diff options
Diffstat (limited to 'sys/amd64')
| -rw-r--r-- | sys/amd64/amd64/mem.c | 4 | ||||
| -rw-r--r-- | sys/amd64/amd64/minidump_machdep.c | 10 | ||||
| -rw-r--r-- | sys/amd64/amd64/pmap.c | 233 | ||||
| -rw-r--r-- | sys/amd64/amd64/trap.c | 2 | ||||
| -rw-r--r-- | sys/amd64/include/param.h | 5 | ||||
| -rw-r--r-- | sys/amd64/include/pmap.h | 20 | ||||
| -rw-r--r-- | sys/amd64/include/vmparam.h | 41 |
7 files changed, 193 insertions, 122 deletions
diff --git a/sys/amd64/amd64/mem.c b/sys/amd64/amd64/mem.c index 413b7c74890e..851f2df0e6e1 100644 --- a/sys/amd64/amd64/mem.c +++ b/sys/amd64/amd64/mem.c @@ -105,8 +105,8 @@ memrw(struct cdev *dev, struct uio *uio, int flags) * PAGE_SIZE, the uiomove() call does not * access past the end of the direct map. */ - if (v >= DMAP_MIN_ADDRESS && - v < DMAP_MIN_ADDRESS + dmaplimit) { + if (v >= kva_layout.dmap_low && + v < kva_layout.dmap_high) { error = uiomove((void *)v, c, uio); break; } diff --git a/sys/amd64/amd64/minidump_machdep.c b/sys/amd64/amd64/minidump_machdep.c index 6d0917e16099..43bf81a991bf 100644 --- a/sys/amd64/amd64/minidump_machdep.c +++ b/sys/amd64/amd64/minidump_machdep.c @@ -186,7 +186,7 @@ cpu_minidumpsys(struct dumperinfo *di, const struct minidumpstate *state) * tables, so care must be taken to read each entry only once. */ pmapsize = 0; - for (va = VM_MIN_KERNEL_ADDRESS; va < kva_end; ) { + for (va = kva_layout.km_low; va < kva_end; ) { /* * We always write a page, even if it is zero. Each * page written corresponds to 1GB of space @@ -279,9 +279,9 @@ cpu_minidumpsys(struct dumperinfo *di, const struct minidumpstate *state) mdhdr.msgbufsize = mbp->msg_size; mdhdr.bitmapsize = round_page(BITSET_SIZE(vm_page_dump_pages)); mdhdr.pmapsize = pmapsize; - mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS; - mdhdr.dmapbase = DMAP_MIN_ADDRESS; - mdhdr.dmapend = DMAP_MAX_ADDRESS; + mdhdr.kernbase = kva_layout.km_low; + mdhdr.dmapbase = kva_layout.dmap_low; + mdhdr.dmapend = kva_layout.dmap_high; mdhdr.dumpavailsize = round_page(sizeof(dump_avail)); dump_init_header(di, &kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, @@ -323,7 +323,7 @@ cpu_minidumpsys(struct dumperinfo *di, const struct minidumpstate *state) /* Dump kernel page directory pages */ bzero(fakepd, sizeof(fakepd)); - for (va = VM_MIN_KERNEL_ADDRESS; va < kva_end; va += NBPDP) { + for (va = kva_layout.km_low; va < kva_end; va += NBPDP) { ii = pmap_pml4e_index(va); pml4 = (uint64_t *)PHYS_TO_DMAP(KPML4phys) + ii; pdp = (uint64_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME); diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c index fd02d1ad6690..2c7777e608b9 100644 --- a/sys/amd64/amd64/pmap.c +++ b/sys/amd64/amd64/pmap.c @@ -415,7 +415,7 @@ SYSCTL_INT(_machdep, OID_AUTO, nkpt, CTLFLAG_RD, &nkpt, 0, static int ndmpdp; vm_paddr_t dmaplimit; -vm_offset_t kernel_vm_end = VM_MIN_KERNEL_ADDRESS; +vm_offset_t kernel_vm_end = VM_MIN_KERNEL_ADDRESS_LA48; pt_entry_t pg_nx; static SYSCTL_NODE(_vm, OID_AUTO, pmap, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, @@ -475,11 +475,36 @@ _Static_assert(DMPML4I + NDMPML4E <= KMSANSHADPML4I, "direct map overflow"); static pml4_entry_t *kernel_pml4; static u_int64_t DMPDphys; /* phys addr of direct mapped level 2 */ static u_int64_t DMPDPphys; /* phys addr of direct mapped level 3 */ +static u_int64_t DMPML4phys; /* ... level 4, for la57 */ static int ndmpdpphys; /* number of DMPDPphys pages */ vm_paddr_t kernphys; /* phys addr of start of bootstrap data */ vm_paddr_t KERNend; /* and the end */ +struct kva_layout_s kva_layout = { + .kva_min = KV4ADDR(PML4PML4I, 0, 0, 0), + .dmap_low = KV4ADDR(DMPML4I, 0, 0, 0), + .dmap_high = KV4ADDR(DMPML4I + NDMPML4E, 0, 0, 0), + .lm_low = KV4ADDR(LMSPML4I, 0, 0, 0), + .lm_high = KV4ADDR(LMEPML4I + 1, 0, 0, 0), + .km_low = KV4ADDR(KPML4BASE, 0, 0, 0), + .km_high = KV4ADDR(KPML4BASE + NKPML4E - 1, NPDPEPG - 1, + NPDEPG - 1, NPTEPG - 1), + .rec_pt = KV4ADDR(PML4PML4I, 0, 0, 0), +}; + +struct kva_layout_s kva_layout_la57 = { + .kva_min = KV5ADDR(NPML5EPG / 2, 0, 0, 0, 0), /* == rec_pt */ + .dmap_low = KV5ADDR(DMPML5I, 0, 0, 0, 0), + .dmap_high = KV5ADDR(DMPML5I + NDMPML5E, 0, 0, 0, 0), + .lm_low = KV4ADDR(LMSPML4I, 0, 0, 0), + .lm_high = KV4ADDR(LMEPML4I + 1, 0, 0, 0), + .km_low = KV4ADDR(KPML4BASE, 0, 0, 0), + .km_high = KV4ADDR(KPML4BASE + NKPML4E - 1, NPDPEPG - 1, + NPDEPG - 1, NPTEPG - 1), + .rec_pt = KV5ADDR(PML5PML5I, 0, 0, 0, 0), +}; + /* * pmap_mapdev support pre initialization (i.e. console) */ @@ -549,8 +574,8 @@ static int pmap_flags = PMAP_PDE_SUPERPAGE; /* flags for x86 pmaps */ static vmem_t *large_vmem; static u_int lm_ents; -#define PMAP_ADDRESS_IN_LARGEMAP(va) ((va) >= LARGEMAP_MIN_ADDRESS && \ - (va) < LARGEMAP_MIN_ADDRESS + NBPML4 * (u_long)lm_ents) +#define PMAP_ADDRESS_IN_LARGEMAP(va) ((va) >= kva_layout.lm_low && \ + (va) < kva_layout.lm_high) int pmap_pcid_enabled = 1; SYSCTL_INT(_vm_pmap, OID_AUTO, pcid_enabled, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, @@ -1336,7 +1361,7 @@ static pdp_entry_t *pmap_pti_pdpe(vm_offset_t va); static pd_entry_t *pmap_pti_pde(vm_offset_t va); static void pmap_pti_wire_pte(void *pte); static int pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, - bool remove_pt, struct spglist *free, struct rwlock **lockp); + bool demote_kpde, struct spglist *free, struct rwlock **lockp); static int pmap_remove_pte(pmap_t pmap, pt_entry_t *ptq, vm_offset_t sva, pd_entry_t ptepde, struct spglist *free, struct rwlock **lockp); static vm_page_t pmap_remove_pt_page(pmap_t pmap, vm_offset_t va); @@ -1722,7 +1747,7 @@ create_pagetables(vm_paddr_t *firstaddr) { pd_entry_t *pd_p; pdp_entry_t *pdp_p; - pml4_entry_t *p4_p; + pml4_entry_t *p4_p, *p4d_p; pml5_entry_t *p5_p; uint64_t DMPDkernphys; vm_paddr_t pax; @@ -1732,7 +1757,7 @@ create_pagetables(vm_paddr_t *firstaddr) vm_offset_t kasankernbase; int kasankpdpi, kasankpdi, nkasanpte; #endif - int i, j, ndm1g, nkpdpe, nkdmpde; + int i, j, ndm1g, nkpdpe, nkdmpde, ndmpml4phys; TSENTER(); /* Allocate page table pages for the direct map */ @@ -1740,15 +1765,30 @@ create_pagetables(vm_paddr_t *firstaddr) if (ndmpdp < 4) /* Minimum 4GB of dirmap */ ndmpdp = 4; ndmpdpphys = howmany(ndmpdp, NPDPEPG); - if (ndmpdpphys > NDMPML4E) { - /* - * Each NDMPML4E allows 512 GB, so limit to that, - * and then readjust ndmpdp and ndmpdpphys. - */ - printf("NDMPML4E limits system to %d GB\n", NDMPML4E * 512); - Maxmem = atop(NDMPML4E * NBPML4); - ndmpdpphys = NDMPML4E; - ndmpdp = NDMPML4E * NPDEPG; + if (la57) { + ndmpml4phys = howmany(ndmpdpphys, NPML4EPG); + if (ndmpml4phys > NDMPML5E) { + printf("NDMPML5E limits system to %ld GB\n", + (u_long)NDMPML5E * NBPML5 / 1024 / 1024 / 1024); + Maxmem = atop(NDMPML5E * NBPML5); + ndmpml4phys = NDMPML5E; + ndmpdpphys = ndmpml4phys * NPML4EPG; + ndmpdp = ndmpdpphys * NPDEPG; + } + DMPML4phys = allocpages(firstaddr, ndmpml4phys); + } else { + if (ndmpdpphys > NDMPML4E) { + /* + * Each NDMPML4E allows 512 GB, so limit to + * that, and then readjust ndmpdp and + * ndmpdpphys. + */ + printf("NDMPML4E limits system to %d GB\n", + NDMPML4E * 512); + Maxmem = atop(NDMPML4E * NBPML4); + ndmpdpphys = NDMPML4E; + ndmpdp = NDMPML4E * NPDEPG; + } } DMPDPphys = allocpages(firstaddr, ndmpdpphys); ndm1g = 0; @@ -1773,7 +1813,13 @@ create_pagetables(vm_paddr_t *firstaddr) dmaplimit = (vm_paddr_t)ndmpdp << PDPSHIFT; /* Allocate pages. */ + if (la57) { + KPML5phys = allocpages(firstaddr, 1); + p5_p = (pml5_entry_t *)KPML5phys; + } KPML4phys = allocpages(firstaddr, 1); + p4_p = (pml4_entry_t *)KPML4phys; + KPDPphys = allocpages(firstaddr, NKPML4E); #ifdef KASAN KASANPDPphys = allocpages(firstaddr, NKASANPML4E); @@ -1893,6 +1939,16 @@ create_pagetables(vm_paddr_t *firstaddr) } /* + * Connect the Direct Map slots up to the PML4. + * pml5 entries for DMAP are handled below in global pml5 loop. + */ + p4d_p = la57 ? (pml4_entry_t *)DMPML4phys : &p4_p[DMPML4I]; + for (i = 0; i < ndmpdpphys; i++) { + p4d_p[i] = (DMPDPphys + ptoa(i)) | X86_PG_RW | X86_PG_V | + pg_nx; + } + + /* * Instead of using a 1G page for the memory containing the kernel, * use 2M pages with read-only and no-execute permissions. (If using 1G * pages, this will partially overwrite the PDPEs above.) @@ -1911,11 +1967,6 @@ create_pagetables(vm_paddr_t *firstaddr) } } - /* And recursively map PML4 to itself in order to get PTmap */ - p4_p = (pml4_entry_t *)KPML4phys; - p4_p[PML4PML4I] = KPML4phys; - p4_p[PML4PML4I] |= X86_PG_RW | X86_PG_V | pg_nx; - #ifdef KASAN /* Connect the KASAN shadow map slots up to the PML4. */ for (i = 0; i < NKASANPML4E; i++) { @@ -1938,25 +1989,15 @@ create_pagetables(vm_paddr_t *firstaddr) } #endif - /* Connect the Direct Map slots up to the PML4. */ - for (i = 0; i < ndmpdpphys; i++) { - p4_p[DMPML4I + i] = DMPDPphys + ptoa(i); - p4_p[DMPML4I + i] |= X86_PG_RW | X86_PG_V | pg_nx; - } - /* Connect the KVA slots up to the PML4 */ for (i = 0; i < NKPML4E; i++) { p4_p[KPML4BASE + i] = KPDPphys + ptoa(i); p4_p[KPML4BASE + i] |= X86_PG_RW | X86_PG_V; } - kernel_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(KPML4phys); - if (la57) { /* XXXKIB bootstrap KPML5phys page is lost */ - KPML5phys = allocpages(firstaddr, 1); - for (i = 0, p5_p = (pml5_entry_t *)KPML5phys; i < NPML5EPG; - i++) { + for (i = 0; i < NPML5EPG; i++) { if (i == PML5PML5I) { /* * Recursively map PML5 to itself in @@ -1964,6 +2005,10 @@ create_pagetables(vm_paddr_t *firstaddr) */ p5_p[i] = KPML5phys | X86_PG_RW | X86_PG_A | X86_PG_M | X86_PG_V | pg_nx; + } else if (i >= DMPML5I && i < DMPML5I + NDMPML5E) { + /* Connect DMAP pml4 pages to PML5. */ + p5_p[i] = (DMPML4phys + ptoa(i - DMPML5I)) | + X86_PG_RW | X86_PG_V | pg_nx; } else if (i == pmap_pml5e_index(UPT_MAX_ADDRESS)) { p5_p[i] = KPML4phys | X86_PG_RW | X86_PG_A | X86_PG_M | X86_PG_V; @@ -1971,6 +2016,10 @@ create_pagetables(vm_paddr_t *firstaddr) p5_p[i] = 0; } } + } else { + /* Recursively map PML4 to itself in order to get PTmap */ + p4_p[PML4PML4I] = KPML4phys; + p4_p[PML4PML4I] |= X86_PG_RW | X86_PG_V | pg_nx; } TSEXIT(); } @@ -2024,7 +2073,7 @@ pmap_bootstrap(vm_paddr_t *firstaddr) */ virtual_avail = (vm_offset_t)KERNSTART + round_2mpage(KERNend - (vm_paddr_t)kernphys); - virtual_end = VM_MAX_KERNEL_ADDRESS; + virtual_end = kva_layout.km_high; /* * Enable PG_G global pages, then switch to the kernel page @@ -2046,9 +2095,13 @@ pmap_bootstrap(vm_paddr_t *firstaddr) * Initialize the kernel pmap (which is statically allocated). * Count bootstrap data as being resident in case any of this data is * later unmapped (using pmap_remove()) and freed. + * + * DMAP_TO_PHYS()/PHYS_TO_DMAP() are functional only after + * kva_layout is fixed. */ PMAP_LOCK_INIT(kernel_pmap); if (la57) { + kva_layout = kva_layout_la57; vtoptem = ((1ul << (NPTEPGSHIFT + NPDEPGSHIFT + NPDPEPGSHIFT + NPML4EPGSHIFT + NPML5EPGSHIFT)) - 1) << 3; PTmap = (vm_offset_t)P5Tmap; @@ -2059,6 +2112,7 @@ pmap_bootstrap(vm_paddr_t *firstaddr) kernel_pmap->pm_cr3 = KPML5phys; pmap_pt_page_count_adj(kernel_pmap, 1); /* top-level page */ } else { + kernel_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(KPML4phys); kernel_pmap->pm_pmltop = kernel_pml4; kernel_pmap->pm_cr3 = KPML4phys; } @@ -2420,6 +2474,7 @@ pmap_init(void) { struct pmap_preinit_mapping *ppim; vm_page_t m, mpte; + pml4_entry_t *pml4e; int error, i, ret, skz63; /* L1TF, reserve page @0 unconditionally */ @@ -2559,18 +2614,19 @@ pmap_init(void) printf("pmap: large map %u PML4 slots (%lu GB)\n", lm_ents, (u_long)lm_ents * (NBPML4 / 1024 / 1024 / 1024)); if (lm_ents != 0) { - large_vmem = vmem_create("large", LARGEMAP_MIN_ADDRESS, - (vmem_size_t)lm_ents * NBPML4, PAGE_SIZE, 0, M_WAITOK); + large_vmem = vmem_create("large", kva_layout.lm_low, + (vmem_size_t)kva_layout.lm_high - kva_layout.lm_low, + PAGE_SIZE, 0, M_WAITOK); if (large_vmem == NULL) { printf("pmap: cannot create large map\n"); lm_ents = 0; } for (i = 0; i < lm_ents; i++) { m = pmap_large_map_getptp_unlocked(); - /* XXXKIB la57 */ - kernel_pml4[LMSPML4I + i] = X86_PG_V | - X86_PG_RW | X86_PG_A | X86_PG_M | pg_nx | - VM_PAGE_TO_PHYS(m); + pml4e = pmap_pml4e(kernel_pmap, kva_layout.lm_low + + (u_long)i * NBPML4); + *pml4e = X86_PG_V | X86_PG_RW | X86_PG_A | X86_PG_M | + pg_nx | VM_PAGE_TO_PHYS(m); } } } @@ -3899,7 +3955,7 @@ pmap_kextract(vm_offset_t va) pd_entry_t pde; vm_paddr_t pa; - if (va >= DMAP_MIN_ADDRESS && va < DMAP_MAX_ADDRESS) { + if (va >= kva_layout.dmap_low && va < kva_layout.dmap_high) { pa = DMAP_TO_PHYS(va); } else if (PMAP_ADDRESS_IN_LARGEMAP(va)) { pa = pmap_large_map_kextract(va); @@ -4040,7 +4096,7 @@ pmap_qremove(vm_offset_t sva, int count) * enough to one of those pmap_enter() calls for it to * be caught up in a promotion. */ - KASSERT(va >= VM_MIN_KERNEL_ADDRESS, ("usermode va %lx", va)); + KASSERT(va >= kva_layout.km_low, ("usermode va %lx", va)); KASSERT((*vtopde(va) & X86_PG_PS) == 0, ("pmap_qremove on promoted va %#lx", va)); @@ -4328,21 +4384,13 @@ void pmap_pinit_pml5(vm_page_t pml5pg) { pml5_entry_t *pm_pml5; + int i; pm_pml5 = (pml5_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pml5pg)); - - /* - * Add pml5 entry at top of KVA pointing to existing pml4 table, - * entering all existing kernel mappings into level 5 table. - */ - pm_pml5[pmap_pml5e_index(UPT_MAX_ADDRESS)] = KPML4phys | X86_PG_V | - X86_PG_RW | X86_PG_A | X86_PG_M; - - /* - * Install self-referential address mapping entry. - */ - pm_pml5[PML5PML5I] = VM_PAGE_TO_PHYS(pml5pg) | - X86_PG_RW | X86_PG_V | X86_PG_M | X86_PG_A; + for (i = 0; i < NPML5EPG / 2; i++) + pm_pml5[i] = 0; + for (; i < NPML5EPG; i++) + pm_pml5[i] = kernel_pmap->pm_pmltop[i]; } static void @@ -4899,8 +4947,8 @@ pmap_release(pmap_t pmap) m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)pmap->pm_pmltop)); if (pmap_is_la57(pmap)) { - pmap->pm_pmltop[pmap_pml5e_index(UPT_MAX_ADDRESS)] = 0; - pmap->pm_pmltop[PML5PML5I] = 0; + for (i = NPML5EPG / 2; i < NPML5EPG; i++) + pmap->pm_pmltop[i] = 0; } else { for (i = 0; i < NKPML4E; i++) /* KVA */ pmap->pm_pmltop[KPML4BASE + i] = 0; @@ -4942,7 +4990,7 @@ pmap_release(pmap_t pmap) static int kvm_size(SYSCTL_HANDLER_ARGS) { - unsigned long ksize = VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS; + unsigned long ksize = kva_layout.km_high - kva_layout.km_low; return sysctl_handle_long(oidp, &ksize, 0, req); } @@ -4953,7 +5001,7 @@ SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG | CTLFLAG_RD | CTLFLAG_MPSAFE, static int kvm_free(SYSCTL_HANDLER_ARGS) { - unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end; + unsigned long kfree = kva_layout.km_high - kernel_vm_end; return sysctl_handle_long(oidp, &kfree, 0, req); } @@ -5031,7 +5079,7 @@ pmap_page_array_startup(long pages) vm_page_array_size = pages; - start = VM_MIN_KERNEL_ADDRESS; + start = kva_layout.km_low; end = start + pages * sizeof(struct vm_page); for (va = start; va < end; va += NBPDR) { pfn = first_page + (va - start) / sizeof(struct vm_page); @@ -6067,8 +6115,8 @@ pmap_demote_pde_mpte(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, * so the direct map region is the only part of the * kernel address space that must be handled here. */ - KASSERT(!in_kernel || (va >= DMAP_MIN_ADDRESS && - va < DMAP_MAX_ADDRESS), + KASSERT(!in_kernel || (va >= kva_layout.dmap_low && + va < kva_layout.dmap_high), ("pmap_demote_pde: No saved mpte for va %#lx", va)); /* @@ -6165,8 +6213,7 @@ pmap_demote_pde_mpte(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, * pmap_remove_kernel_pde: Remove a kernel superpage mapping. */ static void -pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, - bool remove_pt) +pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va) { pd_entry_t newpde; vm_paddr_t mptepa; @@ -6174,12 +6221,8 @@ pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, KASSERT(pmap == kernel_pmap, ("pmap %p is not kernel_pmap", pmap)); PMAP_LOCK_ASSERT(pmap, MA_OWNED); - if (remove_pt) - mpte = pmap_remove_pt_page(pmap, va); - else - mpte = vm_radix_lookup(&pmap->pm_root, pmap_pde_pindex(va)); - if (mpte == NULL) - panic("pmap_remove_kernel_pde: Missing pt page."); + mpte = pmap_remove_pt_page(pmap, va); + KASSERT(mpte != NULL, ("pmap_remove_kernel_pde: missing pt page")); mptepa = VM_PAGE_TO_PHYS(mpte); newpde = mptepa | X86_PG_M | X86_PG_A | X86_PG_RW | X86_PG_V; @@ -6209,7 +6252,7 @@ pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, * pmap_remove_pde: do the things to unmap a superpage in a process */ static int -pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, bool remove_pt, +pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, bool demote_kpde, struct spglist *free, struct rwlock **lockp) { struct md_page *pvh; @@ -6249,9 +6292,7 @@ pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, bool remove_pt, pmap_delayed_invl_page(m); } } - if (pmap == kernel_pmap) { - pmap_remove_kernel_pde(pmap, pdq, sva, remove_pt); - } else { + if (pmap != kernel_pmap) { mpte = pmap_remove_pt_page(pmap, sva); if (mpte != NULL) { KASSERT(vm_page_any_valid(mpte), @@ -6262,6 +6303,14 @@ pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, bool remove_pt, mpte->ref_count = 0; pmap_add_delayed_free_list(mpte, free, false); } + } else if (demote_kpde) { + pmap_remove_kernel_pde(pmap, pdq, sva); + } else { + mpte = vm_radix_lookup(&pmap->pm_root, pmap_pde_pindex(sva)); + if (vm_page_any_valid(mpte)) { + mpte->valid = 0; + pmap_zero_page(mpte); + } } return (pmap_unuse_pt(pmap, sva, *pmap_pdpe(pmap, sva), free)); } @@ -7183,7 +7232,7 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, PG_RW = pmap_rw_bit(pmap); va = trunc_page(va); - KASSERT(va <= VM_MAX_KERNEL_ADDRESS, ("pmap_enter: toobig")); + KASSERT(va <= kva_layout.km_high, ("pmap_enter: toobig")); KASSERT(va < UPT_MIN_ADDRESS || va >= UPT_MAX_ADDRESS, ("pmap_enter: invalid to pmap_enter page table pages (va: 0x%lx)", va)); @@ -7573,8 +7622,8 @@ pmap_enter_pde(pmap_t pmap, vm_offset_t va, pd_entry_t newpde, u_int flags, * the mapping is not from kernel_pmap, then * a reserved PT page could be freed. */ - (void)pmap_remove_pde(pmap, pde, va, - pmap != kernel_pmap, &free, lockp); + (void)pmap_remove_pde(pmap, pde, va, false, &free, + lockp); if ((oldpde & PG_G) == 0) pmap_invalidate_pde_page(pmap, va, oldpde); } else { @@ -7584,10 +7633,9 @@ pmap_enter_pde(pmap_t pmap, vm_offset_t va, pd_entry_t newpde, u_int flags, * before any changes to mappings are * made. Abort on failure. */ - mt = PHYS_TO_VM_PAGE(*pde & PG_FRAME); - if (pmap_insert_pt_page(pmap, mt, false, false)) { - if (pdpg != NULL) - pdpg->ref_count--; + mt = PHYS_TO_VM_PAGE(oldpde & PG_FRAME); + if (pmap_insert_pt_page(pmap, mt, false, + false)) { CTR1(KTR_PMAP, "pmap_enter_pde: cannot ins kern ptp va %#lx", va); @@ -9550,7 +9598,7 @@ pmap_unmapdev(void *p, vm_size_t size) va = (vm_offset_t)p; /* If we gave a direct map region in pmap_mapdev, do nothing */ - if (va >= DMAP_MIN_ADDRESS && va < DMAP_MAX_ADDRESS) + if (va >= kva_layout.dmap_low && va < kva_layout.dmap_high) return; offset = va & PAGE_MASK; size = round_page(offset + size); @@ -9729,7 +9777,7 @@ pmap_change_prot(vm_offset_t va, vm_size_t size, vm_prot_t prot) int error; /* Only supported within the kernel map. */ - if (va < VM_MIN_KERNEL_ADDRESS) + if (va < kva_layout.km_low) return (EINVAL); PMAP_LOCK(kernel_pmap); @@ -9760,7 +9808,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, * Only supported on kernel virtual addresses, including the direct * map but excluding the recursive map. */ - if (base < DMAP_MIN_ADDRESS) + if (base < kva_layout.dmap_low) return (EINVAL); /* @@ -9783,7 +9831,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, pte_bits |= X86_PG_RW; } if ((prot & VM_PROT_EXECUTE) == 0 || - va < VM_MIN_KERNEL_ADDRESS) { + va < kva_layout.km_low) { pde_bits |= pg_nx; pte_bits |= pg_nx; } @@ -9879,7 +9927,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, pmap_pte_props(pdpe, pde_bits, pde_mask); changed = true; } - if (tmpva >= VM_MIN_KERNEL_ADDRESS && + if (tmpva >= kva_layout.km_low && (*pdpe & PG_PS_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ @@ -9909,7 +9957,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, pmap_pte_props(pde, pde_bits, pde_mask); changed = true; } - if (tmpva >= VM_MIN_KERNEL_ADDRESS && + if (tmpva >= kva_layout.km_low && (*pde & PG_PS_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ @@ -9937,7 +9985,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, pmap_pte_props(pte, pte_bits, pte_mask); changed = true; } - if (tmpva >= VM_MIN_KERNEL_ADDRESS && + if (tmpva >= kva_layout.km_low && (*pte & PG_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ @@ -10909,8 +10957,8 @@ pmap_large_unmap(void *svaa, vm_size_t len) struct spglist spgf; sva = (vm_offset_t)svaa; - if (len == 0 || sva + len < sva || (sva >= DMAP_MIN_ADDRESS && - sva + len <= DMAP_MIN_ADDRESS + dmaplimit)) + if (len == 0 || sva + len < sva || (sva >= kva_layout.dmap_low && + sva + len < kva_layout.dmap_high)) return; SLIST_INIT(&spgf); @@ -11156,11 +11204,10 @@ pmap_large_map_wb(void *svap, vm_size_t len) sva = (vm_offset_t)svap; eva = sva + len; pmap_large_map_wb_fence(); - if (sva >= DMAP_MIN_ADDRESS && eva <= DMAP_MIN_ADDRESS + dmaplimit) { + if (sva >= kva_layout.dmap_low && eva < kva_layout.dmap_high) { pmap_large_map_flush_range(sva, len); } else { - KASSERT(sva >= LARGEMAP_MIN_ADDRESS && - eva <= LARGEMAP_MIN_ADDRESS + lm_ents * NBPML4, + KASSERT(sva >= kva_layout.lm_low && eva < kva_layout.lm_high, ("pmap_large_map_wb: not largemap %#lx %#lx", sva, len)); pmap_large_map_wb_large(sva, eva); } @@ -11201,8 +11248,8 @@ pmap_pti_init(void) VM_OBJECT_WLOCK(pti_obj); pml4_pg = pmap_pti_alloc_page(); pti_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pml4_pg)); - for (va = VM_MIN_KERNEL_ADDRESS; va <= VM_MAX_KERNEL_ADDRESS && - va >= VM_MIN_KERNEL_ADDRESS && va > NBPML4; va += NBPML4) { + for (va = kva_layout.km_low; va <= kva_layout.km_high && + va >= kva_layout.km_low && va > NBPML4; va += NBPML4) { pdpe = pmap_pti_pdpe(va); pmap_pti_wire_pte(pdpe); } diff --git a/sys/amd64/amd64/trap.c b/sys/amd64/amd64/trap.c index 09ac0a67dbef..eefddad2f142 100644 --- a/sys/amd64/amd64/trap.c +++ b/sys/amd64/amd64/trap.c @@ -769,7 +769,7 @@ trap_pfault(struct trapframe *frame, bool usermode, int *signo, int *ucode) return (-1); } } - if (eva >= VM_MIN_KERNEL_ADDRESS) { + if (eva >= kva_layout.km_low) { /* * Don't allow user-mode faults in kernel address space. */ diff --git a/sys/amd64/include/param.h b/sys/amd64/include/param.h index 8db314fa034d..1bbb302259d6 100644 --- a/sys/amd64/include/param.h +++ b/sys/amd64/include/param.h @@ -146,8 +146,9 @@ #define amd64_btop(x) ((unsigned long)(x) >> PAGE_SHIFT) #define amd64_ptob(x) ((unsigned long)(x) << PAGE_SHIFT) -#define INKERNEL(va) (((va) >= DMAP_MIN_ADDRESS && (va) < DMAP_MAX_ADDRESS) \ - || ((va) >= VM_MIN_KERNEL_ADDRESS && (va) < VM_MAX_KERNEL_ADDRESS)) +#define INKERNEL(va) \ + (((va) >= kva_layout.dmap_low && (va) < kva_layout.dmap_high) || \ + ((va) >= kva_layout.km_low && (va) < kva_layout.km_high)) #ifdef SMP #define SC_TABLESIZE 1024 /* Must be power of 2. */ diff --git a/sys/amd64/include/pmap.h b/sys/amd64/include/pmap.h index 7d3e91bcd9b9..08e96027a5ed 100644 --- a/sys/amd64/include/pmap.h +++ b/sys/amd64/include/pmap.h @@ -169,11 +169,12 @@ * the recursive page table map. */ #define NDMPML4E 8 +#define NDMPML5E 32 /* - * These values control the layout of virtual memory. The starting address - * of the direct map, which is controlled by DMPML4I, must be a multiple of - * its size. (See the PHYS_TO_DMAP() and DMAP_TO_PHYS() macros.) + * These values control the layout of virtual memory. The starting + * address of the direct map is controlled by DMPML4I on LA48 and + * DMPML5I on LA57. * * Note: KPML4I is the index of the (single) level 4 page that maps * the KVA that holds KERNBASE, while KPML4BASE is the index of the @@ -191,6 +192,7 @@ #define KPML4BASE (NPML4EPG-NKPML4E) /* KVM at highest addresses */ #define DMPML4I rounddown(KPML4BASE-NDMPML4E, NDMPML4E) /* Below KVM */ +#define DMPML5I (NPML5EPG / 2 + 1) #define KPML4I (NPML4EPG-1) #define KPDPI (NPDPEPG-2) /* kernbase at -2GB */ @@ -548,6 +550,18 @@ pmap_pml5e_index(vm_offset_t va) return ((va >> PML5SHIFT) & ((1ul << NPML5EPGSHIFT) - 1)); } +struct kva_layout_s { + vm_offset_t kva_min; + vm_offset_t dmap_low; /* DMAP_MIN_ADDRESS */ + vm_offset_t dmap_high; /* DMAP_MAX_ADDRESS */ + vm_offset_t lm_low; /* LARGEMAP_MIN_ADDRESS */ + vm_offset_t lm_high; /* LARGEMAP_MAX_ADDRESS */ + vm_offset_t km_low; /* VM_MIN_KERNEL_ADDRESS */ + vm_offset_t km_high; /* VM_MAX_KERNEL_ADDRESS */ + vm_offset_t rec_pt; +}; +extern struct kva_layout_s kva_layout; + #endif /* !LOCORE */ #endif /* !_MACHINE_PMAP_H_ */ diff --git a/sys/amd64/include/vmparam.h b/sys/amd64/include/vmparam.h index 0cd9bb4fa7a4..59053665dc40 100644 --- a/sys/amd64/include/vmparam.h +++ b/sys/amd64/include/vmparam.h @@ -163,6 +163,7 @@ * Virtual addresses of things. Derived from the page directory and * page table indexes from pmap.h for precision. * + * LA48: * 0x0000000000000000 - 0x00007fffffffffff user map * 0x0000800000000000 - 0xffff7fffffffffff does not exist (hole) * 0xffff800000000000 - 0xffff804020100fff recursive page table (512GB slot) @@ -175,18 +176,29 @@ * 0xfffffc0000000000 - 0xfffffdffffffffff 2TB KMSAN shadow map, optional * 0xfffffe0000000000 - 0xffffffffffffffff 2TB kernel map * + * LA57: + * 0x0000000000000000 - 0x00ffffffffffffff user map + * 0x0100000000000000 - 0xf0ffffffffffffff does not exist (hole) + * 0xff00000000000000 - 0xff00ffffffffffff recursive page table (2048TB slot) + * 0xff01000000000000 - 0xff20ffffffffffff direct map (32 x 2048TB slots) + * 0xff21000000000000 - 0xffff807fffffffff unused + * 0xffff808000000000 - 0xffff847fffffffff large map (can be tuned up) + * 0xffff848000000000 - 0xfffff77fffffffff unused (large map extends there) + * 0xfffff60000000000 - 0xfffff7ffffffffff 2TB KMSAN origin map, optional + * 0xfffff78000000000 - 0xfffff7bfffffffff 512GB KASAN shadow map, optional + * 0xfffff80000000000 - 0xfffffbffffffffff 4TB unused + * 0xfffffc0000000000 - 0xfffffdffffffffff 2TB KMSAN shadow map, optional + * 0xfffffe0000000000 - 0xffffffffffffffff 2TB kernel map + * * Within the kernel map: * * 0xfffffe0000000000 vm_page_array * 0xffffffff80000000 KERNBASE */ -#define VM_MIN_KERNEL_ADDRESS KV4ADDR(KPML4BASE, 0, 0, 0) -#define VM_MAX_KERNEL_ADDRESS KV4ADDR(KPML4BASE + NKPML4E - 1, \ - NPDPEPG-1, NPDEPG-1, NPTEPG-1) - -#define DMAP_MIN_ADDRESS KV4ADDR(DMPML4I, 0, 0, 0) -#define DMAP_MAX_ADDRESS KV4ADDR(DMPML4I + NDMPML4E, 0, 0, 0) +#define VM_MIN_KERNEL_ADDRESS_LA48 KV4ADDR(KPML4BASE, 0, 0, 0) +#define VM_MIN_KERNEL_ADDRESS kva_layout.km_low +#define VM_MAX_KERNEL_ADDRESS kva_layout.km_high #define KASAN_MIN_ADDRESS KV4ADDR(KASANPML4I, 0, 0, 0) #define KASAN_MAX_ADDRESS KV4ADDR(KASANPML4I + NKASANPML4E, 0, 0, 0) @@ -199,9 +211,6 @@ #define KMSAN_ORIG_MAX_ADDRESS KV4ADDR(KMSANORIGPML4I + NKMSANORIGPML4E, \ 0, 0, 0) -#define LARGEMAP_MIN_ADDRESS KV4ADDR(LMSPML4I, 0, 0, 0) -#define LARGEMAP_MAX_ADDRESS KV4ADDR(LMEPML4I + 1, 0, 0, 0) - /* * Formally kernel mapping starts at KERNBASE, but kernel linker * script leaves first PDE reserved. For legacy BIOS boot, kernel is @@ -239,21 +248,21 @@ * vt fb startup needs to be reworked. */ #define PHYS_IN_DMAP(pa) (dmaplimit == 0 || (pa) < dmaplimit) -#define VIRT_IN_DMAP(va) ((va) >= DMAP_MIN_ADDRESS && \ - (va) < (DMAP_MIN_ADDRESS + dmaplimit)) +#define VIRT_IN_DMAP(va) \ + ((va) >= kva_layout.dmap_low && (va) < kva_layout.dmap_high) #define PMAP_HAS_DMAP 1 -#define PHYS_TO_DMAP(x) ({ \ +#define PHYS_TO_DMAP(x) __extension__ ({ \ KASSERT(PHYS_IN_DMAP(x), \ ("physical address %#jx not covered by the DMAP", \ (uintmax_t)x)); \ - (x) | DMAP_MIN_ADDRESS; }) + (x) + kva_layout.dmap_low; }) -#define DMAP_TO_PHYS(x) ({ \ +#define DMAP_TO_PHYS(x) __extension__ ({ \ KASSERT(VIRT_IN_DMAP(x), \ ("virtual address %#jx not covered by the DMAP", \ (uintmax_t)x)); \ - (x) & ~DMAP_MIN_ADDRESS; }) + (x) - kva_layout.dmap_low; }) /* * amd64 maps the page array into KVA so that it can be more easily @@ -274,7 +283,7 @@ */ #ifndef VM_KMEM_SIZE_MAX #define VM_KMEM_SIZE_MAX ((VM_MAX_KERNEL_ADDRESS - \ - VM_MIN_KERNEL_ADDRESS + 1) * 3 / 5) + kva_layout.km_low + 1) * 3 / 5) #endif /* initial pagein size of beginning of executable file */ |