1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | #include <linux/kernel.h> |
3 | |
4 | #include <asm/desc.h> |
5 | #include <asm/fred.h> |
6 | #include <asm/msr.h> |
7 | #include <asm/tlbflush.h> |
8 | #include <asm/traps.h> |
9 | |
10 | /* #DB in the kernel would imply the use of a kernel debugger. */ |
11 | #define FRED_DB_STACK_LEVEL 1UL |
12 | #define FRED_NMI_STACK_LEVEL 2UL |
13 | #define FRED_MC_STACK_LEVEL 2UL |
14 | /* |
15 | * #DF is the highest level because a #DF means "something went wrong |
16 | * *while delivering an exception*." The number of cases for which that |
17 | * can happen with FRED is drastically reduced and basically amounts to |
18 | * "the stack you pointed me to is broken." Thus, always change stacks |
19 | * on #DF, which means it should be at the highest level. |
20 | */ |
21 | #define FRED_DF_STACK_LEVEL 3UL |
22 | |
23 | #define FRED_STKLVL(vector, lvl) ((lvl) << (2 * (vector))) |
24 | |
25 | DEFINE_PER_CPU(unsigned long, fred_rsp0); |
26 | EXPORT_PER_CPU_SYMBOL(fred_rsp0); |
27 | |
28 | void cpu_init_fred_exceptions(void) |
29 | { |
30 | /* When FRED is enabled by default, remove this log message */ |
31 | pr_info("Initialize FRED on CPU%d\n" , smp_processor_id()); |
32 | |
33 | /* |
34 | * If a kernel event is delivered before a CPU goes to user level for |
35 | * the first time, its SS is NULL thus NULL is pushed into the SS field |
36 | * of the FRED stack frame. But before ERETS is executed, the CPU may |
37 | * context switch to another task and go to user level. Then when the |
38 | * CPU comes back to kernel mode, SS is changed to __KERNEL_DS. Later |
39 | * when ERETS is executed to return from the kernel event handler, a #GP |
40 | * fault is generated because SS doesn't match the SS saved in the FRED |
41 | * stack frame. |
42 | * |
43 | * Initialize SS to __KERNEL_DS when enabling FRED to avoid such #GPs. |
44 | */ |
45 | loadsegment(ss, __KERNEL_DS); |
46 | |
47 | wrmsrq(MSR_IA32_FRED_CONFIG, |
48 | /* Reserve for CALL emulation */ |
49 | FRED_CONFIG_REDZONE | |
50 | FRED_CONFIG_INT_STKLVL(0) | |
51 | FRED_CONFIG_ENTRYPOINT(asm_fred_entrypoint_user)); |
52 | |
53 | wrmsrq(MSR_IA32_FRED_STKLVLS, val: 0); |
54 | |
55 | /* |
56 | * Ater a CPU offline/online cycle, the FRED RSP0 MSR should be |
57 | * resynchronized with its per-CPU cache. |
58 | */ |
59 | wrmsrq(MSR_IA32_FRED_RSP0, __this_cpu_read(fred_rsp0)); |
60 | |
61 | wrmsrq(MSR_IA32_FRED_RSP1, val: 0); |
62 | wrmsrq(MSR_IA32_FRED_RSP2, val: 0); |
63 | wrmsrq(MSR_IA32_FRED_RSP3, val: 0); |
64 | |
65 | /* Enable FRED */ |
66 | cr4_set_bits(X86_CR4_FRED); |
67 | /* Any further IDT use is a bug */ |
68 | idt_invalidate(); |
69 | |
70 | /* Use int $0x80 for 32-bit system calls in FRED mode */ |
71 | setup_clear_cpu_cap(X86_FEATURE_SYSENTER32); |
72 | setup_clear_cpu_cap(X86_FEATURE_SYSCALL32); |
73 | } |
74 | |
75 | /* Must be called after setup_cpu_entry_areas() */ |
76 | void cpu_init_fred_rsps(void) |
77 | { |
78 | /* |
79 | * The purpose of separate stacks for NMI, #DB and #MC *in the kernel* |
80 | * (remember that user space faults are always taken on stack level 0) |
81 | * is to avoid overflowing the kernel stack. |
82 | */ |
83 | wrmsrq(MSR_IA32_FRED_STKLVLS, |
84 | FRED_STKLVL(X86_TRAP_DB, FRED_DB_STACK_LEVEL) | |
85 | FRED_STKLVL(X86_TRAP_NMI, FRED_NMI_STACK_LEVEL) | |
86 | FRED_STKLVL(X86_TRAP_MC, FRED_MC_STACK_LEVEL) | |
87 | FRED_STKLVL(X86_TRAP_DF, FRED_DF_STACK_LEVEL)); |
88 | |
89 | /* The FRED equivalents to IST stacks... */ |
90 | wrmsrq(MSR_IA32_FRED_RSP1, __this_cpu_ist_top_va(DB)); |
91 | wrmsrq(MSR_IA32_FRED_RSP2, __this_cpu_ist_top_va(NMI)); |
92 | wrmsrq(MSR_IA32_FRED_RSP3, __this_cpu_ist_top_va(DF)); |
93 | } |
94 | |