drivers/net/ethernet/mellanox/mlx4/crdump.c

   1 /*
   2  * Copyright (c) 2018, Mellanox Technologies. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #include "mlx4.h"
  34
  35 #define BAD_ACCESS                      0xBADACCE5
  36 #define HEALTH_BUFFER_SIZE              0x40
  37 #define CR_ENABLE_BIT                   swab32(BIT(6))
  38 #define CR_ENABLE_BIT_OFFSET            0xF3F04
  39 #define MAX_NUM_OF_DUMPS_TO_STORE       (8)
  40
  41 static const char *region_cr_space_str = "cr-space";
  42 static const char *region_fw_health_str = "fw-health";
  43
  44 /* Set to true in case cr enable bit was set to true before crdump */
  45 static bool crdump_enbale_bit_set;
  46
  47 static void crdump_enable_crspace_access(struct mlx4_dev *dev,
  48                                          u8 __iomem *cr_space)
  49 {
  50         /* Get current enable bit value */
  51         crdump_enbale_bit_set =
  52                 readl(cr_space + CR_ENABLE_BIT_OFFSET) & CR_ENABLE_BIT;
  53
  54         /* Enable FW CR filter (set bit6 to 0) */
  55         if (crdump_enbale_bit_set)
  56                 writel(readl(cr_space + CR_ENABLE_BIT_OFFSET) & ~CR_ENABLE_BIT,
  57                        cr_space + CR_ENABLE_BIT_OFFSET);
  58
  59         /* Enable block volatile crspace accesses */
  60         writel(swab32(1), cr_space + dev->caps.health_buffer_addrs +
  61                HEALTH_BUFFER_SIZE);
  62 }
  63
  64 static void crdump_disable_crspace_access(struct mlx4_dev *dev,
  65                                           u8 __iomem *cr_space)
  66 {
  67         /* Disable block volatile crspace accesses */
  68         writel(0, cr_space + dev->caps.health_buffer_addrs +
  69                HEALTH_BUFFER_SIZE);
  70
  71         /* Restore FW CR filter value (set bit6 to original value) */
  72         if (crdump_enbale_bit_set)
  73                 writel(readl(cr_space + CR_ENABLE_BIT_OFFSET) | CR_ENABLE_BIT,
  74                        cr_space + CR_ENABLE_BIT_OFFSET);
  75 }
  76
  77 static void mlx4_crdump_collect_crspace(struct mlx4_dev *dev,
  78                                         u8 __iomem *cr_space,
  79                                         u32 id)
  80 {
  81         struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
  82         struct pci_dev *pdev = dev->persist->pdev;
  83         unsigned long cr_res_size;
  84         u8 *crspace_data;
  85         int offset;
  86         int err;
  87
  88         if (!crdump->region_crspace) {
  89                 mlx4_err(dev, "crdump: cr-space region is NULL\n");
  90                 return;
  91         }
  92
  93         /* Try to collect CR space */
  94         cr_res_size = pci_resource_len(pdev, 0);
  95         crspace_data = kvmalloc(cr_res_size, GFP_KERNEL);
  96         if (crspace_data) {
  97                 for (offset = 0; offset < cr_res_size; offset += 4)
  98                         *(u32 *)(crspace_data + offset) =
  99                                         readl(cr_space + offset);
 100
 101                 err = devlink_region_snapshot_create(crdump->region_crspace,
 102                                                      cr_res_size, crspace_data,
 103                                                      id, &kvfree);
 104                 if (err) {
 105                         kvfree(crspace_data);
 106                         mlx4_warn(dev, "crdump: devlink create %s snapshot id %d err %d\n",
 107                                   region_cr_space_str, id, err);
 108                 } else {
 109                         mlx4_info(dev, "crdump: added snapshot %d to devlink region %s\n",
 110                                   id, region_cr_space_str);
 111                 }
 112         } else {
 113                 mlx4_err(dev, "crdump: Failed to allocate crspace buffer\n");
 114         }
 115 }
 116
 117 static void mlx4_crdump_collect_fw_health(struct mlx4_dev *dev,
 118                                           u8 __iomem *cr_space,
 119                                           u32 id)
 120 {
 121         struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
 122         u8 *health_data;
 123         int offset;
 124         int err;
 125
 126         if (!crdump->region_fw_health) {
 127                 mlx4_err(dev, "crdump: fw-health region is NULL\n");
 128                 return;
 129         }
 130
 131         /* Try to collect health buffer */
 132         health_data = kvmalloc(HEALTH_BUFFER_SIZE, GFP_KERNEL);
 133         if (health_data) {
 134                 u8 __iomem *health_buf_start =
 135                                 cr_space + dev->caps.health_buffer_addrs;
 136
 137                 for (offset = 0; offset < HEALTH_BUFFER_SIZE; offset += 4)
 138                         *(u32 *)(health_data + offset) =
 139                                         readl(health_buf_start + offset);
 140
 141                 err = devlink_region_snapshot_create(crdump->region_fw_health,
 142                                                      HEALTH_BUFFER_SIZE,
 143                                                      health_data,
 144                                                      id, &kvfree);
 145                 if (err) {
 146                         kvfree(health_data);
 147                         mlx4_warn(dev, "crdump: devlink create %s snapshot id %d err %d\n",
 148                                   region_fw_health_str, id, err);
 149                 } else {
 150                         mlx4_info(dev, "crdump: added snapshot %d to devlink region %s\n",
 151                                   id, region_fw_health_str);
 152                 }
 153         } else {
 154                 mlx4_err(dev, "crdump: Failed to allocate health buffer\n");
 155         }
 156 }
 157
 158 int mlx4_crdump_collect(struct mlx4_dev *dev)
 159 {
 160         struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
 161         struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
 162         struct pci_dev *pdev = dev->persist->pdev;
 163         unsigned long cr_res_size;
 164         u8 __iomem *cr_space;
 165         u32 id;
 166
 167         if (!dev->caps.health_buffer_addrs) {
 168                 mlx4_info(dev, "crdump: FW doesn't support health buffer access, skipping\n");
 169                 return 0;
 170         }
 171
 172         if (!crdump->snapshot_enable) {
 173                 mlx4_info(dev, "crdump: devlink snapshot disabled, skipping\n");
 174                 return 0;
 175         }
 176
 177         cr_res_size = pci_resource_len(pdev, 0);
 178
 179         cr_space = ioremap(pci_resource_start(pdev, 0), cr_res_size);
 180         if (!cr_space) {
 181                 mlx4_err(dev, "crdump: Failed to map pci cr region\n");
 182                 return -ENODEV;
 183         }
 184
 185         crdump_enable_crspace_access(dev, cr_space);
 186
 187         /* Get the available snapshot ID for the dumps */
 188         id = devlink_region_shapshot_id_get(devlink);
 189
 190         /* Try to capture dumps */
 191         mlx4_crdump_collect_crspace(dev, cr_space, id);
 192         mlx4_crdump_collect_fw_health(dev, cr_space, id);
 193
 194         crdump_disable_crspace_access(dev, cr_space);
 195
 196         iounmap(cr_space);
 197         return 0;
 198 }
 199
 200 int mlx4_crdump_init(struct mlx4_dev *dev)
 201 {
 202         struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
 203         struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
 204         struct pci_dev *pdev = dev->persist->pdev;
 205
 206         crdump->snapshot_enable = false;
 207
 208         /* Create cr-space region */
 209         crdump->region_crspace =
 210                 devlink_region_create(devlink,
 211                                       region_cr_space_str,
 212                                       MAX_NUM_OF_DUMPS_TO_STORE,
 213                                       pci_resource_len(pdev, 0));
 214         if (IS_ERR(crdump->region_crspace))
 215                 mlx4_warn(dev, "crdump: create devlink region %s err %ld\n",
 216                           region_cr_space_str,
 217                           PTR_ERR(crdump->region_crspace));
 218
 219         /* Create fw-health region */
 220         crdump->region_fw_health =
 221                 devlink_region_create(devlink,
 222                                       region_fw_health_str,
 223                                       MAX_NUM_OF_DUMPS_TO_STORE,
 224                                       HEALTH_BUFFER_SIZE);
 225         if (IS_ERR(crdump->region_fw_health))
 226                 mlx4_warn(dev, "crdump: create devlink region %s err %ld\n",
 227                           region_fw_health_str,
 228                           PTR_ERR(crdump->region_fw_health));
 229
 230         return 0;
 231 }
 232
 233 void mlx4_crdump_end(struct mlx4_dev *dev)
 234 {
 235         struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
 236
 237         devlink_region_destroy(crdump->region_fw_health);
 238         devlink_region_destroy(crdump->region_crspace);
 239 }