xref: /linux/arch/x86/kernel/fred.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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/tlbflush.h>
7 #include <asm/traps.h>
8 
9 /* #DB in the kernel would imply the use of a kernel debugger. */
10 #define FRED_DB_STACK_LEVEL		1UL
11 #define FRED_NMI_STACK_LEVEL		2UL
12 #define FRED_MC_STACK_LEVEL		2UL
13 /*
14  * #DF is the highest level because a #DF means "something went wrong
15  * *while delivering an exception*." The number of cases for which that
16  * can happen with FRED is drastically reduced and basically amounts to
17  * "the stack you pointed me to is broken." Thus, always change stacks
18  * on #DF, which means it should be at the highest level.
19  */
20 #define FRED_DF_STACK_LEVEL		3UL
21 
22 #define FRED_STKLVL(vector, lvl)	((lvl) << (2 * (vector)))
23 
24 DEFINE_PER_CPU(unsigned long, fred_rsp0);
25 EXPORT_PER_CPU_SYMBOL(fred_rsp0);
26 
27 void cpu_init_fred_exceptions(void)
28 {
29 	/* When FRED is enabled by default, remove this log message */
30 	pr_info("Initialize FRED on CPU%d\n", smp_processor_id());
31 
32 	/*
33 	 * If a kernel event is delivered before a CPU goes to user level for
34 	 * the first time, its SS is NULL thus NULL is pushed into the SS field
35 	 * of the FRED stack frame.  But before ERETS is executed, the CPU may
36 	 * context switch to another task and go to user level.  Then when the
37 	 * CPU comes back to kernel mode, SS is changed to __KERNEL_DS.  Later
38 	 * when ERETS is executed to return from the kernel event handler, a #GP
39 	 * fault is generated because SS doesn't match the SS saved in the FRED
40 	 * stack frame.
41 	 *
42 	 * Initialize SS to __KERNEL_DS when enabling FRED to avoid such #GPs.
43 	 */
44 	loadsegment(ss, __KERNEL_DS);
45 
46 	wrmsrl(MSR_IA32_FRED_CONFIG,
47 	       /* Reserve for CALL emulation */
48 	       FRED_CONFIG_REDZONE |
49 	       FRED_CONFIG_INT_STKLVL(0) |
50 	       FRED_CONFIG_ENTRYPOINT(asm_fred_entrypoint_user));
51 
52 	wrmsrl(MSR_IA32_FRED_STKLVLS, 0);
53 	wrmsrl(MSR_IA32_FRED_RSP0, 0);
54 	wrmsrl(MSR_IA32_FRED_RSP1, 0);
55 	wrmsrl(MSR_IA32_FRED_RSP2, 0);
56 	wrmsrl(MSR_IA32_FRED_RSP3, 0);
57 
58 	/* Enable FRED */
59 	cr4_set_bits(X86_CR4_FRED);
60 	/* Any further IDT use is a bug */
61 	idt_invalidate();
62 
63 	/* Use int $0x80 for 32-bit system calls in FRED mode */
64 	setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
65 	setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
66 }
67 
68 /* Must be called after setup_cpu_entry_areas() */
69 void cpu_init_fred_rsps(void)
70 {
71 	/*
72 	 * The purpose of separate stacks for NMI, #DB and #MC *in the kernel*
73 	 * (remember that user space faults are always taken on stack level 0)
74 	 * is to avoid overflowing the kernel stack.
75 	 */
76 	wrmsrl(MSR_IA32_FRED_STKLVLS,
77 	       FRED_STKLVL(X86_TRAP_DB,  FRED_DB_STACK_LEVEL) |
78 	       FRED_STKLVL(X86_TRAP_NMI, FRED_NMI_STACK_LEVEL) |
79 	       FRED_STKLVL(X86_TRAP_MC,  FRED_MC_STACK_LEVEL) |
80 	       FRED_STKLVL(X86_TRAP_DF,  FRED_DF_STACK_LEVEL));
81 
82 	/* The FRED equivalents to IST stacks... */
83 	wrmsrl(MSR_IA32_FRED_RSP1, __this_cpu_ist_top_va(DB));
84 	wrmsrl(MSR_IA32_FRED_RSP2, __this_cpu_ist_top_va(NMI));
85 	wrmsrl(MSR_IA32_FRED_RSP3, __this_cpu_ist_top_va(DF));
86 }
87