xref: /freebsd/sys/dev/amd_ecc_inject/ecc_inject.c (revision 95ee2897e98f5d444f26ed2334cc7c439f9c16c6)
1f5aac907SAndriy Gapon /*-
2f5aac907SAndriy Gapon  * Copyright (c) 2017 Andriy Gapon
3f5aac907SAndriy Gapon  * All rights reserved.
4f5aac907SAndriy Gapon  *
5f5aac907SAndriy Gapon  * Redistribution and use in source and binary forms, with or without
6f5aac907SAndriy Gapon  * modification, are permitted provided that the following conditions
7f5aac907SAndriy Gapon  * are met:
8f5aac907SAndriy Gapon  * 1. Redistributions of source code must retain the above copyright
9f5aac907SAndriy Gapon  *    notice, this list of conditions and the following disclaimer.
10f5aac907SAndriy Gapon  * 2. Redistributions in binary form must reproduce the above copyright
11f5aac907SAndriy Gapon  *    notice, this list of conditions and the following disclaimer in the
12f5aac907SAndriy Gapon  *    documentation and/or other materials provided with the distribution.
13f5aac907SAndriy Gapon  *
14f5aac907SAndriy Gapon  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15f5aac907SAndriy Gapon  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16f5aac907SAndriy Gapon  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17f5aac907SAndriy Gapon  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18f5aac907SAndriy Gapon  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19f5aac907SAndriy Gapon  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20f5aac907SAndriy Gapon  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21f5aac907SAndriy Gapon  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22f5aac907SAndriy Gapon  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23f5aac907SAndriy Gapon  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24f5aac907SAndriy Gapon  * SUCH DAMAGE.
25f5aac907SAndriy Gapon  */
26f5aac907SAndriy Gapon 
27f5aac907SAndriy Gapon #include <sys/param.h>
28f5aac907SAndriy Gapon #include <sys/systm.h>
29f5aac907SAndriy Gapon #include <sys/bus.h>
30f5aac907SAndriy Gapon #include <sys/kernel.h>
31f5aac907SAndriy Gapon #include <sys/conf.h>
32f5aac907SAndriy Gapon #include <sys/malloc.h>
33f5aac907SAndriy Gapon #include <sys/module.h>
34f5aac907SAndriy Gapon #include <sys/sysctl.h>
35f5aac907SAndriy Gapon #include <sys/types.h>
36f5aac907SAndriy Gapon 
37f5aac907SAndriy Gapon #include <dev/pci/pcivar.h>
38f5aac907SAndriy Gapon 
39f5aac907SAndriy Gapon #include <vm/vm.h>
40f5aac907SAndriy Gapon #include <vm/vm_extern.h>
41f5aac907SAndriy Gapon #include <vm/vm_kern.h>
42f5aac907SAndriy Gapon 
43f5aac907SAndriy Gapon #include <machine/cputypes.h>
44f5aac907SAndriy Gapon #include <machine/md_var.h>
45f5aac907SAndriy Gapon 
46f5aac907SAndriy Gapon /*
47f5aac907SAndriy Gapon  * See BKDG for AMD Family 15h Models 00h-0Fh Processors
48f5aac907SAndriy Gapon  * (publication 42301 Rev 3.08 - March 12, 2012):
49f5aac907SAndriy Gapon  * - 2.13.3.1 DRAM Error Injection
50f5aac907SAndriy Gapon  * - D18F3xB8 NB Array Address
51f5aac907SAndriy Gapon  * - D18F3xBC NB Array Data Port
52f5aac907SAndriy Gapon  * - D18F3xBC_x8 DRAM ECC
53f5aac907SAndriy Gapon  */
54f5aac907SAndriy Gapon #define	NB_MCA_CFG		0x44
55f5aac907SAndriy Gapon #define		DRAM_ECC_EN	(1 << 22)
56f5aac907SAndriy Gapon #define	NB_MCA_EXTCFG		0x180
57f5aac907SAndriy Gapon #define		ECC_SYMB_SZ	(1 << 25)
58f5aac907SAndriy Gapon #define	NB_ARRAY_ADDR		0xb8
59f5aac907SAndriy Gapon #define		DRAM_ECC_SEL	(0x8 << 28)
60f5aac907SAndriy Gapon #define		QUADRANT_SHIFT	1
61f5aac907SAndriy Gapon #define		QUADRANT_MASK	0x3
62f5aac907SAndriy Gapon #define	NB_ARRAY_PORT		0xbc
63f5aac907SAndriy Gapon #define		INJ_WORD_SHIFT	20
64f5aac907SAndriy Gapon #define		INJ_WORD_MASK	0x1ff
65f5aac907SAndriy Gapon #define		DRAM_ERR_EN	(1 << 18)
66f5aac907SAndriy Gapon #define		DRAM_WR_REQ	(1 << 17)
67f5aac907SAndriy Gapon #define		DRAM_RD_REQ	(1 << 16)
68f5aac907SAndriy Gapon #define		INJ_VECTOR_MASK	0xffff
69f5aac907SAndriy Gapon 
70f5aac907SAndriy Gapon static void ecc_ei_inject(int);
71f5aac907SAndriy Gapon 
72f5aac907SAndriy Gapon static device_t nbdev;
73f5aac907SAndriy Gapon static int delay_ms = 0;
74f5aac907SAndriy Gapon static int quadrant = 0;	/* 0 - 3 */
75f5aac907SAndriy Gapon static int word_mask = 0x001;	/* 9 bits: 8 + 1 for ECC */
76f5aac907SAndriy Gapon static int bit_mask = 0x0001;	/* 16 bits */
77f5aac907SAndriy Gapon 
78f5aac907SAndriy Gapon static int
sysctl_int_with_max(SYSCTL_HANDLER_ARGS)79f5aac907SAndriy Gapon sysctl_int_with_max(SYSCTL_HANDLER_ARGS)
80f5aac907SAndriy Gapon {
81f5aac907SAndriy Gapon 	u_int value;
82f5aac907SAndriy Gapon 	int error;
83f5aac907SAndriy Gapon 
84f5aac907SAndriy Gapon 	value = *(u_int *)arg1;
85f5aac907SAndriy Gapon 	error = sysctl_handle_int(oidp, &value, 0, req);
86f5aac907SAndriy Gapon 	if (error || req->newptr == NULL)
87f5aac907SAndriy Gapon 		return (error);
88f5aac907SAndriy Gapon 	if (value > arg2)
89f5aac907SAndriy Gapon 		return (EINVAL);
90f5aac907SAndriy Gapon 	*(u_int *)arg1 = value;
91f5aac907SAndriy Gapon 	return (0);
92f5aac907SAndriy Gapon }
93f5aac907SAndriy Gapon 
94f5aac907SAndriy Gapon static int
sysctl_nonzero_int_with_max(SYSCTL_HANDLER_ARGS)95f5aac907SAndriy Gapon sysctl_nonzero_int_with_max(SYSCTL_HANDLER_ARGS)
96f5aac907SAndriy Gapon {
97f5aac907SAndriy Gapon 	u_int value;
98f5aac907SAndriy Gapon 	int error;
99f5aac907SAndriy Gapon 
100f5aac907SAndriy Gapon 	value = *(u_int *)arg1;
101f5aac907SAndriy Gapon 	error = sysctl_int_with_max(oidp, &value, arg2, req);
102f5aac907SAndriy Gapon 	if (error || req->newptr == NULL)
103f5aac907SAndriy Gapon 		return (error);
104f5aac907SAndriy Gapon 	if (value == 0)
105f5aac907SAndriy Gapon 		return (EINVAL);
106f5aac907SAndriy Gapon 	*(u_int *)arg1 = value;
107f5aac907SAndriy Gapon 	return (0);
108f5aac907SAndriy Gapon }
109f5aac907SAndriy Gapon 
110f5aac907SAndriy Gapon static int
sysctl_proc_inject(SYSCTL_HANDLER_ARGS)111f5aac907SAndriy Gapon sysctl_proc_inject(SYSCTL_HANDLER_ARGS)
112f5aac907SAndriy Gapon {
113f5aac907SAndriy Gapon 	int error;
114f5aac907SAndriy Gapon 	int i;
115f5aac907SAndriy Gapon 
116f5aac907SAndriy Gapon 	i = 0;
117f5aac907SAndriy Gapon 	error = sysctl_handle_int(oidp, &i, 0, req);
118f5aac907SAndriy Gapon 	if (error)
119f5aac907SAndriy Gapon 		return (error);
120f5aac907SAndriy Gapon 	if (i != 0)
121f5aac907SAndriy Gapon 		ecc_ei_inject(i);
122f5aac907SAndriy Gapon 	return (0);
123f5aac907SAndriy Gapon }
124f5aac907SAndriy Gapon 
1257029da5cSPawel Biernacki static SYSCTL_NODE(_hw, OID_AUTO, error_injection,
1267029da5cSPawel Biernacki     CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
127f5aac907SAndriy Gapon     "Hardware error injection");
1287029da5cSPawel Biernacki static SYSCTL_NODE(_hw_error_injection, OID_AUTO, dram_ecc,
1297029da5cSPawel Biernacki     CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
130f5aac907SAndriy Gapon     "DRAM ECC error injection");
131f5aac907SAndriy Gapon SYSCTL_UINT(_hw_error_injection_dram_ecc, OID_AUTO, delay,
132f5aac907SAndriy Gapon     CTLTYPE_UINT | CTLFLAG_RW, &delay_ms, 0,
133f5aac907SAndriy Gapon     "Delay in milliseconds between error injections");
134f5aac907SAndriy Gapon SYSCTL_PROC(_hw_error_injection_dram_ecc, OID_AUTO, quadrant,
1357029da5cSPawel Biernacki     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &quadrant, QUADRANT_MASK,
136f5aac907SAndriy Gapon     sysctl_int_with_max, "IU",
137f5aac907SAndriy Gapon     "Index of 16-byte quadrant within 64-byte line where errors "
138f5aac907SAndriy Gapon     "should be injected");
139f5aac907SAndriy Gapon SYSCTL_PROC(_hw_error_injection_dram_ecc, OID_AUTO, word_mask,
1407029da5cSPawel Biernacki     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &word_mask, INJ_WORD_MASK,
141f5aac907SAndriy Gapon     sysctl_nonzero_int_with_max, "IU",
142f5aac907SAndriy Gapon     "9-bit mask of words where errors should be injected (8 data + 1 ECC)");
143f5aac907SAndriy Gapon SYSCTL_PROC(_hw_error_injection_dram_ecc, OID_AUTO, bit_mask,
1447029da5cSPawel Biernacki     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &bit_mask, INJ_VECTOR_MASK,
145f5aac907SAndriy Gapon     sysctl_nonzero_int_with_max, "IU",
146f5aac907SAndriy Gapon     "16-bit mask of bits within each selected word where errors "
147f5aac907SAndriy Gapon     "should be injected");
148f5aac907SAndriy Gapon SYSCTL_PROC(_hw_error_injection_dram_ecc, OID_AUTO, inject,
149f5aac907SAndriy Gapon     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, sysctl_proc_inject, "I",
150f5aac907SAndriy Gapon     "Inject a number of errors according to configured parameters");
151f5aac907SAndriy Gapon 
152f5aac907SAndriy Gapon static void
ecc_ei_inject_one(void * arg,size_t size)153f5aac907SAndriy Gapon ecc_ei_inject_one(void *arg, size_t size)
154f5aac907SAndriy Gapon {
155f5aac907SAndriy Gapon 	volatile uint64_t *memory = arg;
156f5aac907SAndriy Gapon 	uint32_t val;
157f5aac907SAndriy Gapon 	int i;
158f5aac907SAndriy Gapon 
159f5aac907SAndriy Gapon 	val = DRAM_ECC_SEL | (quadrant << QUADRANT_SHIFT);
160f5aac907SAndriy Gapon 	pci_write_config(nbdev, NB_ARRAY_ADDR, val, 4);
161f5aac907SAndriy Gapon 
162f5aac907SAndriy Gapon 	val = (word_mask << INJ_WORD_SHIFT) | DRAM_WR_REQ | bit_mask;
163f5aac907SAndriy Gapon 	pci_write_config(nbdev, NB_ARRAY_PORT, val, 4);
164f5aac907SAndriy Gapon 
165f5aac907SAndriy Gapon 	for (i = 0; i < size / sizeof(uint64_t); i++) {
166f5aac907SAndriy Gapon 		memory[i] = 0;
167f5aac907SAndriy Gapon 		val = pci_read_config(nbdev, NB_ARRAY_PORT, 4);
168f5aac907SAndriy Gapon 		if ((val & DRAM_WR_REQ) == 0)
169f5aac907SAndriy Gapon 			break;
170f5aac907SAndriy Gapon 	}
171f5aac907SAndriy Gapon 	for (i = 0; i < size / sizeof(uint64_t); i++)
172f5aac907SAndriy Gapon 		memory[0] = memory[i];
173f5aac907SAndriy Gapon }
174f5aac907SAndriy Gapon 
175f5aac907SAndriy Gapon static void
ecc_ei_inject(int count)176f5aac907SAndriy Gapon ecc_ei_inject(int count)
177f5aac907SAndriy Gapon {
178*f49fd63aSJohn Baldwin 	void *memory;
179f5aac907SAndriy Gapon 	int injected;
180f5aac907SAndriy Gapon 
181f5aac907SAndriy Gapon 	KASSERT((quadrant & ~QUADRANT_MASK) == 0,
182f5aac907SAndriy Gapon 	    ("quadrant value is outside of range: %u", quadrant));
183f5aac907SAndriy Gapon 	KASSERT(word_mask != 0 && (word_mask & ~INJ_WORD_MASK) == 0,
184f5aac907SAndriy Gapon 	    ("word mask value is outside of range: 0x%x", word_mask));
185f5aac907SAndriy Gapon 	KASSERT(bit_mask != 0 && (bit_mask & ~INJ_VECTOR_MASK) == 0,
186f5aac907SAndriy Gapon 	    ("bit mask value is outside of range: 0x%x", bit_mask));
187f5aac907SAndriy Gapon 
188db7c2a48SAlan Cox 	memory = kmem_alloc_attr(PAGE_SIZE, M_WAITOK, 0, ~0,
189f5aac907SAndriy Gapon 	    VM_MEMATTR_UNCACHEABLE);
190f5aac907SAndriy Gapon 
191f5aac907SAndriy Gapon 	for (injected = 0; injected < count; injected++) {
192*f49fd63aSJohn Baldwin 		ecc_ei_inject_one(memory, PAGE_SIZE);
193f5aac907SAndriy Gapon 		if (delay_ms != 0 && injected != count - 1)
194f5aac907SAndriy Gapon 			pause_sbt("ecc_ei_inject", delay_ms * SBT_1MS, 0, 0);
195f5aac907SAndriy Gapon 	}
196f5aac907SAndriy Gapon 
19749bfa624SAlan Cox 	kmem_free(memory, PAGE_SIZE);
198f5aac907SAndriy Gapon }
199f5aac907SAndriy Gapon 
200f5aac907SAndriy Gapon static int
ecc_ei_load(void)201f5aac907SAndriy Gapon ecc_ei_load(void)
202f5aac907SAndriy Gapon {
203f5aac907SAndriy Gapon 	uint32_t val;
204f5aac907SAndriy Gapon 
205100eced5SKonstantin Belousov 	if ((cpu_vendor_id != CPU_VENDOR_AMD || CPUID_TO_FAMILY(cpu_id) < 0x10) &&
206100eced5SKonstantin Belousov 	    cpu_vendor_id != CPU_VENDOR_HYGON) {
207f5aac907SAndriy Gapon 		printf("DRAM ECC error injection is not supported\n");
208f5aac907SAndriy Gapon 		return (ENXIO);
209f5aac907SAndriy Gapon 	}
210f5aac907SAndriy Gapon 	nbdev = pci_find_bsf(0, 24, 3);
211f5aac907SAndriy Gapon 	if (nbdev == NULL) {
212f5aac907SAndriy Gapon 		printf("Couldn't find NB PCI device\n");
213f5aac907SAndriy Gapon 		return (ENXIO);
214f5aac907SAndriy Gapon 	}
215f5aac907SAndriy Gapon 	val = pci_read_config(nbdev, NB_MCA_CFG, 4);
216f5aac907SAndriy Gapon 	if ((val & DRAM_ECC_EN) == 0) {
217f5aac907SAndriy Gapon 		printf("DRAM ECC is not supported or disabled\n");
218f5aac907SAndriy Gapon 		return (ENXIO);
219f5aac907SAndriy Gapon 	}
220f5aac907SAndriy Gapon 	printf("DRAM ECC error injection support loaded\n");
221f5aac907SAndriy Gapon 	return (0);
222f5aac907SAndriy Gapon }
223f5aac907SAndriy Gapon 
224f5aac907SAndriy Gapon static int
tsc_modevent(module_t mod __unused,int type,void * data __unused)225f5aac907SAndriy Gapon tsc_modevent(module_t mod __unused, int type, void *data __unused)
226f5aac907SAndriy Gapon {
227f5aac907SAndriy Gapon 	int error;
228f5aac907SAndriy Gapon 
229f5aac907SAndriy Gapon 	error = 0;
230f5aac907SAndriy Gapon 	switch (type) {
231f5aac907SAndriy Gapon 	case MOD_LOAD:
232f5aac907SAndriy Gapon 		error = ecc_ei_load();
233f5aac907SAndriy Gapon 		break;
234f5aac907SAndriy Gapon 	case MOD_UNLOAD:
235f5aac907SAndriy Gapon 	case MOD_SHUTDOWN:
236f5aac907SAndriy Gapon 		break;
237f5aac907SAndriy Gapon 	default:
238ab9d2f02SMateusz Guzik 		error = EOPNOTSUPP;
239f5aac907SAndriy Gapon 	}
240ab9d2f02SMateusz Guzik 	return (error);
241f5aac907SAndriy Gapon }
242f5aac907SAndriy Gapon 
243f5aac907SAndriy Gapon DEV_MODULE(tsc, tsc_modevent, NULL);
244