[PATCH 2/2] drm/amdgpu: correct the calculation of RAS bad page

Zhang, Hawking Hawking.Zhang at amd.com
Tue Dec 3 06:29:36 UTC 2024


[AMD Official Use Only - AMD Internal Distribution Only]

Series is

Reviewed-by: Hawking Zhang <Hawking.Zhang at amd.com>

Regards,
Hawking

-----Original Message-----
From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Tao Zhou
Sent: Monday, December 2, 2024 18:25
To: amd-gfx at lists.freedesktop.org
Cc: Zhou1, Tao <Tao.Zhou1 at amd.com>
Subject: [PATCH 2/2] drm/amdgpu: correct the calculation of RAS bad page

After the introduction of NPS RAS, one bad page record on eeprom may be related to 1 or 16 bad pages, so the bad page record and bad page are two different concepts, define a new variable to store bad page number.

Signed-off-by: Tao Zhou <tao.zhou1 at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c       | 10 +----
 .../gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c    | 40 +++++++++++++------
 .../gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h    |  5 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c       |  2 +-
 4 files changed, 35 insertions(+), 22 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index da072ab3fb5c..d03e4ae708dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2934,13 +2934,7 @@ int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev,
        mutex_lock(&con->recovery_lock);
        control = &con->eeprom_control;
        data = con->eh_data;
-       bad_page_num = control->ras_num_recs;
-       /* one record on eeprom stands for all pages in one memory row
-        * in this mode
-        */
-       if (control->rec_type == AMDGPU_RAS_EEPROM_REC_MCA)
-               bad_page_num = control->ras_num_recs * adev->umc.retire_unit;
-
+       bad_page_num = control->ras_num_bad_pages;
        save_count = data->count - bad_page_num;
        mutex_unlock(&con->recovery_lock);

@@ -3432,7 +3426,7 @@ int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev)
                        return ret;

                amdgpu_dpm_send_hbm_bad_pages_num(
-                       adev, control->ras_num_recs);
+                       adev, control->ras_num_bad_pages);

                if (con->update_channel_flag == true) {
                        amdgpu_dpm_send_hbm_bad_channel_flag(
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
index 9dae4ac2f5d0..0b15f0370b67 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
@@ -470,9 +470,10 @@ int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
                res = __write_table_ras_info(control);

        control->ras_num_recs = 0;
+       control->ras_num_bad_pages = 0;
        control->ras_fri = 0;

-       amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_recs);
+       amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_bad_pages);

        control->bad_channel_bitmap = 0;
        amdgpu_dpm_send_hbm_bad_channel_flag(adev, control->bad_channel_bitmap); @@ -559,7 +560,7 @@ bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
        if (con->eeprom_control.tbl_hdr.header == RAS_TABLE_HDR_BAD) {
                if (amdgpu_bad_page_threshold == -1) {
                        dev_warn(adev->dev, "RAS records:%d exceed threshold:%d",
-                               con->eeprom_control.ras_num_recs, con->bad_page_cnt_threshold);
+                               con->eeprom_control.ras_num_bad_pages,
+con->bad_page_cnt_threshold);
                        dev_warn(adev->dev,
                                "But GPU can be operated due to bad_page_threshold = -1.\n");
                        return false;
@@ -621,6 +622,7 @@ amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
                               const u32 num)
 {
        struct amdgpu_ras *con = amdgpu_ras_get_context(to_amdgpu_device(control));
+       struct amdgpu_device *adev = to_amdgpu_device(control);
        u32 a, b, i;
        u8 *buf, *pp;
        int res;
@@ -723,6 +725,12 @@ amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
        control->ras_num_recs = 1 + (control->ras_max_record_count + b
                                     - control->ras_fri)
                % control->ras_max_record_count;
+
+       if (control->rec_type == AMDGPU_RAS_EEPROM_REC_PA)
+               control->ras_num_bad_pages = control->ras_num_recs;
+       else
+               control->ras_num_bad_pages =
+                       control->ras_num_recs * adev->umc.retire_unit;
 Out:
        kfree(buf);
        return res;
@@ -740,10 +748,10 @@ amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
        /* Modify the header if it exceeds.
         */
        if (amdgpu_bad_page_threshold != 0 &&
-           control->ras_num_recs >= ras->bad_page_cnt_threshold) {
+           control->ras_num_bad_pages >= ras->bad_page_cnt_threshold) {
                dev_warn(adev->dev,
                        "Saved bad pages %d reaches threshold value %d\n",
-                       control->ras_num_recs, ras->bad_page_cnt_threshold);
+                       control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
                control->tbl_hdr.header = RAS_TABLE_HDR_BAD;
                if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1) {
                        control->tbl_rai.rma_status = GPU_RETIRED__ECC_REACH_THRESHOLD; @@ -798,9 +806,9 @@ amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
         */
        if (amdgpu_bad_page_threshold != 0 &&
            control->tbl_hdr.version == RAS_TABLE_VER_V2_1 &&
-           control->ras_num_recs < ras->bad_page_cnt_threshold)
+           control->ras_num_bad_pages < ras->bad_page_cnt_threshold)
                control->tbl_rai.health_percent = ((ras->bad_page_cnt_threshold -
-                                                  control->ras_num_recs) * 100) /
+                                                  control->ras_num_bad_pages) * 100) /
                                                   ras->bad_page_cnt_threshold;

        /* Recalc the checksum.
@@ -1402,9 +1410,15 @@ int amdgpu_ras_eeprom_check(struct amdgpu_ras_eeprom_control *control)
        if (!__get_eeprom_i2c_addr(adev, control))
                return -EINVAL;

+       if (control->rec_type == AMDGPU_RAS_EEPROM_REC_PA)
+               control->ras_num_bad_pages = control->ras_num_recs;
+       else
+               control->ras_num_bad_pages =
+                       control->ras_num_recs * adev->umc.retire_unit;
+
        if (hdr->header == RAS_TABLE_HDR_VAL) {
                DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
-                                control->ras_num_recs);
+                                control->ras_num_bad_pages);

                if (hdr->version == RAS_TABLE_VER_V2_1) {
                        res = __read_table_ras_info(control); @@ -1419,9 +1433,9 @@ int amdgpu_ras_eeprom_check(struct amdgpu_ras_eeprom_control *control)

                /* Warn if we are at 90% of the threshold or above
                 */
-               if (10 * control->ras_num_recs >= 9 * ras->bad_page_cnt_threshold)
+               if (10 * control->ras_num_bad_pages >= 9 *
+ras->bad_page_cnt_threshold)
                        dev_warn(adev->dev, "RAS records:%u exceeds 90%% of threshold:%d",
-                                       control->ras_num_recs,
+                                       control->ras_num_bad_pages,
                                        ras->bad_page_cnt_threshold);
        } else if (hdr->header == RAS_TABLE_HDR_BAD &&
                   amdgpu_bad_page_threshold != 0) {
@@ -1435,7 +1449,7 @@ int amdgpu_ras_eeprom_check(struct amdgpu_ras_eeprom_control *control)
                if (res)
                        DRM_ERROR("RAS Table incorrect checksum or error:%d\n",
                                  res);
-               if (ras->bad_page_cnt_threshold > control->ras_num_recs) {
+               if (ras->bad_page_cnt_threshold > control->ras_num_bad_pages) {
                        /* This means that, the threshold was increased since
                         * the last time the system was booted, and now,
                         * ras->bad_page_cnt_threshold - control->num_recs > 0, @@ -1445,13 +1459,13 @@ int amdgpu_ras_eeprom_check(struct amdgpu_ras_eeprom_control *control)
                        dev_info(adev->dev,
                                 "records:%d threshold:%d, resetting "
                                 "RAS table header signature",
-                                control->ras_num_recs,
+                                control->ras_num_bad_pages,
                                 ras->bad_page_cnt_threshold);
                        res = amdgpu_ras_eeprom_correct_header_tag(control,
                                                                   RAS_TABLE_HDR_VAL);
                } else {
                        dev_err(adev->dev, "RAS records:%d exceed threshold:%d",
-                               control->ras_num_recs, ras->bad_page_cnt_threshold);
+                               control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
                        if (amdgpu_bad_page_threshold == -1) {
                                dev_warn(adev->dev, "GPU will be initialized due to bad_page_threshold = -1.");
                                res = 0;
@@ -1460,7 +1474,7 @@ int amdgpu_ras_eeprom_check(struct amdgpu_ras_eeprom_control *control)
                                dev_err(adev->dev,
                                        "RAS records:%d exceed threshold:%d, "
                                        "GPU will not be initialized. Replace this GPU or increase the threshold",
-                                       control->ras_num_recs, ras->bad_page_cnt_threshold);
+                                       control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
                        }
                }
        } else {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h
index b87422df52fd..81d55cb7b397 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h
@@ -95,6 +95,11 @@ struct amdgpu_ras_eeprom_control {
         */
        u32 ras_num_recs;

+       /* the bad page number is ras_num_recs or
+        * ras_num_recs * umc.retire_unit
+        */
+       u32 ras_num_bad_pages;
+
        /* First record index to read, 0-based.
         * Range is [0, num_recs-1]. This is
         * an absolute index, starting right after diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
index 1881bcd040c7..cffe22b86118 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
@@ -169,7 +169,7 @@ void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
                                                err_data->err_addr_cnt, false);
                        amdgpu_ras_save_bad_pages(adev, &err_count);

-                       amdgpu_dpm_send_hbm_bad_pages_num(adev, con->eeprom_control.ras_num_recs);
+                       amdgpu_dpm_send_hbm_bad_pages_num(adev,
+con->eeprom_control.ras_num_bad_pages);

                        if (con->update_channel_flag == true) {
                                amdgpu_dpm_send_hbm_bad_channel_flag(adev, con->eeprom_control.bad_channel_bitmap);
--
2.34.1



More information about the amd-gfx mailing list