linux/arch/s390/mm/pgalloc.c

372 lines
9.0 KiB
C
Raw Normal View History

/*
* Page table allocation functions
*
* Copyright IBM Corp. 2016
* Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
*/
#include <linux/mm.h>
#include <linux/sysctl.h>
#include <asm/mmu_context.h>
#include <asm/pgalloc.h>
#include <asm/gmap.h>
#include <asm/tlb.h>
#include <asm/tlbflush.h>
#ifdef CONFIG_PGSTE
static int page_table_allocate_pgste_min = 0;
static int page_table_allocate_pgste_max = 1;
int page_table_allocate_pgste = 0;
EXPORT_SYMBOL(page_table_allocate_pgste);
static struct ctl_table page_table_sysctl[] = {
{
.procname = "allocate_pgste",
.data = &page_table_allocate_pgste,
.maxlen = sizeof(int),
.mode = S_IRUGO | S_IWUSR,
.proc_handler = proc_dointvec,
.extra1 = &page_table_allocate_pgste_min,
.extra2 = &page_table_allocate_pgste_max,
},
{ }
};
static struct ctl_table page_table_sysctl_dir[] = {
{
.procname = "vm",
.maxlen = 0,
.mode = 0555,
.child = page_table_sysctl,
},
{ }
};
static int __init page_table_register_sysctl(void)
{
return register_sysctl_table(page_table_sysctl_dir) ? 0 : -ENOMEM;
}
__initcall(page_table_register_sysctl);
#endif /* CONFIG_PGSTE */
unsigned long *crst_table_alloc(struct mm_struct *mm)
{
struct page *page = alloc_pages(GFP_KERNEL, 2);
if (!page)
return NULL;
arch_set_page_dat(page, 2);
return (unsigned long *) page_to_phys(page);
}
void crst_table_free(struct mm_struct *mm, unsigned long *table)
{
free_pages((unsigned long) table, 2);
}
static void __crst_table_upgrade(void *arg)
{
struct mm_struct *mm = arg;
if (current->active_mm == mm) {
clear_user_asce();
set_user_asce(mm);
}
__tlb_flush_local();
}
int crst_table_upgrade(struct mm_struct *mm, unsigned long end)
{
unsigned long *table, *pgd;
int rc, notify;
/* upgrade should only happen from 3 to 4, 3 to 5, or 4 to 5 levels */
VM_BUG_ON(mm->context.asce_limit < _REGION2_SIZE);
if (end >= TASK_SIZE_MAX)
return -ENOMEM;
rc = 0;
notify = 0;
while (mm->context.asce_limit < end) {
table = crst_table_alloc(mm);
if (!table) {
rc = -ENOMEM;
break;
}
spin_lock_bh(&mm->page_table_lock);
pgd = (unsigned long *) mm->pgd;
if (mm->context.asce_limit == _REGION2_SIZE) {
crst_table_init(table, _REGION2_ENTRY_EMPTY);
p4d_populate(mm, (p4d_t *) table, (pud_t *) pgd);
mm->pgd = (pgd_t *) table;
mm->context.asce_limit = _REGION1_SIZE;
mm->context.asce = __pa(mm->pgd) | _ASCE_TABLE_LENGTH |
_ASCE_USER_BITS | _ASCE_TYPE_REGION2;
} else {
crst_table_init(table, _REGION1_ENTRY_EMPTY);
pgd_populate(mm, (pgd_t *) table, (p4d_t *) pgd);
mm->pgd = (pgd_t *) table;
mm->context.asce_limit = -PAGE_SIZE;
mm->context.asce = __pa(mm->pgd) | _ASCE_TABLE_LENGTH |
_ASCE_USER_BITS | _ASCE_TYPE_REGION1;
}
notify = 1;
spin_unlock_bh(&mm->page_table_lock);
}
if (notify)
on_each_cpu(__crst_table_upgrade, mm, 0);
return rc;
}
s390/mm: fix asce_bits handling with dynamic pagetable levels There is a race with multi-threaded applications between context switch and pagetable upgrade. In switch_mm() a new user_asce is built from mm->pgd and mm->context.asce_bits, w/o holding any locks. A concurrent mmap with a pagetable upgrade on another thread in crst_table_upgrade() could already have set new asce_bits, but not yet the new mm->pgd. This would result in a corrupt user_asce in switch_mm(), and eventually in a kernel panic from a translation exception. Fix this by storing the complete asce instead of just the asce_bits, which can then be read atomically from switch_mm(), so that it either sees the old value or the new value, but no mixture. Both cases are OK. Having the old value would result in a page fault on access to the higher level memory, but the fault handler would see the new mm->pgd, if it was a valid access after the mmap on the other thread has completed. So as worst-case scenario we would have a page fault loop for the racing thread until the next time slice. Also remove dead code and simplify the upgrade/downgrade path, there are no upgrades from 2 levels, and only downgrades from 3 levels for compat tasks. There are also no concurrent upgrades, because the mmap_sem is held with down_write() in do_mmap, so the flush and table checks during upgrade can be removed. Reported-by: Michael Munday <munday@ca.ibm.com> Reviewed-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2016-04-15 16:38:40 +02:00
void crst_table_downgrade(struct mm_struct *mm)
{
pgd_t *pgd;
s390/mm: fix asce_bits handling with dynamic pagetable levels There is a race with multi-threaded applications between context switch and pagetable upgrade. In switch_mm() a new user_asce is built from mm->pgd and mm->context.asce_bits, w/o holding any locks. A concurrent mmap with a pagetable upgrade on another thread in crst_table_upgrade() could already have set new asce_bits, but not yet the new mm->pgd. This would result in a corrupt user_asce in switch_mm(), and eventually in a kernel panic from a translation exception. Fix this by storing the complete asce instead of just the asce_bits, which can then be read atomically from switch_mm(), so that it either sees the old value or the new value, but no mixture. Both cases are OK. Having the old value would result in a page fault on access to the higher level memory, but the fault handler would see the new mm->pgd, if it was a valid access after the mmap on the other thread has completed. So as worst-case scenario we would have a page fault loop for the racing thread until the next time slice. Also remove dead code and simplify the upgrade/downgrade path, there are no upgrades from 2 levels, and only downgrades from 3 levels for compat tasks. There are also no concurrent upgrades, because the mmap_sem is held with down_write() in do_mmap, so the flush and table checks during upgrade can be removed. Reported-by: Michael Munday <munday@ca.ibm.com> Reviewed-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2016-04-15 16:38:40 +02:00
/* downgrade should only happen from 3 to 2 levels (compat only) */
VM_BUG_ON(mm->context.asce_limit != _REGION2_SIZE);
s390/mm: fix asce_bits handling with dynamic pagetable levels There is a race with multi-threaded applications between context switch and pagetable upgrade. In switch_mm() a new user_asce is built from mm->pgd and mm->context.asce_bits, w/o holding any locks. A concurrent mmap with a pagetable upgrade on another thread in crst_table_upgrade() could already have set new asce_bits, but not yet the new mm->pgd. This would result in a corrupt user_asce in switch_mm(), and eventually in a kernel panic from a translation exception. Fix this by storing the complete asce instead of just the asce_bits, which can then be read atomically from switch_mm(), so that it either sees the old value or the new value, but no mixture. Both cases are OK. Having the old value would result in a page fault on access to the higher level memory, but the fault handler would see the new mm->pgd, if it was a valid access after the mmap on the other thread has completed. So as worst-case scenario we would have a page fault loop for the racing thread until the next time slice. Also remove dead code and simplify the upgrade/downgrade path, there are no upgrades from 2 levels, and only downgrades from 3 levels for compat tasks. There are also no concurrent upgrades, because the mmap_sem is held with down_write() in do_mmap, so the flush and table checks during upgrade can be removed. Reported-by: Michael Munday <munday@ca.ibm.com> Reviewed-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2016-04-15 16:38:40 +02:00
if (current->active_mm == mm) {
clear_user_asce();
__tlb_flush_mm(mm);
}
s390/mm: fix asce_bits handling with dynamic pagetable levels There is a race with multi-threaded applications between context switch and pagetable upgrade. In switch_mm() a new user_asce is built from mm->pgd and mm->context.asce_bits, w/o holding any locks. A concurrent mmap with a pagetable upgrade on another thread in crst_table_upgrade() could already have set new asce_bits, but not yet the new mm->pgd. This would result in a corrupt user_asce in switch_mm(), and eventually in a kernel panic from a translation exception. Fix this by storing the complete asce instead of just the asce_bits, which can then be read atomically from switch_mm(), so that it either sees the old value or the new value, but no mixture. Both cases are OK. Having the old value would result in a page fault on access to the higher level memory, but the fault handler would see the new mm->pgd, if it was a valid access after the mmap on the other thread has completed. So as worst-case scenario we would have a page fault loop for the racing thread until the next time slice. Also remove dead code and simplify the upgrade/downgrade path, there are no upgrades from 2 levels, and only downgrades from 3 levels for compat tasks. There are also no concurrent upgrades, because the mmap_sem is held with down_write() in do_mmap, so the flush and table checks during upgrade can be removed. Reported-by: Michael Munday <munday@ca.ibm.com> Reviewed-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2016-04-15 16:38:40 +02:00
pgd = mm->pgd;
mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
mm->context.asce_limit = _REGION3_SIZE;
s390/mm: fix asce_bits handling with dynamic pagetable levels There is a race with multi-threaded applications between context switch and pagetable upgrade. In switch_mm() a new user_asce is built from mm->pgd and mm->context.asce_bits, w/o holding any locks. A concurrent mmap with a pagetable upgrade on another thread in crst_table_upgrade() could already have set new asce_bits, but not yet the new mm->pgd. This would result in a corrupt user_asce in switch_mm(), and eventually in a kernel panic from a translation exception. Fix this by storing the complete asce instead of just the asce_bits, which can then be read atomically from switch_mm(), so that it either sees the old value or the new value, but no mixture. Both cases are OK. Having the old value would result in a page fault on access to the higher level memory, but the fault handler would see the new mm->pgd, if it was a valid access after the mmap on the other thread has completed. So as worst-case scenario we would have a page fault loop for the racing thread until the next time slice. Also remove dead code and simplify the upgrade/downgrade path, there are no upgrades from 2 levels, and only downgrades from 3 levels for compat tasks. There are also no concurrent upgrades, because the mmap_sem is held with down_write() in do_mmap, so the flush and table checks during upgrade can be removed. Reported-by: Michael Munday <munday@ca.ibm.com> Reviewed-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2016-04-15 16:38:40 +02:00
mm->context.asce = __pa(mm->pgd) | _ASCE_TABLE_LENGTH |
_ASCE_USER_BITS | _ASCE_TYPE_SEGMENT;
crst_table_free(mm, (unsigned long *) pgd);
if (current->active_mm == mm)
set_user_asce(mm);
}
static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
{
unsigned int old, new;
do {
old = atomic_read(v);
new = old ^ bits;
} while (atomic_cmpxchg(v, old, new) != old);
return new;
}
s390/mm: add shadow gmap support For a nested KVM guest the outer KVM host needs to create shadow page tables for the nested guest. This patch adds the basic support to the guest address space (gmap) code. For each guest address space the inner KVM host creates, the first outer KVM host needs to create shadow page tables. The address space is identified by the ASCE loaded into the control register 1 at the time the inner SIE instruction for the second nested KVM guest is executed. The outer KVM host creates the shadow tables starting with the table identified by the ASCE on a on-demand basis. The outer KVM host will get repeated faults for all the shadow tables needed to run the second KVM guest. While a shadow page table for the second KVM guest is active the access to the origin region, segment and page tables needs to be restricted for the first KVM guest. For region and segment and page tables the first KVM guest may read the memory, but write attempt has to lead to an unshadow. This is done using the page invalid and read-only bits in the page table of the first KVM guest. If the first guest re-accesses one of the origin pages of a shadow, it gets a fault and the affected parts of the shadow page table hierarchy needs to be removed again. PGSTE tables don't have to be shadowed, as all interpretation assist can't deal with the invalid bits in the shadow pte being set differently than the original ones provided by the first KVM guest. Many bug fixes and improvements by David Hildenbrand. Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
2016-03-08 12:12:18 +01:00
#ifdef CONFIG_PGSTE
struct page *page_table_alloc_pgste(struct mm_struct *mm)
{
struct page *page;
unsigned long *table;
page = alloc_page(GFP_KERNEL);
s390/mm: add shadow gmap support For a nested KVM guest the outer KVM host needs to create shadow page tables for the nested guest. This patch adds the basic support to the guest address space (gmap) code. For each guest address space the inner KVM host creates, the first outer KVM host needs to create shadow page tables. The address space is identified by the ASCE loaded into the control register 1 at the time the inner SIE instruction for the second nested KVM guest is executed. The outer KVM host creates the shadow tables starting with the table identified by the ASCE on a on-demand basis. The outer KVM host will get repeated faults for all the shadow tables needed to run the second KVM guest. While a shadow page table for the second KVM guest is active the access to the origin region, segment and page tables needs to be restricted for the first KVM guest. For region and segment and page tables the first KVM guest may read the memory, but write attempt has to lead to an unshadow. This is done using the page invalid and read-only bits in the page table of the first KVM guest. If the first guest re-accesses one of the origin pages of a shadow, it gets a fault and the affected parts of the shadow page table hierarchy needs to be removed again. PGSTE tables don't have to be shadowed, as all interpretation assist can't deal with the invalid bits in the shadow pte being set differently than the original ones provided by the first KVM guest. Many bug fixes and improvements by David Hildenbrand. Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
2016-03-08 12:12:18 +01:00
if (page) {
table = (unsigned long *) page_to_phys(page);
clear_table(table, _PAGE_INVALID, PAGE_SIZE/2);
clear_table(table + PTRS_PER_PTE, 0, PAGE_SIZE/2);
}
return page;
}
void page_table_free_pgste(struct page *page)
{
__free_page(page);
}
#endif /* CONFIG_PGSTE */
/*
* page table entry allocation/free routines.
*/
unsigned long *page_table_alloc(struct mm_struct *mm)
{
unsigned long *table;
struct page *page;
unsigned int mask, bit;
/* Try to get a fragment of a 4K page as a 2K page table */
if (!mm_alloc_pgste(mm)) {
table = NULL;
spin_lock_bh(&mm->context.lock);
if (!list_empty(&mm->context.pgtable_list)) {
page = list_first_entry(&mm->context.pgtable_list,
struct page, lru);
mask = atomic_read(&page->_mapcount);
mask = (mask | (mask >> 4)) & 3;
if (mask != 3) {
table = (unsigned long *) page_to_phys(page);
bit = mask & 1; /* =1 -> second 2K */
if (bit)
table += PTRS_PER_PTE;
atomic_xor_bits(&page->_mapcount, 1U << bit);
list_del(&page->lru);
}
}
spin_unlock_bh(&mm->context.lock);
if (table)
return table;
}
/* Allocate a fresh page */
page = alloc_page(GFP_KERNEL);
if (!page)
return NULL;
if (!pgtable_page_ctor(page)) {
__free_page(page);
return NULL;
}
arch_set_page_dat(page, 0);
/* Initialize page table */
table = (unsigned long *) page_to_phys(page);
if (mm_alloc_pgste(mm)) {
/* Return 4K page table with PGSTEs */
atomic_set(&page->_mapcount, 3);
clear_table(table, _PAGE_INVALID, PAGE_SIZE/2);
clear_table(table + PTRS_PER_PTE, 0, PAGE_SIZE/2);
} else {
/* Return the first 2K fragment of the page */
atomic_set(&page->_mapcount, 1);
clear_table(table, _PAGE_INVALID, PAGE_SIZE);
spin_lock_bh(&mm->context.lock);
list_add(&page->lru, &mm->context.pgtable_list);
spin_unlock_bh(&mm->context.lock);
}
return table;
}
void page_table_free(struct mm_struct *mm, unsigned long *table)
{
struct page *page;
unsigned int bit, mask;
page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
if (!mm_alloc_pgste(mm)) {
/* Free 2K page table fragment of a 4K page */
bit = (__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t));
spin_lock_bh(&mm->context.lock);
mask = atomic_xor_bits(&page->_mapcount, 1U << bit);
if (mask & 3)
list_add(&page->lru, &mm->context.pgtable_list);
else
list_del(&page->lru);
spin_unlock_bh(&mm->context.lock);
if (mask != 0)
return;
}
pgtable_page_dtor(page);
atomic_set(&page->_mapcount, -1);
__free_page(page);
}
void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table,
unsigned long vmaddr)
{
struct mm_struct *mm;
struct page *page;
unsigned int bit, mask;
mm = tlb->mm;
page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
if (mm_alloc_pgste(mm)) {
gmap_unlink(mm, table, vmaddr);
table = (unsigned long *) (__pa(table) | 3);
tlb_remove_table(tlb, table);
return;
}
bit = (__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t));
spin_lock_bh(&mm->context.lock);
mask = atomic_xor_bits(&page->_mapcount, 0x11U << bit);
if (mask & 3)
list_add_tail(&page->lru, &mm->context.pgtable_list);
else
list_del(&page->lru);
spin_unlock_bh(&mm->context.lock);
table = (unsigned long *) (__pa(table) | (1U << bit));
tlb_remove_table(tlb, table);
}
static void __tlb_remove_table(void *_table)
{
unsigned int mask = (unsigned long) _table & 3;
void *table = (void *)((unsigned long) _table ^ mask);
struct page *page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
switch (mask) {
case 0: /* pmd, pud, or p4d */
free_pages((unsigned long) table, 2);
break;
case 1: /* lower 2K of a 4K page table */
case 2: /* higher 2K of a 4K page table */
if (atomic_xor_bits(&page->_mapcount, mask << 4) != 0)
break;
/* fallthrough */
case 3: /* 4K page table with pgstes */
pgtable_page_dtor(page);
atomic_set(&page->_mapcount, -1);
__free_page(page);
break;
}
}
static void tlb_remove_table_smp_sync(void *arg)
{
/* Simply deliver the interrupt */
}
static void tlb_remove_table_one(void *table)
{
/*
* This isn't an RCU grace period and hence the page-tables cannot be
* assumed to be actually RCU-freed.
*
* It is however sufficient for software page-table walkers that rely
* on IRQ disabling. See the comment near struct mmu_table_batch.
*/
smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
__tlb_remove_table(table);
}
static void tlb_remove_table_rcu(struct rcu_head *head)
{
struct mmu_table_batch *batch;
int i;
batch = container_of(head, struct mmu_table_batch, rcu);
for (i = 0; i < batch->nr; i++)
__tlb_remove_table(batch->tables[i]);
free_page((unsigned long)batch);
}
void tlb_table_flush(struct mmu_gather *tlb)
{
struct mmu_table_batch **batch = &tlb->batch;
if (*batch) {
call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
*batch = NULL;
}
}
void tlb_remove_table(struct mmu_gather *tlb, void *table)
{
struct mmu_table_batch **batch = &tlb->batch;
tlb->mm->context.flush_mm = 1;
if (*batch == NULL) {
*batch = (struct mmu_table_batch *)
__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
if (*batch == NULL) {
__tlb_flush_mm_lazy(tlb->mm);
tlb_remove_table_one(table);
return;
}
(*batch)->nr = 0;
}
(*batch)->tables[(*batch)->nr++] = table;
if ((*batch)->nr == MAX_TABLE_BATCH)
tlb_flush_mmu(tlb);
}