git: fd6cb031f577 - main - arm64 pmap: Add ATTR_CONTIGUOUS support [Part 1]
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Sat, 30 Mar 2024 18:46:23 UTC
The branch main has been updated by alc:
URL: https://cgit.FreeBSD.org/src/commit/?id=fd6cb031f577a449894e73daa8f6bd309ba27c73
commit fd6cb031f577a449894e73daa8f6bd309ba27c73
Author: Eliot Solomon <ehs3@rice.edu>
AuthorDate: 2024-03-24 19:01:47 +0000
Commit: Alan Cox <alc@FreeBSD.org>
CommitDate: 2024-03-30 18:37:17 +0000
arm64 pmap: Add ATTR_CONTIGUOUS support [Part 1]
The ATTR_CONTIGUOUS bit within an L3 page table entry designates that
L3 page as being part of an aligned, physically contiguous collection
of L3 pages. For example, 16 aligned, physically contiguous 4 KB pages
can form a 64 KB superpage, occupying a single TLB entry. While this
change only creates ATTR_CONTIGUOUS mappings in a few places,
specifically, the direct map and pmap_kenter{,_device}(), it adds all
of the necessary code for handling them once they exist, including
demotion, protection, and removal. Consequently, new ATTR_CONTIGUOUS
usage can be added (and tested) incrementally.
Modify the implementation of sysctl vm.pmap.kernel_maps so that it
correctly reports the number of ATTR_CONTIGUOUS mappings on machines
configured to use a 16 KB base page size, where an ATTR_CONTIGUOUS
mapping consists of 128 base pages.
Additionally, this change adds support for creating L2 superpage
mappings to pmap_kenter{,_device}().
Reviewed by: markj
Tested by: gallatin
Differential Revision: https://reviews.freebsd.org/D42737
---
sys/arm64/arm64/pmap.c | 767 +++++++++++++++++++++++++++++++++++++++++++++---
sys/arm64/include/pte.h | 21 ++
2 files changed, 740 insertions(+), 48 deletions(-)
diff --git a/sys/arm64/arm64/pmap.c b/sys/arm64/arm64/pmap.c
index ba72f1dac8d0..12e5e1d73b38 100644
--- a/sys/arm64/arm64/pmap.c
+++ b/sys/arm64/arm64/pmap.c
@@ -461,18 +461,33 @@ static bool pmap_activate_int(pmap_t pmap);
static void pmap_alloc_asid(pmap_t pmap);
static int pmap_change_props_locked(vm_offset_t va, vm_size_t size,
vm_prot_t prot, int mode, bool skip_unmapped);
+static bool pmap_copy_l3c(pmap_t pmap, pt_entry_t *l3p, vm_offset_t va,
+ pt_entry_t l3e, vm_page_t ml3, struct rwlock **lockp);
static pt_entry_t *pmap_demote_l1(pmap_t pmap, pt_entry_t *l1, vm_offset_t va);
static pt_entry_t *pmap_demote_l2_locked(pmap_t pmap, pt_entry_t *l2,
vm_offset_t va, struct rwlock **lockp);
static pt_entry_t *pmap_demote_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t va);
+static bool pmap_demote_l3c(pmap_t pmap, pt_entry_t *l3p, vm_offset_t va);
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 int pmap_enter_l2(pmap_t pmap, vm_offset_t va, pd_entry_t new_l2,
u_int flags, vm_page_t m, struct rwlock **lockp);
+static bool pmap_every_pte_zero(vm_paddr_t pa);
+static int pmap_insert_pt_page(pmap_t pmap, vm_page_t mpte, bool promoted,
+ bool all_l3e_AF_set);
+static pt_entry_t pmap_load_l3c(pt_entry_t *l3p);
+static void pmap_mask_set_l3c(pmap_t pmap, pt_entry_t *l3p, vm_offset_t va,
+ vm_offset_t *vap, vm_offset_t va_next, pt_entry_t mask, pt_entry_t nbits);
+static bool pmap_pv_insert_l3c(pmap_t pmap, vm_offset_t va, vm_page_t m,
+ struct rwlock **lockp);
+static void pmap_remove_kernel_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t va);
static int pmap_remove_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t sva,
pd_entry_t l1e, struct spglist *free, struct rwlock **lockp);
static int pmap_remove_l3(pmap_t pmap, pt_entry_t *l3, vm_offset_t sva,
pd_entry_t l2e, struct spglist *free, struct rwlock **lockp);
+static bool pmap_remove_l3c(pmap_t pmap, pt_entry_t *l3p, vm_offset_t va,
+ vm_offset_t *vap, vm_offset_t va_next, vm_page_t ml3, struct spglist *free,
+ struct rwlock **lockp);
static void pmap_reset_asid_set(pmap_t pmap);
static bool pmap_try_insert_pv_entry(pmap_t pmap, vm_offset_t va,
vm_page_t m, struct rwlock **lockp);
@@ -483,6 +498,8 @@ static vm_page_t _pmap_alloc_l3(pmap_t pmap, vm_pindex_t ptepindex,
static void _pmap_unwire_l3(pmap_t pmap, vm_offset_t va, vm_page_t m,
struct spglist *free);
static int pmap_unuse_pt(pmap_t, vm_offset_t, pd_entry_t, struct spglist *);
+static void pmap_update_entry(pmap_t pmap, pd_entry_t *pte, pd_entry_t newpte,
+ vm_offset_t va, vm_size_t size);
static __inline vm_page_t pmap_remove_pt_page(pmap_t pmap, vm_offset_t va);
static uma_zone_t pmap_bti_ranges_zone;
@@ -1121,19 +1138,20 @@ pmap_bootstrap_l2_block(struct pmap_bootstrap_state *state, int i)
static void
pmap_bootstrap_l3_page(struct pmap_bootstrap_state *state, int i)
{
+ pt_entry_t contig;
u_int l3_slot;
bool first;
- if ((physmap[i + 1] - state->pa) < L3_SIZE)
+ if (physmap[i + 1] - state->pa < L3_SIZE)
return;
/* Make sure there is a valid L2 table */
pmap_bootstrap_l2_table(state);
MPASS((state->va & L3_OFFSET) == 0);
- for (first = true;
+ for (first = true, contig = 0;
state->va < DMAP_MAX_ADDRESS &&
- (physmap[i + 1] - state->pa) >= L3_SIZE;
+ physmap[i + 1] - state->pa >= L3_SIZE;
state->va += L3_SIZE, state->pa += L3_SIZE) {
/*
* Stop if we are about to walk off the end of what the
@@ -1142,13 +1160,27 @@ pmap_bootstrap_l3_page(struct pmap_bootstrap_state *state, int i)
if (!first && (state->pa & L2_OFFSET) == 0)
break;
+ /*
+ * If we have an aligned, contiguous chunk of L3C_ENTRIES
+ * L3 pages, set the contiguous bit within each PTE so that
+ * the chunk can be cached using only one TLB entry.
+ */
+ if ((state->pa & L3C_OFFSET) == 0) {
+ if (state->va + L3C_SIZE < DMAP_MAX_ADDRESS &&
+ physmap[i + 1] - state->pa >= L3C_SIZE) {
+ contig = ATTR_CONTIGUOUS;
+ } else {
+ contig = 0;
+ }
+ }
+
first = false;
l3_slot = pmap_l3_index(state->va);
MPASS((state->pa & L3_OFFSET) == 0);
MPASS(state->l3[l3_slot] == 0);
pmap_store(&state->l3[l3_slot], PHYS_TO_PTE(state->pa) |
ATTR_DEFAULT | ATTR_S1_XN | ATTR_KERN_GP |
- ATTR_S1_IDX(VM_MEMATTR_WRITE_BACK) | L3_PAGE);
+ ATTR_S1_IDX(VM_MEMATTR_WRITE_BACK) | contig | L3_PAGE);
}
MPASS(state->va == (state->pa - dmap_phys_base + DMAP_MIN_ADDRESS));
}
@@ -1649,6 +1681,17 @@ static u_long pmap_l2_promotions;
SYSCTL_ULONG(_vm_pmap_l2, OID_AUTO, promotions, CTLFLAG_RD,
&pmap_l2_promotions, 0, "2MB page promotions");
+static SYSCTL_NODE(_vm_pmap, OID_AUTO, l3c, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
+ "64KB page mapping counters");
+
+static u_long pmap_l3c_demotions;
+SYSCTL_ULONG(_vm_pmap_l3c, OID_AUTO, demotions, CTLFLAG_RD,
+ &pmap_l3c_demotions, 0, "64KB page demotions");
+
+static u_long pmap_l3c_mappings;
+SYSCTL_ULONG(_vm_pmap_l3c, OID_AUTO, mappings, CTLFLAG_RD,
+ &pmap_l3c_mappings, 0, "64KB page mappings");
+
/*
* If the given value for "final_only" is false, then any cached intermediate-
* level entries, i.e., L{0,1,2}_TABLE entries, are invalidated in addition to
@@ -2016,7 +2059,8 @@ pmap_kenter(vm_offset_t sva, vm_size_t size, vm_paddr_t pa, int mode)
pd_entry_t *pde;
pt_entry_t attr, old_l3e, *pte;
vm_offset_t va;
- int lvl;
+ vm_page_t mpte;
+ int error, lvl;
KASSERT((pa & L3_OFFSET) == 0,
("pmap_kenter: Invalid physical address"));
@@ -2026,7 +2070,7 @@ pmap_kenter(vm_offset_t sva, vm_size_t size, vm_paddr_t pa, int mode)
("pmap_kenter: Mapping is not page-sized"));
attr = ATTR_DEFAULT | ATTR_S1_AP(ATTR_S1_AP_RW) | ATTR_S1_XN |
- ATTR_KERN_GP | ATTR_S1_IDX(mode) | L3_PAGE;
+ ATTR_KERN_GP | ATTR_S1_IDX(mode);
old_l3e = 0;
va = sva;
while (size != 0) {
@@ -2035,8 +2079,55 @@ pmap_kenter(vm_offset_t sva, vm_size_t size, vm_paddr_t pa, int mode)
("pmap_kenter: Invalid page entry, va: 0x%lx", va));
KASSERT(lvl == 2, ("pmap_kenter: Invalid level %d", lvl));
+ /*
+ * If we have an aligned, contiguous chunk of L2_SIZE, try
+ * to create an L2_BLOCK mapping.
+ */
+ if ((va & L2_OFFSET) == 0 && size >= L2_SIZE &&
+ (pa & L2_OFFSET) == 0 && vm_initialized) {
+ mpte = PHYS_TO_VM_PAGE(PTE_TO_PHYS(pmap_load(pde)));
+ KASSERT(pmap_every_pte_zero(VM_PAGE_TO_PHYS(mpte)),
+ ("pmap_kenter: Unexpected mapping"));
+ PMAP_LOCK(kernel_pmap);
+ error = pmap_insert_pt_page(kernel_pmap, mpte, false,
+ false);
+ if (error == 0) {
+ attr &= ~ATTR_CONTIGUOUS;
+
+ /*
+ * Although the page table page "mpte" should
+ * be devoid of mappings, the TLB might hold
+ * intermediate entries that reference it, so
+ * we perform a single-page invalidation.
+ */
+ pmap_update_entry(kernel_pmap, pde,
+ PHYS_TO_PTE(pa) | attr | L2_BLOCK, va,
+ PAGE_SIZE);
+ }
+ PMAP_UNLOCK(kernel_pmap);
+ if (error == 0) {
+ va += L2_SIZE;
+ pa += L2_SIZE;
+ size -= L2_SIZE;
+ continue;
+ }
+ }
+
+ /*
+ * If we have an aligned, contiguous chunk of L3C_ENTRIES
+ * L3 pages, set the contiguous bit within each PTE so that
+ * the chunk can be cached using only one TLB entry.
+ */
+ if ((va & L3C_OFFSET) == 0 && (pa & L3C_OFFSET) == 0) {
+ if (size >= L3C_SIZE)
+ attr |= ATTR_CONTIGUOUS;
+ else
+ attr &= ~ATTR_CONTIGUOUS;
+ }
+
pte = pmap_l2_to_l3(pde, va);
- old_l3e |= pmap_load_store(pte, PHYS_TO_PTE(pa) | attr);
+ old_l3e |= pmap_load_store(pte, PHYS_TO_PTE(pa) | attr |
+ L3_PAGE);
va += PAGE_SIZE;
pa += PAGE_SIZE;
@@ -2069,6 +2160,8 @@ pmap_kremove(vm_offset_t va)
pt_entry_t *pte;
pte = pmap_pte_exists(kernel_pmap, va, 3, __func__);
+ KASSERT((pmap_load(pte) & ATTR_CONTIGUOUS) == 0,
+ ("pmap_kremove: unexpected ATTR_CONTIGUOUS"));
pmap_clear(pte);
pmap_s1_invalidate_page(kernel_pmap, va, true);
}
@@ -2083,8 +2176,9 @@ pmap_kremove(vm_offset_t va)
void
pmap_kremove_device(vm_offset_t sva, vm_size_t size)
{
- pt_entry_t *pte;
+ pt_entry_t *ptep, *ptep_end;
vm_offset_t va;
+ int lvl;
KASSERT((sva & L3_OFFSET) == 0,
("pmap_kremove_device: Invalid virtual address"));
@@ -2093,13 +2187,55 @@ pmap_kremove_device(vm_offset_t sva, vm_size_t size)
va = sva;
while (size != 0) {
- pte = pmap_pte_exists(kernel_pmap, va, 3, __func__);
- pmap_clear(pte);
+ ptep = pmap_pte(kernel_pmap, va, &lvl);
+ KASSERT(ptep != NULL, ("Invalid page table, va: 0x%lx", va));
+ switch (lvl) {
+ case 2:
+ KASSERT((va & L2_OFFSET) == 0,
+ ("Unaligned virtual address"));
+ KASSERT(size >= L2_SIZE, ("Insufficient size"));
- va += PAGE_SIZE;
- size -= PAGE_SIZE;
+ if (va != sva) {
+ pmap_s1_invalidate_range(kernel_pmap, sva, va,
+ true);
+ }
+ pmap_clear(ptep);
+ pmap_s1_invalidate_page(kernel_pmap, va, true);
+ PMAP_LOCK(kernel_pmap);
+ pmap_remove_kernel_l2(kernel_pmap, ptep, va);
+ PMAP_UNLOCK(kernel_pmap);
+
+ va += L2_SIZE;
+ sva = va;
+ size -= L2_SIZE;
+ break;
+ case 3:
+ if ((pmap_load(ptep) & ATTR_CONTIGUOUS) != 0) {
+ KASSERT((va & L3C_OFFSET) == 0,
+ ("Unaligned L3C virtual address"));
+ KASSERT(size >= L3C_SIZE,
+ ("Insufficient L3C size"));
+
+ ptep_end = ptep + L3C_ENTRIES;
+ for (; ptep < ptep_end; ptep++)
+ pmap_clear(ptep);
+
+ va += L3C_SIZE;
+ size -= L3C_SIZE;
+ break;
+ }
+ pmap_clear(ptep);
+
+ va += PAGE_SIZE;
+ size -= PAGE_SIZE;
+ break;
+ default:
+ __assert_unreachable();
+ break;
+ }
}
- pmap_s1_invalidate_range(kernel_pmap, sva, va, true);
+ if (va != sva)
+ pmap_s1_invalidate_range(kernel_pmap, sva, va, true);
}
/*
@@ -2961,6 +3097,8 @@ reclaim_pv_chunk_domain(pmap_t locked_pmap, struct rwlock **lockp, int domain)
tpte = pmap_load(pte);
if ((tpte & ATTR_SW_WIRED) != 0)
continue;
+ if ((tpte & ATTR_CONTIGUOUS) != 0)
+ (void)pmap_demote_l3c(pmap, pte, va);
tpte = pmap_load_clear(pte);
m = PHYS_TO_VM_PAGE(PTE_TO_PHYS(tpte));
if (pmap_pte_dirty(pmap, tpte))
@@ -3443,6 +3581,44 @@ pmap_pv_insert_l2(pmap_t pmap, vm_offset_t va, pd_entry_t l2e, u_int flags,
return (true);
}
+/*
+ * Conditionally creates the PV entries for a L3C superpage mapping if
+ * the required memory can be allocated without resorting to reclamation.
+ */
+static bool
+pmap_pv_insert_l3c(pmap_t pmap, vm_offset_t va, vm_page_t m,
+ struct rwlock **lockp)
+{
+ pv_entry_t pv;
+ vm_offset_t tva;
+ vm_paddr_t pa __diagused;
+ vm_page_t mt;
+
+ PMAP_LOCK_ASSERT(pmap, MA_OWNED);
+ KASSERT((va & L3C_OFFSET) == 0,
+ ("pmap_pv_insert_l3c: va is not aligned"));
+ pa = VM_PAGE_TO_PHYS(m);
+ KASSERT((pa & L3C_OFFSET) == 0,
+ ("pmap_pv_insert_l3c: pa is not aligned"));
+ CHANGE_PV_LIST_LOCK_TO_VM_PAGE(lockp, m);
+ for (mt = m, tva = va; mt < &m[L3C_ENTRIES]; mt++, tva += L3_SIZE) {
+ /* Pass NULL instead of lockp to disable reclamation. */
+ pv = get_pv_entry(pmap, NULL);
+ if (__predict_false(pv == NULL)) {
+ while (tva > va) {
+ mt--;
+ tva -= L3_SIZE;
+ pmap_pvh_free(&mt->md, pmap, tva);
+ }
+ return (false);
+ }
+ pv->pv_va = tva;
+ TAILQ_INSERT_TAIL(&mt->md.pv_list, pv, pv_next);
+ mt->md.pv_gen++;
+ }
+ return (true);
+}
+
static void
pmap_remove_kernel_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t va)
{
@@ -3547,6 +3723,9 @@ pmap_remove_l3(pmap_t pmap, pt_entry_t *l3, vm_offset_t va,
vm_page_t m;
PMAP_LOCK_ASSERT(pmap, MA_OWNED);
+ old_l3 = pmap_load(l3);
+ if ((old_l3 & ATTR_CONTIGUOUS) != 0)
+ (void)pmap_demote_l3c(pmap, l3, va);
old_l3 = pmap_load_clear(l3);
pmap_s1_invalidate_page(pmap, va, true);
if (old_l3 & ATTR_SW_WIRED)
@@ -3570,6 +3749,95 @@ pmap_remove_l3(pmap_t pmap, pt_entry_t *l3, vm_offset_t va,
return (pmap_unuse_pt(pmap, va, l2e, free));
}
+/*
+ * Removes the specified L3C superpage mapping. Requests TLB invalidations
+ * to be performed by the caller through the returned "*vap". Returns true
+ * if the level 3 table "ml3" was unmapped and added to the spglist "free".
+ * Otherwise, returns false.
+ */
+static bool
+pmap_remove_l3c(pmap_t pmap, pt_entry_t *l3p, vm_offset_t va, vm_offset_t *vap,
+ vm_offset_t va_next, vm_page_t ml3, struct spglist *free,
+ struct rwlock **lockp)
+{
+ struct md_page *pvh;
+ struct rwlock *new_lock;
+ pt_entry_t first_l3e, l3e, *tl3p;
+ vm_offset_t tva;
+ vm_page_t m, mt;
+
+ PMAP_LOCK_ASSERT(pmap, MA_OWNED);
+ KASSERT(((uintptr_t)l3p & ((L3C_ENTRIES * sizeof(pt_entry_t)) - 1)) ==
+ 0, ("pmap_remove_l3c: l3p is not aligned"));
+ KASSERT((va & L3C_OFFSET) == 0,
+ ("pmap_remove_l3c: va is not aligned"));
+
+ /*
+ * Hardware accessed and dirty bit maintenance might only update a
+ * single L3 entry, so we must combine the accessed and dirty bits
+ * from this entire set of contiguous L3 entries.
+ */
+ first_l3e = pmap_load_clear(l3p);
+ for (tl3p = l3p + 1; tl3p < &l3p[L3C_ENTRIES]; tl3p++) {
+ l3e = pmap_load_clear(tl3p);
+ KASSERT((l3e & ATTR_CONTIGUOUS) != 0,
+ ("pmap_remove_l3c: l3e is missing ATTR_CONTIGUOUS"));
+ if ((l3e & (ATTR_SW_DBM | ATTR_S1_AP_RW_BIT)) ==
+ (ATTR_SW_DBM | ATTR_S1_AP(ATTR_S1_AP_RW)))
+ first_l3e &= ~ATTR_S1_AP_RW_BIT;
+ first_l3e |= l3e & ATTR_AF;
+ }
+ if ((first_l3e & ATTR_SW_WIRED) != 0)
+ pmap->pm_stats.wired_count -= L3C_ENTRIES;
+ pmap_resident_count_dec(pmap, L3C_ENTRIES);
+ if ((first_l3e & ATTR_SW_MANAGED) != 0) {
+ m = PHYS_TO_VM_PAGE(PTE_TO_PHYS(first_l3e));
+ new_lock = VM_PAGE_TO_PV_LIST_LOCK(m);
+ if (new_lock != *lockp) {
+ if (*lockp != NULL) {
+ /*
+ * Pending TLB invalidations must be
+ * performed before the PV list lock is
+ * released. Otherwise, a concurrent
+ * pmap_remove_all() on a physical page
+ * could return while a stale TLB entry
+ * still provides access to that page.
+ */
+ if (*vap != va_next) {
+ pmap_invalidate_range(pmap, *vap, va,
+ true);
+ *vap = va_next;
+ }
+ rw_wunlock(*lockp);
+ }
+ *lockp = new_lock;
+ rw_wlock(*lockp);
+ }
+ pvh = page_to_pvh(m);
+ for (mt = m, tva = va; mt < &m[L3C_ENTRIES]; mt++, tva +=
+ L3_SIZE) {
+ if (pmap_pte_dirty(pmap, first_l3e))
+ vm_page_dirty(mt);
+ if ((first_l3e & ATTR_AF) != 0)
+ vm_page_aflag_set(mt, PGA_REFERENCED);
+ pmap_pvh_free(&mt->md, pmap, tva);
+ if (TAILQ_EMPTY(&mt->md.pv_list) &&
+ TAILQ_EMPTY(&pvh->pv_list))
+ vm_page_aflag_clear(mt, PGA_WRITEABLE);
+ }
+ }
+ if (*vap == va_next)
+ *vap = va;
+ if (ml3 != NULL) {
+ ml3->ref_count -= L3C_ENTRIES;
+ if (ml3->ref_count == 0) {
+ _pmap_unwire_l3(pmap, va, ml3, free);
+ return (true);
+ }
+ }
+ return (false);
+}
+
/*
* Remove the specified range of addresses from the L3 page table that is
* identified by the given L2 entry.
@@ -3595,13 +3863,35 @@ pmap_remove_l3_range(pmap_t pmap, pd_entry_t l2e, vm_offset_t sva,
l3pg = !ADDR_IS_KERNEL(sva) ? PHYS_TO_VM_PAGE(PTE_TO_PHYS(l2e)) : NULL;
va = eva;
for (l3 = pmap_l2_to_l3(&l2e, sva); sva != eva; l3++, sva += L3_SIZE) {
- if (!pmap_l3_valid(pmap_load(l3))) {
+ old_l3 = pmap_load(l3);
+ if (!pmap_l3_valid(old_l3)) {
if (va != eva) {
pmap_invalidate_range(pmap, va, sva, true);
va = eva;
}
continue;
}
+ if ((old_l3 & ATTR_CONTIGUOUS) != 0) {
+ /*
+ * Is this entire set of contiguous L3 entries being
+ * removed? Handle the possibility that "eva" is zero
+ * because of address wraparound.
+ */
+ if ((sva & L3C_OFFSET) == 0 &&
+ sva + L3C_OFFSET <= eva - 1) {
+ if (pmap_remove_l3c(pmap, l3, sva, &va, eva,
+ l3pg, free, lockp)) {
+ /* The L3 table was unmapped. */
+ sva += L3C_SIZE;
+ break;
+ }
+ l3 += L3C_ENTRIES - 1;
+ sva += L3C_SIZE - L3_SIZE;
+ continue;
+ }
+
+ (void)pmap_demote_l3c(pmap, l3, sva);
+ }
old_l3 = pmap_load_clear(l3);
if ((old_l3 & ATTR_SW_WIRED) != 0)
pmap->pm_stats.wired_count--;
@@ -3857,6 +4147,9 @@ retry:
tpde = pmap_load(pde);
pte = pmap_l2_to_l3(pde, pv->pv_va);
+ tpte = pmap_load(pte);
+ if ((tpte & ATTR_CONTIGUOUS) != 0)
+ (void)pmap_demote_l3c(pmap, pte, pv->pv_va);
tpte = pmap_load_clear(pte);
if (tpte & ATTR_SW_WIRED)
pmap->pm_stats.wired_count--;
@@ -3929,6 +4222,54 @@ pmap_protect_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t sva, pt_entry_t mask,
pmap_s1_invalidate_page(pmap, sva, true);
}
+/*
+ * Masks and sets bits in the specified L3C superpage mapping.
+ *
+ * Requests TLB invalidations to be performed by the caller through the
+ * returned "*vap".
+ */
+static void
+pmap_mask_set_l3c(pmap_t pmap, pt_entry_t *l3p, vm_offset_t va,
+ vm_offset_t *vap, vm_offset_t va_next, pt_entry_t mask, pt_entry_t nbits)
+{
+ pt_entry_t l3e, *tl3p;
+ vm_page_t m, mt;
+ bool dirty;
+
+ PMAP_LOCK_ASSERT(pmap, MA_OWNED);
+ KASSERT(((uintptr_t)l3p & ((L3C_ENTRIES * sizeof(pt_entry_t)) - 1)) ==
+ 0, ("pmap_mask_set_l3c: l3p is not aligned"));
+ KASSERT((va & L3C_OFFSET) == 0,
+ ("pmap_mask_set_l3c: va is not aligned"));
+ dirty = false;
+ for (tl3p = l3p; tl3p < &l3p[L3C_ENTRIES]; tl3p++) {
+ l3e = pmap_load(tl3p);
+ KASSERT((l3e & ATTR_CONTIGUOUS) != 0,
+ ("pmap_mask_set_l3c: l3e is missing ATTR_CONTIGUOUS"));
+ while (!atomic_fcmpset_64(tl3p, &l3e, (l3e & ~mask) | nbits))
+ cpu_spinwait();
+ if ((l3e & (ATTR_SW_DBM | ATTR_S1_AP_RW_BIT)) ==
+ (ATTR_SW_DBM | ATTR_S1_AP(ATTR_S1_AP_RW)))
+ dirty = true;
+ }
+
+ /*
+ * When a dirty read/write superpage mapping is write protected,
+ * update the dirty field of each of the superpage's constituent 4KB
+ * pages.
+ */
+ if ((l3e & ATTR_SW_MANAGED) != 0 &&
+ (nbits & ATTR_S1_AP(ATTR_S1_AP_RO)) != 0 &&
+ dirty) {
+ m = PHYS_TO_VM_PAGE(PTE_TO_PHYS(pmap_load(l3p)));
+ for (mt = m; mt < &m[L3C_ENTRIES]; mt++)
+ vm_page_dirty(mt);
+ }
+
+ if (*vap == va_next)
+ *vap = va;
+}
+
/*
* Masks and sets bits in last level page table entries in the specified
* pmap and range
@@ -4013,9 +4354,36 @@ pmap_mask_set_locked(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, pt_entry_t m
va, sva, true);
va = va_next;
}
+ if ((l3 & ATTR_CONTIGUOUS) != 0) {
+ l3p += L3C_ENTRIES - 1;
+ sva += L3C_SIZE - L3_SIZE;
+ }
continue;
}
+ if ((l3 & ATTR_CONTIGUOUS) != 0) {
+ /*
+ * Is this entire set of contiguous L3 entries
+ * being protected? Handle the possibility
+ * that "va_next" is zero because of address
+ * wraparound.
+ */
+ if ((sva & L3C_OFFSET) == 0 &&
+ sva + L3C_OFFSET <= va_next - 1) {
+ pmap_mask_set_l3c(pmap, l3p, sva, &va,
+ va_next, mask, nbits);
+ l3p += L3C_ENTRIES - 1;
+ sva += L3C_SIZE - L3_SIZE;
+ continue;
+ }
+
+ (void)pmap_demote_l3c(pmap, l3p, sva);
+
+ /*
+ * The L3 entry's accessed bit may have changed.
+ */
+ l3 = pmap_load(l3p);
+ }
while (!atomic_fcmpset_64(l3p, &l3, (l3 & ~mask) |
nbits))
cpu_spinwait();
@@ -4141,9 +4509,10 @@ pmap_remove_pt_page(pmap_t pmap, vm_offset_t va)
* inconsistent state.
*/
static void
-pmap_update_entry(pmap_t pmap, pd_entry_t *pte, pd_entry_t newpte,
+pmap_update_entry(pmap_t pmap, pd_entry_t *ptep, pd_entry_t newpte,
vm_offset_t va, vm_size_t size)
{
+ pd_entry_t *lip, *ptep_end;
register_t intr;
PMAP_LOCK_ASSERT(pmap, MA_OWNED);
@@ -4151,6 +4520,11 @@ pmap_update_entry(pmap_t pmap, pd_entry_t *pte, pd_entry_t newpte,
if ((newpte & ATTR_SW_NO_PROMOTE) != 0)
panic("%s: Updating non-promote pte", __func__);
+ if (size == L3C_SIZE)
+ ptep_end = ptep + L3C_ENTRIES;
+ else
+ ptep_end = ptep + 1;
+
/*
* Ensure we don't get switched out with the page table in an
* inconsistent state. We also need to ensure no interrupts fire
@@ -4163,7 +4537,8 @@ pmap_update_entry(pmap_t pmap, pd_entry_t *pte, pd_entry_t newpte,
* unchanged, so that a lockless, concurrent pmap_kextract() can still
* lookup the physical address.
*/
- pmap_clear_bits(pte, ATTR_DESCR_VALID);
+ for (lip = ptep; lip < ptep_end; lip++)
+ pmap_clear_bits(lip, ATTR_DESCR_VALID);
/*
* When promoting, the L{1,2}_TABLE entry that is being replaced might
@@ -4173,7 +4548,10 @@ pmap_update_entry(pmap_t pmap, pd_entry_t *pte, pd_entry_t newpte,
pmap_s1_invalidate_range(pmap, va, va + size, false);
/* Create the new mapping */
- pmap_store(pte, newpte);
+ for (lip = ptep; lip < ptep_end; lip++) {
+ pmap_store(lip, newpte);
+ newpte += PAGE_SIZE;
+ }
dsb(ishst);
intr_restore(intr);
@@ -4321,8 +4699,7 @@ setl3:
goto setl3;
oldl3 &= ~ATTR_SW_DBM;
}
- if ((oldl3 & (ATTR_MASK & ~ATTR_AF)) != (newl2 & (ATTR_MASK &
- ~ATTR_AF))) {
+ if ((oldl3 & ATTR_PROMOTE) != (newl2 & ATTR_PROMOTE)) {
atomic_add_long(&pmap_l2_p_failures, 1);
CTR2(KTR_PMAP, "pmap_promote_l2: failure for va %#lx"
" in pmap %p", va, pmap);
@@ -4338,7 +4715,7 @@ setl3:
* such as pmap_enter_quick(), don't automatically mark the
* underlying pages as referenced.
*/
- newl2 &= ~ATTR_AF | all_l3e_AF;
+ newl2 &= ~(ATTR_CONTIGUOUS | ATTR_AF | ATTR_DESCR_MASK) | all_l3e_AF;
/*
* Save the page table page in its current state until the L2
@@ -4363,10 +4740,7 @@ setl3:
if ((newl2 & ATTR_SW_MANAGED) != 0)
pmap_pv_promote_l2(pmap, va, PTE_TO_PHYS(newl2), lockp);
- newl2 &= ~ATTR_DESCR_MASK;
- newl2 |= L2_BLOCK;
-
- pmap_update_entry(pmap, l2, newl2, va & ~L2_OFFSET, L2_SIZE);
+ pmap_update_entry(pmap, l2, newl2 | L2_BLOCK, va & ~L2_OFFSET, L2_SIZE);
atomic_add_long(&pmap_l2_promotions, 1);
CTR2(KTR_PMAP, "pmap_promote_l2: success for va %#lx in pmap %p", va,
@@ -4666,6 +5040,8 @@ havel3:
* The physical page has changed. Temporarily invalidate
* the mapping.
*/
+ if ((orig_l3 & ATTR_CONTIGUOUS) != 0)
+ (void)pmap_demote_l3c(pmap, l3, va);
orig_l3 = pmap_load_clear(l3);
KASSERT(PTE_TO_PHYS(orig_l3) == opa,
("pmap_enter: unexpected pa update for %#lx", va));
@@ -4752,6 +5128,8 @@ validate:
KASSERT(opa == pa, ("pmap_enter: invalid update"));
if ((orig_l3 & ~ATTR_AF) != (new_l3 & ~ATTR_AF)) {
/* same PA, different attributes */
+ if ((orig_l3 & ATTR_CONTIGUOUS) != 0)
+ (void)pmap_demote_l3c(pmap, l3, va);
orig_l3 = pmap_load_store(l3, new_l3);
pmap_invalidate_page(pmap, va, true);
if ((orig_l3 & ATTR_SW_MANAGED) != 0 &&
@@ -5278,6 +5656,7 @@ pmap_unwire(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
vm_offset_t va_next;
pd_entry_t *l0, *l1, *l2;
pt_entry_t *l3;
+ bool partial_l3c;
PMAP_LOCK(pmap);
for (; sva < eva; sva = va_next) {
@@ -5340,10 +5719,26 @@ pmap_unwire(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
if (va_next > eva)
va_next = eva;
- for (l3 = pmap_l2_to_l3(l2, sva); sva != va_next; l3++,
- sva += L3_SIZE) {
+ for (partial_l3c = true, l3 = pmap_l2_to_l3(l2, sva);
+ sva != va_next; l3++, sva += L3_SIZE) {
if (pmap_load(l3) == 0)
continue;
+ if ((pmap_load(l3) & ATTR_CONTIGUOUS) != 0) {
+ /*
+ * Avoid demotion for whole-page unwiring.
+ */
+ if ((sva & L3C_OFFSET) == 0) {
+ /*
+ * Handle the possibility that
+ * "va_next" is zero because of
+ * address wraparound.
+ */
+ partial_l3c = sva + L3C_OFFSET >
+ va_next - 1;
+ }
+ if (partial_l3c)
+ (void)pmap_demote_l3c(pmap, l3, sva);
+ }
if ((pmap_load(l3) & ATTR_SW_WIRED) == 0)
panic("pmap_unwire: l3 %#jx is missing "
"ATTR_SW_WIRED", (uintmax_t)pmap_load(l3));
@@ -5360,6 +5755,58 @@ pmap_unwire(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
PMAP_UNLOCK(pmap);
}
+/*
+ * This function requires that the caller has already added one to ml3's
+ * ref_count in anticipation of creating a 4KB page mapping.
+ */
+static bool
+pmap_copy_l3c(pmap_t pmap, pt_entry_t *l3p, vm_offset_t va, pt_entry_t l3e,
+ vm_page_t ml3, struct rwlock **lockp)
+{
+ pt_entry_t *tl3p;
+
+ PMAP_LOCK_ASSERT(pmap, MA_OWNED);
+ KASSERT((va & L3C_OFFSET) == 0,
+ ("pmap_copy_l3c: va is not aligned"));
+ KASSERT((l3e & ATTR_SW_MANAGED) != 0,
+ ("pmap_copy_l3c: l3e is not managed"));
+
+ /*
+ * Abort if a mapping already exists.
+ */
+ for (tl3p = l3p; tl3p < &l3p[L3C_ENTRIES]; tl3p++)
+ if (pmap_load(tl3p) != 0) {
+ if (ml3 != NULL)
+ ml3->ref_count--;
+ return (false);
+ }
+
+ if (!pmap_pv_insert_l3c(pmap, va, PHYS_TO_VM_PAGE(PTE_TO_PHYS(l3e)),
+ lockp)) {
+ if (ml3 != NULL)
+ pmap_abort_ptp(pmap, va, ml3);
+ return (false);
+ }
+ ml3->ref_count += L3C_ENTRIES - 1;
+
+ /*
+ * Clear the wired and accessed bits. However, leave the dirty bit
+ * unchanged because read/write superpage mappings are required to be
+ * dirty.
+ */
+ l3e &= ~(ATTR_SW_WIRED | ATTR_AF);
+
+ for (tl3p = l3p; tl3p < &l3p[L3C_ENTRIES]; tl3p++) {
+ pmap_store(tl3p, l3e);
+ l3e += L3_SIZE;
+ }
+ pmap_resident_count_inc(pmap, L3C_ENTRIES);
+ atomic_add_long(&pmap_l3c_mappings, 1);
+ CTR2(KTR_PMAP, "pmap_copy_l3c: success for va %#lx in pmap %p",
+ va, pmap);
+ return (true);
+}
+
/*
* Copy the range specified by src_addr/len
* from the source map to the range dst_addr/len
@@ -5507,14 +5954,25 @@ pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr, vm_size_t len,
dst_pte = (pt_entry_t *)
PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dstmpte));
dst_pte = &dst_pte[pmap_l3_index(addr)];
- if (pmap_load(dst_pte) == 0 &&
+ if ((ptetemp & ATTR_CONTIGUOUS) != 0 && (addr &
+ L3C_OFFSET) == 0 && addr + L3C_OFFSET <=
+ va_next - 1) {
+ if (!pmap_copy_l3c(dst_pmap, dst_pte, addr,
+ ptetemp, dstmpte, &lock))
+ goto out;
+ addr += L3C_SIZE - PAGE_SIZE;
+ src_pte += L3C_ENTRIES - 1;
+ } else if (pmap_load(dst_pte) == 0 &&
pmap_try_insert_pv_entry(dst_pmap, addr,
PHYS_TO_VM_PAGE(PTE_TO_PHYS(ptetemp)), &lock)) {
/*
- * Clear the wired, modified, and accessed
- * (referenced) bits during the copy.
+ * Clear the wired, contiguous, modified, and
+ * accessed bits from the destination PTE.
+ * The contiguous bit is cleared because we
+ * are not copying the entire L3C superpage.
*/
- mask = ATTR_AF | ATTR_SW_WIRED;
+ mask = ATTR_SW_WIRED | ATTR_CONTIGUOUS |
+ ATTR_AF;
nbits = 0;
if ((ptetemp & ATTR_SW_DBM) != 0)
nbits |= ATTR_S1_AP_RW_BIT;
@@ -5879,9 +6337,13 @@ pmap_remove_pages(pmap_t pmap)
lvl);
}
-/*
- * We cannot remove wired pages from a process' mapping at this time
- */
+ /*
+ * We cannot remove wired mappings at this time.
+ *
+ * For L3C superpages, all of the constituent PTEs
+ * should have the wired bit set, so we don't
+ * check for ATTR_CONTIGUOUS here.
+ */
if (tpte & ATTR_SW_WIRED) {
allfree = 0;
continue;
@@ -5912,6 +6374,11 @@ pmap_remove_pages(pmap_t pmap)
/*
* Update the vm_page_t clean/reference bits.
+ *
+ * We don't check for ATTR_CONTIGUOUS here
+ * because writeable L3C superpages are expected
+ * to be dirty, i.e., every constituent PTE
+ * should be dirty.
*/
if (pmap_pte_dirty(pmap, tpte)) {
switch (lvl) {
@@ -6000,7 +6467,7 @@ pmap_page_test_mappings(vm_page_t m, bool accessed, bool modified)
struct rwlock *lock;
pv_entry_t pv;
struct md_page *pvh;
- pt_entry_t *pte, mask, value;
+ pt_entry_t l3e, mask, *pte, value;
pmap_t pmap;
int md_gen, pvh_gen;
bool rv;
@@ -6033,8 +6500,11 @@ restart:
mask |= ATTR_AF | ATTR_DESCR_MASK;
value |= ATTR_AF | L3_PAGE;
}
- rv = (pmap_load(pte) & mask) == value;
+ l3e = pmap_load(pte);
+ if ((l3e & ATTR_CONTIGUOUS) != 0)
+ l3e = pmap_load_l3c(pte);
PMAP_UNLOCK(pmap);
+ rv = (l3e & mask) == value;
if (rv)
goto out;
}
@@ -6205,6 +6675,15 @@ retry:
pte = pmap_pte_exists(pmap, pv->pv_va, 3, __func__);
oldpte = pmap_load(pte);
if ((oldpte & ATTR_SW_DBM) != 0) {
+ if ((oldpte & ATTR_CONTIGUOUS) != 0) {
+ (void)pmap_demote_l3c(pmap, pte, pv->pv_va);
+
+ /*
+ * The L3 entry's accessed bit may have
+ * changed.
+ */
+ oldpte = pmap_load(pte);
+ }
if (pmap->pm_stage == PM_STAGE1) {
set = ATTR_S1_AP_RW_BIT;
clear = 0;
@@ -6362,11 +6841,23 @@ small_mappings:
vm_page_dirty(m);
if ((tpte & ATTR_AF) != 0) {
if ((tpte & ATTR_SW_WIRED) == 0) {
+ /*
+ * Clear the accessed bit in this L3 entry
+ * regardless of the contiguous bit.
+ */
pmap_clear_bits(pte, ATTR_AF);
pmap_invalidate_page(pmap, pv->pv_va, true);
cleared++;
} else
not_cleared++;
+ } else if ((tpte & ATTR_CONTIGUOUS) != 0 &&
+ (pmap_load_l3c(pte) & ATTR_AF) != 0) {
+ /*
+ * An L3C superpage mapping is regarded as accessed
+ * until the accessed bit has been cleared in all
+ * of its constituent entries.
+ */
+ not_cleared++;
}
PMAP_UNLOCK(pmap);
/* Rotate the PV list if it has more than one entry. */
@@ -6392,10 +6883,10 @@ void
pmap_advise(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, int advice)
{
struct rwlock *lock;
- vm_offset_t va, va_next;
+ vm_offset_t va, va_next, dva;
vm_page_t m;
pd_entry_t *l0, *l1, *l2, oldl2;
- pt_entry_t *l3, oldl3;
+ pt_entry_t *l3, *dl3, oldl3;
PMAP_ASSERT_STAGE1(pmap);
@@ -6490,13 +6981,59 @@ pmap_advise(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, int advice)
m = PHYS_TO_VM_PAGE(PTE_TO_PHYS(oldl3));
vm_page_dirty(m);
}
- while (!atomic_fcmpset_long(l3, &oldl3,
- (oldl3 & ~ATTR_AF) |
- ATTR_S1_AP(ATTR_S1_AP_RO)))
- cpu_spinwait();
- } else if ((oldl3 & ATTR_AF) != 0)
+ if ((oldl3 & ATTR_CONTIGUOUS) != 0) {
+ /*
+ * Unconditionally demote the L3C
+ * superpage because we do not allow
+ * writeable, clean superpages.
+ */
+ (void)pmap_demote_l3c(pmap, l3, sva);
+
+ /*
+ * Destroy the final mapping before the
+ * next L3C boundary or va_next,
+ * whichever comes first, so that a
+ * subsequent access may act as a
+ * repromotion trigger.
+ */
+ if ((oldl3 & ATTR_SW_WIRED) == 0) {
+ dva = MIN((sva & ~L3C_OFFSET) +
+ L3C_SIZE - PAGE_SIZE,
+ va_next - PAGE_SIZE);
+ dl3 = pmap_l2_to_l3(l2, dva);
+ KASSERT(pmap_load(dl3) != 0,
+ ("pmap_advise: invalid PTE"));
+ lock = NULL;
+ pmap_remove_l3(pmap, dl3, dva,
+ pmap_load(l2), NULL, &lock);
+ if (lock != NULL)
+ rw_wunlock(lock);
+ }
+
+ /*
+ * The L3 entry's accessed bit may have
+ * changed.
+ */
+ oldl3 = pmap_load(l3);
+ }
+
+ /*
+ * Check that we did not just destroy this entry so
+ * we avoid corrupting the page able.
+ */
*** 242 LINES SKIPPED ***