xref: /linux/arch/powerpc/kernel/syscall.c (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 
3 #include <linux/compat.h>
4 #include <linux/context_tracking.h>
5 #include <linux/randomize_kstack.h>
6 
7 #include <asm/interrupt.h>
8 #include <asm/kup.h>
9 #include <asm/syscall.h>
10 #include <asm/time.h>
11 #include <asm/tm.h>
12 #include <asm/unistd.h>
13 
14 
15 /* Has to run notrace because it is entered not completely "reconciled" */
16 notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
17 {
18 	long ret;
19 	syscall_fn f;
20 
21 	kuap_lock();
22 
23 	add_random_kstack_offset();
24 
25 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
26 		BUG_ON(irq_soft_mask_return() != IRQS_ALL_DISABLED);
27 
28 	trace_hardirqs_off(); /* finish reconciling */
29 
30 	CT_WARN_ON(ct_state() == CONTEXT_KERNEL);
31 	user_exit_irqoff();
32 
33 	BUG_ON(regs_is_unrecoverable(regs));
34 	BUG_ON(!user_mode(regs));
35 	BUG_ON(arch_irq_disabled_regs(regs));
36 
37 #ifdef CONFIG_PPC_PKEY
38 	if (mmu_has_feature(MMU_FTR_PKEY)) {
39 		unsigned long amr, iamr;
40 		bool flush_needed = false;
41 		/*
42 		 * When entering from userspace we mostly have the AMR/IAMR
43 		 * different from kernel default values. Hence don't compare.
44 		 */
45 		amr = mfspr(SPRN_AMR);
46 		iamr = mfspr(SPRN_IAMR);
47 		regs->amr  = amr;
48 		regs->iamr = iamr;
49 		if (mmu_has_feature(MMU_FTR_KUAP)) {
50 			mtspr(SPRN_AMR, AMR_KUAP_BLOCKED);
51 			flush_needed = true;
52 		}
53 		if (mmu_has_feature(MMU_FTR_BOOK3S_KUEP)) {
54 			mtspr(SPRN_IAMR, AMR_KUEP_BLOCKED);
55 			flush_needed = true;
56 		}
57 		if (flush_needed)
58 			isync();
59 	} else
60 #endif
61 		kuap_assert_locked();
62 
63 	booke_restore_dbcr0();
64 
65 	account_cpu_user_entry();
66 
67 	account_stolen_time();
68 
69 	/*
70 	 * This is not required for the syscall exit path, but makes the
71 	 * stack frame look nicer. If this was initialised in the first stack
72 	 * frame, or if the unwinder was taught the first stack frame always
73 	 * returns to user with IRQS_ENABLED, this store could be avoided!
74 	 */
75 	irq_soft_mask_regs_set_state(regs, IRQS_ENABLED);
76 
77 	/*
78 	 * If system call is called with TM active, set _TIF_RESTOREALL to
79 	 * prevent RFSCV being used to return to userspace, because POWER9
80 	 * TM implementation has problems with this instruction returning to
81 	 * transactional state. Final register values are not relevant because
82 	 * the transaction will be aborted upon return anyway. Or in the case
83 	 * of unsupported_scv SIGILL fault, the return state does not much
84 	 * matter because it's an edge case.
85 	 */
86 	if (IS_ENABLED(CONFIG_PPC_TRANSACTIONAL_MEM) &&
87 			unlikely(MSR_TM_TRANSACTIONAL(regs->msr)))
88 		set_bits(_TIF_RESTOREALL, &current_thread_info()->flags);
89 
90 	/*
91 	 * If the system call was made with a transaction active, doom it and
92 	 * return without performing the system call. Unless it was an
93 	 * unsupported scv vector, in which case it's treated like an illegal
94 	 * instruction.
95 	 */
96 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
97 	if (unlikely(MSR_TM_TRANSACTIONAL(regs->msr)) &&
98 	    !trap_is_unsupported_scv(regs)) {
99 		/* Enable TM in the kernel, and disable EE (for scv) */
100 		hard_irq_disable();
101 		mtmsr(mfmsr() | MSR_TM);
102 
103 		/* tabort, this dooms the transaction, nothing else */
104 		asm volatile(".long 0x7c00071d | ((%0) << 16)"
105 				:: "r"(TM_CAUSE_SYSCALL|TM_CAUSE_PERSISTENT));
106 
107 		/*
108 		 * Userspace will never see the return value. Execution will
109 		 * resume after the tbegin. of the aborted transaction with the
110 		 * checkpointed register state. A context switch could occur
111 		 * or signal delivered to the process before resuming the
112 		 * doomed transaction context, but that should all be handled
113 		 * as expected.
114 		 */
115 		return -ENOSYS;
116 	}
117 #endif // CONFIG_PPC_TRANSACTIONAL_MEM
118 
119 	local_irq_enable();
120 
121 	if (unlikely(read_thread_flags() & _TIF_SYSCALL_DOTRACE)) {
122 		if (unlikely(trap_is_unsupported_scv(regs))) {
123 			/* Unsupported scv vector */
124 			_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
125 			return regs->gpr[3];
126 		}
127 		/*
128 		 * We use the return value of do_syscall_trace_enter() as the
129 		 * syscall number. If the syscall was rejected for any reason
130 		 * do_syscall_trace_enter() returns an invalid syscall number
131 		 * and the test against NR_syscalls will fail and the return
132 		 * value to be used is in regs->gpr[3].
133 		 */
134 		r0 = do_syscall_trace_enter(regs);
135 		if (unlikely(r0 >= NR_syscalls))
136 			return regs->gpr[3];
137 
138 	} else if (unlikely(r0 >= NR_syscalls)) {
139 		if (unlikely(trap_is_unsupported_scv(regs))) {
140 			/* Unsupported scv vector */
141 			_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
142 			return regs->gpr[3];
143 		}
144 		return -ENOSYS;
145 	}
146 
147 	/* May be faster to do array_index_nospec? */
148 	barrier_nospec();
149 
150 #ifdef CONFIG_ARCH_HAS_SYSCALL_WRAPPER
151 	// No COMPAT if we have SYSCALL_WRAPPER, see Kconfig
152 	f = (void *)sys_call_table[r0];
153 	ret = f(regs);
154 #else
155 	if (unlikely(is_compat_task())) {
156 		unsigned long r3, r4, r5, r6, r7, r8;
157 
158 		f = (void *)compat_sys_call_table[r0];
159 
160 		r3 = regs->gpr[3] & 0x00000000ffffffffULL;
161 		r4 = regs->gpr[4] & 0x00000000ffffffffULL;
162 		r5 = regs->gpr[5] & 0x00000000ffffffffULL;
163 		r6 = regs->gpr[6] & 0x00000000ffffffffULL;
164 		r7 = regs->gpr[7] & 0x00000000ffffffffULL;
165 		r8 = regs->gpr[8] & 0x00000000ffffffffULL;
166 
167 		ret = f(r3, r4, r5, r6, r7, r8);
168 	} else {
169 		f = (void *)sys_call_table[r0];
170 
171 		ret = f(regs->gpr[3], regs->gpr[4], regs->gpr[5],
172 			regs->gpr[6], regs->gpr[7], regs->gpr[8]);
173 	}
174 #endif
175 
176 	/*
177 	 * Ultimately, this value will get limited by KSTACK_OFFSET_MAX(),
178 	 * so the maximum stack offset is 1k bytes (10 bits).
179 	 *
180 	 * The actual entropy will be further reduced by the compiler when
181 	 * applying stack alignment constraints: the powerpc architecture
182 	 * may have two kinds of stack alignment (16-bytes and 8-bytes).
183 	 *
184 	 * So the resulting 6 or 7 bits of entropy is seen in SP[9:4] or SP[9:3].
185 	 */
186 	choose_random_kstack_offset(mftb());
187 
188 	return ret;
189 }
190