Denys Fedoryshchenko reported a bootup crash when he upgraded
his system from 3GB to 4GB RAM:
http://lkml.org/lkml/2008/1/7/9
the bug is due to HIGHMEM4G && SPARSEMEM kernels making pfn_to_page()
to return an invalid pointer when the pfn is in a memory hole. The
256 MB PCI aperture at the end of RAM was not mapped by sparsemem,
and hence the pfn was not valid. But set_highmem_pages_init() iterated
this range without checking the pfn's validity first.
this bug was probably present in the sparsemem code ever since sparsemem
has been introduced in v2.6.13. It was masked due to HIGHMEM64G using
larger memory regions in sparsemem_32.h:
#ifdef CONFIG_X86_PAE
#define SECTION_SIZE_BITS 30
#define MAX_PHYSADDR_BITS 36
#define MAX_PHYSMEM_BITS 36
#else
#define SECTION_SIZE_BITS 26
#define MAX_PHYSADDR_BITS 32
#define MAX_PHYSMEM_BITS 32
#endif
which creates 1GB sparsemem regions instead of 64MB sparsemem regions.
So in practice we only ever created true sparsemem holes on x86 with
HIGHMEM4G - but that was rarely used by distros.
( btw., we could probably save 2MB of mem_map[]s on X86_PAE if we reduced
the sparsemem region size to 256 MB. )
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
if CONFIG_PAGEALLOC is enabled then X86_FEATURE_PSE is disabled and all
the kernel physical RAM pagetables are set up as 4K pages. This is
needed so that CONFIG_PAGEALLOC can do finegrained mapping and unmapping
of pages.
as a side-effect though, the total size of memory allocated as kernel
pagetables increases significantly. All these pagetables are allocated
via alloc_bootmem_low_pages(), straight out of the lowmem DMA pool. If
the system has enough RAM and a large kernel image then almost all of
the 16 MB lowmem DMA pool is allocated to the image and to pagetables -
leaving no space for __GFP_DMA allocations.
this results in drivers failing and the bootup hanging:
swapper invoked oom-killer: gfp_mask=0x80d1, order=0, oomkilladj=0
[<4015059f>] out_of_memory+0x17f/0x1c0
[<40151f3c>] __alloc_pages+0x37c/0x3a0
[<40168cd7>] slob_new_page+0x37/0x50
[<40168dff>] slob_alloc+0x10f/0x190
[<40169010>] __kmalloc_node+0x80/0x90
[<405a17e3>] scsi_host_alloc+0x33/0x2c0
[<405a1a82>] scsi_register+0x12/0x60
[<40d5889e>] aha1542_detect+0x9e/0x940
[<405c5ba5>] ultrastor_detect+0x265/0x5f0
[<401352f5>] getnstimeofday+0x35/0xf0
[<40d58751>] init_this_scsi_driver+0x41/0xf0
[<40d0b856>] kernel_init+0x136/0x310
[<40d58710>] init_this_scsi_driver+0x0/0xf0
[<40d0b720>] kernel_init+0x0/0x310
[<40105547>] kernel_thread_helper+0x7/0x10
=======================
the fix is to first allocate from above the DMA pool, and if that fails
(for example due to it being a machine with less than 16 MB of RAM),
allocate from the DMA pool as a fallback.
With this fix applied i was able to boot a PAGEALLOC=y kernel that would
hang before.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
* 'xen-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/jeremy/xen:
xfs: eagerly remove vmap mappings to avoid upsetting Xen
xen: add some debug output for failed multicalls
xen: fix incorrect vcpu_register_vcpu_info hypercall argument
xen: ask the hypervisor how much space it needs reserved
xen: lock pte pages while pinning/unpinning
xen: deal with stale cr3 values when unpinning pagetables
xen: add batch completion callbacks
xen: yield to IPI target if necessary
Clean up duplicate includes in arch/i386/xen/
remove dead code in pgtable_cache_init
paravirt: clean up lazy mode handling
paravirt: refactor struct paravirt_ops into smaller pv_*_ops
The conversion from using a slab cache to quicklist left some residual
dead code.
I note that in the conversion it now always allocates a whole page for
the pgd, rather than the 32 bytes needed for a PAE pgd. Was this
intended?
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@suse.de>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Now, arch dependent code around CONFIG_MEMORY_HOTREMOVE is a mess.
This patch cleans up them. This is against 2.6.23-rc6-mm1.
- fix compile failure on ia64/ CONFIG_MEMORY_HOTPLUG && !CONFIG_MEMORY_HOTREMOVE case.
- For !CONFIG_MEMORY_HOTREMOVE, add generic no-op remove_memory(),
which returns -EINVAL.
- removed remove_pages() only used in powerpc.
- removed no-op remove_memory() in i386, sh, sparc64, x86_64.
- only powerpc returns -ENOSYS at memory hot remove(no-op). changes it
to return -EINVAL.
Note:
Currently, only ia64 supports CONFIG_MEMORY_HOTREMOVE. I welcome other
archs if there are requirements and testers.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ingo Molnar