platform/kernel/linux-starfive.git
5 years agopowerpc/64: Adjust order in pcibios_init()
Sam Bobroff [Fri, 16 Aug 2019 04:48:05 +0000 (14:48 +1000)]
powerpc/64: Adjust order in pcibios_init()

The pcibios_init() function for PowerPC 64 currently calls
pci_bus_add_devices() before pcibios_resource_survey(). This means
that at boot time, when the pcibios_bus_add_device() hooks are called
by pci_bus_add_devices(), device resources have not been allocated and
they are unable to perform EEH setup, so a separate pass is needed.

This patch adjusts that order so that it will become possible to
consolidate the EEH setup work into a single location.

The only functional change is to execute pcibios_resource_survey()
(excepting ppc_md.pcibios_fixup(), see below) before
pci_bus_add_devices() instead of after it.

Because pcibios_scan_phb() and pci_bus_add_devices() are called
together in a loop, this must be broken into one loop for each call.
Then the call to pcibios_resource_survey() is moved up in between
them. This changes the ordering but because pcibios_resource_survey()
also calls ppc_md.pcibios_fixup(), that call is extracted out into
pcibios_init() to where pcibios_resource_survey() was, so that it is
not moved.

The only other caller of pcibios_resource_survey() is the PowerPC 32
version of pcibios_init(), and therefore, that is modified to call
ppc_md.pcibios_fixup() right after pcibios_resource_survey() so that
there is no functional change there at all.

The re-arrangement will cause very few side-effects because at this
stage in the boot, pci_bus_add_devices() does very little:
- pci_create_sysfs_dev_files() does nothing (no sysfs yet)
- pci_proc_attach_device() does nothing (no proc yet)
- device_attach() does nothing (no drivers yet)
This leaves only the pci_final_fixup calls, D3 support, and marking
the device as added. Of those, only the pci_final_fixup calls have the
potential to be affected by resource allocation.

The only pci_final_fixup handlers that touch resources seem to be one
for x86 (pci_amd_enable_64bit_bar()), and a PowerPC 32 platform driver
(quirk_final_uli1575()), neither of which use this pcibios_init()
function. Even if they did, it would almost certainly be a bug, under
the current ordering, to rely on or make changes to resources before
they were allocated.

Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4506b0489eabd0921a3587d90bd44c7683f3472d.1565930772.git.sbobroff@linux.ibm.com
5 years agopowerpc: remove meaningless KBUILD_ARFLAGS addition
Masahiro Yamada [Sat, 13 Jul 2019 03:21:06 +0000 (12:21 +0900)]
powerpc: remove meaningless KBUILD_ARFLAGS addition

The KBUILD_ARFLAGS addition in arch/powerpc/Makefile has never worked
in a useful way because it is always overridden by the following code
in the top Makefile:

  # use the deterministic mode of AR if available
  KBUILD_ARFLAGS := $(call ar-option,D)

The code in the top Makefile was added in 2011, by commit 40df759e2b9e
("kbuild: Fix build with binutils <= 2.19").

The KBUILD_ARFLAGS addition for ppc has always been dead code from the
beginning.

Nobody has reported a problem since 43c9127d94d6 ("powerpc: Add option
to use thin archives"), so this code was unneeded.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190713032106.8509-1-yamada.masahiro@socionext.com
5 years agopowerpc: add machine check safe copy_to_user
Santosh Sivaraj [Tue, 20 Aug 2019 08:13:52 +0000 (13:43 +0530)]
powerpc: add machine check safe copy_to_user

Use  memcpy_mcsafe() implementation to define copy_to_user_mcsafe()

Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820081352.8641-8-santosh@fossix.org
5 years agopowerpc/memcpy: Add memcpy_mcsafe for pmem
Balbir Singh [Tue, 20 Aug 2019 08:13:51 +0000 (13:43 +0530)]
powerpc/memcpy: Add memcpy_mcsafe for pmem

The pmem infrastructure uses memcpy_mcsafe in the pmem layer so as to
convert machine check exceptions into a return value on failure in case
a machine check exception is encountered during the memcpy. The return
value is the number of bytes remaining to be copied.

This patch largely borrows from the copyuser_power7 logic and does not add
the VMX optimizations, largely to keep the patch simple. If needed those
optimizations can be folded in.

Signed-off-by: Balbir Singh <bsingharora@gmail.com>
[arbab@linux.ibm.com: Added symbol export]
Co-developed-by: Santosh Sivaraj <santosh@fossix.org>
Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820081352.8641-7-santosh@fossix.org
5 years agopowerpc/mce: Handle UE event for memcpy_mcsafe
Balbir Singh [Tue, 20 Aug 2019 08:13:50 +0000 (13:43 +0530)]
powerpc/mce: Handle UE event for memcpy_mcsafe

If we take a UE on one of the instructions with a fixup entry, set nip
to continue execution at the fixup entry. Stop processing the event
further or print it.

Co-developed-by: Reza Arbab <arbab@linux.ibm.com>
Signed-off-by: Reza Arbab <arbab@linux.ibm.com>
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
Reviewed-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820081352.8641-6-santosh@fossix.org
5 years agoextable: Add function to search only kernel exception table
Santosh Sivaraj [Tue, 20 Aug 2019 08:13:49 +0000 (13:43 +0530)]
extable: Add function to search only kernel exception table

Certain architecture specific operating modes (e.g., in powerpc machine
check handler that is unable to access vmalloc memory), the
search_exception_tables cannot be called because it also searches the
module exception tables if entry is not found in the kernel exception
table.

Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820081352.8641-5-santosh@fossix.org
5 years agopowerpc/mce: Make machine_check_ue_event() static
Reza Arbab [Tue, 20 Aug 2019 08:13:48 +0000 (13:43 +0530)]
powerpc/mce: Make machine_check_ue_event() static

The function doesn't get used outside this file, so make it static.

Signed-off-by: Reza Arbab <arbab@linux.ibm.com>
Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820081352.8641-4-santosh@fossix.org
5 years agopowerpc/mce: Fix MCE handling for huge pages
Balbir Singh [Tue, 20 Aug 2019 08:13:47 +0000 (13:43 +0530)]
powerpc/mce: Fix MCE handling for huge pages

The current code would fail on huge pages addresses, since the shift would
be incorrect. Use the correct page shift value returned by
__find_linux_pte() to get the correct physical address. The code is more
generic and can handle both regular and compound pages.

Fixes: ba41e1e1ccb9 ("powerpc/mce: Hookup derror (load/store) UE errors")
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
[arbab@linux.ibm.com: Fixup pseries_do_memory_failure()]
Signed-off-by: Reza Arbab <arbab@linux.ibm.com>
Tested-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Cc: stable@vger.kernel.org # v4.15+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820081352.8641-3-santosh@fossix.org
5 years agopowerpc/mce: Schedule work from irq_work
Santosh Sivaraj [Tue, 20 Aug 2019 08:13:46 +0000 (13:43 +0530)]
powerpc/mce: Schedule work from irq_work

schedule_work() cannot be called from MCE exception context as MCE can
interrupt even in interrupt disabled context.

Fixes: 733e4a4c4467 ("powerpc/mce: hookup memory_failure for UE errors")
Cc: stable@vger.kernel.org # v4.15+
Reviewed-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820081352.8641-2-santosh@fossix.org
5 years agopowerpc/pseries/mobility: use cond_resched when updating device tree
Nathan Lynch [Fri, 2 Aug 2019 19:29:26 +0000 (14:29 -0500)]
powerpc/pseries/mobility: use cond_resched when updating device tree

After a partition migration, pseries_devicetree_update() processes
changes to the device tree communicated from the platform to
Linux. This is a relatively heavyweight operation, with multiple
device tree searches, memory allocations, and conversations with
partition firmware.

There's a few levels of nested loops which are bounded only by
decisions made by the platform, outside of Linux's control, and indeed
we have seen RCU stalls on large systems while executing this call
graph. Use cond_resched() in these loops so that the cpu is yielded
when needed.

Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802192926.19277-4-nathanl@linux.ibm.com
5 years agopowerpc/rtas: allow rescheduling while changing cpu states
Nathan Lynch [Fri, 2 Aug 2019 19:29:25 +0000 (14:29 -0500)]
powerpc/rtas: allow rescheduling while changing cpu states

rtas_cpu_state_change_mask() potentially operates on scores of cpus,
so explicitly allow rescheduling in the loop body.

Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802192926.19277-3-nathanl@linux.ibm.com
5 years agopowerpc/rtas: use device model APIs and serialization during LPM
Nathan Lynch [Fri, 2 Aug 2019 19:29:24 +0000 (14:29 -0500)]
powerpc/rtas: use device model APIs and serialization during LPM

The LPAR migration implementation and userspace-initiated cpu hotplug
can interleave their executions like so:

1. Set cpu 7 offline via sysfs.

2. Begin a partition migration, whose implementation requires the OS
   to ensure all present cpus are online; cpu 7 is onlined:

     rtas_ibm_suspend_me -> rtas_online_cpus_mask -> cpu_up

   This sets cpu 7 online in all respects except for the cpu's
   corresponding struct device; dev->offline remains true.

3. Set cpu 7 online via sysfs. _cpu_up() determines that cpu 7 is
   already online and returns success. The driver core (device_online)
   sets dev->offline = false.

4. The migration completes and restores cpu 7 to offline state:

     rtas_ibm_suspend_me -> rtas_offline_cpus_mask -> cpu_down

This leaves cpu7 in a state where the driver core considers the cpu
device online, but in all other respects it is offline and
unused. Attempts to online the cpu via sysfs appear to succeed but the
driver core actually does not pass the request to the lower-level
cpuhp support code. This makes the cpu unusable until the cpu device
is manually set offline and then online again via sysfs.

Instead of directly calling cpu_up/cpu_down, the migration code should
use the higher-level device core APIs to maintain consistent state and
serialize operations.

Fixes: 120496ac2d2d ("powerpc: Bring all threads online prior to migration/hibernation")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802192926.19277-2-nathanl@linux.ibm.com
5 years agopowerpc/603: Fix handling of the DIRTY flag
Christophe Leroy [Mon, 19 Aug 2019 06:40:25 +0000 (06:40 +0000)]
powerpc/603: Fix handling of the DIRTY flag

If a page is already mapped RW without the DIRTY flag, the DIRTY
flag is never set and a TLB store miss exception is taken forever.

This is easily reproduced with the following app:

void main(void)
{
volatile char *ptr = mmap(0, 4096, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0);

*ptr = *ptr;
}

When DIRTY flag is not set, bail out of TLB miss handler and take
a minor page fault which will set the DIRTY flag.

Fixes: f8b58c64eaef ("powerpc/603: let's handle PAGE_DIRTY directly")
Cc: stable@vger.kernel.org # v5.1+
Reported-by: Doug Crawford <doug.crawford@intelight-its.com>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/80432f71194d7ee75b2f5043ecf1501cf1cca1f3.1566196646.git.christophe.leroy@c-s.fr
5 years agopowerpc/64s/radix: Remove redundant pfn_pte bitop, add VM_BUG_ON
Nicholas Piggin [Wed, 24 Jul 2019 08:46:38 +0000 (18:46 +1000)]
powerpc/64s/radix: Remove redundant pfn_pte bitop, add VM_BUG_ON

pfn_pte is never given a pte above the addressable physical memory
limit, so the masking is redundant. In case of a software bug, it
is not obviously better to silently truncate the pfn than to corrupt
the pte (either one will result in memory corruption or crashes),
so there is no reason to add this to the fast path.

Add VM_BUG_ON to catch cases where the pfn is invalid. These would
catch the create_section_mapping bug fixed by a previous commit.

  [16885.256466] ------------[ cut here ]------------
  [16885.256492] kernel BUG at arch/powerpc/include/asm/book3s/64/pgtable.h:612!
  cpu 0x0: Vector: 700 (Program Check) at [c0000000ee0a36d0]
      pc: c000000000080738: __map_kernel_page+0x248/0x6f0
      lr: c000000000080ac0: __map_kernel_page+0x5d0/0x6f0
      sp: c0000000ee0a3960
     msr: 9000000000029033
    current = 0xc0000000ec63b400
    paca    = 0xc0000000017f0000   irqmask: 0x03   irq_happened: 0x01
      pid   = 85, comm = sh
  kernel BUG at arch/powerpc/include/asm/book3s/64/pgtable.h:612!
  Linux version 5.3.0-rc1-00001-g0fe93e5f3394
  enter ? for help
  [c0000000ee0a3a00c000000000d37378 create_physical_mapping+0x260/0x360
  [c0000000ee0a3b10c000000000d370bc create_section_mapping+0x1c/0x3c
  [c0000000ee0a3b30c000000000071f54 arch_add_memory+0x74/0x130

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190724084638.24982-5-npiggin@gmail.com
5 years agopowerpc/64: Add VIRTUAL_BUG_ON checks for __va and __pa addresses
Nicholas Piggin [Wed, 24 Jul 2019 08:46:37 +0000 (18:46 +1000)]
powerpc/64: Add VIRTUAL_BUG_ON checks for __va and __pa addresses

Ensure __va is given a physical address below PAGE_OFFSET, and __pa is
given a virtual address above PAGE_OFFSET.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190724084638.24982-4-npiggin@gmail.com
5 years agopowerpc/perf: fix imc allocation failure handling
Nicholas Piggin [Wed, 24 Jul 2019 08:46:36 +0000 (18:46 +1000)]
powerpc/perf: fix imc allocation failure handling

The alloc_pages_node return value should be tested for failure
before being passed to page_address.

Tested-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190724084638.24982-3-npiggin@gmail.com
5 years agopowerpc/64s/radix: Fix memory hot-unplug page table split
Nicholas Piggin [Wed, 24 Jul 2019 08:46:35 +0000 (18:46 +1000)]
powerpc/64s/radix: Fix memory hot-unplug page table split

create_physical_mapping expects physical addresses, but splitting
these mapping on hot unplug is supplying virtual (effective)
addresses.

Fixes: 4dd5f8a99e791 ("powerpc/mm/radix: Split linear mapping on hot-unplug")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190724084638.24982-2-npiggin@gmail.com
5 years agopowerpc/64s/radix: Fix memory hotplug section page table creation
Nicholas Piggin [Wed, 24 Jul 2019 08:46:34 +0000 (18:46 +1000)]
powerpc/64s/radix: Fix memory hotplug section page table creation

create_physical_mapping expects physical addresses, but creating and
splitting these mappings after boot is supplying virtual (effective)
addresses. This can be irritated by booting with mem= to limit memory
then probing an unused physical memory range:

  echo <addr> > /sys/devices/system/memory/probe

This mostly works by accident, firstly because __va(__va(x)) == __va(x)
so the virtual address does not get corrupted. Secondly because pfn_pte
masks out the upper bits of the pfn beyond the physical address limit,
so a pfn constructed with a 0xc000000000000000 virtual linear address
will be masked back to the correct physical address in the pte.

Fixes: 6cc27341b21a8 ("powerpc/mm: add radix__create_section_mapping()")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190724084638.24982-1-npiggin@gmail.com
5 years agopowerpc/64: allow compiler to cache 'current'
Nicholas Piggin [Wed, 12 Jun 2019 14:03:17 +0000 (00:03 +1000)]
powerpc/64: allow compiler to cache 'current'

current may be cached by the compiler, so remove the volatile asm
restriction. This results in better generated code, as well as being
smaller and fewer dependent loads, it can avoid store-hit-load flushes
like this one that shows up in irq_exit():

    preempt_count_sub(HARDIRQ_OFFSET);
    if (!in_interrupt() && ...)

Which ends up as:

    ((struct thread_info *)current)->preempt_count -= HARDIRQ_OFFSET;
    if (((struct thread_info *)current)->preempt_count ...

Evaluating current twice presently means it has to be loaded twice, and
here gcc happens to pick a different register each time, then
preempt_count is accessed via that base register:

    1058:       ld      r10,2392(r13)     <-- current
    105c:       lwz     r9,0(r10)         <-- preempt_count
    1060:       addis   r9,r9,-1
    1064:       stw     r9,0(r10)         <-- preempt_count
    1068:       ld      r9,2392(r13)      <-- current
    106c:       lwz     r9,0(r9)          <-- preempt_count
    1070:       rlwinm. r9,r9,0,11,23
    1074:       bne     1090 <irq_exit+0x60>

This can frustrate store-hit-load detection heuristics and cause
flushes. Allowing the compiler to cache current in a reigster with this
patch results in the same base register being used for all accesses,
which is more likely to be detected as an alias:

    1058:       ld      r31,2392(r13)
    ...
    1070:       lwz     r9,0(r31)
    1074:       addis   r9,r9,-1
    1078:       stw     r9,0(r31)
    107c:       lwz     r9,0(r31)
    1080:       rlwinm. r9,r9,0,11,23
    1084:       bne     10a0 <irq_exit+0x60>

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190612140317.24490-1-npiggin@gmail.com
5 years agopowerpc/32: Add warning on misaligned copy_page() or clear_page()
Christophe Leroy [Fri, 16 Aug 2019 07:52:20 +0000 (07:52 +0000)]
powerpc/32: Add warning on misaligned copy_page() or clear_page()

copy_page() and clear_page() expect page aligned destination, and
use dcbz instruction to clear entire cache lines based on the
assumption that the destination is cache aligned.

As shown during analysis of a bug in BTRFS filesystem, a misaligned
copy_page() can create bugs that are difficult to locate (see Link).

Add an explicit WARNING when copy_page() or clear_page() are called
with misaligned destination.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=204371
Link: https://lore.kernel.org/r/c6cea38f90480268d439ca44a645647e260fff09.1565941808.git.christophe.leroy@c-s.fr
5 years agopowerpc/mm: ppc 603 doesn't need update_mmu_cache()
Christophe Leroy [Fri, 16 Aug 2019 05:41:44 +0000 (05:41 +0000)]
powerpc/mm: ppc 603 doesn't need update_mmu_cache()

On powerpc 603, there is no hash table so get out of
update_mmu_cache() early.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/6133e0f115d955fac4061536dab0fa7480a1c433.1565933217.git.christophe.leroy@c-s.fr
5 years agopowerpc/mm: Simplify update_mmu_cache() on BOOK3S32
Christophe Leroy [Fri, 16 Aug 2019 05:41:43 +0000 (05:41 +0000)]
powerpc/mm: Simplify update_mmu_cache() on BOOK3S32

On BOOK3S32, hash_preload() neither use is_exec nor trap,
so drop those parameters and simplify update_mmu_cached().

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/35f143c6fe29f9fd25c7f3cd4448ae401029ce3c.1565933217.git.christophe.leroy@c-s.fr
5 years agopowerpc/mm: move update_mmu_cache() into book3s hash utils.
Christophe Leroy [Fri, 16 Aug 2019 05:41:42 +0000 (05:41 +0000)]
powerpc/mm: move update_mmu_cache() into book3s hash utils.

update_mmu_cache() is only for BOOK3S, and can be simplified for
BOOK3S32.

Move it out of mem.c into respective BOOK3S32 and BOOK3S64 files
containing hash utils.

BOOK3S64 version of hash_preload() is only used locally, declare it
static.

Remove the radix_enabled() stuff in BOOK3S32 version.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/107aaf43583a5f5d09e0d4e84c4c4390ecfcd512.1565933217.git.christophe.leroy@c-s.fr
5 years agopowerpc/mm: move FSL_BOOK3 version of update_mmu_cache()
Christophe Leroy [Fri, 16 Aug 2019 05:41:41 +0000 (05:41 +0000)]
powerpc/mm: move FSL_BOOK3 version of update_mmu_cache()

Move FSL_BOOK3E version of update_mmu_cache() at the same
place as book3e_hugetlb_preload() as update_mmu_cache() is
the only user of book3e_hugetlb_preload().

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4d69fdc86df9c74adc71a60331a86f6afb8b5e9e.1565933217.git.christophe.leroy@c-s.fr
5 years agopowerpc/mm: define empty update_mmu_cache() as static inline
Christophe Leroy [Fri, 16 Aug 2019 05:41:40 +0000 (05:41 +0000)]
powerpc/mm: define empty update_mmu_cache() as static inline

Only BOOK3S and FSL_BOOK3E have a usefull update_mmu_cache().

For the others, just define it static inline.

In the meantime, simplify the FSL_BOOK3E related ifdef as
book3e_hugetlb_preload() only exists when CONFIG_PPC_FSL_BOOK3E
is selected.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/668aba4db6b9af6d8a151174e11a4289f1a6bbcd.1565933217.git.christophe.leroy@c-s.fr
5 years agopowerpc/mm: don't display empty early ioremap area
Christophe Leroy [Wed, 14 Aug 2019 14:36:10 +0000 (14:36 +0000)]
powerpc/mm: don't display empty early ioremap area

On the 8xx, the layout displayed at boot is:

[    0.000000] Memory: 121856K/131072K available (5728K kernel code, 592K rwdata, 1248K rodata, 560K init, 448K bss, 9216K reserved, 0K cma-reserved)
[    0.000000] Kernel virtual memory layout:
[    0.000000]   * 0xffefc000..0xffffc000  : fixmap
[    0.000000]   * 0xffefc000..0xffefc000  : early ioremap
[    0.000000]   * 0xc9000000..0xffefc000  : vmalloc & ioremap
[    0.000000] SLUB: HWalign=16, Order=0-3, MinObjects=0, CPUs=1, Nodes=1

Remove display of an empty early ioremap.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f6267226038cb25a839b567319e240576e3f8565.1565793287.git.christophe.leroy@c-s.fr
5 years agopowerpc/32s: Fix boot failure with DEBUG_PAGEALLOC without KASAN.
Christophe Leroy [Wed, 14 Aug 2019 10:02:20 +0000 (10:02 +0000)]
powerpc/32s: Fix boot failure with DEBUG_PAGEALLOC without KASAN.

When KASAN is selected, the definitive hash table has to be
set up later, but there is already an early temporary one.

When KASAN is not selected, there is no early hash table,
so the setup of the definitive hash table cannot be delayed.

Fixes: 72f208c6a8f7 ("powerpc/32s: move hash code patching out of MMU_init_hw()")
Cc: stable@vger.kernel.org # v5.2+
Reported-by: Jonathan Neuschafer <j.neuschaefer@gmx.net>
Tested-by: Jonathan Neuschafer <j.neuschaefer@gmx.net>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b7860c5e1e784d6b96ba67edf47dd6cbc2e78ab6.1565776892.git.christophe.leroy@c-s.fr
5 years agopowerpc/futex: Fix warning: 'oldval' may be used uninitialized in this function
Christophe Leroy [Wed, 14 Aug 2019 09:25:52 +0000 (09:25 +0000)]
powerpc/futex: Fix warning: 'oldval' may be used uninitialized in this function

We see warnings such as:
  kernel/futex.c: In function 'do_futex':
  kernel/futex.c:1676:17: warning: 'oldval' may be used uninitialized in this function [-Wmaybe-uninitialized]
     return oldval == cmparg;
                   ^
  kernel/futex.c:1651:6: note: 'oldval' was declared here
    int oldval, ret;
        ^

This is because arch_futex_atomic_op_inuser() only sets *oval if ret
is 0 and GCC doesn't see that it will only use it when ret is 0.

Anyway, the non-zero ret path is an error path that won't suffer from
setting *oval, and as *oval is a local var in futex_atomic_op_inuser()
it will have no impact.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: reword change log slightly]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/86b72f0c134367b214910b27b9a6dd3321af93bb.1565774657.git.christophe.leroy@c-s.fr
5 years agopowerpc/kasan: Fix shadow area set up for modules.
Christophe Leroy [Fri, 9 Aug 2019 14:58:10 +0000 (14:58 +0000)]
powerpc/kasan: Fix shadow area set up for modules.

When loading modules, from time to time an Oops is encountered during
the init of shadow area for globals. This is due to the last page not
always being mapped depending on the exact distance between the start
and the end of the shadow area and the alignment with the page
addresses.

Fix this by aligning the starting address with the page address.

Fixes: 2edb16efc899 ("powerpc/32: Add KASAN support")
Cc: stable@vger.kernel.org # v5.2+
Reported-by: Erhard F. <erhard_f@mailbox.org>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4f887e9b77d0d725cbb52035c7ece485c1c5fc14.1565361881.git.christophe.leroy@c-s.fr
5 years agopowerpc/kasan: Fix parallel loading of modules.
Christophe Leroy [Fri, 9 Aug 2019 14:58:09 +0000 (14:58 +0000)]
powerpc/kasan: Fix parallel loading of modules.

Parallel loading of modules may lead to bad setup of shadow page table
entries.

First, lets align modules so that two modules never share the same
shadow page.

Second, ensure that two modules cannot allocate two page tables for
the same PMD entry at the same time. This is done by using
init_mm.page_table_lock in the same way as __pte_alloc_kernel()

Fixes: 2edb16efc899 ("powerpc/32: Add KASAN support")
Cc: stable@vger.kernel.org # v5.2+
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c97284f912128cbc3f2fe09d68e90e65fb3e6026.1565361876.git.christophe.leroy@c-s.fr
5 years agopowerpc/hw_breakpoint: move instruction stepping out of hw_breakpoint_handler()
Christophe Leroy [Fri, 28 Jun 2019 15:55:52 +0000 (15:55 +0000)]
powerpc/hw_breakpoint: move instruction stepping out of hw_breakpoint_handler()

On 8xx, breakpoints stop after executing the instruction, so
stepping/emulation is not needed. Move it into a sub-function and
remove the #ifdefs.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f8cdc3f1c66ad3c43ebc568abcc6c39ed4676284.1561737231.git.christophe.leroy@c-s.fr
5 years agopowerpc/ptdump: drop non vital #ifdefs
Christophe Leroy [Wed, 14 Aug 2019 12:36:13 +0000 (12:36 +0000)]
powerpc/ptdump: drop non vital #ifdefs

hashpagetable.c is only compiled when CONFIG_PPC_BOOK3S_64 is
defined, so drop the test and its 'else' branch.

Use IS_ENABLED(CONFIG_PPC_PSERIES) instead of #ifdef, this allows the
code to be checked at any build. It is still optimised out by GCC.

Use IS_ENABLED(CONFIG_PPC_64K_PAGES) instead of #ifdef.

Use IS_ENABLED(CONFIG_SPARSEMEN_VMEMMAP) instead of #ifdef.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c8998ed32e4e3954b56a8dacecfe43319a2a0483.1565786091.git.christophe.leroy@c-s.fr
5 years agopowerpc/ptdump: get out of note_prot_wx() when CONFIG_PPC_DEBUG_WX is not selected.
Christophe Leroy [Wed, 14 Aug 2019 12:36:12 +0000 (12:36 +0000)]
powerpc/ptdump: get out of note_prot_wx() when CONFIG_PPC_DEBUG_WX is not selected.

When CONFIG_PPC_DEBUG_WX, note_prot_wx() is useless.

Get out of it early and inconditionnally in that case,
so that GCC can kick all the code out.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ff6c8f631bd4ce3a10e0cc241eb569816187bc20.1565786091.git.christophe.leroy@c-s.fr
5 years agopowerpc/ptdump: drop dummy KERN_VIRT_START on PPC32
Christophe Leroy [Wed, 14 Aug 2019 12:36:11 +0000 (12:36 +0000)]
powerpc/ptdump: drop dummy KERN_VIRT_START on PPC32

PPC32 doesn't have KERN_VIRT_START. Make PAGE_OFFSET the
default starting address for the dump, and drop the dummy
definition of KERN_VIRT_START. Only use KERN_VIRT_START for
non radix PPC64.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/334632b1df4775b0ccf3bdc8d6b201d14e3daedd.1565786091.git.christophe.leroy@c-s.fr
5 years agopowerpc/ptdump: fix walk_pagetables() address mismatch
Christophe Leroy [Wed, 14 Aug 2019 12:36:10 +0000 (12:36 +0000)]
powerpc/ptdump: fix walk_pagetables() address mismatch

walk_pagetables() always walk the entire pgdir from address 0
but considers PAGE_OFFSET or KERN_VIRT_START as the starting
address of the walk, resulting in a possible mismatch in the
displayed addresses.

Ex: on PPC32, when KERN_VIRT_START was locally defined as
PAGE_OFFSET, ptdump displayed 0x80000000
instead of 0xc0000000 for the first kernel page,
because 0xc0000000 + 0xc0000000 = 0x80000000

Start the walk at st->start_address instead of starting at 0.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5aa2ac513295f594cce8ddb1c649f61947bd063d.1565786091.git.christophe.leroy@c-s.fr
5 years agopowerpc/ptdump: Fix addresses display on PPC32
Christophe Leroy [Wed, 14 Aug 2019 12:36:09 +0000 (12:36 +0000)]
powerpc/ptdump: Fix addresses display on PPC32

Commit 453d87f6a8ae ("powerpc/mm: Warn if W+X pages found on boot")
wrongly changed KERN_VIRT_START from 0 to PAGE_OFFSET, leading to a
shift in the displayed addresses.

Lets revert that change to resync walk_pagetables()'s addr val and
pgd_t pointer for PPC32.

Fixes: 453d87f6a8ae ("powerpc/mm: Warn if W+X pages found on boot")
Cc: stable@vger.kernel.org # v5.2+
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/eb4d626514e22f85814830012642329018ef6af9.1565786091.git.christophe.leroy@c-s.fr
5 years agopowerpc/xmon: Add a dump of all XIVE interrupts
Cédric Le Goater [Wed, 14 Aug 2019 15:47:54 +0000 (17:47 +0200)]
powerpc/xmon: Add a dump of all XIVE interrupts

Modify the xmon 'dxi' command to query all interrupts if no IRQ number
is specified.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190814154754.23682-4-clg@kaod.org
5 years agopowerpc/xive: Fix dump of XIVE interrupt under pseries
Cédric Le Goater [Wed, 14 Aug 2019 15:47:53 +0000 (17:47 +0200)]
powerpc/xive: Fix dump of XIVE interrupt under pseries

The xmon 'dxi' command calls OPAL to query the XIVE configuration of a
interrupt. This can only be done on baremetal (PowerNV) and it will
crash a pseries machine.

Introduce a new XIVE get_irq_config() operation which implements a
different query depending on the platform, PowerNV or pseries, and
modify xmon to use a top level wrapper.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190814154754.23682-3-clg@kaod.org
5 years agopowerpc/xmon: Check for HV mode when dumping XIVE info from OPAL
Cédric Le Goater [Wed, 14 Aug 2019 15:47:52 +0000 (17:47 +0200)]
powerpc/xmon: Check for HV mode when dumping XIVE info from OPAL

Currently, the xmon 'dx' command calls OPAL to dump the XIVE state in
the OPAL logs and also outputs some of the fields of the internal XIVE
structures in Linux. The OPAL calls can only be done on baremetal
(PowerNV) and they crash a pseries machine. Fix by checking the
hypervisor feature of the CPU.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190814154754.23682-2-clg@kaod.org
5 years agopowerpc/powernv/ioda2: Create bigger default window with 64k IOMMU pages
Alexey Kardashevskiy [Thu, 18 Jul 2019 05:11:39 +0000 (15:11 +1000)]
powerpc/powernv/ioda2: Create bigger default window with 64k IOMMU pages

At the moment we create a small window only for 32bit devices, the window
maps 0..2GB of the PCI space only. For other devices we either use
a sketchy bypass or hardware bypass but the former can only work if
the amount of RAM is no bigger than the device's DMA mask and the latter
requires devices to support at least 59bit DMA.

This extends the default DMA window to the maximum size possible to allow
a wider DMA mask than just 32bit. The default window size is now limited
by the the iommu_table::it_map allocation bitmap which is a contiguous
array, 1 bit per an IOMMU page.

This increases the default IOMMU page size from hard coded 4K to
the system page size to allow wider DMA masks.

This increases the level number to not exceed the max order allocation
limit per TCE level. By the same time, this keeps minimal levels number
as 2 in order to save memory.

As the extended window now overlaps the 32bit MMIO region, this adds
an area reservation to iommu_init_table().

After this change the default window size is 0x80000000000==1<<43 so
devices limited to DMA mask smaller than the amount of system RAM can
still use more than just 2GB of memory for DMA.

This is an optimization and not a bug fix for DMA API usage.

With the on-demand allocation of indirect TCE table levels enabled and
2 levels, the first TCE level size is just
1<<ceil((log2(0x7ffffffffff+1)-16)/2)=16384 TCEs or 2 system pages.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190718051139.74787-5-aik@ozlabs.ru
5 years agopowerpc/powernv/ioda2: Allocate TCE table levels on demand for default DMA window
Alexey Kardashevskiy [Thu, 18 Jul 2019 05:11:38 +0000 (15:11 +1000)]
powerpc/powernv/ioda2: Allocate TCE table levels on demand for default DMA window

We allocate only the first level of multilevel TCE tables for KVM
already (alloc_userspace_copy==true), and the rest is allocated on demand.
This is not enabled though for bare metal.

This removes the KVM limitation (implicit, via the alloc_userspace_copy
parameter) and always allocates just the first level. The on-demand
allocation of missing levels is already implemented.

As from now on DMA map might happen with disabled interrupts, this
allocates TCEs with GFP_ATOMIC; otherwise lockdep reports errors 1].
In practice just a single page is allocated there so chances for failure
are quite low.

To save time when creating a new clean table, this skips non-allocated
indirect TCE entries in pnv_tce_free just like we already do in
the VFIO IOMMU TCE driver.

This changes the default level number from 1 to 2 to reduce the amount
of memory required for the default 32bit DMA window at the boot time.
The default window size is up to 2GB which requires 4MB of TCEs which is
unlikely to be used entirely or at all as most devices these days are
64bit capable so by switching to 2 levels by default we save 4032KB of
RAM per a device.

While at this, add __GFP_NOWARN to alloc_pages_node() as the userspace
can trigger this path via VFIO, see the failure and try creating a table
again with different parameters which might succeed.

[1]:
===
BUG: sleeping function called from invalid context at mm/page_alloc.c:4596
in_atomic(): 1, irqs_disabled(): 1, pid: 1038, name: scsi_eh_1
2 locks held by scsi_eh_1/1038:
 #0: 000000005efd659a (&host->eh_mutex){+.+.}, at: ata_eh_acquire+0x34/0x80
 #1: 0000000006cf56a6 (&(&host->lock)->rlock){....}, at: ata_exec_internal_sg+0xb0/0x5c0
irq event stamp: 500
hardirqs last  enabled at (499): [<c000000000cb8a74>] _raw_spin_unlock_irqrestore+0x94/0xd0
hardirqs last disabled at (500): [<c000000000cb85c4>] _raw_spin_lock_irqsave+0x44/0x120
softirqs last  enabled at (0): [<c000000000101120>] copy_process.isra.4.part.5+0x640/0x1a80
softirqs last disabled at (0): [<0000000000000000>] 0x0
CPU: 73 PID: 1038 Comm: scsi_eh_1 Not tainted 5.2.0-rc6-le_nv2_aikATfstn1-p1 #634
Call Trace:
[c000003d064cef50] [c000000000c8e6c4] dump_stack+0xe8/0x164 (unreliable)
[c000003d064cefa0] [c00000000014ed78] ___might_sleep+0x2f8/0x310
[c000003d064cf020] [c0000000003ca084] __alloc_pages_nodemask+0x2a4/0x1560
[c000003d064cf220] [c0000000000c2530] pnv_alloc_tce_level.isra.0+0x90/0x130
[c000003d064cf290] [c0000000000c2888] pnv_tce+0x128/0x3b0
[c000003d064cf360] [c0000000000c2c00] pnv_tce_build+0xb0/0xf0
[c000003d064cf3c0] [c0000000000bbd9c] pnv_ioda2_tce_build+0x3c/0xb0
[c000003d064cf400] [c00000000004cfe0] ppc_iommu_map_sg+0x210/0x550
[c000003d064cf510] [c00000000004b7a4] dma_iommu_map_sg+0x74/0xb0
[c000003d064cf530] [c000000000863944] ata_qc_issue+0x134/0x470
[c000003d064cf5b0] [c000000000863ec4] ata_exec_internal_sg+0x244/0x5c0
[c000003d064cf700] [c0000000008642d0] ata_exec_internal+0x90/0xe0
[c000003d064cf780] [c0000000008650ac] ata_dev_read_id+0x2ec/0x640
[c000003d064cf8d0] [c000000000878e28] ata_eh_recover+0x948/0x16d0
[c000003d064cfa10] [c00000000087d760] sata_pmp_error_handler+0x480/0xbf0
[c000003d064cfbc0] [c000000000884624] ahci_error_handler+0x74/0xe0
[c000003d064cfbf0] [c000000000879fa8] ata_scsi_port_error_handler+0x2d8/0x7c0
[c000003d064cfca0] [c00000000087a544] ata_scsi_error+0xb4/0x100
[c000003d064cfd00] [c000000000802450] scsi_error_handler+0x120/0x510
[c000003d064cfdb0] [c000000000140c48] kthread+0x1b8/0x1c0
[c000003d064cfe20] [c00000000000bd8c] ret_from_kernel_thread+0x5c/0x70
ata1: SATA link up 6.0 Gbps (SStatus 133 SControl 300)
irq event stamp: 2305

========================================================
hardirqs last  enabled at (2305): [<c00000000000e4c8>] fast_exc_return_irq+0x28/0x34
hardirqs last disabled at (2303): [<c000000000cb9fd0>] __do_softirq+0x4a0/0x654
WARNING: possible irq lock inversion dependency detected
5.2.0-rc6-le_nv2_aikATfstn1-p1 #634 Tainted: G        W
softirqs last  enabled at (2304): [<c000000000cba054>] __do_softirq+0x524/0x654
softirqs last disabled at (2297): [<c00000000010f278>] irq_exit+0x128/0x180
--------------------------------------------------------
swapper/0/0 just changed the state of lock:
0000000006cf56a6 (&(&host->lock)->rlock){-...}, at: ahci_single_level_irq_intr+0xac/0x120
but this lock took another, HARDIRQ-unsafe lock in the past:
 (fs_reclaim){+.+.}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
 Possible interrupt unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(fs_reclaim);
                               local_irq_disable();
                               lock(&(&host->lock)->rlock);
                               lock(fs_reclaim);
  <Interrupt>
    lock(&(&host->lock)->rlock);

 *** DEADLOCK ***

no locks held by swapper/0/0.

the shortest dependencies between 2nd lock and 1st lock:
 -> (fs_reclaim){+.+.} ops: 167579 {
    HARDIRQ-ON-W at:
                      lock_acquire+0xf8/0x2a0
                      fs_reclaim_acquire.part.23+0x44/0x60
                      kmem_cache_alloc_node_trace+0x80/0x590
                      alloc_desc+0x64/0x270
                      __irq_alloc_descs+0x2e4/0x3a0
                      irq_domain_alloc_descs+0xb0/0x150
                      irq_create_mapping+0x168/0x2c0
                      xics_smp_probe+0x2c/0x98
                      pnv_smp_probe+0x40/0x9c
                      smp_prepare_cpus+0x524/0x6c4
                      kernel_init_freeable+0x1b4/0x650
                      kernel_init+0x2c/0x148
                      ret_from_kernel_thread+0x5c/0x70
    SOFTIRQ-ON-W at:
                      lock_acquire+0xf8/0x2a0
                      fs_reclaim_acquire.part.23+0x44/0x60
                      kmem_cache_alloc_node_trace+0x80/0x590
                      alloc_desc+0x64/0x270
                      __irq_alloc_descs+0x2e4/0x3a0
                      irq_domain_alloc_descs+0xb0/0x150
                      irq_create_mapping+0x168/0x2c0
                      xics_smp_probe+0x2c/0x98
                      pnv_smp_probe+0x40/0x9c
                      smp_prepare_cpus+0x524/0x6c4
                      kernel_init_freeable+0x1b4/0x650
                      kernel_init+0x2c/0x148
                      ret_from_kernel_thread+0x5c/0x70
    INITIAL USE at:
                     lock_acquire+0xf8/0x2a0
                     fs_reclaim_acquire.part.23+0x44/0x60
                     kmem_cache_alloc_node_trace+0x80/0x590
                     alloc_desc+0x64/0x270
                     __irq_alloc_descs+0x2e4/0x3a0
                     irq_domain_alloc_descs+0xb0/0x150
                     irq_create_mapping+0x168/0x2c0
                     xics_smp_probe+0x2c/0x98
                     pnv_smp_probe+0x40/0x9c
                     smp_prepare_cpus+0x524/0x6c4
                     kernel_init_freeable+0x1b4/0x650
                     kernel_init+0x2c/0x148
                     ret_from_kernel_thread+0x5c/0x70
  }
===

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190718051139.74787-4-aik@ozlabs.ru
5 years agopowerpc/iommu: Allow bypass-only for DMA
Alexey Kardashevskiy [Thu, 18 Jul 2019 05:11:37 +0000 (15:11 +1000)]
powerpc/iommu: Allow bypass-only for DMA

POWER8 and newer support a bypass mode which maps all host memory to
PCI buses so an IOMMU table is not always required. However if we fail to
create such a table, the DMA setup fails and the kernel does not boot.

This skips the 32bit DMA setup check if the bypass is selected.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190718051139.74787-3-aik@ozlabs.ru
5 years agopowerpc/powernv/ioda: Fix race in TCE level allocation
Alexey Kardashevskiy [Thu, 18 Jul 2019 05:11:36 +0000 (15:11 +1000)]
powerpc/powernv/ioda: Fix race in TCE level allocation

pnv_tce() returns a pointer to a TCE entry and originally a TCE table
would be pre-allocated. For the default case of 2GB window the table
needs only a single level and that is fine. However if more levels are
requested, it is possible to get a race when 2 threads want a pointer
to a TCE entry from the same page of TCEs.

This adds cmpxchg to handle the race. Note that once TCE is non-zero,
it cannot become zero again.

Fixes: a68bd1267b72 ("powerpc/powernv/ioda: Allocate indirect TCE levels on demand")
CC: stable@vger.kernel.org # v4.19+
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190718051139.74787-2-aik@ozlabs.ru
5 years agopowerpc/pseries: Fix cpu_hotplug_lock acquisition in resize_hpt()
Gautham R. Shenoy [Wed, 15 May 2019 07:45:52 +0000 (13:15 +0530)]
powerpc/pseries: Fix cpu_hotplug_lock acquisition in resize_hpt()

The calls to arch_add_memory()/arch_remove_memory() are always made
with the read-side cpu_hotplug_lock acquired via memory_hotplug_begin().
On pSeries, arch_add_memory()/arch_remove_memory() eventually call
resize_hpt() which in turn calls stop_machine() which acquires the
read-side cpu_hotplug_lock again, thereby resulting in the recursive
acquisition of this lock.

In the absence of CONFIG_PROVE_LOCKING, we hadn't observed a system
lockup during a memory hotplug operation because cpus_read_lock() is a
per-cpu rwsem read, which, in the fast-path (in the absence of the
writer, which in our case is a CPU-hotplug operation) simply
increments the read_count on the semaphore. Thus a recursive read in
the fast-path doesn't cause any problems.

However, we can hit this problem in practice if there is a concurrent
CPU-Hotplug operation in progress which is waiting to acquire the
write-side of the lock. This will cause the second recursive read to
block until the writer finishes. While the writer is blocked since the
first read holds the lock. Thus both the reader as well as the writers
fail to make any progress thereby blocking both CPU-Hotplug as well as
Memory Hotplug operations.

Memory-Hotplug CPU-Hotplug
CPU 0 CPU 1
------                                  ------

1. down_read(cpu_hotplug_lock.rw_sem)
   [memory_hotplug_begin]
2. down_write(cpu_hotplug_lock.rw_sem)
[cpu_up/cpu_down]
3. down_read(cpu_hotplug_lock.rw_sem)
   [stop_machine()]

Lockdep complains as follows in these code-paths.

 swapper/0/1 is trying to acquire lock:
 (____ptrval____) (cpu_hotplug_lock.rw_sem){++++}, at: stop_machine+0x2c/0x60

but task is already holding lock:
(____ptrval____) (cpu_hotplug_lock.rw_sem){++++}, at: mem_hotplug_begin+0x20/0x50

 other info that might help us debug this:
  Possible unsafe locking scenario:

        CPU0
        ----
   lock(cpu_hotplug_lock.rw_sem);
   lock(cpu_hotplug_lock.rw_sem);

  *** DEADLOCK ***

  May be due to missing lock nesting notation

 3 locks held by swapper/0/1:
  #0: (____ptrval____) (&dev->mutex){....}, at: __driver_attach+0x12c/0x1b0
  #1: (____ptrval____) (cpu_hotplug_lock.rw_sem){++++}, at: mem_hotplug_begin+0x20/0x50
  #2: (____ptrval____) (mem_hotplug_lock.rw_sem){++++}, at: percpu_down_write+0x54/0x1a0

stack backtrace:
 CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.0.0-rc5-58373-gbc99402235f3-dirty #166
 Call Trace:
   dump_stack+0xe8/0x164 (unreliable)
   __lock_acquire+0x1110/0x1c70
   lock_acquire+0x240/0x290
   cpus_read_lock+0x64/0xf0
   stop_machine+0x2c/0x60
   pseries_lpar_resize_hpt+0x19c/0x2c0
   resize_hpt_for_hotplug+0x70/0xd0
   arch_add_memory+0x58/0xfc
   devm_memremap_pages+0x5e8/0x8f0
   pmem_attach_disk+0x764/0x830
   nvdimm_bus_probe+0x118/0x240
   really_probe+0x230/0x4b0
   driver_probe_device+0x16c/0x1e0
   __driver_attach+0x148/0x1b0
   bus_for_each_dev+0x90/0x130
   driver_attach+0x34/0x50
   bus_add_driver+0x1a8/0x360
   driver_register+0x108/0x170
   __nd_driver_register+0xd0/0xf0
   nd_pmem_driver_init+0x34/0x48
   do_one_initcall+0x1e0/0x45c
   kernel_init_freeable+0x540/0x64c
   kernel_init+0x2c/0x160
   ret_from_kernel_thread+0x5c/0x68

Fix this issue by
  1) Requiring all the calls to pseries_lpar_resize_hpt() be made
     with cpu_hotplug_lock held.

  2) In pseries_lpar_resize_hpt() invoke stop_machine_cpuslocked()
     as a consequence of 1)

  3) To satisfy 1), in hpt_order_set(), call mmu_hash_ops.resize_hpt()
     with cpu_hotplug_lock held.

Fixes: dbcf929c0062 ("powerpc/pseries: Add support for hash table resizing")
Cc: stable@vger.kernel.org # v4.11+
Reported-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1557906352-29048-1-git-send-email-ego@linux.vnet.ibm.com
5 years agoMerge branch 'topic/ppc-kvm' into next
Michael Ellerman [Mon, 19 Aug 2019 03:19:43 +0000 (13:19 +1000)]
Merge branch 'topic/ppc-kvm' into next

Merge our ppc-kvm topic branch. This contains several fixes for the XIVE
interrupt controller that we are sharing with the KVM tree.

5 years agoMerge branch 'fixes' into next
Michael Ellerman [Mon, 19 Aug 2019 02:43:13 +0000 (12:43 +1000)]
Merge branch 'fixes' into next

Merge in our fixes branch, which brings in clone3() as well as some
implicit fallthrough fixes we want in next.

5 years agopowerpc/xive: Implement get_irqchip_state method for XIVE to fix shutdown race
Paul Mackerras [Tue, 13 Aug 2019 10:06:48 +0000 (20:06 +1000)]
powerpc/xive: Implement get_irqchip_state method for XIVE to fix shutdown race

Testing has revealed the existence of a race condition where a XIVE
interrupt being shut down can be in one of the XIVE interrupt queues
(of which there are up to 8 per CPU, one for each priority) at the
point where free_irq() is called.  If this happens, can return an
interrupt number which has been shut down.  This can lead to various
symptoms:

- irq_to_desc(irq) can be NULL.  In this case, no end-of-interrupt
  function gets called, resulting in the CPU's elevated interrupt
  priority (numerically lowered CPPR) never gets reset.  That then
  means that the CPU stops processing interrupts, causing device
  timeouts and other errors in various device drivers.

- The irq descriptor or related data structures can be in the process
  of being freed as the interrupt code is using them.  This typically
  leads to crashes due to bad pointer dereferences.

This race is basically what commit 62e0468650c3 ("genirq: Add optional
hardware synchronization for shutdown", 2019-06-28) is intended to
fix, given a get_irqchip_state() method for the interrupt controller
being used.  It works by polling the interrupt controller when an
interrupt is being freed until the controller says it is not pending.

With XIVE, the PQ bits of the interrupt source indicate the state of
the interrupt source, and in particular the P bit goes from 0 to 1 at
the point where the hardware writes an entry into the interrupt queue
that this interrupt is directed towards.  Normally, the code will then
process the interrupt and do an end-of-interrupt (EOI) operation which
will reset PQ to 00 (assuming another interrupt hasn't been generated
in the meantime).  However, there are situations where the code resets
P even though a queue entry exists (for example, by setting PQ to 01,
which disables the interrupt source), and also situations where the
code leaves P at 1 after removing the queue entry (for example, this
is done for escalation interrupts so they cannot fire again until
they are explicitly re-enabled).

The code already has a 'saved_p' flag for the interrupt source which
indicates that a queue entry exists, although it isn't maintained
consistently.  This patch adds a 'stale_p' flag to indicate that
P has been left at 1 after processing a queue entry, and adds code
to set and clear saved_p and stale_p as necessary to maintain a
consistent indication of whether a queue entry may or may not exist.

With this, we can implement xive_get_irqchip_state() by looking at
stale_p, saved_p and the ESB PQ bits for the interrupt.

There is some additional code to handle escalation interrupts
properly; because they are enabled and disabled in KVM assembly code,
which does not have access to the xive_irq_data struct for the
escalation interrupt.  Hence, stale_p may be incorrect when the
escalation interrupt is freed in kvmppc_xive_{,native_}cleanup_vcpu().
Fortunately, we can fix it up by looking at vcpu->arch.xive_esc_on,
with some careful attention to barriers in order to ensure the correct
result if xive_esc_irq() races with kvmppc_xive_cleanup_vcpu().

Finally, this adds code to make noise on the console (pr_crit and
WARN_ON(1)) if we find an interrupt queue entry for an interrupt
which does not have a descriptor.  While this won't catch the race
reliably, if it does get triggered it will be an indication that
the race is occurring and needs to be debugged.

Fixes: 243e25112d06 ("powerpc/xive: Native exploitation of the XIVE interrupt controller")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190813100648.GE9567@blackberry
5 years agoKVM: PPC: Book3S HV: Don't push XIVE context when not using XIVE device
Paul Mackerras [Tue, 13 Aug 2019 10:01:00 +0000 (20:01 +1000)]
KVM: PPC: Book3S HV: Don't push XIVE context when not using XIVE device

At present, when running a guest on POWER9 using HV KVM but not using
an in-kernel interrupt controller (XICS or XIVE), for example if QEMU
is run with the kernel_irqchip=off option, the guest entry code goes
ahead and tries to load the guest context into the XIVE hardware, even
though no context has been set up.

To fix this, we check that the "CAM word" is non-zero before pushing
it to the hardware.  The CAM word is initialized to a non-zero value
in kvmppc_xive_connect_vcpu() and kvmppc_xive_native_connect_vcpu(),
and is now cleared in kvmppc_xive_{,native_}cleanup_vcpu.

Fixes: 5af50993850a ("KVM: PPC: Book3S HV: Native usage of the XIVE interrupt controller")
Cc: stable@vger.kernel.org # v4.12+
Reported-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190813100100.GC9567@blackberry
5 years agoKVM: PPC: Book3S HV: Fix race in re-enabling XIVE escalation interrupts
Paul Mackerras [Tue, 13 Aug 2019 10:03:49 +0000 (20:03 +1000)]
KVM: PPC: Book3S HV: Fix race in re-enabling XIVE escalation interrupts

Escalation interrupts are interrupts sent to the host by the XIVE
hardware when it has an interrupt to deliver to a guest VCPU but that
VCPU is not running anywhere in the system.  Hence we disable the
escalation interrupt for the VCPU being run when we enter the guest
and re-enable it when the guest does an H_CEDE hypercall indicating
it is idle.

It is possible that an escalation interrupt gets generated just as we
are entering the guest.  In that case the escalation interrupt may be
using a queue entry in one of the interrupt queues, and that queue
entry may not have been processed when the guest exits with an H_CEDE.
The existing entry code detects this situation and does not clear the
vcpu->arch.xive_esc_on flag as an indication that there is a pending
queue entry (if the queue entry gets processed, xive_esc_irq() will
clear the flag).  There is a comment in the code saying that if the
flag is still set on H_CEDE, we have to abort the cede rather than
re-enabling the escalation interrupt, lest we end up with two
occurrences of the escalation interrupt in the interrupt queue.

However, the exit code doesn't do that; it aborts the cede in the sense
that vcpu->arch.ceded gets cleared, but it still enables the escalation
interrupt by setting the source's PQ bits to 00.  Instead we need to
set the PQ bits to 10, indicating that an interrupt has been triggered.
We also need to avoid setting vcpu->arch.xive_esc_on in this case
(i.e. vcpu->arch.xive_esc_on seen to be set on H_CEDE) because
xive_esc_irq() will run at some point and clear it, and if we race with
that we may end up with an incorrect result (i.e. xive_esc_on set when
the escalation interrupt has just been handled).

It is extremely unlikely that having two queue entries would cause
observable problems; theoretically it could cause queue overflow, but
the CPU would have to have thousands of interrupts targetted to it for
that to be possible.  However, this fix will also make it possible to
determine accurately whether there is an unhandled escalation
interrupt in the queue, which will be needed by the following patch.

Fixes: 9b9b13a6d153 ("KVM: PPC: Book3S HV: Keep XIVE escalation interrupt masked unless ceded")
Cc: stable@vger.kernel.org # v4.16+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190813100349.GD9567@blackberry
5 years agoKVM: PPC: Book3S HV: XIVE: Free escalation interrupts before disabling the VP
Cédric Le Goater [Tue, 6 Aug 2019 17:25:38 +0000 (19:25 +0200)]
KVM: PPC: Book3S HV: XIVE: Free escalation interrupts before disabling the VP

When a vCPU is brought done, the XIVE VP (Virtual Processor) is first
disabled and then the event notification queues are freed. When freeing
the queues, we check for possible escalation interrupts and free them
also.

But when a XIVE VP is disabled, the underlying XIVE ENDs also are
disabled in OPAL. When an END (Event Notification Descriptor) is
disabled, its ESB pages (ESn and ESe) are disabled and loads return all
1s. Which means that any access on the ESB page of the escalation
interrupt will return invalid values.

When an interrupt is freed, the shutdown handler computes a 'saved_p'
field from the value returned by a load in xive_do_source_set_mask().
This value is incorrect for escalation interrupts for the reason
described above.

This has no impact on Linux/KVM today because we don't make use of it
but we will introduce in future changes a xive_get_irqchip_state()
handler. This handler will use the 'saved_p' field to return the state
of an interrupt and 'saved_p' being incorrect, softlockup will occur.

Fix the vCPU cleanup sequence by first freeing the escalation interrupts
if any, then disable the XIVE VP and last free the queues.

Fixes: 90c73795afa2 ("KVM: PPC: Book3S HV: Add a new KVM device for the XIVE native exploitation mode")
Fixes: 5af50993850a ("KVM: PPC: Book3S HV: Native usage of the XIVE interrupt controller")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190806172538.5087-1-clg@kaod.org
5 years agopowerpc/64s: Make boot look nice(r)
Nicholas Piggin [Thu, 16 May 2019 02:04:37 +0000 (12:04 +1000)]
powerpc/64s: Make boot look nice(r)

Radix boot looks like this:

 -----------------------------------------------------
 phys_mem_size     = 0x200000000
 dcache_bsize      = 0x80
 icache_bsize      = 0x80
 cpu_features      = 0x0000c06f8f5fb1a7
   possible        = 0x0000fbffcf5fb1a7
   always          = 0x00000003800081a1
 cpu_user_features = 0xdc0065c2 0xaee00000
 mmu_features      = 0xbc006041
 firmware_features = 0x0000000010000000
 hash-mmu: ppc64_pft_size    = 0x0
 hash-mmu: kernel vmalloc start   = 0xc008000000000000
 hash-mmu: kernel IO start        = 0xc00a000000000000
 hash-mmu: kernel vmemmap start   = 0xc00c000000000000
 -----------------------------------------------------

Fix:

 -----------------------------------------------------
 phys_mem_size     = 0x200000000
 dcache_bsize      = 0x80
 icache_bsize      = 0x80
 cpu_features      = 0x0000c06f8f5fb1a7
   possible        = 0x0000fbffcf5fb1a7
   always          = 0x00000003800081a1
 cpu_user_features = 0xdc0065c2 0xaee00000
 mmu_features      = 0xbc006041
 firmware_features = 0x0000000010000000
 vmalloc start     = 0xc008000000000000
 IO start          = 0xc00a000000000000
 vmemmap start     = 0xc00c000000000000
 -----------------------------------------------------

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190516020437.11783-1-npiggin@gmail.com
5 years agopowerpc/xive: Add a check for memory allocation failure
Christophe JAILLET [Thu, 1 Aug 2019 08:32:42 +0000 (10:32 +0200)]
powerpc/xive: Add a check for memory allocation failure

The result of this kzalloc is not checked. Add a check and corresponding
error handling code.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/cc53462734dfeaf15b6bad0e626b483de18656b4.1564647619.git.christophe.jaillet@wanadoo.fr
5 years agopowerpc/xive: Use GFP_KERNEL instead of GFP_ATOMIC in 'xive_irq_bitmap_add()'
Christophe JAILLET [Thu, 1 Aug 2019 08:32:31 +0000 (10:32 +0200)]
powerpc/xive: Use GFP_KERNEL instead of GFP_ATOMIC in 'xive_irq_bitmap_add()'

There is no need to use GFP_ATOMIC here. GFP_KERNEL should be enough.
GFP_KERNEL is also already used for another allocation just a few lines
below.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/85d5d247ce753befd6aa63c473f7823de6520ccd.1564647619.git.christophe.jaillet@wanadoo.fr
5 years agopowerpc/64e: Drop stale call to smp_processor_id() which hangs SMP startup
Christophe Leroy [Thu, 8 Aug 2019 12:48:26 +0000 (12:48 +0000)]
powerpc/64e: Drop stale call to smp_processor_id() which hangs SMP startup

Commit ebb9d30a6a74 ("powerpc/mm: any thread in one core can be the
first to setup TLB1") removed the need to know the cpu_id in
early_init_this_mmu(), but the call to smp_processor_id() which was
marked __maybe_used remained.

Since commit ed1cd6deb013 ("powerpc: Activate CONFIG_THREAD_INFO_IN_TASK")
thread_info cannot be reached before MMU is properly set up.

Drop this stale call to smp_processor_id() which makes SMP hang when
CONFIG_PREEMPT is set.

Fixes: ebb9d30a6a74 ("powerpc/mm: any thread in one core can be the first to setup TLB1")
Fixes: ed1cd6deb013 ("powerpc: Activate CONFIG_THREAD_INFO_IN_TASK")
Cc: stable@vger.kernel.org # v5.1+
Reported-by: Chris Packham <Chris.Packham@alliedtelesis.co.nz>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Tested-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/bef479514f4c08329fa649f67735df8918bc0976.1565268248.git.christophe.leroy@c-s.fr
5 years agopowerpc/pseries/hotplug-memory.c: Replace nested ifs by switch-case
Leonardo Bras [Thu, 1 Aug 2019 22:52:51 +0000 (19:52 -0300)]
powerpc/pseries/hotplug-memory.c: Replace nested ifs by switch-case

I noticed these nested ifs can be easily replaced by switch-cases,
which can improve readability.

Signed-off-by: Leonardo Bras <leonardo@linux.ibm.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190801225251.17864-1-leonardo@linux.ibm.com
5 years agopowerpc/xive: Update comment referencing magic loads from an ESB
Jordan Niethe [Fri, 2 Aug 2019 00:08:35 +0000 (10:08 +1000)]
powerpc/xive: Update comment referencing magic loads from an ESB

The comment above xive_esb_read() references magic loads from an ESB as
described xive.h. This has been inaccurate since commit 12c1f339cd49
("powerpc/xive: Move definition of ESB bits") which moved the
description. Update the comment to reference the new location of the
description in xive-regs.h

Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Acked-by: Stewart Smith <stewart@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802000835.26191-1-jniethe5@gmail.com
5 years agopowerpc/32: activate ARCH_HAS_PMEM_API and ARCH_HAS_UACCESS_FLUSHCACHE
Christophe Leroy [Wed, 31 Jul 2019 06:31:41 +0000 (06:31 +0000)]
powerpc/32: activate ARCH_HAS_PMEM_API and ARCH_HAS_UACCESS_FLUSHCACHE

PPC32 also have flush_dcache_range() so it can also support
ARCH_HAS_PMEM_API and ARCH_HAS_UACCESS_FLUSHCACHE without changes.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/a682a2f9db308c5cfe77e45aa3352e41bc9f4e33.1564554634.git.christophe.leroy@c-s.fr
5 years agopowerpc/powernv: Restrict OPAL symbol map to only be readable by root
Andrew Donnellan [Fri, 3 May 2019 07:52:53 +0000 (17:52 +1000)]
powerpc/powernv: Restrict OPAL symbol map to only be readable by root

Currently the OPAL symbol map is globally readable, which seems bad as
it contains physical addresses.

Restrict it to root.

Fixes: c8742f85125d ("powerpc/powernv: Expose OPAL firmware symbol map")
Cc: stable@vger.kernel.org # v3.19+
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190503075253.22798-1-ajd@linux.ibm.com
5 years agopowerpc/configs: Disable SCOM_DEBUGFS in powernv_defconfig
Andrew Donnellan [Thu, 9 May 2019 05:11:19 +0000 (15:11 +1000)]
powerpc/configs: Disable SCOM_DEBUGFS in powernv_defconfig

SCOM_DEBUGFS is really not needed for anything other than low-level
hardware debugging.

mpe: It also introduces a large and poorly documented/understood
attack surface. Although the interface is only available to root, the
kernel still aspires to restrict root to accessing hardware through
well defined interfaces, which this is not.

opal-prd uses its own interface (/dev/prd) for SCOM access, so it
doesn't need SCOM_DEBUGFS.

At some point in the future we'll introduce a debug config fragment
where this can go instead.

Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190509051119.7694-5-ajd@linux.ibm.com
5 years agopowerpc/powernv: Fix checkpatch warnings in opal-xscom.c
Andrew Donnellan [Thu, 9 May 2019 05:11:18 +0000 (15:11 +1000)]
powerpc/powernv: Fix checkpatch warnings in opal-xscom.c

Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190509051119.7694-4-ajd@linux.ibm.com
5 years agopowerpc/powernv: Get rid of old scom_controller abstraction
Andrew Donnellan [Thu, 9 May 2019 05:11:17 +0000 (15:11 +1000)]
powerpc/powernv: Get rid of old scom_controller abstraction

Once upon a time, the SCOM access code was used by the WSP platform as
well as powernv. Thus it made sense to have a generic SCOM access
interface to abstract between different platforms.

Now that it's just powernv, with no other platforms currently on the
horizon, let's rip out scom_controller and make everything much
simpler and more direct.

While we're here, fix up the comment block at the top.

Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190509051119.7694-3-ajd@linux.ibm.com
5 years agopowerpc/powernv: Remove dead SCOM access code
Andrew Donnellan [Thu, 9 May 2019 05:11:16 +0000 (15:11 +1000)]
powerpc/powernv: Remove dead SCOM access code

Nothing is using scom_map_device() or scom_find_parent(). Remove them.

Also don't export scom_controller, there are no other users of it.

Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190509051119.7694-2-ajd@linux.ibm.com
5 years agopowerpc/powernv: Move SCOM access code into powernv platform
Andrew Donnellan [Thu, 9 May 2019 05:11:15 +0000 (15:11 +1000)]
powerpc/powernv: Move SCOM access code into powernv platform

The powernv platform is the only one that directly accesses SCOMs.
Move the support code to platforms/powernv, and get rid of the
PPC_SCOM Kconfig option, as SCOM support is always selected when
compiling for powernv.

This also means that the Kconfig item for CONFIG_SCOM_DEBUGFS will
show up in menuconfig in the platform menu, rather than at the root,
which is a much better location.

Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190509051119.7694-1-ajd@linux.ibm.com
5 years agoPCI: rpaphp: Avoid a sometimes-uninitialized warning
Nathan Chancellor [Mon, 3 Jun 2019 22:11:58 +0000 (15:11 -0700)]
PCI: rpaphp: Avoid a sometimes-uninitialized warning

When building with -Wsometimes-uninitialized, clang warns:

drivers/pci/hotplug/rpaphp_core.c:243:14: warning: variable 'fndit' is
used uninitialized whenever 'for' loop exits because its condition is
false [-Wsometimes-uninitialized]
        for (j = 0; j < entries; j++) {
                    ^~~~~~~~~~~
drivers/pci/hotplug/rpaphp_core.c:256:6: note: uninitialized use occurs
here
        if (fndit)
            ^~~~~
drivers/pci/hotplug/rpaphp_core.c:243:14: note: remove the condition if
it is always true
        for (j = 0; j < entries; j++) {
                    ^~~~~~~~~~~
drivers/pci/hotplug/rpaphp_core.c:233:14: note: initialize the variable
'fndit' to silence this warning
        int j, fndit;
                    ^
                     = 0

fndit is only used to gate a sprintf call, which can be moved into the
loop to simplify the code and eliminate the local variable, which will
fix this warning.

Fixes: 2fcf3ae508c2 ("hotplug/drc-info: Add code to search ibm,drc-info property")
Suggested-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Acked-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Acked-by: Joel Savitz <jsavitz@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://github.com/ClangBuiltLinux/linux/issues/504
Link: https://lore.kernel.org/r/20190603221157.58502-1-natechancellor@gmail.com
5 years agopowerpc/rtas: Unexport rtas_online_cpus_mask, rtas_offline_cpus_mask
Nathan Lynch [Thu, 18 Jul 2019 16:22:14 +0000 (11:22 -0500)]
powerpc/rtas: Unexport rtas_online_cpus_mask, rtas_offline_cpus_mask

These aren't used by modular code, nor should they be.

Fixes: 120496ac2d2d ("powerpc: Bring all threads online prior to migration/hibernation")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190718162214.5694-1-nathanl@linux.ibm.com
5 years agoRevert "powerpc: slightly improve cache helpers"
Michael Ellerman [Mon, 29 Jul 2019 12:28:54 +0000 (22:28 +1000)]
Revert "powerpc: slightly improve cache helpers"

This reverts commit 6c5875843b87c3adea2beade9d1b8b3d4523900a.

It triggers a probable compiler bug on clang which leads to crashes.

With GCC it allows the compiler to use a more efficient register
allocation but current GCC versions never do that at any of the current
call sites, so there's no benefit.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/kasan: fix early boot failure on PPC32
Christophe Leroy [Wed, 31 Jul 2019 06:01:42 +0000 (06:01 +0000)]
powerpc/kasan: fix early boot failure on PPC32

Due to commit 4a6d8cf90017 ("powerpc/mm: don't use pte_alloc_kernel()
until slab is available on PPC32"), pte_alloc_kernel() cannot be used
during early KASAN init.

Fix it by using memblock_alloc() instead.

Fixes: 2edb16efc899 ("powerpc/32: Add KASAN support")
Cc: stable@vger.kernel.org # v5.2+
Reported-by: Erhard F. <erhard_f@mailbox.org>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/da89670093651437f27d2975224712e0a130b055.1564552796.git.christophe.leroy@c-s.fr
5 years agodrivers/macintosh/smu.c: Mark expected switch fall-through
Stephen Rothwell [Tue, 30 Jul 2019 04:37:04 +0000 (14:37 +1000)]
drivers/macintosh/smu.c: Mark expected switch fall-through

Mark switch cases where we are expecting to fall through.

This patch fixes the following warning (Building: powerpc):

  drivers/macintosh/smu.c: In function 'smu_queue_i2c':
  drivers/macintosh/smu.c:854:21: warning: this statement may fall through [-Wimplicit-fallthrough=]
     cmd->info.devaddr &= 0xfe;
     ~~~~~~~~~~~~~~~~~~^~~~~~~
  drivers/macintosh/smu.c:855:2: note: here
    case SMU_I2C_TRANSFER_STDSUB:
    ^~~~

Fixes: 0365ba7fb1fa ("[PATCH] ppc64: SMU driver update & i2c support")
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190730143704.060a2606@canb.auug.org.au
5 years agopowerpc/spe: Mark expected switch fall-throughs
Michael Ellerman [Tue, 30 Jul 2019 14:00:15 +0000 (00:00 +1000)]
powerpc/spe: Mark expected switch fall-throughs

Mark switch cases where we are expecting to fall through.

Fixes errors such as below, seen with mpc85xx_defconfig:

  arch/powerpc/kernel/align.c: In function 'emulate_spe':
  arch/powerpc/kernel/align.c:178:8: error: this statement may fall through
    ret |= __get_user_inatomic(temp.v[3], p++);
        ^~

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190730141917.21817-1-mpe@ellerman.id.au
5 years agopowerpc/nvdimm: Pick nearby online node if the device node is not online
Aneesh Kumar K.V [Mon, 29 Jul 2019 09:51:28 +0000 (15:21 +0530)]
powerpc/nvdimm: Pick nearby online node if the device node is not online

Currently, nvdimm subsystem expects the device numa node for SCM device to be
an online node. It also doesn't try to bring the device numa node online. Hence
if we use a non-online numa node as device node we hit crashes like below. This
is because we try to access uninitialized NODE_DATA in different code paths.

cpu 0x0: Vector: 300 (Data Access) at [c0000000fac53170]
    pc: c0000000004bbc50: ___slab_alloc+0x120/0xca0
    lr: c0000000004bc834: __slab_alloc+0x64/0xc0
    sp: c0000000fac53400
   msr: 8000000002009033
   dar: 73e8
 dsisr: 80000
  current = 0xc0000000fabb6d80
  paca    = 0xc000000003870000   irqmask: 0x03   irq_happened: 0x01
    pid   = 7, comm = kworker/u16:0
Linux version 5.2.0-06234-g76bd729b2644 (kvaneesh@ltc-boston123) (gcc version 7.4.0 (Ubuntu 7.4.0-1ubuntu1~18.04.1)) #135 SMP Thu Jul 11 05:36:30 CDT 2019
enter ? for help
[link register   ] c0000000004bc834 __slab_alloc+0x64/0xc0
[c0000000fac53400c0000000fac53480 (unreliable)
[c0000000fac53500c0000000004bc818 __slab_alloc+0x48/0xc0
[c0000000fac53560c0000000004c30a0 __kmalloc_node_track_caller+0x3c0/0x6b0
[c0000000fac535d0c000000000cfafe4 devm_kmalloc+0x74/0xc0
[c0000000fac53600c000000000d69434 nd_region_activate+0x144/0x560
[c0000000fac536d0c000000000d6b19c nd_region_probe+0x17c/0x370
[c0000000fac537b0c000000000d6349c nvdimm_bus_probe+0x10c/0x230
[c0000000fac53840c000000000cf3cc4 really_probe+0x254/0x4e0
[c0000000fac538d0c000000000cf429c driver_probe_device+0x16c/0x1e0
[c0000000fac53950c000000000cf0b44 bus_for_each_drv+0x94/0x130
[c0000000fac539b0c000000000cf392c __device_attach+0xdc/0x200
[c0000000fac53a50c000000000cf231c bus_probe_device+0x4c/0xf0
[c0000000fac53a90c000000000ced268 device_add+0x528/0x810
[c0000000fac53b60c000000000d62a58 nd_async_device_register+0x28/0xa0
[c0000000fac53bd0c0000000001ccb8c async_run_entry_fn+0xcc/0x1f0
[c0000000fac53c50c0000000001bcd9c process_one_work+0x46c/0x860
[c0000000fac53d20c0000000001bd4f4 worker_thread+0x364/0x5f0
[c0000000fac53db0c0000000001c7260 kthread+0x1b0/0x1c0
[c0000000fac53e20c00000000000b954 ret_from_kernel_thread+0x5c/0x68

The patch tries to fix this by picking the nearest online node as the SCM node.
This does have a problem of us losing the information that SCM node is
equidistant from two other online nodes. If applications need to understand these
fine-grained details we should express then like x86 does via
/sys/devices/system/node/nodeX/accessY/initiators/

With the patch we get

 # numactl -H
available: 2 nodes (0-1)
node 0 cpus:
node 0 size: 0 MB
node 0 free: 0 MB
node 1 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
node 1 size: 130865 MB
node 1 free: 129130 MB
node distances:
node   0   1
  0:  10  20
  1:  20  10
 # cat /sys/bus/nd/devices/region0/numa_node
0
 # dmesg | grep papr_scm
[   91.332305] papr_scm ibm,persistent-memory:ibm,pmemory@44104001: Region registered with target node 2 and online node 0

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190729095128.23707-1-aneesh.kumar@linux.ibm.com
5 years agopowerpc/kvm: Fall through switch case explicitly
Santosh Sivaraj [Mon, 29 Jul 2019 05:55:36 +0000 (11:25 +0530)]
powerpc/kvm: Fall through switch case explicitly

Implicit fallthrough warning was enabled globally which broke
the build. Make it explicit with a `fall through` comment.

Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Reviewed-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190729055536.25591-1-santosh@fossix.org
5 years agopowerpc: Wire up clone3 syscall
Michael Ellerman [Mon, 22 Jul 2019 12:26:56 +0000 (22:26 +1000)]
powerpc: Wire up clone3 syscall

Wire up the new clone3 syscall added in commit 7f192e3cd316 ("fork:
add clone3").

This requires a ppc_clone3 wrapper, in order to save the non-volatile
GPRs before calling into the generic syscall code. Otherwise we hit
the BUG_ON in CHECK_FULL_REGS in copy_thread().

Lightly tested using Christian's test code on a Power8 LE VM.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Christian Brauner <christian@brauner.io>
Link: https://lore.kernel.org/r/20190724140259.23554-1-mpe@ellerman.id.au
5 years agoLinux 5.3-rc2
Linus Torvalds [Sun, 28 Jul 2019 19:47:02 +0000 (12:47 -0700)]
Linux 5.3-rc2

5 years agoMerge tag 'meminit-v5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/kees...
Linus Torvalds [Sun, 28 Jul 2019 19:33:15 +0000 (12:33 -0700)]
Merge tag 'meminit-v5.3-rc2' of git://git./linux/kernel/git/kees/linux

Pull structleak fix from Kees Cook:
 "Disable gcc-based stack variable auto-init under KASAN (Arnd
  Bergmann).

  This fixes a bunch of build warnings under KASAN and the
  gcc-plugin-based stack auto-initialization features (which are
  arguably redundant, so better to let KASAN control this)"

* tag 'meminit-v5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux:
  structleak: disable STRUCTLEAK_BYREF in combination with KASAN_STACK

5 years agoMerge tag 'kbuild-fixes-v5.3' of git://git.kernel.org/pub/scm/linux/kernel/git/masahi...
Linus Torvalds [Sun, 28 Jul 2019 17:35:04 +0000 (10:35 -0700)]
Merge tag 'kbuild-fixes-v5.3' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - add compile_commands.json to .gitignore

 - fix false-positive warning from gen_compile_commands.py after
   allnoconfig build

 - remove unused code

* tag 'kbuild-fixes-v5.3' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  kbuild: remove unused single-used-m
  gen_compile_commands: lower the entry count threshold
  .gitignore: Add compilation database file
  kbuild: remove unused objectify macro

5 years agoMerge tag 'char-misc-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 28 Jul 2019 17:26:10 +0000 (10:26 -0700)]
Merge tag 'char-misc-5.3-rc2' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc driver fixes from Greg KH:
 "Here are some small char and misc driver fixes for 5.3-rc2 to resolve
  some reported issues.

  Nothing major at all, some binder bugfixes for issues found, some new
  mei device ids, firmware building warning fixes, habanalabs fixes, a
  few other build fixes, and a MAINTAINERS update.

  All of these have been in linux-next with no reported issues"

* tag 'char-misc-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  test_firmware: fix a memory leak bug
  hpet: Fix division by zero in hpet_time_div()
  eeprom: make older eeprom drivers select NVMEM_SYSFS
  vmw_balloon: Remove Julien from the maintainers list
  fpga-manager: altera-ps-spi: Fix build error
  mei: me: add mule creek canyon (EHL) device ids
  binder: prevent transactions to context manager from its own process.
  binder: Set end of SG buffer area properly.
  firmware: Fix missing inline
  firmware: fix build errors in paged buffer handling code
  habanalabs: don't reset device when getting VRHOT
  habanalabs: use %pad for printing a dma_addr_t

5 years agoMerge tag 'tty-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
Linus Torvalds [Sun, 28 Jul 2019 17:18:33 +0000 (10:18 -0700)]
Merge tag 'tty-5.3-rc2' of git://git./linux/kernel/git/gregkh/tty

Pull tty fixes from Greg KH:
 "Here are two tty/vt fixes:

   - delete the netx-serial driver as the arch has been removed, no need
     to keep the serial driver for it around either.

   - vt console_lock fix to resolve a reported noisy warning at runtime

  Both of these have been in linux-next with no reported issues"

* tag 'tty-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty:
  vt: Grab console_lock around con_is_bound in show_bind
  tty: serial: netx: Delete driver

5 years agoMerge tag 'spdx-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 28 Jul 2019 17:00:06 +0000 (10:00 -0700)]
Merge tag 'spdx-5.3-rc2' of git://git./linux/kernel/git/gregkh/spdx

Pull SPDX fixes from Greg KH:
 "Here are some small SPDX fixes for 5.3-rc2 for things that came in
  during the 5.3-rc1 merge window that we previously missed.

  Only three small patches here:

   - two uapi patches to resolve some SPDX tags that were not correct

   - fix an invalid SPDX tag in the iomap Makefile file

  All have been properly reviewed on the public mailing lists"

* tag 'spdx-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/spdx:
  iomap: fix Invalid License ID
  treewide: remove SPDX "WITH Linux-syscall-note" from kernel-space headers again
  treewide: add "WITH Linux-syscall-note" to SPDX tag of uapi headers

5 years agoMerge tag 'usb-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sun, 28 Jul 2019 16:52:35 +0000 (09:52 -0700)]
Merge tag 'usb-5.3-rc2' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are some small fixes for 5.3-rc2. All of these resolve some
  reported issues, some more than others :)

  Included in here is:

   - xhci fix for an annoying issue with odd devices

   - reversion of some usb251xb patches that should not have been merged

   - usb pci quirk additions and fixups

   - usb storage fix

   - usb host controller error test fix

  All of these have been in linux-next with no reported issues"

* tag 'usb-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  xhci: Fix crash if scatter gather is used with Immediate Data Transfer (IDT).
  usb: usb251xb: Reallow swap-dx-lanes to apply to the upstream port
  Revert "usb: usb251xb: Add US port lanes inversion property"
  Revert "usb: usb251xb: Add US lanes inversion dts-bindings"
  usb: wusbcore: fix unbalanced get/put cluster_id
  usb/hcd: Fix a NULL vs IS_ERR() bug in usb_hcd_setup_local_mem()
  usb-storage: Add a limitation for blk_queue_max_hw_sectors()
  usb: pci-quirks: Minor cleanup for AMD PLL quirk
  usb: pci-quirks: Correct AMD PLL quirk detection

5 years agoMerge tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
Linus Torvalds [Sun, 28 Jul 2019 16:38:55 +0000 (09:38 -0700)]
Merge tag 'armsoc-fixes' of git://git./linux/kernel/git/soc/soc

Pull ARM SoC fixes from Olof Johansson:
 "Here's the first batch of fixes for this release cycle.

  Main diffstat here is the re-deletion of netx. I messed up and most
  likely didn't remove the files from the index when I test-merged this
  and saw conflicts, and from there on out 'git rerere' remembered the
  mistake and I missed checking it. Here it's done again as expected.

  Besides that:

   - A defconfig refresh + enabling of new drivers for u8500

   - i.MX fixlets for i2c/SAI/pinmux

   - sleep.S build fix for Davinci

   - Broadcom devicetree build/warning fix"

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
  ARM: defconfig: u8500: Add new drivers
  ARM: defconfig: u8500: Refresh defconfig
  ARM: dts: bcm: bcm47094: add missing #cells for mdio-bus-mux
  ARM: davinci: fix sleep.S build error on ARMv4
  arm64: dts: imx8mq: fix SAI compatible
  arm64: dts: imx8mm: Correct SAI3 RXC/TXFS pin's mux option #1
  ARM: dts: imx6ul: fix clock frequency property name of I2C buses
  ARM: Delete netx a second time
  ARM: dts: imx7ulp: Fix usb-phy unit address format

5 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Jul 2019 04:46:43 +0000 (21:46 -0700)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "A set of x86 fixes and functional updates:

   - Prevent stale huge I/O TLB mappings on 32bit. A long standing bug
     which got exposed by KPTI support for 32bit

   - Prevent bogus access_ok() warnings in arch_stack_walk_user()

   - Add display quirks for Lenovo devices which have height and width
     swapped

   - Add the missing CR2 fixup for 32 bit async pagefaults. Fallout of
     the CR2 bug fix series.

   - Unbreak handling of force enabled HPET by moving the 'is HPET
     counting' check back to the original place.

   - A more accurate check for running on a hypervisor platform in the
     MDS mitigation code. Not perfect, but more accurate than the
     previous one.

   - Update a stale and confusing comment vs. IRQ stacks"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/speculation/mds: Apply more accurate check on hypervisor platform
  x86/hpet: Undo the early counter is counting check
  x86/entry/32: Pass cr2 to do_async_page_fault()
  x86/irq/64: Update stale comment
  x86/sysfb_efi: Add quirks for some devices with swapped width and height
  x86/stacktrace: Prevent access_ok() warnings in arch_stack_walk_user()
  mm/vmalloc: Sync unmappings in __purge_vmap_area_lazy()
  x86/mm: Sync also unmappings in vmalloc_sync_all()
  x86/mm: Check for pfn instead of page in vmalloc_sync_one()

5 years agoMerge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Jul 2019 04:22:33 +0000 (21:22 -0700)]
Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixes from Thomas Gleixner:
 "Two fixes for the fair scheduling class:

   - Prevent freeing memory which is accessible by concurrent readers

   - Make the RCU annotations for numa groups consistent"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/fair: Use RCU accessors consistently for ->numa_group
  sched/fair: Don't free p->numa_faults with concurrent readers

5 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Jul 2019 04:17:56 +0000 (21:17 -0700)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Thomas Gleixner:
 "A pile of perf related fixes:

  Kernel:
   - Fix SLOTS PEBS event constraints for Icelake CPUs

   - Add the missing mask bit to allow counting hardware generated
     prefetches on L3 for Icelake CPUs

   - Make the test for hypervisor platforms more accurate (as far as
     possible)

   - Handle PMUs correctly which override event->cpu

   - Yet another missing fallthrough annotation

  Tools:
     perf.data:
        - Fix loading of compressed data split across adjacent records
        - Fix buffer size setting for processing CPU topology perf.data
          header.

     perf stat:
        - Fix segfault for event group in repeat mode
        - Always separate "stalled cycles per insn" line, it was being
          appended to the "instructions" line.

     perf script:
        - Fix --max-blocks man page description.
        - Improve man page description of metrics.
        - Fix off by one in brstackinsn IPC computation.

     perf probe:
        - Avoid calling freeing routine multiple times for same pointer.

     perf build:
        - Do not use -Wshadow on gcc < 4.8, avoiding too strict warnings
          treated as errors, breaking the build"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/x86/intel: Mark expected switch fall-throughs
  perf/core: Fix creating kernel counters for PMUs that override event->cpu
  perf/x86: Apply more accurate check on hypervisor platform
  perf/x86/intel: Fix invalid Bit 13 for Icelake MSR_OFFCORE_RSP_x register
  perf/x86/intel: Fix SLOTS PEBS event constraint
  perf build: Do not use -Wshadow on gcc < 4.8
  perf probe: Avoid calling freeing routine multiple times for same pointer
  perf probe: Set pev->nargs to zero after freeing pev->args entries
  perf session: Fix loading of compressed data split across adjacent records
  perf stat: Always separate stalled cycles per insn
  perf stat: Fix segfault for event group in repeat mode
  perf tools: Fix proper buffer size for feature processing
  perf script: Fix off by one in brstackinsn IPC computation
  perf script: Improve man page description of metrics
  perf script: Fix --max-blocks man page description

5 years agoMerge branch 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Jul 2019 04:10:26 +0000 (21:10 -0700)]
Merge branch 'locking-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull locking fixes from Thomas Gleixner:
 "A set of locking fixes:

   - Address the fallout of the rwsem rework. Missing ACQUIREs and a
     sanity check to prevent a use-after-free

   - Add missing checks for unitialized mutexes when mutex debugging is
     enabled.

   - Remove the bogus code in the generic SMP variant of
     arch_futex_atomic_op_inuser()

   - Fixup the #ifdeffery in lockdep to prevent compile warnings"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/mutex: Test for initialized mutex
  locking/lockdep: Clean up #ifdef checks
  locking/lockdep: Hide unused 'class' variable
  locking/rwsem: Add ACQUIRE comments
  tty/ldsem, locking/rwsem: Add missing ACQUIRE to read_failed sleep loop
  lcoking/rwsem: Add missing ACQUIRE to read_slowpath sleep loop
  locking/rwsem: Add missing ACQUIRE to read_slowpath exit when queue is empty
  locking/rwsem: Don't call owner_on_cpu() on read-owner
  futex: Cleanup generic SMP variant of arch_futex_atomic_op_inuser()

5 years agoMerge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Jul 2019 03:49:43 +0000 (20:49 -0700)]
Merge branch 'core-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull objtool fix from Thomas Gleixner:
 "A single robustness fix for objtool to handle unbalanced CLAC
  invocations under all circumstances"

* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  objtool: Improve UACCESS coverage

5 years agoMerge tag 'Wimplicit-fallthrough-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kerne...
Linus Torvalds [Sat, 27 Jul 2019 18:04:18 +0000 (11:04 -0700)]
Merge tag 'Wimplicit-fallthrough-5.3-rc2' of git://git./linux/kernel/git/gustavoars/linux

Pull Wimplicit-fallthrough enablement from Gustavo A. R. Silva:
 "This marks switch cases where we are expecting to fall through, and
  globally enables the -Wimplicit-fallthrough option in the main
  Makefile.

  Finally, some missing-break fixes that have been tagged for -stable:

   - drm/amdkfd: Fix missing break in switch statement

   - drm/amdgpu/gfx10: Fix missing break in switch statement

  With these changes, we completely get rid of all the fall-through
  warnings in the kernel"

* tag 'Wimplicit-fallthrough-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gustavoars/linux:
  Makefile: Globally enable fall-through warning
  drm/i915: Mark expected switch fall-throughs
  drm/amd/display: Mark expected switch fall-throughs
  drm/amdkfd/kfd_mqd_manager_v10: Avoid fall-through warning
  drm/amdgpu/gfx10: Fix missing break in switch statement
  drm/amdkfd: Fix missing break in switch statement
  perf/x86/intel: Mark expected switch fall-throughs
  mtd: onenand_base: Mark expected switch fall-through
  afs: fsclient: Mark expected switch fall-throughs
  afs: yfsclient: Mark expected switch fall-throughs
  can: mark expected switch fall-throughs
  firewire: mark expected switch fall-throughs

5 years agoMerge tag 's390-5.3-3' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Sat, 27 Jul 2019 15:58:04 +0000 (08:58 -0700)]
Merge tag 's390-5.3-3' of git://git./linux/kernel/git/s390/linux

Pull s390 updates from Heiko Carstens:

 - Add ABI to kernel image file which allows e.g. the file utility to
   figure out the kernel version.

 - Wire up clone3 system call.

 - Add support for kasan bitops instrumentation.

 - uapi header cleanup: use __u{16,32,64} instead of uint{16,32,64}_t.

 - Provide proper ARCH_ZONE_DMA_BITS so the s390 DMA zone is correctly
   defined with 2 GB instead of the default value of 1 MB.

 - Farhan Ali leaves the group of vfio-ccw maintainers.

 - Various small vfio-ccw fixes.

 - Add missing locking for airq_areas array in virtio code.

 - Minor qdio improvements.

* tag 's390-5.3-3' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  MAINTAINERS: vfio-ccw: Remove myself as the maintainer
  s390/mm: use shared variables for sysctl range check
  virtio/s390: fix race on airq_areas[]
  s390/dma: provide proper ARCH_ZONE_DMA_BITS value
  s390/kasan: add bitops instrumentation
  s390/bitops: make test functions return bool
  s390: wire up clone3 system call
  kbuild: enable arch/s390/include/uapi/asm/zcrypt.h for uapi header test
  s390: use __u{16,32,64} instead of uint{16,32,64}_t in uapi header
  s390/hypfs: fix a typo in the name of a function
  s390/qdio: restrict QAOB usage to IQD unicast queues
  s390/qdio: add sanity checks to the fast-requeue path
  s390: enable detection of kernel version from bzImage
  Documentation: fix vfio-ccw doc
  vfio-ccw: Update documentation for csch/hsch
  vfio-ccw: Don't call cp_free if we are processing a channel program
  vfio-ccw: Set pa_nr to 0 if memory allocation fails for pa_iova_pfn
  vfio-ccw: Fix memory leak and don't call cp_free in cp_init
  vfio-ccw: Fix misleading comment when setting orb.cmd.c64

5 years agoMerge tag 'devicetree-fixes-for-5.3-2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 27 Jul 2019 15:49:19 +0000 (08:49 -0700)]
Merge tag 'devicetree-fixes-for-5.3-2' of git://git./linux/kernel/git/robh/linux

Pull Devicetree fixes from Rob Herring:
 "The nvmem changes would typically go thru Greg's tree, but they were
  missed in the merge window. [ Acked by Greg ]

  Summary:

   - Fix mismatches in $id values and actual filenames. Now checked by
     tools.

   - Convert nvmem binding to DT schema

   - Fix a typo in of_property_read_bool() kerneldoc

   - Remove some redundant description in al-fic interrupt-controller"

* tag 'devicetree-fixes-for-5.3-2' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  dt-bindings: Fix more $id value mismatches filenames
  dt-bindings: nvmem: SID: Fix the examples node names
  dt-bindings: nvmem: Add YAML schemas for the generic NVMEM bindings
  of: Fix typo in kerneldoc
  dt-bindings: interrupt-controller: al-fic: remove redundant binding
  dt-bindings: clk: allwinner,sun4i-a10-ccu: Correct path in $id

5 years agoMerge tag 'libnvdimm-fixes-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 27 Jul 2019 15:25:51 +0000 (08:25 -0700)]
Merge tag 'libnvdimm-fixes-5.3-rc2' of git://git./linux/kernel/git/nvdimm/nvdimm

Pull libnvdimm fixes from Dan Williams:
 "A collection of locking and async operations fixes for v5.3-rc2. These
  had been soaking in a branch targeting the merge window, but missed
  due to a regression hunt. This fixed up version has otherwise been in
  -next this past week with no reported issues.

  In order to gain confidence in the locking changes the pull also
  includes a debug / instrumentation patch to enable lockdep coverage
  for libnvdimm subsystem operations that depend on the device_lock for
  exclusion. As mentioned in the changelog it is a hack, but it works
  and documents the locking expectations of the sub-system in a way that
  others can use lockdep to verify. The driver core touches got an ack
  from Greg.

  Summary:

   - Fix duplicate device_unregister() calls (multiple threads competing
     to do unregister work when scheduling device removal from a sysfs
     attribute of the self-same device).

   - Fix badblocks registration order bug. Ensure region badblocks are
     initialized in advance of namespace registration.

   - Fix a deadlock between the bus lock and probe operations.

   - Export device-core infrastructure to coordinate async operations
     via the device ->dead state.

   - Add device-core infrastructure to validate device_lock() usage with
     lockdep"

* tag 'libnvdimm-fixes-5.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  driver-core, libnvdimm: Let device subsystems add local lockdep coverage
  libnvdimm/bus: Fix wait_nvdimm_bus_probe_idle() ABBA deadlock
  libnvdimm/bus: Stop holding nvdimm_bus_list_mutex over __nd_ioctl()
  libnvdimm/bus: Prepare the nd_ioctl() path to be re-entrant
  libnvdimm/region: Register badblocks before namespaces
  libnvdimm/bus: Prevent duplicate device_unregister() calls
  drivers/base: Introduce kill_device()

5 years agokbuild: remove unused single-used-m
Masahiro Yamada [Fri, 26 Jul 2019 02:17:38 +0000 (11:17 +0900)]
kbuild: remove unused single-used-m

This is unused since commit 9f69a496f100 ("kbuild: split out *.mod out
of {single,multi}-used-m rules").

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
5 years agogen_compile_commands: lower the entry count threshold
Masahiro Yamada [Sat, 27 Jul 2019 03:01:10 +0000 (12:01 +0900)]
gen_compile_commands: lower the entry count threshold

Running gen_compile_commands.py after building the kernel with
allnoconfig gave this:

$ ./scripts/gen_compile_commands.py
WARNING: Found 449 entries. Have you compiled the kernel?

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
5 years ago.gitignore: Add compilation database file
Toru Komatsu [Wed, 24 Jul 2019 00:22:33 +0000 (09:22 +0900)]
.gitignore: Add compilation database file

This file is used by clangd to use language server protocol.
It can be generated at each compile using scripts/gen_compile_commands.py.
Therefore it is different depending on the environment and should be
ignored.

Signed-off-by: Toru Komatsu <k0ma@utam0k.jp>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
5 years agokbuild: remove unused objectify macro
Masahiro Yamada [Tue, 23 Jul 2019 04:11:26 +0000 (13:11 +0900)]
kbuild: remove unused objectify macro

Commit 415008af3219 ("docs-rst: convert lsm from DocBook to ReST")
removed the last users of this macro.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
5 years agoMerge tag 'for-linus-20190726-2' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 27 Jul 2019 02:20:34 +0000 (19:20 -0700)]
Merge tag 'for-linus-20190726-2' of git://git.kernel.dk/linux-block

Pull block DMA segment fix from Jens Axboe:
 "Here's the virtual boundary segment size fix"

* tag 'for-linus-20190726-2' of git://git.kernel.dk/linux-block:
  block: fix max segment size handling in blk_queue_virt_boundary

5 years agoMerge tag 'selinux-pr-20190726' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 27 Jul 2019 02:13:38 +0000 (19:13 -0700)]
Merge tag 'selinux-pr-20190726' of git://git./linux/kernel/git/pcmoore/selinux

Pull selinux fix from Paul Moore:
 "One small SELinux patch to add some proper bounds/overflow checking
  when adding a new sid/secid"

* tag 'selinux-pr-20190726' of git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/selinux:
  selinux: check sidtab limit before adding a new entry

5 years agodt-bindings: Fix more $id value mismatches filenames
Rob Herring [Fri, 26 Jul 2019 23:36:52 +0000 (17:36 -0600)]
dt-bindings: Fix more $id value mismatches filenames

The path in the schema '$id' values are wrong. Fix them.

Signed-off-by: Rob Herring <robh@kernel.org>
5 years agodt-bindings: nvmem: SID: Fix the examples node names
Maxime Ripard [Wed, 3 Jul 2019 09:54:21 +0000 (11:54 +0200)]
dt-bindings: nvmem: SID: Fix the examples node names

Now that the examples are validated, the examples in the SID binding
generates an error since the node names aren't one of the valid ones.

Let's switch for one that is ok.

Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com>
Signed-off-by: Rob Herring <robh@kernel.org>
5 years agodt-bindings: nvmem: Add YAML schemas for the generic NVMEM bindings
Maxime Ripard [Thu, 27 Jun 2019 15:10:37 +0000 (16:10 +0100)]
dt-bindings: nvmem: Add YAML schemas for the generic NVMEM bindings

The nvmem providers and consumers have a bunch of generic properties that
are needed in a device tree. Add a YAML schemas for those.

Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com>
[Srini: Changed licence to (GPL-2.0 OR BSD-2-Clause)]
Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Signed-off-by: Rob Herring <robh@kernel.org>
5 years agoof: Fix typo in kerneldoc
Thierry Reding [Fri, 26 Jul 2019 10:17:44 +0000 (12:17 +0200)]
of: Fix typo in kerneldoc

"Findfrom" is not a word. Replace the function synopsis by something
that makes sense.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Rob Herring <robh@kernel.org>
5 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Fri, 26 Jul 2019 21:20:28 +0000 (14:20 -0700)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Nine fixes: The most important core one is the dma_max_mapping_size
  fix that corrects the boot problem Gunter Roeck was having. A couple
  of other driver only fixes are significant, like the cxgbi selector
  support addition, the alua 2 second delay and the fdomain build fix"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: scsi_dh_alua: always use a 2 second delay before retrying RTPG
  scsi: ibmvfc: fix WARN_ON during event pool release
  scsi: fcoe: fix a typo
  scsi: megaraid_sas: Make some functions static
  scsi: megaraid_sas: fix panic on loading firmware crashdump
  scsi: megaraid_sas: fix spelling mistake "megarid_sas" -> "megaraid_sas"
  scsi: core: fix the dma_max_mapping_size call
  scsi: fdomain: fix building pcmcia front-end
  scsi: target: cxgbit: add support for IEEE_8021QAZ_APP_SEL_STREAM selector