| Message ID | 20210118155242.7172-2-jbx6244@gmail.com |
|---|---|
| State | Accepted |
| Commit | 1034e2b6b8435758c0d699b77761365fd09a5f4a |
| Headers | show |
| Series | None | expand |
On Mon, Jan 18, 2021 at 4:52 PM Johan Jonker <jbx6244@gmail.com> wrote: > > Recently introduced async probe on mmc devices can shuffle block IDs. > Pin them to fixed values to ease booting in environments where UUIDs are > not practical. Use newly introduced aliases for mmcblk devices from [1]. > The sort order is based on reg address. > > [1] https://patchwork.kernel.org/patch/11747669/ I just saw this in the pull request: > Signed-off-by: Johan Jonker <jbx6244@gmail.com> > --- > arch/arm/boot/dts/rv1108.dtsi | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/arch/arm/boot/dts/rv1108.dtsi b/arch/arm/boot/dts/rv1108.dtsi > index bec47e0be..a754851f4 100644 > --- a/arch/arm/boot/dts/rv1108.dtsi > +++ b/arch/arm/boot/dts/rv1108.dtsi > @@ -19,6 +19,9 @@ > i2c1 = &i2c1; > i2c2 = &i2c2; > i2c3 = &i2c3; > + mmc0 = &emmc; > + mmc1 = &sdio; > + mmc2 = &sdmmc; > serial0 = &uart0; > serial1 = &uart1; > serial2 = &uart2; Please don't put these aliases into a .dtsi file, as not every board will provide each instance. The entire point of the aliases is to have sane enumeration, so you should start at index 0 for the first one that is actually present and count up from there. I would suggest you move these aliases into the .dts files for the existing boards for the next cycle, and then make sure only the ones that are present have an alias. It might actually be a good idea to have a warning in dtc when there is an alias pointing to a status="disabled" device, but I suspect there would be a lot of fallout from that. Arnd
Am Dienstag, 9. Februar 2021, 23:25:40 CET schrieb Arnd Bergmann: > On Mon, Jan 18, 2021 at 4:52 PM Johan Jonker <jbx6244@gmail.com> wrote: > > > > Recently introduced async probe on mmc devices can shuffle block IDs. > > Pin them to fixed values to ease booting in environments where UUIDs are > > not practical. Use newly introduced aliases for mmcblk devices from [1]. > > The sort order is based on reg address. > > > > [1] https://patchwork.kernel.org/patch/11747669/ > > I just saw this in the pull request: > > > Signed-off-by: Johan Jonker <jbx6244@gmail.com> > > --- > > arch/arm/boot/dts/rv1108.dtsi | 3 +++ > > 1 file changed, 3 insertions(+) > > > > diff --git a/arch/arm/boot/dts/rv1108.dtsi b/arch/arm/boot/dts/rv1108.dtsi > > index bec47e0be..a754851f4 100644 > > --- a/arch/arm/boot/dts/rv1108.dtsi > > +++ b/arch/arm/boot/dts/rv1108.dtsi > > @@ -19,6 +19,9 @@ > > i2c1 = &i2c1; > > i2c2 = &i2c2; > > i2c3 = &i2c3; > > + mmc0 = &emmc; > > + mmc1 = &sdio; > > + mmc2 = &sdmmc; > > serial0 = &uart0; > > serial1 = &uart1; > > serial2 = &uart2; > > Please don't put these aliases into a .dtsi file, as not every board > will provide each instance. The entire point of the aliases is to > have sane enumeration, so you should start at index 0 for the > first one that is actually present and count up from there. Hmm, right now I don't see the disadvantage of missing mmc numbers. As similarly we count i2c and serial numbers for a long time, even though not all of them appear on every board. Especially as the main goal is to simply have stable numbers and not having the mmc devices swap numbers on every boot. So right now we're not using them from a userspace POV but instead agreed on following the address ordering of the soc. so when ordering mmc controllers by their io-address, mmc0 is the first one, then mmc1, etc. So just for my understanding, what is different for mmc? I guess to guarantee ongoing numbering similar to sd{a,b,c,...} Or should all aliases be duplicted in each board dts and not live in any soc dtsi? Heiko > I would suggest you move these aliases into the .dts files for > the existing boards for the next cycle, and then make sure > only the ones that are present have an alias. > > It might actually be a good idea to have a warning in dtc when > there is an alias pointing to a status="disabled" device, but I > suspect there would be a lot of fallout from that. > > Arnd >
On Wed, Feb 10, 2021 at 12:50 AM Heiko Stübner <heiko@sntech.de> wrote: > Am Dienstag, 9. Februar 2021, 23:25:40 CET schrieb Arnd Bergmann: > > Hmm, right now I don't see the disadvantage of missing mmc numbers. It's inconsistent with the normal use of these aliases across other platforms. > As similarly we count i2c and serial numbers for a long time, even though > not all of them appear on every board. Yes, that is a similar mistake. > Especially as the main goal is to simply have stable numbers and > not having the mmc devices swap numbers on every boot. > > So right now we're not using them from a userspace POV but > instead agreed on following the address ordering of the soc. > so when ordering mmc controllers by their io-address, mmc0 > is the first one, then mmc1, etc. > > So just for my understanding, what is different for mmc? > I guess to guarantee ongoing numbering similar to sd{a,b,c,...} > Or should all aliases be duplicted in each board dts and not > live in any soc dtsi? Each board should have its own aliases node that describes exactly which of the devices are wired up on that board, and in which order. If there are connectors on the board that are labeled in some form, then the aliases are meant to match what is written on the board or in its documentation. Arnd
Am Mittwoch, 10. Februar 2021, 11:34:41 CET schrieb Arnd Bergmann: > On Wed, Feb 10, 2021 at 12:50 AM Heiko Stübner <heiko@sntech.de> wrote: > > Am Dienstag, 9. Februar 2021, 23:25:40 CET schrieb Arnd Bergmann: > > > > Hmm, right now I don't see the disadvantage of missing mmc numbers. > > It's inconsistent with the normal use of these aliases across other > platforms. > > > As similarly we count i2c and serial numbers for a long time, even though > > not all of them appear on every board. > > Yes, that is a similar mistake. > > > Especially as the main goal is to simply have stable numbers and > > not having the mmc devices swap numbers on every boot. > > > > So right now we're not using them from a userspace POV but > > instead agreed on following the address ordering of the soc. > > so when ordering mmc controllers by their io-address, mmc0 > > is the first one, then mmc1, etc. > > > > So just for my understanding, what is different for mmc? > > I guess to guarantee ongoing numbering similar to sd{a,b,c,...} > > Or should all aliases be duplicted in each board dts and not > > live in any soc dtsi? > > Each board should have its own aliases node that describes > exactly which of the devices are wired up on that board, and > in which order. If there are connectors on the board that > are labeled in some form, then the aliases are meant to > match what is written on the board or in its documentation. Then we're at least in the clear for i2c, serial and the rest ... as these are numbered in the soc documentation, and all boards I've seen so far use these number also to identify these in schematics. So an i2c2 is always i2c2 even if i2c1 is not populated. And of course doing i2c0 = &i2c2 would definitly confuse people to no end. So I guess we'll just move the mmc stuff over to board files.
On Wed, Feb 10, 2021 at 11:49 AM Heiko Stübner <heiko@sntech.de> wrote: > Am Mittwoch, 10. Februar 2021, 11:34:41 CET schrieb Arnd Bergmann: > > On Wed, Feb 10, 2021 at 12:50 AM Heiko Stübner <heiko@sntech.de> wrote: > > > Am Dienstag, 9. Februar 2021, 23:25:40 CET schrieb Arnd Bergmann: > > Each board should have its own aliases node that describes > > exactly which of the devices are wired up on that board, and > > in which order. If there are connectors on the board that > > are labeled in some form, then the aliases are meant to > > match what is written on the board or in its documentation. > > Then we're at least in the clear for i2c, serial and the rest ... as these > are numbered in the soc documentation, and all boards I've seen so > far use these number also to identify these in schematics. Ok, that is helpful. It would still be good to ensure that only aliases exist for nodes that are actually enabled. > So an i2c2 is always i2c2 even if i2c1 is not populated. > > And of course doing > i2c0 = &i2c2 > would definitly confuse people to no end. I think that's just an unfortunate choice of the labels here, it's sometimes easier to just refer to the node by the full path instead of the label. Arnd
diff --git a/arch/arm/boot/dts/rv1108.dtsi b/arch/arm/boot/dts/rv1108.dtsi index bec47e0be..a754851f4 100644 --- a/arch/arm/boot/dts/rv1108.dtsi +++ b/arch/arm/boot/dts/rv1108.dtsi @@ -19,6 +19,9 @@ i2c1 = &i2c1; i2c2 = &i2c2; i2c3 = &i2c3; + mmc0 = &emmc; + mmc1 = &sdio; + mmc2 = &sdmmc; serial0 = &uart0; serial1 = &uart1; serial2 = &uart2;
Recently introduced async probe on mmc devices can shuffle block IDs. Pin them to fixed values to ease booting in environments where UUIDs are not practical. Use newly introduced aliases for mmcblk devices from [1]. The sort order is based on reg address. [1] https://patchwork.kernel.org/patch/11747669/ Signed-off-by: Johan Jonker <jbx6244@gmail.com> --- arch/arm/boot/dts/rv1108.dtsi | 3 +++ 1 file changed, 3 insertions(+)