| Message ID | 1559747630-28065-2-git-send-email-suzuki.poulose@arm.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [01/13] acpi: utils: Cleanup acpi_dev_match_cb | expand |
On Wed, Jun 5, 2019 at 5:14 PM Suzuki K Poulose <suzuki.poulose@arm.com> wrote: > > acpi_dev_match_cb match function modifies the "data" argument > to pass on a result which could be easily deduced from the result > of the bus_find_device() call at the caller site. Clean this > up in preparation to convert the "match" argument for bus_find_device > to accept a "const" data pointer, similar to class_find_device. This > would allow consolidating the match routines for these two APIs. This changelog can be improved IMO. In fact, the final goal here is to pass (const void *) as the second argument to acpi_dev_match_cb() (which you could do right away in this patch if I'm not mistaken) which is because you want to modify the prototype of bus_find_device(). So why don't you write something like this in the changelog: "The prototype of bus_find_device() will be unified with that of class_find_device() subsequently, but for this purpose the callback functions passed to it need to take (const void *) as the second argument. Consequently, they cannot modify the memory pointed to by that argument which currently is not the case for acpi_dev_match_cb(). However, acpi_dev_match_cb() really need not modify the "match" object passed to it, because acpi_dev_get_first_match_dev() which uses it via bus_find_device() can easily convert the result of bus_find_device() into the pointer to return. For this reason, update acpi_dev_match_cb() to avoid the redundant memory updates and change the type of its second argument to (const void *)." > > Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> > Cc: Len Brown <lenb@kernel.org> > Cc: linux-acpi@vger.kernel.org > Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com> > --- > drivers/acpi/utils.c | 7 +------ > 1 file changed, 1 insertion(+), 6 deletions(-) > > diff --git a/drivers/acpi/utils.c b/drivers/acpi/utils.c > index 7def63a..1391b63 100644 > --- a/drivers/acpi/utils.c > +++ b/drivers/acpi/utils.c > @@ -725,8 +725,6 @@ bool acpi_dev_found(const char *hid) > EXPORT_SYMBOL(acpi_dev_found); > > struct acpi_dev_match_info { > - const char *dev_name; > - struct acpi_device *adev; > struct acpi_device_id hid[2]; > const char *uid; > s64 hrv; > @@ -746,9 +744,6 @@ static int acpi_dev_match_cb(struct device *dev, void *data) And why not to change the type of the second arg to "const void *data" here? > strcmp(adev->pnp.unique_id, match->uid))) > return 0; > > - match->dev_name = acpi_dev_name(adev); > - match->adev = adev; > - > if (match->hrv == -1) > return 1; > > @@ -818,7 +813,7 @@ acpi_dev_get_first_match_dev(const char *hid, const char *uid, s64 hrv) > match.hrv = hrv; > > dev = bus_find_device(&acpi_bus_type, NULL, &match, acpi_dev_match_cb); > - return dev ? match.adev : NULL; > + return dev ? to_acpi_device(dev) : NULL; > } > EXPORT_SYMBOL(acpi_dev_get_first_match_dev); > > --
Hi Rafael, On 06/06/2019 10:14, Rafael J. Wysocki wrote: > On Wed, Jun 5, 2019 at 5:14 PM Suzuki K Poulose <suzuki.poulose@arm.com> wrote: >> >> acpi_dev_match_cb match function modifies the "data" argument >> to pass on a result which could be easily deduced from the result >> of the bus_find_device() call at the caller site. Clean this >> up in preparation to convert the "match" argument for bus_find_device >> to accept a "const" data pointer, similar to class_find_device. This >> would allow consolidating the match routines for these two APIs. > > This changelog can be improved IMO. I agree. Will update it. > > In fact, the final goal here is to pass (const void *) as the second > argument to acpi_dev_match_cb() (which you could do right away in this > patch if I'm not mistaken) which is because you want to modify the > prototype of bus_find_device(). > > So why don't you write something like this in the changelog: > > "The prototype of bus_find_device() will be unified with that of > class_find_device() subsequently, but for this purpose the callback > functions passed to it need to take (const void *) as the second > argument. Consequently, they cannot modify the memory pointed to by > that argument which currently is not the case for acpi_dev_match_cb(). > However, acpi_dev_match_cb() really need not modify the "match" object > passed to it, because acpi_dev_get_first_match_dev() which uses it via > bus_find_device() can easily convert the result of bus_find_device() > into the pointer to return. Sure. > For this reason, update acpi_dev_match_cb() to avoid the redundant > memory updates and change the type of its second argument to (const > void *)." We can't do that quite yet, until we unify the prototype of the bus_find_device(). >> >> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> >> Cc: Len Brown <lenb@kernel.org> >> Cc: linux-acpi@vger.kernel.org >> Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com> >> --- >> drivers/acpi/utils.c | 7 +------ >> 1 file changed, 1 insertion(+), 6 deletions(-) >> >> diff --git a/drivers/acpi/utils.c b/drivers/acpi/utils.c >> index 7def63a..1391b63 100644 >> --- a/drivers/acpi/utils.c >> +++ b/drivers/acpi/utils.c >> @@ -725,8 +725,6 @@ bool acpi_dev_found(const char *hid) >> EXPORT_SYMBOL(acpi_dev_found); >> >> struct acpi_dev_match_info { >> - const char *dev_name; >> - struct acpi_device *adev; >> struct acpi_device_id hid[2]; >> const char *uid; >> s64 hrv; >> @@ -746,9 +744,6 @@ static int acpi_dev_match_cb(struct device *dev, void *data) > > And why not to change the type of the second arg to "const void *data" here? Because, that would conflict with what bus_find_device() expects now. We make the change only later. Since this change was a bit more intrusive than simply changing the type of the parameter, I kept it as a preparatory patch. Thanks for the review ! Suzuki
diff --git a/drivers/acpi/utils.c b/drivers/acpi/utils.c index 7def63a..1391b63 100644 --- a/drivers/acpi/utils.c +++ b/drivers/acpi/utils.c @@ -725,8 +725,6 @@ bool acpi_dev_found(const char *hid) EXPORT_SYMBOL(acpi_dev_found); struct acpi_dev_match_info { - const char *dev_name; - struct acpi_device *adev; struct acpi_device_id hid[2]; const char *uid; s64 hrv; @@ -746,9 +744,6 @@ static int acpi_dev_match_cb(struct device *dev, void *data) strcmp(adev->pnp.unique_id, match->uid))) return 0; - match->dev_name = acpi_dev_name(adev); - match->adev = adev; - if (match->hrv == -1) return 1; @@ -818,7 +813,7 @@ acpi_dev_get_first_match_dev(const char *hid, const char *uid, s64 hrv) match.hrv = hrv; dev = bus_find_device(&acpi_bus_type, NULL, &match, acpi_dev_match_cb); - return dev ? match.adev : NULL; + return dev ? to_acpi_device(dev) : NULL; } EXPORT_SYMBOL(acpi_dev_get_first_match_dev);
acpi_dev_match_cb match function modifies the "data" argument to pass on a result which could be easily deduced from the result of the bus_find_device() call at the caller site. Clean this up in preparation to convert the "match" argument for bus_find_device to accept a "const" data pointer, similar to class_find_device. This would allow consolidating the match routines for these two APIs. Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com> --- drivers/acpi/utils.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-) -- 2.7.4