linux/arch/ia64/mm/init.c

720 lines
19 KiB
C
Raw Normal View History

/*
* Initialize MMU support.
*
* Copyright (C) 1998-2003 Hewlett-Packard Co
* David Mosberger-Tang <davidm@hpl.hp.com>
*/
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/bootmem.h>
#include <linux/efi.h>
#include <linux/elf.h>
#include <linux/mm.h>
#include <linux/mmzone.h>
#include <linux/module.h>
#include <linux/personality.h>
#include <linux/reboot.h>
#include <linux/slab.h>
#include <linux/swap.h>
#include <linux/proc_fs.h>
#include <linux/bitops.h>
#include <linux/kexec.h>
#include <asm/dma.h>
#include <asm/io.h>
#include <asm/machvec.h>
#include <asm/numa.h>
#include <asm/patch.h>
#include <asm/pgalloc.h>
#include <asm/sal.h>
#include <asm/sections.h>
#include <asm/system.h>
#include <asm/tlb.h>
#include <asm/uaccess.h>
#include <asm/unistd.h>
#include <asm/mca.h>
#include <asm/paravirt.h>
DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
extern void ia64_tlb_init (void);
unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
#ifdef CONFIG_VIRTUAL_MEM_MAP
unsigned long VMALLOC_END = VMALLOC_END_INIT;
EXPORT_SYMBOL(VMALLOC_END);
struct page *vmem_map;
EXPORT_SYMBOL(vmem_map);
#endif
[IA64] Percpu quicklist for combined allocator for pgd/pmd/pte. This patch introduces using the quicklists for pgd, pmd, and pte levels by combining the alloc and free functions into a common set of routines. This greatly simplifies the reading of this header file. This patch is simple but necessary for large numa configurations. It simply ensures that only pages from the local node are added to a cpus quicklist. This prevents the trapping of pages on a remote nodes quicklist by starting a process, touching a large number of pages to fill pmd and pte entries, migrating to another node, and then unmapping or exiting. With those conditions, the pages get trapped and if the machine has more than 100 nodes of the same size, the calculation of the pgtable high water mark will be larger than any single node so page table cache flushing will never occur. I ran lmbench lat_proc fork and lat_proc exec on a zx1 with and without this patch and did not notice any change. On an sn2 machine, there was a slight improvement which is possibly due to pages from other nodes trapped on the test node before starting the run. I did not investigate further. This patch shrinks the quicklist based upon free memory on the node instead of the high/low water marks. I have written it to enable preemption periodically and recalculate the amount to shrink every time we have freed enough pages that the quicklist size should have grown. I rescan the nodes zones each pass because other processess may be draining node memory at the same time as we are adding. Signed-off-by: Robin Holt <holt@sgi.com> Signed-off-by: Tony Luck <tony.luck@intel.com>
2005-04-25 20:13:16 +00:00
struct page *zero_page_memmap_ptr; /* map entry for zero page */
EXPORT_SYMBOL(zero_page_memmap_ptr);
void
__ia64_sync_icache_dcache (pte_t pte)
{
unsigned long addr;
struct page *page;
page = pte_page(pte);
addr = (unsigned long) page_address(page);
if (test_bit(PG_arch_1, &page->flags))
return; /* i-cache is already coherent with d-cache */
flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page)));
set_bit(PG_arch_1, &page->flags); /* mark page as clean */
}
/*
* Since DMA is i-cache coherent, any (complete) pages that were written via
* DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
* flush them when they get mapped into an executable vm-area.
*/
void
dma_mark_clean(void *addr, size_t size)
{
unsigned long pg_addr, end;
pg_addr = PAGE_ALIGN((unsigned long) addr);
end = (unsigned long) addr + size;
while (pg_addr + PAGE_SIZE <= end) {
struct page *page = virt_to_page(pg_addr);
set_bit(PG_arch_1, &page->flags);
pg_addr += PAGE_SIZE;
}
}
inline void
ia64_set_rbs_bot (void)
{
unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
if (stack_size > MAX_USER_STACK_SIZE)
stack_size = MAX_USER_STACK_SIZE;
current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
}
/*
* This performs some platform-dependent address space initialization.
* On IA-64, we want to setup the VM area for the register backing
* store (which grows upwards) and install the gateway page which is
* used for signal trampolines, etc.
*/
void
ia64_init_addr_space (void)
{
struct vm_area_struct *vma;
ia64_set_rbs_bot();
/*
* If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
* the problem. When the process attempts to write to the register backing store
* for the first time, it will get a SEGFAULT in this case.
*/
vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
if (vma) {
mm: change anon_vma linking to fix multi-process server scalability issue The old anon_vma code can lead to scalability issues with heavily forking workloads. Specifically, each anon_vma will be shared between the parent process and all its child processes. In a workload with 1000 child processes and a VMA with 1000 anonymous pages per process that get COWed, this leads to a system with a million anonymous pages in the same anon_vma, each of which is mapped in just one of the 1000 processes. However, the current rmap code needs to walk them all, leading to O(N) scanning complexity for each page. This can result in systems where one CPU is walking the page tables of 1000 processes in page_referenced_one, while all other CPUs are stuck on the anon_vma lock. This leads to catastrophic failure for a benchmark like AIM7, where the total number of processes can reach in the tens of thousands. Real workloads are still a factor 10 less process intensive than AIM7, but they are catching up. This patch changes the way anon_vmas and VMAs are linked, which allows us to associate multiple anon_vmas with a VMA. At fork time, each child process gets its own anon_vmas, in which its COWed pages will be instantiated. The parents' anon_vma is also linked to the VMA, because non-COWed pages could be present in any of the children. This reduces rmap scanning complexity to O(1) for the pages of the 1000 child processes, with O(N) complexity for at most 1/N pages in the system. This reduces the average scanning cost in heavily forking workloads from O(N) to 2. The only real complexity in this patch stems from the fact that linking a VMA to anon_vmas now involves memory allocations. This means vma_adjust can fail, if it needs to attach a VMA to anon_vma structures. This in turn means error handling needs to be added to the calling functions. A second source of complexity is that, because there can be multiple anon_vmas, the anon_vma linking in vma_adjust can no longer be done under "the" anon_vma lock. To prevent the rmap code from walking up an incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h to make sure it is impossible to compile a kernel that needs both symbolic values for the same bitflag. Some test results: Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test box with 16GB RAM and not quite enough IO), the system ends up running >99% in system time, with every CPU on the same anon_vma lock in the pageout code. With these changes, AIM7 hits the cross-over point around 29.7k users. This happens with ~99% IO wait time, there never seems to be any spike in system time. The anon_vma lock contention appears to be resolved. [akpm@linux-foundation.org: cleanups] Signed-off-by: Rik van Riel <riel@redhat.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Larry Woodman <lwoodman@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com> Cc: Minchan Kim <minchan.kim@gmail.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 21:42:07 +00:00
INIT_LIST_HEAD(&vma->anon_vma_chain);
vma->vm_mm = current->mm;
vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
vma->vm_end = vma->vm_start + PAGE_SIZE;
vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
down_write(&current->mm->mmap_sem);
if (insert_vm_struct(current->mm, vma)) {
up_write(&current->mm->mmap_sem);
kmem_cache_free(vm_area_cachep, vma);
return;
}
up_write(&current->mm->mmap_sem);
}
/* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
if (!(current->personality & MMAP_PAGE_ZERO)) {
vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
if (vma) {
mm: change anon_vma linking to fix multi-process server scalability issue The old anon_vma code can lead to scalability issues with heavily forking workloads. Specifically, each anon_vma will be shared between the parent process and all its child processes. In a workload with 1000 child processes and a VMA with 1000 anonymous pages per process that get COWed, this leads to a system with a million anonymous pages in the same anon_vma, each of which is mapped in just one of the 1000 processes. However, the current rmap code needs to walk them all, leading to O(N) scanning complexity for each page. This can result in systems where one CPU is walking the page tables of 1000 processes in page_referenced_one, while all other CPUs are stuck on the anon_vma lock. This leads to catastrophic failure for a benchmark like AIM7, where the total number of processes can reach in the tens of thousands. Real workloads are still a factor 10 less process intensive than AIM7, but they are catching up. This patch changes the way anon_vmas and VMAs are linked, which allows us to associate multiple anon_vmas with a VMA. At fork time, each child process gets its own anon_vmas, in which its COWed pages will be instantiated. The parents' anon_vma is also linked to the VMA, because non-COWed pages could be present in any of the children. This reduces rmap scanning complexity to O(1) for the pages of the 1000 child processes, with O(N) complexity for at most 1/N pages in the system. This reduces the average scanning cost in heavily forking workloads from O(N) to 2. The only real complexity in this patch stems from the fact that linking a VMA to anon_vmas now involves memory allocations. This means vma_adjust can fail, if it needs to attach a VMA to anon_vma structures. This in turn means error handling needs to be added to the calling functions. A second source of complexity is that, because there can be multiple anon_vmas, the anon_vma linking in vma_adjust can no longer be done under "the" anon_vma lock. To prevent the rmap code from walking up an incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h to make sure it is impossible to compile a kernel that needs both symbolic values for the same bitflag. Some test results: Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test box with 16GB RAM and not quite enough IO), the system ends up running >99% in system time, with every CPU on the same anon_vma lock in the pageout code. With these changes, AIM7 hits the cross-over point around 29.7k users. This happens with ~99% IO wait time, there never seems to be any spike in system time. The anon_vma lock contention appears to be resolved. [akpm@linux-foundation.org: cleanups] Signed-off-by: Rik van Riel <riel@redhat.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Larry Woodman <lwoodman@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com> Cc: Minchan Kim <minchan.kim@gmail.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 21:42:07 +00:00
INIT_LIST_HEAD(&vma->anon_vma_chain);
vma->vm_mm = current->mm;
vma->vm_end = PAGE_SIZE;
vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | VM_RESERVED;
down_write(&current->mm->mmap_sem);
if (insert_vm_struct(current->mm, vma)) {
up_write(&current->mm->mmap_sem);
kmem_cache_free(vm_area_cachep, vma);
return;
}
up_write(&current->mm->mmap_sem);
}
}
}
void
free_initmem (void)
{
unsigned long addr, eaddr;
addr = (unsigned long) ia64_imva(__init_begin);
eaddr = (unsigned long) ia64_imva(__init_end);
while (addr < eaddr) {
ClearPageReserved(virt_to_page(addr));
init_page_count(virt_to_page(addr));
free_page(addr);
++totalram_pages;
addr += PAGE_SIZE;
}
printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n",
(__init_end - __init_begin) >> 10);
}
void __init
free_initrd_mem (unsigned long start, unsigned long end)
{
struct page *page;
/*
* EFI uses 4KB pages while the kernel can use 4KB or bigger.
* Thus EFI and the kernel may have different page sizes. It is
* therefore possible to have the initrd share the same page as
* the end of the kernel (given current setup).
*
* To avoid freeing/using the wrong page (kernel sized) we:
* - align up the beginning of initrd
* - align down the end of initrd
*
* | |
* |=============| a000
* | |
* | |
* | | 9000
* |/////////////|
* |/////////////|
* |=============| 8000
* |///INITRD////|
* |/////////////|
* |/////////////| 7000
* | |
* |KKKKKKKKKKKKK|
* |=============| 6000
* |KKKKKKKKKKKKK|
* |KKKKKKKKKKKKK|
* K=kernel using 8KB pages
*
* In this example, we must free page 8000 ONLY. So we must align up
* initrd_start and keep initrd_end as is.
*/
start = PAGE_ALIGN(start);
end = end & PAGE_MASK;
if (start < end)
printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
for (; start < end; start += PAGE_SIZE) {
if (!virt_addr_valid(start))
continue;
page = virt_to_page(start);
ClearPageReserved(page);
init_page_count(page);
free_page(start);
++totalram_pages;
}
}
/*
* This installs a clean page in the kernel's page table.
*/
static struct page * __init
put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
{
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
if (!PageReserved(page))
printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
page_address(page));
pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
{
pud = pud_alloc(&init_mm, pgd, address);
if (!pud)
goto out;
pmd = pmd_alloc(&init_mm, pud, address);
if (!pmd)
goto out;
[PATCH] mm: init_mm without ptlock First step in pushing down the page_table_lock. init_mm.page_table_lock has been used throughout the architectures (usually for ioremap): not to serialize kernel address space allocation (that's usually vmlist_lock), but because pud_alloc,pmd_alloc,pte_alloc_kernel expect caller holds it. Reverse that: don't lock or unlock init_mm.page_table_lock in any of the architectures; instead rely on pud_alloc,pmd_alloc,pte_alloc_kernel to take and drop it when allocating a new one, to check lest a racing task already did. Similarly no page_table_lock in vmalloc's map_vm_area. Some temporary ugliness in __pud_alloc and __pmd_alloc: since they also handle user mms, which are converted only by a later patch, for now they have to lock differently according to whether or not it's init_mm. If sources get muddled, there's a danger that an arch source taking init_mm.page_table_lock will be mixed with common source also taking it (or neither take it). So break the rules and make another change, which should break the build for such a mismatch: remove the redundant mm arg from pte_alloc_kernel (ppc64 scrapped its distinct ioremap_mm in 2.6.13). Exceptions: arm26 used pte_alloc_kernel on user mm, now pte_alloc_map; ia64 used pte_alloc_map on init_mm, now pte_alloc_kernel; parisc had bad args to pmd_alloc and pte_alloc_kernel in unused USE_HPPA_IOREMAP code; ppc64 map_io_page forgot to unlock on failure; ppc mmu_mapin_ram and ppc64 im_free took page_table_lock for no good reason. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:21 +00:00
pte = pte_alloc_kernel(pmd, address);
if (!pte)
goto out;
[PATCH] mm: init_mm without ptlock First step in pushing down the page_table_lock. init_mm.page_table_lock has been used throughout the architectures (usually for ioremap): not to serialize kernel address space allocation (that's usually vmlist_lock), but because pud_alloc,pmd_alloc,pte_alloc_kernel expect caller holds it. Reverse that: don't lock or unlock init_mm.page_table_lock in any of the architectures; instead rely on pud_alloc,pmd_alloc,pte_alloc_kernel to take and drop it when allocating a new one, to check lest a racing task already did. Similarly no page_table_lock in vmalloc's map_vm_area. Some temporary ugliness in __pud_alloc and __pmd_alloc: since they also handle user mms, which are converted only by a later patch, for now they have to lock differently according to whether or not it's init_mm. If sources get muddled, there's a danger that an arch source taking init_mm.page_table_lock will be mixed with common source also taking it (or neither take it). So break the rules and make another change, which should break the build for such a mismatch: remove the redundant mm arg from pte_alloc_kernel (ppc64 scrapped its distinct ioremap_mm in 2.6.13). Exceptions: arm26 used pte_alloc_kernel on user mm, now pte_alloc_map; ia64 used pte_alloc_map on init_mm, now pte_alloc_kernel; parisc had bad args to pmd_alloc and pte_alloc_kernel in unused USE_HPPA_IOREMAP code; ppc64 map_io_page forgot to unlock on failure; ppc mmu_mapin_ram and ppc64 im_free took page_table_lock for no good reason. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:21 +00:00
if (!pte_none(*pte))
goto out;
set_pte(pte, mk_pte(page, pgprot));
}
[PATCH] mm: init_mm without ptlock First step in pushing down the page_table_lock. init_mm.page_table_lock has been used throughout the architectures (usually for ioremap): not to serialize kernel address space allocation (that's usually vmlist_lock), but because pud_alloc,pmd_alloc,pte_alloc_kernel expect caller holds it. Reverse that: don't lock or unlock init_mm.page_table_lock in any of the architectures; instead rely on pud_alloc,pmd_alloc,pte_alloc_kernel to take and drop it when allocating a new one, to check lest a racing task already did. Similarly no page_table_lock in vmalloc's map_vm_area. Some temporary ugliness in __pud_alloc and __pmd_alloc: since they also handle user mms, which are converted only by a later patch, for now they have to lock differently according to whether or not it's init_mm. If sources get muddled, there's a danger that an arch source taking init_mm.page_table_lock will be mixed with common source also taking it (or neither take it). So break the rules and make another change, which should break the build for such a mismatch: remove the redundant mm arg from pte_alloc_kernel (ppc64 scrapped its distinct ioremap_mm in 2.6.13). Exceptions: arm26 used pte_alloc_kernel on user mm, now pte_alloc_map; ia64 used pte_alloc_map on init_mm, now pte_alloc_kernel; parisc had bad args to pmd_alloc and pte_alloc_kernel in unused USE_HPPA_IOREMAP code; ppc64 map_io_page forgot to unlock on failure; ppc mmu_mapin_ram and ppc64 im_free took page_table_lock for no good reason. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:21 +00:00
out:
/* no need for flush_tlb */
return page;
}
static void __init
setup_gate (void)
{
void *gate_section;
struct page *page;
/*
* Map the gate page twice: once read-only to export the ELF
* headers etc. and once execute-only page to enable
* privilege-promotion via "epc":
*/
gate_section = paravirt_get_gate_section();
page = virt_to_page(ia64_imva(gate_section));
put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
#ifdef HAVE_BUGGY_SEGREL
page = virt_to_page(ia64_imva(gate_section + PAGE_SIZE));
put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
#else
put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
/* Fill in the holes (if any) with read-only zero pages: */
{
unsigned long addr;
for (addr = GATE_ADDR + PAGE_SIZE;
addr < GATE_ADDR + PERCPU_PAGE_SIZE;
addr += PAGE_SIZE)
{
put_kernel_page(ZERO_PAGE(0), addr,
PAGE_READONLY);
put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
PAGE_READONLY);
}
}
#endif
ia64_patch_gate();
}
void __devinit
ia64_mmu_init (void *my_cpu_data)
{
unsigned long pta, impl_va_bits;
extern void __devinit tlb_init (void);
#ifdef CONFIG_DISABLE_VHPT
# define VHPT_ENABLE_BIT 0
#else
# define VHPT_ENABLE_BIT 1
#endif
/*
* Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
* address space. The IA-64 architecture guarantees that at least 50 bits of
* virtual address space are implemented but if we pick a large enough page size
* (e.g., 64KB), the mapped address space is big enough that it will overlap with
* VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
* IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
* problem in practice. Alternatively, we could truncate the top of the mapped
* address space to not permit mappings that would overlap with the VMLPT.
* --davidm 00/12/06
*/
# define pte_bits 3
# define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
/*
* The virtual page table has to cover the entire implemented address space within
* a region even though not all of this space may be mappable. The reason for
* this is that the Access bit and Dirty bit fault handlers perform
* non-speculative accesses to the virtual page table, so the address range of the
* virtual page table itself needs to be covered by virtual page table.
*/
# define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
# define POW2(n) (1ULL << (n))
impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
if (impl_va_bits < 51 || impl_va_bits > 61)
panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
/*
* mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
* which must fit into "vmlpt_bits - pte_bits" slots. Second half of
* the test makes sure that our mapped space doesn't overlap the
* unimplemented hole in the middle of the region.
*/
if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
(mapped_space_bits > impl_va_bits - 1))
panic("Cannot build a big enough virtual-linear page table"
" to cover mapped address space.\n"
" Try using a smaller page size.\n");
/* place the VMLPT at the end of each page-table mapped region: */
pta = POW2(61) - POW2(vmlpt_bits);
/*
* Set the (virtually mapped linear) page table address. Bit
* 8 selects between the short and long format, bits 2-7 the
* size of the table, and bit 0 whether the VHPT walker is
* enabled.
*/
ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
ia64_tlb_init();
#ifdef CONFIG_HUGETLB_PAGE
ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
ia64_srlz_d();
#endif
}
#ifdef CONFIG_VIRTUAL_MEM_MAP
int vmemmap_find_next_valid_pfn(int node, int i)
{
unsigned long end_address, hole_next_pfn;
unsigned long stop_address;
pg_data_t *pgdat = NODE_DATA(node);
end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
end_address = PAGE_ALIGN(end_address);
stop_address = (unsigned long) &vmem_map[
pgdat->node_start_pfn + pgdat->node_spanned_pages];
do {
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
pgd = pgd_offset_k(end_address);
if (pgd_none(*pgd)) {
end_address += PGDIR_SIZE;
continue;
}
pud = pud_offset(pgd, end_address);
if (pud_none(*pud)) {
end_address += PUD_SIZE;
continue;
}
pmd = pmd_offset(pud, end_address);
if (pmd_none(*pmd)) {
end_address += PMD_SIZE;
continue;
}
pte = pte_offset_kernel(pmd, end_address);
retry_pte:
if (pte_none(*pte)) {
end_address += PAGE_SIZE;
pte++;
if ((end_address < stop_address) &&
(end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
goto retry_pte;
continue;
}
/* Found next valid vmem_map page */
break;
} while (end_address < stop_address);
end_address = min(end_address, stop_address);
end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
hole_next_pfn = end_address / sizeof(struct page);
return hole_next_pfn - pgdat->node_start_pfn;
}
int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
{
unsigned long address, start_page, end_page;
struct page *map_start, *map_end;
int node;
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
start_page = (unsigned long) map_start & PAGE_MASK;
end_page = PAGE_ALIGN((unsigned long) map_end);
node = paddr_to_nid(__pa(start));
for (address = start_page; address < end_page; address += PAGE_SIZE) {
pgd = pgd_offset_k(address);
if (pgd_none(*pgd))
pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pud = pud_offset(pgd, address);
if (pud_none(*pud))
pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pmd = pmd_offset(pud, address);
if (pmd_none(*pmd))
pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pte = pte_offset_kernel(pmd, address);
if (pte_none(*pte))
set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
PAGE_KERNEL));
}
return 0;
}
struct memmap_init_callback_data {
struct page *start;
struct page *end;
int nid;
unsigned long zone;
};
static int __meminit
virtual_memmap_init(u64 start, u64 end, void *arg)
{
struct memmap_init_callback_data *args;
struct page *map_start, *map_end;
args = (struct memmap_init_callback_data *) arg;
map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
if (map_start < args->start)
map_start = args->start;
if (map_end > args->end)
map_end = args->end;
/*
* We have to initialize "out of bounds" struct page elements that fit completely
* on the same pages that were allocated for the "in bounds" elements because they
* may be referenced later (and found to be "reserved").
*/
map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
/ sizeof(struct page));
if (map_start < map_end)
memmap_init_zone((unsigned long)(map_end - map_start),
args->nid, args->zone, page_to_pfn(map_start),
MEMMAP_EARLY);
return 0;
}
void __meminit
memmap_init (unsigned long size, int nid, unsigned long zone,
unsigned long start_pfn)
{
if (!vmem_map)
memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
else {
struct page *start;
struct memmap_init_callback_data args;
start = pfn_to_page(start_pfn);
args.start = start;
args.end = start + size;
args.nid = nid;
args.zone = zone;
efi_memmap_walk(virtual_memmap_init, &args);
}
}
int
ia64_pfn_valid (unsigned long pfn)
{
char byte;
struct page *pg = pfn_to_page(pfn);
return (__get_user(byte, (char __user *) pg) == 0)
&& ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
|| (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
}
EXPORT_SYMBOL(ia64_pfn_valid);
int __init find_largest_hole(u64 start, u64 end, void *arg)
{
u64 *max_gap = arg;
static u64 last_end = PAGE_OFFSET;
/* NOTE: this algorithm assumes efi memmap table is ordered */
if (*max_gap < (start - last_end))
*max_gap = start - last_end;
last_end = end;
return 0;
}
#endif /* CONFIG_VIRTUAL_MEM_MAP */
int __init register_active_ranges(u64 start, u64 len, int nid)
{
u64 end = start + len;
#ifdef CONFIG_KEXEC
if (start > crashk_res.start && start < crashk_res.end)
start = crashk_res.end;
if (end > crashk_res.start && end < crashk_res.end)
end = crashk_res.start;
#endif
if (start < end)
add_active_range(nid, __pa(start) >> PAGE_SHIFT,
__pa(end) >> PAGE_SHIFT);
return 0;
}
static int __init
count_reserved_pages(u64 start, u64 end, void *arg)
{
unsigned long num_reserved = 0;
unsigned long *count = arg;
for (; start < end; start += PAGE_SIZE)
if (PageReserved(virt_to_page(start)))
++num_reserved;
*count += num_reserved;
return 0;
}
[IA64] min_low_pfn and max_low_pfn calculation fix We have seen bad_pte_print when testing crashdump on an SN machine in recent 2.6.20 kernel. There are tons of bad pte print (pfn < max_low_pfn) reports when the crash kernel boots up, all those reported bad pages are inside initmem range; That is because if the crash kernel code and data happens to be at the beginning of the 1st node. build_node_maps in discontig.c will bypass reserved regions with filter_rsvd_memory. Since min_low_pfn is calculated in build_node_map, so in this case, min_low_pfn will be greater than kernel code and data. Because pages inside initmem are freed and reused later, we saw pfn_valid check fail on those pages. I think this theoretically happen on a normal kernel. When I check min_low_pfn and max_low_pfn calculation in contig.c and discontig.c. I found more issues than this. 1. min_low_pfn and max_low_pfn calculation is inconsistent between contig.c and discontig.c, min_low_pfn is calculated as the first page number of boot memmap in contig.c (Why? Though this may work at the most of the time, I don't think it is the right logic). It is calculated as the lowest physical memory page number bypass reserved regions in discontig.c. max_low_pfn is calculated include reserved regions in contig.c. It is calculated exclude reserved regions in discontig.c. 2. If kernel code and data region is happen to be at the begin or the end of physical memory, when min_low_pfn and max_low_pfn calculation is bypassed kernel code and data, pages in initmem will report bad. 3. initrd is also in reserved regions, if it is at the begin or at the end of physical memory, kernel will refuse to reuse the memory. Because the virt_addr_valid check in free_initrd_mem. So it is better to fix and clean up those issues. Calculate min_low_pfn and max_low_pfn in a consistent way. Signed-off-by: Zou Nan hai <nanhai.zou@intel.com> Acked-by: Jay Lan <jlan@sgi.com> Signed-off-by: Tony Luck <tony.luck@intel.com>
2007-03-20 20:41:57 +00:00
int
find_max_min_low_pfn (u64 start, u64 end, void *arg)
[IA64] min_low_pfn and max_low_pfn calculation fix We have seen bad_pte_print when testing crashdump on an SN machine in recent 2.6.20 kernel. There are tons of bad pte print (pfn < max_low_pfn) reports when the crash kernel boots up, all those reported bad pages are inside initmem range; That is because if the crash kernel code and data happens to be at the beginning of the 1st node. build_node_maps in discontig.c will bypass reserved regions with filter_rsvd_memory. Since min_low_pfn is calculated in build_node_map, so in this case, min_low_pfn will be greater than kernel code and data. Because pages inside initmem are freed and reused later, we saw pfn_valid check fail on those pages. I think this theoretically happen on a normal kernel. When I check min_low_pfn and max_low_pfn calculation in contig.c and discontig.c. I found more issues than this. 1. min_low_pfn and max_low_pfn calculation is inconsistent between contig.c and discontig.c, min_low_pfn is calculated as the first page number of boot memmap in contig.c (Why? Though this may work at the most of the time, I don't think it is the right logic). It is calculated as the lowest physical memory page number bypass reserved regions in discontig.c. max_low_pfn is calculated include reserved regions in contig.c. It is calculated exclude reserved regions in discontig.c. 2. If kernel code and data region is happen to be at the begin or the end of physical memory, when min_low_pfn and max_low_pfn calculation is bypassed kernel code and data, pages in initmem will report bad. 3. initrd is also in reserved regions, if it is at the begin or at the end of physical memory, kernel will refuse to reuse the memory. Because the virt_addr_valid check in free_initrd_mem. So it is better to fix and clean up those issues. Calculate min_low_pfn and max_low_pfn in a consistent way. Signed-off-by: Zou Nan hai <nanhai.zou@intel.com> Acked-by: Jay Lan <jlan@sgi.com> Signed-off-by: Tony Luck <tony.luck@intel.com>
2007-03-20 20:41:57 +00:00
{
unsigned long pfn_start, pfn_end;
#ifdef CONFIG_FLATMEM
pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
#else
pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
#endif
min_low_pfn = min(min_low_pfn, pfn_start);
max_low_pfn = max(max_low_pfn, pfn_end);
return 0;
}
/*
* Boot command-line option "nolwsys" can be used to disable the use of any light-weight
* system call handler. When this option is in effect, all fsyscalls will end up bubbling
* down into the kernel and calling the normal (heavy-weight) syscall handler. This is
* useful for performance testing, but conceivably could also come in handy for debugging
* purposes.
*/
static int nolwsys __initdata;
static int __init
nolwsys_setup (char *s)
{
nolwsys = 1;
return 1;
}
__setup("nolwsys", nolwsys_setup);
void __init
mem_init (void)
{
long reserved_pages, codesize, datasize, initsize;
pg_data_t *pgdat;
int i;
[IA64] Percpu quicklist for combined allocator for pgd/pmd/pte. This patch introduces using the quicklists for pgd, pmd, and pte levels by combining the alloc and free functions into a common set of routines. This greatly simplifies the reading of this header file. This patch is simple but necessary for large numa configurations. It simply ensures that only pages from the local node are added to a cpus quicklist. This prevents the trapping of pages on a remote nodes quicklist by starting a process, touching a large number of pages to fill pmd and pte entries, migrating to another node, and then unmapping or exiting. With those conditions, the pages get trapped and if the machine has more than 100 nodes of the same size, the calculation of the pgtable high water mark will be larger than any single node so page table cache flushing will never occur. I ran lmbench lat_proc fork and lat_proc exec on a zx1 with and without this patch and did not notice any change. On an sn2 machine, there was a slight improvement which is possibly due to pages from other nodes trapped on the test node before starting the run. I did not investigate further. This patch shrinks the quicklist based upon free memory on the node instead of the high/low water marks. I have written it to enable preemption periodically and recalculate the amount to shrink every time we have freed enough pages that the quicklist size should have grown. I rescan the nodes zones each pass because other processess may be draining node memory at the same time as we are adding. Signed-off-by: Robin Holt <holt@sgi.com> Signed-off-by: Tony Luck <tony.luck@intel.com>
2005-04-25 20:13:16 +00:00
BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
#ifdef CONFIG_PCI
/*
* This needs to be called _after_ the command line has been parsed but _before_
* any drivers that may need the PCI DMA interface are initialized or bootmem has
* been freed.
*/
platform_dma_init();
#endif
#ifdef CONFIG_FLATMEM
BUG_ON(!mem_map);
max_mapnr = max_low_pfn;
#endif
high_memory = __va(max_low_pfn * PAGE_SIZE);
for_each_online_pgdat(pgdat)
if (pgdat->bdata->node_bootmem_map)
totalram_pages += free_all_bootmem_node(pgdat);
reserved_pages = 0;
efi_memmap_walk(count_reserved_pages, &reserved_pages);
codesize = (unsigned long) _etext - (unsigned long) _stext;
datasize = (unsigned long) _edata - (unsigned long) _etext;
initsize = (unsigned long) __init_end - (unsigned long) __init_begin;
printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, "
"%luk data, %luk init)\n", nr_free_pages() << (PAGE_SHIFT - 10),
num_physpages << (PAGE_SHIFT - 10), codesize >> 10,
reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10);
/*
* For fsyscall entrpoints with no light-weight handler, use the ordinary
* (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
* code can tell them apart.
*/
for (i = 0; i < NR_syscalls; ++i) {
extern unsigned long sys_call_table[NR_syscalls];
unsigned long *fsyscall_table = paravirt_get_fsyscall_table();
if (!fsyscall_table[i] || nolwsys)
fsyscall_table[i] = sys_call_table[i] | 1;
}
setup_gate();
}
#ifdef CONFIG_MEMORY_HOTPLUG
int arch_add_memory(int nid, u64 start, u64 size)
{
pg_data_t *pgdat;
struct zone *zone;
unsigned long start_pfn = start >> PAGE_SHIFT;
unsigned long nr_pages = size >> PAGE_SHIFT;
int ret;
pgdat = NODE_DATA(nid);
zone = pgdat->node_zones + ZONE_NORMAL;
mm: show node to memory section relationship with symlinks in sysfs Show node to memory section relationship with symlinks in sysfs Add /sys/devices/system/node/nodeX/memoryY symlinks for all the memory sections located on nodeX. For example: /sys/devices/system/node/node1/memory135 -> ../../memory/memory135 indicates that memory section 135 resides on node1. Also revises documentation to cover this change as well as updating Documentation/ABI/testing/sysfs-devices-memory to include descriptions of memory hotremove files 'phys_device', 'phys_index', and 'state' that were previously not described there. In addition to it always being a good policy to provide users with the maximum possible amount of physical location information for resources that can be hot-added and/or hot-removed, the following are some (but likely not all) of the user benefits provided by this change. Immediate: - Provides information needed to determine the specific node on which a defective DIMM is located. This will reduce system downtime when the node or defective DIMM is swapped out. - Prevents unintended onlining of a memory section that was previously offlined due to a defective DIMM. This could happen during node hot-add when the user or node hot-add assist script onlines _all_ offlined sections due to user or script inability to identify the specific memory sections located on the hot-added node. The consequences of reintroducing the defective memory could be ugly. - Provides information needed to vary the amount and distribution of memory on specific nodes for testing or debugging purposes. Future: - Will provide information needed to identify the memory sections that need to be offlined prior to physical removal of a specific node. Symlink creation during boot was tested on 2-node x86_64, 2-node ppc64, and 2-node ia64 systems. Symlink creation during physical memory hot-add tested on a 2-node x86_64 system. Signed-off-by: Gary Hade <garyhade@us.ibm.com> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Acked-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:14 +00:00
ret = __add_pages(nid, zone, start_pfn, nr_pages);
if (ret)
printk("%s: Problem encountered in __add_pages() as ret=%d\n",
__func__, ret);
return ret;
}
#endif
/*
* Even when CONFIG_IA32_SUPPORT is not enabled it is
* useful to have the Linux/x86 domain registered to
* avoid an attempted module load when emulators call
* personality(PER_LINUX32). This saves several milliseconds
* on each such call.
*/
static struct exec_domain ia32_exec_domain;
static int __init
per_linux32_init(void)
{
ia32_exec_domain.name = "Linux/x86";
ia32_exec_domain.handler = NULL;
ia32_exec_domain.pers_low = PER_LINUX32;
ia32_exec_domain.pers_high = PER_LINUX32;
ia32_exec_domain.signal_map = default_exec_domain.signal_map;
ia32_exec_domain.signal_invmap = default_exec_domain.signal_invmap;
register_exec_domain(&ia32_exec_domain);
return 0;
}
__initcall(per_linux32_init);