]> git.karo-electronics.de Git - karo-tx-linux.git/blobdiff - arch/x86/xen/enlighten.c
xen: avoid hypercalls when updating unpinned pud/pmd
[karo-tx-linux.git] / arch / x86 / xen / enlighten.c
index c0388220cf9728a8e5b47c6032b0baeef1ed12f9..ed9f04b3836df4f6307a03237e1e2b8cb8dee728 100644 (file)
@@ -75,13 +75,13 @@ DEFINE_PER_CPU(unsigned long, xen_current_cr3);      /* actual vcpu cr3 */
 struct start_info *xen_start_info;
 EXPORT_SYMBOL_GPL(xen_start_info);
 
-static /* __initdata */ struct shared_info dummy_shared_info;
+struct shared_info xen_dummy_shared_info;
 
 /*
  * Point at some empty memory to start with. We map the real shared_info
  * page as soon as fixmap is up and running.
  */
-struct shared_info *HYPERVISOR_shared_info = (void *)&dummy_shared_info;
+struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
 
 /*
  * Flag to determine whether vcpu info placement is available on all
@@ -104,7 +104,7 @@ static void __init xen_vcpu_setup(int cpu)
        int err;
        struct vcpu_info *vcpup;
 
-       BUG_ON(HYPERVISOR_shared_info == &dummy_shared_info);
+       BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
        per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
 
        if (!have_vcpu_info_placement)
@@ -155,7 +155,8 @@ static void xen_cpuid(unsigned int *ax, unsigned int *bx,
        if (*ax == 1)
                maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
                            (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
-                           (1 << X86_FEATURE_SEP)  |  /* disable SEP */
+                           (1 << X86_FEATURE_MCE)  |  /* disable MCE */
+                           (1 << X86_FEATURE_MCA)  |  /* disable MCA */
                            (1 << X86_FEATURE_ACC));   /* thermal monitoring */
 
        asm(XEN_EMULATE_PREFIX "cpuid"
@@ -234,13 +235,13 @@ static void xen_irq_enable(void)
 {
        struct vcpu_info *vcpu;
 
-       /* There's a one instruction preempt window here.  We need to
-          make sure we're don't switch CPUs between getting the vcpu
-          pointer and updating the mask. */
-       preempt_disable();
+       /* We don't need to worry about being preempted here, since
+          either a) interrupts are disabled, so no preemption, or b)
+          the caller is confused and is trying to re-enable interrupts
+          on an indeterminate processor. */
+
        vcpu = x86_read_percpu(xen_vcpu);
        vcpu->evtchn_upcall_mask = 0;
-       preempt_enable_no_resched();
 
        /* Doesn't matter if we get preempted here, because any
           pending event will get dealt with anyway. */
@@ -253,7 +254,7 @@ static void xen_irq_enable(void)
 static void xen_safe_halt(void)
 {
        /* Blocking includes an implicit local_irq_enable(). */
-       if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
+       if (HYPERVISOR_sched_op(SCHEDOP_block, NULL) != 0)
                BUG();
 }
 
@@ -531,26 +532,37 @@ static void xen_apic_write(unsigned long reg, u32 val)
 static void xen_flush_tlb(void)
 {
        struct mmuext_op *op;
-       struct multicall_space mcs = xen_mc_entry(sizeof(*op));
+       struct multicall_space mcs;
+
+       preempt_disable();
+
+       mcs = xen_mc_entry(sizeof(*op));
 
        op = mcs.args;
        op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
        MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 
        xen_mc_issue(PARAVIRT_LAZY_MMU);
+
+       preempt_enable();
 }
 
 static void xen_flush_tlb_single(unsigned long addr)
 {
        struct mmuext_op *op;
-       struct multicall_space mcs = xen_mc_entry(sizeof(*op));
+       struct multicall_space mcs;
+
+       preempt_disable();
 
+       mcs = xen_mc_entry(sizeof(*op));
        op = mcs.args;
        op->cmd = MMUEXT_INVLPG_LOCAL;
        op->arg1.linear_addr = addr & PAGE_MASK;
        MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 
        xen_mc_issue(PARAVIRT_LAZY_MMU);
+
+       preempt_enable();
 }
 
 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
@@ -595,6 +607,30 @@ static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
        xen_mc_issue(PARAVIRT_LAZY_MMU);
 }
 
+static void xen_clts(void)
+{
+       struct multicall_space mcs;
+
+       mcs = xen_mc_entry(0);
+
+       MULTI_fpu_taskswitch(mcs.mc, 0);
+
+       xen_mc_issue(PARAVIRT_LAZY_CPU);
+}
+
+static void xen_write_cr0(unsigned long cr0)
+{
+       struct multicall_space mcs;
+
+       /* Only pay attention to cr0.TS; everything else is
+          ignored. */
+       mcs = xen_mc_entry(0);
+
+       MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
+
+       xen_mc_issue(PARAVIRT_LAZY_CPU);
+}
+
 static void xen_write_cr2(unsigned long cr2)
 {
        x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
@@ -612,8 +648,10 @@ static unsigned long xen_read_cr2_direct(void)
 
 static void xen_write_cr4(unsigned long cr4)
 {
-       /* Just ignore cr4 changes; Xen doesn't allow us to do
-          anything anyway. */
+       cr4 &= ~X86_CR4_PGE;
+       cr4 &= ~X86_CR4_PSE;
+
+       native_write_cr4(cr4);
 }
 
 static unsigned long xen_read_cr3(void)
@@ -655,15 +693,17 @@ static void xen_write_cr3(unsigned long cr3)
 
 /* Early in boot, while setting up the initial pagetable, assume
    everything is pinned. */
-static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn)
+static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn)
 {
+#ifdef CONFIG_FLATMEM
        BUG_ON(mem_map);        /* should only be used early */
+#endif
        make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
 }
 
-/* Early release_pt assumes that all pts are pinned, since there's
+/* Early release_pte assumes that all pts are pinned, since there's
    only init_mm and anything attached to that is pinned. */
-static void xen_release_pt_init(u32 pfn)
+static void xen_release_pte_init(u32 pfn)
 {
        make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
 }
@@ -697,12 +737,12 @@ static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
        }
 }
 
-static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
+static void xen_alloc_pte(struct mm_struct *mm, u32 pfn)
 {
        xen_alloc_ptpage(mm, pfn, PT_PTE);
 }
 
-static void xen_alloc_pd(struct mm_struct *mm, u32 pfn)
+static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn)
 {
        xen_alloc_ptpage(mm, pfn, PT_PMD);
 }
@@ -722,12 +762,12 @@ static void xen_release_ptpage(u32 pfn, unsigned level)
        }
 }
 
-static void xen_release_pt(u32 pfn)
+static void xen_release_pte(u32 pfn)
 {
        xen_release_ptpage(pfn, PT_PTE);
 }
 
-static void xen_release_pd(u32 pfn)
+static void xen_release_pmd(u32 pfn)
 {
        xen_release_ptpage(pfn, PT_PMD);
 }
@@ -771,38 +811,35 @@ static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
 static __init void xen_pagetable_setup_start(pgd_t *base)
 {
        pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
+       int i;
 
        /* special set_pte for pagetable initialization */
        pv_mmu_ops.set_pte = xen_set_pte_init;
 
        init_mm.pgd = base;
        /*
-        * copy top-level of Xen-supplied pagetable into place.  For
-        * !PAE we can use this as-is, but for PAE it is a stand-in
-        * while we copy the pmd pages.
+        * copy top-level of Xen-supplied pagetable into place.  This
+        * is a stand-in while we copy the pmd pages.
         */
        memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
 
-       if (PTRS_PER_PMD > 1) {
-               int i;
-               /*
-                * For PAE, need to allocate new pmds, rather than
-                * share Xen's, since Xen doesn't like pmd's being
-                * shared between address spaces.
-                */
-               for (i = 0; i < PTRS_PER_PGD; i++) {
-                       if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
-                               pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
+       /*
+        * For PAE, need to allocate new pmds, rather than
+        * share Xen's, since Xen doesn't like pmd's being
+        * shared between address spaces.
+        */
+       for (i = 0; i < PTRS_PER_PGD; i++) {
+               if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
+                       pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 
-                               memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
-                                      PAGE_SIZE);
+                       memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
+                              PAGE_SIZE);
 
-                               make_lowmem_page_readonly(pmd);
+                       make_lowmem_page_readonly(pmd);
 
-                               set_pgd(&base[i], __pgd(1 + __pa(pmd)));
-                       } else
-                               pgd_clear(&base[i]);
-               }
+                       set_pgd(&base[i], __pgd(1 + __pa(pmd)));
+               } else
+                       pgd_clear(&base[i]);
        }
 
        /* make sure zero_page is mapped RO so we can use it in pagetables */
@@ -820,7 +857,7 @@ static __init void xen_pagetable_setup_start(pgd_t *base)
                          PFN_DOWN(__pa(xen_start_info->pt_base)));
 }
 
-static __init void setup_shared_info(void)
+void xen_setup_shared_info(void)
 {
        if (!xen_feature(XENFEAT_auto_translated_physmap)) {
                unsigned long addr = fix_to_virt(FIX_PARAVIRT_BOOTMAP);
@@ -843,37 +880,37 @@ static __init void setup_shared_info(void)
        /* In UP this is as good a place as any to set up shared info */
        xen_setup_vcpu_info_placement();
 #endif
+
+       xen_setup_mfn_list_list();
 }
 
 static __init void xen_pagetable_setup_done(pgd_t *base)
 {
        /* This will work as long as patching hasn't happened yet
           (which it hasn't) */
-       pv_mmu_ops.alloc_pt = xen_alloc_pt;
-       pv_mmu_ops.alloc_pd = xen_alloc_pd;
-       pv_mmu_ops.release_pt = xen_release_pt;
-       pv_mmu_ops.release_pd = xen_release_pd;
+       pv_mmu_ops.alloc_pte = xen_alloc_pte;
+       pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
+       pv_mmu_ops.release_pte = xen_release_pte;
+       pv_mmu_ops.release_pmd = xen_release_pmd;
        pv_mmu_ops.set_pte = xen_set_pte;
 
-       setup_shared_info();
+       xen_setup_shared_info();
 
        /* Actually pin the pagetable down, but we can't set PG_pinned
           yet because the page structures don't exist yet. */
-       {
-               unsigned level;
+       pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(base)));
+}
 
-#ifdef CONFIG_X86_PAE
-               level = MMUEXT_PIN_L3_TABLE;
-#else
-               level = MMUEXT_PIN_L2_TABLE;
-#endif
+static __init void xen_post_allocator_init(void)
+{
+       pv_mmu_ops.set_pmd = xen_set_pmd;
+       pv_mmu_ops.set_pud = xen_set_pud;
 
-               pin_pagetable_pfn(level, PFN_DOWN(__pa(base)));
-       }
+       xen_mark_init_mm_pinned();
 }
 
 /* This is called once we have the cpu_possible_map */
-void __init xen_setup_vcpu_info_placement(void)
+void xen_setup_vcpu_info_placement(void)
 {
        int cpu;
 
@@ -959,7 +996,7 @@ static const struct pv_init_ops xen_init_ops __initdata = {
        .banner = xen_banner,
        .memory_setup = xen_memory_setup,
        .arch_setup = xen_arch_setup,
-       .post_allocator_init = xen_mark_init_mm_pinned,
+       .post_allocator_init = xen_post_allocator_init,
 };
 
 static const struct pv_time_ops xen_time_ops __initdata = {
@@ -977,10 +1014,10 @@ static const struct pv_cpu_ops xen_cpu_ops __initdata = {
        .set_debugreg = xen_set_debugreg,
        .get_debugreg = xen_get_debugreg,
 
-       .clts = native_clts,
+       .clts = xen_clts,
 
        .read_cr0 = native_read_cr0,
-       .write_cr0 = native_write_cr0,
+       .write_cr0 = xen_write_cr0,
 
        .read_cr4 = native_read_cr4,
        .read_cr4_safe = native_read_cr4_safe,
@@ -994,7 +1031,7 @@ static const struct pv_cpu_ops xen_cpu_ops __initdata = {
        .read_pmc = native_read_pmc,
 
        .iret = xen_iret,
-       .irq_enable_syscall_ret = NULL,  /* never called */
+       .irq_enable_syscall_ret = xen_sysexit,
 
        .load_tr_desc = paravirt_nop,
        .set_ldt = xen_set_ldt,
@@ -1059,11 +1096,11 @@ static const struct pv_mmu_ops xen_mmu_ops __initdata = {
        .pte_update = paravirt_nop,
        .pte_update_defer = paravirt_nop,
 
-       .alloc_pt = xen_alloc_pt_init,
-       .release_pt = xen_release_pt_init,
-       .alloc_pd = xen_alloc_pt_init,
-       .alloc_pd_clone = paravirt_nop,
-       .release_pd = xen_release_pt_init,
+       .alloc_pte = xen_alloc_pte_init,
+       .release_pte = xen_release_pte_init,
+       .alloc_pmd = xen_alloc_pte_init,
+       .alloc_pmd_clone = paravirt_nop,
+       .release_pmd = xen_release_pte_init,
 
 #ifdef CONFIG_HIGHPTE
        .kmap_atomic_pte = xen_kmap_atomic_pte,
@@ -1071,24 +1108,23 @@ static const struct pv_mmu_ops xen_mmu_ops __initdata = {
 
        .set_pte = NULL,        /* see xen_pagetable_setup_* */
        .set_pte_at = xen_set_pte_at,
-       .set_pmd = xen_set_pmd,
+       .set_pmd = xen_set_pmd_hyper,
 
        .pte_val = xen_pte_val,
+       .pte_flags = native_pte_val,
        .pgd_val = xen_pgd_val,
 
        .make_pte = xen_make_pte,
        .make_pgd = xen_make_pgd,
 
-#ifdef CONFIG_X86_PAE
        .set_pte_atomic = xen_set_pte_atomic,
        .set_pte_present = xen_set_pte_at,
-       .set_pud = xen_set_pud,
+       .set_pud = xen_set_pud_hyper,
        .pte_clear = xen_pte_clear,
        .pmd_clear = xen_pmd_clear,
 
        .make_pmd = xen_make_pmd,
        .pmd_val = xen_pmd_val,
-#endif /* PAE */
 
        .activate_mm = xen_activate_mm,
        .dup_mmap = xen_dup_mmap,
@@ -1115,11 +1151,13 @@ static const struct smp_ops xen_smp_ops __initdata = {
 
 static void xen_reboot(int reason)
 {
+       struct sched_shutdown r = { .reason = reason };
+
 #ifdef CONFIG_SMP
        smp_send_stop();
 #endif
 
-       if (HYPERVISOR_sched_op(SCHEDOP_shutdown, reason))
+       if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
                BUG();
 }
 
@@ -1193,7 +1231,7 @@ asmlinkage void __init xen_start_kernel(void)
 
        /* Get mfn list */
        if (!xen_feature(XENFEAT_auto_translated_physmap))
-               phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
+               xen_build_dynamic_phys_to_machine();
 
        pgd = (pgd_t *)xen_start_info->pt_base;
 
@@ -1228,8 +1266,11 @@ asmlinkage void __init xen_start_kernel(void)
                ? __pa(xen_start_info->mod_start) : 0;
        boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
 
-       if (!is_initial_xendomain())
+       if (!is_initial_xendomain()) {
+               add_preferred_console("xenboot", 0, NULL);
+               add_preferred_console("tty", 0, NULL);
                add_preferred_console("hvc", 0, NULL);
+       }
 
        /* Start the world */
        start_kernel();