drm/amdgpu: RAS EEPROM table is now in debugfs
authorLuben Tuikov <luben.tuikov@amd.com>
Thu, 8 Apr 2021 15:34:26 +0000 (11:34 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Thu, 1 Jul 2021 04:24:41 +0000 (00:24 -0400)
Add "ras_eeprom_size" file in debugfs, which
reports the maximum size allocated to the RAS
table in EEROM, as the number of bytes and the
number of records it could store. For instance,

$cat /sys/kernel/debug/dri/0/ras/ras_eeprom_size
262144 bytes or 10921 records
$_

Add "ras_eeprom_table" file in debugfs, which
dumps the RAS table stored EEPROM, in a formatted
way. For instance,

$cat ras_eeprom_table
 Signature    Version  FirstOffs       Size   Checksum
0x414D4452 0x00010000 0x00000014 0x000000EC 0x000000DA
Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage
    0 0x00014      ue    0x00 0x00000000607608DC 0x000000000000   0x00    0x00 0x000000000000
    1 0x0002C      ue    0x00 0x00000000607608DC 0x000000001000   0x00    0x00 0x000000000001
    2 0x00044      ue    0x00 0x00000000607608DC 0x000000002000   0x00    0x00 0x000000000002
    3 0x0005C      ue    0x00 0x00000000607608DC 0x000000003000   0x00    0x00 0x000000000003
    4 0x00074      ue    0x00 0x00000000607608DC 0x000000004000   0x00    0x00 0x000000000004
    5 0x0008C      ue    0x00 0x00000000607608DC 0x000000005000   0x00    0x00 0x000000000005
    6 0x000A4      ue    0x00 0x00000000607608DC 0x000000006000   0x00    0x00 0x000000000006
    7 0x000BC      ue    0x00 0x00000000607608DC 0x000000007000   0x00    0x00 0x000000000007
    8 0x000D4      ue    0x00 0x00000000607608DD 0x000000008000   0x00    0x00 0x000000000008
$_

Cc: Alexander Deucher <Alexander.Deucher@amd.com>
Cc: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Cc: John Clements <john.clements@amd.com>
Cc: Hawking Zhang <Hawking.Zhang@amd.com>
Cc: Xinhui Pan <xinhui.pan@amd.com>
Signed-off-by: Luben Tuikov <luben.tuikov@amd.com>
Acked-by: Alexander Deucher <Alexander.Deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h

index fd63e5f..53a54ff 100644 (file)
@@ -404,9 +404,9 @@ static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f,
                /* umc ce/ue error injection for a bad page is not allowed */
                if ((data.head.block == AMDGPU_RAS_BLOCK__UMC) &&
                    amdgpu_ras_check_bad_page(adev, data.inject.address)) {
-                       dev_warn(adev->dev, "RAS WARN: 0x%llx has been marked "
-                                       "as bad before error injection!\n",
-                                       data.inject.address);
+                       dev_warn(adev->dev, "RAS WARN: inject: 0x%llx has "
+                                "already been marked as bad!\n",
+                                data.inject.address);
                        break;
                }
 
@@ -1301,6 +1301,12 @@ static struct dentry *amdgpu_ras_debugfs_create_ctrl_node(struct amdgpu_device *
                           &con->bad_page_cnt_threshold);
        debugfs_create_x32("ras_hw_enabled", 0444, dir, &adev->ras_hw_enabled);
        debugfs_create_x32("ras_enabled", 0444, dir, &adev->ras_enabled);
+       debugfs_create_file("ras_eeprom_size", S_IRUGO, dir, adev,
+                           &amdgpu_ras_debugfs_eeprom_size_ops);
+       con->de_ras_eeprom_table = debugfs_create_file("ras_eeprom_table",
+                                                      S_IRUGO, dir, adev,
+                                                      &amdgpu_ras_debugfs_eeprom_table_ops);
+       amdgpu_ras_debugfs_set_ret_size(&con->eeprom_control);
 
        /*
         * After one uncorrectable error happens, usually GPU recovery will
index 256cea5..283afd7 100644 (file)
@@ -318,6 +318,7 @@ struct amdgpu_ras {
        /* sysfs */
        struct device_attribute features_attr;
        struct bin_attribute badpages_attr;
+       struct dentry *de_ras_eeprom_table;
        /* block array */
        struct ras_manager *objs;
 
index dc4a845..677e379 100644 (file)
@@ -27,6 +27,8 @@
 #include <linux/bits.h>
 #include "atom.h"
 #include "amdgpu_eeprom.h"
+#include <linux/debugfs.h>
+#include <linux/uaccess.h>
 
 #define EEPROM_I2C_MADDR_VEGA20         0x0
 #define EEPROM_I2C_MADDR_ARCTURUS       0x40000
 #define RAS_OFFSET_TO_INDEX(_C, _O) (((_O) - \
                                      (_C)->ras_record_offset) / RAS_TABLE_RECORD_SIZE)
 
+/* Given a 0-based relative record index, 0, 1, 2, ..., etc., off
+ * of "fri", return the absolute record index off of the end of
+ * the table header.
+ */
+#define RAS_RI_TO_AI(_C, _I) (((_I) + (_C)->ras_fri) % \
+                             (_C)->ras_max_record_count)
+
 #define RAS_NUM_RECS(_tbl_hdr)  (((_tbl_hdr)->tbl_size - \
                                  RAS_TABLE_HEADER_SIZE) / RAS_TABLE_RECORD_SIZE)
 
 
 static bool __is_ras_eeprom_supported(struct amdgpu_device *adev)
 {
-       if ((adev->asic_type == CHIP_VEGA20) ||
-           (adev->asic_type == CHIP_ARCTURUS) ||
-           (adev->asic_type == CHIP_SIENNA_CICHLID) ||
-           (adev->asic_type == CHIP_ALDEBARAN))
-               return true;
-
-       return false;
+       return  adev->asic_type == CHIP_VEGA20 ||
+               adev->asic_type == CHIP_ARCTURUS ||
+               adev->asic_type == CHIP_SIENNA_CICHLID ||
+               adev->asic_type == CHIP_ALDEBARAN;
 }
 
 static bool __get_eeprom_i2c_addr_arct(struct amdgpu_device *adev,
@@ -258,6 +264,8 @@ int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
        control->ras_num_recs = 0;
        control->ras_fri = 0;
 
+       amdgpu_ras_debugfs_set_ret_size(control);
+
        mutex_unlock(&control->ras_tbl_mutex);
 
        return res;
@@ -591,6 +599,8 @@ int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
        res = amdgpu_ras_eeprom_append_table(control, record, num);
        if (!res)
                res = amdgpu_ras_eeprom_update_header(control);
+       if (!res)
+               amdgpu_ras_debugfs_set_ret_size(control);
 
        mutex_unlock(&control->ras_tbl_mutex);
        return res;
@@ -734,6 +744,223 @@ inline uint32_t amdgpu_ras_eeprom_max_record_count(void)
        return RAS_MAX_RECORD_COUNT;
 }
 
+static ssize_t
+amdgpu_ras_debugfs_eeprom_size_read(struct file *f, char __user *buf,
+                                   size_t size, loff_t *pos)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
+       struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+       struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
+       u8 data[50];
+       int res;
+
+       if (!size)
+               return size;
+
+       if (!ras || !control) {
+               res = snprintf(data, sizeof(data), "Not supported\n");
+       } else {
+               res = snprintf(data, sizeof(data), "%d bytes or %d records\n",
+                              RAS_TBL_SIZE_BYTES, control->ras_max_record_count);
+       }
+
+       if (*pos >= res)
+               return 0;
+
+       res -= *pos;
+       res = min_t(size_t, res, size);
+
+       if (copy_to_user(buf, &data[*pos], res))
+               return -EINVAL;
+
+       *pos += res;
+
+       return res;
+}
+
+const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops = {
+       .owner = THIS_MODULE,
+       .read = amdgpu_ras_debugfs_eeprom_size_read,
+       .write = NULL,
+       .llseek = default_llseek,
+};
+
+static const char *tbl_hdr_str = " Signature    Version  FirstOffs       Size   Checksum\n";
+static const char *tbl_hdr_fmt = "0x%08X 0x%08X 0x%08X 0x%08X 0x%08X\n";
+#define tbl_hdr_fmt_size (5 * (2+8) + 4 + 1)
+static const char *rec_hdr_str = "Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage\n";
+static const char *rec_hdr_fmt = "%5d 0x%05X %7s    0x%02X 0x%016llX 0x%012llX   0x%02X    0x%02X 0x%012llX\n";
+#define rec_hdr_fmt_size (5 + 1 + 7 + 1 + 7 + 1 + 7 + 1 + 18 + 1 + 14 + 1 + 6 + 1 + 7 + 1 + 14 + 1)
+
+static const char *record_err_type_str[AMDGPU_RAS_EEPROM_ERR_COUNT] = {
+       "ignore",
+       "re",
+       "ue",
+};
+
+static loff_t amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control *control)
+{
+       return strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
+               strlen(rec_hdr_str) + rec_hdr_fmt_size * control->ras_num_recs;
+}
+
+void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
+{
+       struct amdgpu_ras *ras = container_of(control, struct amdgpu_ras,
+                                             eeprom_control);
+       struct dentry *de = ras->de_ras_eeprom_table;
+
+       if (de)
+               d_inode(de)->i_size = amdgpu_ras_debugfs_table_size(control);
+}
+
+static ssize_t amdgpu_ras_debugfs_table_read(struct file *f, char __user *buf,
+                                            size_t size, loff_t *pos)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
+       struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+       struct amdgpu_ras_eeprom_control *control = &ras->eeprom_control;
+       const size_t orig_size = size;
+       int res = -EINVAL;
+       size_t data_len;
+
+       mutex_lock(&control->ras_tbl_mutex);
+
+       /* We want *pos - data_len > 0, which means there's
+        * bytes to be printed from data.
+        */
+       data_len = strlen(tbl_hdr_str);
+       if (*pos < data_len) {
+               data_len -= *pos;
+               data_len = min_t(size_t, data_len, size);
+               if (copy_to_user(buf, &tbl_hdr_str[*pos], data_len))
+                       goto Out;
+               buf += data_len;
+               size -= data_len;
+               *pos += data_len;
+       }
+
+       data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size;
+       if (*pos < data_len && size > 0) {
+               u8 data[tbl_hdr_fmt_size + 1];
+               loff_t lpos;
+
+               snprintf(data, sizeof(data), tbl_hdr_fmt,
+                        control->tbl_hdr.header,
+                        control->tbl_hdr.version,
+                        control->tbl_hdr.first_rec_offset,
+                        control->tbl_hdr.tbl_size,
+                        control->tbl_hdr.checksum);
+
+               data_len -= *pos;
+               data_len = min_t(size_t, data_len, size);
+               lpos = *pos - strlen(tbl_hdr_str);
+               if (copy_to_user(buf, &data[lpos], data_len))
+                       goto Out;
+               buf += data_len;
+               size -= data_len;
+               *pos += data_len;
+       }
+
+       data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size + strlen(rec_hdr_str);
+       if (*pos < data_len && size > 0) {
+               loff_t lpos;
+
+               data_len -= *pos;
+               data_len = min_t(size_t, data_len, size);
+               lpos = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size;
+               if (copy_to_user(buf, &rec_hdr_str[lpos], data_len))
+                       goto Out;
+               buf += data_len;
+               size -= data_len;
+               *pos += data_len;
+       }
+
+       data_len = amdgpu_ras_debugfs_table_size(control);
+       if (*pos < data_len && size > 0) {
+               u8 dare[RAS_TABLE_RECORD_SIZE];
+               u8 data[rec_hdr_fmt_size + 1];
+               /* Find the starting record index
+                */
+               int s = (*pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
+                        strlen(rec_hdr_str)) / rec_hdr_fmt_size;
+               int r = (*pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
+                        strlen(rec_hdr_str)) % rec_hdr_fmt_size;
+               struct eeprom_table_record record;
+
+               for ( ; size > 0 && s < control->ras_num_recs; s++) {
+                       u32 ai = RAS_RI_TO_AI(control, s);
+                       /* Read a single record
+                        */
+                       res = __amdgpu_ras_eeprom_read(control, dare, ai, 1);
+                       if (res)
+                               goto Out;
+                       __decode_table_record_from_buf(control, &record, dare);
+                       snprintf(data, sizeof(data), rec_hdr_fmt,
+                                s,
+                                RAS_INDEX_TO_OFFSET(control, ai),
+                                record_err_type_str[record.err_type],
+                                record.bank,
+                                record.ts,
+                                record.offset,
+                                record.mem_channel,
+                                record.mcumc_id,
+                                record.retired_page);
+
+                       data_len = min_t(size_t, rec_hdr_fmt_size - r, size);
+                       if (copy_to_user(buf, &data[r], data_len))
+                               return -EINVAL;
+                       buf += data_len;
+                       size -= data_len;
+                       *pos += data_len;
+                       r = 0;
+               }
+       }
+       res = 0;
+Out:
+       mutex_unlock(&control->ras_tbl_mutex);
+       return res < 0 ? res : orig_size - size;
+}
+
+static ssize_t
+amdgpu_ras_debugfs_eeprom_table_read(struct file *f, char __user *buf,
+                                    size_t size, loff_t *pos)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
+       struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+       struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
+       u8 data[81];
+       int res;
+
+       if (!size)
+               return size;
+
+       if (!ras || !control) {
+               res = snprintf(data, sizeof(data), "Not supported\n");
+               if (*pos >= res)
+                       return 0;
+
+               res -= *pos;
+               res = min_t(size_t, res, size);
+
+               if (copy_to_user(buf, &data[*pos], res))
+                       return -EINVAL;
+
+               *pos += res;
+
+               return res;
+       } else {
+               return amdgpu_ras_debugfs_table_read(f, buf, size, pos);
+       }
+}
+
+const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops = {
+       .owner = THIS_MODULE,
+       .read = amdgpu_ras_debugfs_eeprom_table_read,
+       .write = NULL,
+       .llseek = default_llseek,
+};
+
 /**
  * __verify_ras_table_checksum -- verify the RAS EEPROM table checksum
  * @control: pointer to control structure
index e26af82..f95fc61 100644 (file)
 struct amdgpu_device;
 
 enum amdgpu_ras_eeprom_err_type {
-       AMDGPU_RAS_EEPROM_ERR_PLACE_HOLDER,
+       AMDGPU_RAS_EEPROM_ERR_NA,
        AMDGPU_RAS_EEPROM_ERR_RECOVERABLE,
-       AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE
+       AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE,
+       AMDGPU_RAS_EEPROM_ERR_COUNT,
 };
 
 struct amdgpu_ras_eeprom_table_header {
@@ -121,4 +122,9 @@ int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
 
 inline uint32_t amdgpu_ras_eeprom_max_record_count(void);
 
+void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control);
+
+extern const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops;
+extern const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops;
+
 #endif // _AMDGPU_RAS_EEPROM_H