linux/arch/s390/mm/hugetlbpage.c
Martin Schwidefsky 1b948d6cae s390/mm,tlb: optimize TLB flushing for zEC12
The zEC12 machines introduced the local-clearing control for the IDTE
and IPTE instruction. If the control is set only the TLB of the local
CPU is cleared of entries, either all entries of a single address space
for IDTE, or the entry for a single page-table entry for IPTE.
Without the local-clearing control the TLB flush is broadcasted to all
CPUs in the configuration, which is expensive.

The reset of the bit mask of the CPUs that need flushing after a
non-local IDTE is tricky. As TLB entries for an address space remain
in the TLB even if the address space is detached a new bit field is
required to keep track of attached CPUs vs. CPUs in the need of a
flush. After a non-local flush with IDTE the bit-field of attached CPUs
is copied to the bit-field of CPUs in need of a flush. The ordering
of operations on cpu_attach_mask, attach_count and mm_cpumask(mm) is
such that an underindication in mm_cpumask(mm) is prevented but an
overindication in mm_cpumask(mm) is possible.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-04-03 14:31:00 +02:00

241 lines
5.7 KiB
C

/*
* IBM System z Huge TLB Page Support for Kernel.
*
* Copyright IBM Corp. 2007
* Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com>
*/
#include <linux/mm.h>
#include <linux/hugetlb.h>
static inline pmd_t __pte_to_pmd(pte_t pte)
{
int none, young, prot;
pmd_t pmd;
/*
* Convert encoding pte bits pmd bits
* .IR...wrdytp ..R...I...y.
* empty .10...000000 -> ..0...1...0.
* prot-none, clean, old .11...000001 -> ..0...1...1.
* prot-none, clean, young .11...000101 -> ..1...1...1.
* prot-none, dirty, old .10...001001 -> ..0...1...1.
* prot-none, dirty, young .10...001101 -> ..1...1...1.
* read-only, clean, old .11...010001 -> ..1...1...0.
* read-only, clean, young .01...010101 -> ..1...0...1.
* read-only, dirty, old .11...011001 -> ..1...1...0.
* read-only, dirty, young .01...011101 -> ..1...0...1.
* read-write, clean, old .11...110001 -> ..0...1...0.
* read-write, clean, young .01...110101 -> ..0...0...1.
* read-write, dirty, old .10...111001 -> ..0...1...0.
* read-write, dirty, young .00...111101 -> ..0...0...1.
* Huge ptes are dirty by definition, a clean pte is made dirty
* by the conversion.
*/
if (pte_present(pte)) {
pmd_val(pmd) = pte_val(pte) & PAGE_MASK;
if (pte_val(pte) & _PAGE_INVALID)
pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
none = (pte_val(pte) & _PAGE_PRESENT) &&
!(pte_val(pte) & _PAGE_READ) &&
!(pte_val(pte) & _PAGE_WRITE);
prot = (pte_val(pte) & _PAGE_PROTECT) &&
!(pte_val(pte) & _PAGE_WRITE);
young = pte_val(pte) & _PAGE_YOUNG;
if (none || young)
pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
if (prot || (none && young))
pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
} else
pmd_val(pmd) = _SEGMENT_ENTRY_INVALID;
return pmd;
}
static inline pte_t __pmd_to_pte(pmd_t pmd)
{
pte_t pte;
/*
* Convert encoding pmd bits pte bits
* ..R...I...y. .IR...wrdytp
* empty ..0...1...0. -> .10...000000
* prot-none, old ..0...1...1. -> .10...001001
* prot-none, young ..1...1...1. -> .10...001101
* read-only, old ..1...1...0. -> .11...011001
* read-only, young ..1...0...1. -> .01...011101
* read-write, old ..0...1...0. -> .10...111001
* read-write, young ..0...0...1. -> .00...111101
* Huge ptes are dirty by definition
*/
if (pmd_present(pmd)) {
pte_val(pte) = _PAGE_PRESENT | _PAGE_LARGE | _PAGE_DIRTY |
(pmd_val(pmd) & PAGE_MASK);
if (pmd_val(pmd) & _SEGMENT_ENTRY_INVALID)
pte_val(pte) |= _PAGE_INVALID;
if (pmd_prot_none(pmd)) {
if (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT)
pte_val(pte) |= _PAGE_YOUNG;
} else {
pte_val(pte) |= _PAGE_READ;
if (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT)
pte_val(pte) |= _PAGE_PROTECT;
else
pte_val(pte) |= _PAGE_WRITE;
if (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG)
pte_val(pte) |= _PAGE_YOUNG;
}
} else
pte_val(pte) = _PAGE_INVALID;
return pte;
}
void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
pte_t *ptep, pte_t pte)
{
pmd_t pmd;
pmd = __pte_to_pmd(pte);
if (!MACHINE_HAS_HPAGE) {
pmd_val(pmd) &= ~_SEGMENT_ENTRY_ORIGIN;
pmd_val(pmd) |= pte_page(pte)[1].index;
} else
pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_CO;
*(pmd_t *) ptep = pmd;
}
pte_t huge_ptep_get(pte_t *ptep)
{
unsigned long origin;
pmd_t pmd;
pmd = *(pmd_t *) ptep;
if (!MACHINE_HAS_HPAGE && pmd_present(pmd)) {
origin = pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN;
pmd_val(pmd) &= ~_SEGMENT_ENTRY_ORIGIN;
pmd_val(pmd) |= *(unsigned long *) origin;
}
return __pmd_to_pte(pmd);
}
pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
unsigned long addr, pte_t *ptep)
{
pmd_t *pmdp = (pmd_t *) ptep;
pte_t pte = huge_ptep_get(ptep);
pmdp_flush_direct(mm, addr, pmdp);
pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
return pte;
}
int arch_prepare_hugepage(struct page *page)
{
unsigned long addr = page_to_phys(page);
pte_t pte;
pte_t *ptep;
int i;
if (MACHINE_HAS_HPAGE)
return 0;
ptep = (pte_t *) pte_alloc_one(&init_mm, addr);
if (!ptep)
return -ENOMEM;
pte_val(pte) = addr;
for (i = 0; i < PTRS_PER_PTE; i++) {
set_pte_at(&init_mm, addr + i * PAGE_SIZE, ptep + i, pte);
pte_val(pte) += PAGE_SIZE;
}
page[1].index = (unsigned long) ptep;
return 0;
}
void arch_release_hugepage(struct page *page)
{
pte_t *ptep;
if (MACHINE_HAS_HPAGE)
return;
ptep = (pte_t *) page[1].index;
if (!ptep)
return;
clear_table((unsigned long *) ptep, _PAGE_INVALID,
PTRS_PER_PTE * sizeof(pte_t));
page_table_free(&init_mm, (unsigned long *) ptep);
page[1].index = 0;
}
pte_t *huge_pte_alloc(struct mm_struct *mm,
unsigned long addr, unsigned long sz)
{
pgd_t *pgdp;
pud_t *pudp;
pmd_t *pmdp = NULL;
pgdp = pgd_offset(mm, addr);
pudp = pud_alloc(mm, pgdp, addr);
if (pudp)
pmdp = pmd_alloc(mm, pudp, addr);
return (pte_t *) pmdp;
}
pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
{
pgd_t *pgdp;
pud_t *pudp;
pmd_t *pmdp = NULL;
pgdp = pgd_offset(mm, addr);
if (pgd_present(*pgdp)) {
pudp = pud_offset(pgdp, addr);
if (pud_present(*pudp))
pmdp = pmd_offset(pudp, addr);
}
return (pte_t *) pmdp;
}
int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep)
{
return 0;
}
struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
int write)
{
return ERR_PTR(-EINVAL);
}
int pmd_huge(pmd_t pmd)
{
if (!MACHINE_HAS_HPAGE)
return 0;
return !!(pmd_val(pmd) & _SEGMENT_ENTRY_LARGE);
}
int pud_huge(pud_t pud)
{
return 0;
}
int pmd_huge_support(void)
{
return 1;
}
struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
pmd_t *pmdp, int write)
{
struct page *page;
if (!MACHINE_HAS_HPAGE)
return NULL;
page = pmd_page(*pmdp);
if (page)
page += ((address & ~HPAGE_MASK) >> PAGE_SHIFT);
return page;
}