cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

cpu_entry_area.c (7227B)


      1// SPDX-License-Identifier: GPL-2.0
      2
      3#include <linux/spinlock.h>
      4#include <linux/percpu.h>
      5#include <linux/kallsyms.h>
      6#include <linux/kcore.h>
      7#include <linux/pgtable.h>
      8
      9#include <asm/cpu_entry_area.h>
     10#include <asm/fixmap.h>
     11#include <asm/desc.h>
     12
     13static DEFINE_PER_CPU_PAGE_ALIGNED(struct entry_stack_page, entry_stack_storage);
     14
     15#ifdef CONFIG_X86_64
     16static DEFINE_PER_CPU_PAGE_ALIGNED(struct exception_stacks, exception_stacks);
     17DEFINE_PER_CPU(struct cea_exception_stacks*, cea_exception_stacks);
     18#endif
     19
     20#ifdef CONFIG_X86_32
     21DECLARE_PER_CPU_PAGE_ALIGNED(struct doublefault_stack, doublefault_stack);
     22#endif
     23
     24/* Is called from entry code, so must be noinstr */
     25noinstr struct cpu_entry_area *get_cpu_entry_area(int cpu)
     26{
     27	unsigned long va = CPU_ENTRY_AREA_PER_CPU + cpu * CPU_ENTRY_AREA_SIZE;
     28	BUILD_BUG_ON(sizeof(struct cpu_entry_area) % PAGE_SIZE != 0);
     29
     30	return (struct cpu_entry_area *) va;
     31}
     32EXPORT_SYMBOL(get_cpu_entry_area);
     33
     34void cea_set_pte(void *cea_vaddr, phys_addr_t pa, pgprot_t flags)
     35{
     36	unsigned long va = (unsigned long) cea_vaddr;
     37	pte_t pte = pfn_pte(pa >> PAGE_SHIFT, flags);
     38
     39	/*
     40	 * The cpu_entry_area is shared between the user and kernel
     41	 * page tables.  All of its ptes can safely be global.
     42	 * _PAGE_GLOBAL gets reused to help indicate PROT_NONE for
     43	 * non-present PTEs, so be careful not to set it in that
     44	 * case to avoid confusion.
     45	 */
     46	if (boot_cpu_has(X86_FEATURE_PGE) &&
     47	    (pgprot_val(flags) & _PAGE_PRESENT))
     48		pte = pte_set_flags(pte, _PAGE_GLOBAL);
     49
     50	set_pte_vaddr(va, pte);
     51}
     52
     53static void __init
     54cea_map_percpu_pages(void *cea_vaddr, void *ptr, int pages, pgprot_t prot)
     55{
     56	for ( ; pages; pages--, cea_vaddr+= PAGE_SIZE, ptr += PAGE_SIZE)
     57		cea_set_pte(cea_vaddr, per_cpu_ptr_to_phys(ptr), prot);
     58}
     59
     60static void __init percpu_setup_debug_store(unsigned int cpu)
     61{
     62#ifdef CONFIG_CPU_SUP_INTEL
     63	unsigned int npages;
     64	void *cea;
     65
     66	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
     67		return;
     68
     69	cea = &get_cpu_entry_area(cpu)->cpu_debug_store;
     70	npages = sizeof(struct debug_store) / PAGE_SIZE;
     71	BUILD_BUG_ON(sizeof(struct debug_store) % PAGE_SIZE != 0);
     72	cea_map_percpu_pages(cea, &per_cpu(cpu_debug_store, cpu), npages,
     73			     PAGE_KERNEL);
     74
     75	cea = &get_cpu_entry_area(cpu)->cpu_debug_buffers;
     76	/*
     77	 * Force the population of PMDs for not yet allocated per cpu
     78	 * memory like debug store buffers.
     79	 */
     80	npages = sizeof(struct debug_store_buffers) / PAGE_SIZE;
     81	for (; npages; npages--, cea += PAGE_SIZE)
     82		cea_set_pte(cea, 0, PAGE_NONE);
     83#endif
     84}
     85
     86#ifdef CONFIG_X86_64
     87
     88#define cea_map_stack(name) do {					\
     89	npages = sizeof(estacks->name## _stack) / PAGE_SIZE;		\
     90	cea_map_percpu_pages(cea->estacks.name## _stack,		\
     91			estacks->name## _stack, npages, PAGE_KERNEL);	\
     92	} while (0)
     93
     94static void __init percpu_setup_exception_stacks(unsigned int cpu)
     95{
     96	struct exception_stacks *estacks = per_cpu_ptr(&exception_stacks, cpu);
     97	struct cpu_entry_area *cea = get_cpu_entry_area(cpu);
     98	unsigned int npages;
     99
    100	BUILD_BUG_ON(sizeof(exception_stacks) % PAGE_SIZE != 0);
    101
    102	per_cpu(cea_exception_stacks, cpu) = &cea->estacks;
    103
    104	/*
    105	 * The exceptions stack mappings in the per cpu area are protected
    106	 * by guard pages so each stack must be mapped separately. DB2 is
    107	 * not mapped; it just exists to catch triple nesting of #DB.
    108	 */
    109	cea_map_stack(DF);
    110	cea_map_stack(NMI);
    111	cea_map_stack(DB);
    112	cea_map_stack(MCE);
    113
    114	if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT)) {
    115		if (cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT)) {
    116			cea_map_stack(VC);
    117			cea_map_stack(VC2);
    118		}
    119	}
    120}
    121#else
    122static inline void percpu_setup_exception_stacks(unsigned int cpu)
    123{
    124	struct cpu_entry_area *cea = get_cpu_entry_area(cpu);
    125
    126	cea_map_percpu_pages(&cea->doublefault_stack,
    127			     &per_cpu(doublefault_stack, cpu), 1, PAGE_KERNEL);
    128}
    129#endif
    130
    131/* Setup the fixmap mappings only once per-processor */
    132static void __init setup_cpu_entry_area(unsigned int cpu)
    133{
    134	struct cpu_entry_area *cea = get_cpu_entry_area(cpu);
    135#ifdef CONFIG_X86_64
    136	/* On 64-bit systems, we use a read-only fixmap GDT and TSS. */
    137	pgprot_t gdt_prot = PAGE_KERNEL_RO;
    138	pgprot_t tss_prot = PAGE_KERNEL_RO;
    139#else
    140	/*
    141	 * On native 32-bit systems, the GDT cannot be read-only because
    142	 * our double fault handler uses a task gate, and entering through
    143	 * a task gate needs to change an available TSS to busy.  If the
    144	 * GDT is read-only, that will triple fault.  The TSS cannot be
    145	 * read-only because the CPU writes to it on task switches.
    146	 *
    147	 * On Xen PV, the GDT must be read-only because the hypervisor
    148	 * requires it.
    149	 */
    150	pgprot_t gdt_prot = boot_cpu_has(X86_FEATURE_XENPV) ?
    151		PAGE_KERNEL_RO : PAGE_KERNEL;
    152	pgprot_t tss_prot = PAGE_KERNEL;
    153#endif
    154
    155	cea_set_pte(&cea->gdt, get_cpu_gdt_paddr(cpu), gdt_prot);
    156
    157	cea_map_percpu_pages(&cea->entry_stack_page,
    158			     per_cpu_ptr(&entry_stack_storage, cpu), 1,
    159			     PAGE_KERNEL);
    160
    161	/*
    162	 * The Intel SDM says (Volume 3, 7.2.1):
    163	 *
    164	 *  Avoid placing a page boundary in the part of the TSS that the
    165	 *  processor reads during a task switch (the first 104 bytes). The
    166	 *  processor may not correctly perform address translations if a
    167	 *  boundary occurs in this area. During a task switch, the processor
    168	 *  reads and writes into the first 104 bytes of each TSS (using
    169	 *  contiguous physical addresses beginning with the physical address
    170	 *  of the first byte of the TSS). So, after TSS access begins, if
    171	 *  part of the 104 bytes is not physically contiguous, the processor
    172	 *  will access incorrect information without generating a page-fault
    173	 *  exception.
    174	 *
    175	 * There are also a lot of errata involving the TSS spanning a page
    176	 * boundary.  Assert that we're not doing that.
    177	 */
    178	BUILD_BUG_ON((offsetof(struct tss_struct, x86_tss) ^
    179		      offsetofend(struct tss_struct, x86_tss)) & PAGE_MASK);
    180	BUILD_BUG_ON(sizeof(struct tss_struct) % PAGE_SIZE != 0);
    181	/*
    182	 * VMX changes the host TR limit to 0x67 after a VM exit. This is
    183	 * okay, since 0x67 covers the size of struct x86_hw_tss. Make sure
    184	 * that this is correct.
    185	 */
    186	BUILD_BUG_ON(offsetof(struct tss_struct, x86_tss) != 0);
    187	BUILD_BUG_ON(sizeof(struct x86_hw_tss) != 0x68);
    188
    189	cea_map_percpu_pages(&cea->tss, &per_cpu(cpu_tss_rw, cpu),
    190			     sizeof(struct tss_struct) / PAGE_SIZE, tss_prot);
    191
    192#ifdef CONFIG_X86_32
    193	per_cpu(cpu_entry_area, cpu) = cea;
    194#endif
    195
    196	percpu_setup_exception_stacks(cpu);
    197
    198	percpu_setup_debug_store(cpu);
    199}
    200
    201static __init void setup_cpu_entry_area_ptes(void)
    202{
    203#ifdef CONFIG_X86_32
    204	unsigned long start, end;
    205
    206	/* The +1 is for the readonly IDT: */
    207	BUILD_BUG_ON((CPU_ENTRY_AREA_PAGES+1)*PAGE_SIZE != CPU_ENTRY_AREA_MAP_SIZE);
    208	BUILD_BUG_ON(CPU_ENTRY_AREA_TOTAL_SIZE != CPU_ENTRY_AREA_MAP_SIZE);
    209	BUG_ON(CPU_ENTRY_AREA_BASE & ~PMD_MASK);
    210
    211	start = CPU_ENTRY_AREA_BASE;
    212	end = start + CPU_ENTRY_AREA_MAP_SIZE;
    213
    214	/* Careful here: start + PMD_SIZE might wrap around */
    215	for (; start < end && start >= CPU_ENTRY_AREA_BASE; start += PMD_SIZE)
    216		populate_extra_pte(start);
    217#endif
    218}
    219
    220void __init setup_cpu_entry_areas(void)
    221{
    222	unsigned int cpu;
    223
    224	setup_cpu_entry_area_ptes();
    225
    226	for_each_possible_cpu(cpu)
    227		setup_cpu_entry_area(cpu);
    228
    229	/*
    230	 * This is the last essential update to swapper_pgdir which needs
    231	 * to be synchronized to initial_page_table on 32bit.
    232	 */
    233	sync_initial_page_table();
    234}