From 2cc42bac1c795f75fcc062b95c6ca7ac1b84d5d8 Mon Sep 17 00:00:00 2001 From: Jan Beulich Date: Mon, 18 Dec 2017 09:37:45 -0700 Subject: [PATCH 1/2] x86-64/Xen: eliminate W+X mappings A few thousand such pages are usually left around due to the re-use of L1 tables having been provided by the hypervisor (Dom0) or tool stack (DomU). Set NX in the direct map variant, which needs to be done in L2 due to the dual use of the re-used L1s. For x86_configure_nx() to actually do what it is supposed to do, call get_cpu_cap() first. This was broken by commit 4763ed4d45 ("x86, mm: Clean up and simplify NX enablement") when switching away from the direct EFER read. Signed-off-by: Jan Beulich Reviewed-by: Juergen Gross Signed-off-by: Boris Ostrovsky --- arch/x86/xen/enlighten_pv.c | 3 +++ arch/x86/xen/mmu_pv.c | 12 ++++++++++++ 2 files changed, 15 insertions(+) diff --git a/arch/x86/xen/enlighten_pv.c b/arch/x86/xen/enlighten_pv.c index 69b9deff7e5c..86f26ea99324 100644 --- a/arch/x86/xen/enlighten_pv.c +++ b/arch/x86/xen/enlighten_pv.c @@ -87,6 +87,8 @@ #include "multicalls.h" #include "pmu.h" +#include "../kernel/cpu/cpu.h" /* get_cpu_cap() */ + void *xen_initial_gdt; static int xen_cpu_up_prepare_pv(unsigned int cpu); @@ -1249,6 +1251,7 @@ asmlinkage __visible void __init xen_start_kernel(void) __userpte_alloc_gfp &= ~__GFP_HIGHMEM; /* Work out if we support NX */ + get_cpu_cap(&boot_cpu_data); x86_configure_nx(); /* Get mfn list */ diff --git a/arch/x86/xen/mmu_pv.c b/arch/x86/xen/mmu_pv.c index 9d9cc3870722..7118f776cd49 100644 --- a/arch/x86/xen/mmu_pv.c +++ b/arch/x86/xen/mmu_pv.c @@ -1916,6 +1916,18 @@ void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) /* Graft it onto L4[511][510] */ copy_page(level2_kernel_pgt, l2); + /* + * Zap execute permission from the ident map. Due to the sharing of + * L1 entries we need to do this in the L2. + */ + if (__supported_pte_mask & _PAGE_NX) { + for (i = 0; i < PTRS_PER_PMD; ++i) { + if (pmd_none(level2_ident_pgt[i])) + continue; + level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX); + } + } + /* Copy the initial P->M table mappings if necessary. */ i = pgd_index(xen_start_info->mfn_list); if (i && i < pgd_index(__START_KERNEL_map)) From b3cf8528bb21febb650a7ecbf080d0647be40b9f Mon Sep 17 00:00:00 2001 From: Boris Ostrovsky Date: Tue, 12 Dec 2017 15:08:21 -0500 Subject: [PATCH 2/2] xen/balloon: Mark unallocated host memory as UNUSABLE Commit f5775e0b6116 ("x86/xen: discard RAM regions above the maximum reservation") left host memory not assigned to dom0 as available for memory hotplug. Unfortunately this also meant that those regions could be used by others. Specifically, commit fa564ad96366 ("x86/PCI: Enable a 64bit BAR on AMD Family 15h (Models 00-1f, 30-3f, 60-7f)") may try to map those addresses as MMIO. To prevent this mark unallocated host memory as E820_TYPE_UNUSABLE (thus effectively reverting f5775e0b6116) and keep track of that region as a hostmem resource that can be used for the hotplug. Signed-off-by: Boris Ostrovsky Reviewed-by: Juergen Gross --- arch/x86/xen/enlighten.c | 81 ++++++++++++++++++++++++++++++++++++++++ arch/x86/xen/setup.c | 6 +-- drivers/xen/balloon.c | 65 +++++++++++++++++++++++++++----- include/xen/balloon.h | 5 +++ 4 files changed, 144 insertions(+), 13 deletions(-) diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c index d669e9d89001..c9081c6671f0 100644 --- a/arch/x86/xen/enlighten.c +++ b/arch/x86/xen/enlighten.c @@ -1,8 +1,12 @@ +#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG +#include +#endif #include #include #include #include +#include #include #include @@ -331,3 +335,80 @@ void xen_arch_unregister_cpu(int num) } EXPORT_SYMBOL(xen_arch_unregister_cpu); #endif + +#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG +void __init arch_xen_balloon_init(struct resource *hostmem_resource) +{ + struct xen_memory_map memmap; + int rc; + unsigned int i, last_guest_ram; + phys_addr_t max_addr = PFN_PHYS(max_pfn); + struct e820_table *xen_e820_table; + const struct e820_entry *entry; + struct resource *res; + + if (!xen_initial_domain()) + return; + + xen_e820_table = kmalloc(sizeof(*xen_e820_table), GFP_KERNEL); + if (!xen_e820_table) + return; + + memmap.nr_entries = ARRAY_SIZE(xen_e820_table->entries); + set_xen_guest_handle(memmap.buffer, xen_e820_table->entries); + rc = HYPERVISOR_memory_op(XENMEM_machine_memory_map, &memmap); + if (rc) { + pr_warn("%s: Can't read host e820 (%d)\n", __func__, rc); + goto out; + } + + last_guest_ram = 0; + for (i = 0; i < memmap.nr_entries; i++) { + if (xen_e820_table->entries[i].addr >= max_addr) + break; + if (xen_e820_table->entries[i].type == E820_TYPE_RAM) + last_guest_ram = i; + } + + entry = &xen_e820_table->entries[last_guest_ram]; + if (max_addr >= entry->addr + entry->size) + goto out; /* No unallocated host RAM. */ + + hostmem_resource->start = max_addr; + hostmem_resource->end = entry->addr + entry->size; + + /* + * Mark non-RAM regions between the end of dom0 RAM and end of host RAM + * as unavailable. The rest of that region can be used for hotplug-based + * ballooning. + */ + for (; i < memmap.nr_entries; i++) { + entry = &xen_e820_table->entries[i]; + + if (entry->type == E820_TYPE_RAM) + continue; + + if (entry->addr >= hostmem_resource->end) + break; + + res = kzalloc(sizeof(*res), GFP_KERNEL); + if (!res) + goto out; + + res->name = "Unavailable host RAM"; + res->start = entry->addr; + res->end = (entry->addr + entry->size < hostmem_resource->end) ? + entry->addr + entry->size : hostmem_resource->end; + rc = insert_resource(hostmem_resource, res); + if (rc) { + pr_warn("%s: Can't insert [%llx - %llx) (%d)\n", + __func__, res->start, res->end, rc); + kfree(res); + goto out; + } + } + + out: + kfree(xen_e820_table); +} +#endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */ diff --git a/arch/x86/xen/setup.c b/arch/x86/xen/setup.c index ac55c02f98e9..e9011e1ee3de 100644 --- a/arch/x86/xen/setup.c +++ b/arch/x86/xen/setup.c @@ -807,7 +807,6 @@ char * __init xen_memory_setup(void) addr = xen_e820_table.entries[0].addr; size = xen_e820_table.entries[0].size; while (i < xen_e820_table.nr_entries) { - bool discard = false; chunk_size = size; type = xen_e820_table.entries[i].type; @@ -823,11 +822,10 @@ char * __init xen_memory_setup(void) xen_add_extra_mem(pfn_s, n_pfns); xen_max_p2m_pfn = pfn_s + n_pfns; } else - discard = true; + type = E820_TYPE_UNUSABLE; } - if (!discard) - xen_align_and_add_e820_region(addr, chunk_size, type); + xen_align_and_add_e820_region(addr, chunk_size, type); addr += chunk_size; size -= chunk_size; diff --git a/drivers/xen/balloon.c b/drivers/xen/balloon.c index f77e499afddd..065f0b607373 100644 --- a/drivers/xen/balloon.c +++ b/drivers/xen/balloon.c @@ -257,10 +257,25 @@ static void release_memory_resource(struct resource *resource) kfree(resource); } +/* + * Host memory not allocated to dom0. We can use this range for hotplug-based + * ballooning. + * + * It's a type-less resource. Setting IORESOURCE_MEM will make resource + * management algorithms (arch_remove_reservations()) look into guest e820, + * which we don't want. + */ +static struct resource hostmem_resource = { + .name = "Host RAM", +}; + +void __attribute__((weak)) __init arch_xen_balloon_init(struct resource *res) +{} + static struct resource *additional_memory_resource(phys_addr_t size) { - struct resource *res; - int ret; + struct resource *res, *res_hostmem; + int ret = -ENOMEM; res = kzalloc(sizeof(*res), GFP_KERNEL); if (!res) @@ -269,13 +284,42 @@ static struct resource *additional_memory_resource(phys_addr_t size) res->name = "System RAM"; res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; - ret = allocate_resource(&iomem_resource, res, - size, 0, -1, - PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL); - if (ret < 0) { - pr_err("Cannot allocate new System RAM resource\n"); - kfree(res); - return NULL; + res_hostmem = kzalloc(sizeof(*res), GFP_KERNEL); + if (res_hostmem) { + /* Try to grab a range from hostmem */ + res_hostmem->name = "Host memory"; + ret = allocate_resource(&hostmem_resource, res_hostmem, + size, 0, -1, + PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL); + } + + if (!ret) { + /* + * Insert this resource into iomem. Because hostmem_resource + * tracks portion of guest e820 marked as UNUSABLE noone else + * should try to use it. + */ + res->start = res_hostmem->start; + res->end = res_hostmem->end; + ret = insert_resource(&iomem_resource, res); + if (ret < 0) { + pr_err("Can't insert iomem_resource [%llx - %llx]\n", + res->start, res->end); + release_memory_resource(res_hostmem); + res_hostmem = NULL; + res->start = res->end = 0; + } + } + + if (ret) { + ret = allocate_resource(&iomem_resource, res, + size, 0, -1, + PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL); + if (ret < 0) { + pr_err("Cannot allocate new System RAM resource\n"); + kfree(res); + return NULL; + } } #ifdef CONFIG_SPARSEMEM @@ -287,6 +331,7 @@ static struct resource *additional_memory_resource(phys_addr_t size) pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n", pfn, limit); release_memory_resource(res); + release_memory_resource(res_hostmem); return NULL; } } @@ -765,6 +810,8 @@ static int __init balloon_init(void) set_online_page_callback(&xen_online_page); register_memory_notifier(&xen_memory_nb); register_sysctl_table(xen_root); + + arch_xen_balloon_init(&hostmem_resource); #endif #ifdef CONFIG_XEN_PV diff --git a/include/xen/balloon.h b/include/xen/balloon.h index 8906361bb50c..d0adfc78dcbd 100644 --- a/include/xen/balloon.h +++ b/include/xen/balloon.h @@ -43,3 +43,8 @@ static inline void xen_balloon_init(void) { } #endif + +#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG +struct resource; +void arch_xen_balloon_init(struct resource *hostmem_resource); +#endif