platform/kernel/linux-rpi.git
6 years agoKVM: PPC: Book3S: Fix guest DMA when guest partially backed by THP pages
Paul Mackerras [Thu, 23 Aug 2018 00:08:58 +0000 (10:08 +1000)]
KVM: PPC: Book3S: Fix guest DMA when guest partially backed by THP pages

Commit 76fa4975f3ed ("KVM: PPC: Check if IOMMU page is contained in
the pinned physical page", 2018-07-17) added some checks to ensure
that guest DMA mappings don't attempt to map more than the guest is
entitled to access. However, errors in the logic mean that legitimate
guest requests to map pages for DMA are being denied in some
situations. Specifically, if the first page of the range passed to
mm_iommu_get() is mapped with a normal page, and subsequent pages are
mapped with transparent huge pages, we end up with mem->pageshift ==
0. That means that the page size checks in mm_iommu_ua_to_hpa() and
mm_iommu_up_to_hpa_rm() will always fail for every page in that
region, and thus the guest can never map any memory in that region for
DMA, typically leading to a flood of error messages like this:

  qemu-system-ppc64: VFIO_MAP_DMA: -22
  qemu-system-ppc64: vfio_dma_map(0x10005f47780, 0x800000000000000, 0x10000, 0x7fff63ff0000) = -22 (Invalid argument)

The logic errors in mm_iommu_get() are:

  (a) use of 'ua' not 'ua + (i << PAGE_SHIFT)' in the find_linux_pte()
      call (meaning that find_linux_pte() returns the pte for the
      first address in the range, not the address we are currently up
      to);
  (b) use of 'pageshift' as the variable to receive the hugepage shift
      returned by find_linux_pte() - for a normal page this gets set
      to 0, leading to us setting mem->pageshift to 0 when we conclude
      that the pte returned by find_linux_pte() didn't match the page
      we were looking at;
  (c) comparing 'compshift', which is a page order, i.e. log base 2 of
      the number of pages, with 'pageshift', which is a log base 2 of
      the number of bytes.

To fix these problems, this patch introduces 'cur_ua' to hold the
current user address and uses that in the find_linux_pte() call;
introduces 'pteshift' to hold the hugepage shift found by
find_linux_pte(); and compares 'pteshift' with 'compshift +
PAGE_SHIFT' rather than 'compshift'.

The patch also moves the local_irq_restore to the point after the PTE
pointer returned by find_linux_pte() has been dereferenced because
otherwise the PTE could change underneath us, and adds a check to
avoid doing the find_linux_pte() call once mem->pageshift has been
reduced to PAGE_SHIFT, as an optimization.

Fixes: 76fa4975f3ed ("KVM: PPC: Check if IOMMU page is contained in the pinned physical page")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm/radix: Only need the Nest MMU workaround for R -> RW transition
Aneesh Kumar K.V [Wed, 22 Aug 2018 17:16:05 +0000 (22:46 +0530)]
powerpc/mm/radix: Only need the Nest MMU workaround for R -> RW transition

The Nest MMU workaround is only needed for RW upgrades. Avoid doing
that for other PTE updates.

We also avoid clearing the PTE while marking it invalid. This is
because other page table walkers will find this PTE none and can
result in unexpected behaviour due to that. Instead we clear
_PAGE_PRESENT and set the software PTE bit _PAGE_INVALID.
pte_present() is already updated to check for both bits. This makes
sure page table walkers will find the PTE present and things like
pte_pfn(pte) returns the right value.

Based on an original patch from Benjamin Herrenschmidt <benh@kernel.crashing.org>

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm/books3s: Add new pte bit to mark pte temporarily invalid.
Aneesh Kumar K.V [Wed, 22 Aug 2018 17:16:04 +0000 (22:46 +0530)]
powerpc/mm/books3s: Add new pte bit to mark pte temporarily invalid.

When splitting a huge pmd pte, we need to mark the pmd entry invalid. We
can do that by clearing _PAGE_PRESENT bit. But then that will be taken as a
swap pte. In order to differentiate between the two use a software pte bit
when invalidating.

For regular pte, due to bd5050e38aec ("powerpc/mm/radix: Change pte relax
sequence to handle nest MMU hang") we need to mark the pte entry invalid when
relaxing access permission. Instead of marking pte_none which can result in
different page table walk routines possibly skipping this pte entry, invalidate
it but still keep it marked present.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/nohash: fix pte_access_permitted()
Christophe Leroy [Tue, 21 Aug 2018 13:03:23 +0000 (13:03 +0000)]
powerpc/nohash: fix pte_access_permitted()

Commit 5769beaf180a8 ("powerpc/mm: Add proper pte access check helper
for other platforms") replaced generic pte_access_permitted() by an
arch specific one.

The generic one is defined as
(pte_present(pte) && (!(write) || pte_write(pte)))

The arch specific one is open coded checking that _PAGE_USER and
_PAGE_WRITE (_PAGE_RW) flags are set, but lacking to check that
_PAGE_RO and _PAGE_PRIVILEGED are unset, leading to a useless test
on targets like the 8xx which defines _PAGE_RW and _PAGE_USER as 0.

Commit 5fa5b16be5b31 ("powerpc/mm/hugetlb: Use pte_access_permitted
for hugetlb access check") replaced some tests performed with
pte helpers by a call to pte_access_permitted(), leading to the same
issue.

This patch rewrites powerpc/nohash pte_access_permitted()
using pte helpers.

Fixes: 5769beaf180a8 ("powerpc/mm: Add proper pte access check helper for other platforms")
Fixes: 5fa5b16be5b31 ("powerpc/mm/hugetlb: Use pte_access_permitted for hugetlb access check")
Cc: stable@vger.kernel.org # v4.15+
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/topology: Get topology for shared processors at boot
Srikar Dronamraju [Fri, 17 Aug 2018 14:54:39 +0000 (20:24 +0530)]
powerpc/topology: Get topology for shared processors at boot

On a shared LPAR, Phyp will not update the CPU associativity at boot
time. Just after the boot system does recognize itself as a shared
LPAR and trigger a request for correct CPU associativity. But by then
the scheduler would have already created/destroyed its sched domains.

This causes
  - Broken load balance across Nodes causing islands of cores.
  - Performance degradation esp if the system is lightly loaded
  - dmesg to wrongly report all CPUs to be in Node 0.
  - Messages in dmesg saying borken topology.
  - With commit 051f3ca02e46 ("sched/topology: Introduce NUMA identity
    node sched domain"), can cause rcu stalls at boot up.

The sched_domains_numa_masks table which is used to generate cpumasks
is only created at boot time just before creating sched domains and
never updated. Hence, its better to get the topology correct before
the sched domains are created.

For example on 64 core Power 8 shared LPAR, dmesg reports

  Brought up 512 CPUs
  Node 0 CPUs: 0-511
  Node 1 CPUs:
  Node 2 CPUs:
  Node 3 CPUs:
  Node 4 CPUs:
  Node 5 CPUs:
  Node 6 CPUs:
  Node 7 CPUs:
  Node 8 CPUs:
  Node 9 CPUs:
  Node 10 CPUs:
  Node 11 CPUs:
  ...
  BUG: arch topology borken
       the DIE domain not a subset of the NUMA domain
  BUG: arch topology borken
       the DIE domain not a subset of the NUMA domain

numactl/lscpu output will still be correct with cores spreading across
all nodes:

  Socket(s):             64
  NUMA node(s):          12
  Model:                 2.0 (pvr 004d 0200)
  Model name:            POWER8 (architected), altivec supported
  Hypervisor vendor:     pHyp
  Virtualization type:   para
  L1d cache:             64K
  L1i cache:             32K
  NUMA node0 CPU(s): 0-7,32-39,64-71,96-103,176-183,272-279,368-375,464-471
  NUMA node1 CPU(s): 8-15,40-47,72-79,104-111,184-191,280-287,376-383,472-479
  NUMA node2 CPU(s): 16-23,48-55,80-87,112-119,192-199,288-295,384-391,480-487
  NUMA node3 CPU(s): 24-31,56-63,88-95,120-127,200-207,296-303,392-399,488-495
  NUMA node4 CPU(s):     208-215,304-311,400-407,496-503
  NUMA node5 CPU(s):     168-175,264-271,360-367,456-463
  NUMA node6 CPU(s):     128-135,224-231,320-327,416-423
  NUMA node7 CPU(s):     136-143,232-239,328-335,424-431
  NUMA node8 CPU(s):     216-223,312-319,408-415,504-511
  NUMA node9 CPU(s):     144-151,240-247,336-343,432-439
  NUMA node10 CPU(s):    152-159,248-255,344-351,440-447
  NUMA node11 CPU(s):    160-167,256-263,352-359,448-455

Currently on this LPAR, the scheduler detects 2 levels of Numa and
created numa sched domains for all CPUs, but it finds a single DIE
domain consisting of all CPUs. Hence it deletes all numa sched
domains.

To address this, detect the shared processor and update topology soon
after CPUs are setup so that correct topology is updated just before
scheduler creates sched domain.

With the fix, dmesg reports:

  numa: Node 0 CPUs: 0-7 32-39 64-71 96-103 176-183 272-279 368-375 464-471
  numa: Node 1 CPUs: 8-15 40-47 72-79 104-111 184-191 280-287 376-383 472-479
  numa: Node 2 CPUs: 16-23 48-55 80-87 112-119 192-199 288-295 384-391 480-487
  numa: Node 3 CPUs: 24-31 56-63 88-95 120-127 200-207 296-303 392-399 488-495
  numa: Node 4 CPUs: 208-215 304-311 400-407 496-503
  numa: Node 5 CPUs: 168-175 264-271 360-367 456-463
  numa: Node 6 CPUs: 128-135 224-231 320-327 416-423
  numa: Node 7 CPUs: 136-143 232-239 328-335 424-431
  numa: Node 8 CPUs: 216-223 312-319 408-415 504-511
  numa: Node 9 CPUs: 144-151 240-247 336-343 432-439
  numa: Node 10 CPUs: 152-159 248-255 344-351 440-447
  numa: Node 11 CPUs: 160-167 256-263 352-359 448-455

and lscpu also reports:

  Socket(s):             64
  NUMA node(s):          12
  Model:                 2.0 (pvr 004d 0200)
  Model name:            POWER8 (architected), altivec supported
  Hypervisor vendor:     pHyp
  Virtualization type:   para
  L1d cache:             64K
  L1i cache:             32K
  NUMA node0 CPU(s): 0-7,32-39,64-71,96-103,176-183,272-279,368-375,464-471
  NUMA node1 CPU(s): 8-15,40-47,72-79,104-111,184-191,280-287,376-383,472-479
  NUMA node2 CPU(s): 16-23,48-55,80-87,112-119,192-199,288-295,384-391,480-487
  NUMA node3 CPU(s): 24-31,56-63,88-95,120-127,200-207,296-303,392-399,488-495
  NUMA node4 CPU(s):     208-215,304-311,400-407,496-503
  NUMA node5 CPU(s):     168-175,264-271,360-367,456-463
  NUMA node6 CPU(s):     128-135,224-231,320-327,416-423
  NUMA node7 CPU(s):     136-143,232-239,328-335,424-431
  NUMA node8 CPU(s):     216-223,312-319,408-415,504-511
  NUMA node9 CPU(s):     144-151,240-247,336-343,432-439
  NUMA node10 CPU(s):    152-159,248-255,344-351,440-447
  NUMA node11 CPU(s):    160-167,256-263,352-359,448-455

Reported-by: Manjunatha H R <manjuhr1@in.ibm.com>
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
[mpe: Trim / format change log]
Tested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc64/ftrace: Include ftrace.h needed for enable/disable calls
Luke Dashjr [Thu, 16 Aug 2018 21:36:26 +0000 (21:36 +0000)]
powerpc64/ftrace: Include ftrace.h needed for enable/disable calls

this_cpu_disable_ftrace and this_cpu_enable_ftrace are inlines in
ftrace.h Without it included, the build fails.

Fixes: a4bc64d305af ("powerpc64/ftrace: Disable ftrace during kvm entry/exit")
Cc: stable@vger.kernel.org # v4.18+
Signed-off-by: Luke Dashjr <luke-jr+git@utopios.org>
Acked-by: Naveen N. Rao <naveen.n.rao at linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powernv/pci: Work around races in PCI bridge enabling
Benjamin Herrenschmidt [Fri, 17 Aug 2018 07:30:39 +0000 (17:30 +1000)]
powerpc/powernv/pci: Work around races in PCI bridge enabling

The generic code is racy when multiple children of a PCI bridge try to
enable it simultaneously.

This leads to drivers trying to access a device through a
not-yet-enabled bridge, and this EEH errors under various
circumstances when using parallel driver probing.

There is work going on to fix that properly in the PCI core but it
will take some time.

x86 gets away with it because (outside of hotplug), the BIOS enables
all the bridges at boot time.

This patch does the same thing on powernv by enabling all bridges that
have child devices at boot time, thus avoiding subsequent races. It's
suitable for backporting to stable and distros, while the proper PCI
fix will probably be significantly more invasive.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: stable@vger.kernel.org
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/fadump: cleanup crash memory ranges support
Hari Bathini [Fri, 17 Aug 2018 20:40:56 +0000 (02:10 +0530)]
powerpc/fadump: cleanup crash memory ranges support

Commit 1bd6a1c4b80a ("powerpc/fadump: handle crash memory ranges array
index overflow") changed crash memory ranges to a dynamic array that
is reallocated on-demand with krealloc(). The relevant header for this
call was not included. The kernel compiles though. But be cautious and
add the header anyway.

Also, memory allocation logic in fadump_add_crash_memory() takes care
of memory allocation for crash memory ranges in all scenarios. Drop
unnecessary memory allocation in fadump_setup_crash_memory_ranges().

Fixes: 1bd6a1c4b80a ("powerpc/fadump: handle crash memory ranges array index overflow")
Cc: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powernv: provide a console flush operation for opal hvc driver
Nicholas Piggin [Mon, 30 Apr 2018 14:55:58 +0000 (00:55 +1000)]
powerpc/powernv: provide a console flush operation for opal hvc driver

Provide the flush hv_op for the opal hvc driver. This will flush the
firmware console buffers without spinning with interrupts disabled.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/traps: Avoid rate limit messages from show unhandled signals
Michael Ellerman [Thu, 16 Aug 2018 05:27:47 +0000 (15:27 +1000)]
powerpc/traps: Avoid rate limit messages from show unhandled signals

In the recent commit to add an explicit ratelimit state when showing
unhandled signals, commit 35a52a10c3ac ("powerpc/traps: Use an
explicit ratelimit state for show_signal_msg()"), I put the check of
show_unhandled_signals and the ratelimit state before the call to
unhandled_signal() so as to avoid unnecessarily calling the latter
when show_unhandled_signals is false.

However that causes us to check the ratelimit state on every call, so
if we take a lot of *handled* signals that has the effect of making
the ratelimit code print warnings that callbacks have been suppressed
when they haven't.

So rearrange the code so that we check show_unhandled_signals first,
then call unhandled_signal() and finally check the ratelimit state.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
6 years agopowerpc/64s: Fix PACA_IRQ_HARD_DIS accounting in idle_power4()
Nicholas Piggin [Tue, 7 Aug 2018 13:21:56 +0000 (23:21 +1000)]
powerpc/64s: Fix PACA_IRQ_HARD_DIS accounting in idle_power4()

When idle_power4() hard disables interrupts then finds a soft pending
interrupt, it returns with interrupts hard disabled but without
PACA_IRQ_HARD_DIS set. Commit 9b81c0211c ("powerpc/64s: make
PACA_IRQ_HARD_DIS track MSR[EE] closely") added a warning for that
condition (since disabled).

Fix this by adding the PACA_IRQ_HARD_DIS for that case.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm/book3s/radix: Add mapping statistics
Aneesh Kumar K.V [Mon, 13 Aug 2018 05:44:57 +0000 (11:14 +0530)]
powerpc/mm/book3s/radix: Add mapping statistics

Add statistics that show how memory is mapped within the kernel linear mapping.
This is similar to commit 37cd944c8d8f ("s390/pgtable: add mapping statistics")

We don't do this with Hash translation mode. Hash uses one size (mmu_linear_psize)
to map the kernel linear mapping and we print the linear psize during boot as
below.

"Page orders: linear mapping = 24, virtual = 16, io = 16, vmemmap = 24"

A sample output looks like:

DirectMap4k:           0 kB
DirectMap64k:       18432 kB
DirectMap2M:     1030144 kB
DirectMap1G:    11534336 kB

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoMerge branch 'next' of https://git.kernel.org/pub/scm/linux/kernel/git/scottwood...
Michael Ellerman [Mon, 13 Aug 2018 06:06:23 +0000 (16:06 +1000)]
Merge branch 'next' of https://git./linux/kernel/git/scottwood/linux into next

Merge some updates from Scott:

"This contains an 8xx compilation fix, and a dpaa device tree fix."

6 years agoMerge branch 'topic/hvc' into next
Michael Ellerman [Mon, 13 Aug 2018 06:04:10 +0000 (16:04 +1000)]
Merge branch 'topic/hvc' into next

This branch held some hvc related commits (Hypervisor Virtual Console)
so that they could get some wider testing in linux-next before merging.

6 years agoMerge branch 'fixes' into next
Michael Ellerman [Mon, 13 Aug 2018 05:59:06 +0000 (15:59 +1000)]
Merge branch 'fixes' into next

Merge our fixes branch from the 4.18 cycle to resolve some minor
conflicts.

6 years agopowerpc/uaccess: Enable get_user(u64, *p) on 32-bit
Michael Ellerman [Tue, 10 Jul 2018 06:20:56 +0000 (16:20 +1000)]
powerpc/uaccess: Enable get_user(u64, *p) on 32-bit

Currently if you build a 32-bit powerpc kernel and use get_user() to
load a u64 value it will fail to build with eg:

  kernel/rseq.o: In function `rseq_get_rseq_cs':
  kernel/rseq.c:123: undefined reference to `__get_user_bad'

This is hitting the check in __get_user_size() that makes sure the
size we're copying doesn't exceed the size of the destination:

  #define __get_user_size(x, ptr, size, retval)
  do {
   retval = 0;
   __chk_user_ptr(ptr);
   if (size > sizeof(x))
   (x) = __get_user_bad();

Which doesn't immediately make sense because the size of the
destination is u64, but it's not really, because __get_user_check()
etc. internally create an unsigned long and copy into that:

  #define __get_user_check(x, ptr, size)
  ({
   long __gu_err = -EFAULT;
   unsigned long  __gu_val = 0;

The problem being that on 32-bit unsigned long is not big enough to
hold a u64. We can fix this with a trick from hpa in the x86 code, we
statically check the type of x and set the type of __gu_val to either
unsigned long or unsigned long long.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm/hash: Remove unnecessary do { } while(0) loop
Aneesh Kumar K.V [Thu, 9 Aug 2018 13:36:42 +0000 (19:06 +0530)]
powerpc/mm/hash: Remove unnecessary do { } while(0) loop

Avoid coverity false warnings like:

  *** CID 187347:  Control flow issues  (UNREACHABLE)
  /arch/powerpc/mm/hash_native_64.c: 819 in native_flush_hash_range()
  813        slot += hidx & _PTEIDX_GROUP_IX;
  814        hptep = htab_address + slot;
  815        want_v = hpte_encode_avpn(vpn, psize, ssize);
  816        hpte_v = hpte_get_old_v(hptep);
  817
  818        if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
  >>>     CID 187347:  Control flow issues  (UNREACHABLE)

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: move machine check SLB flushing to mm/slb.c
Nicholas Piggin [Fri, 10 Aug 2018 06:42:48 +0000 (16:42 +1000)]
powerpc/64s: move machine check SLB flushing to mm/slb.c

The machine check code that flushes and restores bolted segments in
real mode belongs in mm/slb.c. This will also be used by pseries
machine check and idle code in future changes.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powernv/idle: Fix build error
Aneesh Kumar K.V [Thu, 9 Aug 2018 13:37:20 +0000 (19:07 +0530)]
powerpc/powernv/idle: Fix build error

Fix the below build error using strlcpy instead of strncpy

In function 'pnv_parse_cpuidle_dt',
    inlined from 'pnv_init_idle_states' at arch/powerpc/platforms/powernv/idle.c:840:7,
    inlined from '__machine_initcall_powernv_pnv_init_idle_states' at arch/powerpc/platforms/powernv/idle.c:870:1:
arch/powerpc/platforms/powernv/idle.c:820:3: error: 'strncpy' specified bound 16 equals destination size [-Werror=stringop-truncation]
   strncpy(pnv_idle_states[i].name, temp_string[i],
   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    PNV_IDLE_NAME_LEN);

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm/tlbflush: update the mmu_gather page size while iterating address range
Aneesh Kumar K.V [Thu, 9 Aug 2018 13:36:59 +0000 (19:06 +0530)]
powerpc/mm/tlbflush: update the mmu_gather page size while iterating address range

This patch makes sure we update the mmu_gather page size even if we are
requesting for a fullmm flush. This avoids triggering VM_WARN_ON in code
paths like __tlb_remove_page_size that explicitly check for removing range page
size to be same as mmu gather page size.

Fixes: 5a6099346c41 ("powerpc/64s/radix: tlb do not flush on page size when fullmm")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Acked-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm: remove warning about ‘type’ being set
Mathieu Malaterre [Fri, 22 Jun 2018 19:27:47 +0000 (21:27 +0200)]
powerpc/mm: remove warning about ‘type’ being set

‘type’ is only used when CONFIG_DEBUG_HIGHMEM is set. So add a possibly
unused tag to variable. Remove warning treated as error with W=1:

  arch/powerpc/mm/highmem.c:59:6: error: variable ‘type’ set but not used [-Werror=unused-but-set-variable]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/32: Include setup.h header file to fix warnings
Mathieu Malaterre [Fri, 22 Jun 2018 19:26:53 +0000 (21:26 +0200)]
powerpc/32: Include setup.h header file to fix warnings

Make sure to include setup.h to provide the following prototypes:

  - irqstack_early_init
  - setup_power_save
  - initialize_cache_info

Fix the following warnings (treated as error in W=1):

  arch/powerpc/kernel/setup_32.c:198:13: error: no previous prototype for ‘irqstack_early_init’
  arch/powerpc/kernel/setup_32.c:238:13: error: no previous prototype for ‘setup_power_save’
  arch/powerpc/kernel/setup_32.c:253:13: error: no previous prototype for ‘initialize_cache_info’

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Move `path` variable inside DEBUG_PROM
Mathieu Malaterre [Wed, 4 Apr 2018 20:08:35 +0000 (22:08 +0200)]
powerpc: Move `path` variable inside DEBUG_PROM

Add gcc attribute unused for two variables. Fix warnings treated as errors
with W=1:

  arch/powerpc/kernel/prom_init.c:1388:8: error: variable ‘path’ set but not used [-Werror=unused-but-set-variable]

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powermac: Make some functions static
Mathieu Malaterre [Wed, 28 Mar 2018 19:39:35 +0000 (21:39 +0200)]
powerpc/powermac: Make some functions static

These functions can all be static, make it so. Fix warnings treated as
errors with W=1:

  arch/powerpc/platforms/powermac/pci.c:1022:6: error: no previous prototype for ‘pmac_pci_fixup_ohci’
  arch/powerpc/platforms/powermac/pci.c:1057:6: error: no previous prototype for ‘pmac_pci_fixup_cardbus’
  arch/powerpc/platforms/powermac/pci.c:1094:6: error: no previous prototype for ‘pmac_pci_fixup_pciata’

Remove has_address declaration and assignment since it's not used.
Also add gcc attribute unused to fix a warning treated as error with
W=1:

  arch/powerpc/platforms/powermac/pci.c:784:19: error: variable ‘has_address’ set but not used
  arch/powerpc/platforms/powermac/pci.c:907:22: error: variable ‘ht’ set but not used

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powermac: Remove variable x that's never read
Mathieu Malaterre [Wed, 28 Mar 2018 19:30:28 +0000 (21:30 +0200)]
powerpc/powermac: Remove variable x that's never read

Since the value of x is never intended to be read, remove it. Fix
warning treated as error with W=1:

  arch/powerpc/platforms/powermac/udbg_scc.c:76:9: error: variable ‘x’ set but not used

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Reviewed-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agocxl: remove a dead branch
Mathieu Malaterre [Thu, 22 Mar 2018 21:05:28 +0000 (22:05 +0100)]
cxl: remove a dead branch

In commit 14baf4d9c739 ("cxl: Add guest-specific code") the following code
was added:

if (afu->crs_len < 0) {
dev_err(&afu->dev, "Unexpected configuration record size value\n");
return -EINVAL;
}

However the variable `crs_len` is of type u64 and cannot be compared < 0.
Remove the dead code section. Fix the following warning treated as error
with W=1:

../drivers/misc/cxl/guest.c:919:19: error: comparison of unsigned expression < 0 is always false [-Werror=type-limits]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powermac: Add missing include of header pmac.h
Mathieu Malaterre [Thu, 22 Mar 2018 20:20:02 +0000 (21:20 +0100)]
powerpc/powermac: Add missing include of header pmac.h

The header `pmac.h` was not included, leading to the following warnings,
treated as error with W=1:

  arch/powerpc/platforms/powermac/time.c:69:13: error: no previous prototype for ‘pmac_time_init’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/powermac/time.c:207:15: error: no previous prototype for ‘pmac_get_boot_time’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/powermac/time.c:222:6: error: no previous prototype for ‘pmac_get_rtc_time’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/powermac/time.c:240:5: error: no previous prototype for ‘pmac_set_rtc_time’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/powermac/time.c:259:12: error: no previous prototype for ‘via_calibrate_decr’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/powermac/time.c:311:13: error: no previous prototype for ‘pmac_calibrate_decr’ [-Werror=missing-prototypes]

The function `via_calibrate_decr` was made static to silence a warning.

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/kexec: Use common error handling code in setup_new_fdt()
Markus Elfring [Sun, 11 Mar 2018 08:03:42 +0000 (09:03 +0100)]
powerpc/kexec: Use common error handling code in setup_new_fdt()

Add a jump target so that a bit of exception handling can be better
reused at the end of this function.

This issue was detected by using the Coccinelle software.

Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Reviewed-by: Thiago Jung Bauermann <bauerman@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/xmon: Add address lookup for percpu symbols
Boqun Feng [Tue, 22 Nov 2016 09:20:09 +0000 (17:20 +0800)]
powerpc/xmon: Add address lookup for percpu symbols

Currently, in xmon, there is no obvious way to get an address for a
percpu symbol for a particular cpu. Having such an ability would be
good for debugging the system when percpu variables got involved.

Therefore, this patch introduces a new xmon command "lp" to lookup the
address for percpu symbols. Usage of "lp" is similar to "ls", except
that we could add a cpu number to choose the variable of which cpu we
want to lookup. If no cpu number is given, lookup for current cpu.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm: remove huge_pte_offset_and_shift() prototype
Christophe Leroy [Wed, 8 Aug 2018 15:36:34 +0000 (15:36 +0000)]
powerpc/mm: remove huge_pte_offset_and_shift() prototype

huge_pte_offset_and_shift() has never existed

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/lib: Use patch_site to patch copy_32 functions once cache is enabled
Christophe Leroy [Thu, 9 Aug 2018 08:14:41 +0000 (08:14 +0000)]
powerpc/lib: Use patch_site to patch copy_32 functions once cache is enabled

The symbol memcpy_nocache_branch defined in order to allow patching
of memset function once cache is enabled leads to confusing reports
by perf tool.

Using the new patch_site functionality solves this issue.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/pseries: Fix endianness while restoring of r3 in MCE handler.
Mahesh Salgaonkar [Tue, 7 Aug 2018 14:16:46 +0000 (19:46 +0530)]
powerpc/pseries: Fix endianness while restoring of r3 in MCE handler.

During Machine Check interrupt on pseries platform, register r3 points
RTAS extended event log passed by hypervisor. Since hypervisor uses r3
to pass pointer to rtas log, it stores the original r3 value at the
start of the memory (first 8 bytes) pointed by r3. Since hypervisor
stores this info and rtas log is in BE format, linux should make
sure to restore r3 value in correct endian format.

Without this patch when MCE handler, after recovery, returns to code that
that caused the MCE may end up with Data SLB access interrupt for invalid
address followed by kernel panic or hang.

  Severe Machine check interrupt [Recovered]
    NIP [d00000000ca301b8]: init_module+0x1b8/0x338 [bork_kernel]
    Initiator: CPU
    Error type: SLB [Multihit]
      Effective address: d00000000ca70000
  cpu 0xa: Vector: 380 (Data SLB Access) at [c0000000fc7775b0]
      pc: c0000000009694c0: vsnprintf+0x80/0x480
      lr: c0000000009698e0: vscnprintf+0x20/0x60
      sp: c0000000fc777830
     msr: 8000000002009033
     dar: a803a30c000000d0
    current = 0xc00000000bc9ef00
    paca    = 0xc00000001eca5c00  softe: 3  irq_happened: 0x01
      pid   = 8860, comm = insmod
  vscnprintf+0x20/0x60
  vprintk_emit+0xb4/0x4b0
  vprintk_func+0x5c/0xd0
  printk+0x38/0x4c
  init_module+0x1c0/0x338 [bork_kernel]
  do_one_initcall+0x54/0x230
  do_init_module+0x8c/0x248
  load_module+0x12b8/0x15b0
  sys_finit_module+0xa8/0x110
  system_call+0x58/0x6c
  --- Exception: c00 (System Call) at 00007fff8bda0644
  SP (7fffdfbfe980) is in userspace

This patch fixes this issue.

Fixes: a08a53ea4c97 ("powerpc/le: Enable RTAS events support")
Cc: stable@vger.kernel.org # v3.15+
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/fadump: merge adjacent memory ranges to reduce PT_LOAD segements
Hari Bathini [Mon, 6 Aug 2018 20:42:54 +0000 (02:12 +0530)]
powerpc/fadump: merge adjacent memory ranges to reduce PT_LOAD segements

With dynamic memory allocation support for crash memory ranges array,
there is no hard limit on the no. of crash memory ranges kernel could
export, but program headers count could overflow in the /proc/vmcore
ELF file while exporting each memory range as PT_LOAD segment. Reduce
the likelihood of a such scenario, by folding adjacent crash memory
ranges which minimizes the total number of PT_LOAD segments.

Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
Reviewed-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/fadump: handle crash memory ranges array index overflow
Hari Bathini [Mon, 6 Aug 2018 20:42:45 +0000 (02:12 +0530)]
powerpc/fadump: handle crash memory ranges array index overflow

Crash memory ranges is an array of memory ranges of the crashing kernel
to be exported as a dump via /proc/vmcore file. The size of the array
is set based on INIT_MEMBLOCK_REGIONS, which works alright in most cases
where memblock memory regions count is less than INIT_MEMBLOCK_REGIONS
value. But this count can grow beyond INIT_MEMBLOCK_REGIONS value since
commit 142b45a72e22 ("memblock: Add array resizing support").

On large memory systems with a few DLPAR operations, the memblock memory
regions count could be larger than INIT_MEMBLOCK_REGIONS value. On such
systems, registering fadump results in crash or other system failures
like below:

  task: c00007f39a290010 ti: c00000000b738000 task.ti: c00000000b738000
  NIP: c000000000047df4 LR: c0000000000f9e58 CTR: c00000000010f180
  REGS: c00000000b73b570 TRAP: 0300   Tainted: G          L   X  (4.4.140+)
  MSR: 8000000000009033 <SF,EE,ME,IR,DR,RI,LE>  CR: 22004484  XER: 20000000
  CFAR: c000000000008500 DAR: 000007a450000000 DSISR: 40000000 SOFTE: 0
  ...
  NIP [c000000000047df4] smp_send_reschedule+0x24/0x80
  LR [c0000000000f9e58] resched_curr+0x138/0x160
  Call Trace:
    resched_curr+0x138/0x160 (unreliable)
    check_preempt_curr+0xc8/0xf0
    ttwu_do_wakeup+0x38/0x150
    try_to_wake_up+0x224/0x4d0
    __wake_up_common+0x94/0x100
    ep_poll_callback+0xac/0x1c0
    __wake_up_common+0x94/0x100
    __wake_up_sync_key+0x70/0xa0
    sock_def_readable+0x58/0xa0
    unix_stream_sendmsg+0x2dc/0x4c0
    sock_sendmsg+0x68/0xa0
    ___sys_sendmsg+0x2cc/0x2e0
    __sys_sendmsg+0x5c/0xc0
    SyS_socketcall+0x36c/0x3f0
    system_call+0x3c/0x100

as array index overflow is not checked for while setting up crash memory
ranges causing memory corruption. To resolve this issue, dynamically
allocate memory for crash memory ranges and resize it incrementally,
in units of pagesize, on hitting array size limit.

Fixes: 2df173d9e85d ("fadump: Initialize elfcore header and add PT_LOAD program headers.")
Cc: stable@vger.kernel.org # v3.4+
Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
Reviewed-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
[mpe: Just use PAGE_SIZE directly, fixup variable placement]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/cpm1: fix compilation error with CONFIG_PPC_EARLY_DEBUG_CPM
Christophe Leroy [Mon, 6 Aug 2018 15:09:11 +0000 (15:09 +0000)]
powerpc/cpm1: fix compilation error with CONFIG_PPC_EARLY_DEBUG_CPM

commit e8cb7a55eb8dc ("powerpc: remove superflous inclusions of
asm/fixmap.h") removed inclusion of asm/fixmap.h from files not
including objects from that file.

However, asm/mmu-8xx.h includes  call to __fix_to_virt(). The proper
way would be to include asm/fixmap.h in asm/mmu-8xx.h but it creates
an inclusion loop.

So we have to leave asm/fixmap.h in sysdep/cpm_common.c for
CONFIG_PPC_EARLY_DEBUG_CPM

  CC      arch/powerpc/sysdev/cpm_common.o
In file included from ./arch/powerpc/include/asm/mmu.h:340:0,
                 from ./arch/powerpc/include/asm/reg_8xx.h:8,
                 from ./arch/powerpc/include/asm/reg.h:29,
                 from ./arch/powerpc/include/asm/processor.h:13,
                 from ./arch/powerpc/include/asm/thread_info.h:28,
                 from ./include/linux/thread_info.h:38,
                 from ./arch/powerpc/include/asm/ptrace.h:159,
                 from ./arch/powerpc/include/asm/hw_irq.h:12,
                 from ./arch/powerpc/include/asm/irqflags.h:12,
                 from ./include/linux/irqflags.h:16,
                 from ./include/asm-generic/cmpxchg-local.h:6,
                 from ./arch/powerpc/include/asm/cmpxchg.h:537,
                 from ./arch/powerpc/include/asm/atomic.h:11,
                 from ./include/linux/atomic.h:5,
                 from ./include/linux/mutex.h:18,
                 from ./include/linux/kernfs.h:13,
                 from ./include/linux/sysfs.h:16,
                 from ./include/linux/kobject.h:20,
                 from ./include/linux/device.h:16,
                 from ./include/linux/node.h:18,
                 from ./include/linux/cpu.h:17,
                 from ./include/linux/of_device.h:5,
                 from arch/powerpc/sysdev/cpm_common.c:21:
arch/powerpc/sysdev/cpm_common.c: In function ‘udbg_init_cpm’:
./arch/powerpc/include/asm/mmu-8xx.h:218:25: error: implicit declaration of function ‘__fix_to_virt’ [-Werror=implicit-function-declaration]
 #define VIRT_IMMR_BASE (__fix_to_virt(FIX_IMMR_BASE))
                         ^
arch/powerpc/sysdev/cpm_common.c:75:7: note: in expansion of macro ‘VIRT_IMMR_BASE’
       VIRT_IMMR_BASE);
       ^
./arch/powerpc/include/asm/mmu-8xx.h:218:39: error: ‘FIX_IMMR_BASE’ undeclared (first use in this function)
 #define VIRT_IMMR_BASE (__fix_to_virt(FIX_IMMR_BASE))
                                       ^
arch/powerpc/sysdev/cpm_common.c:75:7: note: in expansion of macro ‘VIRT_IMMR_BASE’
       VIRT_IMMR_BASE);
       ^
./arch/powerpc/include/asm/mmu-8xx.h:218:39: note: each undeclared identifier is reported only once for each function it appears in
 #define VIRT_IMMR_BASE (__fix_to_virt(FIX_IMMR_BASE))
                                       ^
arch/powerpc/sysdev/cpm_common.c:75:7: note: in expansion of macro ‘VIRT_IMMR_BASE’
       VIRT_IMMR_BASE);
       ^
cc1: all warnings being treated as errors
make[1]: *** [arch/powerpc/sysdev/cpm_common.o] Error 1

Fixes: e8cb7a55eb8dc ("powerpc: remove superflous inclusions of asm/fixmap.h")
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Fix size calculation using resource_size()
Dan Carpenter [Wed, 8 Aug 2018 11:57:24 +0000 (14:57 +0300)]
powerpc: Fix size calculation using resource_size()

The problem is the the calculation should be "end - start + 1" but the
plus one is missing in this calculation.

Fixes: 8626816e905e ("powerpc: add support for MPIC message register API")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Tyrel Datwyler <tyreld@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoDocumentation: Update documentation on ppc-memtrace
Rashmica Gupta [Fri, 3 Aug 2018 06:06:01 +0000 (16:06 +1000)]
Documentation: Update documentation on ppc-memtrace

Signed-off-by: Rashmica Gupta <rashmica.g@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powernv: Allow memory that has been hot-removed to be hot-added
Rashmica Gupta [Fri, 3 Aug 2018 06:06:00 +0000 (16:06 +1000)]
powerpc/powernv: Allow memory that has been hot-removed to be hot-added

This patch allows the memory removed by memtrace to be readded to the
kernel. So now you don't have to reboot your system to add the memory
back to the kernel or to have a different amount of memory removed.

Signed-off-by: Rashmica Gupta <rashmica.g@gmail.com>
Tested-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/dts/fsl: t2080rdb: use the Cortina PHY driver compatible
Camelia Groza [Wed, 18 Jul 2018 11:46:51 +0000 (14:46 +0300)]
powerpc/dts/fsl: t2080rdb: use the Cortina PHY driver compatible

The Cortina PHY is not compatible with IEEE 802.3 clause 45.

Signed-off-by: Camelia Groza <camelia.groza@nxp.com>
[scottwood: made commit message about compatibility, not driver choice]
Signed-off-by: Scott Wood <oss@buserror.net>
6 years agopowerpc/dts/fsl: t4240rdb: use the Cortina PHY driver compatible
Camelia Groza [Wed, 18 Jul 2018 11:46:50 +0000 (14:46 +0300)]
powerpc/dts/fsl: t4240rdb: use the Cortina PHY driver compatible

The Cortina PHY is not compatible with IEEE 802.3 clause 45.

Signed-off-by: Camelia Groza <camelia.groza@nxp.com>
[scottwood: made commit message about compatibility, not driver choice]
Signed-off-by: Scott Wood <oss@buserror.net>
6 years agopowerpc/configs/dpaa: enable the Cortina PHY driver
Camelia Groza [Wed, 18 Jul 2018 11:46:49 +0000 (14:46 +0300)]
powerpc/configs/dpaa: enable the Cortina PHY driver

Cortina PHYs are present on T4240RDB and T2080RDB. Enable the driver
by default.

Signed-off-by: Camelia Groza <camelia.groza@nxp.com>
Signed-off-by: Scott Wood <oss@buserror.net>
6 years agopowerpc/cpm1: fix compilation error with CONFIG_PPC_EARLY_DEBUG_CPM
Christophe Leroy [Mon, 6 Aug 2018 15:09:11 +0000 (15:09 +0000)]
powerpc/cpm1: fix compilation error with CONFIG_PPC_EARLY_DEBUG_CPM

commit e8cb7a55eb8dc ("powerpc: remove superflous inclusions of
asm/fixmap.h") removed inclusion of asm/fixmap.h from files not
including objects from that file.

However, asm/mmu-8xx.h includes  call to __fix_to_virt(). The proper
way would be to include asm/fixmap.h in asm/mmu-8xx.h but it creates
an inclusion loop.

So we have to leave asm/fixmap.h in sysdep/cpm_common.c for
CONFIG_PPC_EARLY_DEBUG_CPM

  CC      arch/powerpc/sysdev/cpm_common.o
In file included from ./arch/powerpc/include/asm/mmu.h:340:0,
                 from ./arch/powerpc/include/asm/reg_8xx.h:8,
                 from ./arch/powerpc/include/asm/reg.h:29,
                 from ./arch/powerpc/include/asm/processor.h:13,
                 from ./arch/powerpc/include/asm/thread_info.h:28,
                 from ./include/linux/thread_info.h:38,
                 from ./arch/powerpc/include/asm/ptrace.h:159,
                 from ./arch/powerpc/include/asm/hw_irq.h:12,
                 from ./arch/powerpc/include/asm/irqflags.h:12,
                 from ./include/linux/irqflags.h:16,
                 from ./include/asm-generic/cmpxchg-local.h:6,
                 from ./arch/powerpc/include/asm/cmpxchg.h:537,
                 from ./arch/powerpc/include/asm/atomic.h:11,
                 from ./include/linux/atomic.h:5,
                 from ./include/linux/mutex.h:18,
                 from ./include/linux/kernfs.h:13,
                 from ./include/linux/sysfs.h:16,
                 from ./include/linux/kobject.h:20,
                 from ./include/linux/device.h:16,
                 from ./include/linux/node.h:18,
                 from ./include/linux/cpu.h:17,
                 from ./include/linux/of_device.h:5,
                 from arch/powerpc/sysdev/cpm_common.c:21:
arch/powerpc/sysdev/cpm_common.c: In function ‘udbg_init_cpm’:
./arch/powerpc/include/asm/mmu-8xx.h:218:25: error: implicit declaration of function ‘__fix_to_virt’ [-Werror=implicit-function-declaration]
 #define VIRT_IMMR_BASE (__fix_to_virt(FIX_IMMR_BASE))
                         ^
arch/powerpc/sysdev/cpm_common.c:75:7: note: in expansion of macro ‘VIRT_IMMR_BASE’
       VIRT_IMMR_BASE);
       ^
./arch/powerpc/include/asm/mmu-8xx.h:218:39: error: ‘FIX_IMMR_BASE’ undeclared (first use in this function)
 #define VIRT_IMMR_BASE (__fix_to_virt(FIX_IMMR_BASE))
                                       ^
arch/powerpc/sysdev/cpm_common.c:75:7: note: in expansion of macro ‘VIRT_IMMR_BASE’
       VIRT_IMMR_BASE);
       ^
./arch/powerpc/include/asm/mmu-8xx.h:218:39: note: each undeclared identifier is reported only once for each function it appears in
 #define VIRT_IMMR_BASE (__fix_to_virt(FIX_IMMR_BASE))
                                       ^
arch/powerpc/sysdev/cpm_common.c:75:7: note: in expansion of macro ‘VIRT_IMMR_BASE’
       VIRT_IMMR_BASE);
       ^
cc1: all warnings being treated as errors
make[1]: *** [arch/powerpc/sysdev/cpm_common.o] Error 1

Fixes: e8cb7a55eb8dc ("powerpc: remove superflous inclusions of asm/fixmap.h")
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Scott Wood <oss@buserror.net>
6 years agoselftests/powerpc: Kill child processes on SIGINT
Breno Leitao [Tue, 7 Aug 2018 14:15:39 +0000 (11:15 -0300)]
selftests/powerpc: Kill child processes on SIGINT

There are some powerpc selftests, as tm/tm-unavailable, that run for a long
period (>120 seconds), and if it is interrupted, as pressing CRTL-C
(SIGINT), the foreground process (harness) dies but the child process and
threads continue to execute (with PPID = 1 now) in background.

In this case, you'd think the whole test exited, but there are remaining
threads and processes being executed in background. Sometimes these
zombies processes are doing annoying things, as consuming the whole CPU or
dumping things to STDOUT.

This patch fixes this problem by attaching an empty signal handler to
SIGINT in the harness process. This handler will interrupt (EINTR) the
parent process waitpid() call, letting the code to follow through the
normal flow, which will kill all the processes in the child process group.

This patch also fixes a typo.

Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Gustavo Romero <gromero@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powernv/opal: Use standard interrupts property when available
Benjamin Herrenschmidt [Tue, 10 Apr 2018 07:16:10 +0000 (17:16 +1000)]
powerpc/powernv/opal: Use standard interrupts property when available

For (bad) historical reasons, OPAL used to create a non-standard pair
of properties "opal-interrupts" and "opal-interrupts-names" for
representing the list of interrupts it wants Linux to request on its
behalf.

Among other issues, the opal-interrupts doesn't have a way to carry
the type of interrupts, and they were assumed to be all level
sensitive.

This is wrong on some recent systems where some of them are edge
sensitive causing warnings in the XIVE code and possible misbehaviours
if they need to be retriggered (typically the NPU2 TCE error
interrupts).

This makes Linux switch to using the standard "interrupts" and
"interrupt-names" properties instead when they are available, using
standard of_irq helpers, which can carry all the desired type
information.

Newer versions of OPAL will generate those properties in addition to
the legacy ones.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Fixup prefix logic to check strlen(r->name). Reinstate setting
 of start = 0 in opal_event_shutdown() to avoid double free warnings]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Allow CPU selection of e300core variants
Christophe Leroy [Thu, 7 Jun 2018 10:10:22 +0000 (10:10 +0000)]
powerpc: Allow CPU selection of e300core variants

GCC supports -mcpu=e300c2 and -mcpu=e300c3

This patch gives the opportunity to tune kernel to one of
those two types.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Allow CPU selection also on PPC32
Christophe Leroy [Thu, 7 Jun 2018 10:10:20 +0000 (10:10 +0000)]
powerpc: Allow CPU selection also on PPC32

This patch extends to PPC32 the capability to select the exact
CPU type.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Make CPU selection logic generic in Makefile
Christophe Leroy [Thu, 7 Jun 2018 10:10:18 +0000 (10:10 +0000)]
powerpc: Make CPU selection logic generic in Makefile

At the time being, when adding a new CPU for selection, both
Kconfig.cputype and Makefile have to be modified.

This patch moves into Kconfig.cputype the name of the CPU to me
passed to the -mcpu= argument.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Rename the option to TARGET_CPU to echo the gcc documentation]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/Makefiles: Convert ifeq to ifdef where possible
Rodrigo R. Galvao [Mon, 6 Aug 2018 16:42:03 +0000 (13:42 -0300)]
powerpc/Makefiles: Convert ifeq to ifdef where possible

In Makefiles if we're testing a CONFIG_FOO symbol for equality with 'y'
we can instead just use ifdef. The latter reads easily, so convert to
it where possible.

Signed-off-by: Rodrigo R. Galvao <rosattig@linux.vnet.ibm.com>
Reviewed-by: Mauro S. M. Rodrigues <maurosr@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64: Copy as much as possible in __copy_tofrom_user
Paul Mackerras [Fri, 3 Aug 2018 10:13:06 +0000 (20:13 +1000)]
powerpc/64: Copy as much as possible in __copy_tofrom_user

In __copy_tofrom_user, if we encounter an exception on a store, we
stop copying and return the number of bytes not copied.  However,
if the store is wider than one byte and is to an unaligned address,
it is possible that the store operand overlaps a page boundary
and the exception occurred on the latter part of the store operand,
meaning that it would be possible to copy a few more bytes.  Since
copy_to_user is generally expected to copy as much as possible,
it would be better to copy those extra few bytes.  This adds code
to do that.  Since this edge case is not performance-critical,
the code has been written to be compact rather than as fast as
possible.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoselftests/powerpc/64: Test exception cases in copy_tofrom_user
Michael Ellerman [Fri, 3 Aug 2018 10:13:05 +0000 (20:13 +1000)]
selftests/powerpc/64: Test exception cases in copy_tofrom_user

This adds a set of test cases to test the behaviour of
copy_tofrom_user when exceptions are encountered accessing the
source or destination.  Currently, copy_tofrom_user does not always
copy as many bytes as possible when an exception occurs on a store
to the destination, and that is reflected in failures in these tests.

Based on a test program from Anton Blanchard.

[paulus@ozlabs.org - test all three paths, wrote commit description,
 made EX_TABLE create an exception table.]

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoselftests/powerpc/64: Test all paths through copy routines
Paul Mackerras [Fri, 3 Aug 2018 10:13:04 +0000 (20:13 +1000)]
selftests/powerpc/64: Test all paths through copy routines

The hand-coded assembler 64-bit copy routines include feature sections
that select one code path or another depending on which CPU we are
executing on.  The self-tests for these copy routines end up testing
just one path.  This adds a mechanism for selecting any desired code
path at compile time, and makes 2 or 3 versions of each test, each
using a different code path, so as to cover all the possible paths.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
[mpe: Add -mcpu=power4 to CFLAGS for older compilers]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64: Make exception table clearer in __copy_tofrom_user_base
Paul Mackerras [Fri, 3 Aug 2018 10:13:03 +0000 (20:13 +1000)]
powerpc/64: Make exception table clearer in __copy_tofrom_user_base

This aims to make the generation of exception table entries for the
loads and stores in __copy_tofrom_user_base clearer and easier to
verify.  Instead of having a series of local labels on the loads and
stores, with a series of corresponding labels later for the exception
handlers, we now use macros to generate exception table entries at the
point of each load and store that could potentially trap.  We do this
with the macros lex (load exception) and stex (store exception).
These macros are used right before the load or store to which they
apply.

Some complexity is introduced by the fact that we have some more work
to do after hitting an exception, because we need to calculate and
return the number of bytes not copied.  The code uses r3 as the
current pointer into the destination buffer, that is, the address of
the first byte of the destination that has not been modified.
However, at various points in the copy loops, r3 can be 4, 8, 16 or 24
bytes behind that point.

To express this offset in an understandable way, we define a symbol
r3_offset which is updated at various points so that it equal to the
difference between the address of the first unmodified byte of the
destination and the value in r3.  (In fact it only needs to be
accurate at the point of each lex or stex macro invocation.)

The rules for updating r3_offset are as follows:

* It starts out at 0
* An addi r3,r3,N instruction decreases r3_offset by N
* A store instruction (stb, sth, stw, std) to N(r3)
  increases r3_offset by the width of the store (1, 2, 4, 8)
* A store with update instruction (stbu, sthu, stwu, stdu) to N(r3)
  sets r3_offset to the width of the store.

There is some trickiness to the way that the lex and stex macros and
the associated exception handlers work.  I would have liked to use
the current value of r3_offset in the name of the symbol used as
the exception handler, as in ".Lld_exc_$(r3_offset)" and then
have symbols .Lld_exc_0, .Lld_exc_8, .Lld_exc_16 etc. corresponding
to the offsets that needed to be added to r3.  However, I couldn't
see a way to do that with gas.

Instead, the exception handler address is .Lld_exc - r3_offset or
.Lst_exc - r3_offset, that is, the distance ahead of .Lld_exc/.Lst_exc
that we start executing is equal to the amount that we need to add to
r3.  This works because r3_offset is always a small multiple of 4,
and our instructions are 4 bytes long.  This means that before
.Lld_exc and .Lst_exc, we have a sequence of instructions that
increments r3 by 4, 8, 16 or 24 depending on where we start.  The
sequence increments r3 by 4 per instruction (on average).

We also replace the exception table for the 4k copy loop by a
macro per load or store.  These loads and stores all use exactly
the same exception handler, which simply resets the argument registers
r3, r4 and r5 to there original values and re-does the whole copy
using the slower loop.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powermac: of_node_put() is not needed after iterator
zhong jiang [Sat, 4 Aug 2018 14:25:00 +0000 (22:25 +0800)]
powerpc/powermac: of_node_put() is not needed after iterator

for_each_node_by_name() iterators only exit normally when the loop
cursor is NULL, So there is no need to call of_node_put().

Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Reviewed-by: Tyrel Datwyler <tyreld@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agocrypto/nx: Initialize 842 high and normal RxFIFO control registers
Haren Myneni [Wed, 13 Jun 2018 07:32:40 +0000 (00:32 -0700)]
crypto/nx: Initialize 842 high and normal RxFIFO control registers

NX increments readOffset by FIFO size in receive FIFO control register
when CRB is read. But the index in RxFIFO has to match with the
corresponding entry in FIFO maintained by VAS in kernel. Otherwise NX
may be processing incorrect CRBs and can cause CRB timeout.

VAS FIFO offset is 0 when the receive window is opened during
initialization. When the module is reloaded or in kexec boot, readOffset
in FIFO control register may not match with VAS entry. This patch adds
nx_coproc_init OPAL call to reset readOffset and queued entries in FIFO
control register for both high and normal FIFOs.

Signed-off-by: Haren Myneni <haren@us.ibm.com>
[mpe: Fixup uninitialized variable warning]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powernv: Export opal_check_token symbol
Haren Myneni [Wed, 13 Jun 2018 07:28:57 +0000 (00:28 -0700)]
powerpc/powernv: Export opal_check_token symbol

Export opal_check_token symbol for modules to check the availability
of OPAL calls before using them.

Signed-off-by: Haren Myneni <haren@us.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/platforms/85xx: fix t1042rdb_diu.c build errors & warning
Randy Dunlap [Sun, 15 Jul 2018 17:34:46 +0000 (10:34 -0700)]
powerpc/platforms/85xx: fix t1042rdb_diu.c build errors & warning

Fix build errors and warnings in t1042rdb_diu.c by adding header files
and MODULE_LICENSE().

../arch/powerpc/platforms/85xx/t1042rdb_diu.c:152:1: warning: data definition has no type or storage class
 early_initcall(t1042rdb_diu_init);
../arch/powerpc/platforms/85xx/t1042rdb_diu.c:152:1: error: type defaults to 'int' in declaration of 'early_initcall' [-Werror=implicit-int]
../arch/powerpc/platforms/85xx/t1042rdb_diu.c:152:1: warning: parameter names (without types) in function declaration

and
WARNING: modpost: missing MODULE_LICENSE() in arch/powerpc/platforms/85xx/t1042rdb_diu.o

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Scott Wood <oss@buserror.net>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/perf: Remove sched_task function defined for thread-imc
Anju T Sudhakar [Fri, 18 May 2018 07:35:25 +0000 (13:05 +0530)]
powerpc/perf: Remove sched_task function defined for thread-imc

Call trace observed while running perf-fuzzer:

  CPU: 43 PID: 9088 Comm: perf_fuzzer Not tainted 4.13.0-32-generic #35~lp1746225
  task: c000003f776ac900 task.stack: c000003f77728000
  NIP: c000000000299b70 LR: c0000000002a4534 CTR: c00000000029bb80
  REGS: c000003f7772b760 TRAP: 0700   Not tainted  (4.13.0-32-generic)
  MSR: 900000000282b033 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE>
    CR: 24008822  XER: 00000000
  CFAR: c000000000299a70 SOFTE: 0
  GPR00: c0000000002a4534 c000003f7772b9e0 c000000001606200 c000003fef858908
  GPR04: c000003f776ac900 0000000000000001 ffffffffffffffff 0000003fee730000
  GPR08: 0000000000000000 0000000000000000 c0000000011220d8 0000000000000002
  GPR12: c00000000029bb80 c000000007a3d900 0000000000000000 0000000000000000
  GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
  GPR20: 0000000000000000 0000000000000000 c000003f776ad090 c000000000c71354
  GPR24: c000003fef716780 0000003fee730000 c000003fe69d4200 c000003f776ad330
  GPR28: c0000000011220d8 0000000000000001 c0000000014c6108 c000003fef858900
  NIP [c000000000299b70] perf_pmu_sched_task+0x170/0x180
  LR [c0000000002a4534] __perf_event_task_sched_in+0xc4/0x230
  Call Trace:
    perf_iterate_sb+0x158/0x2a0 (unreliable)
    __perf_event_task_sched_in+0xc4/0x230
    finish_task_switch+0x21c/0x310
    __schedule+0x304/0xb80
    schedule+0x40/0xc0
    do_wait+0x254/0x2e0
    kernel_wait4+0xa0/0x1a0
    SyS_wait4+0x64/0xc0
    system_call+0x58/0x6c
  Instruction dump:
  3beafea0 7faa4800 409eff18 e8010060 eb610028 ebc10040 7c0803a6 38210050
  eb81ffe0 eba1ffe8 ebe1fff8 4e800020 <0fe000004bffffbc 60000000 60420000
  ---[ end trace 8c46856d314c1811 ]---

The context switch call-backs for thread-imc are defined in sched_task function.
So when thread-imc events are grouped with software pmu events,
perf_pmu_sched_task hits the WARN_ON_ONCE condition, since software PMUs are
assumed not to have a sched_task defined.

Patch to move the thread_imc enable/disable opal call back from sched_task to
event_[add/del] function

Fixes: f74c89bd80fb ("powerpc/perf: Add thread IMC PMU support")
Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
Reviewed-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Tested-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Fix page table fragment refcount race vs speculative references
Nicholas Piggin [Fri, 27 Jul 2018 11:48:17 +0000 (21:48 +1000)]
powerpc/64s: Fix page table fragment refcount race vs speculative references

The page table fragment allocator uses the main page refcount racily
with respect to speculative references. A customer observed a BUG due
to page table page refcount underflow in the fragment allocator. This
can be caused by the fragment allocator set_page_count stomping on a
speculative reference, and then the speculative failure handler
decrements the new reference, and the underflow eventually pops when
the page tables are freed.

Fix this by using a dedicated field in the struct page for the page
table fragment allocator.

Fixes: 5c1f6ee9a31c ("powerpc: Reduce PTE table memory wastage")
Cc: stable@vger.kernel.org # v3.10+
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agomisc: cxl: changed asterisk position
Parth Y Shah [Fri, 3 Aug 2018 10:20:38 +0000 (15:50 +0530)]
misc: cxl: changed asterisk position

Resolved <"foo* bar" should be "foo *bar"> error

Signed-off-by: Parth Y Shah <sparth1292@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/pasemi: Use pr_err/pr_warn... for kernel messages
Darren Stevens [Fri, 3 Aug 2018 11:15:10 +0000 (21:15 +1000)]
powerpc/pasemi: Use pr_err/pr_warn... for kernel messages

Pasemi code still uses printk(KERN_ERR/KERN_WARN ... change these to
pr_err(, pr_warn(... to match other powerpc arch code.

No functional changes.

Signed-off-by: Darren Stevens <darren@stevens-zone.net>
[mpe: Unsplit some strings while we're at it]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/traps: Show instructions on exceptions
Murilo Opsfelder Araujo [Wed, 1 Aug 2018 21:33:20 +0000 (18:33 -0300)]
powerpc/traps: Show instructions on exceptions

Call show_user_instructions() in arch/powerpc/kernel/traps.c to dump
instructions at faulty location, useful to debugging.

Before this patch, an unhandled signal message looked like:

  pandafault[10524]: segfault (11) at 100007d0 nip 1000061c lr 7fffbd295100 code 2 in pandafault[10000000+10000]

After this patch, it looks like:

  pandafault[10524]: segfault (11) at 100007d0 nip 1000061c lr 7fffbd295100 code 2 in pandafault[10000000+10000]
  pandafault[10524]: code: 4bfffeec 4bfffee8 3c401002 38427f00 fbe1fff8 f821ffc1 7c3f0b78 3d22fffe
  pandafault[10524]: code: 392988d0 f93f0020 e93f0020 39400048 <9949000039200000 7d234b78 383f0040

Signed-off-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Add show_user_instructions()
Murilo Opsfelder Araujo [Wed, 1 Aug 2018 21:33:19 +0000 (18:33 -0300)]
powerpc: Add show_user_instructions()

show_user_instructions() is a slightly modified version of
show_instructions() that allows userspace instruction dump.

This will be useful within show_signal_msg() to dump userspace
instructions of the faulty location.

Here is a sample of what show_user_instructions() outputs:

  pandafault[10850]: code: 4bfffeec 4bfffee8 3c401002 38427f00 fbe1fff8 f821ffc1 7c3f0b78 3d22fffe
  pandafault[10850]: code: 392988d0 f93f0020 e93f0020 39400048 <9949000039200000 7d234b78 383f0040

The current->comm and current->pid printed can serve as a glue that
links the instructions dump to its originator, allowing messages to be
interleaved in the logs.

Signed-off-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/traps: Print VMA for unhandled signals
Murilo Opsfelder Araujo [Wed, 1 Aug 2018 21:33:18 +0000 (18:33 -0300)]
powerpc/traps: Print VMA for unhandled signals

This adds VMA address in the message printed for unhandled signals,
similarly to what other architectures, like x86, print.

Before this patch, a page fault looked like:

  pandafault[61470]: unhandled signal 11 at 100007d0 nip 1000061c lr 7fff8d185100 code 2

After this patch, a page fault looks like:

  pandafault[6303]: segfault 11 at 100007d0 nip 1000061c lr 7fff93c55100 code 2 in pandafault[10000000+10000]

Signed-off-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/traps: Use %lx format in show_signal_msg()
Murilo Opsfelder Araujo [Wed, 1 Aug 2018 21:33:17 +0000 (18:33 -0300)]
powerpc/traps: Use %lx format in show_signal_msg()

Use %lx format to print registers.  This avoids having two different
formats and avoids checking for MSR_64BIT, improving readability of the
function.

Even though we could have used %px, which is functionally equivalent to %lx
as per Documentation/core-api/printk-formats.rst, it is not semantically
correct because the data printed are not pointers.  And using %px requires
casting data to (void *).

Besides that, %lx matches the format used in show_regs().

Before this patch:

  pandafault[4808]: unhandled signal 11 at 0000000010000718 nip 0000000010000574 lr 00007fff935e7a6c code 2

After this patch:

  pandafault[4732]: unhandled signal 11 at 10000718 nip 10000574 lr 7fff86697a6c code 2

Signed-off-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/traps: Use an explicit ratelimit state for show_signal_msg()
Murilo Opsfelder Araujo [Wed, 1 Aug 2018 21:33:16 +0000 (18:33 -0300)]
powerpc/traps: Use an explicit ratelimit state for show_signal_msg()

Replace printk_ratelimited() by printk() and a default rate limit
burst to limit displaying unhandled signals messages.

This will allow us to call print_vma_addr() in a future patch, which
does not work with printk_ratelimited().

Signed-off-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/traps: Print unhandled signals in a separate function
Murilo Opsfelder Araujo [Wed, 1 Aug 2018 21:33:15 +0000 (18:33 -0300)]
powerpc/traps: Print unhandled signals in a separate function

Isolate the logic of printing unhandled signals out of _exception_pkey().
No functional change, only code rearrangement.

Signed-off-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoselftests/powerpc: Add more version checks to alignment_handler test
Michael Ellerman [Tue, 31 Jul 2018 12:08:42 +0000 (22:08 +1000)]
selftests/powerpc: Add more version checks to alignment_handler test

The alignment_handler is documented to only work on Power8/Power9, but
we can make it run on older CPUs by guarding more of the tests with
feature checks.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
6 years agoselftests/powerpc: Skip earlier in alignment_handler test
Michael Ellerman [Tue, 31 Jul 2018 12:08:41 +0000 (22:08 +1000)]
selftests/powerpc: Skip earlier in alignment_handler test

Currently the alignment_handler test prints "Can't open /dev/fb0"
about 80 times per run, which is a little annoying.

Refactor it to check earlier if it can open /dev/fb0 and skip if not,
this results in each test printing something like:

  test: test_alignment_handler_vsx_206
  tags: git_version:v4.18-rc3-134-gfb21a48904aa
  [SKIP] Test skipped on line 291
  skip: test_alignment_handler_vsx_206

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
6 years agopowerpc/64s: Make rfi_flush_fallback a little more robust
Michael Ellerman [Thu, 26 Jul 2018 12:42:44 +0000 (22:42 +1000)]
powerpc/64s: Make rfi_flush_fallback a little more robust

Because rfi_flush_fallback runs immediately before the return to
userspace it currently runs with the user r1 (stack pointer). This
means if we oops in there we will report a bad kernel stack pointer in
the exception entry path, eg:

  Bad kernel stack pointer 7ffff7150e40 at c0000000000023b4
  Oops: Bad kernel stack pointer, sig: 6 [#1]
  LE SMP NR_CPUS=32 NUMA PowerNV
  Modules linked in:
  CPU: 0 PID: 1246 Comm: klogd Not tainted 4.18.0-rc2-gcc-7.3.1-00175-g0443f8a69ba3 #7
  NIP:  c0000000000023b4 LR: 0000000010053e00 CTR: 0000000000000040
  REGS: c0000000fffe7d40 TRAP: 4100   Not tainted  (4.18.0-rc2-gcc-7.3.1-00175-g0443f8a69ba3)
  MSR:  9000000002803031 <SF,HV,VEC,VSX,FP,ME,IR,DR,LE>  CR: 44000442  XER: 20000000
  CFAR: c00000000000bac8 IRQMASK: c0000000f1e66a80
  GPR00: 0000000002000000 00007ffff7150e40 00007fff93a99900 0000000000000020
  ...
  NIP [c0000000000023b4] rfi_flush_fallback+0x34/0x80
  LR [0000000010053e00] 0x10053e00

Although the NIP tells us where we were, and the TRAP number tells us
what happened, it would still be nicer if we could report the actual
exception rather than barfing about the stack pointer.

We an do that fairly simply by loading the kernel stack pointer on
entry and restoring the user value before returning. That way we see a
regular oops such as:

  Unrecoverable exception 4100 at c00000000000239c
  Oops: Unrecoverable exception, sig: 6 [#1]
  LE SMP NR_CPUS=32 NUMA PowerNV
  Modules linked in:
  CPU: 0 PID: 1251 Comm: klogd Not tainted 4.18.0-rc3-gcc-7.3.1-00097-g4ebfcac65acd-dirty #40
  NIP:  c00000000000239c LR: 0000000010053e00 CTR: 0000000000000040
  REGS: c0000000f1e17bb0 TRAP: 4100   Not tainted  (4.18.0-rc3-gcc-7.3.1-00097-g4ebfcac65acd-dirty)
  MSR:  9000000002803031 <SF,HV,VEC,VSX,FP,ME,IR,DR,LE>  CR: 44000442  XER: 20000000
  CFAR: c00000000000bac8 IRQMASK: 0
  ...
  NIP [c00000000000239c] rfi_flush_fallback+0x3c/0x80
  LR [0000000010053e00] 0x10053e00
  Call Trace:
  [c0000000f1e17e30] [c00000000000b9e4] system_call+0x5c/0x70 (unreliable)

Note this shouldn't make the kernel stack pointer vulnerable to a
meltdown attack, because it should be flushed from the cache before we
return to userspace. The user r1 value will be in the cache, because
we load it in the return path, but that is harmless.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
6 years agopowerpc/powernv: Query firmware for count cache flush settings
Michael Ellerman [Mon, 23 Jul 2018 15:07:56 +0000 (01:07 +1000)]
powerpc/powernv: Query firmware for count cache flush settings

Look for fw-features properties to determine the appropriate settings
for the count cache flush, and then call the generic powerpc code to
set it up based on the security feature flags.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/pseries: Query hypervisor for count cache flush settings
Michael Ellerman [Mon, 23 Jul 2018 15:07:55 +0000 (01:07 +1000)]
powerpc/pseries: Query hypervisor for count cache flush settings

Use the existing hypercall to determine the appropriate settings for
the count cache flush, and then call the generic powerpc code to set
it up based on the security feature flags.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Add support for software count cache flush
Michael Ellerman [Mon, 23 Jul 2018 15:07:54 +0000 (01:07 +1000)]
powerpc/64s: Add support for software count cache flush

Some CPU revisions support a mode where the count cache needs to be
flushed by software on context switch. Additionally some revisions may
have a hardware accelerated flush, in which case the software flush
sequence can be shortened.

If we detect the appropriate flag from firmware we patch a branch
into _switch() which takes us to a count cache flush sequence.

That sequence in turn may be patched to return early if we detect that
the CPU supports accelerating the flush sequence in hardware.

Add debugfs support for reporting the state of the flush, as well as
runtime disabling it.

And modify the spectre_v2 sysfs file to report the state of the
software flush.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Add new security feature flags for count cache flush
Michael Ellerman [Mon, 23 Jul 2018 15:07:53 +0000 (01:07 +1000)]
powerpc/64s: Add new security feature flags for count cache flush

Add security feature flags to indicate the need for software to flush
the count cache on context switch, and for the presence of a hardware
assisted count cache flush.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/asm: Add a patch_site macro & helpers for patching instructions
Michael Ellerman [Mon, 23 Jul 2018 15:07:52 +0000 (01:07 +1000)]
powerpc/asm: Add a patch_site macro & helpers for patching instructions

Add a macro and some helper C functions for patching single asm
instructions.

The gas macro means we can do something like:

  1: nop
   patch_site 1b, patch__foo

Which is less visually distracting than defining a GLOBAL symbol at 1,
and also doesn't pollute the symbol table which can confuse eg. perf.

These are obviously similar to our existing feature sections, but are
not automatically patched based on CPU/MMU features, rather they are
designed to be manually patched by C code at some arbitrary point.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoDocumentation: Add nospectre_v1 parameter
Diana Craciun [Fri, 27 Jul 2018 23:06:39 +0000 (09:06 +1000)]
Documentation: Add nospectre_v1 parameter

Currently only supported on powerpc.

Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/fsl: Sanitize the syscall table for NXP PowerPC 32 bit platforms
Diana Craciun [Fri, 27 Jul 2018 23:06:38 +0000 (09:06 +1000)]
powerpc/fsl: Sanitize the syscall table for NXP PowerPC 32 bit platforms

Used barrier_nospec to sanitize the syscall table.

Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/fsl: Add barrier_nospec implementation for NXP PowerPC Book3E
Diana Craciun [Fri, 27 Jul 2018 23:06:37 +0000 (09:06 +1000)]
powerpc/fsl: Add barrier_nospec implementation for NXP PowerPC Book3E

Implement the barrier_nospec as a isync;sync instruction sequence.
The implementation uses the infrastructure built for BOOK3S 64.

Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64: Make meltdown reporting Book3S 64 specific
Diana Craciun [Fri, 27 Jul 2018 23:06:36 +0000 (09:06 +1000)]
powerpc/64: Make meltdown reporting Book3S 64 specific

In a subsequent patch we will enable building security.c for Book3E.
However the NXP platforms are not vulnerable to Meltdown, so make the
Meltdown vulnerability reporting PPC_BOOK3S_64 specific.

Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64: Call setup_barrier_nospec() from setup_arch()
Michael Ellerman [Fri, 27 Jul 2018 23:06:35 +0000 (09:06 +1000)]
powerpc/64: Call setup_barrier_nospec() from setup_arch()

Currently we require platform code to call setup_barrier_nospec(). But
if we add an empty definition for the !CONFIG_PPC_BARRIER_NOSPEC case
then we can call it in setup_arch().

Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64: Add CONFIG_PPC_BARRIER_NOSPEC
Michael Ellerman [Fri, 27 Jul 2018 23:06:34 +0000 (09:06 +1000)]
powerpc/64: Add CONFIG_PPC_BARRIER_NOSPEC

Add a config symbol to encode which platforms support the
barrier_nospec speculation barrier. Currently this is just Book3S 64
but we will add Book3E in a future patch.

Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64: Make stf barrier PPC_BOOK3S_64 specific.
Diana Craciun [Fri, 27 Jul 2018 23:06:33 +0000 (09:06 +1000)]
powerpc/64: Make stf barrier PPC_BOOK3S_64 specific.

NXP Book3E platforms are not vulnerable to speculative store
bypass, so make the mitigations PPC_BOOK3S_64 specific.

Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64: Disable the speculation barrier from the command line
Diana Craciun [Fri, 27 Jul 2018 23:06:32 +0000 (09:06 +1000)]
powerpc/64: Disable the speculation barrier from the command line

The speculation barrier can be disabled from the command line
with the parameter: "nospectre_v1".

Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Don't use __MASKABLE_EXCEPTION unnecessarily
Michael Ellerman [Thu, 26 Jul 2018 13:07:17 +0000 (23:07 +1000)]
powerpc/64s: Don't use __MASKABLE_EXCEPTION unnecessarily

We only need to use __MASKABLE_EXCEPTION in one of the four cases for
hardware interrupt, so use the helper macros in the other cases.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Drop unused loc parameter to MASKABLE_EXCEPTION macros
Michael Ellerman [Thu, 26 Jul 2018 13:07:16 +0000 (23:07 +1000)]
powerpc/64s: Drop unused loc parameter to MASKABLE_EXCEPTION macros

We pass the "loc" (location) parameter to MASKABLE_EXCEPTION and
friends, but it's not used, so drop it.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Remove PSERIES naming from the MASKABLE macros
Michael Ellerman [Thu, 26 Jul 2018 13:07:15 +0000 (23:07 +1000)]
powerpc/64s: Remove PSERIES naming from the MASKABLE macros

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Drop _MASKABLE_RELON_EXCEPTION_PSERIES()
Michael Ellerman [Thu, 26 Jul 2018 13:07:14 +0000 (23:07 +1000)]
powerpc/64s: Drop _MASKABLE_RELON_EXCEPTION_PSERIES()

_MASKABLE_RELON_EXCEPTION_PSERIES() does nothing useful, update all
callers to use __MASKABLE_RELON_EXCEPTION_PSERIES() directly.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Drop _MASKABLE_EXCEPTION_PSERIES()
Michael Ellerman [Thu, 26 Jul 2018 13:07:13 +0000 (23:07 +1000)]
powerpc/64s: Drop _MASKABLE_EXCEPTION_PSERIES()

_MASKABLE_EXCEPTION_PSERIES() does nothing useful, update all callers
to use __MASKABLE_EXCEPTION_PSERIES() directly.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Rename EXCEPTION_PROLOG_PSERIES to EXCEPTION_PROLOG
Michael Ellerman [Thu, 26 Jul 2018 13:07:12 +0000 (23:07 +1000)]
powerpc/64s: Rename EXCEPTION_PROLOG_PSERIES to EXCEPTION_PROLOG

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Rename EXCEPTION_RELON_PROLOG_PSERIES
Michael Ellerman [Thu, 26 Jul 2018 13:07:11 +0000 (23:07 +1000)]
powerpc/64s: Rename EXCEPTION_RELON_PROLOG_PSERIES

To just EXCEPTION_RELON_PROLOG().

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Rename EXCEPTION_RELON_PROLOG_PSERIES_1
Michael Ellerman [Thu, 26 Jul 2018 13:07:10 +0000 (23:07 +1000)]
powerpc/64s: Rename EXCEPTION_RELON_PROLOG_PSERIES_1

The EXCEPTION_RELON_PROLOG_PSERIES_1() macro does the same job as
EXCEPTION_PROLOG_2 (which we just recently created), except for
"RELON" (relocation on) exceptions.

So rename it as such.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Remove PSERIES from the NORI macros
Michael Ellerman [Thu, 26 Jul 2018 13:07:09 +0000 (23:07 +1000)]
powerpc/64s: Remove PSERIES from the NORI macros

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Rename EXCEPTION_PROLOG_PSERIES_1 to EXCEPTION_PROLOG_2
Michael Ellerman [Thu, 26 Jul 2018 13:07:08 +0000 (23:07 +1000)]
powerpc/64s: Rename EXCEPTION_PROLOG_PSERIES_1 to EXCEPTION_PROLOG_2

As with the other patches in this series, we are removing the
"PSERIES" from the name as it's no longer meaningful.

In this case it's not simply a case of removing the "PSERIES" as that
would result in a clash with the existing EXCEPTION_PROLOG_1.

Instead we name this one EXCEPTION_PROLOG_2, as it's usually used in
sequence after 0 and 1.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Rename STD_RELON_EXCEPTION_PSERIES_OOL to STD_RELON_EXCEPTION_OOL
Michael Ellerman [Thu, 26 Jul 2018 13:07:07 +0000 (23:07 +1000)]
powerpc/64s: Rename STD_RELON_EXCEPTION_PSERIES_OOL to STD_RELON_EXCEPTION_OOL

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Rename STD_RELON_EXCEPTION_PSERIES to STD_RELON_EXCEPTION
Michael Ellerman [Thu, 26 Jul 2018 13:07:06 +0000 (23:07 +1000)]
powerpc/64s: Rename STD_RELON_EXCEPTION_PSERIES to STD_RELON_EXCEPTION

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Rename STD_EXCEPTION_PSERIES_OOL to STD_EXCEPTION_OOL
Michael Ellerman [Thu, 26 Jul 2018 13:07:05 +0000 (23:07 +1000)]
powerpc/64s: Rename STD_EXCEPTION_PSERIES_OOL to STD_EXCEPTION_OOL

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Rename STD_EXCEPTION_PSERIES to STD_EXCEPTION
Michael Ellerman [Thu, 26 Jul 2018 13:07:04 +0000 (23:07 +1000)]
powerpc/64s: Rename STD_EXCEPTION_PSERIES to STD_EXCEPTION

The "PSERIES" in STD_EXCEPTION_PSERIES is to differentiate the macros
from the legacy iSeries versions, which are called
STD_EXCEPTION_ISERIES. It is not anything to do with pseries vs
powernv or powermac etc.

We removed the legacy iSeries code in 2012, in commit 8ee3e0d69623x
("powerpc: Remove the main legacy iSerie platform code").

So remove "PSERIES" from the macros.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Move SET_SCRATCH0() into EXCEPTION_RELON_PROLOG_PSERIES()
Michael Ellerman [Thu, 26 Jul 2018 13:07:03 +0000 (23:07 +1000)]
powerpc/64s: Move SET_SCRATCH0() into EXCEPTION_RELON_PROLOG_PSERIES()

EXCEPTION_RELON_PROLOG_PSERIES() only has two users,
STD_RELON_EXCEPTION_PSERIES() and STD_RELON_EXCEPTION_HV() both of
which "call" SET_SCRATCH0(), so just move SET_SCRATCH0() into
EXCEPTION_RELON_PROLOG_PSERIES().

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/64s: Move SET_SCRATCH0() into EXCEPTION_PROLOG_PSERIES()
Michael Ellerman [Thu, 26 Jul 2018 13:07:02 +0000 (23:07 +1000)]
powerpc/64s: Move SET_SCRATCH0() into EXCEPTION_PROLOG_PSERIES()

EXCEPTION_PROLOG_PSERIES() only has two users, STD_EXCEPTION_PSERIES()
and STD_EXCEPTION_HV() both of which "call" SET_SCRATCH0(), so just
move SET_SCRATCH0() into EXCEPTION_PROLOG_PSERIES().

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/pasemi: Search for PCI root bus by compatible property
Darren Stevens [Wed, 25 Jul 2018 20:55:18 +0000 (21:55 +0100)]
powerpc/pasemi: Search for PCI root bus by compatible property

Pasemi arch code finds the root of the PCI-e bus by searching the
device-tree for a node called 'pxp'. But the root bus has a compatible
property of 'pasemi,rootbus' so search for that instead.

Signed-off-by: Darren Stevens <darren@stevens-zone.net>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoselftests/powerpc: Update strlen() test to test the new assembly function for PPC32
Christophe Leroy [Wed, 1 Aug 2018 09:01:16 +0000 (09:01 +0000)]
selftests/powerpc: Update strlen() test to test the new assembly function for PPC32

This patch adds a test for testing the new assembly strlen() for PPC32

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Fix 64-bit build]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>