Skip to contentSkip to conversation

Please check ufshpb init flow

DISCUSSION 정요한(JOUNG YOHAN) Mobile SE 3 participants 5 messages Nov 19, 2021 – Apr 12, 2022 SCSI

lore
Share & tools
lore
#0 lore raw

Hi daejun

Please check ufshpb init flow.
sending write buffer(0x03) seems spec violation (JESD220 Commands for exceptional behavior on UAC, SAM 5r05) before UAC clear (sd_probe). Anyway hpb reset query(ufshpb_check_hpb_reset_query) seems sufficient.
Why do we need to send write buffer?

void ufshpb_init_hpb_lu(struct ufs_hba *hba, struct scsi_device *sdev) {
out:

    /* All LUs are initialized */
    if (atomic_dec_and_test(&hba->ufshpb_dev.slave_conf_cnt))
	There seems to be a problem with this logic.
	If hpb is set on the last LUN, write buffer command is sent before sd_probe completes.
        ufshpb_hpb_lu_prepared(hba);
}

static void ufshpb_hpb_lu_prepared(struct ufs_hba *hba) {

    init_success = !ufshpb_check_hpb_reset_query(hba);

    shost_for_each_device(sdev, hba->host) {
        hpb = ufshpb_get_hpb_data(sdev);
        if (!hpb)
            continue;

        if (init_success) {
            ufshpb_set_state(hpb, HPB_PRESENT);
            if ((hpb->lu_pinned_end - hpb->lu_pinned_start) > 0)
                queue_work(ufshpb_wq, &hpb->map_work);
            if (!hpb->is_hcm)
                ufshpb_issue_umap_all_req(hpb); 
		This seems unnecessary.
		
        } else {

Thanks
yohan

#1 lore raw

it's for "Inactivating all HPB regions" after reboot

scsi_add_lun()...>ufshpb_issue_umap_all_req(hpb);

if it is a cold reboot on both host side and UFS device side, it is not necessary to issue this write buffer.

Quoted 43 lines · 정요한(JOUNG YOHAN) Mobile SE

Hi daejun

Please check ufshpb init flow.
sending write buffer(0x03) seems spec violation (JESD220 Commands for exceptional behavior on UAC, SAM 5r05) before UAC clear (sd_probe). Anyway hpb reset query(ufshpb_check_hpb_reset_query) seems sufficient.
Why do we need to send write buffer?

void ufshpb_init_hpb_lu(struct ufs_hba *hba, struct scsi_device *sdev)
{
out:
/* All LUs are initialized */
if (atomic_dec_and_test(&hba->ufshpb_dev.slave_conf_cnt))
There seems to be a problem with this logic.
If hpb is set on the last LUN, write buffer command is sent
before sd_probe completes.

        ufshpb_hpb_lu_prepared(hba);
}

static void ufshpb_hpb_lu_prepared(struct ufs_hba *hba) {

    init_success = !ufshpb_check_hpb_reset_query(hba);

    shost_for_each_device(sdev, hba->host) {
        hpb = ufshpb_get_hpb_data(sdev);
        if (!hpb)
            continue;

        if (init_success) {
            ufshpb_set_state(hpb, HPB_PRESENT);
            if ((hpb->lu_pinned_end - hpb->lu_pinned_start) > 0)
                queue_work(ufshpb_wq, &hpb->map_work);
            if (!hpb->is_hcm)
                ufshpb_issue_umap_all_req(hpb); 
		This seems unnecessary.
		
        } else {

Thanks
yohan

#2 lore raw
Quoted 6 lines

it's for "Inactivating all HPB regions" after reboot

scsi_add_lun()...>ufshpb_issue_umap_all_req(hpb);

if it is a cold reboot on both host side and UFS device side, it is not necessary to issue this write buffer.

According to you, is ufshpb_issue_umap_all_req used only for host reset?
During boottime, the problem is that the hpb is set to the last LUN, write buffer command is sent before sd_probe completes. so, instead of performing unmap, UAC is returned.
If ufshpb_issue_umap_all_req is not needed in cold boot, it seems necessary to move it to an appropriate location or remove

Quoted 44 lines · 정요한(JOUNG YOHAN) Mobile SE
Quoted 41 lines

Hi daejun

Please check ufshpb init flow.
sending write buffer(0x03) seems spec violation (JESD220 Commands for exceptional behavior on UAC, SAM 5r05) before UAC clear (sd_probe). Anyway hpb reset query(ufshpb_check_hpb_reset_query) seems sufficient. Why do we need to send write buffer?

void ufshpb_init_hpb_lu(struct ufs_hba *hba, struct scsi_device *sdev)
{
out:
/* All LUs are initialized */
if (atomic_dec_and_test(&hba->ufshpb_dev.slave_conf_cnt))
There seems to be a problem with this logic.
If hpb is set on the last LUN, write buffer command is sent before
sd_probe completes.

        ufshpb_hpb_lu_prepared(hba);
}

static void ufshpb_hpb_lu_prepared(struct ufs_hba *hba) {

    init_success = !ufshpb_check_hpb_reset_query(hba);

    shost_for_each_device(sdev, hba->host) {
        hpb = ufshpb_get_hpb_data(sdev);
        if (!hpb)
            continue;

        if (init_success) {
            ufshpb_set_state(hpb, HPB_PRESENT);
            if ((hpb->lu_pinned_end - hpb->lu_pinned_start) > 0)
                queue_work(ufshpb_wq, &hpb->map_work);
            if (!hpb->is_hcm)
                ufshpb_issue_umap_all_req(hpb);
		This seems unnecessary.

        } else {

Thanks
yohan

#3 lore raw

Hi Yohan and Bean,

I have propose a patch to remove ufshpb_issue_umap_all_req as we also consider this a redundant operation.

Please refer to topic "[PATCH 1/1] scsi: ufs: remove redundant HPB unmap". Feel free to leave comments there.
Thanks~
Po-Wen

Quoted 66 lines · 정요한(JOUNG YOHAN) Mobile SE
Quoted 7 lines

it's for "Inactivating all HPB regions" after reboot

scsi_add_lun()...>ufshpb_issue_umap_all_req(hpb);

if it is a cold reboot on both host side and UFS device side, it is not
necessary to issue this write buffer.

According to you, is ufshpb_issue_umap_all_req used only for host reset?
During boottime, the problem is that the hpb is set to the last LUN, write buffer command is sent before sd_probe completes. so, instead of performing unmap, UAC is returned.
If ufshpb_issue_umap_all_req is not needed in cold boot, it seems necessary to move it to an appropriate location or remove

Quoted 49 lines
Quoted 45 lines · 정요한(JOUNG YOHAN) Mobile SE

Hi daejun

Please check ufshpb init flow.
sending write buffer(0x03) seems spec violation (JESD220 Commands for
exceptional behavior on UAC, SAM 5r05) before UAC clear (sd_probe).
Anyway hpb reset query(ufshpb_check_hpb_reset_query) seems sufficient.
Why do we need to send write buffer?

void ufshpb_init_hpb_lu(struct ufs_hba *hba, struct scsi_device *sdev)
{
out:
/* All LUs are initialized */
if (atomic_dec_and_test(&hba->ufshpb_dev.slave_conf_cnt))
There seems to be a problem with this logic.
If hpb is set on the last LUN, write buffer command is sent
before
sd_probe completes.

        ufshpb_hpb_lu_prepared(hba);
}

static void ufshpb_hpb_lu_prepared(struct ufs_hba *hba) {

    init_success = !ufshpb_check_hpb_reset_query(hba);

    shost_for_each_device(sdev, hba->host) {
        hpb = ufshpb_get_hpb_data(sdev);
        if (!hpb)
            continue;

        if (init_success) {
            ufshpb_set_state(hpb, HPB_PRESENT);
            if ((hpb->lu_pinned_end - hpb->lu_pinned_start) > 0)
                queue_work(ufshpb_wq, &hpb->map_work);
            if (!hpb->is_hcm)
                ufshpb_issue_umap_all_req(hpb);
		This seems unnecessary.

        } else {

Thanks
yohan

#4 lore raw
In reply to 정요한(JOUNG YOHAN) Mobile SE

Hi Yohan and Bean,

I have propose a patch to remove ufshpb_issue_umap_all_req as we also consider this a redundant operation.

Please refer to topic "[PATCH 1/1] scsi: ufs: remove redundant HPB unmap". Feel free to leave comments there. Thanks~

Po-Wen

Quoted 66 lines · 정요한(JOUNG YOHAN) Mobile SE
Quoted 7 lines

it's for "Inactivating all HPB regions" after reboot

scsi_add_lun()...>ufshpb_issue_umap_all_req(hpb);

if it is a cold reboot on both host side and UFS device side, it is not
necessary to issue this write buffer.

According to you, is ufshpb_issue_umap_all_req used only for host reset?
During boottime, the problem is that the hpb is set to the last LUN, write buffer command is sent before sd_probe completes. so, instead of performing unmap, UAC is returned.
If ufshpb_issue_umap_all_req is not needed in cold boot, it seems necessary to move it to an appropriate location or remove

Quoted 49 lines
Quoted 45 lines · 정요한(JOUNG YOHAN) Mobile SE

Hi daejun

Please check ufshpb init flow.
sending write buffer(0x03) seems spec violation (JESD220 Commands for
exceptional behavior on UAC, SAM 5r05) before UAC clear (sd_probe).
Anyway hpb reset query(ufshpb_check_hpb_reset_query) seems sufficient.
Why do we need to send write buffer?

void ufshpb_init_hpb_lu(struct ufs_hba *hba, struct scsi_device *sdev)
{
out:
/* All LUs are initialized */
if (atomic_dec_and_test(&hba->ufshpb_dev.slave_conf_cnt))
There seems to be a problem with this logic.
If hpb is set on the last LUN, write buffer command is sent
before
sd_probe completes.

        ufshpb_hpb_lu_prepared(hba);
}

static void ufshpb_hpb_lu_prepared(struct ufs_hba *hba) {

    init_success = !ufshpb_check_hpb_reset_query(hba);

    shost_for_each_device(sdev, hba->host) {
        hpb = ufshpb_get_hpb_data(sdev);
        if (!hpb)
            continue;

        if (init_success) {
            ufshpb_set_state(hpb, HPB_PRESENT);
            if ((hpb->lu_pinned_end - hpb->lu_pinned_start) > 0)
                queue_work(ufshpb_wq, &hpb->map_work);
            if (!hpb->is_hcm)
                ufshpb_issue_umap_all_req(hpb);
		This seems unnecessary.

        } else {

Thanks
yohan