d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add support for pmap_enter(..., psind=1) to the amd64 pmap. In other words,

Browse Source 
add support for explicitly requesting that pmap_enter() create a 2MB page
mapping.  (Essentially, this feature allows the machine-independent layer to
create superpage mappings preemptively, and not wait for automatic promotion
to occur.)

Export pmap_ps_enabled() to the machine-independent layer.

Add a flag to pmap_pv_insert_pde() that specifies whether it should fail or
reclaim a PV entry when one is not available.

Refactor pmap_enter_pde() into two functions, one by the same name, that is
a general-purpose function for creating PDE PG_PS mappings, and another,
pmap_enter_2mpage(), that is used to prefault 2MB read- and/or execute-only
mappings for execve(2), mmap(2), and shmat(2).

Submitted by:	Yufeng Zhou <yz70@rice.edu> (an earlier version)
Reviewed by:	kib, markj
Tested by:	pho
MFC after:	10 days
Differential Revision:	https://reviews.freebsd.org/D11556
This commit is contained in:
Alan Cox 2017-07-23 06:33:58 +00:00
parent 1d3b9818e7
commit 782e896088
3 changed files with 150 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

199
sys/amd64/amd64/pmap.c
View File

@ -582,6 +582,12 @@ pmap_delayed_invl_page(vm_page_t m)
*/
static caddr_t crashdumpmap;
/*
* Internal flags for pmap_enter()'s helper functions.
*/
#define PMAP_ENTER_NORECLAIM 0x1000000 /* Don't reclaim PV entries. */
#define PMAP_ENTER_NOREPLACE 0x2000000 /* Don't replace mappings. */
static void free_pv_chunk(struct pv_chunk *pc);
static void free_pv_entry(pmap_t pmap, pv_entry_t pv);
static pv_entry_t get_pv_entry(pmap_t pmap, struct rwlock **lockp);
@ -591,8 +597,8 @@ static void reserve_pv_entries(pmap_t pmap, int needed,
struct rwlock **lockp);
static void pmap_pv_demote_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa,
struct rwlock **lockp);
static boolean_t pmap_pv_insert_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa,
struct rwlock **lockp);
static bool pmap_pv_insert_pde(pmap_t pmap, vm_offset_t va, pd_entry_t pde,
u_int flags, struct rwlock **lockp);
static void pmap_pv_promote_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa,
struct rwlock **lockp);
static void pmap_pvh_free(struct md_page *pvh, pmap_t pmap, vm_offset_t va);
@ -605,8 +611,10 @@ static boolean_t pmap_demote_pde_locked(pmap_t pmap, pd_entry_t *pde,
vm_offset_t va, struct rwlock **lockp);
static boolean_t pmap_demote_pdpe(pmap_t pmap, pdp_entry_t *pdpe,
vm_offset_t va);
static boolean_t pmap_enter_pde(pmap_t pmap, vm_offset_t va, vm_page_t m,
vm_prot_t prot, struct rwlock **lockp);
static bool pmap_enter_2mpage(pmap_t pmap, vm_offset_t va, vm_page_t m,
vm_prot_t prot, struct rwlock **lockp);
static int pmap_enter_pde(pmap_t pmap, vm_offset_t va, pd_entry_t newpde,
u_int flags, vm_page_t m, struct rwlock **lockp);
static vm_page_t pmap_enter_quick_locked(pmap_t pmap, vm_offset_t va,
vm_page_t m, vm_prot_t prot, vm_page_t mpte, struct rwlock **lockp);
static void pmap_fill_ptp(pt_entry_t *firstpte, pt_entry_t newpte);
@ -1389,7 +1397,7 @@ pmap_cache_mask(pmap_t pmap, boolean_t is_pde)
return (mask);
}
static __inline boolean_t
bool
pmap_ps_enabled(pmap_t pmap)
{
@ -3393,27 +3401,30 @@ pmap_try_insert_pv_entry(pmap_t pmap, vm_offset_t va, vm_page_t m,
}
/*
* Conditionally create the PV entry for a 2MB page mapping if the required
* memory can be allocated without resorting to reclamation.
* Create the PV entry for a 2MB page mapping. Always returns true unless the
* flag PMAP_ENTER_NORECLAIM is specified. If that flag is specified, returns
* false if the PV entry cannot be allocated without resorting to reclamation.
*/
static boolean_t
pmap_pv_insert_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa,
static bool
pmap_pv_insert_pde(pmap_t pmap, vm_offset_t va, pd_entry_t pde, u_int flags,
struct rwlock **lockp)
{
struct md_page *pvh;
pv_entry_t pv;
vm_paddr_t pa;
PMAP_LOCK_ASSERT(pmap, MA_OWNED);
/* Pass NULL instead of the lock pointer to disable reclamation. */
if ((pv = get_pv_entry(pmap, NULL)) != NULL) {
pv->pv_va = va;
CHANGE_PV_LIST_LOCK_TO_PHYS(lockp, pa);
pvh = pa_to_pvh(pa);
TAILQ_INSERT_TAIL(&pvh->pv_list, pv, pv_next);
pvh->pv_gen++;
return (TRUE);
} else
return (FALSE);
if ((pv = get_pv_entry(pmap, (flags & PMAP_ENTER_NORECLAIM) != 0 ?
NULL : lockp)) == NULL)
return (false);
pv->pv_va = va;
pa = pde & PG_PS_FRAME;
CHANGE_PV_LIST_LOCK_TO_PHYS(lockp, pa);
pvh = pa_to_pvh(pa);
TAILQ_INSERT_TAIL(&pvh->pv_list, pv, pv_next);
pvh->pv_gen++;
return (true);
}
/*
@ -4323,7 +4334,7 @@ pmap_promote_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va,
*/
int
pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
u_int flags, int8_t psind __unused)
u_int flags, int8_t psind)
{
struct rwlock *lock;
pd_entry_t *pde;
@ -4351,6 +4362,8 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
("pmap_enter: managed mapping within the clean submap"));
if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
VM_OBJECT_ASSERT_LOCKED(m->object);
KASSERT((flags & PMAP_ENTER_RESERVED) == 0,
("pmap_enter: flags %u has reserved bits set", flags));
pa = VM_PAGE_TO_PHYS(m);
newpte = (pt_entry_t)(pa | PG_A | PG_V);
if ((flags & VM_PROT_WRITE) != 0)
@ -4367,7 +4380,7 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
newpte |= PG_U;
if (pmap == kernel_pmap)
newpte |= PG_G;
newpte |= pmap_cache_bits(pmap, m->md.pat_mode, 0);
newpte |= pmap_cache_bits(pmap, m->md.pat_mode, psind > 0);
/*
* Set modified bit gratuitously for writeable mappings if
@ -4380,10 +4393,16 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
} else
newpte |= PG_MANAGED;
mpte = NULL;
lock = NULL;
PMAP_LOCK(pmap);
if (psind == 1) {
/* Assert the required virtual and physical alignment. */
KASSERT((va & PDRMASK) == 0, ("pmap_enter: va unaligned"));
KASSERT(m->psind > 0, ("pmap_enter: m->psind < psind"));
rv = pmap_enter_pde(pmap, va, newpte | PG_PS, flags, m, &lock);
goto out;
}
mpte = NULL;
/*
* In the case that a page table page is not
@ -4543,48 +4562,118 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
}
/*
* Tries to create a 2MB page mapping. Returns TRUE if successful and FALSE
* otherwise. Fails if (1) a page table page cannot be allocated without
* blocking, (2) a mapping already exists at the specified virtual address, or
* (3) a pv entry cannot be allocated without reclaiming another pv entry.
* Tries to create a read- and/or execute-only 2MB page mapping. Returns true
* if successful. Returns false if (1) a page table page cannot be allocated
* without sleeping, (2) a mapping already exists at the specified virtual
* address, or (3) a PV entry cannot be allocated without reclaiming another
* PV entry.
*/
static boolean_t
pmap_enter_pde(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
static bool
pmap_enter_2mpage(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
struct rwlock **lockp)
{
pd_entry_t *pde, newpde;
pd_entry_t newpde;
pt_entry_t PG_V;
vm_page_t pdpg;
struct spglist free;
PMAP_LOCK_ASSERT(pmap, MA_OWNED);
PG_V = pmap_valid_bit(pmap);
newpde = VM_PAGE_TO_PHYS(m) | pmap_cache_bits(pmap, m->md.pat_mode, 1) |
PG_PS | PG_V;
if ((m->oflags & VPO_UNMANAGED) == 0)
newpde |= PG_MANAGED;
if ((prot & VM_PROT_EXECUTE) == 0)
newpde |= pg_nx;
if (va < VM_MAXUSER_ADDRESS)
newpde |= PG_U;
return (pmap_enter_pde(pmap, va, newpde, PMAP_ENTER_NOSLEEP |
PMAP_ENTER_NOREPLACE | PMAP_ENTER_NORECLAIM, NULL, lockp) ==
KERN_SUCCESS);
}
/*
* Tries to create the specified 2MB page mapping. Returns KERN_SUCCESS if
* the mapping was created, and either KERN_FAILURE or KERN_RESOURCE_SHORTAGE
* otherwise. Returns KERN_FAILURE if PMAP_ENTER_NOREPLACE was specified and
* a mapping already exists at the specified virtual address. Returns
* KERN_RESOURCE_SHORTAGE if PMAP_ENTER_NOSLEEP was specified and a page table
* page allocation failed. Returns KERN_RESOURCE_SHORTAGE if
* PMAP_ENTER_NORECLAIM was specified and a PV entry allocation failed.
*
* The parameter "m" is only used when creating a managed, writeable mapping.
*/
static int
pmap_enter_pde(pmap_t pmap, vm_offset_t va, pd_entry_t newpde, u_int flags,
vm_page_t m, struct rwlock **lockp)
{
struct spglist free;
pd_entry_t oldpde, *pde;
pt_entry_t PG_G, PG_RW, PG_V;
vm_page_t mt, pdpg;
PG_G = pmap_global_bit(pmap);
PG_RW = pmap_rw_bit(pmap);
KASSERT((newpde & (pmap_modified_bit(pmap) | PG_RW)) != PG_RW,
("pmap_enter_pde: newpde is missing PG_M"));
PG_V = pmap_valid_bit(pmap);
PMAP_LOCK_ASSERT(pmap, MA_OWNED);
if ((pdpg = pmap_allocpde(pmap, va, NULL)) == NULL) {
if ((pdpg = pmap_allocpde(pmap, va, (flags & PMAP_ENTER_NOSLEEP) != 0 ?
NULL : lockp)) == NULL) {
CTR2(KTR_PMAP, "pmap_enter_pde: failure for va %#lx"
" in pmap %p", va, pmap);
return (FALSE);
return (KERN_RESOURCE_SHORTAGE);
}
pde = (pd_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pdpg));
pde = &pde[pmap_pde_index(va)];
if ((*pde & PG_V) != 0) {
oldpde = *pde;
if ((oldpde & PG_V) != 0) {
KASSERT(pdpg->wire_count > 1,
("pmap_enter_pde: pdpg's wire count is too low"));
pdpg->wire_count--;
CTR2(KTR_PMAP, "pmap_enter_pde: failure for va %#lx"
" in pmap %p", va, pmap);
return (FALSE);
if ((flags & PMAP_ENTER_NOREPLACE) != 0) {
pdpg->wire_count--;
CTR2(KTR_PMAP, "pmap_enter_pde: failure for va %#lx"
" in pmap %p", va, pmap);
return (KERN_FAILURE);
}
/* Break the existing mapping(s). */
SLIST_INIT(&free);
if ((oldpde & PG_PS) != 0) {
/*
* The reference to the PD page that was acquired by
* pmap_allocpde() ensures that it won't be freed.
* However, if the PDE resulted from a promotion, then
* a reserved PT page could be freed.
*/
(void)pmap_remove_pde(pmap, pde, va, &free, lockp);
if ((oldpde & PG_G) == 0)
pmap_invalidate_pde_page(pmap, va, oldpde);
} else {
pmap_delayed_invl_started();
if (pmap_remove_ptes(pmap, va, va + NBPDR, pde, &free,
lockp))
pmap_invalidate_all(pmap);
pmap_delayed_invl_finished();
}
pmap_free_zero_pages(&free);
if (va >= VM_MAXUSER_ADDRESS) {
mt = PHYS_TO_VM_PAGE(*pde & PG_FRAME);
if (pmap_insert_pt_page(pmap, mt)) {
/*
* XXX Currently, this can't happen because
* we do not perform pmap_enter(psind == 1)
* on the kernel pmap.
*/
panic("pmap_enter_pde: trie insert failed");
}
} else
KASSERT(*pde == 0, ("pmap_enter_pde: non-zero pde %p",
pde));
}
newpde = VM_PAGE_TO_PHYS(m) | pmap_cache_bits(pmap, m->md.pat_mode, 1) |
PG_PS | PG_V;
if ((m->oflags & VPO_UNMANAGED) == 0) {
newpde |= PG_MANAGED;
if ((newpde & PG_MANAGED) != 0) {
/*
* Abort this mapping if its PV entry could not be created.
*/
if (!pmap_pv_insert_pde(pmap, va, VM_PAGE_TO_PHYS(m),
lockp)) {
if (!pmap_pv_insert_pde(pmap, va, newpde, flags, lockp)) {
SLIST_INIT(&free);
if (pmap_unwire_ptp(pmap, va, pdpg, &free)) {
/*
@ -4598,17 +4687,19 @@ pmap_enter_pde(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
}
CTR2(KTR_PMAP, "pmap_enter_pde: failure for va %#lx"
" in pmap %p", va, pmap);
return (FALSE);
return (KERN_RESOURCE_SHORTAGE);
}
if ((newpde & PG_RW) != 0) {
for (mt = m; mt < &m[NBPDR / PAGE_SIZE]; mt++)
vm_page_aflag_set(mt, PGA_WRITEABLE);
}
}
if ((prot & VM_PROT_EXECUTE) == 0)
newpde |= pg_nx;
if (va < VM_MAXUSER_ADDRESS)
newpde |= PG_U;
/*
* Increment counters.
*/
if ((newpde & PG_W) != 0)
pmap->pm_stats.wired_count += NBPDR / PAGE_SIZE;
pmap_resident_count_inc(pmap, NBPDR / PAGE_SIZE);
/*
@ -4620,7 +4711,7 @@ pmap_enter_pde(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
atomic_add_long(&pmap_pde_mappings, 1);
CTR2(KTR_PMAP, "pmap_enter_pde: success for va %#lx"
" in pmap %p", va, pmap);
return (TRUE);
return (KERN_SUCCESS);
}
/*
@ -4655,7 +4746,7 @@ pmap_enter_object(pmap_t pmap, vm_offset_t start, vm_offset_t end,
va = start + ptoa(diff);
if ((va & PDRMASK) == 0 && va + NBPDR <= end &&
m->psind == 1 && pmap_ps_enabled(pmap) &&
pmap_enter_pde(pmap, va, m, prot, &lock))
pmap_enter_2mpage(pmap, va, m, prot, &lock))
m = &m[NBPDR / PAGE_SIZE - 1];
else
mpte = pmap_enter_quick_locked(pmap, va, m, prot,
@ -5088,8 +5179,8 @@ pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr, vm_size_t len,
PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dst_pdpg));
pde = &pde[pmap_pde_index(addr)];
if (*pde == 0 && ((srcptepaddr & PG_MANAGED) == 0 ||
pmap_pv_insert_pde(dst_pmap, addr, srcptepaddr &
PG_PS_FRAME, &lock))) {
pmap_pv_insert_pde(dst_pmap, addr, srcptepaddr,
PMAP_ENTER_NORECLAIM, &lock))) {
*pde = srcptepaddr & ~PG_W;
pmap_resident_count_inc(dst_pmap, NBPDR / PAGE_SIZE);
atomic_add_long(&pmap_pde_mappings, 1);

1
sys/amd64/include/pmap.h
View File

@ -408,6 +408,7 @@ void *pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
boolean_t pmap_page_is_mapped(vm_page_t m);
void pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
void pmap_pinit_pml4(vm_page_t);
bool pmap_ps_enabled(pmap_t pmap);
void pmap_unmapdev(vm_offset_t, vm_size_t);
void pmap_invalidate_page(pmap_t, vm_offset_t);
void pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);

6
sys/vm/pmap.h
View File

@ -100,9 +100,11 @@ extern vm_offset_t kernel_vm_end;
/*
* Flags for pmap_enter(). The bits in the low-order byte are reserved
* for the protection code (vm_prot_t) that describes the fault type.
* Bits 24 through 31 are reserved for the pmap's internal use.
*/
#define PMAP_ENTER_NOSLEEP 0x0100
#define PMAP_ENTER_WIRED 0x0200
#define PMAP_ENTER_NOSLEEP 0x00000100
#define PMAP_ENTER_WIRED 0x00000200
#define PMAP_ENTER_RESERVED 0xFF000000
/*
* Define the maximum number of machine-dependent reference bits that are