UFS driver general fixes bundle 1
v4 cover UFS driver general fixes bundle 1
This bundle includes 5 general fixes for UFS driver.
Changes since v3:
- Incorporated review comments from Martin K. Petersen.
Changes since v2:
- Incorporated review comments from Mark Salyzyn
Changes since v1:
- Incorporated review comments from Bart Van Assche.
Can Guo (5):
scsi: Adjust DBD setting in mode sense for caching mode page per LLD
scsi: ufs: Use DBD setting in mode sense
scsi: ufs: Release clock if DMA map fails
scsi: ufs: Do not clear the DL layer timers
scsi: ufs: Do not free irq in suspend
drivers/scsi/scsi_lib.c | 2 ++ drivers/scsi/ufs/ufshcd.c | 42 ++++++++++++++++++++++++++++-------------- drivers/scsi/ufs/unipro.h | 11 +++++++++++ include/scsi/scsi_device.h | 1 + 4 files changed, 42 insertions(+), 14 deletions(-)
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
PATCH v4 1/5 scsi: Adjust DBD setting in mode sense for caching mode page per LLD
UFS JEDEC standards require DBD field to be set to 1 in mode sense command. This patch allows LLD to define the setting of DBD if required.
--- drivers/scsi/scsi_lib.c | 2 ++ include/scsi/scsi_device.h | 1 + 2 files changed, 3 insertions(+)
drivers/scsi/scsi_lib.c +2 -0
@@ -2108,6 +2108,8 @@ void scsi_exit_queue(void)memset (data,0 ,sizeof (*data));memset (&cmd[0 ],0 ,12 ); + + dbd = sdev->set_dbd_for_ms ?8 : dbd; cmd[1 ] = dbd &0x18 ;/* allows DBD and LLBA bits */ cmd[2 ] = modepage;
include/scsi/scsi_device.h +1 -0
@@ -172,6 +172,7 @@ struct scsi_device {* because we did a bus reset. */ unsigned use_10_for_rw:1; /* first try 10-byte read / write */ unsigned use_10_for_ms:1; /* first try 10-byte mode sense/select */ + unsigned set_dbd_for_ms:1; /* Set "DBD" field in mode sense */ unsigned no_report_opcodes:1; /* no REPORT SUPPORTED OPERATION CODES */ unsigned no_write_same:1; /* no WRITE SAME command */ unsigned use_16_for_rw:1; /* Use read/write(16) over read/write(10) */
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
PATCH v4 2/5 scsi: ufs: Use DBD setting in mode sense
In reply to Can GuoUFS standards requires DBD field to be set to 1 in mode sense(10).
Some card vendors are more strict and check the DBD field, hence respond with CHECK_CONDITION (Sense key set to ILLEGAL_REQUEST and ASC set to INVALID FIELD IN CDB).
When host send mode sense for page caching, as a result of the CHECK_CONDITION response, host assumes that the device doesn't support the cache feature and doesn't send SYNCHRONIZE_CACHE commands to flush the device cache. This can result in data corruption in case of sudden power down, when there is data stored in the device cache.
This patch fixes the DBD field setting as required in UFS standards.
--- drivers/scsi/ufs/ufshcd.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/scsi/ufs/ufshcd.c +3 -0
@@ -4596,6 +4596,9 @@ static int ufshcd_slave_alloc(struct scsi_device *sdev)/* Mode sense(6) is not supported by UFS, so use Mode sense(10) */ sdev->use_10_for_ms =1 ; +/* DBD field should be set to 1 in mode sense(10) */ + sdev->set_dbd_for_ms =1 ; +/* allow SCSI layer to restart the device in case of errors */ sdev->allow_restart =1 ;
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
PATCH v4 3/5 scsi: ufs: Release clock if DMA map fails
In reply to Can GuoIn queuecommand path, if DMA map fails, it bails out with clock held. In this case, release the clock to keep its usage paired.
--- drivers/scsi/ufs/ufshcd.c | 1 + 1 file changed, 1 insertion(+)
drivers/scsi/ufs/ufshcd.c +1 -0
@@ -2480,6 +2480,7 @@ static int ufshcd_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd)if (err) { lrbp->cmd =NULL ;clear_bit_unlock (tag, &hba->lrb_in_use); +ufshcd_release (hba); goto out; }/* Make sure descriptors are ready before ringing the doorbell */
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
PATCH v4 4/5 scsi: ufs: Do not clear the DL layer timers
In reply to Can GuoDuring power mode change, PACP_PWR_Req frame sends PAPowerModeUserData parameters (and they are considered valid by device if Flags[4] - UserDataValid bit is set in the same frame). Currently we don't set these PAPowerModeUserData parameters and hardware always sets UserDataValid bit which would clear all the DL layer timeout values of the peer device after the power mode change.
This change sets the PAPowerModeUserData[0..5] to UniPro specification recommended default values, in addition we are also setting the relevant DME_LOCAL_* timer attributes as required by UFS HCI specification.
--- drivers/scsi/ufs/ufshcd.c | 20 ++++++++++++++++++++ drivers/scsi/ufs/unipro.h | 11 +++++++++++ 2 files changed, 31 insertions(+)
drivers/scsi/ufs/ufshcd.c +20 -0
@@ -4084,6 +4084,26 @@ static int ufshcd_change_power_mode(struct ufs_hba *hba,ufshcd_dme_set (hba,UIC_ARG_MIB (PA_HSSERIES ), pwr_mode->hs_rate); +ufshcd_dme_set (hba,UIC_ARG_MIB (PA_PWRMODEUSERDATA0 ), + DL_FC0ProtectionTimeOutVal_Default); +ufshcd_dme_set (hba,UIC_ARG_MIB (PA_PWRMODEUSERDATA1 ), + DL_TC0ReplayTimeOutVal_Default); +ufshcd_dme_set (hba,UIC_ARG_MIB (PA_PWRMODEUSERDATA2 ), + DL_AFC0ReqTimeOutVal_Default); +ufshcd_dme_set (hba,UIC_ARG_MIB (PA_PWRMODEUSERDATA3 ), + DL_FC1ProtectionTimeOutVal_Default); +ufshcd_dme_set (hba,UIC_ARG_MIB (PA_PWRMODEUSERDATA4 ), + DL_TC1ReplayTimeOutVal_Default); +ufshcd_dme_set (hba,UIC_ARG_MIB (PA_PWRMODEUSERDATA5 ), + DL_AFC1ReqTimeOutVal_Default); + +ufshcd_dme_set (hba,UIC_ARG_MIB (DME_LocalFC0ProtectionTimeOutVal), + DL_FC0ProtectionTimeOutVal_Default); +ufshcd_dme_set (hba,UIC_ARG_MIB (DME_LocalTC0ReplayTimeOutVal), + DL_TC0ReplayTimeOutVal_Default); +ufshcd_dme_set (hba,UIC_ARG_MIB (DME_LocalAFC0ReqTimeOutVal), + DL_AFC0ReqTimeOutVal_Default); + ret =ufshcd_uic_change_pwr_mode (hba, pwr_mode->pwr_rx <<4 | pwr_mode->pwr_tx);
drivers/scsi/ufs/unipro.h +11 -0
@@ -161,6 +161,17 @@/* PHY Adapter Protocol Constants */ #define PA_MAXDATALANES 4 +#define DL_FC0ProtectionTimeOutVal_Default 8191 +#define DL_TC0ReplayTimeOutVal_Default 65535 +#define DL_AFC0ReqTimeOutVal_Default 32767 +#define DL_FC1ProtectionTimeOutVal_Default 8191 +#define DL_TC1ReplayTimeOutVal_Default 65535 +#define DL_AFC1ReqTimeOutVal_Default 32767 + +#define DME_LocalFC0ProtectionTimeOutVal 0xD041 +#define DME_LocalTC0ReplayTimeOutVal 0xD042 +#define DME_LocalAFC0ReqTimeOutVal 0xD043 +/* PA power modes */ enum {FAST_MODE =1 ,
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
PATCH v4 5/5 scsi: ufs: Do not free irq in suspend
In reply to Can GuoIf PM QoS is enabled and we set request type to PM_QOS_REQ_AFFINE_IRQ then freeing up the irq makes the free_irq() print out warning with call stack. We don't really need to free up irq during suspend, disabling it during suspend and reenabling it during resume should be good enough.
--- drivers/scsi/ufs/ufshcd.c | 18 ++++-------------- 1 file changed, 4 insertions(+), 14 deletions(-)
drivers/scsi/ufs/ufshcd.c +4 -14
@@ -266,26 +266,18 @@ static inline bool ufshcd_valid_tag(struct ufs_hba *hba, int tag)return tag >=0 && tag < hba->nutrs; } -static inline int ufshcd_enable_irq (struct ufs_hba *hba) +static inline void ufshcd_enable_irq (struct ufs_hba *hba) { -int ret =0 ; -if (!hba->is_irq_enabled) { - ret =request_irq (hba->irq, ufshcd_intr,IRQF_SHARED ,UFSHCD , - hba); -if (ret) -dev_err (hba->dev,"%s: request_irq failed, ret=%d\n" , - __func__, ret); +enable_irq (hba->irq); hba->is_irq_enabled = true; } - -return ret; }static inline void ufshcd_disable_irq (struct ufs_hba *hba) {if (hba->is_irq_enabled) { -free_irq (hba->irq, hba); +disable_irq (hba->irq); hba->is_irq_enabled = false; } }
@@ -7930,9 +7922,7 @@ static int ufshcd_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op) goto out;/* enable the host irq as host controller would be active soon */ - ret =ufshcd_enable_irq (hba); -if (ret) - goto disable_irq_and_vops_clks; +ufshcd_enable_irq (hba); ret =ufshcd_vreg_set_hpm (hba);if (ret)
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
RE: [EXT] [PATCH v4 3/5] scsi: ufs: Release clock if DMA map fails
In reply to Can GuoQuoted 5 lines
In queuecommand path, if DMA map fails, it bails out with clock held. In this case, release the clock to keep its usage paired.
Signed-off-by Can Guo <[email protected]>
Re: [PATCH v4 1/5] scsi: Adjust DBD setting in mode sense for caching mode page per LLD
In reply to Can GuoCan,
Quoted 3 lines
UFS JEDEC standards require DBD field to be set to 1 in mode sense command. This patch allows LLD to define the setting of DBD if required.
These changes look OK to me, btw. Please make sure each patch has a changelog which reflects how the review feedback was addressed.
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering
RE: [PATCH v4 4/5] scsi: ufs: Do not clear the DL layer timers
In reply to Can GuoQuoted 9 lines
During power mode change,
PACP_PWR_Reqframe sends PAPowerModeUserData parameters (and they are considered valid by device if Flags[4] - UserDataValid bit is set in the same frame).
Currently we don't set these PAPowerModeUserData parameters and hardware always sets UserDataValid bit which would clear all the DL layer timeout values of the peer device after the power mode change.
Quoted 5 lines
This change sets the PAPowerModeUserData[0..5] to UniPro specification recommended default values, in addition we are also setting the relevant DME_LOCAL_* timer attributes as required by UFS HCI specification.
Signed-off-by Can Guo <[email protected]>
Reviewed-by Avri Altman <[email protected]>
RE: [PATCH v4 5/5] scsi: ufs: Do not free irq in suspend
In reply to Can GuoQuoted 8 lines
If PM QoS is enabled and we set request type to
PM_QOS_REQ_AFFINE_IRQthen freeing up the irq makes thefree_irq()print out warning with call stack. We don't really need to free up irq during suspend, disabling it during suspend and reenabling it during resume should be good enough.Signed-off-by Can Guo <[email protected]>
Your approach seems reasonable,
However I failed to locate in the kernel PM_QOS_REQ_AFFINE_IRQ, Just in codeaurora.
Is the WARN_ON in free_irq still applies?
Thanks,
Avri
RE: [PATCH v4 4/5] scsi: ufs: Do not clear the DL layer timers
In reply to Avri AltmanQuoted 9 lines
Quoted 2 lines
During power mode change,
PACP_PWR_Reqframe sendsPAPowerModeUserData
Quoted 6 lines
parameters (and they are considered valid by device if Flags[4] - UserDataValid bit is set in the same frame).
Currently we don't set these PAPowerModeUserData parameters and hardware always sets UserDataValid bit which would clear all the DL layer timeout values of the peer device after the power mode change.
Quoted 7 lines
Quoted 6 lines
This change sets the PAPowerModeUserData[0..5] to UniPro specification recommended default values, in addition we are also setting the relevant
DME_LOCAL_* timer attributes as required by UFS HCI specification.Signed-off-by Can Guo <[email protected]>Reviewed-by Avri Altman <[email protected]>
BTW, I noticed that you are only updating the TC0 registers. Why not setting the TC1 registers as well?
Thanks,
Avri
Re: [PATCH v4 4/5] scsi: ufs: Do not clear the DL layer timers
Quoted 14 lines · Avri Altman
Quoted 14 lines
Quoted 2 lines
During power mode change,
PACP_PWR_Reqframe sendsPAPowerModeUserData
Quoted 11 lines
parameters (and they are considered valid by device if Flags[4] - UserDataValid bit is set in the same frame).
Currently we don't set these PAPowerModeUserData parameters and hardware always sets UserDataValid bit which would clear all the DL layer timeout values of the peer device after the power mode change.This change sets the PAPowerModeUserData[0..5] to UniPro specification recommended default values, in addition we are also setting the relevant
DME_LOCAL_* timer attributes as required by UFS HCI specification.
Quoted 7 lines
Quoted 2 lines
Quoted 1 line
Signed-off-by Can Guo <[email protected]>Reviewed-by Avri Altman <[email protected]>
BTW, I noticed that you are only updating the TC0 registers. Why not setting the TC1 registers as well?
Thanks,
Avri
Hi Avri,
In the HCI spec, it goes
Currently, UFS uses TC0 only. Therefore, setting the following values are not needed:
DME_ Local_FC1ProtectionTimeOutVal
DME_ Local_TC1ReplayTimeOutVal
DME_ Local_ AFC1ReqTimeOutVal
Best Regards,
Can Guo.
Re: [PATCH v4 5/5] scsi: ufs: Do not free irq in suspend
In reply to Avri AltmanQuoted 7 lines · Avri Altman
Quoted 7 lines
If PM QoS is enabled and we set request type to
PM_QOS_REQ_AFFINE_IRQthen freeing up the irq makes thefree_irq()print out warning with call stack. We don't really need to free up irq during suspend, disabling it during suspend and reenabling it during resume should be good enough.Signed-off-by Can Guo <[email protected]>
Quoted 8 lines
Your approach seems reasonable,
However I failed to locate in the kernelPM_QOS_REQ_AFFINE_IRQ, Just in codeaurora.Is the
WARN_ONinfree_irqstill applies?Thanks,
Avri
Hi Avri,
Thanks for pointing. It seems PM_QOS_REQ_AFFINE_IRQ is not present on upstream yet. But this change is still applicable. How about changing the commit message to below?
"Since ufshcd irq resource is allocated with the device resource management aware IRQ request implementation, we don't really need to free up irq during suspend, disabling it during suspend and reenabling it during resume should be good enough."
Thanks,
Can Guo
RE: [PATCH v4 5/5] scsi: ufs: Do not free irq in suspend
In reply to a message not in the archiveQuoted 22 lines · Avri Altman
Quoted 8 lines
Quoted 8 lines
If PM QoS is enabled and we set request type to
PM_QOS_REQ_AFFINE_IRQthen freeing up the irq makes thefree_irq()print out warning with call stack.
We don't really need to free up irq during suspend, disabling it during suspend and reenabling it during resume should be good enough.Signed-off-by Can Guo <[email protected]>Quoted 8 lines
Your approach seems reasonable,
However I failed to locate in the kernelPM_QOS_REQ_AFFINE_IRQ, Just in codeaurora.Is the
WARN_ONinfree_irqstill applies?Thanks,
AvriHi Avri,
Quoted 8 lines
Thanks for pointing. It seems
PM_QOS_REQ_AFFINE_IRQis not present on upstream yet. But this change is still applicable.
How about changing the commit message to below?"Since ufshcd irq resource is allocated with the device resource management aware IRQ request implementation, we don't really need to free up irq during suspend, disabling it during suspend and reenabling it during resume should be good enough."
Looks good to me.
Thanks,
Avri
Quoted 3 lines
Thanks,
Can Guo
Avri Altman
Reviewed-byRE: [PATCH v4 4/5] scsi: ufs: Do not clear the DL layer timers
In reply to a message not in the archiveQuoted 8 lines
During power mode change,
PACP_PWR_Reqframe sends PAPowerModeUserData parameters (and they are considered valid by device if Flags[4] - UserDataValid bit is set in the same frame).
Currently we don't set these PAPowerModeUserData parameters and hardware always sets UserDataValid bit which would clear all the DL layer timeout values of the peer device after the power mode change.
Quoted 5 lines
This change sets the PAPowerModeUserData[0..5] to UniPro specification recommended default values, in addition we are also setting the relevant DME_LOCAL_* timer attributes as required by UFS HCI specification.
Signed-off-by Can Guo <[email protected]>
RE: [EXT] [PATCH v4 2/5] scsi: ufs: Use DBD setting in mode sense
In reply to Can GuoBean Huo (beanhuo)
Reviewed-byRE: [EXT] RE: [PATCH v4 4/5] scsi: ufs: Do not clear the DL layer timers
In reply to Avri AltmanQuoted 3 lines
Quoted 2 lines
Signed-off-by Can Guo <[email protected]>Reviewed-by Avri Altman <[email protected]>