Date: Mon, 25 Mar 2013 13:33:06 +0000 (UTC) From: "Cherry G. Mathew" <cherry@FreeBSD.org> To: src-committers@freebsd.org, svn-src-projects@freebsd.org Subject: svn commit: r248703 - in projects/amd64_xen_pv/sys/amd64: amd64 include xen Message-ID: <201303251333.r2PDX6a9028719@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: cherry Date: Mon Mar 25 13:33:06 2013 New Revision: 248703 URL: http://svnweb.freebsd.org/changeset/base/248703 Log: This commit combines the following features: i) Context switch now DTRT and maps in userspace into a separate kernel VA ii) Supporting pmap_xxx() functions for boot progress have been added. iii) Boottime \"dummy\" user page tables have been temporarily disabled. Approved by: gibbs(implicit) Modified: projects/amd64_xen_pv/sys/amd64/amd64/cpu_switch.S projects/amd64_xen_pv/sys/amd64/amd64/vm_machdep.c projects/amd64_xen_pv/sys/amd64/include/pmap.h projects/amd64_xen_pv/sys/amd64/xen/mm.c projects/amd64_xen_pv/sys/amd64/xen/pmap.c Modified: projects/amd64_xen_pv/sys/amd64/amd64/cpu_switch.S ============================================================================== --- projects/amd64_xen_pv/sys/amd64/amd64/cpu_switch.S Mon Mar 25 13:23:47 2013 (r248702) +++ projects/amd64_xen_pv/sys/amd64/amd64/cpu_switch.S Mon Mar 25 13:33:06 2013 (r248703) @@ -138,7 +138,6 @@ ctx_switch_xsave: /* Save is done. Now fire up new thread. Leave old vmspace. */ movq TD_PCB(%rsi),%r8 - /* switch address space */ movq PCB_CR3(%r8),%rcx movq %cr3,%rax @@ -155,9 +154,23 @@ swinact: pushq %rdx pushq %r8 - movq %rcx, %rdi - callq xen_pt_switch + /* + * On xen, the hypervisor loads %cr3 for us on return to + * userland. We use a separate "kernel space" for kernel mode, + * which is setup at boot time (see: pmap.c:pmap_bootstrap) + * + * We need to tell the hypervisor via xen_pt_user_switch() + * about the new user pmap. Additionally, we modify the kernel VA + * space by copying in the userland bits of the new pmap, in + * case the kernel needs to access them. + */ + movq TD_PROC(%rsi), %rdx /* newproc */ + movq P_VMSPACE(%rdx), %rdx + addq $VM_PMAP, %rdx + movq %rdx, %rdi + callq pmap_xen_userload + popq %r8 popq %rdx popq %rsi Modified: projects/amd64_xen_pv/sys/amd64/amd64/vm_machdep.c ============================================================================== --- projects/amd64_xen_pv/sys/amd64/amd64/vm_machdep.c Mon Mar 25 13:23:47 2013 (r248702) +++ projects/amd64_xen_pv/sys/amd64/amd64/vm_machdep.c Mon Mar 25 13:33:06 2013 (r248703) @@ -220,7 +220,7 @@ cpu_fork(td1, p2, td2, flags) */ pmap2 = vmspace_pmap(p2->p_vmspace); #ifdef XEN - pcb2->pcb_cr3 = pmap_kextract((vm_offset_t)pmap2->pm_pml4); + pcb2->pcb_cr3 = pmap_kextract_ma((vm_offset_t)pmap2->pm_pml4); #else pcb2->pcb_cr3 = DMAP_TO_PHYS((vm_offset_t)pmap2->pm_pml4); #endif Modified: projects/amd64_xen_pv/sys/amd64/include/pmap.h ============================================================================== --- projects/amd64_xen_pv/sys/amd64/include/pmap.h Mon Mar 25 13:23:47 2013 (r248702) +++ projects/amd64_xen_pv/sys/amd64/include/pmap.h Mon Mar 25 13:33:06 2013 (r248703) @@ -365,6 +365,7 @@ void pmap_invalidate_all(pmap_t); void pmap_invalidate_cache(void); void pmap_invalidate_cache_pages(vm_page_t *pages, int count); void pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva); +void pmap_xen_userload(pmap_t); #endif /* _KERNEL */ Modified: projects/amd64_xen_pv/sys/amd64/xen/mm.c ============================================================================== --- projects/amd64_xen_pv/sys/amd64/xen/mm.c Mon Mar 25 13:23:47 2013 (r248702) +++ projects/amd64_xen_pv/sys/amd64/xen/mm.c Mon Mar 25 13:33:06 2013 (r248703) @@ -163,12 +163,13 @@ xen_invlpg(vm_offset_t va) inline void xen_load_cr3(u_long val) { - xen_pt_switch(val); + xen_pt_switch(xpmap_ptom(val)); } void -xen_pt_switch(vm_paddr_t kpml4phys) +xen_pt_switch(vm_paddr_t kpml4mach) { + printk("%s: kpml4mach == 0x%lx\n", __func__, kpml4mach); struct mmuext_op op; #ifdef INVARIANTS SET_VCPU(); @@ -176,12 +177,12 @@ xen_pt_switch(vm_paddr_t kpml4phys) KASSERT(XPQ_IDX == 0, ("pending operations XPQ_IDX=%d", XPQ_IDX)); #endif op.cmd = MMUEXT_NEW_BASEPTR; - op.arg1.mfn = xpmap_ptom(kpml4phys) >> PAGE_SHIFT; + op.arg1.mfn = kpml4mach >> PAGE_SHIFT; PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0); } void -xen_pt_user_switch(vm_paddr_t upml4phys) +xen_pt_user_switch(vm_paddr_t upml4mach) { struct mmuext_op op; #ifdef INVARIANTS @@ -190,7 +191,7 @@ xen_pt_user_switch(vm_paddr_t upml4phys) KASSERT(XPQ_IDX == 0, ("pending operations XPQ_IDX=%d", XPQ_IDX)); #endif op.cmd = MMUEXT_NEW_USER_BASEPTR; - op.arg1.mfn = xpmap_ptom(upml4phys) >> PAGE_SHIFT; + op.arg1.mfn = upml4mach >> PAGE_SHIFT; PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0); } Modified: projects/amd64_xen_pv/sys/amd64/xen/pmap.c ============================================================================== --- projects/amd64_xen_pv/sys/amd64/xen/pmap.c Mon Mar 25 13:23:47 2013 (r248702) +++ projects/amd64_xen_pv/sys/amd64/xen/pmap.c Mon Mar 25 13:33:06 2013 (r248703) @@ -132,6 +132,9 @@ __FBSDID("$FreeBSD$"); #include <sys/smp.h> #endif +#include <sys/proc.h> +#include <sys/sched.h> + #include <vm/vm.h> #include <vm/vm_extern.h> #include <vm/vm_object.h> @@ -511,7 +514,7 @@ pmap_xen_bootpages(vm_paddr_t *firstaddr HYPERVISOR_shared_info = (void *) va; - +#if 0 /* ii) Userland page table base */ va = vallocpages(firstaddr, 1); bzero((void *)va, PAGE_SIZE); @@ -530,7 +533,8 @@ pmap_xen_bootpages(vm_paddr_t *firstaddr xen_pgdir_pin(phystomach(VTOP(va))); /* Register user page table with Xen */ - xen_pt_user_switch(VTOP(va)); + xen_pt_user_switch(xpmap_ptom(VTOP(va))); +#endif } /* Boot time ptov - xen guarantees bootpages to be offset */ @@ -578,7 +582,7 @@ pmap_bootstrap(vm_paddr_t *firstaddr) create_boot_pagetables(firstaddr); /* Switch to the new kernel tables */ - xen_pt_switch(VTOP(KPML4phys)); + xen_pt_switch(xpmap_ptom(VTOP(KPML4phys))); /* Unpin old page table hierarchy, and mark all its pages r/w */ xen_pgdir_unpin(phystomach(VTOP(xen_start_info->pt_base))); @@ -774,38 +778,22 @@ pmap_pinit(pmap_t pmap) */ pmap->pm_pml4 = (void *) kmem_alloc(kernel_map, PAGE_SIZE); bzero(pmap->pm_pml4, PAGE_SIZE); - + printf("%s: pmap->pm_pml4 == %p\n", __func__, pmap->pm_pml4); /* * We do not wire in kernel space, or the self-referencial - * entry in userspace pmaps for two reasons: - * i) both kernel and userland run in ring3 (same CPU - * privilege level). This means that userland that has kernel - * address space mapped in, can access kernel memory! - * Instead, we make the kernel pmap is exclusive and - * unshared, and we switch to it on *every* kernel - * entry. This is facilitated by the hypervisor. - * ii) we access the user pmap from within kernel VA. The - * self-referencing entry is useful if we access the pmap - * from the *user* VA. - * XXX: review this when userland is up. + * entry in userspace pmaps becase both kernel and userland + * share ring3 privilege. The user/kernel context switch is + * arbitrated by the hypervisor by means of pre-loaded values + * for kernel and user %cr3. The userland parts of kernel VA + * may be conditionally overlaid with the VA of curthread, + * since the kernel occasionally needs to access userland + * process VA space. */ -#if 1 /* XXX: DEBUG ONLY - EXPOSES KERNEL TO USERLAND - TERRIBLE SECURITY RISK! */ - /* Wire in kernel global address entries. */ - pmap->pm_pml4[KPML4I] = phystomach(VTOP(KPDPphys)) | PG_RW | PG_V | PG_U; - - /* Copy over Direct mapping entries, from kernel_pmap. */ - int i; - for (i = 0; i < NDMPML4E; i++) { - pmap->pm_pml4[DMPML4I + i] = ((pdp_entry_t *)KPML4phys)[DMPML4I + i]; - } - pmap_xen_setpages_ro((uintptr_t)pmap->pm_pml4, 1); xen_pgdir_pin(phystomach(ptmb_vtop((uintptr_t)pmap->pm_pml4))); -#endif - pmap->pm_root = NULL; CPU_ZERO(&pmap->pm_active); pmap_pv_pmap_init(pmap); @@ -814,14 +802,33 @@ pmap_pinit(pmap_t pmap) return 1; } +void pmap_xen_userload(pmap_t pmap) +{ + KASSERT(pmap != kernel_pmap, + ("Kernel pmap requested on user load.\n")); + + printf("%s: pmap->pm_pml4 == %p\n", __func__, pmap->pm_pml4); + printf("%s: curthread %s\n", __func__, curthread->td_name); + int i; + for (i = 0; i < NUPML4E; i++) { + pml4_entry_t pml4e; + pml4e = (pmap->pm_pml4[i]); + PT_SET_VA_MA((pml4_entry_t *)KPML4phys + i, pml4e, false); + } + PT_UPDATES_FLUSH(); + + /* Tell xen about user pmap switch */ + xen_pt_user_switch(vtomach(pmap->pm_pml4)); +} + void pmap_release(pmap_t pmap) { KASSERT(0, ("XXX: %s: TODO\n", __func__)); } -pt_entry_t * -vtopte_hold(uintptr_t va, void *addr) +static pt_entry_t * +pmap_vtopte_hold(pmap_t pmap, uintptr_t va, void *addr) { KASSERT(addr != NULL, ("addr == NULL")); @@ -850,16 +857,28 @@ vtopte_hold(uintptr_t va, void *addr) return pte; } -void -vtopte_release(uintptr_t va, void *addr) +pt_entry_t * +vtopte_hold(uintptr_t va, void *addr) +{ + return pmap_vtopte_hold(kernel_pmap, va, addr); +} + +static void +pmap_vtopte_release(pmap_t pmap, uintptr_t va, void *addr) { mmu_map_t tptr = *(mmu_map_t *)addr; - mmu_map_release_va(kernel_pmap, tptr, va); + mmu_map_release_va(pmap, tptr, va); mmu_map_t_fini(tptr); } +void +vtopte_release(uintptr_t va, void *addr) +{ + pmap_vtopte_release(kernel_pmap, va, addr); +} + #ifdef SMP void pmap_lazyfix_action(void); @@ -943,6 +962,7 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_prot_t prot, boolean_t wired) { va = trunc_page(va); + KASSERT(va <= VM_MAX_KERNEL_ADDRESS, ("pmap_enter: toobig")); KASSERT(va < UPT_MIN_ADDRESS || va >= UPT_MAX_ADDRESS, ("pmap_enter: invalid to pmap_enter page table pages (va: 0x%lx)", @@ -998,10 +1018,57 @@ pmap_remove(pmap_t pmap, vm_offset_t sva pmap_qremove(sva, atop(eva - sva)); } +static bool +pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m) +{ + pt_entry_t *pte, tpte; + + char tbuf[tsz]; /* Safe to do this on the stack since tsz is + * effectively const. + */ + + mmu_map_t tptr = tbuf; + + PMAP_LOCK(pmap); + pte = pmap_vtopte_hold(pmap, va, &tptr); + tpte = *pte; + PT_CLEAR_VA(pte, TRUE); + if (tpte & PG_A) + vm_page_aflag_set(m, PGA_REFERENCED); + + /* + * Update the vm_page_t clean and reference bits. + */ + if ((tpte & (PG_M | PG_RW)) == (PG_M | PG_RW)) + vm_page_dirty(m); + + /* XXX: Tell mmu_xxx about backing page */ + pmap_vtopte_release(pmap, va, &tptr); + + pmap_invalidate_page(pmap, va); + PMAP_UNLOCK(pmap); + + return false; +} + +/* + * Routine: pmap_remove_all + * Function: + * Removes this physical page from + * all physical maps in which it resides. + * Reflects back modify bits to the pager. + */ void pmap_remove_all(vm_page_t m) { - KASSERT(0, ("XXX: %s: TODO\n", __func__)); + + KASSERT((m->oflags & VPO_UNMANAGED) == 0, + ("pmap_remove_all: page %p is not managed", m)); + + pmap_pv_iterate(m, pv_remove); + + /* free pv entry from all pmaps */ + pmap_pv_page_unmap(m); } vm_paddr_t @@ -1225,7 +1292,21 @@ pmap_protect(pmap_t pmap, vm_offset_t sv void pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { - KASSERT(0, ("XXX: %s: TODO\n", __func__)); + + vm_offset_t addr; + /* XXX: TODO SMP */ + sched_pin(); + + for (addr = sva; addr < eva; addr += PAGE_SIZE) + invlpg(addr); + + sched_unpin(); +} + +void +pmap_invalidate_page(pmap_t pmap, vm_offset_t va) +{ + pmap_invalidate_range(pmap, va, va + PAGE_SIZE); } void @@ -1288,11 +1369,21 @@ pmap_page_set_memattr(vm_page_t m, vm_me KASSERT(0, ("XXX: %s: TODO\n", __func__)); } +static bool +pv_dummy(pmap_t pmap, vm_offset_t va, vm_page_t m) +{ + printf("%s: va == 0x%lx, pa == 0x%lx\n", + __func__, va, VM_PAGE_TO_PHYS(m)); + return true; /* stop at the first iteration */ +} + boolean_t pmap_page_is_mapped(vm_page_t m) { - KASSERT(0, ("XXX: %s: TODO\n", __func__)); - return 0; + if ((m->oflags & VPO_UNMANAGED) != 0) + return (FALSE); + printf("pmap_pv_iterate(m, pv_dummy) == %d\n", pmap_pv_iterate(m, pv_dummy)); + return pmap_pv_iterate(m, pv_dummy); } boolean_t @@ -1342,10 +1433,55 @@ pmap_clear_reference(vm_page_t m) KASSERT(0, ("XXX: %s: TODO\n", __func__)); } +/* Callback to remove write access on given va and pmap */ +static bool +pv_remove_write(pmap_t pmap, vm_offset_t va, vm_page_t m) +{ + + pt_entry_t oldpte, *pte; + char tbuf[tsz]; /* Safe to do this on the stack since tsz is + * effectively const. + */ + + mmu_map_t tptr = tbuf; + + PMAP_LOCK(pmap); + pte = pmap_vtopte_hold(pmap, va, &tptr); + + oldpte = *pte; + if (oldpte & PG_RW) { + PT_SET_MA(va, oldpte & ~(PG_RW | PG_M)); + if ((oldpte & PG_M) != 0) + vm_page_dirty(m); + pmap_invalidate_page(pmap, va); + } + pmap_vtopte_release(pmap, va, &tptr); + PMAP_UNLOCK(pmap); + + return false; /* Iterate through every mapping */ +} + +/* + * Clear the write and modified bits in each of the given page's mappings. + */ void pmap_remove_write(vm_page_t m) { - KASSERT(0, ("XXX: %s: TODO\n", __func__)); + KASSERT((m->oflags & VPO_UNMANAGED) == 0, + ("pmap_remove_write: page %p is not managed", m)); + + /* + * If the page is not VPO_BUSY, then PGA_WRITEABLE cannot be set by + * another thread while the object is locked. Thus, if PGA_WRITEABLE + * is clear, no page table entries need updating. + */ + VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); + if ((m->oflags & VPO_BUSY) == 0 && + (m->aflags & PGA_WRITEABLE) == 0) + return; + + pmap_pv_iterate(m, pv_remove_write); + vm_page_aflag_clear(m, PGA_WRITEABLE); } int
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201303251333.r2PDX6a9028719>