int i, ret = 0;
        struct i2c_msg *msgs, *msg;
        unsigned char *buffs, *buff;
+       bool sched_ras_recovery = false;
        struct eeprom_table_record *record;
        struct amdgpu_device *adev = to_amdgpu_device(control);
+       struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
 
        if (adev->asic_type != CHIP_VEGA20 && adev->asic_type != CHIP_ARCTURUS)
                return 0;
                goto free_buff;
        }
 
+       /*
+        * If saved bad pages number exceeds the bad page threshold for
+        * the whole VRAM, update table header to mark the BAD GPU tag
+        * and schedule one ras recovery after eeprom write is done,
+        * this can avoid the missing for latest records.
+        *
+        * This new header will be picked up and checked in the bootup
+        * by ras recovery, which may break bootup process to notify
+        * user this GPU is in bad state and to retire such GPU for
+        * further check.
+        */
+       if (write && (amdgpu_bad_page_threshold != 0) &&
+               ((control->num_recs + num) >= ras->bad_page_cnt_threshold)) {
+               dev_warn(adev->dev,
+                       "Saved bad pages(%d) reaches threshold value(%d).\n",
+                       control->num_recs + num, ras->bad_page_cnt_threshold);
+               control->tbl_hdr.header = EEPROM_TABLE_HDR_BAD;
+               sched_ras_recovery = true;
+       }
+
        /* In case of overflow just start from beginning to not lose newest records */
        if (write && (control->next_addr + EEPROM_TABLE_RECORD_SIZE * num > EEPROM_SIZE_BYTES))
                control->next_addr = EEPROM_RECORD_START;
 
-
        /*
         * TODO Currently makes EEPROM writes for each record, this creates
         * internal fragmentation. Optimized the code to do full page write of
                __update_tbl_checksum(control, records, num, old_hdr_byte_sum);
 
                __update_table_header(control, buffs);
+
+               if (sched_ras_recovery) {
+                       /*
+                        * Before scheduling ras recovery, assert the related
+                        * flag first, which shall bypass common bad page
+                        * reservation execution in amdgpu_ras_reset_gpu.
+                        */
+                       amdgpu_ras_get_context(adev)->flags |=
+                               AMDGPU_RAS_FLAG_SKIP_BAD_PAGE_RESV;
+
+                       dev_warn(adev->dev, "Conduct ras recovery due to bad "
+                               "page threshold reached.\n");
+                       amdgpu_ras_reset_gpu(adev);
+               }
        } else if (!__validate_tbl_checksum(control, records, num)) {
                DRM_WARN("EEPROM Table checksum mismatch!");
                /* TODO Uncomment when EEPROM read/write is relliable */