Add support for arbitrary bus address offset. Previously we ignored the
child (PCI) address in the "ranges" property and assumed it was always
zero. That means every host bridge window mapped to PCI bus address zero,
e.g.,
pci_bus 0000:00: root bus resource [mem 0x2000000000000-0x200007fffffff] (bus address [0x00000000-0x7fffffff])
But some systems have host bridge windows with non-zero child addresses, so
parse the child address and compute the offset between the parent (CPU) and
child (PCI) addresses. This allows windows like these:
/pci@305: PCI MEM [mem 0x2000000100000-0x200007effffff] offset 2000000000000
pci_sun4v f02ae7f8: PCI host bridge to bus 0000:00
pci_bus 0000:00: root bus resource [mem 0x2000000100000-0x200007effffff] (bus address [0x00100000-0x7effffff])
[bhelgaas: changelog]
Tested-by: Khalid Aziz <khalid.aziz@oracle.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: sparclinux@vger.kernel.org
We call pcie_aspm_exit_link_state() when we remove a device. If the device
is the last PCIe function to be removed below a bridge and the bridge has
an ASPM link_state struct, we disable ASPM on the link. Disabling ASPM
requires link->downstream (used in pcie_config_aspm_link()).
We previously set link->downstream in pcie_aspm_cap_init(), but only if the
device was not blacklisted. Removing the blacklisted device caused a NULL
pointer dereference in the pcie_aspm_exit_link_state() ->
pcie_config_aspm_link() path:
# echo 1 > /sys/bus/pci/devices/0000\:0b\:00.0/remove
...
BUG: unable to handle kernel NULL pointer dereference at 0000000000000080
IP: pcie_config_aspm_link+0x5d/0x2b0
Call Trace:
pcie_aspm_exit_link_state+0x75/0x130
pci_stop_bus_device+0xa4/0xb0
pci_stop_and_remove_bus_device_locked+0x1a/0x30
remove_store+0x50/0x70
dev_attr_store+0x18/0x30
sysfs_kf_write+0x44/0x60
kernfs_fop_write+0x10e/0x190
__vfs_write+0x28/0x110
? rcu_read_lock_sched_held+0x5d/0x80
? rcu_sync_lockdep_assert+0x2c/0x60
? __sb_start_write+0x173/0x1a0
? vfs_write+0xb3/0x180
vfs_write+0xc4/0x180
SyS_write+0x49/0xa0
do_syscall_64+0xa6/0x1c0
entry_SYSCALL64_slow_path+0x25/0x25
---[ end trace bd187ee0267df5d9 ]---
To avoid this, set link->downstream in alloc_pcie_link_state(), so every
pcie_link_state structure has a valid link->downstream pointer.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rajat Jain <rajatja@google.com>
CC: stable@vger.kernel.org
Found one megaraid_sas HBA probe fails,
[ 187.235190] scsi host2: Avago SAS based MegaRAID driver
[ 191.112365] megaraid_sas 0000:89:00.0: BAR 0: can't reserve [io 0x0000-0x00ff]
[ 191.120548] megaraid_sas 0000:89:00.0: IO memory region busy!
and the card has resource like,
[ 125.097714] pci 0000:89:00.0: [1000:005d] type 00 class 0x010400
[ 125.104446] pci 0000:89:00.0: reg 0x10: [io 0x0000-0x00ff]
[ 125.110686] pci 0000:89:00.0: reg 0x14: [mem 0xce400000-0xce40ffff 64bit]
[ 125.118286] pci 0000:89:00.0: reg 0x1c: [mem 0xce300000-0xce3fffff 64bit]
[ 125.125891] pci 0000:89:00.0: reg 0x30: [mem 0xce200000-0xce2fffff pref]
that does not io port resource allocated from BIOS, and kernel can not
assign one as io port shortage.
The driver is only looking for MEM, and should not fail.
It turns out megasas_init_fw() etc are using bar index as mask. index 1
is used as mask 1, so that pci_request_selected_regions() is trying to
request BAR0 instead of BAR1.
Fix all related reference.
Fixes: b6d5d8808b ("megaraid_sas: Use lowest memory bar for SR-IOV VF support")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Acked-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Currently the kernel image physical address randomization's lower
boundary is the original kernel load address.
For bootloaders that load kernels into very high memory (e.g. kexec),
this means randomization takes place in a very small window at the
top of memory, ignoring the large region of physical memory below
the load address.
Since mem_avoid[] is already correctly tracking the regions that must be
avoided, this patch changes the minimum address to whatever is less:
512M (to conservatively avoid unknown things in lower memory) or the
load address. Now, for example, if the kernel is loaded at 8G, [512M,
8G) will be added to the list of possible physical memory positions.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
[ Rewrote the changelog, refactored the code to use min(). ]
Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Baoquan He <bhe@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: H.J. Lu <hjl.tools@gmail.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1464216334-17200-6-git-send-email-keescook@chromium.org
[ Edited the changelog some more, plus the code comment as well. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The powerpc-specific __pci_mmap_set_pgprot() does two things:
1) Disables write combining for I/O port space mappings
This only affects procfs mappings. The pci_mmap_resource() sysfs path
only requests write combining for resources with IORESOURCE_PREFETCH
set, which doesn't include I/O resources.
The only way to request write combining for I/O port space mappings
was via the PCIIOC_WRITE_COMBINE ioctl and the proc_bus_pci_mmap()
path, and we recently changed that path to ignore write combining for
I/O, so this code in powerpc is no longer needed.
2) Automatically enables write combining for mappings of prefetchable
resources, even if not requested by the user
Both procfs (via PCIIOC_MMAP_IS_MEM and PCIIOC_WRITE_COMBINE ioctls)
and sysfs (via "resourceN_wc" files, which are created for resources
with IORESOURCE_PREFETCH) provide ways for the user to map PCI memory
space with write combining.
Users that desire write combining should use one of those ways instead
of relying on powerpc-specific behavior.
Remove the powerpc-specific __pci_mmap_set_pgprot().
The user-visible effect of this change is that powerpc users mapping
prefetchable PCI memory space via procfs without PCIIOC_WRITE_COMBINE or
via sysfs "resourceN" (not "resourceN_wc") will get regular uncacheable
mappings instead of the write combining mappings they used to get.
The new behavior matches the behavior on all other arches that support
write combining mapping.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Currently, the "run_size" variable holds the total kernel size
(size of code plus brk and bss) and is calculated via the shell script
arch/x86/tools/calc_run_size.sh. It gets the file offset and mem size
of the .bss and .brk sections from the vmlinux, and adds them as follows:
run_size = $(( $offsetA + $sizeA + $sizeB ))
However, this is not correct (it is too large). To illustrate, here's
a walk-through of the script's calculation, compared to the correct way
to find it.
First, offsetA is found as the starting address of the first .bss or
.brk section seen in the ELF file. The sizeA and sizeB values are the
respective section sizes.
[bhe@x1 linux]$ objdump -h vmlinux
vmlinux: file format elf64-x86-64
Sections:
Idx Name Size VMA LMA File off Algn
27 .bss 00170000 ffffffff81ec8000 0000000001ec8000 012c8000 2**12
ALLOC
28 .brk 00027000 ffffffff82038000 0000000002038000 012c8000 2**0
ALLOC
Here, offsetA is 0x012c8000, with sizeA at 0x00170000 and sizeB at
0x00027000. The resulting run_size is 0x145f000:
0x012c8000 + 0x00170000 + 0x00027000 = 0x145f000
However, if we instead examine the ELF LOAD program headers, we see a
different picture.
[bhe@x1 linux]$ readelf -l vmlinux
Elf file type is EXEC (Executable file)
Entry point 0x1000000
There are 5 program headers, starting at offset 64
Program Headers:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
LOAD 0x0000000000200000 0xffffffff81000000 0x0000000001000000
0x0000000000b5e000 0x0000000000b5e000 R E 200000
LOAD 0x0000000000e00000 0xffffffff81c00000 0x0000000001c00000
0x0000000000145000 0x0000000000145000 RW 200000
LOAD 0x0000000001000000 0x0000000000000000 0x0000000001d45000
0x0000000000018158 0x0000000000018158 RW 200000
LOAD 0x000000000115e000 0xffffffff81d5e000 0x0000000001d5e000
0x000000000016a000 0x0000000000301000 RWE 200000
NOTE 0x000000000099bcac 0xffffffff8179bcac 0x000000000179bcac
0x00000000000001bc 0x00000000000001bc 4
Section to Segment mapping:
Segment Sections...
00 .text .notes __ex_table .rodata __bug_table .pci_fixup .tracedata
__ksymtab __ksymtab_gpl __ksymtab_strings __init_rodata __param
__modver
01 .data .vvar
02 .data..percpu
03 .init.text .init.data .x86_cpu_dev.init .parainstructions
.altinstructions .altinstr_replacement .iommu_table .apicdrivers
.exit.text .smp_locks .bss .brk
04 .notes
As mentioned, run_size needs to be the size of the running kernel
including .bss and .brk. We can see from the Section/Segment mapping
above that .bss and .brk are included in segment 03 (which corresponds
to the final LOAD program header). To find the run_size, we calculate
the end of the LOAD segment from its PhysAddr start (0x0000000001d5e000)
and its MemSiz (0x0000000000301000), minus the physical load address of
the kernel (the first LOAD segment's PhysAddr: 0x0000000001000000). The
resulting run_size is 0x105f000:
0x0000000001d5e000 + 0x0000000000301000 - 0x0000000001000000 = 0x105f000
So, from this we can see that the existing run_size calculation is
0x400000 too high. And, as it turns out, the correct run_size is
actually equal to VO_end - VO_text, which is certainly easier to calculate.
_end: 0xffffffff8205f000
_text:0xffffffff81000000
0xffffffff8205f000 - 0xffffffff81000000 = 0x105f000
As a result, run_size is a simple constant, so we don't need to pass it
around; we already have voffset.h for such things. We can share voffset.h
between misc.c and header.S instead of getting run_size in other ways.
This patch moves voffset.h creation code to boot/compressed/Makefile,
and switches misc.c to use the VO_end - VO_text calculation for run_size.
Dependence before:
boot/header.S ==> boot/voffset.h ==> vmlinux
boot/header.S ==> compressed/vmlinux ==> compressed/misc.c
Dependence after:
boot/header.S ==> compressed/vmlinux ==> compressed/misc.c ==> boot/voffset.h ==> vmlinux
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Baoquan He <bhe@redhat.com>
[ Rewrote the changelog. ]
Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Young <dyoung@redhat.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Junjie Mao <eternal.n08@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: lasse.collin@tukaani.org
Fixes: e6023367d7 ("x86, kaslr: Prevent .bss from overlaping initrd")
Link: http://lkml.kernel.org/r/1461888548-32439-5-git-send-email-keescook@chromium.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Currently z_extract_offset is calculated in boot/compressed/mkpiggy.c.
This doesn't work well because mkpiggy.c doesn't know the details of the
decompressor in use. As a result, it can only make an estimation, which
has risks:
- output + output_len (VO) could be much bigger than input + input_len
(ZO). In this case, the decompressed kernel plus relocs could overwrite
the decompression code while it is running.
- The head code of ZO could be bigger than z_extract_offset. In this case
an overwrite could happen when the head code is running to move ZO to
the end of buffer. Though currently the size of the head code is very
small it's still a potential risk. Since there is no rule to limit the
size of the head code of ZO, it runs the risk of suddenly becoming a
(hard to find) bug.
Instead, this moves the z_extract_offset calculation into header.S, and
makes adjustments to be sure that the above two cases can never happen,
and further corrects the comments describing the calculations.
Since we have (in the previous patch) made ZO always be located against
the end of decompression buffer, z_extract_offset is only used here to
calculate an appropriate buffer size (INIT_SIZE), and is not longer used
elsewhere. As such, it can be removed from voffset.h.
Additionally clean up #if/#else #define to improve readability.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Baoquan He <bhe@redhat.com>
[ Rewrote the changelog and comments. ]
Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Young <dyoung@redhat.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: lasse.collin@tukaani.org
Link: http://lkml.kernel.org/r/1461888548-32439-4-git-send-email-keescook@chromium.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This change makes later calculations about where the kernel is located
easier to reason about. To better understand this change, we must first
clarify what 'VO' and 'ZO' are. These values were introduced in commits
by hpa:
77d1a49995 ("x86, boot: make symbols from the main vmlinux available")
37ba7ab5e3 ("x86, boot: make kernel_alignment adjustable; new bzImage fields")
Specifically:
All names prefixed with 'VO_':
- relate to the uncompressed kernel image
- the size of the VO image is: VO__end-VO__text ("VO_INIT_SIZE" define)
All names prefixed with 'ZO_':
- relate to the bootable compressed kernel image (boot/compressed/vmlinux),
which is composed of the following memory areas:
- head text
- compressed kernel (VO image and relocs table)
- decompressor code
- the size of the ZO image is: ZO__end - ZO_startup_32 ("ZO_INIT_SIZE" define, though see below)
The 'INIT_SIZE' value is used to find the larger of the two image sizes:
#define ZO_INIT_SIZE (ZO__end - ZO_startup_32 + ZO_z_extract_offset)
#define VO_INIT_SIZE (VO__end - VO__text)
#if ZO_INIT_SIZE > VO_INIT_SIZE
# define INIT_SIZE ZO_INIT_SIZE
#else
# define INIT_SIZE VO_INIT_SIZE
#endif
The current code uses extract_offset to decide where to position the
copied ZO (i.e. ZO starts at extract_offset). (This is why ZO_INIT_SIZE
currently includes the extract_offset.)
Why does z_extract_offset exist? It's needed because we are trying to minimize
the amount of RAM used for the whole act of creating an uncompressed, executable,
properly relocation-linked kernel image in system memory. We do this so that
kernels can be booted on even very small systems.
To achieve the goal of minimal memory consumption we have implemented an in-place
decompression strategy: instead of cleanly separating the VO and ZO images and
also allocating some memory for the decompression code's runtime needs, we instead
create this elaborate layout of memory buffers where the output (decompressed)
stream, as it progresses, overlaps with and destroys the input (compressed)
stream. This can only be done safely if the ZO image is placed to the end of the
VO range, plus a certain amount of safety distance to make sure that when the last
bytes of the VO range are decompressed, the compressed stream pointer is safely
beyond the end of the VO range.
z_extract_offset is calculated in arch/x86/boot/compressed/mkpiggy.c during
the build process, at a point when we know the exact compressed and
uncompressed size of the kernel images and can calculate this safe minimum
offset value. (Note that the mkpiggy.c calculation is not perfect, because
we don't know the decompressor used at that stage, so the z_extract_offset
calculation is necessarily imprecise and is mostly based on gzip internals -
we'll improve that in the next patch.)
When INIT_SIZE is bigger than VO_INIT_SIZE (uncommon but possible),
the copied ZO occupies the memory from extract_offset to the end of
decompression buffer. It overlaps with the soon-to-be-uncompressed kernel
like this:
|-----compressed kernel image------|
V V
0 extract_offset +INIT_SIZE
|-----------|---------------|-------------------------|--------|
| | | |
VO__text startup_32 of ZO VO__end ZO__end
^ ^
|-------uncompressed kernel image---------|
When INIT_SIZE is equal to VO_INIT_SIZE (likely) there's still space
left from end of ZO to the end of decompressing buffer, like below.
|-compressed kernel image-|
V V
0 extract_offset +INIT_SIZE
|-----------|---------------|-------------------------|--------|
| | | |
VO__text startup_32 of ZO ZO__end VO__end
^ ^
|------------uncompressed kernel image-------------|
To simplify calculations and avoid special cases, it is cleaner to
always place the compressed kernel image in memory so that ZO__end
is at the end of the decompression buffer, instead of placing t at
the start of extract_offset as is currently done.
This patch adds BP_init_size (which is the INIT_SIZE as passed in from
the boot_params) into asm-offsets.c to make it visible to the assembly
code.
Then when moving the ZO, it calculates the starting position of
the copied ZO (via BP_init_size and the ZO run size) so that the VO__end
will be at the end of the decompression buffer. To make the position
calculation safe, the end of ZO is page aligned (and a comment is added
to the existing VO alignment for good measure).
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
[ Rewrote changelog and comments. ]
Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Baoquan He <bhe@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Young <dyoung@redhat.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: lasse.collin@tukaani.org
Link: http://lkml.kernel.org/r/1461888548-32439-3-git-send-email-keescook@chromium.org
[ Rewrote the changelog some more. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
David reported that a T5-8 sparc system failed to boot with:
pci_sun4v f02dbcfc: PCI host bridge to bus 0000:00
pci_bus 0000:00: root bus resource [io 0x804000000000-0x80400fffffff] (bus address [0x0000-0xfffffff])
pci_bus 0000:00: root bus resource [mem 0x800000000000-0x80007effffff] (bus address [0x00000000-0x7effffff])
pci 0000:00:01.0: can't claim BAR 15 [mem 0x100000000-0x4afffffff pref]: no compatible bridge window
Note that we don't know about a host bridge aperture that contains
BAR 15. OF does report a MEM64 aperture, but before this patch,
pci_determine_mem_io_space() ignored it.
Add support for host bridge apertures with 64-bit PCI addresses. Also
set IORESOURCE_MEM_64 for PCI device and bridge resources in PCI 64-bit
memory space.
Sparc doesn't actually print the device and bridge resources, but after
this patch, we should have the equivalent of this:
pci_sun4v f02dbcfc: PCI host bridge to bus 0000:00
pci_bus 0000:00: root bus resource [io 0x804000000000-0x80400fffffff] (bus address [0x0000-0xfffffff])
pci_bus 0000:00: root bus resource [mem 0x800000000000-0x80007effffff] (bus address [0x00000000-0x7effffff])
pci_bus 0000:00: root bus resource [mem 0x800100000000-0x8007ffffffff] (bus address [0x100000000-0x7ffffffff])
pci 0000:00:01.0: bridge window [mem 0x800100000000-0x8004afffffff 64bit pref]
[bhelgaas: changelog, URL to David's report]
Fixes: d63e2e1f3d ("sparc/PCI: Clip bridge windows to fit in upstream windows")
Link: http://lkml.kernel.org/r/5514391F.2030300@oracle.com
Reported-by: David Ahern <david.ahern@oracle.com>
Tested-by: David Ahern <david.ahern@oracle.com>
Tested-by: Khalid Aziz <khalid.aziz@oracle.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Tony found on his setup, if memory block size 512M will cause crash
during booting.
BUG: unable to handle kernel paging request at ffffea0074000020
IP: [<ffffffff81670527>] get_nid_for_pfn+0x17/0x40
PGD 128ffcb067 PUD 128ffc9067 PMD 0
Oops: 0000 [#1] SMP
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.2.0-rc8 #1
...
Call Trace:
[<ffffffff81453b56>] ? register_mem_sect_under_node+0x66/0xe0
[<ffffffff81453eeb>] register_one_node+0x17b/0x240
[<ffffffff81b1f1ed>] ? pci_iommu_alloc+0x6e/0x6e
[<ffffffff81b1f229>] topology_init+0x3c/0x95
[<ffffffff8100213d>] do_one_initcall+0xcd/0x1f0
The system has non continuous RAM address:
BIOS-e820: [mem 0x0000001300000000-0x0000001cffffffff] usable
BIOS-e820: [mem 0x0000001d70000000-0x0000001ec7ffefff] usable
BIOS-e820: [mem 0x0000001f00000000-0x0000002bffffffff] usable
BIOS-e820: [mem 0x0000002c18000000-0x0000002d6fffefff] usable
BIOS-e820: [mem 0x0000002e00000000-0x00000039ffffffff] usable
So there are start sections in memory block not present.
For example:
memory block : [0x2c18000000, 0x2c20000000) 512M
first three sections are not present.
Current register_mem_sect_under_node() assume first section is present,
but memory block section number range [start_section_nr, end_section_nr]
would include not present section.
For arch that support vmemmap, we don't setup memmap for struct page area
within not present sections area.
So skip the pfn range that belong to absent section.
Also fixes unregister_mem_sect_under_nodes() that assume one section per
memory block.
Reported-by: Tony Luck <tony.luck@intel.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Fixes: bdee237c03 ("x86: mm: Use 2GB memory block size on large memory x86-64 systems")
Fixes: 982792c782 ("x86, mm: probe memory block size for generic x86 64bit")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: stable@vger.kernel.org #v3.15
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When loading x86 64bit kernel above 4GiB with patched grub2, got kernel
gunzip error.
| early console in decompress_kernel
| decompress_kernel:
| input: [0x807f2143b4-0x807ff61aee]
| output: [0x807cc00000-0x807f3ea29b] 0x027ea29c: output_len
| boot via startup_64
| KASLR using RDTSC...
| new output: [0x46fe000000-0x470138cfff] 0x0338d000: output_run_size
| decompress: [0x46fe000000-0x47007ea29b] <=== [0x807f2143b4-0x807ff61aee]
|
| Decompressing Linux... gz...
|
| uncompression error
|
| -- System halted
the new buffer is at 0x46fe000000ULL, decompressor_gzip is using
0xffffffb901ffffff as out_len. gunzip in lib/zlib_inflate/inflate.c cap
that len to 0x01ffffff and decompress fails later.
We could hit this problem with crashkernel booting that uses kexec loading
kernel above 4GiB.
We have decompress_* support:
1. inbuf[]/outbuf[] for kernel preboot.
2. inbuf[]/flush() for initramfs
3. fill()/flush() for initrd.
This bug only affect kernel preboot path that use outbuf[].
Add __decompress and take real out_buf_len for gunzip instead of guessing
wrong buf size.
Fixes: 1431574a1c (lib/decompressors: fix "no limit" output buffer length)
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Alexandre Courbot <acourbot@nvidia.com>
Cc: Jon Medhurst <tixy@linaro.org>
Cc: Stephen Warren <swarren@wwwdotorg.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Tony Luck found on his setup, if memory block size 512M will cause crash
during booting.
BUG: unable to handle kernel paging request at ffffea0074000020
IP: get_nid_for_pfn+0x17/0x40
PGD 128ffcb067 PUD 128ffc9067 PMD 0
Oops: 0000 [#1] SMP
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.2.0-rc8 #1
...
Call Trace:
? register_mem_sect_under_node+0x66/0xe0
register_one_node+0x17b/0x240
? pci_iommu_alloc+0x6e/0x6e
topology_init+0x3c/0x95
do_one_initcall+0xcd/0x1f0
The system has non continuous RAM address:
BIOS-e820: [mem 0x0000001300000000-0x0000001cffffffff] usable
BIOS-e820: [mem 0x0000001d70000000-0x0000001ec7ffefff] usable
BIOS-e820: [mem 0x0000001f00000000-0x0000002bffffffff] usable
BIOS-e820: [mem 0x0000002c18000000-0x0000002d6fffefff] usable
BIOS-e820: [mem 0x0000002e00000000-0x00000039ffffffff] usable
So there are start sections in memory block not present. For example:
memory block : [0x2c18000000, 0x2c20000000) 512M
first three sections are not present.
The current register_mem_sect_under_node() assume first section is
present, but memory block section number range [start_section_nr,
end_section_nr] would include not present section.
For arch that support vmemmap, we don't setup memmap for struct page
area within not present sections area.
So skip the pfn range that belong to absent section.
[akpm@linux-foundation.org: simplification]
[rientjes@google.com: more simplification]
Fixes: bdee237c03 ("x86: mm: Use 2GB memory block size on large memory x86-64 systems")
Fixes: 982792c782 ("x86, mm: probe memory block size for generic x86 64bit")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Reported-by: Tony Luck <tony.luck@intel.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Cc: Greg KH <greg@kroah.com>
Cc: Ingo Molnar <mingo@elte.hu>
Tested-by: David Rientjes <rientjes@google.com>
Cc: <stable@vger.kernel.org> [3.15+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In d74b9027a4 ("PCI: Consider additional PF's IOV BAR alignment in sizing
and assigning"), we store additional alignment in realloc_head and take
this into consideration for assignment.
In __assign_resources_sorted(), we changed dev_res->res->start, then used
resource_start() (which depends on res->start), so the recomputed res->end
was completely bogus. Even if we'd had the correct size, the end would
have been off by one.
Preserve the resource size when we adjust its alignment.
[bhelgaas: changelog]
Fixes: d74b9027a4 ("PCI: Consider additional PF's IOV BAR alignment in sizing and assigning")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Wei Yang <weiyang@linux.vnet.ibm.com>
CC: Benjamin Herrenschmidt <benh@kernel.crashing.org>
David Ahern reported that d63e2e1f3d ("sparc/PCI: Clip bridge windows
to fit in upstream windows") fails to boot on sparc/T5-8:
pci 0000:06:00.0: reg 0x184: can't handle BAR above 4GB (bus address 0x110204000)
The problem is that sparc64 assumed that dma_addr_t only needed to hold DMA
addresses, i.e., bus addresses returned via the DMA API (dma_map_single(),
etc.), while the PCI core assumed dma_addr_t could hold *any* bus address,
including raw BAR values. On sparc64, all DMA addresses fit in 32 bits, so
dma_addr_t is a 32-bit type. However, BAR values can be 64 bits wide, so
they don't fit in a dma_addr_t. d63e2e1f3d added new checking that
tripped over this mismatch.
Add pci_bus_addr_t, which is wide enough to hold any PCI bus address,
including both raw BAR values and DMA addresses. This will be 64 bits
on 64-bit platforms and on platforms with a 64-bit dma_addr_t. Then
dma_addr_t only needs to be wide enough to hold addresses from the DMA API.
[bhelgaas: changelog, bugzilla, Kconfig to ensure pci_bus_addr_t is at
least as wide as dma_addr_t, documentation]
Fixes: d63e2e1f3d ("sparc/PCI: Clip bridge windows to fit in upstream windows")
Fixes: 23b13bc76f ("PCI: Fail safely if we can't handle BARs larger than 4GB")
Link: http://lkml.kernel.org/r/CAE9FiQU1gJY1LYrxs+ma5LCTEEe4xmtjRG0aXJ9K_Tsu+m9Wuw@mail.gmail.com
Link: http://lkml.kernel.org/r/1427857069-6789-1-git-send-email-yinghai@kernel.org
Link: https://bugzilla.kernel.org/show_bug.cgi?id=96231
Reported-by: David Ahern <david.ahern@oracle.com>
Tested-by: David Ahern <david.ahern@oracle.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: David S. Miller <davem@davemloft.net>
CC: stable@vger.kernel.org # v3.19+
While adding support loading kernel and initrd above 4G to grub2 in legacy
mode, I was referring to efi_high_alloc().
That will allocate buffer for kernel and then initrd, and initrd will
use kernel buffer start as limit.
During testing found two buffers will be overlapped when initrd size is
very big like 400M.
It turns out efi_high_alloc() boundary checking is not right.
end - size will be the new start, and should not compare new
start with max, we need to make sure end is smaller than max.
[ Basically, with the current efi_high_alloc() code it's possible to
allocate memory above 'max', because efi_high_alloc() doesn't check
that the tail of the allocation is below 'max'.
If you have an EFI memory map with a single entry that looks like so,
[0xc0000000-0xc0004000]
And want to allocate 0x3000 bytes below 0xc0003000 the current code
will allocate [0xc0001000-0xc0004000], not [0xc0000000-0xc0003000]
like you would expect. - Matt ]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Every PCI-PCI bridge window should fit inside an upstream bridge window
because orphaned address space is unreachable from the primary side of the
upstream bridge. If we inherit invalid bridge windows that overlap an
upstream window from firmware, clip them to fit and update the bridge
accordingly.
[bhelgaas: changelog]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=85491
Reported-by: Marek Kordik <kordikmarek@gmail.com>
Fixes: 5b28541552 ("PCI: Restrict 64-bit prefetchable bridge windows to 64-bit resources")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: "James E.J. Bottomley" <jejb@parisc-linux.org>
CC: Helge Deller <deller@gmx.de>
CC: linux-parisc@vger.kernel.org
Every PCI-PCI bridge window should fit inside an upstream bridge window
because orphaned address space is unreachable from the primary side of the
upstream bridge. If we inherit invalid bridge windows that overlap an
upstream window from firmware, clip them to fit and update the bridge
accordingly.
[bhelgaas: changelog]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=85491
Reported-by: Marek Kordik <kordikmarek@gmail.com>
Fixes: 5b28541552 ("PCI: Restrict 64-bit prefetchable bridge windows to 64-bit resources")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: Michal Simek <monstr@monstr.eu>
CC: Benjamin Herrenschmidt <benh@kernel.crashing.org>
CC: Sebastian Ott <sebott@linux.vnet.ibm.com>
Every PCI-PCI bridge window should fit inside an upstream bridge window
because orphaned address space is unreachable from the primary side of the
upstream bridge. If we inherit invalid bridge windows that overlap an
upstream window from firmware, clip them to fit and update the bridge
accordingly.
[bhelgaas: changelog]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=85491
Reported-by: Marek Kordik <kordikmarek@gmail.com>
Fixes: 5b28541552 ("PCI: Restrict 64-bit prefetchable bridge windows to 64-bit resources")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: David Howells <dhowells@redhat.com>
CC: Paul Gortmaker <paul.gortmaker@windriver.com>
Add pci_claim_bridge_resource() to claim a PCI-PCI bridge window. This is
like regular pci_claim_resource(), except that if we fail to claim the
window, we check to see if we can reduce the size of the window and try
again.
This is for scenarios like this:
pci_bus 0000:00: root bus resource [mem 0xc0000000-0xffffffff]
pci 0000:00:01.0: bridge window [mem 0xbdf00000-0xddefffff 64bit pref]
pci 0000:01:00.0: reg 0x10: [mem 0xc0000000-0xcfffffff pref]
The 00:01.0 window is illegal: it starts before the host bridge window, so
we have to assume the [0xbdf00000-0xbfffffff] region is inaccessible. We
can make it legal by clipping it to [mem 0xc0000000-0xddefffff 64bit pref].
Previously we discarded the 00:01.0 window and tried to reassign that part
of the hierarchy from scratch. That is a problem because Linux doesn't
always assign things optimally. For example, in this case, BIOS put the
01:00.0 device in a prefetchable window below 4GB, but after 5b28541552,
Linux puts the prefetchable window above 4GB where the 32-bit 01:00.0
device can't use it.
Clipping the 00:01.0 window is less intrusive than completely reassigning
things and is sufficient to let us use most of the BIOS configuration. Of
course, it's possible that devices below 00:01.0 will no longer fit. If
that's the case, we'll have to reassign things. But that's a separate
problem.
[bhelgaas: changelog, split into separate patch]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=85491
Reported-by: Marek Kordik <kordikmarek@gmail.com>
Fixes: 5b28541552 ("PCI: Restrict 64-bit prefetchable bridge windows to 64-bit resources")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v3.16+
Add pci_bus_clip_resource(). If a PCI-PCI bridge window overlaps an
upstream bridge window but is not completely contained by it, this clips
the downstream window so it fits inside the upstream one.
No functional change (this adds the function but no callers).
[bhelgaas: changelog, split into separate patch]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=85491
Reported-by: Marek Kordik <kordikmarek@gmail.com>
Fixes: 5b28541552 ("PCI: Restrict 64-bit prefetchable bridge windows to 64-bit resources")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v3.16+
pci_setup_bridge_io(), pci_setup_bridge_mmio(), and
pci_setup_bridge_mmio_pref() program the windows of PCI-PCI bridges.
Previously they accepted a pointer to the pci_bus of the secondary bus,
then looked up the bridge leading to that bus. Pass the bridge directly,
which will make it more convenient for future callers.
No functional change.
[bhelgaas: changelog, split into separate patch]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=85491
Reported-by: Marek Kordik <kordikmarek@gmail.com>
Fixes: 5b28541552 ("PCI: Restrict 64-bit prefetchable bridge windows to 64-bit resources")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v3.16+
Aaron reported that a 32-bit x86 kernel with Physical Address Extension
(PAE) support complains about bridge prefetchable memory windows above 4GB:
pci_bus 0000:00: root bus resource [mem 0x380000000000-0x383fffffffff]
...
pci 0000:03:00.0: reg 0x10: [mem 0x383fffc00000-0x383fffdfffff 64bit pref]
pci 0000:03:00.0: reg 0x20: [mem 0x383fffe04000-0x383fffe07fff 64bit pref]
pci 0000:03:00.1: reg 0x10: [mem 0x383fffa00000-0x383fffbfffff 64bit pref]
pci 0000:03:00.1: reg 0x20: [mem 0x383fffe00000-0x383fffe03fff 64bit pref]
pci 0000:00:02.2: PCI bridge to [bus 03-04]
pci 0000:00:02.2: bridge window [io 0x1000-0x1fff]
pci 0000:00:02.2: bridge window [mem 0x91900000-0x91cfffff]
pci 0000:00:02.2: can't handle 64-bit address space for bridge
In this kernel, unsigned long is 32 bits and dma_addr_t is 64 bits.
Previously we used "unsigned long" to hold the bridge window address. But
this is a bus address, so we should use dma_addr_t instead.
Use dma_addr_t to hold the bridge window base and limit.
The question of whether the CPU can actually *address* the window is
separate and depends on what the physical address space of the CPU is and
whether the host bridge does any address translation.
[bhelgaas: fix "shift count > width of type", changelog, stable tag]
Fixes: d56dbf5bab ("PCI: Allocate 64-bit BARs above 4G when possible")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=88131
Reported-by: Aaron Ma <mapengyu@gmail.com>
Tested-by: Aaron Ma <mapengyu@gmail.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v3.14+
Previously we applied _HPX type 2 record Link Control register settings
only to bridges with a subordinate bus. But it's better to apply them to
all devices with a link because if the subordinate bus has not been
allocated yet, we won't apply settings to the device.
Use pcie_cap_has_lnkctl() to determine whether the device has a Link
Control register instead of looking at dev->subordinate.
[bhelgaas: changelog]
Fixes: 6cd33649fa ("PCI: Add pci_configure_device() during enumeration")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
acpi_pci_get_bridge_handle() returns the ACPI handle for the bridge device
(either a host bridge or a PCI-to-PCI bridge) leading to a PCI bus. But
SR-IOV virtual functions can be on a virtual bus with no bridge leading to
it. Return a NULL acpi_handle in this case instead of trying to
dereference the NULL pointer to the bridge.
This fixes a NULL pointer dereference oops in pci_get_hp_params() when
adding SR-IOV VF devices on virtual buses.
[bhelgaas: changelog, add comment in code]
Fixes: 6cd33649fa ("PCI: Add pci_configure_device() during enumeration")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=87591
Reported-by: Chao Zhou <chao.zhou@intel.com>
Reported-by: Joerg Roedel <joro@8bytes.org>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
In 5b28541552 ("PCI: Restrict 64-bit prefetchable bridge windows to
64-bit resources"), we added IORESOURCE_MEM_64 to the mask in
pci_assign_unassigned_root_bus_resources(), but not to the mask in
pci_assign_unassigned_bridge_resources().
Add IORESOURCE_MEM_64 to the pci_assign_unassigned_bridge_resources() type
mask.
Fixes: 5b28541552 ("PCI: Restrict 64-bit prefetchable bridge windows to 64-bit resources")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v3.16+
During pciehp initialization, we previously wrote two hotplug commands:
pciehp_probe
pcie_init
pcie_disable_notification
pcie_write_cmd # command 1
pcie_init_notification
pcie_enable_notification
pcie_write_cmd # command 2
For controllers with errata like Intel CF118, we previously waited for a
timeout before issuing the second hotplug command because the first command
only updates interrupt enable bits and is not a "real" hotplug command, so
the controller doesn't report Command Completed for it.
But there's no need to disable notifications in the first place. If BIOS
left them enabled, we could easily take an interrupt before disabling them,
so there's no benefit in disabling them for the tiny window before we
enable them.
Drop the unnecessary pcie_disable_notification() call.
[bhelgaas: changelog]
Link: http://www.intel.com/content/www/us/en/processors/xeon/xeon-e7-v2-spec-update.html
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Add more Slot Control debug output and move one print after
pcie_write_cmd() to be consistent with other debug output.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
When we warned about a timeout on a hotplug command, we previously printed
the time between calls to pcie_write_cmd(), without accounting for any time
spent actually waiting. Consider this sequence:
pcie_write_cmd
write SLTCTL
cmd_started = jiffies # T1
pcie_write_cmd
pcie_wait_cmd
now = jiffies # T2
wait_event_timeout # we may wait here
if (timeout)
ctrl_info("Timeout on command issued %u msec ago",
jiffies_to_msecs(now - cmd_started))
We previously printed (T2 - T1), but that doesn't include the time spent in
wait_event_timeout().
Fix this by using the current jiffies value, not the one cached before
calling wait_event_timeout().
[bhelgaas: changelog, use current jiffies instead of adding timeout]
Fixes: 40b960831c ("PCI: pciehp: Compute timeout from hotplug command start time")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
pcie_poll_cmd() take msecs instead of jiffies, so convert timeout to msecs.
Fixes: 40b960831c ("PCI: pciehp: Compute timeout from hotplug command start time")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Mantas found that after commit 4bf7111f50 ("x86/efi: Support initrd
loaded above 4G"), the kernel freezes at the earliest possible moment
when trying to boot via UEFI on Asus laptop.
Revert to old way to load initrd under 4G on first try, second try will
use above 4G buffer when initrd is too big and does not fit under 4G.
[ The cause of the freeze appears to be a firmware bug when reading
file data into buffers above 4GB, though the exact reason is unknown.
Mantas reports that the hang can be avoid if the file size is a
multiple of 512 bytes, but I've seen some ASUS firmware simply
corrupting the file data rather than freezing.
Laszlo fixed an issue in the upstream EDK2 DiskIO code in Aug 2013
which may possibly be related, commit 4e39b75e ("MdeModulePkg/DiskIoDxe:
fix source/destination pointer of overrun transfer").
Whatever the cause, it's unlikely that a fix will be forthcoming
from the vendor, hence the workaround - Matt ]
Cc: Laszlo Ersek <lersek@redhat.com>
Reported-by: Mantas Mikulėnas <grawity@gmail.com>
Reported-by: Harald Hoyer <harald@redhat.com>
Tested-by: Anders Darander <anders@chargestorm.se>
Tested-by: Calvin Walton <calvin.walton@kepstin.ca>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Now with 64bit bzImage and kexec tools, we support ramdisk that size is
bigger than 2g, as we could put it above 4G.
Found compressed initramfs image could not be decompressed properly. It
turns out that image length is int during decompress detection, and it
will become < 0 when length is more than 2G. Furthermore, during
decompressing len as int is used for inbuf count, that has problem too.
Change len to long, that should be ok as on 32 bit platform long is
32bits.
Tested with following compressed initramfs image as root with kexec.
gzip, bzip2, xz, lzma, lzop, lz4.
run time for populate_rootfs():
size name Nehalem-EX Westmere-EX Ivybridge-EX
9034400256 root_img : 26s 24s 30s
3561095057 root_img.lz4 : 28s 27s 27s
3459554629 root_img.lzo : 29s 29s 28s
3219399480 root_img.gz : 64s 62s 49s
2251594592 root_img.xz : 262s 260s 183s
2226366598 root_img.lzma: 386s 376s 277s
2901482513 root_img.bz2 : 635s 599s
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Rashika Kheria <rashika.kheria@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Kyungsik Lee <kyungsik.lee@lge.com>
Cc: P J P <ppandit@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc: "Daniel M. Weeks" <dan@danweeks.net>
Cc: Alexandre Courbot <acourbot@nvidia.com>
Cc: Jan Beulich <JBeulich@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When initrd (compressed or not) is used, kernel report data corrupted with
/dev/ram0.
The root cause:
During initramfs checking, if it is initrd, it will be transferred to
/initrd.image with sys_write.
sys_write only support 2G-4K write, so if the initrd ram is more than
that, /initrd.image will not complete at all.
Add local xwrite to loop calling sys_write to workaround the problem.
Also need to use xwrite in write_buffer() to handle:
image is uncompressed cpio and there is one big file (>2G) in it.
unpack_to_rootfs ===> write_buffer ===> actions[]/do_copy
At the same time, we don't need to worry about sys_read/sys_write in
do_mounts_rd.c::crd_load. As decompressor will have fill/flush and local
buffer that is smaller than 2G.
Test with uncompressed initrd, and compressed ones with gz, bz2, lzma,xz,
lzop.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc: "Daniel M. Weeks" <dan@danweeks.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
During testing initrd (>2G) support, find decompress/lz4 does not work
with initrd at all.
decompress_* should support:
1. inbuf[]/outbuf[] for kernel preboot.
2. inbuf[]/flush() for initramfs
3. fill()/flush() for initrd.
in the unlz4 does not handle case 3, as input len is passed as 0, and it
failed in first try.
Fix that add one extra if (fill) checking, and get out if EOF from the
fill().
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Kyungsik Lee <kyungsik.lee@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Increase the maximum BAR size from 8GB to 128GB.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
For boot efi kernel directly without bootloader.
If the kernel support XLF_CAN_BE_LOADED_ABOVE_4G, we should
not limit initrd under hdr->initrd_add_max.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
On system with 2TiB ram, current x86_64 have 128M as section size, and
one memory_block only include one section. So will have 16400 entries
under /sys/devices/system/memory/.
Current code try to use block id to find block pointer in /sys for any
section, and reuse that block pointer. that finding will take some time
even after commit 7c243c7168 ("mm: speedup in __early_pfn_to_nid")
that will skip the search in that case during booting up.
So solution could be increase block size just like SGI UV system did.
(harded code to 2g).
This patch is trying to probe the block size to make it match mmio remap
size. for example, Intel Nehalem later system will have memory range [0,
TOML), [4g, TOMH]. If the memory hole is 2g and total is 128g, TOM will
be 2g, and TOM2 will be 130g.
We could use 2g as block size instead of default 128M. That will reduce
number of entries in /sys/devices/system/memory/
On system 6TiB system will reduce boot time by 35 seconds.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
check_irq_vectors_for_cpu_disable() can overestimate the number of
available interrupt vectors, so the check for cpu down succeeds, but
the actual cpu removal fails.
It iterates from FIRST_EXTERNAL_VECTOR to NR_VECTORS, which is wrong
because the systems vectors are not taken into account.
Limit the search to first_system_vector instead of NR_VECTORS.
The second indicator for vector availability the used_vectors bitmap
is not taken into account at all. So system vectors,
e.g. IA32_SYSCALL_VECTOR (0x80) and IRQ_MOVE_CLEANUP_VECTOR (0x20),
are accounted as available.
Add a check for the used_vectors bitmap and do not account vectors
which are marked there.
[ tglx: Simplified code. Rewrote changelog and code comments. ]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Acked-by: Prarit Bhargava <prarit@redhat.com>
Cc: Seiji Aguchi <seiji.aguchi@hds.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: K. Y. Srinivasan <kys@microsoft.com>
Cc: Steven Rostedt (Red Hat) <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Elliott, Robert (Server Storage)" <Elliott@hp.com>
Cc: x86@kernel.org
Link: http://lkml.kernel.org/r/1400160305-17774-2-git-send-email-prarit@redhat.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Assign PCI resources before pci_bus_add_device(). The resources must be
assigned before a driver can claim the device.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
This patch changes the way we handle 64-bit prefetchable bridge windows to
make it more likely that we can assign space to all devices.
Previously we put all prefetchable resources in the prefetchable bridge
window. If any of those resources was 32-bit only, we restricted the
window to be below 4GB.
After this patch, we only put 64-bit prefetchable resources in a 64-bit
prefetchable window. We put all 32-bit prefetchable resources in the
non-prefetchable window, even if there are no 64-bit prefetchable
resources.
With the previous approach, if there was a 32-bit prefetchable resource
behind a bridge, we forced the bridge's prefetchable window below 4GB,
which meant that even if there was plenty of space above 4GB available, we
couldn't use it, and assignment of large 64-bit resources could fail, as
in the bugzilla below.
The new strategy is:
1) If the prefetchable window is 64 bits wide, we put only 64-bit
prefetchable resources in it. Any 32-bit prefetchable resources go in
the non-prefetchable window.
2) If the prefetchable window is 32 bits wide, we put both 32- and 64-bit
prefetchable resources in it.
3) If there is no prefetchable window, all MMIO resources go in the
non-prefetchable window.
This reduces performance for 32-bit prefetchable resources below a bridge
with a 64-bit prefetchable window. We previously assigned prefetchable
space, but now we'll assign non-prefetchable space. This is the case even
if there are no 64-bit prefetchable resources, or if they would all fit
below 4GB. In those cases, the old strategy would work and would have
better performance.
[bhelgaas: write changelog, add bugzilla link, fold in mem64_mask removal]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=74151
Tested-by: Guo Chao <yan@linux.vnet.ibm.com>
Tested-by: Wei Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
In slit init code, too many pxm_to_node() function calls are done.
We can store from_node/to_node instead of keep calling
pxm_to_node().
- Before this patch: pxm_to_node() is called n*(1+n*3) times.
- After this patch: pxm_to_node() is called n*(1+n) times.
for 8 sockets, it will be 72 instead of 200.
for 32 sockets, it will be 1056 instead of 3104.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/1390770102-4007-1-git-send-email-yinghai@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Now we have memblock_virt_alloc_low to replace original bootmem api in
swiotlb.
But we should not use BOOTMEM_LOW_LIMIT for arch that does not support
CONFIG_NOBOOTMEM, as old api take 0.
| #define alloc_bootmem_low(x) \
| __alloc_bootmem_low(x, SMP_CACHE_BYTES, 0)
|#define alloc_bootmem_low_pages_nopanic(x) \
| __alloc_bootmem_low_nopanic(x, PAGE_SIZE, 0)
and we have
#define BOOTMEM_LOW_LIMIT __pa(MAX_DMA_ADDRESS)
for CONFIG_NOBOOTMEM.
Restore goal to 0 to fix ia64 crash, that Tony found.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Reported-by: Tony Luck <tony.luck@gmail.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Dave reported big numa system booting is broken.
It turns out that commit 5b6e529521 ("x86: memblock: set current limit
to max low memory address") sets the limit to low wrongly.
max_low_pfn_mapped is different from max_pfn_mapped.
max_low_pfn_mapped is always under 4G.
That will memblock_alloc_nid all go under 4G.
Revert 5b6e529521 to fix a no-boot regression which was triggered by
457ff1de2d ("lib/swiotlb.c: use memblock apis for early memory
allocations").
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Reported-by: Dave Hansen <dave.hansen@intel.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When allocating space for 32-bit BARs, we previously limited RESOURCE
addresses so they would fit in 32 bits. However, the BUS address need not
be the same as the resource address, and it's the bus address that must fit
in the 32-bit BAR.
This patch adds:
- pci_clip_resource_to_region(), which clips a resource so it contains
only the range that maps to the specified bus address region, e.g., to
clip a resource to 32-bit bus addresses, and
- pci_bus_alloc_from_region(), which allocates space for a resource from
the specified bus address region,
and changes pci_bus_alloc_resource() to allocate space for 64-bit BARs from
the entire bus address region, and space for 32-bit BARs from only the bus
address region below 4GB.
If we had this window:
pci_root HWP0002:0a: host bridge window [mem 0xf0180000000-0xf01fedfffff] (bus address [0x80000000-0xfedfffff])
we previously could not put a 32-bit BAR there, because the CPU addresses
don't fit in 32 bits. This patch fixes this, so we can use this space for
32-bit BARs.
It's also possible (though unlikely) to have resources with 32-bit CPU
addresses but bus addresses above 4GB. In this case the previous code
would allocate space that a 32-bit BAR could not map.
Remove PCIBIOS_MAX_MEM_32, which is no longer used.
[bhelgaas: reworked starting from http://lkml.kernel.org/r/1386658484-15774-3-git-send-email-yinghai@kernel.org]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Per the Intel 915G/915GV/... Chipset spec (document number 301467-005),
GMADR is a standard PCI BAR.
The PCI core reads GMADR at enumeration-time. Use pci_bus_address()
instead of reading it again in the driver. This works correctly for both
32-bit and 64-bit BARs. The spec above only mentions 32-bit GMADR, but
Yinghai's patch (link below) indicates some devices have a 64-bit GMADR.
[bhelgaas: reworked starting from http://lkml.kernel.org/r/1385851238-21085-13-git-send-email-yinghai@kernel.org]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Convert pci/ioapic.c to be builtin only, with no module option, so we can
support IO-APIC hotplug. Also make it depend on X86_IO_APIC.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
These interfaces:
pcibios_resource_to_bus(struct pci_dev *dev, *bus_region, *resource)
pcibios_bus_to_resource(struct pci_dev *dev, *resource, *bus_region)
took a pci_dev, but they really depend only on the pci_bus. And we want to
use them in resource allocation paths where we have the bus but not a
device, so this patch converts them to take the pci_bus instead of the
pci_dev:
pcibios_resource_to_bus(struct pci_bus *bus, *bus_region, *resource)
pcibios_bus_to_resource(struct pci_bus *bus, *resource, *bus_region)
In fact, with standard PCI-PCI bridges, they only depend on the host
bridge, because that's the only place address translation occurs, but
we aren't going that far yet.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Previously we removed the pci_dev from the bus_list and released its
resources in pci_destroy_dev(). But that's too early: it's possible to
call pci_destroy_dev() twice for the same device (e.g., via sysfs), and
that will cause an oops when we try to remove it from bus_list the second
time.
We should remove it from the bus_list only when the last reference to the
pci_dev has been released, i.e., in pci_release_dev().
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
4f535093cf ("PCI: Put pci_dev in device tree as early as possible")
moved pci_proc_attach_device() from pci_bus_add_device() to
pci_device_add().
This moves it back to pci_bus_add_device(), essentially reverting that
part of 4f535093cf. This makes it symmetric with pci_stop_dev(),
where we call pci_proc_detach_device() and pci_remove_sysfs_dev_files()
and set dev->is_added = 0.
[bhelgaas: changelog, create sysfs then attach proc for symmetry]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
To be consistent with 4bff674990 ("PCI: Move device_del() from
pci_stop_dev() to pci_destroy_dev()", this changes pci_stop_root_bus()
to use device_release_driver() instead of device_del().
This also changes pci_remove_root_bus() to use device_unregister()
instead of put_device() so it corresponds with the device_register()
call in pci_create_root_bus().
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The pcie_portdrv .probe() method calls pci_enable_device() once, in
pcie_port_device_register(), but the .remove() method calls
pci_disable_device() twice, in pcie_port_device_remove() and in
pcie_portdrv_remove().
That causes a "disabling already-disabled device" warning when removing a
PCIe port device. This happens all the time when removing Thunderbolt
devices, but is also easy to reproduce with, e.g.,
"echo 0000:00:1c.3 > /sys/bus/pci/drivers/pcieport/unbind"
This patch removes the disable from pcie_portdrv_remove().
[bhelgaas: changelog, tag for stable]
Reported-by: David Bulkow <David.Bulkow@stratus.com>
Reported-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v2.6.32+
Ben Herrenschmidt found that commit 928bea9648 ("PCI: Delay enabling
bridges until they're needed") breaks PCI in some powerpc environments.
The reason is that the PCIe port driver will call pci_enable_device() on
the bridge, so the device is enabled, but skips pci_set_master because
pcie_port_auto and no acpi on powerpc.
Because of that, pci_enable_bridge() later on (called as a result of the
child device driver doing pci_enable_device) will see the bridge as
already enabled and will not call pci_set_master() on it.
Fixed by add checking in pci_enable_bridge, and call pci_set_master
if driver skip that.
That will make the code more robot and wade off problem for missing
pci_set_master in drivers.
Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ben Herrenschmidt found that commit 928bea9648 ("PCI: Delay enabling
bridges until they're needed") breaks PCI in some powerpc environments.
The reason is that the PCIe port driver will call pci_enable_device() on
the bridge, so the device is enabled, but skips pci_set_master because
pcie_port_auto and no acpi on powerpc.
Because of that, pci_enable_bridge() later on (called as a result of the
child device driver doing pci_enable_device) will see the bridge as
already enabled and will not call pci_set_master() on it.
Fixed by add checking in pci_enable_bridge, and call pci_set_master
if driver skip that.
That will make the code more robot and wade off problem for missing
pci_set_master in drivers.
Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Current ACPI tables in initrd is limited to 10, that is too small.
64 should be good enough as we have 35 sigs and could have several
SSDT.
Two problems in current code prevent us from increasing limit:
1. The cpio file info array is put in stack, as every element is 32
bytes, could run out of stack if we have that array size to 64.
We can move it out from stack, make it global and put it into the
__initdata section.
2. early_ioremap() only can remap 256k one time. Current code maps
10 tables at a time. If we increased that limit, the whole size
could be more than 256k, so early_ioremap() would fail with that.
We can map chunks one by one during copying, instead of mapping
all of them together.
Signed-off-by: Yinghai <yinghai@kernel.org>
Acked-by: Tejun Heo <tj@kernel.org>
Tested-by: Thomas Renninger <trenn@suse.de>
Reviewed-by: Tang Chen <tangchen@cn.fujitsu.com>
Tested-by: Tang Chen <tangchen@cn.fujitsu.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If the hw supports intel_pstate and acpi_cpufreq, intel_pstate will
get loaded first.
acpi_cpufreq_init() will call acpi_cpufreq_early_init()
and that will allocate perf data and init those perf data in ACPI core,
(that will cover all CPUs). But later it will free them as
cpufreq_register_driver(acpi_cpufreq) will fail as intel_pstate is
already registered
Use cpufreq_get_current_driver() to check if we can skip the
acpi_cpufreq loading.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
On systems that support intel_pstate, acpi_cpufreq fails to load, and
udev keeps trying until trace gets filled up and kernel crashes.
The root cause is driver return ret from cpufreq_register_driver(),
because when some other driver takes over before, it will return
EBUSY and then udev will keep trying ...
cpufreq_register_driver() should return EEXIST instead so that the
system can boot without appending intel_pstate=disable and still use
intel_pstate.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Current early_pfn_to_nid() on arch that support memblock go over
memblock.memory one by one, so will take too many try near the end.
We can use existing memblock_search to find the node id for given pfn,
that could save some time on bigger system that have many entries
memblock.memory array.
Here are the timing differences for several machines. In each case with
the patch less time was spent in __early_pfn_to_nid().
3.11-rc5 with patch difference (%)
-------- ---------- --------------
UV1: 256 nodes 9TB: 411.66 402.47 -9.19 (2.23%)
UV2: 255 nodes 16TB: 1141.02 1138.12 -2.90 (0.25%)
UV2: 64 nodes 2TB: 128.15 126.53 -1.62 (1.26%)
UV2: 32 nodes 2TB: 121.87 121.07 -0.80 (0.66%)
Time in seconds.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Acked-by: Russ Anderson <rja@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We should not check loop+1 with loop end in loop body. Just duplicate two
lines code to avoid it.
That will help a bit when we have huge amount of pages on system with
16TiB memory.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Dave Hansen reported that systems between 500G and 600G RAM
crash early if DEBUG_PAGEALLOC is selected.
> [ 0.000000] init_memory_mapping: [mem 0x00000000-0x000fffff]
> [ 0.000000] [mem 0x00000000-0x000fffff] page 4k
> [ 0.000000] BRK [0x02086000, 0x02086fff] PGTABLE
> [ 0.000000] BRK [0x02087000, 0x02087fff] PGTABLE
> [ 0.000000] BRK [0x02088000, 0x02088fff] PGTABLE
> [ 0.000000] init_memory_mapping: [mem 0xe80ee00000-0xe80effffff]
> [ 0.000000] [mem 0xe80ee00000-0xe80effffff] page 4k
> [ 0.000000] BRK [0x02089000, 0x02089fff] PGTABLE
> [ 0.000000] BRK [0x0208a000, 0x0208afff] PGTABLE
> [ 0.000000] Kernel panic - not syncing: alloc_low_page: ran out of memory
It turns out that we missed increasing needed pages in BRK to
mapping initial 2M and [0,1M) when we switched to use the #PF
handler to set memory mappings:
> commit 8170e6bed4
> Author: H. Peter Anvin <hpa@zytor.com>
> Date: Thu Jan 24 12:19:52 2013 -0800
>
> x86, 64bit: Use a #PF handler to materialize early mappings on demand
Before that, we had the maping from [0,512M) in head_64.S, and we
can spare two pages [0-1M). After that change, we can not reuse
pages anymore.
When we have more than 512M ram, we need an extra page for pgd page
with [512G, 1024g).
Increase pages in BRK for page table to solve the boot crash.
Reported-by: Dave Hansen <dave.hansen@intel.com>
Bisected-by: Dave Hansen <dave.hansen@intel.com>
Tested-by: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: <stable@vger.kernel.org> # v3.9 and later
Link: http://lkml.kernel.org/r/1376351004-4015-1-git-send-email-yinghai@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Ben Herrenschmidt reported the following problem:
- The bus has space for all desired MMIO resources, including optional
space for SR-IOV devices
- We attempt to allocate I/O port space, but it fails because the bus
has no I/O space
- Because of the I/O allocation failure, we retry MMIO allocation,
requesting only the required space, without the optional SR-IOV space
This means we don't allocate the optional SR-IOV space, even though we
could.
This is related to 0c5be0cb0e ("PCI: Retry on IORESOURCE_IO type
allocations").
This patch changes how we handle allocation failures. We will now retry
allocation of only the resource type that failed. If MMIO allocation
fails, we'll retry only MMIO allocation. If I/O port allocation fails,
we'll retry only I/O port allocation.
[bhelgaas: changelog]
Reference: https://lkml.kernel.org/r/1367712653.11982.19.camel@pasglop
Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Tested-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v3.10+
Hot-removing a device with SR-IOV enabled causes a null pointer dereference
in v3.9 and v3.10.
This is a regression caused by ba518e3c17 ("PCI: pciehp: Iterate over all
devices in slot, not functions 0-7"). When we iterate over the
bus->devices list, we first remove the PF, which also removes all the VFs
from the list. Then the list iterator blows up because more than just the
current entry was removed from the list.
ac205b7bb7 ("PCI: make sriov work with hotplug remove") works around a
similar problem in pci_stop_bus_devices() by iterating over the list in
reverse, so the VFs are stopped and removed from the list first, before the
PF.
This patch changes pciehp_unconfigure_device() to iterate over the list in
reverse, too.
[bhelgaas: bugzilla, changelog]
Reference: https://bugzilla.kernel.org/show_bug.cgi?id=60604
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Yijing Wang <wangyijing@huawei.com>
CC: stable@vger.kernel.org # v3.9+
When hot-adding an ACPI host bridge, use
pci_assign_unassigned_root_bus_resources() instead of
pci_assign_unassigned_bus_resources().
The former is more aggressive and will release and reassign existing
resources if necessary. This is safe at hot-add time because no drivers
are bound to devices below the new host bridge yet.
[bhelgaas: changelog, split __init changes out for reviewability]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Resource reallocation is currently done only at boot-time, but will
soon be done when host bridge is hot-added. This patch removes the
__init annotations so the code will still be present after boot.
[bhelgaas: split __init changes out]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
We currently enable PCI bridges after scanning a bus and assigning
resources. This is often done in arch code.
This patch changes this so we don't enable a bridge until necessary, i.e.,
until we enable a PCI device behind the bridge. We do this in the generic
pci_enable_device() path, so this also removes the arch-specific code to
enable bridges.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Previously, we did resource assignment globally. This patch splits up
pci_assign_unassigned_resources() so assignment is done for each root bus
in turn. We check each root bus individually to see whether it needs any
reassignment, and if it does, we assign resources for just that bus.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
pci_realloc_detect() turns on automatic resource allocation when it finds
unassigned SR-IOV resources. Previously it did this on a global basis, so
we enabled reallocation if any PCI device anywhere had an unassigned SR-IOV
resource.
This patch changes pci_realloc_detect() so it looks at a single bus, so we
can do this when a host bridge is hot-added.
[bhelgaas: changelog]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Previously we did not turn on automatic PCI resource reallocation for
unassigned IOV resources behind a host bridge with address offset. This
patch fixes that bug.
The intent was that "!r->start" would check for a BAR containing zero. But
that check is incorrect for host bridges that apply an offset, because in
that case the resource address is not the same as the bus address.
This patch fixes that by converting the resource address back to a bus
address before checking for zero.
[bhelgaas: changelog]
Suggested-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
When CONFIG_PCI_REALLOC_ENABLE_AUTO=y, pci_realloc_detect() looks at PCI
devices to see if any have SR-IOV resources that need to be assigned. If
it finds any, it turns on automatic resource reallocation.
This patch changes pci_realloc_detect() so it uses pci_walk_bus() on
each root bus instead of using for_each_pci_dev(). This is a step
toward doing reallocation on a per-bus basis, so we can do it for
a hot-added host bridge.
[bhelgaas: changelog, rename callback to iov_resources_unassigned(), use
boolean for "unassigned"]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Drop the "bus" temporary variable. No functional change, but simplifies
later patch slightly.
[bhelgaas: changelog, make same change in
pci_assign_unassigned_bridge_resources() to keep it parallel with
pci_assign_unassigned_resources()]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
On one sytem that mtrr range is more then 44bits, in dmesg we have
[ 0.000000] MTRR default type: write-back
[ 0.000000] MTRR fixed ranges enabled:
[ 0.000000] 00000-9FFFF write-back
[ 0.000000] A0000-BFFFF uncachable
[ 0.000000] C0000-DFFFF write-through
[ 0.000000] E0000-FFFFF write-protect
[ 0.000000] MTRR variable ranges enabled:
[ 0.000000] 0 [000080000000-0000FFFFFFFF] mask 3FFF80000000 uncachable
[ 0.000000] 1 [380000000000-38FFFFFFFFFF] mask 3F0000000000 uncachable
[ 0.000000] 2 [000099000000-000099FFFFFF] mask 3FFFFF000000 write-through
[ 0.000000] 3 [00009A000000-00009AFFFFFF] mask 3FFFFF000000 write-through
[ 0.000000] 4 [381FFA000000-381FFBFFFFFF] mask 3FFFFE000000 write-through
[ 0.000000] 5 [381FFC000000-381FFC0FFFFF] mask 3FFFFFF00000 write-through
[ 0.000000] 6 [0000AD000000-0000ADFFFFFF] mask 3FFFFF000000 write-through
[ 0.000000] 7 [0000BD000000-0000BDFFFFFF] mask 3FFFFF000000 write-through
[ 0.000000] 8 disabled
[ 0.000000] 9 disabled
but /proc/mtrr report wrong:
reg00: base=0x080000000 ( 2048MB), size= 2048MB, count=1: uncachable
reg01: base=0x80000000000 (8388608MB), size=1048576MB, count=1: uncachable
reg02: base=0x099000000 ( 2448MB), size= 16MB, count=1: write-through
reg03: base=0x09a000000 ( 2464MB), size= 16MB, count=1: write-through
reg04: base=0x81ffa000000 (8519584MB), size= 32MB, count=1: write-through
reg05: base=0x81ffc000000 (8519616MB), size= 1MB, count=1: write-through
reg06: base=0x0ad000000 ( 2768MB), size= 16MB, count=1: write-through
reg07: base=0x0bd000000 ( 3024MB), size= 16MB, count=1: write-through
reg08: base=0x09b000000 ( 2480MB), size= 16MB, count=1: write-combining
so bit 44 and bit 45 get cut off.
We have problems in arch/x86/kernel/cpu/mtrr/generic.c::generic_get_mtrr().
1. for base, we miss cast base_lo to 64bit before shifting.
Fix that by adding u64 casting.
2. for size, it only can handle 44 bits aka 32bits + page_shift
Fix that with 64bit mask instead of 32bit mask_lo, then range could be
more than 44bits.
At the same time, we need to update size_or_mask for old cpus that does
support cpuid 0x80000008 to get phys_addr. Need to set high 32bits
to all 1s, otherwise will not get correct size for them.
Also fix mtrr_add_page: it should check base and (base + size - 1)
instead of base and size, as base and size could be small but
base + size could bigger enough to be out of boundary. We can
use boot_cpu_data.x86_phys_bits directly to avoid size_or_mask.
So When are we going to have size more than 44bits? that is 16TiB.
after patch we have right ouput:
reg00: base=0x080000000 ( 2048MB), size= 2048MB, count=1: uncachable
reg01: base=0x380000000000 (58720256MB), size=1048576MB, count=1: uncachable
reg02: base=0x099000000 ( 2448MB), size= 16MB, count=1: write-through
reg03: base=0x09a000000 ( 2464MB), size= 16MB, count=1: write-through
reg04: base=0x381ffa000000 (58851232MB), size= 32MB, count=1: write-through
reg05: base=0x381ffc000000 (58851264MB), size= 1MB, count=1: write-through
reg06: base=0x0ad000000 ( 2768MB), size= 16MB, count=1: write-through
reg07: base=0x0bd000000 ( 3024MB), size= 16MB, count=1: write-through
reg08: base=0x09b000000 ( 2480MB), size= 16MB, count=1: write-combining
-v2: simply checking in mtrr_add_page according to hpa.
[ hpa: This probably wants to go into -stable only after having sat in
mainline for a bit. It is not a regression. ]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1371162815-29931-1-git-send-email-yinghai@kernel.org
Cc: <stable@vger.kernel.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Joshua reported: Commit cd7b304dfaf1 (x86, range: fix missing merge
during add range) broke mtrr cleanup on his setup in 3.9.5.
corresponding commit in upstream is fbe06b7bae.
The reason is add_range_with_merge could generate blank spot.
We could avoid that by searching new expanded start/end, that
new range should include all connected ranges in range array.
At last add the new expanded start/end to the range array.
Also move up left array so do not add new blank slot in the
range array.
-v2: move left array to avoid enhance add_range()
-v3: include fix from Joshua about memmove declaring when
DYN_DEBUG is used.
Reported-by: Joshua Covington <joshuacov@googlemail.com>
Tested-by: Joshua Covington <joshuacov@googlemail.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1371154622-8929-3-git-send-email-yinghai@kernel.org
Cc: <stable@vger.kernel.org> v3.9
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Joshua reported: Commit cd7b304dfaf1 (x86, range: fix missing merge
during add range) broke mtrr cleanup on his setup in 3.9.5.
corresponding commit in upstream is fbe06b7bae.
*BAD*gran_size: 64K chunk_size: 16M num_reg: 6 lose cover RAM: -0G
https://bugzilla.kernel.org/show_bug.cgi?id=59491
So it rejects new var mtrr layout.
It turns out we have some problem with initial mtrr range retrieval.
The current sequence is:
x86_get_mtrr_mem_range
==> bunchs of add_range_with_merge
==> bunchs of subract_range
==> clean_sort_range
add_range_with_merge for [0,1M)
sort_range()
add_range_with_merge could have blank slots, so we can not just
sort only, that will have final result have extra blank slot in head.
So move that calling add_range_with_merge for [0,1M), with that we
could avoid extra clean_sort_range calling.
Reported-by: Joshua Covington <joshuacov@googlemail.com>
Tested-by: Joshua Covington <joshuacov@googlemail.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1371154622-8929-2-git-send-email-yinghai@kernel.org
Cc: <stable@vger.kernel.org> v3.9
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Commit
8d57470d x86, mm: setup page table in top-down
causes a kernel panic while setting mem=2G.
[mem 0x00000000-0x000fffff] page 4k
[mem 0x7fe00000-0x7fffffff] page 1G
[mem 0x7c000000-0x7fdfffff] page 1G
[mem 0x00100000-0x001fffff] page 4k
[mem 0x00200000-0x7bffffff] page 2M
for last entry is not what we want, we should have
[mem 0x00200000-0x3fffffff] page 2M
[mem 0x40000000-0x7bffffff] page 1G
Actually we merge the continuous ranges with same page size too early.
in this case, before merging we have
[mem 0x00200000-0x3fffffff] page 2M
[mem 0x40000000-0x7bffffff] page 2M
after merging them, will get
[mem 0x00200000-0x7bffffff] page 2M
even we can use 1G page to map
[mem 0x40000000-0x7bffffff]
that will cause problem, because we already map
[mem 0x7fe00000-0x7fffffff] page 1G
[mem 0x7c000000-0x7fdfffff] page 1G
with 1G page, aka [0x40000000-0x7fffffff] is mapped with 1G page already.
During phys_pud_init() for [0x40000000-0x7bffffff], it will not
reuse existing that pud page, and allocate new one then try to use
2M page to map it instead, as page_size_mask does not include
PG_LEVEL_1G. At end will have [7c000000-0x7fffffff] not mapped, loop
in phys_pmd_init stop mapping at 0x7bffffff.
That is right behavoir, it maps exact range with exact page size that
we ask, and we should explicitly call it to map [7c000000-0x7fffffff]
before or after mapping 0x40000000-0x7bffffff.
Anyway we need to make sure ranges' page_size_mask correct and consistent
after split_mem_range for each range.
Fix that by calling adjust_range_size_mask before merging range
with same page size.
-v2: update change log.
-v3: add more explanation why [7c000000-0x7fffffff] is not mapped, and
it causes panic.
Bisected-by: "Xie, ChanglongX" <changlongx.xie@intel.com>
Bisected-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reported-and-tested-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1370015587-20835-1-git-send-email-yinghai@kernel.org
Cc: <stable@vger.kernel.org> v3.9
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
When a PCI host bridge device receives a Bus Check notification, we
must re-enumerate starting with the bridge to discover changes (devices
that have been added or removed).
Prior to 668192b678 ("PCI: acpiphp: Move host bridge hotplug to
pci_root.c"), this happened in _handle_hotplug_event_bridge(). After that
commit, _handle_hotplug_event_bridge() is not installed for host bridges,
and the host bridge notify handler, _handle_hotplug_event_root() did not
re-enumerate.
This patch adds re-enumeration to _handle_hotplug_event_root().
This fixes cases where we don't notice the addition or removal of
PCI devices, e.g., the PCI-to-USB ExpressCard in the bugzilla below.
[bhelgaas: changelog, references]
Reference: https://lkml.kernel.org/r/CAAh6nkmbKR3HTqm5ommevsBwhL_u0N8Rk7Wsms_LfP=nBgKNew@mail.gmail.com
Reference: https://bugzilla.kernel.org/show_bug.cgi?id=57961
Reported-by: Gavin Guo <tuffkidtt@gmail.com>
Tested-by: Gavin Guo <tuffkidtt@gmail.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Jiang Liu <jiang.liu@huawei.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
CC: stable@vger.kernel.org # v3.9+
Christian found v3.9 does not work with E350 with EFI is enabled.
[ 1.658832] Trying to unpack rootfs image as initramfs...
[ 1.679935] BUG: unable to handle kernel paging request at ffff88006e3fd000
[ 1.686940] IP: [<ffffffff813661df>] memset+0x1f/0xb0
[ 1.692010] PGD 1f77067 PUD 1f7a067 PMD 61420067 PTE 0
but early memtest report all memory could be accessed without problem.
early page table is set in following sequence:
[ 0.000000] init_memory_mapping: [mem 0x00000000-0x000fffff]
[ 0.000000] init_memory_mapping: [mem 0x6e600000-0x6e7fffff]
[ 0.000000] init_memory_mapping: [mem 0x6c000000-0x6e5fffff]
[ 0.000000] init_memory_mapping: [mem 0x00100000-0x6bffffff]
[ 0.000000] init_memory_mapping: [mem 0x6e800000-0x6ea07fff]
but later efi_enter_virtual_mode try set mapping again wrongly.
[ 0.010644] pid_max: default: 32768 minimum: 301
[ 0.015302] init_memory_mapping: [mem 0x640c5000-0x6e3fcfff]
that means it fails with pfn_range_is_mapped.
It turns out that we have a bug in add_range_with_merge and it does not
merge range properly when new add one fill the hole between two exsiting
ranges. In the case when [mem 0x00100000-0x6bffffff] is the hole between
[mem 0x00000000-0x000fffff] and [mem 0x6c000000-0x6e7fffff].
Fix the add_range_with_merge by calling itself recursively.
Reported-by: "Christian König" <christian.koenig@amd.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/CAE9FiQVofGoSk7q5-0irjkBxemqK729cND4hov-1QCBJDhxpgQ@mail.gmail.com
Cc: <stable@vger.kernel.org> v3.9
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Commit 4f535093cf "PCI: Put pci_dev in device tree as early as possible"
moved final fixups from pci_bus_add_device() to pci_device_add(). But
pci_device_add() happens before resource assignment, so BARs may not be
valid yet.
Typical flow for hot-add:
pciehp_configure_device
pci_scan_slot
pci_scan_single_device
pci_device_add
pci_fixup_device(pci_fixup_final, dev) # previous location
# resource assignment happens here
pci_bus_add_devices
pci_bus_add_device
pci_fixup_device(pci_fixup_final, dev) # new location
[bhelgaas: changelog, move fixups to pci_bus_add_device()]
Reference: https://lkml.kernel.org/r/20130415182614.GB9224@xanatos
Reported-by: David Bulkow <David.Bulkow@stratus.com>
Tested-by: David Bulkow <David.Bulkow@stratus.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v3.9+
Per hpa, use crashkernel=X,high crashkernel=Y,low instead of
crashkernel_hign=X crashkernel_low=Y. As that could be extensible.
-v2: according to Vivek, change delimiter to ;
-v3: let hign and low only handle simple form and it conforms to
description in kernel-parameters.txt
still keep crashkernel=X override any crashkernel=X,high
crashkernel=Y,low
-v4: update get_last_crashkernel returning and add more strict
checking in parse_crashkernel_simple() found by HATAYAMA.
-v5: Change delimiter back to , according to HPA.
also separate parse_suffix from parse_simper according to vivek.
so we can avoid @pos in that path.
-v6: Tight the checking about crashkernel=X,highblahblah,high
found by HTYAYAMA.
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1366089828-19692-5-git-send-email-yinghai@kernel.org
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Vivek found old kexec-tools does not work new kernel anymore.
So change back crashkernel= back to old behavoir, and add crashkernel_high=
to let user decide if buffer could be above 4G, and also new kexec-tools will
be needed.
-v2: let crashkernel=X override crashkernel_high=
update description about _high will be ignored by crashkernel=X
-v3: update description about kernel-parameters.txt according to Vivek.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1366089828-19692-4-git-send-email-yinghai@kernel.org
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Chao said that kdump does does work well on his system on 3.8
without extra parameter, even iommu does not work with kdump.
And now have to append crashkernel_low=Y in first kernel to make
kdump work.
We have now modified crashkernel=X to allocate memory beyong 4G (if
available) and do not allocate low range for crashkernel if the user
does not specify that with crashkernel_low=Y. This causes regression
if iommu is not enabled. Without iommu, swiotlb needs to be setup in
first 4G and there is no low memory available to second kernel.
Set crashkernel_low automatically if the user does not specify that.
For system that does support IOMMU with kdump properly, user could
specify crashkernel_low=0 to save that 72M low ram.
-v3: add swiotlb_size() according to Konrad.
-v4: add comments what 8M is for according to hpa.
also update more crashkernel_low= in kernel-parameters.txt
-v5: update changelog according to Vivek.
-v6: Change description about swiotlb referring according to HATAYAMA.
Reported-by: WANG Chao <chaowang@redhat.com>
Tested-by: WANG Chao <chaowang@redhat.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1366089828-19692-2-git-send-email-yinghai@kernel.org
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Yinghai Lu