1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright IBM Corp. 1999, 2023
4  */
5 
6 #include <linux/cpuhotplug.h>
7 #include <linux/sched/task.h>
8 #include <linux/errno.h>
9 #include <linux/init.h>
10 #include <linux/irq.h>
11 #include <asm/asm-extable.h>
12 #include <asm/asm-offsets.h>
13 #include <asm/pfault.h>
14 #include <asm/diag.h>
15 
16 #define __SUBCODE_MASK 0x0600
17 #define __PF_RES_FIELD 0x8000000000000000UL
18 
19 /*
20  * 'pfault' pseudo page faults routines.
21  */
22 static int pfault_disable;
23 
24 static int __init nopfault(char *str)
25 {
26 	pfault_disable = 1;
27 	return 1;
28 }
29 early_param("nopfault", nopfault);
30 
31 struct pfault_refbk {
32 	u16 refdiagc;
33 	u16 reffcode;
34 	u16 refdwlen;
35 	u16 refversn;
36 	u64 refgaddr;
37 	u64 refselmk;
38 	u64 refcmpmk;
39 	u64 reserved;
40 };
41 
42 static struct pfault_refbk pfault_init_refbk = {
43 	.refdiagc = 0x258,
44 	.reffcode = 0,
45 	.refdwlen = 5,
46 	.refversn = 2,
47 	.refgaddr = __LC_LPP,
48 	.refselmk = 1UL << 48,
49 	.refcmpmk = 1UL << 48,
50 	.reserved = __PF_RES_FIELD
51 };
52 
53 int __pfault_init(void)
54 {
55 	int rc = -EOPNOTSUPP;
56 
57 	if (pfault_disable)
58 		return rc;
59 	diag_stat_inc(DIAG_STAT_X258);
60 	asm_inline volatile(
61 		"	diag	%[refbk],%[rc],0x258\n"
62 		"0:	nopr	%%r7\n"
63 		EX_TABLE(0b, 0b)
64 		: [rc] "+d" (rc)
65 		: [refbk] "a" (&pfault_init_refbk), "m" (pfault_init_refbk)
66 		: "cc");
67 	return rc;
68 }
69 
70 static struct pfault_refbk pfault_fini_refbk = {
71 	.refdiagc = 0x258,
72 	.reffcode = 1,
73 	.refdwlen = 5,
74 	.refversn = 2,
75 };
76 
77 void __pfault_fini(void)
78 {
79 	if (pfault_disable)
80 		return;
81 	diag_stat_inc(DIAG_STAT_X258);
82 	asm_inline volatile(
83 		"	diag	%[refbk],0,0x258\n"
84 		"0:	nopr	%%r7\n"
85 		EX_TABLE(0b, 0b)
86 		:
87 		: [refbk] "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk)
88 		: "cc");
89 }
90 
91 static DEFINE_SPINLOCK(pfault_lock);
92 static LIST_HEAD(pfault_list);
93 
94 #define PF_COMPLETE	0x0080
95 
96 /*
97  * The mechanism of our pfault code: if Linux is running as guest, runs a user
98  * space process and the user space process accesses a page that the host has
99  * paged out we get a pfault interrupt.
100  *
101  * This allows us, within the guest, to schedule a different process. Without
102  * this mechanism the host would have to suspend the whole virtual cpu until
103  * the page has been paged in.
104  *
105  * So when we get such an interrupt then we set the state of the current task
106  * to uninterruptible and also set the need_resched flag. Both happens within
107  * interrupt context(!). If we later on want to return to user space we
108  * recognize the need_resched flag and then call schedule().  It's not very
109  * obvious how this works...
110  *
111  * Of course we have a lot of additional fun with the completion interrupt (->
112  * host signals that a page of a process has been paged in and the process can
113  * continue to run). This interrupt can arrive on any cpu and, since we have
114  * virtual cpus, actually appear before the interrupt that signals that a page
115  * is missing.
116  */
117 static void pfault_interrupt(struct ext_code ext_code,
118 			     unsigned int param32, unsigned long param64)
119 {
120 	struct task_struct *tsk;
121 	__u16 subcode;
122 	pid_t pid;
123 
124 	/*
125 	 * Get the external interruption subcode & pfault initial/completion
126 	 * signal bit. VM stores this in the 'cpu address' field associated
127 	 * with the external interrupt.
128 	 */
129 	subcode = ext_code.subcode;
130 	if ((subcode & 0xff00) != __SUBCODE_MASK)
131 		return;
132 	inc_irq_stat(IRQEXT_PFL);
133 	/* Get the token (= pid of the affected task). */
134 	pid = param64 & LPP_PID_MASK;
135 	rcu_read_lock();
136 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
137 	if (tsk)
138 		get_task_struct(tsk);
139 	rcu_read_unlock();
140 	if (!tsk)
141 		return;
142 	spin_lock(&pfault_lock);
143 	if (subcode & PF_COMPLETE) {
144 		/* signal bit is set -> a page has been swapped in by VM */
145 		if (tsk->thread.pfault_wait == 1) {
146 			/*
147 			 * Initial interrupt was faster than the completion
148 			 * interrupt. pfault_wait is valid. Set pfault_wait
149 			 * back to zero and wake up the process. This can
150 			 * safely be done because the task is still sleeping
151 			 * and can't produce new pfaults.
152 			 */
153 			tsk->thread.pfault_wait = 0;
154 			list_del(&tsk->thread.list);
155 			wake_up_process(tsk);
156 			put_task_struct(tsk);
157 		} else {
158 			/*
159 			 * Completion interrupt was faster than initial
160 			 * interrupt. Set pfault_wait to -1 so the initial
161 			 * interrupt doesn't put the task to sleep.
162 			 * If the task is not running, ignore the completion
163 			 * interrupt since it must be a leftover of a PFAULT
164 			 * CANCEL operation which didn't remove all pending
165 			 * completion interrupts.
166 			 */
167 			if (task_is_running(tsk))
168 				tsk->thread.pfault_wait = -1;
169 		}
170 	} else {
171 		/* signal bit not set -> a real page is missing. */
172 		if (WARN_ON_ONCE(tsk != current))
173 			goto out;
174 		if (tsk->thread.pfault_wait == 1) {
175 			/* Already on the list with a reference: put to sleep */
176 			goto block;
177 		} else if (tsk->thread.pfault_wait == -1) {
178 			/*
179 			 * Completion interrupt was faster than the initial
180 			 * interrupt (pfault_wait == -1). Set pfault_wait
181 			 * back to zero and exit.
182 			 */
183 			tsk->thread.pfault_wait = 0;
184 		} else {
185 			/*
186 			 * Initial interrupt arrived before completion
187 			 * interrupt. Let the task sleep.
188 			 * An extra task reference is needed since a different
189 			 * cpu may set the task state to TASK_RUNNING again
190 			 * before the scheduler is reached.
191 			 */
192 			get_task_struct(tsk);
193 			tsk->thread.pfault_wait = 1;
194 			list_add(&tsk->thread.list, &pfault_list);
195 block:
196 			/*
197 			 * Since this must be a userspace fault, there
198 			 * is no kernel task state to trample. Rely on the
199 			 * return to userspace schedule() to block.
200 			 */
201 			__set_current_state(TASK_UNINTERRUPTIBLE);
202 			set_tsk_need_resched(tsk);
203 			set_preempt_need_resched();
204 		}
205 	}
206 out:
207 	spin_unlock(&pfault_lock);
208 	put_task_struct(tsk);
209 }
210 
211 static int pfault_cpu_dead(unsigned int cpu)
212 {
213 	struct thread_struct *thread, *next;
214 	struct task_struct *tsk;
215 
216 	spin_lock_irq(&pfault_lock);
217 	list_for_each_entry_safe(thread, next, &pfault_list, list) {
218 		thread->pfault_wait = 0;
219 		list_del(&thread->list);
220 		tsk = container_of(thread, struct task_struct, thread);
221 		wake_up_process(tsk);
222 		put_task_struct(tsk);
223 	}
224 	spin_unlock_irq(&pfault_lock);
225 	return 0;
226 }
227 
228 static int __init pfault_irq_init(void)
229 {
230 	int rc;
231 
232 	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
233 	if (rc)
234 		goto out_extint;
235 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
236 	if (rc)
237 		goto out_pfault;
238 	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
239 	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
240 				  NULL, pfault_cpu_dead);
241 	return 0;
242 
243 out_pfault:
244 	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
245 out_extint:
246 	pfault_disable = 1;
247 	return rc;
248 }
249 early_initcall(pfault_irq_init);
250