2021-10-31 12:38:07

by Avri Altman

Subject: [PATCH] scsi: ufshpb: Properly handle max-single-cmd

The spec recommends that for transfer length larger than the
max-single-cmd attribute (bMAX_ DATA_SIZE_FOR_HPB_SINGLE_CMD) it is
possible to couple pre-reqs with the HPB-READ command. Being a
recommendation, using pre-reqs can be perceived merely as a mean of
optimization. A common practice was to send pre-reqs for chunks within
some interval, and leave the READ10 untouched if larger.

Anyway, now that the pre-reqs flows have been opt-out, all the commands
are single commands. So properly handle this attribute and do not send
HPB-READ for transfer lengths larger than max-single-cmd.

Fixes: 09d9e4d04187 (scsi: ufs: ufshpb: Remove HPB2.0 flows)

Signed-off-by: Avri Altman <[email protected]>
---
drivers/scsi/ufs/ufshpb.c | 29 +++++++++++++++--------------
drivers/scsi/ufs/ufshpb.h | 1 -
2 files changed, 15 insertions(+), 15 deletions(-)

diff --git a/drivers/scsi/ufs/ufshpb.c b/drivers/scsi/ufs/ufshpb.c
index 026a133149dc..40e62d9e2c89 100644
--- a/drivers/scsi/ufs/ufshpb.c
+++ b/drivers/scsi/ufs/ufshpb.c
@@ -394,8 +394,6 @@ int ufshpb_prep(struct ufs_hba *hba, struct ufshcd_lrb *lrbp)
if (!ufshpb_is_supported_chunk(hpb, transfer_len))
return 0;

- WARN_ON_ONCE(transfer_len > HPB_MULTI_CHUNK_HIGH);
-
if (hpb->is_hcm) {
/*
* in host control mode, reads are the main source for
@@ -1572,7 +1570,7 @@ static void ufshpb_lu_parameter_init(struct ufs_hba *hba,
if (ufshpb_is_legacy(hba))
hpb->pre_req_max_tr_len = HPB_LEGACY_CHUNK_HIGH;
else
- hpb->pre_req_max_tr_len = HPB_MULTI_CHUNK_HIGH;
+ hpb->pre_req_max_tr_len = hpb_dev_info->max_hpb_single_cmd;

hpb->lu_pinned_start = hpb_lu_info->pinned_start;
hpb->lu_pinned_end = hpb_lu_info->num_pinned ?
@@ -2582,7 +2580,7 @@ void ufshpb_get_dev_info(struct ufs_hba *hba, u8 *desc_buf)
{
struct ufshpb_dev_info *hpb_dev_info = &hba->ufshpb_dev;
int version, ret;
- u32 max_hpb_single_cmd = HPB_MULTI_CHUNK_LOW;
+ int max_single_cmd;

hpb_dev_info->control_mode = desc_buf[DEVICE_DESC_PARAM_HPB_CONTROL];

@@ -2598,21 +2596,24 @@ void ufshpb_get_dev_info(struct ufs_hba *hba, u8 *desc_buf)
if (version == HPB_SUPPORT_LEGACY_VERSION)
hpb_dev_info->is_legacy = true;

- pm_runtime_get_sync(hba->dev);
- ret = ufshcd_query_attr_retry(hba, UPIU_QUERY_OPCODE_READ_ATTR,
- QUERY_ATTR_IDN_MAX_HPB_SINGLE_CMD, 0, 0, &max_hpb_single_cmd);
- pm_runtime_put_sync(hba->dev);
-
- if (ret)
- dev_err(hba->dev, "%s: idn: read max size of single hpb cmd query request failed",
- __func__);
- hpb_dev_info->max_hpb_single_cmd = max_hpb_single_cmd;
-
/*
* Get the number of user logical unit to check whether all
* scsi_device finish initialization
*/
hpb_dev_info->num_lu = desc_buf[DEVICE_DESC_PARAM_NUM_LU];
+
+ if (hpb_dev_info->is_legacy)
+ return;
+
+ pm_runtime_get_sync(hba->dev);
+ ret = ufshcd_query_attr_retry(hba, UPIU_QUERY_OPCODE_READ_ATTR,
+ QUERY_ATTR_IDN_MAX_HPB_SINGLE_CMD, 0, 0, &max_single_cmd);
+ pm_runtime_put_sync(hba->dev);
+
+ if (ret)
+ hpb_dev_info->max_hpb_single_cmd = HPB_LEGACY_CHUNK_HIGH;
+ else
+ hpb_dev_info->max_hpb_single_cmd = min(max_single_cmd + 1, HPB_MULTI_CHUNK_HIGH);
}

void ufshpb_init(struct ufs_hba *hba)
diff --git a/drivers/scsi/ufs/ufshpb.h b/drivers/scsi/ufs/ufshpb.h
index f15d8fdbce2e..b475dbd78988 100644
--- a/drivers/scsi/ufs/ufshpb.h
+++ b/drivers/scsi/ufs/ufshpb.h
@@ -31,7 +31,6 @@

/* hpb support chunk size */
#define HPB_LEGACY_CHUNK_HIGH 1
-#define HPB_MULTI_CHUNK_LOW 7
#define HPB_MULTI_CHUNK_HIGH 255

/* hpb vender defined opcode */
--
2.17.1

2021-11-01 07:47:56

by Daejun Park

Subject: RE: [PATCH] scsi: ufshpb: Properly handle max-single-cmd

Hi Avri,

> The spec recommends that for transfer length larger than the
> max-single-cmd attribute (bMAX_ DATA_SIZE_FOR_HPB_SINGLE_CMD) it is
> possible to couple pre-reqs with the HPB-READ command. Being a
> recommendation, using pre-reqs can be perceived merely as a mean of
> optimization. A common practice was to send pre-reqs for chunks within
> some interval, and leave the READ10 untouched if larger.
>
> Anyway, now that the pre-reqs flows have been opt-out, all the commands
> are single commands. So properly handle this attribute and do not send
> HPB-READ for transfer lengths larger than max-single-cmd.
>
> Fixes: 09d9e4d04187 (scsi: ufs: ufshpb: Remove HPB2.0 flows)
>
> Signed-off-by: Avri Altman <[email protected]>
> ---
> drivers/scsi/ufs/ufshpb.c | 29 +++++++++++++++--------------
> drivers/scsi/ufs/ufshpb.h | 1 -
> 2 files changed, 15 insertions(+), 15 deletions(-)

Reviewed-by: Daejun Park <[email protected]>