[PATCH v3] scsi: ufs: mcq: Limit the amount of inflight requests

[PATCH v3] scsi: ufs: mcq: Limit the amount of inflight requests

[Avri Altman]

in UFS, each request is designated via the triplet <iid, lun, task tag>.

In UFS4.0 the Initiator ID field is 8 bits wide, comprised of the
EXT_IID and IID fields. Together with the task tag (single byte), they
limit the driver's hw queues capacity.

The purpose of this patch is to document the ufs spec restrictions.
practically, it impose no functional change because a) the current hw
supports much less (merely 64 requests), and b) since shost->host_tagset
is set - then the number of inflight IOs is limited by just the HW queue
depth.

---
v2 -> v3:
fix smatch warnings (Dan Carpenter)
elaborate commit log (John Garry)

v1 -> v2:
Attend Johannes's and Bart's comments

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Avri Altman <avri.altman@wdc.com>
---
 drivers/ufs/core/ufshcd.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index 03c47f9a2750..9cebec6be35c 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -8468,6 +8468,12 @@ static int ufshcd_alloc_mcq(struct ufs_hba *hba)
 	if (ret)
 		goto err;
 
+	if (hba->nutrs * hba->nr_hw_queues > SZ_64K - 1) {
+		dev_info(hba->dev, "there can be at most 64K inflight requests\n");
+		ret = -EINVAL;
+		goto err;
+	}
+
 	/*
 	 * Previously allocated memory for nutrs may not be enough in MCQ mode.
 	 * Number of supported tags in MCQ mode may be larger than SDB mode.
-- 
2.17.1

Re: [PATCH v3] scsi: ufs: mcq: Limit the amount of inflight requests

[John Garry]

On 20/04/2023 11:11, Avri Altman wrote:
> in UFS, each request is designated via the triplet <iid, lun, task tag>.
> 
> In UFS4.0 the Initiator ID field is 8 bits wide, comprised of the
> EXT_IID and IID fields. Together with the task tag (single byte), they
> limit the driver's hw queues capacity.
> 
> The purpose of this patch is to document the ufs spec restrictions.
> practically, it impose no functional change because a) the current hw
> supports much less (merely 64 requests), and b) since shost->host_tagset
> is set - then the number of inflight IOs is limited by just the HW queue
> depth.
> 

I don't know why effectively it has no change, as it seems to me to be 
checking and failing something valid which could happen.

If it really cannot happen, then we generally don't add what is 
effectively dead code just to provide practical illustrations of the 
spec. And it is misleading to others who may use your driver as a reference.

A code comment would be a better choice, if anything, IMHO.

Thanks,
John

> ---
> v2 -> v3:
> fix smatch warnings (Dan Carpenter)
> elaborate commit log (John Garry)
> 
> v1 -> v2:
> Attend Johannes's and Bart's comments

This is the wrong place for these

> 
> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> Signed-off-by: Avri Altman <avri.altman@wdc.com>
> ---
>   drivers/ufs/core/ufshcd.c | 6 ++++++
>   1 file changed, 6 insertions(+)
> 
> diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
> index 03c47f9a2750..9cebec6be35c 100644
> --- a/drivers/ufs/core/ufshcd.c
> +++ b/drivers/ufs/core/ufshcd.c
> @@ -8468,6 +8468,12 @@ static int ufshcd_alloc_mcq(struct ufs_hba *hba)
>   	if (ret)
>   		goto err;
>   
> +	if (hba->nutrs * hba->nr_hw_queues > SZ_64K - 1) {
> +		dev_info(hba->dev, "there can be at most 64K inflight requests\n");
> +		ret = -EINVAL;
> +		goto err;
> +	}
> +
>   	/*
>   	 * Previously allocated memory for nutrs may not be enough in MCQ mode.
>   	 * Number of supported tags in MCQ mode may be larger than SDB mode.

RE: [PATCH v3] scsi: ufs: mcq: Limit the amount of inflight requests

[Avri Altman]

> On 20/04/2023 11:11, Avri Altman wrote:
> > in UFS, each request is designated via the triplet <iid, lun, task tag>.
> >
> > In UFS4.0 the Initiator ID field is 8 bits wide, comprised of the
> > EXT_IID and IID fields. Together with the task tag (single byte), they
> > limit the driver's hw queues capacity.
> >
> > The purpose of this patch is to document the ufs spec restrictions.
> > practically, it impose no functional change because a) the current hw
> > supports much less (merely 64 requests), and b) since
> > shost->host_tagset is set - then the number of inflight IOs is limited
> > by just the HW queue depth.
> >
> 
> I don't know why effectively it has no change, as it seems to me to be checking
> and failing something valid which could happen.
> 
> If it really cannot happen, then we generally don't add what is effectively
> dead code just to provide practical illustrations of the spec. And it is
> misleading to others who may use your driver as a reference.
Better to just drop it then.

Thanks,
Avri

> 
> A code comment would be a better choice, if anything, IMHO.
> 
> Thanks,
> John
> 
> > ---
> > v2 -> v3:
> > fix smatch warnings (Dan Carpenter)
> > elaborate commit log (John Garry)
> >
> > v1 -> v2:
> > Attend Johannes's and Bart's comments
> 
> This is the wrong place for these
> 
> >
> > Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> > Signed-off-by: Avri Altman <avri.altman@wdc.com>
> > ---
> >   drivers/ufs/core/ufshcd.c | 6 ++++++
> >   1 file changed, 6 insertions(+)
> >
> > diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
> > index 03c47f9a2750..9cebec6be35c 100644
> > --- a/drivers/ufs/core/ufshcd.c
> > +++ b/drivers/ufs/core/ufshcd.c
> > @@ -8468,6 +8468,12 @@ static int ufshcd_alloc_mcq(struct ufs_hba *hba)
> >       if (ret)
> >               goto err;
> >
> > +     if (hba->nutrs * hba->nr_hw_queues > SZ_64K - 1) {
> > +             dev_info(hba->dev, "there can be at most 64K inflight requests\n");
> > +             ret = -EINVAL;
> > +             goto err;
> > +     }
> > +
> >       /*
> >        * Previously allocated memory for nutrs may not be enough in MCQ
> mode.
> >        * Number of supported tags in MCQ mode may be larger than SDB
> mode.