| Message ID | 1402990723-28138-5-git-send-email-boris.brezillon@free-electrons.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, Jun 17, 2014 at 09:38:40AM +0200, Boris BREZILLON wrote: > The init_data and of_node fields of the axp2xx_matches tables are filled > at each device probe by the axp20x_regulator_parse_dt function (which then > calls the of_regulator_match function). > This means we can probe a new device and consider data initialized during > the probe of another device as valid. > > Reset init_data and of_node field to NULL before each probe in order to > avoid this kind of issue. > > Signed-off-by: Boris BREZILLON <boris.brezillon@free-electrons.com> > --- > drivers/regulator/axp20x-regulator.c | 9 +++++++++ > 1 file changed, 9 insertions(+) > > diff --git a/drivers/regulator/axp20x-regulator.c b/drivers/regulator/axp20x-regulator.c > index 7a30f49..d42bf6d 100644 > --- a/drivers/regulator/axp20x-regulator.c > +++ b/drivers/regulator/axp20x-regulator.c > @@ -324,6 +324,15 @@ static int axp20x_regulator_probe(struct platform_device *pdev) > nregulators = AXP20X_REG_ID_MAX; > } > > + /* > + * Reset matches table (this table might have been modified by a > + * previous AXP2xx device probe). > + */ > + for (i = 0; i < nmatches; i++) { > + matches[i].init_data = NULL; > + matches[i].of_node = NULL; > + } > + That looks rather hackish, especially since we've never been in such a case yet, since we have a single PMIC in our system. Can't you just use memzero here? Maxime
On 17/06/2014 22:44, Maxime Ripard wrote: > On Tue, Jun 17, 2014 at 09:38:40AM +0200, Boris BREZILLON wrote: >> The init_data and of_node fields of the axp2xx_matches tables are filled >> at each device probe by the axp20x_regulator_parse_dt function (which then >> calls the of_regulator_match function). >> This means we can probe a new device and consider data initialized during >> the probe of another device as valid. >> >> Reset init_data and of_node field to NULL before each probe in order to >> avoid this kind of issue. >> >> Signed-off-by: Boris BREZILLON <boris.brezillon@free-electrons.com> >> --- >> drivers/regulator/axp20x-regulator.c | 9 +++++++++ >> 1 file changed, 9 insertions(+) >> >> diff --git a/drivers/regulator/axp20x-regulator.c b/drivers/regulator/axp20x-regulator.c >> index 7a30f49..d42bf6d 100644 >> --- a/drivers/regulator/axp20x-regulator.c >> +++ b/drivers/regulator/axp20x-regulator.c >> @@ -324,6 +324,15 @@ static int axp20x_regulator_probe(struct platform_device *pdev) >> nregulators = AXP20X_REG_ID_MAX; >> } >> >> + /* >> + * Reset matches table (this table might have been modified by a >> + * previous AXP2xx device probe). >> + */ >> + for (i = 0; i < nmatches; i++) { >> + matches[i].init_data = NULL; >> + matches[i].of_node = NULL; >> + } >> + > That looks rather hackish, especially since we've never been in such a > case yet, since we have a single PMIC in our system. Even if something is unlikely to happen, it doesn't mean it's impossible. I'm pretty sure there are (or will be) some systems containing several identical PMICs in the wild, and fixing this possible bug now prevents us (or other developers) from having a big headache debugging this in the future. BTW, what is hackish in this code ? > > Can't you just use memzero here? We can't just memset the whole matches table to zero, because some fields (name and driver_data) are statically assigned. > > Maxime >
Hi, On 06/18/2014 09:11 AM, Boris BREZILLON wrote: > > On 17/06/2014 22:44, Maxime Ripard wrote: >> On Tue, Jun 17, 2014 at 09:38:40AM +0200, Boris BREZILLON wrote: >>> The init_data and of_node fields of the axp2xx_matches tables are filled >>> at each device probe by the axp20x_regulator_parse_dt function (which then >>> calls the of_regulator_match function). >>> This means we can probe a new device and consider data initialized during >>> the probe of another device as valid. >>> >>> Reset init_data and of_node field to NULL before each probe in order to >>> avoid this kind of issue. >>> >>> Signed-off-by: Boris BREZILLON <boris.brezillon@free-electrons.com> >>> --- >>> drivers/regulator/axp20x-regulator.c | 9 +++++++++ >>> 1 file changed, 9 insertions(+) >>> >>> diff --git a/drivers/regulator/axp20x-regulator.c b/drivers/regulator/axp20x-regulator.c >>> index 7a30f49..d42bf6d 100644 >>> --- a/drivers/regulator/axp20x-regulator.c >>> +++ b/drivers/regulator/axp20x-regulator.c >>> @@ -324,6 +324,15 @@ static int axp20x_regulator_probe(struct platform_device *pdev) >>> nregulators = AXP20X_REG_ID_MAX; >>> } >>> >>> + /* >>> + * Reset matches table (this table might have been modified by a >>> + * previous AXP2xx device probe). >>> + */ >>> + for (i = 0; i < nmatches; i++) { >>> + matches[i].init_data = NULL; >>> + matches[i].of_node = NULL; >>> + } >>> + >> That looks rather hackish, especially since we've never been in such a >> case yet, since we have a single PMIC in our system. > > Even if something is unlikely to happen, it doesn't mean it's impossible. > I'm pretty sure there are (or will be) some systems containing several > identical PMICs in the wild, and fixing this possible bug now prevents > us (or other developers) from having a big headache debugging this in > the future. If you're really worried about this, you should also be worried about 2 probes running at the same time racing against each other (I know the bus level code will not do that now, but what about the future). If you cannot treat / use the global struct as const, then you really should have a local copy, and const-ify the global version and use it as a template to initialize the local copy. > BTW, what is hackish in this code ? See above, changing a global struct, and then re-initializing it on the next probe just is not pretty. TBH this raised my eyebrows the first time you posted it already, but I decided to let it be. But since we're discussing this now anyways I have to agree with Maxime. Regards, Hans
On Wed, Jun 18, 2014 at 09:11:59AM +0200, Boris BREZILLON wrote: > > On 17/06/2014 22:44, Maxime Ripard wrote: > > On Tue, Jun 17, 2014 at 09:38:40AM +0200, Boris BREZILLON wrote: > >> The init_data and of_node fields of the axp2xx_matches tables are filled > >> at each device probe by the axp20x_regulator_parse_dt function (which then > >> calls the of_regulator_match function). > >> This means we can probe a new device and consider data initialized during > >> the probe of another device as valid. > >> > >> Reset init_data and of_node field to NULL before each probe in order to > >> avoid this kind of issue. > >> > >> Signed-off-by: Boris BREZILLON <boris.brezillon@free-electrons.com> > >> --- > >> drivers/regulator/axp20x-regulator.c | 9 +++++++++ > >> 1 file changed, 9 insertions(+) > >> > >> diff --git a/drivers/regulator/axp20x-regulator.c b/drivers/regulator/axp20x-regulator.c > >> index 7a30f49..d42bf6d 100644 > >> --- a/drivers/regulator/axp20x-regulator.c > >> +++ b/drivers/regulator/axp20x-regulator.c > >> @@ -324,6 +324,15 @@ static int axp20x_regulator_probe(struct platform_device *pdev) > >> nregulators = AXP20X_REG_ID_MAX; > >> } > >> > >> + /* > >> + * Reset matches table (this table might have been modified by a > >> + * previous AXP2xx device probe). > >> + */ > >> + for (i = 0; i < nmatches; i++) { > >> + matches[i].init_data = NULL; > >> + matches[i].of_node = NULL; > >> + } > >> + > > That looks rather hackish, especially since we've never been in such a > > case yet, since we have a single PMIC in our system. > > Even if something is unlikely to happen, it doesn't mean it's impossible. > I'm pretty sure there are (or will be) some systems containing several > identical PMICs in the wild, and fixing this possible bug now prevents > us (or other developers) from having a big headache debugging this in > the future. > > BTW, what is hackish in this code ? Pretty what Hans was saying, either you think that there will only be one single instance of the driver, and using a global definition is fine, or you can have several instances of the driver, and in this case you'll use a dynamic allocation, but you seem to be stuck in between. I understand that you might not want to redeclare by hand the whole match content, so maybe you can just use memcpy from the global definition then. Maxime
diff --git a/drivers/regulator/axp20x-regulator.c b/drivers/regulator/axp20x-regulator.c index 7a30f49..d42bf6d 100644 --- a/drivers/regulator/axp20x-regulator.c +++ b/drivers/regulator/axp20x-regulator.c @@ -324,6 +324,15 @@ static int axp20x_regulator_probe(struct platform_device *pdev) nregulators = AXP20X_REG_ID_MAX; } + /* + * Reset matches table (this table might have been modified by a + * previous AXP2xx device probe). + */ + for (i = 0; i < nmatches; i++) { + matches[i].init_data = NULL; + matches[i].of_node = NULL; + } + ret = axp20x_regulator_parse_dt(pdev, matches, nmatches); if (ret) return ret;
The init_data and of_node fields of the axp2xx_matches tables are filled at each device probe by the axp20x_regulator_parse_dt function (which then calls the of_regulator_match function). This means we can probe a new device and consider data initialized during the probe of another device as valid. Reset init_data and of_node field to NULL before each probe in order to avoid this kind of issue. Signed-off-by: Boris BREZILLON <boris.brezillon@free-electrons.com> --- drivers/regulator/axp20x-regulator.c | 9 +++++++++ 1 file changed, 9 insertions(+)