linux-toradex.git/arch/arm64/mm/init.c, branch v4.9.120

arm64: fix vmemmap BUILD_BUG_ON() triggering on !vmemmap setups

2018-08-03T05:55:12+00:00

commit 7b0eb6b41a08fa1fa0d04b1c53becd62b5fbfaee upstream.

Arnd reports the following arm64 randconfig build error with the PSI
patches that add another page flag:

  /git/arm-soc/arch/arm64/mm/init.c: In function 'mem_init':
  /git/arm-soc/include/linux/compiler.h:357:38: error: call to
  '__compiletime_assert_618' declared with attribute error: BUILD_BUG_ON
  failed: sizeof(struct page) > (1 << STRUCT_PAGE_MAX_SHIFT)

The additional page flag causes other information stored in
page->flags to get bumped into their own struct page member:

  #if SECTIONS_WIDTH+ZONES_WIDTH+NODES_SHIFT+LAST_CPUPID_SHIFT <=
  BITS_PER_LONG - NR_PAGEFLAGS
  #define LAST_CPUPID_WIDTH LAST_CPUPID_SHIFT
  #else
  #define LAST_CPUPID_WIDTH 0
  #endif

  #if defined(CONFIG_NUMA_BALANCING) && LAST_CPUPID_WIDTH == 0
  #define LAST_CPUPID_NOT_IN_PAGE_FLAGS
  #endif

which in turn causes the struct page size to exceed the size set in
STRUCT_PAGE_MAX_SHIFT. This value is an an estimate used to size the
VMEMMAP page array according to address space and struct page size.

However, the check is performed - and triggers here - on a !VMEMMAP
config, which consumes an additional 22 page bits for the sparse
section id. When VMEMMAP is enabled, those bits are returned, cpupid
doesn't need its own member, and the page passes the VMEMMAP check.

Restrict that check to the situation it was meant to check: that we
are sizing the VMEMMAP page array correctly.

Says Arnd:

    Further experiments show that the build error already existed before,
    but was only triggered with larger values of CONFIG_NR_CPU and/or
    CONFIG_NODES_SHIFT that might be used in actual configurations but
    not in randconfig builds.

    With longer CPU and node masks, I could recreate the problem with
    kernels as old as linux-4.7 when arm64 NUMA support got added.

Reported-by: Arnd Bergmann 
Tested-by: Arnd Bergmann 
Cc: stable@vger.kernel.org
Fixes: 1a2db300348b ("arm64, numa: Add NUMA support for arm64 platforms.")
Fixes: 3e1907d5bf5a ("arm64: mm: move vmemmap region right below the linear region")
Signed-off-by: Johannes Weiner 
Signed-off-by: Will Deacon 
Signed-off-by: Greg Kroah-Hartman

arm64: Initialise high_memory global variable earlier

2017-12-25T13:23:36+00:00

commit f24e5834a2c3f6c5f814a417f858226f0a010ade upstream.

The high_memory global variable is used by
cma_declare_contiguous(.) before it is defined.

We don't notice this as we compute __pa(high_memory - 1), and it looks
like we're processing a VA from the direct linear map.

This problem becomes apparent when we flip the kernel virtual address
space and the linear map is moved to the bottom of the kernel VA space.

This patch moves the initialisation of high_memory before it used.

Fixes: f7426b983a6a ("mm: cma: adjust address limit to avoid hitting low/high memory boundary")
Signed-off-by: Steve Capper 
Signed-off-by: Will Deacon 
Signed-off-by: Greg Kroah-Hartman

swiotlb: Convert swiotlb_force from int to enum

2017-01-26T07:24:44+00:00

commit ae7871be189cb41184f1e05742b4a99e2c59774d upstream.

Convert the flag swiotlb_force from an int to an enum, to prepare for
the advent of more possible values.

Suggested-by: Konrad Rzeszutek Wilk 
Signed-off-by: Geert Uytterhoeven 
Signed-off-by: Konrad Rzeszutek Wilk 
Signed-off-by: Greg Kroah-Hartman

arm64: Fix swiotlb fallback allocation

2017-01-26T07:24:44+00:00

commit 524dabe1c68e0bca25ce7b108099e5d89472a101 upstream.

Commit b67a8b29df introduced logic to skip swiotlb allocation when all memory
is DMA accessible anyway.

While this is a great idea, __dma_alloc still calls swiotlb code unconditionally
to allocate memory when there is no CMA memory available. The swiotlb code is
called to ensure that we at least try get_free_pages().

Without initialization, swiotlb allocation code tries to access io_tlb_list
which is NULL. That results in a stack trace like this:

  Unable to handle kernel NULL pointer dereference at virtual address 00000000
  [...]
  [] swiotlb_tbl_map_single+0xd0/0x2b0
  [] swiotlb_alloc_coherent+0x10c/0x198
  [] __dma_alloc+0x68/0x1a8
  [] drm_gem_cma_create+0x98/0x108 [drm]
  [] drm_fbdev_cma_create_with_funcs+0xbc/0x368 [drm_kms_helper]
  [] drm_fbdev_cma_create+0x2c/0x40 [drm_kms_helper]
  [] drm_fb_helper_initial_config+0x238/0x410 [drm_kms_helper]
  [] drm_fbdev_cma_init_with_funcs+0x98/0x160 [drm_kms_helper]
  [] drm_fbdev_cma_init+0x40/0x58 [drm_kms_helper]
  [] vc4_kms_load+0x90/0xf0 [vc4]
  [] vc4_drm_bind+0xec/0x168 [vc4]
  [...]

Thankfully swiotlb code just learned how to not do allocations with the FORCE_NO
option. This patch configures the swiotlb code to use that if we decide not to
initialize the swiotlb framework.

Fixes: b67a8b29df ("arm64: mm: only initialize swiotlb when necessary")
Signed-off-by: Alexander Graf 
CC: Jisheng Zhang 
CC: Geert Uytterhoeven 
CC: Konrad Rzeszutek Wilk 
Signed-off-by: Catalin Marinas 
Signed-off-by: Greg Kroah-Hartman

arm64: remove pr_cont abuse from mem_init

2016-10-20T14:27:56+00:00

All the lines printed by mem_init are independent, with each ending with
a newline. While they logically form a large block, none are actually
continuations of previous lines.

The kernel-side printk code and the userspace demsg tool differ in their
handling of KERN_CONT following a newline, and while this isn't always a
problem kernel-side, it does cause difficulty for userspace. Using
pr_cont causes the userspace tool to not print line prefix (e.g.
timestamps) even when following a newline, mis-aligning the output and
making it harder to read, e.g.

[    0.000000] Virtual kernel memory layout:
[    0.000000]     modules : 0xffff000000000000 - 0xffff000008000000   (   128 MB)
    vmalloc : 0xffff000008000000 - 0xffff7dffbfff0000   (129022 GB)
      .text : 0xffff000008080000 - 0xffff0000088b0000   (  8384 KB)
    .rodata : 0xffff0000088b0000 - 0xffff000008c50000   (  3712 KB)
      .init : 0xffff000008c50000 - 0xffff000008d50000   (  1024 KB)
      .data : 0xffff000008d50000 - 0xffff000008e25200   (   853 KB)
       .bss : 0xffff000008e25200 - 0xffff000008e6bec0   (   284 KB)
    fixed   : 0xffff7dfffe7fd000 - 0xffff7dfffec00000   (  4108 KB)
    PCI I/O : 0xffff7dfffee00000 - 0xffff7dffffe00000   (    16 MB)
    vmemmap : 0xffff7e0000000000 - 0xffff800000000000   (  2048 GB maximum)
              0xffff7e0000000000 - 0xffff7e0026000000   (   608 MB actual)
    memory  : 0xffff800000000000 - 0xffff800980000000   ( 38912 MB)
[    0.000000] SLUB: HWalign=64, Order=0-3, MinObjects=0, CPUs=6, Nodes=1

Fix this by using pr_notice consistently for all lines, which both the
kernel and userspace are happy with.

Signed-off-by: Mark Rutland 
Cc: Catalin Marinas 
Cc: James Morse 
Cc: Kefeng Wang 
Cc: Will Deacon 
Signed-off-by: Will Deacon

arm64: mm: drop fixup_init() and mm.h

2016-09-06T18:09:38+00:00

There is only fixup_init() in mm.h , and it is only called
in free_initmem(), so move the codes from fixup_init() into
free_initmem(), then drop fixup_init() and mm.h.

Acked-by: Mark Rutland 
Signed-off-by: Kefeng Wang 
Signed-off-by: Will Deacon

arm64: apply __ro_after_init to some objects

2016-08-22T11:32:29+00:00

These objects are set during initialization, thereafter are read only.

Previously I only want to mark vdso_pages, vdso_spec, vectors_page and
cpu_ops as __read_mostly from performance point of view. Then inspired
by Kees's patch[1] to apply more __ro_after_init for arm, I think it's
better to mark them as __ro_after_init. What's more, I find some more
objects are also read only after init. So apply __ro_after_init to all
of them.

This patch also removes global vdso_pagelist and tries to clean up
vdso_spec[] assignment code.

[1] http://www.spinics.net/lists/arm-kernel/msg523188.html

Acked-by: Mark Rutland 
Signed-off-by: Jisheng Zhang 
Signed-off-by: Will Deacon

arm64:acpi: fix the acpi alignment exception when 'mem=' specified

2016-07-28T23:07:41+00:00

When booting an ACPI enabled kernel with 'mem=x', there is the
possibility that ACPI data regions from the firmware will lie above the
memory limit.  Ordinarily these will be removed by
memblock_enforce_memory_limit(.).

Unfortunately, this means that these regions will then be mapped by
acpi_os_ioremap(.) as device memory (instead of normal) thus unaligned
accessess will then provoke alignment faults.

In this patch we adopt memblock_mem_limit_remove_map instead, and this
preserves these ACPI data regions (marked NOMAP) thus ensuring that
these regions are not mapped as device memory.

For example, below is an alignment exception observed on ARM platform
when booting the kernel with 'acpi=on mem=8G':

  ...
  Unable to handle kernel paging request at virtual address ffff0000080521e7
  pgd = ffff000008aa0000
  [ffff0000080521e7] *pgd=000000801fffe003, *pud=000000801fffd003, *pmd=000000801fffc003, *pte=00e80083ff1c1707
  Internal error: Oops: 96000021 [#1] PREEMPT SMP
  Modules linked in:
  CPU: 1 PID: 1 Comm: swapper/0 Not tainted 4.7.0-rc3-next-20160616+ #172
  Hardware name: AMD Overdrive/Supercharger/Default string, BIOS ROD1001A 02/09/2016
  task: ffff800001ef0000 ti: ffff800001ef8000 task.ti: ffff800001ef8000
  PC is at acpi_ns_lookup+0x520/0x734
  LR is at acpi_ns_lookup+0x4a4/0x734
  pc : [] lr : [] pstate: 60000045
  sp : ffff800001efb8b0
  x29: ffff800001efb8c0 x28: 000000000000001b
  x27: 0000000000000001 x26: 0000000000000000
  x25: ffff800001efb9e8 x24: ffff000008a10000
  x23: 0000000000000001 x22: 0000000000000001
  x21: ffff000008724000 x20: 000000000000001b
  x19: ffff0000080521e7 x18: 000000000000000d
  x17: 00000000000038ff x16: 0000000000000002
  x15: 0000000000000007 x14: 0000000000007fff
  x13: ffffff0000000000 x12: 0000000000000018
  x11: 000000001fffd200 x10: 00000000ffffff76
  x9 : 000000000000005f x8 : ffff000008725fa8
  x7 : ffff000008a8df70 x6 : ffff000008a8df70
  x5 : ffff000008a8d000 x4 : 0000000000000010
  x3 : 0000000000000010 x2 : 000000000000000c
  x1 : 0000000000000006 x0 : 0000000000000000
  ...
    acpi_ns_lookup+0x520/0x734
    acpi_ds_load1_begin_op+0x174/0x4fc
    acpi_ps_build_named_op+0xf8/0x220
    acpi_ps_create_op+0x208/0x33c
    acpi_ps_parse_loop+0x204/0x838
    acpi_ps_parse_aml+0x1bc/0x42c
    acpi_ns_one_complete_parse+0x1e8/0x22c
    acpi_ns_parse_table+0x8c/0x128
    acpi_ns_load_table+0xc0/0x1e8
    acpi_tb_load_namespace+0xf8/0x2e8
    acpi_load_tables+0x7c/0x110
    acpi_init+0x90/0x2c0
    do_one_initcall+0x38/0x12c
    kernel_init_freeable+0x148/0x1ec
    kernel_init+0x10/0xec
    ret_from_fork+0x10/0x40
  Code: b9009fbc 2a00037b 36380057 3219037b (b9400260)
  ---[ end trace 03381e5eb0a24de4 ]---
  Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b

With 'efi=debug', we can see those ACPI regions loaded by firmware on
that board as:

  efi:   0x0083ff185000-0x0083ff1b4fff [Reserved           |   |  |  |  |  |  |  |   |WB|WT|WC|UC]*
  efi:   0x0083ff1b5000-0x0083ff1c2fff [ACPI Reclaim Memory|   |  |  |  |  |  |  |   |WB|WT|WC|UC]*
  efi:   0x0083ff223000-0x0083ff224fff [ACPI Memory NVS    |   |  |  |  |  |  |  |   |WB|WT|WC|UC]*

Link: http://lkml.kernel.org/r/1468475036-5852-3-git-send-email-dennis.chen@arm.com
Acked-by: Steve Capper 
Signed-off-by: Dennis Chen 
Cc: Catalin Marinas 
Cc: Ard Biesheuvel 
Cc: Pekka Enberg 
Cc: Mel Gorman 
Cc: Tang Chen 
Cc: Tony Luck 
Cc: Ingo Molnar 
Cc: Rafael J. Wysocki 
Cc: Will Deacon 
Cc: Mark Rutland 
Cc: Matt Fleming 
Cc: Kaly Xin 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

arm64: mm: fix location of _etext

2016-06-27T17:21:27+00:00

As Kees Cook notes in the ARM counterpart of this patch [0]:

  The _etext position is defined to be the end of the kernel text code,
  and should not include any part of the data segments. This interferes
  with things that might check memory ranges and expect executable code
  up to _etext.

In particular, Kees is referring to the HARDENED_USERCOPY patch set [1],
which rejects attempts to call copy_to_user() on kernel ranges containing
executable code, but does allow access to the .rodata segment. Regardless
of whether one may or may not agree with the distinction, it makes sense
for _etext to have the same meaning across architectures.

So let's put _etext where it belongs, between .text and .rodata, and fix
up existing references to use __init_begin instead, which unlike _end_rodata
includes the exception and notes sections as well.

The _etext references in kaslr.c are left untouched, since its references
to [_stext, _etext) are meant to capture potential jump instruction targets,
and so disregarding .rodata is actually an improvement here.

[0] http://article.gmane.org/gmane.linux.kernel/2245084
[1] http://thread.gmane.org/gmane.linux.kernel.hardened.devel/2502

Reported-by: Kees Cook 
Reviewed-by: Mark Rutland 
Signed-off-by: Ard Biesheuvel 
Reviewed-by: Kees Cook 
Signed-off-by: Catalin Marinas

arm64: mm: simplify memblock numa node extraction

2016-06-27T17:05:39+00:00

We currently open-code extracting the NUMA node of a memblock region,
which requires an ifdef to cater for !CONFIG_NUMA builds where the
memblock_region::nid field does not exist.

The generic memblock_get_region_node helper is intended to cater for
this. For CONFIG_HAVE_MEMBLOCK_NODE_MAP, builds this returns reg->nid,
and for for !CONFIG_HAVE_MEMBLOCK_NODE_MAP builds this is a static
inline that returns 0. Note that for arm64,
CONFIG_HAVE_MEMBLOCK_NODE_MAP is selected iff CONFIG_NUMA is.

This patch makes use of memblock_get_region_node to simplify the arm64
code. At the same time, we can move the nid variable definition into the
loop, as this is the only place it is used.

Signed-off-by: Mark Rutland 
Reviewed-by: Steve Capper 
Acked-by: Will Deacon 
Signed-off-by: Catalin Marinas