| Message ID | 1483627851-17996-2-git-send-email-t.remmet@phytec.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
* Teresa Remmet <t.remmet@phytec.de> [170105 06:57]: > To improve NAND safety we updated the partition layout. > Added barebox backup partition and removed kernel and oftree > partition. They are kept in ubi now. What about the users with earlier partition tables? Please read about "flag day" changes, typically they are not acceptable. Regards, Tony
* Tony Lindgren <tony@atomide.com> [170105 07:37]: > * Teresa Remmet <t.remmet@phytec.de> [170105 06:57]: > > To improve NAND safety we updated the partition layout. > > Added barebox backup partition and removed kernel and oftree > > partition. They are kept in ubi now. > > What about the users with earlier partition tables? > > Please read about "flag day" changes, typically they are not > acceptable. Adding Brian and Adam to Cc. Can you guys come up with some solution on this? I'm suggesting we leave the kernel nodes empty and let u-boot populate them, so maybe you guys can discuss this on the related lists. The rest of the series looks fine to me so applying it into omap-for-v4.11/dt. Cheers, Tony
On Thu, Jan 05, 2017 at 09:18:45AM -0800, Tony Lindgren wrote: > * Tony Lindgren <tony@atomide.com> [170105 07:37]: > > * Teresa Remmet <t.remmet@phytec.de> [170105 06:57]: > > > To improve NAND safety we updated the partition layout. > > > Added barebox backup partition and removed kernel and oftree > > > partition. They are kept in ubi now. > > > > What about the users with earlier partition tables? > > > > Please read about "flag day" changes, typically they are not > > acceptable. > > Adding Brian and Adam to Cc. Can you guys come up with some > solution on this? I don't have much context for this thread, and no I don't plan to solve your problems for you. But I can provide tips! > I'm suggesting we leave the kernel nodes empty and let u-boot > populate them, so maybe you guys can discuss this on the related > lists. That's an option. I've worked with platforms that did something like this, and that's really one of the only ways you can handle putting partition information in the device tree. You're really hamstringing yourself when you put all the partition information in the device tree. And it's just dumb once that gets codified in the kernel source tree. The best solution would be to try to migrate away from static device tree representations of partition info entirely. UBI volumes are best where possible. If not, then some other kind of on-flash data structures (along the lines of a GPT) with a corresponding MTD partition parser is an OK alternative. Unfortunately, there isn't any good standard format for this on MTD, so it's typically all custom -- and so people use the easiest approach: device tree. And it's even more difficult with NAND, which has reliability problems, especially with static data (e.g., read disturb). Anyway, the parser solution is helpful only if one can properly fix the "flag day" first. And it requires a little bit more work to be generally useful; I posted some work for this over a year ago, but bikeshedding brought it down. > The rest of the series looks fine to me so applying it into > omap-for-v4.11/dt. Brian
Hello Brian, Am Donnerstag, den 05.01.2017, 09:56 -0800 schrieb Brian Norris: > On Thu, Jan 05, 2017 at 09:18:45AM -0800, Tony Lindgren wrote: > > > > * Tony Lindgren <tony@atomide.com> [170105 07:37]: > > > > > > * Teresa Remmet <t.remmet@phytec.de> [170105 06:57]: > > > > > > > > To improve NAND safety we updated the partition layout. > > > > Added barebox backup partition and removed kernel and oftree > > > > partition. They are kept in ubi now. > > > What about the users with earlier partition tables? > > > > > > Please read about "flag day" changes, typically they are not > > > acceptable. > > Adding Brian and Adam to Cc. Can you guys come up with some > > solution on this? > I don't have much context for this thread, and no I don't plan to > solve > your problems for you. But I can provide tips! > > > > > I'm suggesting we leave the kernel nodes empty and let u-boot > > populate them, so maybe you guys can discuss this on the related > > lists. > That's an option. I've worked with platforms that did something like > this, and that's really one of the only ways you can handle putting > partition information in the device tree. You're really hamstringing > yourself when you put all the partition information in the device > tree. > And it's just dumb once that gets codified in the kernel source tree. > In our case the bootloader does pass the partition table to the kernel. So it gets overwritten anyway. This was just more for backup, if someone uses a different bootloader. But I'm fine with removing the nand partition table completely from the kernel device tree. Same with the SPI nor partition table. I will send patches for this. Regards, Teresa > The best solution would be to try to migrate away from static device > tree representations of partition info entirely. UBI volumes are best > where possible. If not, then some other kind of on-flash data > structures > (along the lines of a GPT) with a corresponding MTD partition parser > is > an OK alternative. Unfortunately, there isn't any good standard > format > for this on MTD, so it's typically all custom -- and so people use > the > easiest approach: device tree. And it's even more difficult with > NAND, > which has reliability problems, especially with static data (e.g., > read > disturb). > > Anyway, the parser solution is helpful only if one can properly fix > the > "flag day" first. And it requires a little bit more work to be > generally > useful; I posted some work for this over a year ago, but bikeshedding > brought it down. > > > > > The rest of the series looks fine to me so applying it into > > omap-for-v4.11/dt. > Brian > > _______________________________________________ > linux-arm-kernel mailing list > linux-arm-kernel@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
* Teresa Remmet <t.remmet@phytec.de> [170106 01:28]: > Hello Brian, > > Am Donnerstag, den 05.01.2017, 09:56 -0800 schrieb Brian Norris: > > On Thu, Jan 05, 2017 at 09:18:45AM -0800, Tony Lindgren wrote: > > > > > > * Tony Lindgren <tony@atomide.com> [170105 07:37]: > > > > > > > > * Teresa Remmet <t.remmet@phytec.de> [170105 06:57]: > > > > > > > > > > To improve NAND safety we updated the partition layout. > > > > > Added barebox backup partition and removed kernel and oftree > > > > > partition. They are kept in ubi now. > > > > What about the users with earlier partition tables? > > > > > > > > Please read about "flag day" changes, typically they are not > > > > acceptable. > > > Adding Brian and Adam to Cc. Can you guys come up with some > > > solution on this? > > I don't have much context for this thread, and no I don't plan to > > solve > > your problems for you. But I can provide tips! > > > > > > > > I'm suggesting we leave the kernel nodes empty and let u-boot > > > populate them, so maybe you guys can discuss this on the related > > > lists. > > That's an option. I've worked with platforms that did something like > > this, and that's really one of the only ways you can handle putting > > partition information in the device tree. You're really hamstringing > > yourself when you put all the partition information in the device > > tree. > > And it's just dumb once that gets codified in the kernel source tree. > > > > In our case the bootloader does pass the partition table to the kernel. > So it gets overwritten anyway. This was just more for backup, > if someone uses a different bootloader. But I'm fine with removing the > nand partition table completely from the kernel device tree. > Same with the SPI nor partition table. > > I will send patches for this. OK thanks! Also thank you Brian for your comments. Regards, Tony
On Fri, Jan 6, 2017 at 10:02 AM, Tony Lindgren <tony@atomide.com> wrote: > * Teresa Remmet <t.remmet@phytec.de> [170106 01:28]: >> Hello Brian, >> >> Am Donnerstag, den 05.01.2017, 09:56 -0800 schrieb Brian Norris: >> > On Thu, Jan 05, 2017 at 09:18:45AM -0800, Tony Lindgren wrote: >> > > >> > > * Tony Lindgren <tony@atomide.com> [170105 07:37]: >> > > > >> > > > * Teresa Remmet <t.remmet@phytec.de> [170105 06:57]: >> > > > > >> > > > > To improve NAND safety we updated the partition layout. >> > > > > Added barebox backup partition and removed kernel and oftree >> > > > > partition. They are kept in ubi now. >> > > > What about the users with earlier partition tables? >> > > > >> > > > Please read about "flag day" changes, typically they are not >> > > > acceptable. >> > > Adding Brian and Adam to Cc. Can you guys come up with some >> > > solution on this? >> > I don't have much context for this thread, and no I don't plan to >> > solve >> > your problems for you. But I can provide tips! >> > >> > > >> > > I'm suggesting we leave the kernel nodes empty and let u-boot >> > > populate them, so maybe you guys can discuss this on the related >> > > lists. >> > That's an option. I've worked with platforms that did something like >> > this, and that's really one of the only ways you can handle putting >> > partition information in the device tree. You're really hamstringing >> > yourself when you put all the partition information in the device >> > tree. >> > And it's just dumb once that gets codified in the kernel source tree. >> > >> >> In our case the bootloader does pass the partition table to the kernel. >> So it gets overwritten anyway. This was just more for backup, >> if someone uses a different bootloader. But I'm fine with removing the >> nand partition table completely from the kernel device tree. >> Same with the SPI nor partition table. >> >> I will send patches for this. > > OK thanks! Also thank you Brian for your comments. > Tony - I tested leaving the partition info as-is with the updates from the bootloader and it works. Would you prefer that I match Brian and remove the partition table completely, or should I just leave my board alone? I am good either way. > Regards, > > Tony adam
* Adam Ford <aford173@gmail.com> [170106 08:06]: > On Fri, Jan 6, 2017 at 10:02 AM, Tony Lindgren <tony@atomide.com> wrote: > > * Teresa Remmet <t.remmet@phytec.de> [170106 01:28]: > >> Hello Brian, > >> > >> Am Donnerstag, den 05.01.2017, 09:56 -0800 schrieb Brian Norris: > >> > On Thu, Jan 05, 2017 at 09:18:45AM -0800, Tony Lindgren wrote: > >> > > > >> > > * Tony Lindgren <tony@atomide.com> [170105 07:37]: > >> > > > > >> > > > * Teresa Remmet <t.remmet@phytec.de> [170105 06:57]: > >> > > > > > >> > > > > To improve NAND safety we updated the partition layout. > >> > > > > Added barebox backup partition and removed kernel and oftree > >> > > > > partition. They are kept in ubi now. > >> > > > What about the users with earlier partition tables? > >> > > > > >> > > > Please read about "flag day" changes, typically they are not > >> > > > acceptable. > >> > > Adding Brian and Adam to Cc. Can you guys come up with some > >> > > solution on this? > >> > I don't have much context for this thread, and no I don't plan to > >> > solve > >> > your problems for you. But I can provide tips! > >> > > >> > > > >> > > I'm suggesting we leave the kernel nodes empty and let u-boot > >> > > populate them, so maybe you guys can discuss this on the related > >> > > lists. > >> > That's an option. I've worked with platforms that did something like > >> > this, and that's really one of the only ways you can handle putting > >> > partition information in the device tree. You're really hamstringing > >> > yourself when you put all the partition information in the device > >> > tree. > >> > And it's just dumb once that gets codified in the kernel source tree. > >> > > >> > >> In our case the bootloader does pass the partition table to the kernel. > >> So it gets overwritten anyway. This was just more for backup, > >> if someone uses a different bootloader. But I'm fine with removing the > >> nand partition table completely from the kernel device tree. > >> Same with the SPI nor partition table. > >> > >> I will send patches for this. > > > > OK thanks! Also thank you Brian for your comments. > > > > Tony - I tested leaving the partition info as-is with the updates from > the bootloader and it works. Would you prefer that I match Brian and > remove the partition table completely, or should I just leave my board > alone? > > I am good either way. OK. How about let's remove the partitions from kernel for v4.11 as clean-up then for cases where the bootloader might change them? Regards, Tony
On Thu, Jan 05, 2017 at 09:56:20AM -0800, Brian Norris wrote: [snip] > The best solution would be to try to migrate away from static device > tree representations of partition info entirely. UBI volumes are best > where possible. If not, then some other kind of on-flash data structures > (along the lines of a GPT) with a corresponding MTD partition parser is > an OK alternative. Unfortunately, there isn't any good standard format > for this on MTD, so it's typically all custom -- and so people use the > easiest approach: device tree. And it's even more difficult with NAND, > which has reliability problems, especially with static data (e.g., read > disturb). Just as a side note, there is some work in this area: https://www.mail-archive.com/u-boot@lists.denx.de/msg232759.html > Anyway, the parser solution is helpful only if one can properly fix the > "flag day" first. And it requires a little bit more work to be generally > useful; I posted some work for this over a year ago, but bikeshedding > brought it down. Best regards, ladis
On Fri, Jan 06, 2017 at 08:14:22AM -0800, Tony Lindgren wrote: > * Adam Ford <aford173@gmail.com> [170106 08:06]: > > On Fri, Jan 6, 2017 at 10:02 AM, Tony Lindgren <tony@atomide.com> wrote: > > > * Teresa Remmet <t.remmet@phytec.de> [170106 01:28]: > > >> Am Donnerstag, den 05.01.2017, 09:56 -0800 schrieb Brian Norris: > > >> > On Thu, Jan 05, 2017 at 09:18:45AM -0800, Tony Lindgren wrote: > > >> > > I'm suggesting we leave the kernel nodes empty and let u-boot > > >> > > populate them, so maybe you guys can discuss this on the related > > >> > > lists. > > >> > That's an option. I've worked with platforms that did something like > > >> > this, and that's really one of the only ways you can handle putting > > >> > partition information in the device tree. You're really hamstringing > > >> > yourself when you put all the partition information in the device > > >> > tree. > > >> > And it's just dumb once that gets codified in the kernel source tree. > > >> > > > >> > > >> In our case the bootloader does pass the partition table to the kernel. > > >> So it gets overwritten anyway. This was just more for backup, > > >> if someone uses a different bootloader. But I'm fine with removing the > > >> nand partition table completely from the kernel device tree. > > >> Same with the SPI nor partition table. Ah, well if these are essentially just a backup, then why do they need changed at all? To be more precise about my "dumb" statement: it seems unwise to assume that all systems using a particular board will have the same partition layout. So *relying* on the in-kernel DTS(I) files to have a universal partition layout would likely cause problems. I don't have much problem with keeping a sample layout there as a backup, if you find it useful. But it seems like it will be hard to argue that you can ever change it in the future. Anyway, the ofpart mechanism is there, and it's fine to use it, but it's up to you to understand the implications and manage the backwards compatibility problems :) > > >> I will send patches for this. > > > > > > OK thanks! Also thank you Brian for your comments. > > > > > > > Tony - I tested leaving the partition info as-is with the updates from > > the bootloader and it works. Would you prefer that I match Brian and > > remove the partition table completely, or should I just leave my board > > alone? > > > > I am good either way. > > OK. How about let's remove the partitions from kernel for v4.11 as > clean-up then for cases where the bootloader might change them? I don't have much stake in this but...if there were any systems using the backup layout (i.e., the in-kernel partition layout), wouldn't that break them? Is there a problem with just leaving them alone? Brian
* Brian Norris <briannorris@chromium.org> [170106 10:21]: > On Fri, Jan 06, 2017 at 08:14:22AM -0800, Tony Lindgren wrote: > > * Adam Ford <aford173@gmail.com> [170106 08:06]: > > > On Fri, Jan 6, 2017 at 10:02 AM, Tony Lindgren <tony@atomide.com> wrote: > > > > * Teresa Remmet <t.remmet@phytec.de> [170106 01:28]: > > > >> Am Donnerstag, den 05.01.2017, 09:56 -0800 schrieb Brian Norris: > > > >> > On Thu, Jan 05, 2017 at 09:18:45AM -0800, Tony Lindgren wrote: > > > >> > > I'm suggesting we leave the kernel nodes empty and let u-boot > > > >> > > populate them, so maybe you guys can discuss this on the related > > > >> > > lists. > > > >> > That's an option. I've worked with platforms that did something like > > > >> > this, and that's really one of the only ways you can handle putting > > > >> > partition information in the device tree. You're really hamstringing > > > >> > yourself when you put all the partition information in the device > > > >> > tree. > > > >> > And it's just dumb once that gets codified in the kernel source tree. > > > >> > > > > >> > > > >> In our case the bootloader does pass the partition table to the kernel. > > > >> So it gets overwritten anyway. This was just more for backup, > > > >> if someone uses a different bootloader. But I'm fine with removing the > > > >> nand partition table completely from the kernel device tree. > > > >> Same with the SPI nor partition table. > > Ah, well if these are essentially just a backup, then why do they need > changed at all? To be more precise about my "dumb" statement: it seems > unwise to assume that all systems using a particular board will have the > same partition layout. So *relying* on the in-kernel DTS(I) files to > have a universal partition layout would likely cause problems. > > I don't have much problem with keeping a sample layout there as a > backup, if you find it useful. But it seems like it will be hard to > argue that you can ever change it in the future. > > Anyway, the ofpart mechanism is there, and it's fine to use it, but it's > up to you to understand the implications and manage the backwards > compatibility problems :) > > > > >> I will send patches for this. > > > > > > > > OK thanks! Also thank you Brian for your comments. > > > > > > > > > > Tony - I tested leaving the partition info as-is with the updates from > > > the bootloader and it works. Would you prefer that I match Brian and > > > remove the partition table completely, or should I just leave my board > > > alone? > > > > > > I am good either way. > > > > OK. How about let's remove the partitions from kernel for v4.11 as > > clean-up then for cases where the bootloader might change them? > > I don't have much stake in this but...if there were any systems using > the backup layout (i.e., the in-kernel partition layout), wouldn't that > break them? Is there a problem with just leaving them alone? Yeah I guess it depends on the device and it's bootloader versions. Regards, Tony
diff --git a/arch/arm/boot/dts/am335x-phycore-som.dtsi b/arch/arm/boot/dts/am335x-phycore-som.dtsi index 75e24ad..ef3fe5a 100644 --- a/arch/arm/boot/dts/am335x-phycore-som.dtsi +++ b/arch/arm/boot/dts/am335x-phycore-som.dtsi @@ -222,21 +222,19 @@ label = "barebox"; reg = <0x80000 0x80000>; }; + partition@5 { - label = "bareboxenv"; - reg = <0x100000 0x40000>; + label = "barebox_backup"; + reg = <0x100000 0x80000>; }; + partition@6 { - label = "oftree"; - reg = <0x140000 0x40000>; + label = "bareboxenv"; + reg = <0x180000 0x40000>; }; partition@7 { - label = "kernel"; - reg = <0x180000 0x800000>; - }; - partition@8 { label = "root"; - reg = <0x980000 0x0>; + reg = <0x1C0000 0x0>; }; }; };
To improve NAND safety we updated the partition layout. Added barebox backup partition and removed kernel and oftree partition. They are kept in ubi now. Signed-off-by: Teresa Remmet <t.remmet@phytec.de> --- arch/arm/boot/dts/am335x-phycore-som.dtsi | 16 +++++++--------- 1 file changed, 7 insertions(+), 9 deletions(-)