1314638Savg/*- 2314638Savg * Copyright (c) 2017 Andriy Gapon 3314638Savg * All rights reserved. 4314638Savg * 5314638Savg * Redistribution and use in source and binary forms, with or without 6314638Savg * modification, are permitted provided that the following conditions 7314638Savg * are met: 8314638Savg * 1. Redistributions of source code must retain the above copyright 9314638Savg * notice, this list of conditions and the following disclaimer. 10314638Savg * 2. Redistributions in binary form must reproduce the above copyright 11314638Savg * notice, this list of conditions and the following disclaimer in the 12314638Savg * documentation and/or other materials provided with the distribution. 13314638Savg * 14314638Savg * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15314638Savg * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16314638Savg * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17314638Savg * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18314638Savg * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19314638Savg * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20314638Savg * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21314638Savg * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22314638Savg * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23314638Savg * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24314638Savg * SUCH DAMAGE. 25314638Savg * 26314638Savg * $FreeBSD: stable/11/sys/dev/amd_ecc_inject/ecc_inject.c 314638 2017-03-03 22:51:04Z avg $ 27314638Savg */ 28314638Savg 29314638Savg#include <sys/param.h> 30314638Savg#include <sys/systm.h> 31314638Savg#include <sys/bus.h> 32314638Savg#include <sys/kernel.h> 33314638Savg#include <sys/conf.h> 34314638Savg#include <sys/malloc.h> 35314638Savg#include <sys/module.h> 36314638Savg#include <sys/sysctl.h> 37314638Savg#include <sys/types.h> 38314638Savg 39314638Savg#include <dev/pci/pcivar.h> 40314638Savg 41314638Savg#include <vm/vm.h> 42314638Savg#include <vm/vm_extern.h> 43314638Savg#include <vm/vm_kern.h> 44314638Savg 45314638Savg#include <machine/cputypes.h> 46314638Savg#include <machine/md_var.h> 47314638Savg 48314638Savg 49314638Savg/* 50314638Savg * See BKDG for AMD Family 15h Models 00h-0Fh Processors 51314638Savg * (publication 42301 Rev 3.08 - March 12, 2012): 52314638Savg * - 2.13.3.1 DRAM Error Injection 53314638Savg * - D18F3xB8 NB Array Address 54314638Savg * - D18F3xBC NB Array Data Port 55314638Savg * - D18F3xBC_x8 DRAM ECC 56314638Savg */ 57314638Savg#define NB_MCA_CFG 0x44 58314638Savg#define DRAM_ECC_EN (1 << 22) 59314638Savg#define NB_MCA_EXTCFG 0x180 60314638Savg#define ECC_SYMB_SZ (1 << 25) 61314638Savg#define NB_ARRAY_ADDR 0xb8 62314638Savg#define DRAM_ECC_SEL (0x8 << 28) 63314638Savg#define QUADRANT_SHIFT 1 64314638Savg#define QUADRANT_MASK 0x3 65314638Savg#define NB_ARRAY_PORT 0xbc 66314638Savg#define INJ_WORD_SHIFT 20 67314638Savg#define INJ_WORD_MASK 0x1ff 68314638Savg#define DRAM_ERR_EN (1 << 18) 69314638Savg#define DRAM_WR_REQ (1 << 17) 70314638Savg#define DRAM_RD_REQ (1 << 16) 71314638Savg#define INJ_VECTOR_MASK 0xffff 72314638Savg 73314638Savgstatic void ecc_ei_inject(int); 74314638Savg 75314638Savgstatic device_t nbdev; 76314638Savgstatic int delay_ms = 0; 77314638Savgstatic int quadrant = 0; /* 0 - 3 */ 78314638Savgstatic int word_mask = 0x001; /* 9 bits: 8 + 1 for ECC */ 79314638Savgstatic int bit_mask = 0x0001; /* 16 bits */ 80314638Savg 81314638Savgstatic int 82314638Savgsysctl_int_with_max(SYSCTL_HANDLER_ARGS) 83314638Savg{ 84314638Savg u_int value; 85314638Savg int error; 86314638Savg 87314638Savg value = *(u_int *)arg1; 88314638Savg error = sysctl_handle_int(oidp, &value, 0, req); 89314638Savg if (error || req->newptr == NULL) 90314638Savg return (error); 91314638Savg if (value > arg2) 92314638Savg return (EINVAL); 93314638Savg *(u_int *)arg1 = value; 94314638Savg return (0); 95314638Savg} 96314638Savg 97314638Savgstatic int 98314638Savgsysctl_nonzero_int_with_max(SYSCTL_HANDLER_ARGS) 99314638Savg{ 100314638Savg u_int value; 101314638Savg int error; 102314638Savg 103314638Savg value = *(u_int *)arg1; 104314638Savg error = sysctl_int_with_max(oidp, &value, arg2, req); 105314638Savg if (error || req->newptr == NULL) 106314638Savg return (error); 107314638Savg if (value == 0) 108314638Savg return (EINVAL); 109314638Savg *(u_int *)arg1 = value; 110314638Savg return (0); 111314638Savg} 112314638Savg 113314638Savgstatic int 114314638Savgsysctl_proc_inject(SYSCTL_HANDLER_ARGS) 115314638Savg{ 116314638Savg int error; 117314638Savg int i; 118314638Savg 119314638Savg i = 0; 120314638Savg error = sysctl_handle_int(oidp, &i, 0, req); 121314638Savg if (error) 122314638Savg return (error); 123314638Savg if (i != 0) 124314638Savg ecc_ei_inject(i); 125314638Savg return (0); 126314638Savg} 127314638Savg 128314638Savgstatic SYSCTL_NODE(_hw, OID_AUTO, error_injection, CTLFLAG_RD, NULL, 129314638Savg "Hardware error injection"); 130314638Savgstatic SYSCTL_NODE(_hw_error_injection, OID_AUTO, dram_ecc, CTLFLAG_RD, NULL, 131314638Savg "DRAM ECC error injection"); 132314638SavgSYSCTL_UINT(_hw_error_injection_dram_ecc, OID_AUTO, delay, 133314638Savg CTLTYPE_UINT | CTLFLAG_RW, &delay_ms, 0, 134314638Savg "Delay in milliseconds between error injections"); 135314638SavgSYSCTL_PROC(_hw_error_injection_dram_ecc, OID_AUTO, quadrant, 136314638Savg CTLTYPE_UINT | CTLFLAG_RW, &quadrant, QUADRANT_MASK, 137314638Savg sysctl_int_with_max, "IU", 138314638Savg "Index of 16-byte quadrant within 64-byte line where errors " 139314638Savg "should be injected"); 140314638SavgSYSCTL_PROC(_hw_error_injection_dram_ecc, OID_AUTO, word_mask, 141314638Savg CTLTYPE_UINT | CTLFLAG_RW, &word_mask, INJ_WORD_MASK, 142314638Savg sysctl_nonzero_int_with_max, "IU", 143314638Savg "9-bit mask of words where errors should be injected (8 data + 1 ECC)"); 144314638SavgSYSCTL_PROC(_hw_error_injection_dram_ecc, OID_AUTO, bit_mask, 145314638Savg CTLTYPE_UINT | CTLFLAG_RW, &bit_mask, INJ_VECTOR_MASK, 146314638Savg sysctl_nonzero_int_with_max, "IU", 147314638Savg "16-bit mask of bits within each selected word where errors " 148314638Savg "should be injected"); 149314638SavgSYSCTL_PROC(_hw_error_injection_dram_ecc, OID_AUTO, inject, 150314638Savg CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, sysctl_proc_inject, "I", 151314638Savg "Inject a number of errors according to configured parameters"); 152314638Savg 153314638Savgstatic void 154314638Savgecc_ei_inject_one(void *arg, size_t size) 155314638Savg{ 156314638Savg volatile uint64_t *memory = arg; 157314638Savg uint32_t val; 158314638Savg int i; 159314638Savg 160314638Savg val = DRAM_ECC_SEL | (quadrant << QUADRANT_SHIFT); 161314638Savg pci_write_config(nbdev, NB_ARRAY_ADDR, val, 4); 162314638Savg 163314638Savg val = (word_mask << INJ_WORD_SHIFT) | DRAM_WR_REQ | bit_mask; 164314638Savg pci_write_config(nbdev, NB_ARRAY_PORT, val, 4); 165314638Savg 166314638Savg for (i = 0; i < size / sizeof(uint64_t); i++) { 167314638Savg memory[i] = 0; 168314638Savg val = pci_read_config(nbdev, NB_ARRAY_PORT, 4); 169314638Savg if ((val & DRAM_WR_REQ) == 0) 170314638Savg break; 171314638Savg } 172314638Savg for (i = 0; i < size / sizeof(uint64_t); i++) 173314638Savg memory[0] = memory[i]; 174314638Savg} 175314638Savg 176314638Savgstatic void 177314638Savgecc_ei_inject(int count) 178314638Savg{ 179314638Savg vm_offset_t memory; 180314638Savg int injected; 181314638Savg 182314638Savg KASSERT((quadrant & ~QUADRANT_MASK) == 0, 183314638Savg ("quadrant value is outside of range: %u", quadrant)); 184314638Savg KASSERT(word_mask != 0 && (word_mask & ~INJ_WORD_MASK) == 0, 185314638Savg ("word mask value is outside of range: 0x%x", word_mask)); 186314638Savg KASSERT(bit_mask != 0 && (bit_mask & ~INJ_VECTOR_MASK) == 0, 187314638Savg ("bit mask value is outside of range: 0x%x", bit_mask)); 188314638Savg 189314638Savg memory = kmem_alloc_attr(kernel_arena, PAGE_SIZE, M_WAITOK, 0, ~0, 190314638Savg VM_MEMATTR_UNCACHEABLE); 191314638Savg 192314638Savg for (injected = 0; injected < count; injected++) { 193314638Savg ecc_ei_inject_one((void*)memory, PAGE_SIZE); 194314638Savg if (delay_ms != 0 && injected != count - 1) 195314638Savg pause_sbt("ecc_ei_inject", delay_ms * SBT_1MS, 0, 0); 196314638Savg } 197314638Savg 198314638Savg kmem_free(kernel_arena, memory, PAGE_SIZE); 199314638Savg} 200314638Savg 201314638Savgstatic int 202314638Savgecc_ei_load(void) 203314638Savg{ 204314638Savg uint32_t val; 205314638Savg 206314638Savg if (cpu_vendor_id != CPU_VENDOR_AMD || CPUID_TO_FAMILY(cpu_id) < 0x10) { 207314638Savg printf("DRAM ECC error injection is not supported\n"); 208314638Savg return (ENXIO); 209314638Savg } 210314638Savg nbdev = pci_find_bsf(0, 24, 3); 211314638Savg if (nbdev == NULL) { 212314638Savg printf("Couldn't find NB PCI device\n"); 213314638Savg return (ENXIO); 214314638Savg } 215314638Savg val = pci_read_config(nbdev, NB_MCA_CFG, 4); 216314638Savg if ((val & DRAM_ECC_EN) == 0) { 217314638Savg printf("DRAM ECC is not supported or disabled\n"); 218314638Savg return (ENXIO); 219314638Savg } 220314638Savg printf("DRAM ECC error injection support loaded\n"); 221314638Savg return (0); 222314638Savg} 223314638Savg 224314638Savgstatic int 225314638Savgtsc_modevent(module_t mod __unused, int type, void *data __unused) 226314638Savg{ 227314638Savg int error; 228314638Savg 229314638Savg error = 0; 230314638Savg switch (type) { 231314638Savg case MOD_LOAD: 232314638Savg error = ecc_ei_load(); 233314638Savg break; 234314638Savg case MOD_UNLOAD: 235314638Savg case MOD_SHUTDOWN: 236314638Savg break; 237314638Savg default: 238314638Savg return (EOPNOTSUPP); 239314638Savg } 240314638Savg return (0); 241314638Savg} 242314638Savg 243314638SavgDEV_MODULE(tsc, tsc_modevent, NULL); 244