Message ID | 20220917104311.1878250-7-yanaijie@huawei.com |
---|---|
State | New |
Headers | show |
Series | scsi: libsas: sas address comparation refactor | expand |
On 17/09/2022 11:43, Jason Yan wrote: > The sas address comparation of domain device and expander phy is open > coded. Now we can replace it with dev_and_phy_addr_same(). > > Signed-off-by: Jason Yan <yanaijie@huawei.com> > --- > drivers/scsi/pm8001/pm8001_sas.c | 3 +-- > 1 file changed, 1 insertion(+), 2 deletions(-) > > diff --git a/drivers/scsi/pm8001/pm8001_sas.c b/drivers/scsi/pm8001/pm8001_sas.c > index 8e3f2f9ddaac..bb1b1722f3ee 100644 > --- a/drivers/scsi/pm8001/pm8001_sas.c > +++ b/drivers/scsi/pm8001/pm8001_sas.c > @@ -649,8 +649,7 @@ static int pm8001_dev_found_notify(struct domain_device *dev) > for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; This code seems the same between many libsas LLDDs - could we factor it out into libsas? If so, then maybe those new helpers could be put in sas_internal.h Thanks, John > phy_id++) { > phy = &parent_dev->ex_dev.ex_phy[phy_id]; > - if (SAS_ADDR(phy->attached_sas_addr) > - == SAS_ADDR(dev->sas_addr)) { > + if (dev_and_phy_addr_same(dev, phy)) { > pm8001_device->attached_phy = phy_id; > break; > }
On 2022/9/22 22:24, John Garry wrote: > On 17/09/2022 11:43, Jason Yan wrote: >> The sas address comparation of domain device and expander phy is open >> coded. Now we can replace it with dev_and_phy_addr_same(). >> >> Signed-off-by: Jason Yan <yanaijie@huawei.com> >> --- >> drivers/scsi/pm8001/pm8001_sas.c | 3 +-- >> 1 file changed, 1 insertion(+), 2 deletions(-) >> >> diff --git a/drivers/scsi/pm8001/pm8001_sas.c >> b/drivers/scsi/pm8001/pm8001_sas.c >> index 8e3f2f9ddaac..bb1b1722f3ee 100644 >> --- a/drivers/scsi/pm8001/pm8001_sas.c >> +++ b/drivers/scsi/pm8001/pm8001_sas.c >> @@ -649,8 +649,7 @@ static int pm8001_dev_found_notify(struct >> domain_device *dev) >> for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; > > This code seems the same between many libsas LLDDs - could we factor it > out into libsas? Sure we can. I will try to factor it out in the next revision. Thanks, Jason If so, then maybe those new helpers could be put in > sas_internal.h > > Thanks, > John > >> phy_id++) { >> phy = &parent_dev->ex_dev.ex_phy[phy_id]; >> - if (SAS_ADDR(phy->attached_sas_addr) >> - == SAS_ADDR(dev->sas_addr)) { >> + if (dev_and_phy_addr_same(dev, phy)) { >> pm8001_device->attached_phy = phy_id; >> break; >> } > > .
On 2022/9/22 22:24, John Garry wrote: > On 17/09/2022 11:43, Jason Yan wrote: >> The sas address comparation of domain device and expander phy is open >> coded. Now we can replace it with dev_and_phy_addr_same(). >> >> Signed-off-by: Jason Yan <yanaijie@huawei.com> >> --- >> drivers/scsi/pm8001/pm8001_sas.c | 3 +-- >> 1 file changed, 1 insertion(+), 2 deletions(-) >> >> diff --git a/drivers/scsi/pm8001/pm8001_sas.c >> b/drivers/scsi/pm8001/pm8001_sas.c >> index 8e3f2f9ddaac..bb1b1722f3ee 100644 >> --- a/drivers/scsi/pm8001/pm8001_sas.c >> +++ b/drivers/scsi/pm8001/pm8001_sas.c >> @@ -649,8 +649,7 @@ static int pm8001_dev_found_notify(struct >> domain_device *dev) >> for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; > > This code seems the same between many libsas LLDDs - could we factor it > out into libsas? If so, then maybe those new helpers could be put in > sas_internal.h For the part of putting helpers in sas_internal.h, this needs to make the helpers exported. I think it's not worth to do this because they are very small. I'd still like to make them inline functions in libsas.h such as: diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h index 2dbead74a2af..e9e76c898287 100644 --- a/include/scsi/libsas.h +++ b/include/scsi/libsas.h @@ -648,6 +648,22 @@ static inline bool sas_is_internal_abort(struct sas_task *task) return task->task_proto == SAS_PROTOCOL_INTERNAL_ABORT; } +static inline int sas_find_attathed_phy(struct expander_device *ex_dev, + struct domain_device *dev) +{ + struct ex_phy *phy; + int phy_id; + + for (phy_id = 0; phy_id < ex_dev->num_phys; phy_id++) { + phy = &ex_dev->ex_phy[phy_id]; + if (SAS_ADDR(phy->attached_sas_addr) + == SAS_ADDR(dev->sas_addr)) + return phy_id; + } + + return ex_dev->num_phys; +} + struct sas_domain_function_template { /* The class calls these to notify the LLDD of an event. */ void (*lldd_port_formed)(struct asd_sas_phy *); And the LLDDs change like: diff --git a/drivers/scsi/pm8001/pm8001_sas.c b/drivers/scsi/pm8001/pm8001_sas.c index 8e3f2f9ddaac..4e7350609b3d 100644 --- a/drivers/scsi/pm8001/pm8001_sas.c +++ b/drivers/scsi/pm8001/pm8001_sas.c @@ -645,16 +645,8 @@ static int pm8001_dev_found_notify(struct domain_device *dev) pm8001_device->dcompletion = &completion; if (parent_dev && dev_is_expander(parent_dev->dev_type)) { int phy_id; - struct ex_phy *phy; - for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; - phy_id++) { - phy = &parent_dev->ex_dev.ex_phy[phy_id]; - if (SAS_ADDR(phy->attached_sas_addr) - == SAS_ADDR(dev->sas_addr)) { - pm8001_device->attached_phy = phy_id; - break; - } - } + + phy_id = sas_find_attathed_phy(&parent_dev->ex_dev, dev); if (phy_id == parent_dev->ex_dev.num_phys) { pm8001_dbg(pm8001_ha, FAIL, "Error: no attached dev:%016llx at ex:%016llx.\n", @@ -662,6 +654,7 @@ static int pm8001_dev_found_notify(struct domain_device *dev) SAS_ADDR(parent_dev->sas_addr)); res = -1; } + pm8001_device->attached_phy = phy_id; } else { if (dev->dev_type == SAS_SATA_DEV) { pm8001_device->attached_phy = So I wonder if you have any reasons to insist exporting the helper? > > Thanks, > John > >> phy_id++) { >> phy = &parent_dev->ex_dev.ex_phy[phy_id]; >> - if (SAS_ADDR(phy->attached_sas_addr) >> - == SAS_ADDR(dev->sas_addr)) { >> + if (dev_and_phy_addr_same(dev, phy)) { >> pm8001_device->attached_phy = phy_id; >> break; >> } > > .
On 23/09/2022 10:44, Jason Yan wrote: >>> for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; >> >> This code seems the same between many libsas LLDDs - could we factor >> it out into libsas? If so, then maybe those new helpers could be put >> in sas_internal.h > > For the part of putting helpers in sas_internal.h, this needs to make > the helpers exported. Please explain why. I would assume that if those helpers were only used in libsas code (and not LLDDs) then they could be put in sas_internal.h and no need for export > I think it's not worth to do this because they are > very small. I'd still like to make them inline functions in libsas.h > such as: > > > diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h > index 2dbead74a2af..e9e76c898287 100644 > --- a/include/scsi/libsas.h > +++ b/include/scsi/libsas.h > @@ -648,6 +648,22 @@ static inline bool sas_is_internal_abort(struct > sas_task *task) > return task->task_proto == SAS_PROTOCOL_INTERNAL_ABORT; > } > > +static inline int sas_find_attathed_phy(struct expander_device *ex_dev, > + struct domain_device *dev) > +{ > + struct ex_phy *phy; > + int phy_id; > + > + for (phy_id = 0; phy_id < ex_dev->num_phys; phy_id++) { > + phy = &ex_dev->ex_phy[phy_id]; > + if (SAS_ADDR(phy->attached_sas_addr) > + == SAS_ADDR(dev->sas_addr)) > + return phy_id; > + } > + > + return ex_dev->num_phys; I will note that this code does not use your new helpers > +} > + > struct sas_domain_function_template { > /* The class calls these to notify the LLDD of an event. */ > void (*lldd_port_formed)(struct asd_sas_phy *); > > > > And the LLDDs change like: > > > diff --git a/drivers/scsi/pm8001/pm8001_sas.c > b/drivers/scsi/pm8001/pm8001_sas.c > index 8e3f2f9ddaac..4e7350609b3d 100644 > --- a/drivers/scsi/pm8001/pm8001_sas.c > +++ b/drivers/scsi/pm8001/pm8001_sas.c > @@ -645,16 +645,8 @@ static int pm8001_dev_found_notify(struct > domain_device *dev) > pm8001_device->dcompletion = &completion; > if (parent_dev && dev_is_expander(parent_dev->dev_type)) { > int phy_id; > - struct ex_phy *phy; > - for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; > - phy_id++) { > - phy = &parent_dev->ex_dev.ex_phy[phy_id]; > - if (SAS_ADDR(phy->attached_sas_addr) > - == SAS_ADDR(dev->sas_addr)) { > - pm8001_device->attached_phy = phy_id; > - break; > - } > - } > + > + phy_id = sas_find_attathed_phy(&parent_dev->ex_dev, dev); > if (phy_id == parent_dev->ex_dev.num_phys) { > pm8001_dbg(pm8001_ha, FAIL, > "Error: no attached dev:%016llx at > ex:%016llx.\n", > @@ -662,6 +654,7 @@ static int pm8001_dev_found_notify(struct > domain_device *dev) > SAS_ADDR(parent_dev->sas_addr)); > res = -1; > } > + pm8001_device->attached_phy = phy_id; > } else { > if (dev->dev_type == SAS_SATA_DEV) { > pm8001_device->attached_phy = > > > So I wonder if you have any reasons to insist exporting the helper Thanks, John
On 2022/9/23 18:00, John Garry wrote: > On 23/09/2022 10:44, Jason Yan wrote: >>>> for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; >>> >>> This code seems the same between many libsas LLDDs - could we factor >>> it out into libsas? If so, then maybe those new helpers could be put >>> in sas_internal.h >> >> For the part of putting helpers in sas_internal.h, this needs to make >> the helpers exported. > > Please explain why. > > I would assume that if those helpers were only used in libsas code (and > not LLDDs) then they could be put in sas_internal.h and no need for export > >> I think it's not worth to do this because they are very small. I'd >> still like to make them inline functions in libsas.h such as: >> >> >> diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h >> index 2dbead74a2af..e9e76c898287 100644 >> --- a/include/scsi/libsas.h >> +++ b/include/scsi/libsas.h >> @@ -648,6 +648,22 @@ static inline bool sas_is_internal_abort(struct >> sas_task *task) >> return task->task_proto == SAS_PROTOCOL_INTERNAL_ABORT; >> } >> >> +static inline int sas_find_attathed_phy(struct expander_device *ex_dev, >> + struct domain_device *dev) >> +{ >> + struct ex_phy *phy; >> + int phy_id; >> + >> + for (phy_id = 0; phy_id < ex_dev->num_phys; phy_id++) { >> + phy = &ex_dev->ex_phy[phy_id]; >> + if (SAS_ADDR(phy->attached_sas_addr) >> + == SAS_ADDR(dev->sas_addr)) >> + return phy_id; >> + } >> + >> + return ex_dev->num_phys; > > I will note that this code does not use your new helpers NO, this code above will use my new helpers.(Unless we skip this part) diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h index 635039557bdb..9283462704f0 100644 --- a/include/scsi/libsas.h +++ b/include/scsi/libsas.h @@ -673,8 +673,7 @@ static inline int sas_find_attathed_phy(struct expander_device *ex_dev, for (phy_id = 0; phy_id < ex_dev->num_phys; phy_id++) { phy = &ex_dev->ex_phy[phy_id]; - if (SAS_ADDR(phy->attached_sas_addr) - == SAS_ADDR(dev->sas_addr)) + if (sas_phy_match_dev_addr(dev, phy)) return phy_id; > >> +} >> + >> struct sas_domain_function_template { >> /* The class calls these to notify the LLDD of an event. */ >> void (*lldd_port_formed)(struct asd_sas_phy *); >> >> >> >> And the LLDDs change like: >> >> >> diff --git a/drivers/scsi/pm8001/pm8001_sas.c >> b/drivers/scsi/pm8001/pm8001_sas.c >> index 8e3f2f9ddaac..4e7350609b3d 100644 >> --- a/drivers/scsi/pm8001/pm8001_sas.c >> +++ b/drivers/scsi/pm8001/pm8001_sas.c >> @@ -645,16 +645,8 @@ static int pm8001_dev_found_notify(struct >> domain_device *dev) >> pm8001_device->dcompletion = &completion; >> if (parent_dev && dev_is_expander(parent_dev->dev_type)) { >> int phy_id; >> - struct ex_phy *phy; >> - for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; >> - phy_id++) { >> - phy = &parent_dev->ex_dev.ex_phy[phy_id]; >> - if (SAS_ADDR(phy->attached_sas_addr) >> - == SAS_ADDR(dev->sas_addr)) { >> - pm8001_device->attached_phy = phy_id; >> - break; >> - } >> - } >> + >> + phy_id = sas_find_attathed_phy(&parent_dev->ex_dev, dev); >> if (phy_id == parent_dev->ex_dev.num_phys) { >> pm8001_dbg(pm8001_ha, FAIL, >> "Error: no attached dev:%016llx at >> ex:%016llx.\n", >> @@ -662,6 +654,7 @@ static int pm8001_dev_found_notify(struct >> domain_device *dev) >> SAS_ADDR(parent_dev->sas_addr)); >> res = -1; >> } >> + pm8001_device->attached_phy = phy_id; >> } else { >> if (dev->dev_type == SAS_SATA_DEV) { >> pm8001_device->attached_phy = >> >> >> So I wonder if you have any reasons to insist exporting the helper > Thanks, > John > .
On 2022/9/23 18:00, John Garry wrote: > On 23/09/2022 10:44, Jason Yan wrote: >>>> for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; >>> >>> This code seems the same between many libsas LLDDs - could we factor >>> it out into libsas? If so, then maybe those new helpers could be put >>> in sas_internal.h >> >> For the part of putting helpers in sas_internal.h, this needs to make >> the helpers exported. > > Please explain why. > > I would assume that if those helpers were only used in libsas code (and > not LLDDs) then they could be put in sas_internal.h and no need for export > Sorry, I did not make it clear. I mean we need to export sas_find_attathed_phy() below. Not the sas address comparation helpers. >> I think it's not worth to do this because they are very small. I'd >> still like to make them inline functions in libsas.h such as: >> >> >> diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h >> index 2dbead74a2af..e9e76c898287 100644 >> --- a/include/scsi/libsas.h >> +++ b/include/scsi/libsas.h >> @@ -648,6 +648,22 @@ static inline bool sas_is_internal_abort(struct >> sas_task *task) >> return task->task_proto == SAS_PROTOCOL_INTERNAL_ABORT; >> } >> >> +static inline int sas_find_attathed_phy(struct expander_device *ex_dev, >> + struct domain_device *dev) >> +{ >> + struct ex_phy *phy; >> + int phy_id; >> + >> + for (phy_id = 0; phy_id < ex_dev->num_phys; phy_id++) { >> + phy = &ex_dev->ex_phy[phy_id]; >> + if (SAS_ADDR(phy->attached_sas_addr) >> + == SAS_ADDR(dev->sas_addr)) >> + return phy_id; >> + } >> + >> + return ex_dev->num_phys; > > I will note that this code does not use your new helpers > >> +} >> + >> struct sas_domain_function_template { >> /* The class calls these to notify the LLDD of an event. */ >> void (*lldd_port_formed)(struct asd_sas_phy *); >> >> >> >> And the LLDDs change like: >> >> >> diff --git a/drivers/scsi/pm8001/pm8001_sas.c >> b/drivers/scsi/pm8001/pm8001_sas.c >> index 8e3f2f9ddaac..4e7350609b3d 100644 >> --- a/drivers/scsi/pm8001/pm8001_sas.c >> +++ b/drivers/scsi/pm8001/pm8001_sas.c >> @@ -645,16 +645,8 @@ static int pm8001_dev_found_notify(struct >> domain_device *dev) >> pm8001_device->dcompletion = &completion; >> if (parent_dev && dev_is_expander(parent_dev->dev_type)) { >> int phy_id; >> - struct ex_phy *phy; >> - for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; >> - phy_id++) { >> - phy = &parent_dev->ex_dev.ex_phy[phy_id]; >> - if (SAS_ADDR(phy->attached_sas_addr) >> - == SAS_ADDR(dev->sas_addr)) { >> - pm8001_device->attached_phy = phy_id; >> - break; >> - } >> - } >> + >> + phy_id = sas_find_attathed_phy(&parent_dev->ex_dev, dev); >> if (phy_id == parent_dev->ex_dev.num_phys) { >> pm8001_dbg(pm8001_ha, FAIL, >> "Error: no attached dev:%016llx at >> ex:%016llx.\n", >> @@ -662,6 +654,7 @@ static int pm8001_dev_found_notify(struct >> domain_device *dev) >> SAS_ADDR(parent_dev->sas_addr)); >> res = -1; >> } >> + pm8001_device->attached_phy = phy_id; >> } else { >> if (dev->dev_type == SAS_SATA_DEV) { >> pm8001_device->attached_phy = >> >> >> So I wonder if you have any reasons to insist exporting the helper > Thanks, > John > .
On 23/09/2022 11:13, Jason Yan wrote: >> >> Please explain why. >> >> I would assume that if those helpers were only used in libsas code >> (and not LLDDs) then they could be put in sas_internal.h and no need >> for export >> > > > Sorry, I did not make it clear. I mean we need to export > sas_find_attathed_phy() below. Not the sas address comparation helpers. That seems fine to me. About sas_find_attathed_phy() implementation, > +static inline int sas_find_attathed_phy(struct expander_device *ex_dev, > + struct domain_device *dev) > +{ > + struct ex_phy *phy; > + int phy_id; > + > + for (phy_id = 0; phy_id < ex_dev->num_phys; phy_id++) { > + phy = &ex_dev->ex_phy[phy_id]; > + if (SAS_ADDR(phy->attached_sas_addr) > + == SAS_ADDR(dev->sas_addr)) > + return phy_id; > + } > + > + return ex_dev->num_phys; Returning ex_dev->num_phys would seem ok, but then the LLDD needs to check that return against ex_dev->num_phys. It seems ok, but I'm still not 100% comfortable with that. Maybe returning -ENODEV may be better. Or return boolean and pass phy_id as pointer to be filled in when returning true. > +} Thanks, John
On 2022/9/23 18:30, John Garry wrote: > On 23/09/2022 11:13, Jason Yan wrote: >>> >>> Please explain why. >>> >>> I would assume that if those helpers were only used in libsas code >>> (and not LLDDs) then they could be put in sas_internal.h and no need >>> for export >>> >> >> >> Sorry, I did not make it clear. I mean we need to export >> sas_find_attathed_phy() below. Not the sas address comparation helpers. > > That seems fine to me. > > About sas_find_attathed_phy() implementation, > > > +static inline int sas_find_attathed_phy(struct expander_device *ex_dev, > > + struct domain_device *dev) > > +{ > > + struct ex_phy *phy; > > + int phy_id; > > + > > + for (phy_id = 0; phy_id < ex_dev->num_phys; phy_id++) { > > + phy = &ex_dev->ex_phy[phy_id]; > > + if (SAS_ADDR(phy->attached_sas_addr) > > + == SAS_ADDR(dev->sas_addr)) > > + return phy_id; > > + } > > + > > + return ex_dev->num_phys; > > Returning ex_dev->num_phys would seem ok, but then the LLDD needs to > check that return against ex_dev->num_phys. It seems ok, but I'm still > not 100% comfortable with that. Maybe returning -ENODEV may be better. > > Or return boolean and pass phy_id as pointer to be filled in when > returning true. > I've been thinking about this for a while too. Thank you for the advise. Thanks, Jason > > +} > > Thanks, > John > > .
diff --git a/drivers/scsi/pm8001/pm8001_sas.c b/drivers/scsi/pm8001/pm8001_sas.c index 8e3f2f9ddaac..bb1b1722f3ee 100644 --- a/drivers/scsi/pm8001/pm8001_sas.c +++ b/drivers/scsi/pm8001/pm8001_sas.c @@ -649,8 +649,7 @@ static int pm8001_dev_found_notify(struct domain_device *dev) for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys; phy_id++) { phy = &parent_dev->ex_dev.ex_phy[phy_id]; - if (SAS_ADDR(phy->attached_sas_addr) - == SAS_ADDR(dev->sas_addr)) { + if (dev_and_phy_addr_same(dev, phy)) { pm8001_device->attached_phy = phy_id; break; }
The sas address comparation of domain device and expander phy is open coded. Now we can replace it with dev_and_phy_addr_same(). Signed-off-by: Jason Yan <yanaijie@huawei.com> --- drivers/scsi/pm8001/pm8001_sas.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)