is used. This means that we're going to cycle through all LMBs
of the partition even after we're done with what we were going to do.

This patch adds break conditions in both loops to avoid this. The
'lmbs_removed' variable was renamed to 'lmbs_reserved', and it's now
being decremented each time a lmb reservation is removed, indicating
that the operation we're doing (adding back LMBs or releasing DRCs) is
completed.

Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210512202809.95363-4-danielhb413@gmail.com

powerpc/pseries: check DRCONF_MEM_RESERVED in lmb_is_removable()

DRCONF_MEM_RESERVED is a flag that represents the "Reserved Memory"
status in LOPAR v2.10, section 4.2.8. If a LMB is marked as reserved,
quoting LOPAR, "is not to be used or altered by the base OS". This flag
is read only in the kernel, being set by the firmware/hypervisor in the
DT. As an example, QEMU will set this flag in hw/ppc/spapr.c,
spapr_dt_dynamic_memory().

lmb_is_removable() does not check for DRCONF_MEM_RESERVED. This function
is used in dlpar_remove_lmb() as a guard before the removal logic. Since
it is failing to check for !RESERVED, dlpar_remove_lmb() will fail in a
later stage instead of failing in the validation when receiving a
reserved LMB as input.

lmb_is_removable() is also used in dlpar_memory_remove_by_count() to
evaluate if we have enough LMBs to complete the request. The missing
!RESERVED check in this case is causing dlpar_memory_remove_by_count()
to miscalculate the number of elegible LMBs for the removal, and can
make it error out later on instead of failing in the validation with the
'not enough LMBs to satisfy request' message.

Making a DRCONF_MEM_RESERVED check in lmb_is_removable() fixes all these
issues.

Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210512202809.95363-3-danielhb413@gmail.com

powerpc/pseries: Set UNISOLATE on dlpar_memory_remove_by_ic() error

As previously done in dlpar_cpu_remove() for CPUs, this patch changes
dlpar_memory_remove_by_ic() to unisolate the LMB DRC when the LMB is
failed to be removed. The hypervisor, seeing a LMB DRC that was supposed
to be removed being unisolated instead, can do error recovery on its
side.

This change is done in dlpar_memory_remove_by_ic() only because, as of
today, only QEMU is using this code path for error recovery (via the
PSERIES_HP_ELOG_ID_DRC_IC event). phyp treats it as a no-op.

Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210512202809.95363-2-danielhb413@gmail.com

powerpc/pseries/ras: Delete a redundant condition branch

The statement of the last "if (xxx)" branch is the same as the "else"
branch. Delete it to simplify code.

No functional change.

Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210510131924.3907-1-thunder.leizhen@huawei.com

selftests/powerpc: Fix duplicate included pthread.h

Clean up the following includecheck warning:

./tools/testing/selftests/powerpc/tm/tm-vmx-unavail.c: pthread.h is
included more than once.

No functional change.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1620903820-68213-1-git-send-email-jiapeng.chong@linux.alibaba.com

powerpc/pseries/memhotplug: Remove unused inline function dlpar_memory_remove()

dlpar_memory_remove() is never used, so can be removed.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210514071041.17432-1-yuehaibing@huawei.com

powerpc/64s/syscall: Fix ptrace syscall info with scv syscalls

The scv implementation missed updating syscall return value and error
value get/set functions to deal with the changed register ABI. This
broke ptrace PTRACE_GET_SYSCALL_INFO as well as some kernel auditing
and tracing functions.

Fix. tools/testing/selftests/ptrace/get_syscall_info now passes when
scv is used.

Fixes: 7fa95f9adaee ("powerpc/64s: system call support for scv/rfscv instructions")
Cc: stable@vger.kernel.org # v5.9+
Reported-by: "Dmitry V. Levin" <ldv@altlinux.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Dmitry V. Levin <ldv@altlinux.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210520111931.2597127-2-npiggin@gmail.com

powerpc/64s/syscall: Use pt_regs.trap to distinguish syscall ABI difference between sc and scv syscalls

The sc and scv 0 system calls have different ABI conventions, and
ptracers need to know which system call type is being used if they want
to look at the syscall registers.

Document that pt_regs.trap can be used for this, and fix one in-tree user
to work with scv 0 syscalls.

Fixes: 7fa95f9adaee ("powerpc/64s: system call support for scv/rfscv instructions")
Cc: stable@vger.kernel.org # v5.9+
Reported-by: "Dmitry V. Levin" <ldv@altlinux.org>
Suggested-by: "Dmitry V. Levin" <ldv@altlinux.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210520111931.2597127-1-npiggin@gmail.com

powerpc: Fix early setup to make early_ioremap() work

The immediate problem is that after commit
0bd3f9e953bd ("powerpc/legacy_serial: Use early_ioremap()") the kernel
silently reboots on some systems.

The reason is that early_ioremap() returns broken addresses as it uses
slot_virt[] array which initialized with offsets from FIXADDR_TOP ==
IOREMAP_END+FIXADDR_SIZE == KERN_IO_END - FIXADDR_SIZ + FIXADDR_SIZE ==
__kernel_io_end which is 0 when early_ioremap_setup() is called.
__kernel_io_end is initialized little bit later in early_init_mmu().

This fixes the initialization by swapping early_ioremap_setup() and
early_init_mmu().

Fixes: 265c3491c4bc ("powerpc: Add support for GENERIC_EARLY_IOREMAP")
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Drop unrelated cleanup & cleanup change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210520032919.358935-1-aik@ozlabs.ru

selftests/powerpc: Add test of mitigation patching

We recently discovered some of our mitigation patching was not safe
against other CPUs running concurrently.

Add a test which enable/disables all mitigations in a tight loop while
also running some stress load. On an unpatched system this almost always
leads to an oops and panic/reboot, but we also check if the kernel
becomes tainted in case we have a non-fatal oops.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210507064225.1556312-1-mpe@ellerman.id.au

powerpc/32s: Remove asm/book3s/32/hash.h

Move the PAGE bits into pgtable.h to be more similar to book3s/64.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/7f4aaa479569328a1e5b07c96c08fbca0cd7dd88.1620307890.git.christophe.leroy@csgroup.eu