linux-2.6-microblaze.git
2 years agopowerpc/dcr: Use cmplwi instead of 3-argument cmpli
Michael Ellerman [Thu, 14 Oct 2021 02:44:24 +0000 (13:44 +1100)]
powerpc/dcr: Use cmplwi instead of 3-argument cmpli

In dcr-low.S we use cmpli with three arguments, instead of four
arguments as defined in the ISA:

cmpli cr0,r3,1024

This appears to be a PPC440-ism, looking at the "PPC440x5 CPU Core
User’s Manual" it shows cmpli having no L field, but implied to be 0 due
to the core being 32-bit. It mentions that the ISA defines four
arguments and recommends using cmplwi.

It also corresponds to the old POWER instruction set, which had no L
field there, a reserved bit instead.

dcr-low.S is only built 32-bit, because it is only built when
DCR_NATIVE=y, which is only selected by 40x and 44x. Looking at the
generated code (with gcc/gas) we see cmplwi as expected.

Although gas is happy with the 3-argument version when building for
32-bit, the LLVM assembler is not and errors out with:

  arch/powerpc/sysdev/dcr-low.S:27:10: error: invalid operand for instruction
   cmpli 0,%r3,1024; ...
           ^

Switch to the cmplwi extended opcode, which avoids any confusion when
reading the ISA, fixes the issue with the LLVM assembler, and also means
the code could be built 64-bit in future (though that's very unlikely).

Reported-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
BugLink: https://github.com/ClangBuiltLinux/linux/issues/1419
Link: https://lore.kernel.org/r/20211014024424.528848-1-mpe@ellerman.id.au
2 years agopowerpc/security: Use a mutex for interrupt exit code patching
Russell Currey [Wed, 27 Oct 2021 07:24:10 +0000 (17:24 +1000)]
powerpc/security: Use a mutex for interrupt exit code patching

The mitigation-patching.sh script in the powerpc selftests toggles
all mitigations on and off simultaneously, revealing that rfi_flush
and stf_barrier cannot safely operate at the same time due to races
in updating the static key.

On some systems, the static key code throws a warning and the kernel
remains functional.  On others, the kernel will hang or crash.

Fix this by slapping on a mutex.

Fixes: 13799748b957 ("powerpc/64: use interrupt restart table to speed up return from interrupt")
Cc: stable@vger.kernel.org # v5.14+
Signed-off-by: Russell Currey <ruscur@russell.cc>
Acked-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211027072410.40950-1-ruscur@russell.cc
2 years agopowerpc/83xx/mpc8349emitx: Make mcu_gpiochip_remove() return void
Uwe Kleine-König [Thu, 21 Oct 2021 10:56:57 +0000 (12:56 +0200)]
powerpc/83xx/mpc8349emitx: Make mcu_gpiochip_remove() return void

Up to now mcu_gpiochip_remove() returns zero unconditionally. Make it
return void instead which makes it easier to see in the callers that
there is no error to handle.

Also the return value of i2c remove callbacks is ignored anyway.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211021105657.72572-1-u.kleine-koenig@pengutronix.de
2 years agopowerpc/fsl_booke: Fix setting of exec flag when setting TLBCAMs
Christophe Leroy [Tue, 26 Oct 2021 05:39:26 +0000 (07:39 +0200)]
powerpc/fsl_booke: Fix setting of exec flag when setting TLBCAMs

Building tqm8541_defconfig results in:

arch/powerpc/mm/nohash/fsl_book3e.c: In function 'settlbcam':
arch/powerpc/mm/nohash/fsl_book3e.c:126:40: error: '_PAGE_BAP_SX' undeclared (first use in this function)
  126 |         TLBCAM[index].MAS3 |= (flags & _PAGE_BAP_SX) ? MAS3_SX : 0;
      |                                        ^~~~~~~~~~~~
arch/powerpc/mm/nohash/fsl_book3e.c:126:40: note: each undeclared identifier is reported only once for each function it appears in
make[3]: *** [scripts/Makefile.build:277: arch/powerpc/mm/nohash/fsl_book3e.o] Error 1
make[2]: *** [scripts/Makefile.build:540: arch/powerpc/mm/nohash] Error 2
make[1]: *** [scripts/Makefile.build:540: arch/powerpc/mm] Error 2
make: *** [Makefile:1868: arch/powerpc] Error 2

This is because _PAGE_BAP_SX is not defined when using 32 bits PTE.

Now that _PAGE_EXEC contains both _PAGE_BAP_SX and _PAGE_BAP_UX, it can be used instead.

Fixes: 01116e6e98b0 ("powerpc/fsl_booke: Take exec flag into account when setting TLBCAMs")
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/91a0235e7f2a85308b84aa5b9efd8d022e2b899a.1635226743.git.christophe.leroy@csgroup.eu
2 years agopowerpc/book3e: Fix set_memory_x() and set_memory_nx()
Christophe Leroy [Tue, 26 Oct 2021 05:39:25 +0000 (07:39 +0200)]
powerpc/book3e: Fix set_memory_x() and set_memory_nx()

set_memory_x() calls pte_mkexec() which sets _PAGE_EXEC.
set_memory_nx() calls pte_exprotec() which clears _PAGE_EXEC.

Book3e has 2 bits, UX and SX, which defines the exec rights
resp. for user (PR=1) and for kernel (PR=0).

_PAGE_EXEC is defined as UX only.

An executable kernel page is set with either _PAGE_KERNEL_RWX
or _PAGE_KERNEL_ROX, which both have SX set and UX cleared.

So set_memory_nx() call for an executable kernel page does
nothing because UX is already cleared.

And set_memory_x() on a non-executable kernel page makes it
executable for the user and keeps it non-executable for kernel.

Also, pte_exec() always returns 'false' on kernel pages, because
it checks _PAGE_EXEC which doesn't include SX, so for instance
the W+X check doesn't work.

To fix this:
  - change tlb_low_64e.S to use _PAGE_BAP_UX instead of _PAGE_USER
  - sets both UX and SX in _PAGE_EXEC so that pte_exec() returns
    true whenever one of the two bits is set and pte_exprotect()
    clears both bits.
  - Define a book3e specific version of pte_mkexec() which sets
    either SX or UX based on UR.

Fixes: 1f9ad21c3b38 ("powerpc/mm: Implement set_memory() routines")
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c41100f9c144dc5b62e5a751b810190c6b5d42fd.1635226743.git.christophe.leroy@csgroup.eu
2 years agopowerpc/nohash: Fix __ptep_set_access_flags() and ptep_set_wrprotect()
Christophe Leroy [Tue, 26 Oct 2021 05:39:24 +0000 (07:39 +0200)]
powerpc/nohash: Fix __ptep_set_access_flags() and ptep_set_wrprotect()

Commit 26973fa5ac0e ("powerpc/mm: use pte helpers in generic code")
changed those two functions to use pte helpers to determine which
bits to clear and which bits to set.

This change was based on the assumption that bits to be set/cleared
are always the same and can be determined by applying the pte
manipulation helpers on __pte(0).

But on platforms like book3e, the bits depend on whether the page
is a user page or not.

For the time being it more or less works because of _PAGE_EXEC being
used for user pages only and exec right being set at all time on
kernel page. But following patch will clean that and output of
pte_mkexec() will depend on the page being a user or kernel page.

Instead of trying to make an even more complicated helper where bits
would become dependent on the final pte value, come back to a more
static situation like before commit 26973fa5ac0e ("powerpc/mm: use
pte helpers in generic code"), by introducing an 8xx specific
version of __ptep_set_access_flags() and ptep_set_wrprotect().

Fixes: 26973fa5ac0e ("powerpc/mm: use pte helpers in generic 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/922bdab3a220781bae2360ff3dd5adb7fe4d34f1.1635226743.git.christophe.leroy@csgroup.eu
2 years agopowerpc/bpf: Fix write protecting JIT code
Hari Bathini [Mon, 25 Oct 2021 05:56:49 +0000 (11:26 +0530)]
powerpc/bpf: Fix write protecting JIT code

Running program with bpf-to-bpf function calls results in data access
exception (0x300) with the below call trace:

  bpf_int_jit_compile+0x238/0x750 (unreliable)
  bpf_check+0x2008/0x2710
  bpf_prog_load+0xb00/0x13a0
  __sys_bpf+0x6f4/0x27c0
  sys_bpf+0x2c/0x40
  system_call_exception+0x164/0x330
  system_call_vectored_common+0xe8/0x278

as bpf_int_jit_compile() tries writing to write protected JIT code
location during the extra pass.

Fix it by holding off write protection of JIT code until the extra
pass, where branch target addresses fixup happens.

Fixes: 62e3d4210ac9 ("powerpc/bpf: Write protect JIT code")
Cc: stable@vger.kernel.org # v5.14+
Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
Reviewed-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/20211025055649.114728-1-hbathini@linux.ibm.com
2 years agoselftests/powerpc: Use date instead of EPOCHSECONDS in mitigation-patching.sh
Russell Currey [Mon, 25 Oct 2021 10:24:36 +0000 (20:24 +1000)]
selftests/powerpc: Use date instead of EPOCHSECONDS in mitigation-patching.sh

The EPOCHSECONDS environment variable was added in bash 5.0 (released
2019).  Some distributions of the "stable" and "long-term" variety ship
older versions of bash than this, so swap to using the date command
instead.

"%s" was added to coreutils `date` in 1993 so we should be good, but who
knows, it is a GNU extension and not part of the POSIX spec for `date`.

Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211025102436.19177-1-ruscur@russell.cc
2 years agopowerpc/64s/interrupt: Fix check_return_regs_valid() false positive
Nicholas Piggin [Tue, 26 Oct 2021 12:25:31 +0000 (22:25 +1000)]
powerpc/64s/interrupt: Fix check_return_regs_valid() false positive

The check_return_regs_valid() can cause a false positive if the return
regs are marked as norestart and they are an HSRR type interrupt,
because the low bit in the bottom of regs->trap causes interrupt type
matching to fail.

This can occcur for example on bare metal with a HV privileged doorbell
interrupt that causes a signal, but do_signal returns early because
get_signal() fails, and takes the "No signal to deliver" path. In this
case no signal was delivered so the return location is not changed so
return SRRs are not invalidated, yet set_trap_norestart is called, which
messes up the match. Building go-1.16.6 is known to reproduce this.

Fix it by using the TRAP() accessor which masks out the low bit.

Fixes: 6eaaf9de3599 ("powerpc/64s/interrupt: Check and fix srr_valid without crashing")
Cc: stable@vger.kernel.org # v5.14+
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211026122531.3599918-1-npiggin@gmail.com
2 years agopowerpc/boot: Set LC_ALL=C in wrapper script
Christophe Leroy [Tue, 26 Oct 2021 13:48:29 +0000 (15:48 +0200)]
powerpc/boot: Set LC_ALL=C in wrapper script

While trying to build a simple Image for ACADIA platform, I got the
following error:

  WRAP    arch/powerpc/boot/simpleImage.acadia
INFO: Uncompressed kernel (size 0x6ae7d0) overlaps the address of the wrapper(0x400000)
INFO: Fixing the link_address of wrapper to (0x700000)
powerpc64-linux-gnu-ld : mode d'émulation non reconnu : -T
Émulations prises en charge : elf64ppc elf32ppc elf32ppclinux elf32ppcsim elf64lppc elf32lppc elf32lppclinux elf32lppcsim
make[1]: *** [arch/powerpc/boot/Makefile:424 : arch/powerpc/boot/simpleImage.acadia] Erreur 1
make: *** [arch/powerpc/Makefile:285 : simpleImage.acadia] Erreur 2

Trying again with V=1 shows the following command

powerpc64-linux-gnu-ld -m -T arch/powerpc/boot/zImage.lds -Ttext 0x700000 --no-dynamic-linker -o arch/powerpc/boot/simpleImage.acadia -Map wrapper.map arch/powerpc/boot/fixed-head.o arch/powerpc/boot/simpleboot.o ./zImage.3278022.o arch/powerpc/boot/wrapper.a

The argument of '-m' is missing.

This is due to the wrapper script calling 'objdump -p vmlinux' and
looking for 'file format', whereas the output of objdump is:

vmlinux:     format de fichier elf32-powerpc

En-tête de programme:
    LOAD off    0x00010000 vaddr 0xc0000000 paddr 0x00000000 align 2**16
         filesz 0x0069e1d4 memsz 0x006c128c flags rwx
    NOTE off    0x0064591c vaddr 0xc063591c paddr 0x0063591c align 2**2
         filesz 0x00000054 memsz 0x00000054 flags ---

Add LC_ALL=C at the beginning of the wrapper script in order to get the
output expected by the script:

vmlinux:     file format elf32-powerpc

Program Header:
    LOAD off    0x00010000 vaddr 0xc0000000 paddr 0x00000000 align 2**16
         filesz 0x0069e1d4 memsz 0x006c128c flags rwx
    NOTE off    0x0064591c vaddr 0xc063591c paddr 0x0063591c align 2**2
         filesz 0x00000054 memsz 0x00000054 flags ---

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Acked-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/a9ff3bc98035f63b122c051f02dc47c7aed10430.1635256089.git.christophe.leroy@csgroup.eu
2 years agopowerpc/64s: Default to 64K pages for 64 bit book3s
Joel Stanley [Fri, 15 Oct 2021 00:16:49 +0000 (10:46 +1030)]
powerpc/64s: Default to 64K pages for 64 bit book3s

For 64-bit book3s the default should be 64K as that's what modern CPUs
are designed for.

The following defconfigs already set CONFIG_PPC_64K_PAGES:

 cell_defconfig
 pasemi_defconfig
 powernv_defconfig
 ppc64_defconfig
 pseries_defconfig
 skiroot_defconfig

The have the option removed from the defconfig, as it is now the
default.

The defconfigs that now need to set CONFIG_PPC_4K_PAGES to maintain
their existing behaviour are:

 g5_defconfig
 maple_defconfig
 microwatt_defconfig
 ps3_defconfig

Signed-off-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
BugLink: https://github.com/linuxppc/issues/issues/109
Link: https://lore.kernel.org/r/20211015001649.45591-1-joel@jms.id.au
2 years agoRevert "powerpc/audit: Convert powerpc to AUDIT_ARCH_COMPAT_GENERIC"
Michael Ellerman [Wed, 27 Oct 2021 11:29:31 +0000 (22:29 +1100)]
Revert "powerpc/audit: Convert powerpc to AUDIT_ARCH_COMPAT_GENERIC"

This reverts commit 566af8cda399c088763d07464463dc871c943b54.

This caused some conflicts vs the audit tree, and the audit maintainers
would prefer we postpone this to the next merge window so we have more
time for testing.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2 years agopowerpc/pseries/mobility: ignore ibm, platform-facilities updates
Nathan Lynch [Wed, 20 Oct 2021 19:47:03 +0000 (14:47 -0500)]
powerpc/pseries/mobility: ignore ibm, platform-facilities updates

On VMs with NX encryption, compression, and/or RNG offload, these
capabilities are described by nodes in the ibm,platform-facilities device
tree hierarchy:

  $ tree -d /sys/firmware/devicetree/base/ibm,platform-facilities/
  /sys/firmware/devicetree/base/ibm,platform-facilities/
  ├── ibm,compression-v1
  ├── ibm,random-v1
  └── ibm,sym-encryption-v1

  3 directories

The acceleration functions that these nodes describe are not disrupted by
live migration, not even temporarily.

But the post-migration ibm,update-nodes sequence firmware always sends
"delete" messages for this hierarchy, followed by an "add" directive to
reconstruct it via ibm,configure-connector (log with debugging statements
enabled in mobility.c):

  mobility: removing node /ibm,platform-facilities/ibm,random-v1:4294967285
  mobility: removing node /ibm,platform-facilities/ibm,compression-v1:4294967284
  mobility: removing node /ibm,platform-facilities/ibm,sym-encryption-v1:4294967283
  mobility: removing node /ibm,platform-facilities:4294967286
  ...
  mobility: added node /ibm,platform-facilities:4294967286

Note we receive a single "add" message for the entire hierarchy, and what
we receive from the ibm,configure-connector sequence is the top-level
platform-facilities node along with its three children. The debug message
simply reports the parent node and not the whole subtree.

Also, significantly, the nodes added are almost completely equivalent to
the ones removed; even phandles are unchanged. ibm,shared-interrupt-pool in
the leaf nodes is the only property I've observed to differ, and Linux does
not use that. So in practice, the sum of update messages Linux receives for
this hierarchy is equivalent to minor property updates.

We succeed in removing the original hierarchy from the device tree. But the
vio bus code is ignorant of this, and does not unbind or relinquish its
references. The leaf nodes, still reachable through sysfs, of course still
refer to the now-freed ibm,platform-facilities parent node, which makes
use-after-free possible:

  refcount_t: addition on 0; use-after-free.
  WARNING: CPU: 3 PID: 1706 at lib/refcount.c:25 refcount_warn_saturate+0x164/0x1f0
  refcount_warn_saturate+0x160/0x1f0 (unreliable)
  kobject_get+0xf0/0x100
  of_node_get+0x30/0x50
  of_get_parent+0x50/0xb0
  of_fwnode_get_parent+0x54/0x90
  fwnode_count_parents+0x50/0x150
  fwnode_full_name_string+0x30/0x110
  device_node_string+0x49c/0x790
  vsnprintf+0x1c0/0x4c0
  sprintf+0x44/0x60
  devspec_show+0x34/0x50
  dev_attr_show+0x40/0xa0
  sysfs_kf_seq_show+0xbc/0x200
  kernfs_seq_show+0x44/0x60
  seq_read_iter+0x2a4/0x740
  kernfs_fop_read_iter+0x254/0x2e0
  new_sync_read+0x120/0x190
  vfs_read+0x1d0/0x240

Moreover, the "new" replacement subtree is not correctly added to the
device tree, resulting in ibm,platform-facilities parent node without the
appropriate leaf nodes, and broken symlinks in the sysfs device hierarchy:

  $ tree -d /sys/firmware/devicetree/base/ibm,platform-facilities/
  /sys/firmware/devicetree/base/ibm,platform-facilities/

  0 directories

  $ cd /sys/devices/vio ; find . -xtype l -exec file {} +
  ./ibm,sym-encryption-v1/of_node: broken symbolic link to
    ../../../firmware/devicetree/base/ibm,platform-facilities/ibm,sym-encryption-v1
  ./ibm,random-v1/of_node:         broken symbolic link to
    ../../../firmware/devicetree/base/ibm,platform-facilities/ibm,random-v1
  ./ibm,compression-v1/of_node:    broken symbolic link to
    ../../../firmware/devicetree/base/ibm,platform-facilities/ibm,compression-v1

This is because add_dt_node() -> dlpar_attach_node() attaches only the
parent node returned from configure-connector, ignoring any children. This
should be corrected for the general case, but fixing that won't help with
the stale OF node references, which is the more urgent problem.

One way to address that would be to make the drivers respond to node
removal notifications, so that node references can be dropped
appropriately. But this would likely force the drivers to disrupt active
clients for no useful purpose: equivalent nodes are immediately re-added.
And recall that the acceleration capabilities described by the nodes remain
available throughout the whole process.

The solution I believe to be robust for this situation is to convert
remove+add of a node with an unchanged phandle to an update of the node's
properties in the Linux device tree structure. That would involve changing
and adding a fair amount of code, and may take several iterations to land.

Until that can be realized we have a confirmed use-after-free and the
possibility of memory corruption. So add a limited workaround that
discriminates on the node type, ignoring adds and removes. This should be
amenable to backporting in the meantime.

Fixes: 410bccf97881 ("powerpc/pseries: Partition migration in the kernel")
Cc: stable@vger.kernel.org
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211020194703.2613093-1-nathanl@linux.ibm.com
2 years agopowerpc/32: Don't use a struct based type for pte_t
Christophe Leroy [Fri, 17 Sep 2021 13:57:31 +0000 (15:57 +0200)]
powerpc/32: Don't use a struct based type for pte_t

Long time ago we had a config item called STRICT_MM_TYPECHECKS
to build the kernel with pte_t defined as a structure in order
to perform additional build checks or build it with pte_t
defined as a simple type in order to get simpler generated code.

Commit 670eea924198 ("powerpc/mm: Always use STRICT_MM_TYPECHECKS")
made the struct based definition the only one, considering that the
generated code was similar in both cases.

That's right on ppc64 because the ABI is such that the content of a
struct having a single simple type element is passed as register,
but on ppc32 such a structure is passed via the stack like any
structure.

Simple test function:

pte_t test(pte_t pte)
{
return pte;
}

Before this patch we get

c00108ec <test>:
c00108ec: 81 24 00 00  lwz     r9,0(r4)
c00108f0: 91 23 00 00  stw     r9,0(r3)
c00108f4: 4e 80 00 20  blr

So, for PPC32, restore the simple type behaviour we got before
commit 670eea924198, but instead of adding a config option to
activate type check, do it when __CHECKER__ is set so that type
checking is performed by 'sparse' and provides feedback like:

arch/powerpc/mm/pgtable.c:466:16: warning: incorrect type in return expression (different base types)
arch/powerpc/mm/pgtable.c:466:16:    expected unsigned long
arch/powerpc/mm/pgtable.c:466:16:    got struct pte_t [usertype] x

With this patch we now get

c0010890 <test>:
c0010890: 4e 80 00 20  blr

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Define STRICT_MM_TYPECHECKS rather than repeating the condition]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c904599f33aaf6bb7ee2836a9ff8368509e0d78d.1631887042.git.christophe.leroy@csgroup.eu
2 years agopowerpc/breakpoint: Cleanup
Christophe Leroy [Wed, 22 Sep 2021 13:37:18 +0000 (15:37 +0200)]
powerpc/breakpoint: Cleanup

cache_op_size() does exactly the same as l1_dcache_bytes().

Remove it.

MSR_64BIT already exists, no need to enclode the check
around #ifdef __powerpc64__

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/6184b08088312a7d787d450eb902584e4ae77f7a.1632317816.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Activate CONFIG_STRICT_KERNEL_RWX by default
Christophe Leroy [Fri, 24 Sep 2021 17:13:53 +0000 (19:13 +0200)]
powerpc: Activate CONFIG_STRICT_KERNEL_RWX by default

CONFIG_STRICT_KERNEL_RWX should be set by default on every
architectures (See https://github.com/KSPP/linux/issues/4)

On PPC32 we have to find a compromise between performance and/or
memory wasting and selection of strict_kernel_rwx, because it implies
either smaller memory chunks or larger alignment between RO memory
and RW memory.

For instance the 8xx maps memory with 8M pages. So either the limit
between RO and RW must be 8M aligned or it falls back or 512k pages
which implies more pressure on the TLB.

book3s/32 maps memory with BATs as much as possible. BATS can have
any power-of-two size between 128k and 256M but we have only 4 to 8
BATs so the alignment must be good enough to allow efficient use of
the BATs and avoid falling back on standard page mapping which would
kill performance.

So let's go one step forward and make it the default but still allow
users to unset it when wanted.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/057c40164084bfc7d77c0b2ff78d95dbf6a2a21b.1632503622.git.christophe.leroy@csgroup.eu
2 years agopowerpc/8xx: Simplify TLB handling
Christophe Leroy [Fri, 17 Sep 2021 13:57:12 +0000 (15:57 +0200)]
powerpc/8xx: Simplify TLB handling

In the old days, TLB handling for 8xx was using tlbie and tlbia
instructions directly as much as possible.

But commit f048aace29e0 ("powerpc/mm: Add SMP support to no-hash
TLB handling") broke that by introducing out-of-line unnecessary
complex functions for booke/smp which don't have tlbie/tlbia
instructions and require more complex handling.

Restore direct use of tlbie and tlbia for 8xx which is never SMP.

With this patch we now get

c00ecc68 <ptep_clear_flush>:
c00ecc68: 39 00 00 00  li      r8,0
c00ecc6c: 81 46 00 00  lwz     r10,0(r6)
c00ecc70: 91 06 00 00  stw     r8,0(r6)
c00ecc74: 7c 00 2a 64  tlbie   r5,r0
c00ecc78: 7c 00 04 ac  hwsync
c00ecc7c: 91 43 00 00  stw     r10,0(r3)
c00ecc80: 4e 80 00 20  blr

Before it was

c0012880 <local_flush_tlb_page>:
c0012880: 2c 03 00 00  cmpwi   r3,0
c0012884: 41 82 00 54  beq     c00128d8 <local_flush_tlb_page+0x58>
c0012888: 81 22 00 00  lwz     r9,0(r2)
c001288c: 81 43 00 20  lwz     r10,32(r3)
c0012890: 39 29 00 01  addi    r9,r9,1
c0012894: 91 22 00 00  stw     r9,0(r2)
c0012898: 2c 0a 00 00  cmpwi   r10,0
c001289c: 41 82 00 10  beq     c00128ac <local_flush_tlb_page+0x2c>
c00128a0: 81 2a 01 dc  lwz     r9,476(r10)
c00128a4: 2c 09 ff ff  cmpwi   r9,-1
c00128a8: 41 82 00 0c  beq     c00128b4 <local_flush_tlb_page+0x34>
c00128ac: 7c 00 22 64  tlbie   r4,r0
c00128b0: 7c 00 04 ac  hwsync
c00128b4: 81 22 00 00  lwz     r9,0(r2)
c00128b8: 39 29 ff ff  addi    r9,r9,-1
c00128bc: 2c 09 00 00  cmpwi   r9,0
c00128c0: 91 22 00 00  stw     r9,0(r2)
c00128c4: 4c a2 00 20  bclr+   4,eq
c00128c8: 81 22 00 70  lwz     r9,112(r2)
c00128cc: 71 29 00 04  andi.   r9,r9,4
c00128d0: 4d 82 00 20  beqlr
c00128d4: 48 65 76 74  b       c0669f48 <preempt_schedule>
c00128d8: 81 22 00 00  lwz     r9,0(r2)
c00128dc: 39 29 00 01  addi    r9,r9,1
c00128e0: 91 22 00 00  stw     r9,0(r2)
c00128e4: 4b ff ff c8  b       c00128ac <local_flush_tlb_page+0x2c>
...
c00ecdc8 <ptep_clear_flush>:
c00ecdc8: 94 21 ff f0  stwu    r1,-16(r1)
c00ecdcc: 39 20 00 00  li      r9,0
c00ecdd0: 93 c1 00 08  stw     r30,8(r1)
c00ecdd4: 83 c6 00 00  lwz     r30,0(r6)
c00ecdd8: 91 26 00 00  stw     r9,0(r6)
c00ecddc: 93 e1 00 0c  stw     r31,12(r1)
c00ecde0: 7c 08 02 a6  mflr    r0
c00ecde4: 7c 7f 1b 78  mr      r31,r3
c00ecde8: 7c 83 23 78  mr      r3,r4
c00ecdec: 7c a4 2b 78  mr      r4,r5
c00ecdf0: 90 01 00 14  stw     r0,20(r1)
c00ecdf4: 4b f2 5a 8d  bl      c0012880 <local_flush_tlb_page>
c00ecdf8: 93 df 00 00  stw     r30,0(r31)
c00ecdfc: 7f e3 fb 78  mr      r3,r31
c00ece00: 80 01 00 14  lwz     r0,20(r1)
c00ece04: 83 c1 00 08  lwz     r30,8(r1)
c00ece08: 83 e1 00 0c  lwz     r31,12(r1)
c00ece0c: 7c 08 03 a6  mtlr    r0
c00ece10: 38 21 00 10  addi    r1,r1,16
c00ece14: 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/fb324f1c8f2ddb57cf6aad1cea26329558f1c1c0.1631887021.git.christophe.leroy@csgroup.eu
2 years agopowerpc/lib/sstep: Don't use __{get/put}_user() on kernel addresses
Christophe Leroy [Thu, 16 Sep 2021 18:43:36 +0000 (20:43 +0200)]
powerpc/lib/sstep: Don't use __{get/put}_user() on kernel addresses

In the old days, when we didn't have kernel userspace access
protection and had set_fs(), it was wise to use __get_user()
and friends to read kernel memory.

Nowadays, get_user() and put_user() are granting userspace access and
are exclusively for userspace access.

Convert single step emulation functions to user_access_begin() and
friends and use unsafe_get_user() and unsafe_put_user().

When addressing kernel addresses, there is no need to open userspace
access. And for book3s/32 it is particularly important to no try and
open userspace access on kernel address, because that would break the
content of kernel space segment registers. No guard has been put
against that risk in order to avoid degrading performance.

copy_from_kernel_nofault() and copy_to_kernel_nofault() should
be used but they are out-of-line functions which would degrade
performance. Those two functions are making use of
__get_kernel_nofault() and __put_kernel_nofault() macros.
Those two macros are just wrappers behind __get_user_size_goto() and
__put_user_size_goto().

unsafe_get_user() and unsafe_put_user() are also wrappers of
__get_user_size_goto() and __put_user_size_goto(). Use them to
access kernel space. That allows refactoring userspace and
kernelspace access.

Reported-by: Stan Johnson <userm57@yahoo.com>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Depends-on: 4fe5cda9f89d ("powerpc/uaccess: Implement user_read_access_begin and user_write_access_begin")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/22831c9d17f948680a12c5292e7627288b15f713.1631817805.git.christophe.leroy@csgroup.eu
2 years agopowerpc: warn on emulation of dcbz instruction in kernel mode
Christophe Leroy [Thu, 16 Sep 2021 14:52:09 +0000 (16:52 +0200)]
powerpc: warn on emulation of dcbz instruction in kernel mode

dcbz instruction shouldn't be used on non-cached memory. Using
it on non-cached memory can result in alignment exception and
implies a heavy handling.

Instead of silentely emulating the instruction and resulting in high
performance degradation, warn whenever an alignment exception is
taken in kernel mode due to dcbz, so that the user is made aware that
dcbz instruction has been used unexpectedly by the kernel.

Reported-by: Stan Johnson <userm57@yahoo.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/2e3acfe63d289c6fba366e16973c9ab8369e8b75.1631803922.git.christophe.leroy@csgroup.eu
2 years agopowerpc/32: Add support for out-of-line static calls
Christophe Leroy [Wed, 1 Sep 2021 08:30:21 +0000 (08:30 +0000)]
powerpc/32: Add support for out-of-line static calls

Add support for out-of-line static calls on PPC32. This change
improve performance of calls to global function pointers by
using direct calls instead of indirect calls.

The trampoline is initialy populated with a 'blr' or branch to target,
followed by an unreachable long jump sequence.

In order to cater with parallele execution, the trampoline needs to
be updated in a way that ensures it remains consistent at all time.
This means we can't use the traditional lis/addi to load r12 with
the target address, otherwise there would be a window during which
the first instruction contains the upper part of the new target
address while the second instruction still contains the lower part of
the old target address. To avoid that the target address is stored
just after the 'bctr' and loaded from there with a single instruction.

Then, depending on the target distance, arch_static_call_transform()
will either replace the first instruction by a direct 'bl <target>' or
'nop' in order to have the trampoline fall through the long jump
sequence.

For the special case of __static_call_return0(), to avoid the risk of
a far branch, a version of it is inlined at the end of the trampoline.

Performancewise the long jump sequence is probably not better than
the indirect calls set by GCC when we don't use static calls, but
such calls are unlikely to be required on powerpc32: With most
configurations the kernel size is far below 32 Mbytes so only
modules may happen to be too far. And even modules are likely to
be close enough as they are allocated below the kernel core and
as close as possible of the kernel text.

static_call selftest is running successfully with this change.

With this patch, __do_irq() has the following sequence to trace
irq entries:

c0004a00 <__SCT__tp_func_irq_entry>:
c0004a00: 48 00 00 e0  b       c0004ae0 <__traceiter_irq_entry>
c0004a04: 3d 80 c0 00  lis     r12,-16384
c0004a08: 81 8c 4a 1c  lwz     r12,18972(r12)
c0004a0c: 7d 89 03 a6  mtctr   r12
c0004a10: 4e 80 04 20  bctr
c0004a14: 38 60 00 00  li      r3,0
c0004a18: 4e 80 00 20  blr
c0004a1c: 00 00 00 00  .long 0x0
...
c0005654 <__do_irq>:
...
c0005664: 7c 7f 1b 78  mr      r31,r3
...
c00056a0: 81 22 00 00  lwz     r9,0(r2)
c00056a4: 39 29 00 01  addi    r9,r9,1
c00056a8: 91 22 00 00  stw     r9,0(r2)
c00056ac: 3d 20 c0 af  lis     r9,-16209
c00056b0: 81 29 74 cc  lwz     r9,29900(r9)
c00056b4: 2c 09 00 00  cmpwi   r9,0
c00056b8: 41 82 00 10  beq     c00056c8 <__do_irq+0x74>
c00056bc: 80 69 00 04  lwz     r3,4(r9)
c00056c0: 7f e4 fb 78  mr      r4,r31
c00056c4: 4b ff f3 3d  bl      c0004a00 <__SCT__tp_func_irq_entry>

Before this patch, __do_irq() was doing the following to trace irq
entries:

c0005700 <__do_irq>:
...
c0005710: 7c 7e 1b 78  mr      r30,r3
...
c000574c: 93 e1 00 0c  stw     r31,12(r1)
c0005750: 81 22 00 00  lwz     r9,0(r2)
c0005754: 39 29 00 01  addi    r9,r9,1
c0005758: 91 22 00 00  stw     r9,0(r2)
c000575c: 3d 20 c0 af  lis     r9,-16209
c0005760: 83 e9 f4 cc  lwz     r31,-2868(r9)
c0005764: 2c 1f 00 00  cmpwi   r31,0
c0005768: 41 82 00 24  beq     c000578c <__do_irq+0x8c>
c000576c: 81 3f 00 00  lwz     r9,0(r31)
c0005770: 80 7f 00 04  lwz     r3,4(r31)
c0005774: 7d 29 03 a6  mtctr   r9
c0005778: 7f c4 f3 78  mr      r4,r30
c000577c: 4e 80 04 21  bctrl
c0005780: 85 3f 00 0c  lwzu    r9,12(r31)
c0005784: 2c 09 00 00  cmpwi   r9,0
c0005788: 40 82 ff e4  bne     c000576c <__do_irq+0x6c>

Behind the fact of now using a direct 'bl' instead of a
'load/mtctr/bctr' sequence, we can also see that we get one less
register on the stack.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/6ec2a7865ed6a5ec54ab46d026785bafe1d837ea.1630484892.git.christophe.leroy@csgroup.eu
2 years agopowerpc/machdep: Remove stale functions from ppc_md structure
Christophe Leroy [Tue, 31 Aug 2021 08:30:24 +0000 (08:30 +0000)]
powerpc/machdep: Remove stale functions from ppc_md structure

ppc_md.iommu_save() is not set anymore by any platform after
commit c40785ad305b ("powerpc/dart: Use a cachable DART").
So iommu_save() has become a nop and can be removed.

ppc_md.show_percpuinfo() is not set anymore by any platform after
commit 4350147a816b ("[PATCH] ppc64: SMU based macs cpufreq support").

Last users of ppc_md.rtc_read_val() and ppc_md.rtc_write_val() were
removed by commit 0f03a43b8f0f ("[POWERPC] Remove todc code from
ARCH=powerpc")

Last user of kgdb_map_scc() was removed by commit 17ce452f7ea3 ("kgdb,
powerpc: arch specific powerpc kgdb support").

ppc.machine_kexec_prepare() has not been used since
commit 8ee3e0d69623 ("powerpc: Remove the main legacy iSerie platform
code"). This allows the removal of machine_kexec_prepare() and the
rename of default_machine_kexec_prepare() into machine_kexec_prepare()

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Daniel Axtens <dja@axtens.net>
[mpe: Drop prototype for default_machine_kexec_prepare() as noted by dja]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/24d4ca0ada683c9436a5f812a7aeb0a1362afa2b.1630398606.git.christophe.leroy@csgroup.eu
2 years agopowerpc/time: Remove generic_suspend_{dis/en}able_irqs()
Christophe Leroy [Tue, 31 Aug 2021 08:29:35 +0000 (08:29 +0000)]
powerpc/time: Remove generic_suspend_{dis/en}able_irqs()

Commit d75d68cfef49 ("powerpc: Clean up obsolete code relating to
decrementer and timebase") made generic_suspend_enable_irqs() and
generic_suspend_disable_irqs() static.

Fold them into their only caller.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c3f9ec9950394ef939014f7934268e6ee30ca04f.1630398566.git.christophe.leroy@csgroup.eu
2 years agopowerpc/audit: Convert powerpc to AUDIT_ARCH_COMPAT_GENERIC
Christophe Leroy [Tue, 24 Aug 2021 13:36:13 +0000 (13:36 +0000)]
powerpc/audit: Convert powerpc to AUDIT_ARCH_COMPAT_GENERIC

Commit e65e1fc2d24b ("[PATCH] syscall class hookup for all normal
targets") added generic support for AUDIT but that didn't include
support for bi-arch like powerpc.

Commit 4b58841149dc ("audit: Add generic compat syscall support")
added generic support for bi-arch.

Convert powerpc to that bi-arch generic audit support.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/a4b3951d1191d4183d92a07a6097566bde60d00a.1629812058.git.christophe.leroy@csgroup.eu
2 years agopowerpc/32: Don't use lmw/stmw for saving/restoring non volatile regs
Christophe Leroy [Mon, 23 Aug 2021 15:29:12 +0000 (15:29 +0000)]
powerpc/32: Don't use lmw/stmw for saving/restoring non volatile regs

Instructions lmw/stmw are interesting for functions that are rarely
used and not in the cache, because only one instruction is to be
copied into the instruction cache instead of 19. However those
instruction are less performant than 19x raw lwz/stw as they require
synchronisation plus one additional cycle.

SAVE_NVGPRS / REST_NVGPRS are used in only a few places which are
mostly in interrupts entries/exits and in task switch so they are
likely already in the cache.

Using standard lwz improves null_syscall selftest by:
- 10 cycles on mpc832x.
- 2 cycles on mpc8xx.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/316c543b8906712c108985c8463eec09c8db577b.1629732542.git.christophe.leroy@csgroup.eu
2 years agopowerpc/5200: dts: fix memory node unit name
Anatolij Gustschin [Wed, 13 Oct 2021 22:05:31 +0000 (00:05 +0200)]
powerpc/5200: dts: fix memory node unit name

Fixes build warnings:
Warning (unit_address_vs_reg): /memory: node has a reg or ranges property, but no unit name

Signed-off-by: Anatolij Gustschin <agust@denx.de>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211013220532.24759-4-agust@denx.de
2 years agopowerpc/5200: dts: fix pci ranges warnings
Anatolij Gustschin [Wed, 13 Oct 2021 22:05:30 +0000 (00:05 +0200)]
powerpc/5200: dts: fix pci ranges warnings

Fix ranges property warnings:
pci@f0000d00:ranges: 'oneOf' conditional failed, one must be fixed:

Signed-off-by: Anatolij Gustschin <agust@denx.de>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211013220532.24759-3-agust@denx.de
2 years agopowerpc/5200: dts: add missing pci ranges
Anatolij Gustschin [Wed, 13 Oct 2021 22:05:29 +0000 (00:05 +0200)]
powerpc/5200: dts: add missing pci ranges

Add ranges property to fix build warnings:
Warning (pci_bridge): /pci@f0000d00: missing ranges for PCI bridge (or not a bridge)

Signed-off-by: Anatolij Gustschin <agust@denx.de>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211013220532.24759-2-agust@denx.de
2 years agopowerpc/vas: Fix potential NULL pointer dereference
Gustavo A. R. Silva [Fri, 15 Oct 2021 05:03:45 +0000 (00:03 -0500)]
powerpc/vas: Fix potential NULL pointer dereference

(!ptr && !ptr->foo) strikes again. :)

The expression (!ptr && !ptr->foo) is bogus and in case ptr is NULL,
it leads to a NULL pointer dereference: ptr->foo.

Fix this by converting && to ||

This issue was detected with the help of Coccinelle, and audited and
fixed manually.

Fixes: 1a0d0d5ed5e3 ("powerpc/vas: Add platform specific user window operations")
Cc: stable@vger.kernel.org
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Reviewed-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211015050345.GA1161918@embeddedor
2 years agopowerpc/fsl_booke: Enable STRICT_KERNEL_RWX
Christophe Leroy [Fri, 15 Oct 2021 10:02:49 +0000 (12:02 +0200)]
powerpc/fsl_booke: Enable STRICT_KERNEL_RWX

Enable STRICT_KERNEL_RWX on fsl_booke.

For that, we need additional TLBCAMs dedicated to linear mapping,
based on the alignment of _sinittext.

By default, up to 768 Mbytes of memory are mapped.
It uses 3 TLBCAMs of size 256 Mbytes.

With a data alignment of 16, we need up to 9 TLBCAMs:
  16/16/16/16/64/64/64/256/256

With a data alignment of 4, we need up to 12 TLBCAMs:
  4/4/4/4/16/16/16/64/64/64/256/256

With a data alignment of 1, we need up to 15 TLBCAMs:
  1/1/1/1/4/4/4/16/16/16/64/64/64/256/256

By default, set a 16 Mbytes alignment as a compromise between memory
usage and number of TLBCAMs. This can be adjusted manually when needed.

For the time being, it doens't work when the base is randomised.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/29f9e5d2bbbc83ae9ca879265426a6278bf4d5bb.1634292136.git.christophe.leroy@csgroup.eu
2 years agopowerpc/fsl_booke: Update of TLBCAMs after init
Christophe Leroy [Fri, 15 Oct 2021 10:02:48 +0000 (12:02 +0200)]
powerpc/fsl_booke: Update of TLBCAMs after init

After init, set readonly memory as ROX and set readwrite
memory as RWX, if STRICT_KERNEL_RWX is enabled.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/66bef0b9c273e1121706883f3cf5ad0a053d863f.1634292136.git.christophe.leroy@csgroup.eu
2 years agopowerpc/fsl_booke: Allocate separate TLBCAMs for readonly memory
Christophe Leroy [Fri, 15 Oct 2021 10:02:47 +0000 (12:02 +0200)]
powerpc/fsl_booke: Allocate separate TLBCAMs for readonly memory

Reorganise TLBCAM allocation so that when STRICT_KERNEL_RWX is
enabled, TLBCAMs are allocated such that readonly memory uses
different TLBCAMs.

This results in an allocation looking like:

Memory CAM mapping: 4/4/4/1/1/1/1/16/16/16/64/64/64/256/256 Mb, residual: 256Mb

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/8ca169bc288261a0e0558712f979023c3a960ebb.1634292136.git.christophe.leroy@csgroup.eu
2 years agopowerpc/fsl_booke: Tell map_mem_in_cams() if init is done
Christophe Leroy [Fri, 15 Oct 2021 10:02:46 +0000 (12:02 +0200)]
powerpc/fsl_booke: Tell map_mem_in_cams() if init is done

In order to be able to call map_mem_in_cams() once more
after init for STRICT_KERNEL_RWX, add an argument.

For now, map_mem_in_cams() is always called only during init.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/3b69a7e0b393b16984ade882a5eae5d727717459.1634292136.git.christophe.leroy@csgroup.eu
2 years agopowerpc/fsl_booke: Enable reloading of TLBCAM without switching to AS1
Christophe Leroy [Fri, 15 Oct 2021 10:02:45 +0000 (12:02 +0200)]
powerpc/fsl_booke: Enable reloading of TLBCAM without switching to AS1

Avoid switching to AS1 when reloading TLBCAM after init for
STRICT_KERNEL_RWX.

When we setup AS1 we expect the entire accessible memory to be mapped
through one entry, this is not the case anymore at the end of init.

We are not changing the size of TLBCAMs, only flags, so no need to
switch to AS1.

So change loadcam_multi() to not switch to AS1 when the given
temporary tlb entry in 0.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/a9d517fbfbc940f56103c46b323f6eb8f4485571.1634292136.git.christophe.leroy@csgroup.eu
2 years agopowerpc/fsl_booke: Take exec flag into account when setting TLBCAMs
Christophe Leroy [Fri, 15 Oct 2021 10:02:44 +0000 (12:02 +0200)]
powerpc/fsl_booke: Take exec flag into account when setting TLBCAMs

Don't force MAS3_SX and MAS3_UX at all time. Take into account the
exec flag.

While at it, fix a couple of closeby style problems (indent with space
and unnecessary parenthesis), it keeps more 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/5467044e59f27f9fcf709b9661779e3ce5f784f6.1634292136.git.christophe.leroy@csgroup.eu
2 years agopowerpc/fsl_booke: Rename fsl_booke.c to fsl_book3e.c
Christophe Leroy [Fri, 15 Oct 2021 10:02:43 +0000 (12:02 +0200)]
powerpc/fsl_booke: Rename fsl_booke.c to fsl_book3e.c

We have a myriad of CONFIG symbols around different variants
of BOOKEs, which would be worth tidying up one day.

But at least, make file names and CONFIG option match:

We have CONFIG_FSL_BOOKE and CONFIG_PPC_FSL_BOOK3E.

fsl_booke.c is selected by and only by CONFIG_PPC_FSL_BOOK3E.
So rename it fsl_book3e to reduce 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/5dc871db1f67739319bec11f049ca450da1c13a2.1634292136.git.christophe.leroy@csgroup.eu
2 years agopowerpc/booke: Disable STRICT_KERNEL_RWX, DEBUG_PAGEALLOC and KFENCE
Christophe Leroy [Fri, 15 Oct 2021 10:02:42 +0000 (12:02 +0200)]
powerpc/booke: Disable STRICT_KERNEL_RWX, DEBUG_PAGEALLOC and KFENCE

fsl_booke and 44x are not able to map kernel linear memory with
pages, so they can't support DEBUG_PAGEALLOC and KFENCE, and
STRICT_KERNEL_RWX is also a problem for now.

Enable those only on book3s (both 32 and 64 except KFENCE), 8xx and 40x.

Fixes: 88df6e90fa97 ("[POWERPC] DEBUG_PAGEALLOC for 32-bit")
Fixes: 95902e6c8864 ("powerpc/mm: Implement STRICT_KERNEL_RWX on PPC32")
Fixes: 90cbac0e995d ("powerpc: Enable KFENCE for PPC32")
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d1ad9fdd9b27da3fdfa16510bb542ed51fa6e134.1634292136.git.christophe.leroy@csgroup.eu
2 years agopowerpc/kexec_file: Add of_node_put() before goto
Wan Jiabing [Mon, 18 Oct 2021 01:54:16 +0000 (21:54 -0400)]
powerpc/kexec_file: Add of_node_put() before goto

Fix following coccicheck warning:
./arch/powerpc/kexec/file_load_64.c:698:1-22: WARNING: Function
for_each_node_by_type should have of_node_put() before goto

Early exits from for_each_node_by_type should decrement the
node reference counter.

Signed-off-by: Wan Jiabing <wanjiabing@vivo.com>
Acked-by: Hari Bathini <hbathini@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211018015418.10182-1-wanjiabing@vivo.com
2 years agopowerpc/pseries/iommu: Add of_node_put() before break
Wan Jiabing [Thu, 14 Oct 2021 07:56:04 +0000 (03:56 -0400)]
powerpc/pseries/iommu: Add of_node_put() before break

Fix following coccicheck warning:

./arch/powerpc/platforms/pseries/iommu.c:924:1-28: WARNING: Function
for_each_node_with_property should have of_node_put() before break

Early exits from for_each_node_with_property should decrement the
node reference counter.

Signed-off-by: Wan Jiabing <wanjiabing@vivo.com>
Reviewed-by: Leonardo Bras <leobras.c@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211014075624.16344-1-wanjiabing@vivo.com
2 years agopowerpc/s64: Clarify that radix lacks DEBUG_PAGEALLOC
Joel Stanley [Wed, 13 Oct 2021 21:34:38 +0000 (08:04 +1030)]
powerpc/s64: Clarify that radix lacks DEBUG_PAGEALLOC

The page_alloc.c code will call into __kernel_map_pages() when
DEBUG_PAGEALLOC is configured and enabled.

As the implementation assumes hash, this should crash spectacularly if
not for a bit of luck in __kernel_map_pages(). In this function
linear_map_hash_count is always zero, the for loop exits without doing
any damage.

There are no other platforms that determine if they support
debug_pagealloc at runtime. Instead of adding code to mm/page_alloc.c to
do that, this change turns the map/unmap into a noop when in radix
mode and prints a warning once.

Signed-off-by: Joel Stanley <joel@jms.id.au>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Reformat if per Christophe's suggestion]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211013213438.675095-1-joel@jms.id.au
2 years agopowerpc: Mark .opd section read-only
Christophe Leroy [Wed, 13 Oct 2021 14:43:54 +0000 (16:43 +0200)]
powerpc: Mark .opd section read-only

.opd section contains function descriptors used to locate
functions in the kernel. If someone is able to modify a
function descriptor he will be able to run arbitrary
kernel function instead of another.

To avoid that, move .opd section inside read-only memory.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/3cd40b682fb6f75bb40947b55ca0bac20cb3f995.1634136222.git.christophe.leroy@csgroup.eu
2 years agopowerpc/perf: Fix cycles/instructions as PM_CYC/PM_INST_CMPL in power10
Athira Rajeev [Thu, 7 Oct 2021 07:51:21 +0000 (13:21 +0530)]
powerpc/perf: Fix cycles/instructions as PM_CYC/PM_INST_CMPL in power10

On power9 and earlier platforms, the default event used for cyles and
instructions is PM_CYC (0x0001e) and PM_INST_CMPL (0x00002)
respectively. These events use two programmable PMCs and by default will
count irrespective of the run latch state (idle state). But since they
use programmable PMCs, these events can lead to multiplexing with other
events, because there are only 4 programmable PMCs. Hence in power10,
performance monitoring unit (PMU) driver uses performance monitor
counter 5 (PMC5) and performance monitor counter6 (PMC6) for counting
instructions and cycles.

Currently on power10, the event used for cycles is PM_RUN_CYC (0x600F4)
and instructions uses PM_RUN_INST_CMPL (0x500fa). But counting of these
events in idle state is controlled by the CC56RUN bit setting in Monitor
Mode Control Register0 (MMCR0). If the CC56RUN bit is zero, PMC5/6 will
not count when CTRL[RUN] (run latch) is zero. This could lead to missing
some counts if a thread is in idle state during system wide profiling.

To fix it, set the CC56RUN bit in MMCR0 for power10, which makes PMC5
and PMC6 count instructions and cycles regardless of the run latch
state. Since this change make PMC5/6 count as PM_INST_CMPL/PM_CYC,
rename the event code 0x600f4 as PM_CYC instead of PM_RUN_CYC and event
code 0x500fa as PM_INST_CMPL instead of PM_RUN_INST_CMPL. The changes
are only for PMC5/6 event codes and will not affect the behaviour of
PM_RUN_CYC/PM_RUN_INST_CMPL if progammed in other PMC's.

Fixes: a64e697cef23 ("powerpc/perf: power10 Performance Monitoring support")
Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.cm>
Reviewed-by: Madhavan Srinivasan <maddy@linux.ibm.com>
[mpe: Tweak change log wording for style and consistency]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211007075121.28497-1-atrajeev@linux.vnet.ibm.com
2 years agopowerpc/eeh: Fix docstrings in eeh.c
Kai Song [Sat, 9 Oct 2021 04:16:30 +0000 (12:16 +0800)]
powerpc/eeh: Fix docstrings in eeh.c

We fix the following warnings when building kernel with W=1:
arch/powerpc/kernel/eeh.c:598: warning: Function parameter or member 'function' not described in 'eeh_pci_enable'
arch/powerpc/kernel/eeh.c:774: warning: Function parameter or member 'edev' not described in 'eeh_set_dev_freset'
arch/powerpc/kernel/eeh.c:774: warning: expecting prototype for eeh_set_pe_freset(). Prototype was for eeh_set_dev_freset() instead
arch/powerpc/kernel/eeh.c:814: warning: Function parameter or member 'include_passed' not described in 'eeh_pe_reset_full'
arch/powerpc/kernel/eeh.c:944: warning: Function parameter or member 'ops' not described in 'eeh_init'
arch/powerpc/kernel/eeh.c:1451: warning: Function parameter or member 'include_passed' not described in 'eeh_pe_reset'
arch/powerpc/kernel/eeh.c:1526: warning: Function parameter or member 'func' not described in 'eeh_pe_inject_err'
arch/powerpc/kernel/eeh.c:1526: warning: Excess function parameter 'function' described in 'eeh_pe_inject_err'

Signed-off-by: Kai Song <songkai01@inspur.com>
Reviewed-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211009041630.4135-1-songkai01@inspur.com
2 years agopowerpc/boot: Use CONFIG_PPC_POWERNV to compile OPAL support
Cédric Le Goater [Mon, 11 Oct 2021 07:03:56 +0000 (09:03 +0200)]
powerpc/boot: Use CONFIG_PPC_POWERNV to compile OPAL support

CONFIG_PPC64_BOOT_WRAPPER is selected by CPU_LITTLE_ENDIAN which is
used to compile support for other platforms such as Microwatt. There
is no need for OPAL calls on these.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211011070356.99952-1-clg@kaod.org
2 years agopowerpc: Set max_mapnr correctly
Christophe Leroy [Tue, 12 Oct 2021 10:40:37 +0000 (12:40 +0200)]
powerpc: Set max_mapnr correctly

max_mapnr is used by virt_addr_valid() to check if a linear
address is valid.

It must only include lowmem PFNs, like other architectures.

Problem detected on a system with 1G mem (Only 768M are mapped), with
CONFIG_DEBUG_VIRTUAL and CONFIG_TEST_DEBUG_VIRTUAL, it didn't report
virt_to_phys(VMALLOC_START), VMALLOC_START being 0xf1000000.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/77d99037782ac4b3c3b0124fc4ae80ce7b760b05.1634035228.git.christophe.leroy@csgroup.eu
2 years agopowerpc/perf: Expose instruction and data address registers as part of extended regs
Athira Rajeev [Thu, 7 Oct 2021 06:55:04 +0000 (12:25 +0530)]
powerpc/perf: Expose instruction and data address registers as part of extended regs

Patch adds support to include Sampled Instruction Address Register
(SIAR) and Sampled Data Address Register (SDAR) SPRs as part of extended
registers. Update the definition of PERF_REG_PMU_MASK_300/31 and
PERF_REG_EXTENDED_MAX to include these SPR's.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Reviewed-by: Daniel Axtens <dja@axtens.net>
Reviewed-by: Kajol Jain<kjain@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211007065505.27809-4-atrajeev@linux.vnet.ibm.com
2 years agopowerpc/perf: Refactor the code definition of perf reg extended mask
Athira Rajeev [Thu, 7 Oct 2021 06:55:02 +0000 (12:25 +0530)]
powerpc/perf: Refactor the code definition of perf reg extended mask

PERF_REG_PMU_MASK_300 and PERF_REG_PMU_MASK_31 defines the mask
value for extended registers. Current definition of these mask values
uses hex constant and does not use registers by name, making it less
readable. Patch refactor the macro values by or'ing together the actual
register value constants. Also include PERF_REG_EXTENDED_MAX as
part of enum definition.

Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Reviewed-by: Kajol Jain<kjain@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211007065505.27809-2-atrajeev@linux.vnet.ibm.com
2 years agopowerpc/pseries/cpuhp: remove obsolete comment from pseries_cpu_die
Nathan Lynch [Mon, 27 Sep 2021 20:19:33 +0000 (15:19 -0500)]
powerpc/pseries/cpuhp: remove obsolete comment from pseries_cpu_die

This comment likely refers to the obsolete DLPAR workflow where some
resource state transitions were driven more directly from user space
utilities, but it also seems to contradict itself: "Change isolate state to
Isolate [...]" is at odds with the preceding sentences, and it does not
relate at all to the code that follows.

Remove it to prevent confusion.

Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210927201933.76786-5-nathanl@linux.ibm.com
2 years agopowerpc/pseries/cpuhp: delete add/remove_by_count code
Nathan Lynch [Mon, 27 Sep 2021 20:19:32 +0000 (15:19 -0500)]
powerpc/pseries/cpuhp: delete add/remove_by_count code

The core DLPAR code supports two actions (add and remove) and three
subtypes of action:

* By DRC index: the action is attempted on a single specified resource.
  This is the usual case for processors.
* By indexed count: the action is attempted on a range of resources
  beginning at the specified index. This is implemented only by the memory
  DLPAR code.
* By count: the lower layer (CPU or memory) is responsible for locating the
  specified number of resources to which the action can be applied.

I cannot find any evidence of the "by count" subtype being used by drmgr or
qemu for processors. And when I try to exercise this code, the add case
does not work:

  $ ppc64_cpu --smt ; nproc
  SMT=8
  24
  $ printf "cpu remove count 2" > /sys/kernel/dlpar
  $ nproc
  8
  $ printf "cpu add count 2" > /sys/kernel/dlpar
  -bash: printf: write error: Invalid argument
  $ dmesg | tail -2
  pseries-hotplug-cpu: Failed to find enough CPUs (1 of 2) to add
  dlpar: Could not handle DLPAR request "cpu add count 2"
  $ nproc
  8
  $ drmgr -c cpu -a -q 2         # this uses the by-index method
  Validating CPU DLPAR capability...yes.
  CPU 1
  CPU 17
  $ nproc
  24

This is because find_drc_info_cpus_to_add() does not increment drc_index
appropriately during its search.

This is not hard to fix. But the _by_count() functions also have the
property that they attempt to roll back all prior operations if the entire
request cannot be satisfied, even though the rollback itself can encounter
errors. It's not possible to provide transaction-like behavior at this
level, and it's undesirable to have code that can only pretend to do that.
Any users of these functions cannot know what the state of the system is in
the error case. And the error paths are, to my knowledge, impossible to
test without adding custom error injection code.

Summary:

* This code has not worked reliably since its introduction.
* There is no evidence that it is used.
* It contains questionable rollback behaviors in error paths which are
  difficult to test.

So let's remove it.

Fixes: ac71380071d1 ("powerpc/pseries: Add CPU dlpar remove functionality")
Fixes: 90edf184b9b7 ("powerpc/pseries: Add CPU dlpar add functionality")
Fixes: b015f6bc9547 ("powerpc/pseries: Add cpu DLPAR support for drc-info property")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210927201933.76786-4-nathanl@linux.ibm.com
2 years agopowerpc/cpuhp: BUG -> WARN conversion in offline path
Nathan Lynch [Mon, 27 Sep 2021 20:19:31 +0000 (15:19 -0500)]
powerpc/cpuhp: BUG -> WARN conversion in offline path

If, due to bugs elsewhere, we get into unregister_cpu_online() with a CPU
that isn't marked hotpluggable, we can emit a warning and return an
appropriate error instead of crashing.

Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210927201933.76786-3-nathanl@linux.ibm.com
2 years agopowerpc/pseries/cpuhp: cache node corrections
Nathan Lynch [Mon, 27 Sep 2021 20:19:30 +0000 (15:19 -0500)]
powerpc/pseries/cpuhp: cache node corrections

On pseries, cache nodes in the device tree can be added and removed by the
CPU DLPAR code as well as the partition migration (mobility) code. PowerVM
partitions in dedicated processor mode typically have L2 and L3 cache
nodes.

The CPU DLPAR code has the following shortcomings:

* Cache nodes returned as siblings of a new CPU node by
  ibm,configure-connector are silently discarded; only the CPU node is
  added to the device tree.

* Cache nodes which become unreferenced in the processor removal path are
  not removed from the device tree. This can lead to duplicate nodes when
  the post-migration device tree update code replaces cache nodes.

This is long-standing behavior. Presumably it has gone mostly unnoticed
because the two bugs have the property of obscuring each other in common
simple scenarios (e.g. remove a CPU and add it back). Likely you'd notice
only if you cared to inspect the device tree or the sysfs cacheinfo
information.

Booted with two processors:

  $ pwd
  /sys/firmware/devicetree/base/cpus
  $ ls -1d */
  l2-cache@2010/
  l2-cache@2011/
  l3-cache@3110/
  l3-cache@3111/
  PowerPC,POWER9@0/
  PowerPC,POWER9@8/
  $ lsprop */l2-cache
  l2-cache@2010/l2-cache
                 00003110 (12560)
  l2-cache@2011/l2-cache
                 00003111 (12561)
  PowerPC,POWER9@0/l2-cache
                 00002010 (8208)
  PowerPC,POWER9@8/l2-cache
                 00002011 (8209)
  $ ls /sys/devices/system/cpu/cpu0/cache/
  index0  index1  index2  index3

After DLPAR-adding PowerPC,POWER9@10, we see that its associated cache
nodes are absent, its threads' L2+L3 cacheinfo is unpopulated, and it is
missing a cache level in its sched domain hierarchy:

  $ ls -1d */
  l2-cache@2010/
  l2-cache@2011/
  l3-cache@3110/
  l3-cache@3111/
  PowerPC,POWER9@0/
  PowerPC,POWER9@10/
  PowerPC,POWER9@8/
  $ lsprop PowerPC\,POWER9@10/l2-cache
  PowerPC,POWER9@10/l2-cache
                 00002012 (8210)
  $ ls /sys/devices/system/cpu/cpu16/cache/
  index0  index1
  $ grep . /sys/kernel/debug/sched/domains/cpu{0,8,16}/domain*/name
  /sys/kernel/debug/sched/domains/cpu0/domain0/name:SMT
  /sys/kernel/debug/sched/domains/cpu0/domain1/name:CACHE
  /sys/kernel/debug/sched/domains/cpu0/domain2/name:DIE
  /sys/kernel/debug/sched/domains/cpu8/domain0/name:SMT
  /sys/kernel/debug/sched/domains/cpu8/domain1/name:CACHE
  /sys/kernel/debug/sched/domains/cpu8/domain2/name:DIE
  /sys/kernel/debug/sched/domains/cpu16/domain0/name:SMT
  /sys/kernel/debug/sched/domains/cpu16/domain1/name:DIE

When removing PowerPC,POWER9@8, we see that its cache nodes are left
behind:

  $ ls -1d */
  l2-cache@2010/
  l2-cache@2011/
  l3-cache@3110/
  l3-cache@3111/
  PowerPC,POWER9@0/

When DLPAR is combined with VM migration, we can get duplicate nodes. E.g.
removing one processor, then migrating, adding a processor, and then
migrating again can result in warnings from the OF core during
post-migration device tree updates:

  Duplicate name in cpus, renamed to "l2-cache@2011#1"
  Duplicate name in cpus, renamed to "l3-cache@3111#1"

and nodes with duplicated phandles in the tree, making lookup behavior
unpredictable:

  $ lsprop l[23]-cache@*/ibm,phandle
  l2-cache@2010/ibm,phandle
                   00002010 (8208)
  l2-cache@2011#1/ibm,phandle
                   00002011 (8209)
  l2-cache@2011/ibm,phandle
                   00002011 (8209)
  l3-cache@3110/ibm,phandle
                   00003110 (12560)
  l3-cache@3111#1/ibm,phandle
                   00003111 (12561)
  l3-cache@3111/ibm,phandle
                   00003111 (12561)

Address these issues by:

* Correctly processing siblings of the node returned from
  dlpar_configure_connector().
* Removing cache nodes in the CPU remove path when it can be determined
  that they are not associated with other CPUs or caches.

Use the of_changeset API in both cases, which allows us to keep the error
handling in this code from becoming more complex while ensuring that the
device tree cannot become inconsistent.

Fixes: ac71380071d1 ("powerpc/pseries: Add CPU dlpar remove functionality")
Fixes: 90edf184b9b7 ("powerpc/pseries: Add CPU dlpar add functionality")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210927201933.76786-2-nathanl@linux.ibm.com
2 years agopowerpc/paravirt: correct preempt debug splat in vcpu_is_preempted()
Nathan Lynch [Tue, 28 Sep 2021 21:41:47 +0000 (16:41 -0500)]
powerpc/paravirt: correct preempt debug splat in vcpu_is_preempted()

vcpu_is_preempted() can be used outside of preempt-disabled critical
sections, yielding warnings such as:

BUG: using smp_processor_id() in preemptible [00000000] code: systemd-udevd/185
caller is rwsem_spin_on_owner+0x1cc/0x2d0
CPU: 1 PID: 185 Comm: systemd-udevd Not tainted 5.15.0-rc2+ #33
Call Trace:
[c000000012907ac0] [c000000000aa30a8] dump_stack_lvl+0xac/0x108 (unreliable)
[c000000012907b00] [c000000001371f70] check_preemption_disabled+0x150/0x160
[c000000012907b90] [c0000000001e0e8c] rwsem_spin_on_owner+0x1cc/0x2d0
[c000000012907be0] [c0000000001e1408] rwsem_down_write_slowpath+0x478/0x9a0
[c000000012907ca0] [c000000000576cf4] filename_create+0x94/0x1e0
[c000000012907d10] [c00000000057ac08] do_symlinkat+0x68/0x1a0
[c000000012907d70] [c00000000057ae18] sys_symlink+0x58/0x70
[c000000012907da0] [c00000000002e448] system_call_exception+0x198/0x3c0
[c000000012907e10] [c00000000000c54c] system_call_common+0xec/0x250

The result of vcpu_is_preempted() is always used speculatively, and the
function does not access per-cpu resources in a (Linux) preempt-unsafe way.
Use raw_smp_processor_id() to avoid such warnings, adding explanatory
comments.

Fixes: ca3f969dcb11 ("powerpc/paravirt: Use is_kvm_guest() in vcpu_is_preempted()")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210928214147.312412-3-nathanl@linux.ibm.com
2 years agopowerpc/paravirt: vcpu_is_preempted() commentary
Nathan Lynch [Tue, 28 Sep 2021 21:41:46 +0000 (16:41 -0500)]
powerpc/paravirt: vcpu_is_preempted() commentary

Add comments more clearly documenting that this function determines whether
hypervisor-level preemption of the VM has occurred.

Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210928214147.312412-2-nathanl@linux.ibm.com
2 years agopowerpc: fix unbalanced node refcount in check_kvm_guest()
Nathan Lynch [Tue, 28 Sep 2021 12:45:50 +0000 (07:45 -0500)]
powerpc: fix unbalanced node refcount in check_kvm_guest()

When check_kvm_guest() succeeds in looking up a /hypervisor OF node, it
returns without performing a matching put for the lookup, leaving the
node's reference count elevated.

Add the necessary call to of_node_put(), rearranging the code slightly to
avoid repetition or goto.

Fixes: 107c55005fbd ("powerpc/pseries: Add KVM guest doorbell restrictions")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Reviewed-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210928124550.132020-1-nathanl@linux.ibm.com
2 years agovideo: fbdev: chipsfb: use memset_io() instead of memset()
Christophe Leroy [Wed, 15 Sep 2021 13:34:35 +0000 (15:34 +0200)]
video: fbdev: chipsfb: use memset_io() instead of memset()

While investigating a lockup at startup on Powerbook 3400C, it was
identified that the fbdev driver generates alignment exception at
startup:

  --- interrupt: 600 at memset+0x60/0xc0
  NIP:  c0021414 LR: c03fc49c CTR: 00007fff
  REGS: ca021c10 TRAP: 0600   Tainted: G        W          (5.14.2-pmac-00727-g12a41fa69492)
  MSR:  00009032 <EE,ME,IR,DR,RI>  CR: 44008442  XER: 20000100
  DAR: cab80020 DSISR: 00017c07
  GPR00: 00000007 ca021cd0 c14412e0 cab80000 00000000 00100000 cab8001c 00000004
  GPR08: 00100000 00007fff 00000000 00000000 84008442 00000000 c0006fb4 00000000
  GPR16: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00100000
  GPR24: 00000000 81800000 00000320 c15fa400 c14d1878 00000000 c14d1800 c094e19c
  NIP [c0021414] memset+0x60/0xc0
  LR [c03fc49c] chipsfb_pci_init+0x160/0x580
  --- interrupt: 600
  [ca021cd0] [c03fc46c] chipsfb_pci_init+0x130/0x580 (unreliable)
  [ca021d20] [c03a3a70] pci_device_probe+0xf8/0x1b8
  [ca021d50] [c043d584] really_probe.part.0+0xac/0x388
  [ca021d70] [c043d914] __driver_probe_device+0xb4/0x170
  [ca021d90] [c043da18] driver_probe_device+0x48/0x144
  [ca021dc0] [c043e318] __driver_attach+0x11c/0x1c4
  [ca021de0] [c043ad30] bus_for_each_dev+0x88/0xf0
  [ca021e10] [c043c724] bus_add_driver+0x190/0x22c
  [ca021e40] [c043ee94] driver_register+0x9c/0x170
  [ca021e60] [c0006c28] do_one_initcall+0x54/0x1ec
  [ca021ed0] [c08246e4] kernel_init_freeable+0x1c0/0x270
  [ca021f10] [c0006fdc] kernel_init+0x28/0x11c
  [ca021f30] [c0017148] ret_from_kernel_thread+0x14/0x1c
  Instruction dump:
  7d4601a4 39490777 7d4701a4 39490888 7d4801a4 39490999 7d4901a4 39290aaa
  7d2a01a4 4c00012c 4bfffe88 0fe00000 <4bfffe809421fff0 38210010 48001970

This is due to 'dcbz' instruction being used on non-cached memory.
'dcbz' instruction is used by memset() to zeroize a complete
cacheline at once, and memset() is not expected to be used on non
cached memory.

When performing a 'sparse' check on fbdev driver, it also appears
that the use of memset() is unexpected:

  drivers/video/fbdev/chipsfb.c:334:17: warning: incorrect type in argument 1 (different address spaces)
  drivers/video/fbdev/chipsfb.c:334:17:    expected void *
  drivers/video/fbdev/chipsfb.c:334:17:    got char [noderef] __iomem *screen_base
  drivers/video/fbdev/chipsfb.c:334:15: warning: memset with byte count of 1048576

Use fb_memset() instead of memset(). fb_memset() is defined as
memset_io() for powerpc.

Fixes: 8c8709334cec ("[PATCH] ppc32: Remove CONFIG_PMAC_PBOOK")
Reported-by: Stan Johnson <userm57@yahoo.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/884a54f1e5cb774c1d9b4db780209bee5d4f6718.1631712563.git.christophe.leroy@csgroup.eu
2 years agopowerpc/powernv/dump: Fix typo in comment
Vasant Hegde [Tue, 14 Sep 2021 14:38:02 +0000 (20:08 +0530)]
powerpc/powernv/dump: Fix typo in comment

Signed-off-by: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210914143802.54325-1-hegdevasant@linux.vnet.ibm.com
2 years agopowerpc/asm: Remove UPD_CONSTR after GCC 4.9 removal
Nick Desaulniers [Tue, 14 Sep 2021 16:17:04 +0000 (09:17 -0700)]
powerpc/asm: Remove UPD_CONSTR after GCC 4.9 removal

UPD_CONSTR was previously a preprocessor define for an old GCC 4.9
inline asm bug with m<> constraints.

Fixes: 6563139d90ad ("powerpc: remove GCC version check for UPD_CONSTR")
Suggested-by: Nathan Chancellor <nathan@kernel.org>
Suggested-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
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://lore.kernel.org/r/20210914161712.2463458-1-ndesaulniers@google.com
2 years agopowerpc/mem: Fix arch/powerpc/mm/mem.c:53:12: error: no previous prototype for 'creat...
Christophe Leroy [Mon, 13 Sep 2021 15:17:26 +0000 (17:17 +0200)]
powerpc/mem: Fix arch/powerpc/mm/mem.c:53:12: error: no previous prototype for 'create_section_mapping'

Commit 8e11d62e2e87 ("powerpc/mem: Add back missing header to fix 'no
previous prototype' error") was supposed to fix the problem, but in
the meantime commit a927bd6ba952 ("mm: fix phys_to_target_node() and*
memory_add_physaddr_to_nid() exports") moved create_section_mapping()
prototype from asm/sparsemem.h to asm/mmzone.h

Fixes: 8e11d62e2e87 ("powerpc/mem: Add back missing header to fix 'no previous prototype' error")
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/025754fde3d027904ae9d0191f395890bec93369.1631541649.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Drop superfluous pci_dev_is_added() calls
Niklas Schnelle [Fri, 10 Sep 2021 14:19:40 +0000 (16:19 +0200)]
powerpc: Drop superfluous pci_dev_is_added() calls

On powerpc, pci_dev_is_added() is called as part of SR-IOV fixups
that are done under pcibios_add_device() which in turn is only called in
pci_device_add() whih is called when a PCI device is scanned.

pci_dev_assign_added() is called in pci_bus_add_device() which is only
called after scanning the device. Thus pci_dev_is_added() is always
false and can be dropped.

Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
Reviewed-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Oliver O'Halloran <oohall@gmail.com>
[mpe: Tweak change log slightly to reflect Oliver's comments]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210910141940.2598035-2-schnelle@linux.ibm.com
2 years agopowerpc/powermac: Remove stale declaration of pmac_md
Christophe Leroy [Fri, 3 Sep 2021 08:23:52 +0000 (08:23 +0000)]
powerpc/powermac: Remove stale declaration of pmac_md

pmac_md doesn't exist anymore, remove stall declaration.

Fixes: e8222502ee61 ("[PATCH] powerpc: Kill _machine and hard-coded platform numbers")
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b2e52934e5a500f149e6d94db3cfa0569bc35081.1630657402.git.christophe.leroy@csgroup.eu
2 years agopowerpc: Remove unused prototype for of_show_percpuinfo
Daniel Axtens [Fri, 3 Sep 2021 06:32:46 +0000 (16:32 +1000)]
powerpc: Remove unused prototype for of_show_percpuinfo

commit 6d7f58b04d82 ("[PATCH] powerpc: Some minor cleanups to setup_32.c")
removed of_show_percpuinfo but didn't remove the prototype.

Remove it.

Fixes: 6d7f58b04d82 ("[PATCH] powerpc: Some minor cleanups to setup_32.c")
Signed-off-by: Daniel Axtens <dja@axtens.net>
Reviewed-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210903063246.70691-1-dja@axtens.net
2 years agopowerpc/476: Fix sparse report
Christophe Leroy [Sat, 18 Sep 2021 09:22:32 +0000 (11:22 +0200)]
powerpc/476: Fix sparse report

Fix sparse errors:
  arch/powerpc/platforms/44x/ppc476.c:236:17: warning: cast removes address space '__iomem' of expression
  arch/powerpc/platforms/44x/ppc476.c:241:34: warning: incorrect type in argument 1 (different address spaces)
  arch/powerpc/platforms/44x/ppc476.c:241:34:    expected void const volatile [noderef] __iomem *addr
  arch/powerpc/platforms/44x/ppc476.c:241:34:    got unsigned char [usertype] *
  arch/powerpc/platforms/44x/ppc476.c:243:17: warning: incorrect type in argument 1 (different address spaces)
  arch/powerpc/platforms/44x/ppc476.c:243:17:    expected void volatile [noderef] __iomem *addr
  arch/powerpc/platforms/44x/ppc476.c:243:17:    got unsigned char [usertype] *[assigned] fpga

Mark 'fpga' pointer as __iomem.

Fixes: ab9a4183fddf ("powerpc: Update currituck pci/usb fixup for new board revision")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/aa6055769b92a5d8685b8d0adab99c48a0b0ef4b.1631956926.git.christophe.leroy@csgroup.eu
2 years agopowerpc/85xx: fix timebase sync issue when CONFIG_HOTPLUG_CPU=n
Xiaoming Ni [Wed, 29 Sep 2021 03:36:46 +0000 (11:36 +0800)]
powerpc/85xx: fix timebase sync issue when CONFIG_HOTPLUG_CPU=n

When CONFIG_SMP=y, timebase synchronization is required when the second
kernel is started.

arch/powerpc/kernel/smp.c:
  int __cpu_up(unsigned int cpu, struct task_struct *tidle)
  {
   ...
   if (smp_ops->give_timebase)
   smp_ops->give_timebase();
   ...
  }

  void start_secondary(void *unused)
  {
   ...
   if (smp_ops->take_timebase)
   smp_ops->take_timebase();
   ...
  }

When CONFIG_HOTPLUG_CPU=n and CONFIG_KEXEC_CORE=n,
 smp_85xx_ops.give_timebase is NULL,
 smp_85xx_ops.take_timebase is NULL,
As a result, the timebase is not synchronized.

Timebase  synchronization does not depend on CONFIG_HOTPLUG_CPU.

Fixes: 56f1ba280719 ("powerpc/mpc85xx: refactor the PM operations")
Cc: stable@vger.kernel.org # v4.6+
Signed-off-by: Xiaoming Ni <nixiaoming@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210929033646.39630-3-nixiaoming@huawei.com
2 years agopowerpc/85xx: Fix oops when mpc85xx_smp_guts_ids node cannot be found
Xiaoming Ni [Wed, 29 Sep 2021 03:36:45 +0000 (11:36 +0800)]
powerpc/85xx: Fix oops when mpc85xx_smp_guts_ids node cannot be found

When the field described in mpc85xx_smp_guts_ids[] is not configured in
dtb, the mpc85xx_setup_pmc() does not assign a value to the "guts"
variable. As a result, the oops is triggered when
mpc85xx_freeze_time_base() is executed.

Fixes: 56f1ba280719 ("powerpc/mpc85xx: refactor the PM operations")
Cc: stable@vger.kernel.org # v4.6+
Signed-off-by: Xiaoming Ni <nixiaoming@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210929033646.39630-2-nixiaoming@huawei.com
2 years agopowerps/pseries/dma: Add support for 2M IOMMU page size
Alexey Kardashevskiy [Wed, 6 Oct 2021 04:47:35 +0000 (15:47 +1100)]
powerps/pseries/dma: Add support for 2M IOMMU page size

The upcoming PAPR spec adds a 2M page size, bit 23 right after 16G page
size in the "ibm,query-pe-dma-window" call.

This adds support for the new page size. Since the new page size is out
of sorted order, this changes the loop to not assume that shift[] is
sorted.

This has now been tested and is known to work on a pre-release version
of phyp.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Leonardo Bras <leobras.c@gmail.com>
Reviewed-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211006044735.1114669-1-aik@ozlabs.ru
2 years agoLinux 5.15-rc2
Linus Torvalds [Mon, 20 Sep 2021 00:28:22 +0000 (17:28 -0700)]
Linux 5.15-rc2

2 years agopci_iounmap'2: Electric Boogaloo: try to make sense of it all
Linus Torvalds [Mon, 20 Sep 2021 00:13:35 +0000 (17:13 -0700)]
pci_iounmap'2: Electric Boogaloo: try to make sense of it all

Nathan Chancellor reports that the recent change to pci_iounmap in
commit 9caea0007601 ("parisc: Declare pci_iounmap() parisc version only
when CONFIG_PCI enabled") causes build errors on arm64.

It took me about two hours to convince myself that I think I know what
the logic of that mess of #ifdef's in the <asm-generic/io.h> header file
really aim to do, and rewrite it to be easier to follow.

Famous last words.

Anyway, the code has now been lifted from that grotty header file into
lib/pci_iomap.c, and has fairly extensive comments about what the logic
is.  It also avoids indirecting through another confusing (and badly
named) helper function that has other preprocessor config conditionals.

Let's see what odd architecture did something else strange in this area
to break things.  But my arm64 cross build is clean.

Fixes: 9caea0007601 ("parisc: Declare pci_iounmap() parisc version only when CONFIG_PCI enabled")
Reported-by: Nathan Chancellor <nathan@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Ulrich Teichert <krypton@ulrich-teichert.org>
Cc: James Bottomley <James.Bottomley@hansenpartnership.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoMerge tag 'x86_urgent_for_v5.15_rc2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 19 Sep 2021 20:29:36 +0000 (13:29 -0700)]
Merge tag 'x86_urgent_for_v5.15_rc2' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Borislav Petkov:

 - Prevent a infinite loop in the MCE recovery on return to user space,
   which was caused by a second MCE queueing work for the same page and
   thereby creating a circular work list.

 - Make kern_addr_valid() handle existing PMD entries, which are marked
   not present in the higher level page table, correctly instead of
   blindly dereferencing them.

 - Pass a valid address to sanitize_phys(). This was caused by the
   mixture of inclusive and exclusive ranges. memtype_reserve() expect
   'end' being exclusive, but sanitize_phys() wants it inclusive. This
   worked so far, but with end being the end of the physical address
   space the fail is exposed.

 - Increase the maximum supported GPIO numbers for 64bit. Newer SoCs
   exceed the previous maximum.

* tag 'x86_urgent_for_v5.15_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mce: Avoid infinite loop for copy from user recovery
  x86/mm: Fix kern_addr_valid() to cope with existing but not present entries
  x86/platform: Increase maximum GPIO number for X86_64
  x86/pat: Pass valid address to sanitize_phys()

2 years agoMerge tag 'perf-urgent-2021-09-19' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 19 Sep 2021 20:22:40 +0000 (13:22 -0700)]
Merge tag 'perf-urgent-2021-09-19' of git://git./linux/kernel/git/tip/tip

Pull perf event fix from Thomas Gleixner:
 "A single fix for the perf core where a value read with READ_ONCE() was
  checked and then reread which makes all the checks invalid. Reuse the
  already read value instead"

* tag 'perf-urgent-2021-09-19' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  events: Reuse value read using READ_ONCE instead of re-reading it

2 years agoMerge tag 'locking-urgent-2021-09-19' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 19 Sep 2021 20:11:19 +0000 (13:11 -0700)]
Merge tag 'locking-urgent-2021-09-19' of git://git./linux/kernel/git/tip/tip

Pull locking fixes from Thomas Gleixner:
 "A set of updates for the RT specific reader/writer locking base code:

   - Make the fast path reader ordering guarantees correct.

   - Code reshuffling to make the fix simpler"

[ This plays ugly games with atomic_add_return_release() because we
  don't have a plain atomic_add_release(), and should really be cleaned
  up, I think    - Linus ]

* tag 'locking-urgent-2021-09-19' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/rwbase: Take care of ordering guarantee for fastpath reader
  locking/rwbase: Extract __rwbase_write_trylock()
  locking/rwbase: Properly match set_and_save_state() to restore_state()

2 years agoMerge tag 'powerpc-5.15-2' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sun, 19 Sep 2021 20:00:23 +0000 (13:00 -0700)]
Merge tag 'powerpc-5.15-2' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:

 - Fix crashes when scv (System Call Vectored) is used to make a syscall
   when a transaction is active, on Power9 or later.

 - Fix bad interactions between rfscv (Return-from scv) and Power9
   fake-suspend mode.

 - Fix crashes when handling machine checks in LPARs using the Hash MMU.

 - Partly revert a recent change to our XICS interrupt controller code,
   which broke the recently added Microwatt support.

Thanks to Cédric Le Goater, Eirik Fuller, Ganesh Goudar, Gustavo Romero,
Joel Stanley, Nicholas Piggin.

* tag 'powerpc-5.15-2' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/xics: Set the IRQ chip data for the ICS native backend
  powerpc/mce: Fix access error in mce handler
  KVM: PPC: Book3S HV: Tolerate treclaim. in fake-suspend mode changing registers
  powerpc/64s: system call rfscv workaround for TM bugs
  selftests/powerpc: Add scv versions of the basic TM syscall tests
  powerpc/64s: system call scv tabort fix for corrupt irq soft-mask state

2 years agoMerge tag 'kbuild-fixes-v5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/masah...
Linus Torvalds [Sun, 19 Sep 2021 19:55:12 +0000 (12:55 -0700)]
Merge tag 'kbuild-fixes-v5.15' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - Fix bugs in checkkconfigsymbols.py

 - Fix missing sys import in gen_compile_commands.py

 - Fix missing FORCE warning for ARCH=sh builds

 - Fix -Wignored-optimization-argument warnings for Clang builds

 - Turn -Wignored-optimization-argument into an error in order to stop
   building instead of sprinkling warnings

* tag 'kbuild-fixes-v5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  kbuild: Add -Werror=ignored-optimization-argument to CLANG_FLAGS
  x86/build: Do not add -falign flags unconditionally for clang
  kbuild: Fix comment typo in scripts/Makefile.modpost
  sh: Add missing FORCE prerequisites in Makefile
  gen_compile_commands: fix missing 'sys' package
  checkkconfigsymbols.py: Remove skipping of help lines in parse_kconfig_file
  checkkconfigsymbols.py: Forbid passing 'HEAD' to --commit

2 years agoMerge tag 'perf-tools-fixes-for-v5.15-2021-09-18' of git://git.kernel.org/pub/scm...
Linus Torvalds [Sun, 19 Sep 2021 19:49:17 +0000 (12:49 -0700)]
Merge tag 'perf-tools-fixes-for-v5.15-2021-09-18' of git://git./linux/kernel/git/acme/linux

Pull perf tools fixes from Arnaldo Carvalho de Melo:

 - Fix ip display in 'perf script' when output type != attr->type.

 - Ignore deprecation warning when using libbpf'sg btf__get_from_id(),
   fixing the build with libbpf v0.6+.

 - Make use of FD() robust in libperf, fixing a segfault with 'perf stat
   --iostat list'.

 - Initialize addr_location:srcline pointer to NULL when resolving
   callchain addresses.

 - Fix fused instruction logic for assembly functions in 'perf
   annotate'.

* tag 'perf-tools-fixes-for-v5.15-2021-09-18' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux:
  perf bpf: Ignore deprecation warning when using libbpf's btf__get_from_id()
  libperf evsel: Make use of FD robust.
  perf machine: Initialize srcline string member in add_location struct
  perf script: Fix ip display when type != attr->type
  perf annotate: Fix fused instr logic for assembly functions

2 years agodmascc: use proper 'virt_to_bus()' rather than casting to 'int'
Linus Torvalds [Sun, 19 Sep 2021 17:49:42 +0000 (10:49 -0700)]
dmascc: use proper 'virt_to_bus()' rather than casting to 'int'

The old dmascc driver depends on the legacy ISA_DMA_API, and blindly
just casts the kernel virtual address to 'int' for set_dma_addr().

That works only incidentally, and because the high bits of the address
will be ignored anyway. And on 64-bit architectures it causes warnings.

Admittedly, 64-bit architectures with ISA are basically dead - I think
the only example of this is alpha, and nobody would ever use the dmascc
driver there.  But hey, the fix is easy enough, the end result is
cleaner, and it's yet another configuration that now builds without
warnings.

If somebody actually uses this driver on an alpha and this fixes it for
you, please email me.  Because that is just incredibly bizarre.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoalpha: enable GENERIC_PCI_IOMAP unconditionally
Linus Torvalds [Sun, 19 Sep 2021 17:37:00 +0000 (10:37 -0700)]
alpha: enable GENERIC_PCI_IOMAP unconditionally

With the previous commit (9caea0007601: "parisc: Declare pci_iounmap()
parisc version only when CONFIG_PCI enabled") we can now enable
GENERIC_PCI_IOMAP unconditionally on alpha, and if PCI is not enabled we
will just get the nice empty helper functions that allow mixed-bus
drivers to build.

Example driver: the old 3com/3c59x.c driver works with either the PCI or
the EISA version of the 3x59x card, but wouldn't build in an EISA-only
configuration because of missing pci_iomap() and pci_iounmap() dummy
wrappers.

Most of the other PCI infrastructure just becomes empty wrappers even
without GENERIC_PCI_IOMAP, and it's not obvious that the pci_iomap
functionality shouldn't do the same, but this works.

Cc: Ulrich Teichert <krypton@ulrich-teichert.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoparisc: Declare pci_iounmap() parisc version only when CONFIG_PCI enabled
Helge Deller [Sun, 19 Sep 2021 17:36:09 +0000 (10:36 -0700)]
parisc: Declare pci_iounmap() parisc version only when CONFIG_PCI enabled

Linus noticed odd declaration rules for pci_iounmap() in iomap.h and
pci_iomap.h, where it dependend on either NO_GENERIC_PCI_IOPORT_MAP or
GENERIC_IOMAP when CONFIG_PCI was disabled.

Testing on parisc seems to indicate that we need pci_iounmap() only when
CONFIG_PCI is enabled, so the declaration of pci_iounmap() can be moved
cleanly into pci_iomap.h in sync with the declarations of pci_iomap().

Link: https://lore.kernel.org/all/CAHk-=wjRrh98pZoQ+AzfWmsTZacWxTJKXZ9eKU2X_0+jM=O8nw@mail.gmail.com/
Signed-off-by: Helge Deller <deller@gmx.de>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Fixes: 97a29d59fc22 ("[PARISC] fix compile break caused by iomap: make IOPORT/PCI mapping functions conditional")
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Ulrich Teichert <krypton@ulrich-teichert.org>
Cc: James Bottomley <James.Bottomley@hansenpartnership.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoRevert "drm/vc4: hdmi: Remove drm_encoder->crtc usage"
Linus Torvalds [Sun, 19 Sep 2021 17:11:53 +0000 (10:11 -0700)]
Revert "drm/vc4: hdmi: Remove drm_encoder->crtc usage"

This reverts commit 27da370e0fb343a0baf308f503bb3e5dcdfe3362.

Sudip Mukherjee reports that this broke pulseaudio with a NULL pointer
dereference in vc4_hdmi_audio_prepare(), bisected it to this commit, and
confirmed that a revert fixed the problem.

Revert the problematic commit until fixed.

Link: https://lore.kernel.org/all/CADVatmPB9-oKd=ypvj25UYysVo6EZhQ6bCM7EvztQBMyiZfAyw@mail.gmail.com/
Link: https://lore.kernel.org/all/CADVatmN5EpRshGEPS_JozbFQRXg5w_8LFB3OMP1Ai-ghxd3w4g@mail.gmail.com/
Reported-and-tested-by: Sudip Mukherjee <sudipm.mukherjee@gmail.com>
Cc: Maxime Ripard <maxime@cerno.tech>
Cc: Emma Anholt <emma@anholt.net>
Cc: Dave Airlie <airlied@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoRevert drm/vc4 hdmi runtime PM changes
Linus Torvalds [Sun, 19 Sep 2021 17:06:46 +0000 (10:06 -0700)]
Revert drm/vc4 hdmi runtime PM changes

This reverts commits

  9984d6664ce9 ("drm/vc4: hdmi: Make sure the controller is powered in detect")
  411efa18e4b0 ("drm/vc4: hdmi: Move the HSM clock enable to runtime_pm")

as Michael Stapelberg reports that the new runtime PM changes cause his
Raspberry Pi 3 to hang on boot, probably due to interactions with other
changes in the DRM tree (because a bisect points to the merge in commit
e058a84bfddc: "Merge tag 'drm-next-2021-07-01' of git://.../drm").

Revert these two commits until it's been resolved.

Link: https://lore.kernel.org/all/871r5mp7h2.fsf@midna.i-did-not-set--mail-host-address--so-tickle-me/
Reported-and-tested-by: Michael Stapelberg <michael@stapelberg.ch>
Cc: Maxime Ripard <maxime@cerno.tech>
Cc: Dave Stevenson <dave.stevenson@raspberrypi.com>
Cc: Dave Airlie <airlied@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agokbuild: Add -Werror=ignored-optimization-argument to CLANG_FLAGS
Nathan Chancellor [Thu, 16 Sep 2021 18:40:17 +0000 (11:40 -0700)]
kbuild: Add -Werror=ignored-optimization-argument to CLANG_FLAGS

Similar to commit 589834b3a009 ("kbuild: Add
-Werror=unknown-warning-option to CLANG_FLAGS").

Clang ignores certain GCC flags that it has not implemented, only
emitting a warning:

$ echo | clang -fsyntax-only -falign-jumps -x c -
clang-14: warning: optimization flag '-falign-jumps' is not supported
[-Wignored-optimization-argument]

When one of these flags gets added to KBUILD_CFLAGS unconditionally, all
subsequent cc-{disable-warning,option} calls fail because -Werror was
added to these invocations to turn the above warning and the equivalent
-W flag warning into errors.

To catch the presence of these flags earlier, turn
-Wignored-optimization-argument into an error so that the flags can
either be implemented or ignored via cc-option and there are no more
weird errors.

Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
2 years agox86/build: Do not add -falign flags unconditionally for clang
Nathan Chancellor [Thu, 16 Sep 2021 18:40:16 +0000 (11:40 -0700)]
x86/build: Do not add -falign flags unconditionally for clang

clang does not support -falign-jumps and only recently gained support
for -falign-loops. When one of the configuration options that adds these
flags is enabled, clang warns and all cc-{disable-warning,option} that
follow fail because -Werror gets added to test for the presence of this
warning:

clang-14: warning: optimization flag '-falign-jumps=0' is not supported
[-Wignored-optimization-argument]

To resolve this, add a couple of cc-option calls when building with
clang; gcc has supported these options since 3.2 so there is no point in
testing for their support. -falign-functions was implemented in clang-7,
-falign-loops was implemented in clang-14, and -falign-jumps has not
been implemented yet.

Link: https://lore.kernel.org/r/YSQE2f5teuvKLkON@Ryzen-9-3900X.localdomain/
Link: https://lore.kernel.org/r/20210824022640.2170859-2-nathan@kernel.org/
Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
2 years agokbuild: Fix comment typo in scripts/Makefile.modpost
Ramji Jiyani [Thu, 16 Sep 2021 09:21:22 +0000 (09:21 +0000)]
kbuild: Fix comment typo in scripts/Makefile.modpost

Change comment "create one <module>.mod.c file pr. module"
to "create one <module>.mod.c file per module"

Signed-off-by: Ramji Jiyani <ramjiyani@google.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
2 years agosh: Add missing FORCE prerequisites in Makefile
Geert Uytterhoeven [Thu, 16 Sep 2021 08:43:53 +0000 (10:43 +0200)]
sh: Add missing FORCE prerequisites in Makefile

make:

    arch/sh/boot/Makefile:87: FORCE prerequisite is missing

Add the missing FORCE prerequisites for all build targets identified by
"make help".

Fixes: e1f86d7b4b2a5213 ("kbuild: warn if FORCE is missing for if_changed(_dep,_rule) and filechk")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
2 years agogen_compile_commands: fix missing 'sys' package
Kortan [Wed, 8 Sep 2021 03:28:48 +0000 (11:28 +0800)]
gen_compile_commands: fix missing 'sys' package

We need to import the 'sys' package since the script has called
sys.exit() method.

Fixes: 6ad7cbc01527 ("Makefile: Add clang-tidy and static analyzer support to makefile")
Signed-off-by: Kortan <kortanzh@gmail.com>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
2 years agocheckkconfigsymbols.py: Remove skipping of help lines in parse_kconfig_file
Ariel Marcovitch [Wed, 1 Sep 2021 16:49:52 +0000 (19:49 +0300)]
checkkconfigsymbols.py: Remove skipping of help lines in parse_kconfig_file

When parsing Kconfig files to find symbol definitions and references,
lines after a 'help' line are skipped until a new config definition
starts.

However, Kconfig statements can actually be after a help section, as
long as these have shallower indentation. These are skipped by the
parser.

This means that symbols referenced in this kind of statements are
ignored by this function and thus are not considered undefined
references in case the symbol is not defined.

Remove the 'skip' logic entirely, as it is not needed if we just use the
STMT regex to find the end of help lines.

However, this means that keywords that appear as part of the help
message (i.e. with the same indentation as the help lines) it will be
considered as a reference/definition. This can happen now as well, but
only with REGEX_KCONFIG_DEF lines. Also, the keyword must have a SYMBOL
after it, which probably means that someone referenced a config in the
help so it seems like a bonus :)

The real solution is to keep track of the indentation when a the first
help line in encountered and then handle DEF and STMT lines only if the
indentation is shallower.

Signed-off-by: Ariel Marcovitch <arielmarcovitch@gmail.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
2 years agocheckkconfigsymbols.py: Forbid passing 'HEAD' to --commit
Ariel Marcovitch [Wed, 1 Sep 2021 14:52:12 +0000 (17:52 +0300)]
checkkconfigsymbols.py: Forbid passing 'HEAD' to --commit

As opposed to the --diff option, --commit can get ref names instead of
commit hashes.

When using the --commit option, the script resets the working directory
to the commit before the given ref, by adding '~' to the end of the ref.

However, the 'HEAD' ref is relative, and so when the working directory
is reset to 'HEAD~', 'HEAD' points to what was 'HEAD~'. Then when the
script resets to 'HEAD' it actually stays in the same commit. In this
case, the script won't report any cases because there is no diff between
the cases of the two refs.

Prevent the user from using HEAD refs.

A better solution might be to resolve the refs before doing the
reset, but for now just disallow such refs.

Signed-off-by: Ariel Marcovitch <arielmarcovitch@gmail.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
2 years agoalpha: move __udiv_qrnnd library function to arch/alpha/lib/
Linus Torvalds [Sat, 18 Sep 2021 21:45:48 +0000 (14:45 -0700)]
alpha: move __udiv_qrnnd library function to arch/alpha/lib/

We already had the implementation for __udiv_qrnnd (unsigned divide for
multi-precision arithmetic) as part of the alpha math emulation code.

But you can disable the math emulation code - even if you shouldn't -
and then the MPI code that actually wants this functionality (and is
needed by various crypto functions) will fail to build.

So move the extended-precision divide code to be a regular library
function, just like all the regular division code is.  That way ie is
available regardless of math-emulation.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoalpha: mark 'Jensen' platform as no longer broken
Linus Torvalds [Sat, 18 Sep 2021 21:12:39 +0000 (14:12 -0700)]
alpha: mark 'Jensen' platform as no longer broken

Ok, it almost certainly is still broken on actual hardware, but the
immediate reason for it having been marked BROKEN was a build error that
is fixed by just making sure the low-level IO header file is included
sufficiently early that the __EXTERN_INLINE hackery takes effect.

This was marked broken back in 2017 by commit 1883c9f49d02 ("alpha: mark
jensen as broken"), but Ulrich Teichert made me look at it as part of my
cross-build work to make sure -Werror actually does the right thing.

There are lots of alpha configurations that do not build cleanly, but
now it's no longer because Jensen wouldn't be buildable.  That said,
because the Jensen platform doesn't force PCI to be enabled (Jensen only
had EISA), it ends up being somewhat interesting as a source of odd
configs.

Reported-by: Ulrich Teichert <krypton@ulrich-teichert.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoperf bpf: Ignore deprecation warning when using libbpf's btf__get_from_id()
Andrii Nakryiko [Tue, 14 Sep 2021 17:00:04 +0000 (10:00 -0700)]
perf bpf: Ignore deprecation warning when using libbpf's btf__get_from_id()

Perf code re-implements libbpf's btf__load_from_kernel_by_id() API as
a weak function, presumably to dynamically link against old version of
libbpf shared library. Unfortunately this causes compilation warning
when perf is compiled against libbpf v0.6+.

For now, just ignore deprecation warning, but there might be a better
solution, depending on perf's needs.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: kernel-team@fb.com
LPU-Reference: 20210914170004.4185659-1-andrii@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2 years agolibperf evsel: Make use of FD robust.
Ian Rogers [Sat, 18 Sep 2021 05:44:40 +0000 (22:44 -0700)]
libperf evsel: Make use of FD robust.

FD uses xyarray__entry that may return NULL if an index is out of
bounds. If NULL is returned then a segv happens as FD unconditionally
dereferences the pointer. This was happening in a case of with perf
iostat as shown below. The fix is to make FD an "int*" rather than an
int and handle the NULL case as either invalid input or a closed fd.

  $ sudo gdb --args perf stat --iostat  list
  ...
  Breakpoint 1, perf_evsel__alloc_fd (evsel=0x5555560951a0, ncpus=1, nthreads=1) at evsel.c:50
  50      {
  (gdb) bt
   #0  perf_evsel__alloc_fd (evsel=0x5555560951a0, ncpus=1, nthreads=1) at evsel.c:50
   #1  0x000055555585c188 in evsel__open_cpu (evsel=0x5555560951a0, cpus=0x555556093410,
      threads=0x555556086fb0, start_cpu=0, end_cpu=1) at util/evsel.c:1792
   #2  0x000055555585cfb2 in evsel__open (evsel=0x5555560951a0, cpus=0x0, threads=0x555556086fb0)
      at util/evsel.c:2045
   #3  0x000055555585d0db in evsel__open_per_thread (evsel=0x5555560951a0, threads=0x555556086fb0)
      at util/evsel.c:2065
   #4  0x00005555558ece64 in create_perf_stat_counter (evsel=0x5555560951a0,
      config=0x555555c34700 <stat_config>, target=0x555555c2f1c0 <target>, cpu=0) at util/stat.c:590
   #5  0x000055555578e927 in __run_perf_stat (argc=1, argv=0x7fffffffe4a0, run_idx=0)
      at builtin-stat.c:833
   #6  0x000055555578f3c6 in run_perf_stat (argc=1, argv=0x7fffffffe4a0, run_idx=0)
      at builtin-stat.c:1048
   #7  0x0000555555792ee5 in cmd_stat (argc=1, argv=0x7fffffffe4a0) at builtin-stat.c:2534
   #8  0x0000555555835ed3 in run_builtin (p=0x555555c3f540 <commands+288>, argc=3,
      argv=0x7fffffffe4a0) at perf.c:313
   #9  0x0000555555836154 in handle_internal_command (argc=3, argv=0x7fffffffe4a0) at perf.c:365
   #10 0x000055555583629f in run_argv (argcp=0x7fffffffe2ec, argv=0x7fffffffe2e0) at perf.c:409
   #11 0x0000555555836692 in main (argc=3, argv=0x7fffffffe4a0) at perf.c:539
  ...
  (gdb) c
  Continuing.
  Error:
  The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (uncore_iio_0/event=0x83,umask=0x04,ch_mask=0xF,fc_mask=0x07/).
  /bin/dmesg | grep -i perf may provide additional information.

  Program received signal SIGSEGV, Segmentation fault.
  0x00005555559b03ea in perf_evsel__close_fd_cpu (evsel=0x5555560951a0, cpu=1) at evsel.c:166
  166                     if (FD(evsel, cpu, thread) >= 0)

v3. fixes a bug in perf_evsel__run_ioctl where the sense of a branch was
    backward.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20210918054440.2350466-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2 years agoperf machine: Initialize srcline string member in add_location struct
Michael Petlan [Mon, 19 Jul 2021 14:53:32 +0000 (16:53 +0200)]
perf machine: Initialize srcline string member in add_location struct

It's later supposed to be either a correct address or NULL. Without the
initialization, it may contain an undefined value which results in the
following segmentation fault:

  # perf top --sort comm -g --ignore-callees=do_idle

terminates with:

  #0  0x00007ffff56b7685 in __strlen_avx2 () from /lib64/libc.so.6
  #1  0x00007ffff55e3802 in strdup () from /lib64/libc.so.6
  #2  0x00005555558cb139 in hist_entry__init (callchain_size=<optimized out>, sample_self=true, template=0x7fffde7fb110, he=0x7fffd801c250) at util/hist.c:489
  #3  hist_entry__new (template=template@entry=0x7fffde7fb110, sample_self=sample_self@entry=true) at util/hist.c:564
  #4  0x00005555558cb4ba in hists__findnew_entry (hists=hists@entry=0x5555561d9e38, entry=entry@entry=0x7fffde7fb110, al=al@entry=0x7fffde7fb420,
      sample_self=sample_self@entry=true) at util/hist.c:657
  #5  0x00005555558cba1b in __hists__add_entry (hists=hists@entry=0x5555561d9e38, al=0x7fffde7fb420, sym_parent=<optimized out>, bi=bi@entry=0x0, mi=mi@entry=0x0,
      sample=sample@entry=0x7fffde7fb4b0, sample_self=true, ops=0x0, block_info=0x0) at util/hist.c:288
  #6  0x00005555558cbb70 in hists__add_entry (sample_self=true, sample=0x7fffde7fb4b0, mi=0x0, bi=0x0, sym_parent=<optimized out>, al=<optimized out>, hists=0x5555561d9e38)
      at util/hist.c:1056
  #7  iter_add_single_cumulative_entry (iter=0x7fffde7fb460, al=<optimized out>) at util/hist.c:1056
  #8  0x00005555558cc8a4 in hist_entry_iter__add (iter=iter@entry=0x7fffde7fb460, al=al@entry=0x7fffde7fb420, max_stack_depth=<optimized out>, arg=arg@entry=0x7fffffff7db0)
      at util/hist.c:1231
  #9  0x00005555557cdc9a in perf_event__process_sample (machine=<optimized out>, sample=0x7fffde7fb4b0, evsel=<optimized out>, event=<optimized out>, tool=0x7fffffff7db0)
      at builtin-top.c:842
  #10 deliver_event (qe=<optimized out>, qevent=<optimized out>) at builtin-top.c:1202
  #11 0x00005555558a9318 in do_flush (show_progress=false, oe=0x7fffffff80e0) at util/ordered-events.c:244
  #12 __ordered_events__flush (oe=oe@entry=0x7fffffff80e0, how=how@entry=OE_FLUSH__TOP, timestamp=timestamp@entry=0) at util/ordered-events.c:323
  #13 0x00005555558a9789 in __ordered_events__flush (timestamp=<optimized out>, how=<optimized out>, oe=<optimized out>) at util/ordered-events.c:339
  #14 ordered_events__flush (how=OE_FLUSH__TOP, oe=0x7fffffff80e0) at util/ordered-events.c:341
  #15 ordered_events__flush (oe=oe@entry=0x7fffffff80e0, how=how@entry=OE_FLUSH__TOP) at util/ordered-events.c:339
  #16 0x00005555557cd631 in process_thread (arg=0x7fffffff7db0) at builtin-top.c:1114
  #17 0x00007ffff7bb817a in start_thread () from /lib64/libpthread.so.0
  #18 0x00007ffff5656dc3 in clone () from /lib64/libc.so.6

If you look at the frame #2, the code is:

488  if (he->srcline) {
489          he->srcline = strdup(he->srcline);
490          if (he->srcline == NULL)
491              goto err_rawdata;
492  }

If he->srcline is not NULL (it is not NULL if it is uninitialized rubbish),
it gets strdupped and strdupping a rubbish random string causes the problem.

Also, if you look at the commit 1fb7d06a509e, it adds the srcline property
into the struct, but not initializing it everywhere needed.

Committer notes:

Now I see, when using --ignore-callees=do_idle we end up here at line
2189 in add_callchain_ip():

2181         if (al.sym != NULL) {
2182                 if (perf_hpp_list.parent && !*parent &&
2183                     symbol__match_regex(al.sym, &parent_regex))
2184                         *parent = al.sym;
2185                 else if (have_ignore_callees && root_al &&
2186                   symbol__match_regex(al.sym, &ignore_callees_regex)) {
2187                         /* Treat this symbol as the root,
2188                            forgetting its callees. */
2189                         *root_al = al;
2190                         callchain_cursor_reset(cursor);
2191                 }
2192         }

And the al that doesn't have the ->srcline field initialized will be
copied to the root_al, so then, back to:

1211 int hist_entry_iter__add(struct hist_entry_iter *iter, struct addr_location *al,
1212                          int max_stack_depth, void *arg)
1213 {
1214         int err, err2;
1215         struct map *alm = NULL;
1216
1217         if (al)
1218                 alm = map__get(al->map);
1219
1220         err = sample__resolve_callchain(iter->sample, &callchain_cursor, &iter->parent,
1221                                         iter->evsel, al, max_stack_depth);
1222         if (err) {
1223                 map__put(alm);
1224                 return err;
1225         }
1226
1227         err = iter->ops->prepare_entry(iter, al);
1228         if (err)
1229                 goto out;
1230
1231         err = iter->ops->add_single_entry(iter, al);
1232         if (err)
1233                 goto out;
1234

That al at line 1221 is what hist_entry_iter__add() (called from
sample__resolve_callchain()) saw as 'root_al', and then:

        iter->ops->add_single_entry(iter, al);

will go on with al->srcline with a bogus value, I'll add the above
sequence to the cset and apply, thanks!

Signed-off-by: Michael Petlan <mpetlan@redhat.com>
CC: Milian Wolff <milian.wolff@kdab.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Fixes: 1fb7d06a509e ("perf report Use srcline from callchain for hist entries")
Link: https //lore.kernel.org/r/20210719145332.29747-1-mpetlan@redhat.com
Reported-by: Juri Lelli <jlelli@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2 years agoperf script: Fix ip display when type != attr->type
Adrian Hunter [Sat, 11 Sep 2021 13:30:53 +0000 (16:30 +0300)]
perf script: Fix ip display when type != attr->type

set_print_ip_opts() was not being called when type != attr->type
because there is not a one-to-one relationship between output types
and attr->type. That resulted in ip not printing.

The attr_type() function is removed, and the match of attr->type to
output type is corrected.

Example on ADL using taskset to select an atom cpu:

 # perf record -e cpu_atom/cpu-cycles/ taskset 0x1000 uname
 Linux
 [ perf record: Woken up 1 times to write data ]
 [ perf record: Captured and wrote 0.003 MB perf.data (7 samples) ]

 Before:

  # perf script | head
         taskset   428 [-01] 10394.179041:          1 cpu_atom/cpu-cycles/:
         taskset   428 [-01] 10394.179043:          1 cpu_atom/cpu-cycles/:
         taskset   428 [-01] 10394.179044:         11 cpu_atom/cpu-cycles/:
         taskset   428 [-01] 10394.179045:        407 cpu_atom/cpu-cycles/:
         taskset   428 [-01] 10394.179046:      16789 cpu_atom/cpu-cycles/:
         taskset   428 [-01] 10394.179052:     676300 cpu_atom/cpu-cycles/:
           uname   428 [-01] 10394.179278:    4079859 cpu_atom/cpu-cycles/:

 After:

  # perf script | head
         taskset   428 10394.179041:          1 cpu_atom/cpu-cycles/:  ffffffff95a0bb97 __intel_pmu_enable_all.constprop.48+0x47 ([kernel.kallsyms])
         taskset   428 10394.179043:          1 cpu_atom/cpu-cycles/:  ffffffff95a0bb97 __intel_pmu_enable_all.constprop.48+0x47 ([kernel.kallsyms])
         taskset   428 10394.179044:         11 cpu_atom/cpu-cycles/:  ffffffff95a0bb97 __intel_pmu_enable_all.constprop.48+0x47 ([kernel.kallsyms])
         taskset   428 10394.179045:        407 cpu_atom/cpu-cycles/:  ffffffff95a0bb97 __intel_pmu_enable_all.constprop.48+0x47 ([kernel.kallsyms])
         taskset   428 10394.179046:      16789 cpu_atom/cpu-cycles/:  ffffffff95a0bb97 __intel_pmu_enable_all.constprop.48+0x47 ([kernel.kallsyms])
         taskset   428 10394.179052:     676300 cpu_atom/cpu-cycles/:      7f829ef73800 cfree+0x0 (/lib/libc-2.32.so)
           uname   428 10394.179278:    4079859 cpu_atom/cpu-cycles/:  ffffffff95bae912 vma_interval_tree_remove+0x1f2 ([kernel.kallsyms])

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lore.kernel.org/lkml/20210911133053.15682-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2 years agoperf annotate: Fix fused instr logic for assembly functions
Ravi Bangoria [Sat, 11 Sep 2021 04:38:53 +0000 (10:08 +0530)]
perf annotate: Fix fused instr logic for assembly functions

Some x86 microarchitectures fuse a subset of cmp/test/ALU instructions
with branch instructions, and thus perf annotate highlight such valid
pairs as fused.

When annotated with source, perf uses struct disasm_line to contain
either source or instruction line from objdump output. Usually, a C
statement generates multiple instructions which include such
cmp/test/ALU + branch instruction pairs. But in case of assembly
function, each individual assembly source line generate one
instruction.

The 'perf annotate' instruction fusion logic assumes the previous
disasm_line as the previous instruction line, which is wrong because,
for assembly function, previous disasm_line contains source line.  And
thus perf fails to highlight valid fused instruction pairs for assembly
functions.

Fix it by searching backward until we find an instruction line and
consider that disasm_line as fused with current branch instruction.

Before:
         │    cmpq    %rcx, RIP+8(%rsp)
    0.00 │      cmp    %rcx,0x88(%rsp)
         │    je      .Lerror_bad_iret      <--- Source line
    0.14 │   ┌──je     b4                   <--- Instruction line
         │   │movl    %ecx, %eax

After:
         │    cmpq    %rcx, RIP+8(%rsp)
    0.00 │   ┌──cmp    %rcx,0x88(%rsp)
         │   │je      .Lerror_bad_iret
    0.14 │   ├──je     b4
         │   │movl    %ecx, %eax

Reviewed-by: Jin Yao <yao.jin@linux.intel.com>
Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kim Phillips <kim.phillips@amd.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https //lore.kernel.org/r/20210911043854.8373-1-ravi.bangoria@amd.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2 years agoMerge tag 's390-5.15-3' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Sat, 18 Sep 2021 19:46:14 +0000 (12:46 -0700)]
Merge tag 's390-5.15-3' of git://git./linux/kernel/git/s390/linux

Pull s390 fixes from Vasily Gorbik:

 - Fix potential out-of-range access during secure boot facility
   detection.

 - Fully validate the VMA before calling follow_pte() in pci code.

 - Remove arch specific WARN_DYNAMIC_STACK config option.

 - Fix zcrypto kernel doc comments.

 - Update defconfigs.

* tag 's390-5.15-3' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390: remove WARN_DYNAMIC_STACK
  s390/ap: fix kernel doc comments
  s390: update defconfigs
  s390/sclp: fix Secure-IPL facility detection
  s390/pci_mmio: fully validate the VMA before calling follow_pte()

2 years agoMerge tag 'devicetree-fixes-for-5.15-2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 18 Sep 2021 19:40:55 +0000 (12:40 -0700)]
Merge tag 'devicetree-fixes-for-5.15-2' of git://git./linux/kernel/git/robh/linux

Pull devicetree fixes from Rob Herring:

 - Revert fw_devlink tracking 'phy-handle' links. This broke at least a
   few platforms. A better solution is being worked on.

 - Add Samsung UFS binding which fell thru the cracks

 - Doc reference fixes from Mauro

 - Fix for restricted DMA error handling

* tag 'devicetree-fixes-for-5.15-2' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  dt-bindings: arm: Fix Toradex compatible typo
  of: restricted dma: Fix condition for rmem init
  dt-bindings: arm: mediatek: mmsys: update mediatek,mmsys.yaml reference
  dt-bindings: net: dsa: sja1105: update nxp,sja1105.yaml reference
  dt-bindings: ufs: Add bindings for Samsung ufs host
  Revert "of: property: fw_devlink: Add support for "phy-handle" property"

2 years agotgafb: clarify dependencies
Linus Torvalds [Sat, 18 Sep 2021 18:15:01 +0000 (11:15 -0700)]
tgafb: clarify dependencies

The TGA boards were based on the DECchip 21030 PCI graphics accelerator
used mainly for alpha, and existed in a TURBOchannel (TC) version for
the DECstation (MIPS) workstations.

However, the config option for the TGA code is a bit confused, and says

depends on FB && (ALPHA || TC)

because people didn't really want to enable the option for random PCI
environments, so the "ALPHA" stands in for that case (while the TC case
is then the MIPS DECstation case).

So that config dependency is kind of a mixture of architecture and bus
choices.  But it's incorrect, in that there were non-PCI-based alpha
hardware, and then the driver just causes warnings:

  drivers/video/fbdev/tgafb.c:1532:13: error: ‘tgafb_unregister’ defined but not used [-Werror=unused-function]
   1532 | static void tgafb_unregister(struct device *dev)
        |             ^~~~~~~~~~~~~~~~
  drivers/video/fbdev/tgafb.c:1387:12: error: ‘tgafb_register’ defined but not used [-Werror=unused-function]
   1387 | static int tgafb_register(struct device *dev)
        |            ^~~~~~~~~~~~~~

so let's make the config option dependencies a bit more explict:

depends on FB
depends on PCI || TC
depends on ALPHA || TC

where that first "FB" is the software configuration dependency, the
second "PCI || TC" is the hardware bus dependency, while that final
"ALPHA || TC" dependency is the "don't bother asking except for these
situations.

We could make that third case have "COMPILE_TEST" as an option, and mark
the register/unregister functions as __maybe_unused, but I'm not sure
it's really worth it.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoalpha: make 'Jensen' IO functions build again
Linus Torvalds [Sat, 18 Sep 2021 17:57:10 +0000 (10:57 -0700)]
alpha: make 'Jensen' IO functions build again

The Jensen IO functions are overly copmplicated because some of the IO
addresses refer to special 'local IO' ports, and they get accessed
differently.

That then makes gcc not actually inline them, and since they were marked
"extern inline" when included through the regular <asm/io.h> path, and
then only marked "inline" when included from sys_jensen.c, you never
necessarily got a body for the IO functions at all.

The intent of the sys_jensen.c code is to actually get the non-inlined
copy generated, so remove the 'inline' from the magic macro that is
supposed to sort this all out.

Also, do not mix 'extern inline' functions (that may or may not be
inlined and will not generate a function body if they are not) with
'static inline' (that _will_ generate a function body when not inlined).
Because gcc will complain about this situation:

   error: ‘jensen_bus_outb’ is static but used in inline function ‘jensen_outb’ which is not static

because gcc basically doesn't know whether to generate a body for that
static inline function or not for that call site.

So make all of these use that __EXTERN_INLINE marker.  Gcc will
generally not inline these things on use, and then generate the function
body out-of-line in sys_jensen.c.

This makes the core IO functions build for the alpha Jensen config.

Not that the rest then builds, because it turns out Jensen also doesn't
enable PCI, which then makes other drievrs very unhappy, but that's a
separate issue.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agospi: Fix tegra20 build with CONFIG_PM=n
Linus Torvalds [Sat, 18 Sep 2021 17:05:06 +0000 (10:05 -0700)]
spi: Fix tegra20 build with CONFIG_PM=n

Without CONFIG_PM enabled, the SET_RUNTIME_PM_OPS() macro ends up being
empty, and the only use of tegra_slink_runtime_{resume,suspend} goes
away, resulting in

  drivers/spi/spi-tegra20-slink.c:1200:12: error: ‘tegra_slink_runtime_resume’ defined but not used [-Werror=unused-function]
   1200 | static int tegra_slink_runtime_resume(struct device *dev)
        |            ^~~~~~~~~~~~~~~~~~~~~~~~~~
  drivers/spi/spi-tegra20-slink.c:1188:12: error: ‘tegra_slink_runtime_suspend’ defined but not used [-Werror=unused-function]
   1188 | static int tegra_slink_runtime_suspend(struct device *dev)
        |            ^~~~~~~~~~~~~~~~~~~~~~~~~~~

mark the functions __maybe_unused to make the build happy.

This hits the alpha allmodconfig build (and others).

Reported-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agodt-bindings: arm: Fix Toradex compatible typo
David Heidelberg [Sun, 12 Sep 2021 16:51:20 +0000 (18:51 +0200)]
dt-bindings: arm: Fix Toradex compatible typo

Fix board compatible typo reported by dtbs_check.

Fixes: f4d1577e9bc6 ("dt-bindings: arm: Convert Tegra board/soc bindings to json-schema")
Signed-off-by: David Heidelberg <david@ixit.cz>
Link: https://lore.kernel.org/r/20210912165120.188490-1-david@ixit.cz
Signed-off-by: Rob Herring <robh@kernel.org>
2 years agoof: restricted dma: Fix condition for rmem init
David Brazdil [Fri, 17 Sep 2021 13:14:23 +0000 (14:14 +0100)]
of: restricted dma: Fix condition for rmem init

of_dma_set_restricted_buffer fails to handle negative return values from
of_property_count_elems_of_size, e.g. when the property does not exist.
This results in an attempt to assign a non-existent reserved memory
region to the device and a warning being printed. Fix the condition to
take negative values into account.

Fixes: f3cfd136aef0 ("of: restricted dma: Don't fail device probe on rmem init failure")
Cc: Will Deacon <will@kernel.org>
Signed-off-by: David Brazdil <dbrazdil@google.com>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20210917131423.2760155-1-dbrazdil@google.com
Signed-off-by: Rob Herring <robh@kernel.org>
2 years agoMerge tag 'pm-5.15-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Fri, 17 Sep 2021 19:05:04 +0000 (12:05 -0700)]
Merge tag 'pm-5.15-rc2' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "These fix two cpufreq issues, one in the intel_pstate driver and one
  in the core.

  Specifics:

   - Prevent intel_pstate from avoiding to use HWP, even if instructed
     to do so via the kernel command line, when HWP has been enabled
     already by the platform firmware (Doug Smythies).

   - Prevent use-after-free from occurring in the schedutil cpufreq
     governor on exit by fixing a core helper function that attempts to
     access memory associated with a kobject after calling kobject_put()
     on it (James Morse)"

* tag 'pm-5.15-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  cpufreq: schedutil: Destroy mutex before kobject_put() frees the memory
  cpufreq: intel_pstate: Override parameters if HWP forced by BIOS

2 years agoMerge tag 'dma-mapping-5.15-1' of git://git.infradead.org/users/hch/dma-mapping
Linus Torvalds [Fri, 17 Sep 2021 18:54:48 +0000 (11:54 -0700)]
Merge tag 'dma-mapping-5.15-1' of git://git.infradead.org/users/hch/dma-mapping

Pull dma-mapping fixes from Christoph Hellwig:

 - page align size in sparc32 arch_dma_alloc (Andreas Larsson)

 - tone down a new dma-debug message (Hamza Mahfooz)

 - fix the kerneldoc for dma_map_sg_attrs (me)

* tag 'dma-mapping-5.15-1' of git://git.infradead.org/users/hch/dma-mapping:
  sparc32: page align size in arch_dma_alloc
  dma-debug: prevent an error message from causing runtime problems
  dma-mapping: fix the kerneldoc for dma_map_sg_attrs