Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Thu, Oct 15, 2020 at 08:14:39AM +0200, Krzysztof Kozlowski wrote: > On Thu, Oct 15, 2020 at 02:51:05AM +0300, Serge Semin wrote: > > > > > > > So to speak thanks for suggesting it. I'll try it to validate the > > > > proposed > > > > changes. > > > > > > > > Two questions: > > > > 1) Any advise of a good inliner/command to compile all dtbs at once? Of > > > > course I > > > > can get all the updated dtsi'es, then find out all the dts'es which > > > > include > > > > them, then directly use dtc to compile the found dts'es... On the other > > > > hand I > > > > can just compile all dts'es, then compare old and new ones. The diff of > > > > the > > > > non-modified dtb'es will be just empty... > > > > > > > > make dtbs > > > > It's not that easy.) "make dtbs" will build dtbs only for enabled boards, > > which > > first need to be enabled in the kernel config. So I'll need to have a config > > with all the affected dts. The later is the same as if I just found all the > > affected dts and built them one-by-one by directly calling dtc. > > True. Sometimes allyesconfig for given arch might be helpful but not > always (e.g. for ARM it does not select all of ARMv4 and ARMv5 boards). > Most likely your approach is actually faster/more reliable. > > > > > > touch your dts or git stash pop > > > make dtbs > > > compare > > > diff for all unchanged will be simply empty, so easy to spot > > > > > > > 2) What crosc64 is? > > > > > > Ah, just an alias for cross compiling + ccache + kbuild out. I just > > > copied you my helpers, so you need to tweak them. > > > > > > > > > > > > > > > > > 2. Split it per arm architectures (and proper subject prefix - not > > > > > "arch") and subarchitectures so maintainers can pick it up. > > > > > > > > Why? The changes are simple and can be formatted as a single patch. > > > > I've seen > > > > tons of patches submitted like that, accepted and then merged. What you > > > > suggest > > > > is just much more work, which I don't see quite required. > > > > > > > > DTS changes go separate between arm64 and arm. There is nothing > > > unusual here - all changes are submitted like this. > > > Second topic is to split by subarchitectures which is necessary if you > > > want it to be picked up by maintainers. It also makes it easier to > > > review. > > > > The current patches are easy enough for review. The last three patches of > > the > > series is a collection of the one-type changes concerning the same type of > > nodes. So reviewing them won't cause any difficulty. But I assume that's not > > the main point in this discussion. > > > > > Sure, without split ber subarchitectures this could be picked > > > up by SoC folks but you did not even CC them. So if you do not want to > > > split it per subarchitectures for maintainers and you do not CC SoC, > > > then how do you believe this should be picked up? Out of the regular > > > patch submission way? That's not how the changes are handled. > > > > AFAIU there are another ways of merging comprehensive patches. If they get > > to collect > > all the Acked-by tags, they could be merged in, for instance, through Greg' > > or Rob' > > (for dts) repos, if of course they get to agree with doing that. Am I wrong? > > > > My hope was to ask Rob or Greg to get the patches merged in when they get > > to collect all the ackes, since I thought it was an option in such cases. > > So if > > they refuse to do so I'll have no choice but to split the series up into a > > smaller patches as you say. > > This is neither Rob's nor Greg's patch to pick up, but ARM SoC (which was > not CCed here). And most likely they won't pick it up because judging by > contents it is obvious it should go via ARM SoC. > > Sure, if there are dependencies between some patches they can go with > acks through unrelated trees, but this not the usual way. This is an > exception in the process to solve particular dependency problem. It has > drawbacks - increases the chances of annoying conflicts. > > The case here does not fall into this criteria - there is no dependency > of this patch on the others Therefore there is no reason to use the > unusual/exceptional way of handling patches. There is no reason why > this shouldn't go via either specific ARM subarchitecture maintainers or > via ARM SoC. Ok. I see your point. To sum it up I've studied the git log arch/ commit messages and it turns out even Rob has to split the cleanup changes like this ones. So thanks for your patience with stating your point. I'll split the last three patches up to be merged in via the corresponding archs/subarch'es repos. -Sergey > > > > > > 3. The subject title could be more accurate - there is no fix here > > > > > because there was no errors in the first place. Requirement of DWC > > > > > node names comes recently, so it is more alignment with dtschema. > > > > > Otherwise automatic-pickup-stable-bot might want to pick up... and it > > > > > should not go to s
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Thu, Oct 15, 2020 at 01:15:37PM +0300, Felipe Balbi wrote: > Serge Semin writes: > > > On Wed, Oct 14, 2020 at 05:09:37PM +0300, Felipe Balbi wrote: > >> > >> Hi Serge, > >> > >> Serge Semin writes: > >> > In accordance with the DWC USB3 bindings the corresponding node name is > >> > suppose to comply with Generic USB HCD DT schema, which requires the USB > >> > > > >> DWC3 is not a simple HDC, though. > > > > Yeah, strictly speaking it is equipped with a lot of vendor-specific stuff, > > which are tuned by the DWC USB3 driver in the kernel. But after that the > > controller is registered as xhci-hcd device so it's serviced by the xHCI > > driver, > > in Dual-role or host-only builds, that's correct. We can also have > peripheral-only builds (both SW or HW versions) which means xhci isn't > even in the picture. It doesn't really matter though, or at least it does for what the new name might be, but the old one currently used is still pretty bad. The DT spec says that the node name is the class of the device. "usb" as the HCD binding mandates is one, but the current nodes currently have completely different names from one DT to another - which is already an issue - and most of them have dwc3 or some variant of it, which doesn't really qualify for a class name. Maxime signature.asc Description: PGP signature
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Thu, Oct 15, 2020 at 01:15:37PM +0300, Felipe Balbi wrote: > Serge Semin writes: > > > On Wed, Oct 14, 2020 at 05:09:37PM +0300, Felipe Balbi wrote: > >> > >> Hi Serge, > >> > >> Serge Semin writes: > >> > In accordance with the DWC USB3 bindings the corresponding node name is > >> > suppose to comply with Generic USB HCD DT schema, which requires the USB > >> > > > >> DWC3 is not a simple HDC, though. > > > > Yeah, strictly speaking it is equipped with a lot of vendor-specific stuff, > > which are tuned by the DWC USB3 driver in the kernel. But after that the > > controller is registered as xhci-hcd device so it's serviced by the xHCI > > driver, > > in Dual-role or host-only builds, that's correct. We can also have > peripheral-only builds (both SW or HW versions) which means xhci isn't > even in the picture. Hm, good point. In that case perhaps we'll need to apply the xHCI DT schema conditionally. Like this: - allOf: - - $ref: usb-xhci.yaml# + allOf: + - if: + properties: + dr_mode: + const: peripheral + then: + $ref: usb-hcd.yaml# + else: + $ref: usb-xhci.yaml# Note, we need to have the peripheral device being compatible with the usb-hcd.yaml schema to support the maximum-speed, dr_mode properties and to comply with the USB node naming constraint. -Sergey > > -- > balbi
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
Serge Semin writes: > On Wed, Oct 14, 2020 at 05:09:37PM +0300, Felipe Balbi wrote: >> >> Hi Serge, >> >> Serge Semin writes: >> > In accordance with the DWC USB3 bindings the corresponding node name is >> > suppose to comply with Generic USB HCD DT schema, which requires the USB >> > >> DWC3 is not a simple HDC, though. > > Yeah, strictly speaking it is equipped with a lot of vendor-specific stuff, > which are tuned by the DWC USB3 driver in the kernel. But after that the > controller is registered as xhci-hcd device so it's serviced by the xHCI > driver, in Dual-role or host-only builds, that's correct. We can also have peripheral-only builds (both SW or HW versions) which means xhci isn't even in the picture. -- balbi signature.asc Description: PGP signature
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Thu, Oct 15, 2020 at 02:51:05AM +0300, Serge Semin wrote: > > > > > So to speak thanks for suggesting it. I'll try it to validate the proposed > > > changes. > > > > > > Two questions: > > > 1) Any advise of a good inliner/command to compile all dtbs at once? Of > > > course I > > > can get all the updated dtsi'es, then find out all the dts'es which > > > include > > > them, then directly use dtc to compile the found dts'es... On the other > > > hand I > > > can just compile all dts'es, then compare old and new ones. The diff of > > > the > > > non-modified dtb'es will be just empty... > > > > > make dtbs > > It's not that easy.) "make dtbs" will build dtbs only for enabled boards, > which > first need to be enabled in the kernel config. So I'll need to have a config > with all the affected dts. The later is the same as if I just found all the > affected dts and built them one-by-one by directly calling dtc. True. Sometimes allyesconfig for given arch might be helpful but not always (e.g. for ARM it does not select all of ARMv4 and ARMv5 boards). Most likely your approach is actually faster/more reliable. > > > touch your dts or git stash pop > > make dtbs > > compare > > diff for all unchanged will be simply empty, so easy to spot > > > > > 2) What crosc64 is? > > > > Ah, just an alias for cross compiling + ccache + kbuild out. I just > > copied you my helpers, so you need to tweak them. > > > > > > > > > > > > > 2. Split it per arm architectures (and proper subject prefix - not > > > > "arch") and subarchitectures so maintainers can pick it up. > > > > > > Why? The changes are simple and can be formatted as a single patch. I've > > > seen > > > tons of patches submitted like that, accepted and then merged. What you > > > suggest > > > is just much more work, which I don't see quite required. > > > > > DTS changes go separate between arm64 and arm. There is nothing > > unusual here - all changes are submitted like this. > > Second topic is to split by subarchitectures which is necessary if you > > want it to be picked up by maintainers. It also makes it easier to > > review. > > The current patches are easy enough for review. The last three patches of the > series is a collection of the one-type changes concerning the same type of > nodes. So reviewing them won't cause any difficulty. But I assume that's not > the main point in this discussion. > > > Sure, without split ber subarchitectures this could be picked > > up by SoC folks but you did not even CC them. So if you do not want to > > split it per subarchitectures for maintainers and you do not CC SoC, > > then how do you believe this should be picked up? Out of the regular > > patch submission way? That's not how the changes are handled. > > AFAIU there are another ways of merging comprehensive patches. If they get to > collect > all the Acked-by tags, they could be merged in, for instance, through Greg' > or Rob' > (for dts) repos, if of course they get to agree with doing that. Am I wrong? > > My hope was to ask Rob or Greg to get the patches merged in when they get > to collect all the ackes, since I thought it was an option in such cases. So > if > they refuse to do so I'll have no choice but to split the series up into a > smaller patches as you say. This is neither Rob's nor Greg's patch to pick up, but ARM SoC (which was not CCed here). And most likely they won't pick it up because judging by contents it is obvious it should go via ARM SoC. Sure, if there are dependencies between some patches they can go with acks through unrelated trees, but this not the usual way. This is an exception in the process to solve particular dependency problem. It has drawbacks - increases the chances of annoying conflicts. The case here does not fall into this criteria - there is no dependency of this patch on the others Therefore there is no reason to use the unusual/exceptional way of handling patches. There is no reason why this shouldn't go via either specific ARM subarchitecture maintainers or via ARM SoC. > > > > 3. The subject title could be more accurate - there is no fix here > > > > because there was no errors in the first place. Requirement of DWC > > > > node names comes recently, so it is more alignment with dtschema. > > > > Otherwise automatic-pickup-stable-bot might want to pick up... and it > > > > should not go to stable. > > > > > > Actually it is a fix, because the USB DT nodes should have been named > > > with "usb" > > > prefix in the first place. Legacy DWC USB3 bindings didn't define the > > > nodes > > > naming, but implied to be "usb"-prefixed by the USB HCD schema. The > > > Qualcomm > > > DWC3 schema should have defined the sub-nodes as "dwc3@"-prefixed, which > > > was > > > wrong in the first place. > > > > > Not following the naming convention of DT spec which was loosely > > enforced is not an error which should be "fixed". Simply wrong title. > > This is an alignment with dtschema or corre
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Wed, Oct 14, 2020 at 10:04:32PM +0200, Krzysztof Kozlowski wrote: > On Wed, 14 Oct 2020 at 19:16, Serge Semin > wrote: > > > > On Wed, Oct 14, 2020 at 12:33:25PM +0200, Krzysztof Kozlowski wrote: > > > On Wed, 14 Oct 2020 at 12:23, Serge Semin > > > wrote: > > > > > > > > In accordance with the DWC USB3 bindings the corresponding node name is > > > > suppose to comply with Generic USB HCD DT schema, which requires the USB > > > > nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a > > > > lot > > > > of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have > > > > name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will > > > > cause > > > > the dtbs_check procedure failure. Let's fix the nodes naming to be > > > > compatible with the DWC USB3 DT schema to make dtbs_check happy. > > > > > > > > Note we don't change the DWC USB3-compatible nodes names of > > > > arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the > > > > in-source comment says that the nodes name need to be preserved as > > > > "^dwusb@.*" for some backward compatibility. > > > > > > > > Signed-off-by: Serge Semin > > > > > > > > --- > > > > > > > > Please, test the patch out to make sure it doesn't brake the dependent > > > > DTS > > > > files. I did only a manual grepping of the possible nodes dependencies. > > > > > > > > 1. It is you who should compare the decompiled DTS, not us. For example: > > > $ for i in dts-old/*/*dtb dts-old/*/*/*dtb; do echo $i; crosc64 > > > scripts/dtc/dtx_diff ${i} dts-new/${i#dts-old/} ; done > > > > > > $ for i in dts-old/*/*dtb dts-old/*/*/*dtb; do echo $i; crosc64 > > > fdtdump ${i} > ${i}.fdt ; crosc64 fdtdump dts-new/${i#dts-old/} > > > > dts-new/${i#dts-old/}.fdt ; diff -ubB ${i}.fdt > > > dts-new/${i#dts-old/}.fdt ; done > > > > So basically you suggest first to compile the old and new dts files, then to > > de-compile them, then diff old and new fdt's, and visually compare the > > results. > > Personally it isn't that much better than what I did, since each old and new > > dtbs will for sure differ due to the node names change suggested in this > > patch. > > So it will lead to the visual debugging too, which isn't that effective. But > > your approach is still more demonstrative to make sure that I didn't loose > > any > > nodes redefinition, since in the occasion the old and new de-compiled nodes > > will > > differ not only by the node names but with an additional old named node. > > My suggestion is to compare the entire, effective DTS after all > inclusions. Maybe you did it already, I don't know. Only by grepping the dts'es, which include the dtsi'es modified in this patch. So your suggestion of compiling and de-compiling has been indeed relevant. > The point is that > when you change node names in DTSI but you miss one in DTS, you end up > with two nodes. Yep, that's exactly what I meant when I said that your approach was more demonstrative, etc. > This is much easier to spot with dtxdiff or with > fdtdump (which behaves better for node moves). > > Indeed it is still a visual comparison - if you have any ideas how to > automate it (e.g. ignore phandle changes), please share. It would > solve my testings as well. Alas I don't. That's why to save my time of coming up with an automated solution I did a very thorough modification making sure that each affected node isn't updated in the corresponding dts'es and asked to test the patches out. Anyway the approach suggested by you will indeed give us a better understanding of my patches correctness. So I'll use it before sending v3. Thanks. > But asking others to test because you do > not want to check it is not the best way to handle such changes. > > > > > So to speak thanks for suggesting it. I'll try it to validate the proposed > > changes. > > > > Two questions: > > 1) Any advise of a good inliner/command to compile all dtbs at once? Of > > course I > > can get all the updated dtsi'es, then find out all the dts'es which include > > them, then directly use dtc to compile the found dts'es... On the other > > hand I > > can just compile all dts'es, then compare old and new ones. The diff of the > > non-modified dtb'es will be just empty... > > make dtbs It's not that easy.) "make dtbs" will build dtbs only for enabled boards, which first need to be enabled in the kernel config. So I'll need to have a config with all the affected dts. The later is the same as if I just found all the affected dts and built them one-by-one by directly calling dtc. > touch your dts or git stash pop > make dtbs > compare > diff for all unchanged will be simply empty, so easy to spot > > > 2) What crosc64 is? > > Ah, just an alias for cross compiling + ccache + kbuild out. I just > copied you my helpers, so you need to tweak them. > > > > > > > > > 2. Split it per arm architectures (and proper subject prefix - not > > > "arch") and subarchitectures so maintainers can pick it
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Wed, Oct 14, 2020 at 01:35:16PM -0500, Rob Herring wrote: > On Wed, Oct 14, 2020 at 9:37 AM Serge Semin > wrote: > > > > On Wed, Oct 14, 2020 at 05:09:37PM +0300, Felipe Balbi wrote: > > > > > > Hi Serge, > > > > > > Serge Semin writes: > > > > In accordance with the DWC USB3 bindings the corresponding node name is > > > > suppose to comply with Generic USB HCD DT schema, which requires the USB > > > > > > > > DWC3 is not a simple HDC, though. > > > > Yeah, strictly speaking it is equipped with a lot of vendor-specific stuff, > > which are tuned by the DWC USB3 driver in the kernel. But after that the > > controller is registered as xhci-hcd device so it's serviced by the xHCI > > driver, > > which then registers the HCD device so the corresponding DT node is supposed > > to be compatible with the next bindings: usb/usb-hcd.yaml, usb/usb-xhci.yaml > > and usb/snps,dwc3,yaml. I've created the later one so to validate the > > denoted > > compatibility. > > > > > > > > > nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a > > > > lot > > > > of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have > > > > name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will > > > > cause > > > > the dtbs_check procedure failure. Let's fix the nodes naming to be > > > > compatible with the DWC USB3 DT schema to make dtbs_check happy. > > > > > > > > Note we don't change the DWC USB3-compatible nodes names of > > > > arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the > > > > in-source comment says that the nodes name need to be preserved as > > > > "^dwusb@.*" for some backward compatibility. > > > > > > > > interesting, compatibility with what? Some debugfs files, perhaps? :-) > > > > Don't really know.) In my experience the worst type of such compatibility is > > connected with some bootloader magic, which may add/remove/modify properties > > to nodes with pre-defined names. > > I seriously doubt anyone is using the APM machines with DT (even ACPI > is somewhat doubtful). I say change them. Or remove the dts files and > see what happens. Either way it can always be reverted. Ok. I'll change them in v3. -Sergey > > Rob
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Wed, 14 Oct 2020 at 19:16, Serge Semin wrote: > > On Wed, Oct 14, 2020 at 12:33:25PM +0200, Krzysztof Kozlowski wrote: > > On Wed, 14 Oct 2020 at 12:23, Serge Semin > > wrote: > > > > > > In accordance with the DWC USB3 bindings the corresponding node name is > > > suppose to comply with Generic USB HCD DT schema, which requires the USB > > > nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a lot > > > of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have > > > name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will cause > > > the dtbs_check procedure failure. Let's fix the nodes naming to be > > > compatible with the DWC USB3 DT schema to make dtbs_check happy. > > > > > > Note we don't change the DWC USB3-compatible nodes names of > > > arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the > > > in-source comment says that the nodes name need to be preserved as > > > "^dwusb@.*" for some backward compatibility. > > > > > > Signed-off-by: Serge Semin > > > > > > --- > > > > > > Please, test the patch out to make sure it doesn't brake the dependent DTS > > > files. I did only a manual grepping of the possible nodes dependencies. > > > > > 1. It is you who should compare the decompiled DTS, not us. For example: > > $ for i in dts-old/*/*dtb dts-old/*/*/*dtb; do echo $i; crosc64 > > scripts/dtc/dtx_diff ${i} dts-new/${i#dts-old/} ; done > > > > $ for i in dts-old/*/*dtb dts-old/*/*/*dtb; do echo $i; crosc64 > > fdtdump ${i} > ${i}.fdt ; crosc64 fdtdump dts-new/${i#dts-old/} > > > dts-new/${i#dts-old/}.fdt ; diff -ubB ${i}.fdt > > dts-new/${i#dts-old/}.fdt ; done > > So basically you suggest first to compile the old and new dts files, then to > de-compile them, then diff old and new fdt's, and visually compare the > results. > Personally it isn't that much better than what I did, since each old and new > dtbs will for sure differ due to the node names change suggested in this > patch. > So it will lead to the visual debugging too, which isn't that effective. But > your approach is still more demonstrative to make sure that I didn't loose any > nodes redefinition, since in the occasion the old and new de-compiled nodes > will > differ not only by the node names but with an additional old named node. My suggestion is to compare the entire, effective DTS after all inclusions. Maybe you did it already, I don't know. The point is that when you change node names in DTSI but you miss one in DTS, you end up with two nodes. This is much easier to spot with dtxdiff or with fdtdump (which behaves better for node moves). Indeed it is still a visual comparison - if you have any ideas how to automate it (e.g. ignore phandle changes), please share. It would solve my testings as well. But asking others to test because you do not want to check it is not the best way to handle such changes. > > So to speak thanks for suggesting it. I'll try it to validate the proposed > changes. > > Two questions: > 1) Any advise of a good inliner/command to compile all dtbs at once? Of > course I > can get all the updated dtsi'es, then find out all the dts'es which include > them, then directly use dtc to compile the found dts'es... On the other hand I > can just compile all dts'es, then compare old and new ones. The diff of the > non-modified dtb'es will be just empty... make dtbs touch your dts or git stash pop make dtbs compare diff for all unchanged will be simply empty, so easy to spot > 2) What crosc64 is? Ah, just an alias for cross compiling + ccache + kbuild out. I just copied you my helpers, so you need to tweak them. > > > > > 2. Split it per arm architectures (and proper subject prefix - not > > "arch") and subarchitectures so maintainers can pick it up. > > Why? The changes are simple and can be formatted as a single patch. I've seen > tons of patches submitted like that, accepted and then merged. What you > suggest > is just much more work, which I don't see quite required. DTS changes go separate between arm64 and arm. There is nothing unusual here - all changes are submitted like this. Second topic is to split by subarchitectures which is necessary if you want it to be picked up by maintainers. It also makes it easier to review. Sure, without split ber subarchitectures this could be picked up by SoC folks but you did not even CC them. So if you do not want to split it per subarchitectures for maintainers and you do not CC SoC, then how do you believe this should be picked up? Out of the regular patch submission way? That's not how the changes are handled. > > > > > 3. The subject title could be more accurate - there is no fix here > > because there was no errors in the first place. Requirement of DWC > > node names comes recently, so it is more alignment with dtschema. > > Otherwise automatic-pickup-stable-bot might want to pick up... and it > > should not go to stable. > > Actually it is a fix, because the USB DT nodes should have been nam
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Wed, Oct 14, 2020 at 9:37 AM Serge Semin wrote: > > On Wed, Oct 14, 2020 at 05:09:37PM +0300, Felipe Balbi wrote: > > > > Hi Serge, > > > > Serge Semin writes: > > > In accordance with the DWC USB3 bindings the corresponding node name is > > > suppose to comply with Generic USB HCD DT schema, which requires the USB > > > > > DWC3 is not a simple HDC, though. > > Yeah, strictly speaking it is equipped with a lot of vendor-specific stuff, > which are tuned by the DWC USB3 driver in the kernel. But after that the > controller is registered as xhci-hcd device so it's serviced by the xHCI > driver, > which then registers the HCD device so the corresponding DT node is supposed > to be compatible with the next bindings: usb/usb-hcd.yaml, usb/usb-xhci.yaml > and usb/snps,dwc3,yaml. I've created the later one so to validate the denoted > compatibility. > > > > > > nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a lot > > > of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have > > > name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will cause > > > the dtbs_check procedure failure. Let's fix the nodes naming to be > > > compatible with the DWC USB3 DT schema to make dtbs_check happy. > > > > > > Note we don't change the DWC USB3-compatible nodes names of > > > arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the > > > in-source comment says that the nodes name need to be preserved as > > > "^dwusb@.*" for some backward compatibility. > > > > > interesting, compatibility with what? Some debugfs files, perhaps? :-) > > Don't really know.) In my experience the worst type of such compatibility is > connected with some bootloader magic, which may add/remove/modify properties > to nodes with pre-defined names. I seriously doubt anyone is using the APM machines with DT (even ACPI is somewhat doubtful). I say change them. Or remove the dts files and see what happens. Either way it can always be reverted. Rob
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Wed, Oct 14, 2020 at 12:33:25PM +0200, Krzysztof Kozlowski wrote: > On Wed, 14 Oct 2020 at 12:23, Serge Semin > wrote: > > > > In accordance with the DWC USB3 bindings the corresponding node name is > > suppose to comply with Generic USB HCD DT schema, which requires the USB > > nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a lot > > of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have > > name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will cause > > the dtbs_check procedure failure. Let's fix the nodes naming to be > > compatible with the DWC USB3 DT schema to make dtbs_check happy. > > > > Note we don't change the DWC USB3-compatible nodes names of > > arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the > > in-source comment says that the nodes name need to be preserved as > > "^dwusb@.*" for some backward compatibility. > > > > Signed-off-by: Serge Semin > > > > --- > > > > Please, test the patch out to make sure it doesn't brake the dependent DTS > > files. I did only a manual grepping of the possible nodes dependencies. > > 1. It is you who should compare the decompiled DTS, not us. For example: > $ for i in dts-old/*/*dtb dts-old/*/*/*dtb; do echo $i; crosc64 > scripts/dtc/dtx_diff ${i} dts-new/${i#dts-old/} ; done > > $ for i in dts-old/*/*dtb dts-old/*/*/*dtb; do echo $i; crosc64 > fdtdump ${i} > ${i}.fdt ; crosc64 fdtdump dts-new/${i#dts-old/} > > dts-new/${i#dts-old/}.fdt ; diff -ubB ${i}.fdt > dts-new/${i#dts-old/}.fdt ; done So basically you suggest first to compile the old and new dts files, then to de-compile them, then diff old and new fdt's, and visually compare the results. Personally it isn't that much better than what I did, since each old and new dtbs will for sure differ due to the node names change suggested in this patch. So it will lead to the visual debugging too, which isn't that effective. But your approach is still more demonstrative to make sure that I didn't loose any nodes redefinition, since in the occasion the old and new de-compiled nodes will differ not only by the node names but with an additional old named node. So to speak thanks for suggesting it. I'll try it to validate the proposed changes. Two questions: 1) Any advise of a good inliner/command to compile all dtbs at once? Of course I can get all the updated dtsi'es, then find out all the dts'es which include them, then directly use dtc to compile the found dts'es... On the other hand I can just compile all dts'es, then compare old and new ones. The diff of the non-modified dtb'es will be just empty... 2) What crosc64 is? > > 2. Split it per arm architectures (and proper subject prefix - not > "arch") and subarchitectures so maintainers can pick it up. Why? The changes are simple and can be formatted as a single patch. I've seen tons of patches submitted like that, accepted and then merged. What you suggest is just much more work, which I don't see quite required. > > 3. The subject title could be more accurate - there is no fix here > because there was no errors in the first place. Requirement of DWC > node names comes recently, so it is more alignment with dtschema. > Otherwise automatic-pickup-stable-bot might want to pick up... and it > should not go to stable. Actually it is a fix, because the USB DT nodes should have been named with "usb" prefix in the first place. Legacy DWC USB3 bindings didn't define the nodes naming, but implied to be "usb"-prefixed by the USB HCD schema. The Qualcomm DWC3 schema should have defined the sub-nodes as "dwc3@"-prefixed, which was wrong in the first place. Regarding automatic-pickup-stable-bot if it exists I don't think it scans all the emails, but most likely the sta...@vger.kernel.org list only or the emails having the "Fixes:" tag. If I am wrong please give me a link to the bot sources or refer to a doc where I can read about the way it works, to take it into account in future commits. Just to note I submitted patches which did some fixes, had the word "fix" in the subject but weren't selected to be backported to the stable kernel. Anyway I don't really care that much about the subject text using the word "fix" or some else. So if you suggest some better alternative, I'd be glad to consider it. -Sergey > > Best regards, > Krzysztof > > > arch/arm/boot/dts/armada-375.dtsi | 2 +- > > arch/arm/boot/dts/exynos5250.dtsi | 2 +- > > arch/arm/boot/dts/exynos54xx.dtsi | 4 ++-- > > arch/arm/boot/dts/keystone-k2e.dtsi| 4 ++-- > > arch/arm/boot/dts/keystone.dtsi| 2 +- > > arch/arm/boot/dts/ls1021a.dtsi | 2 +- > > arch/arm/boot/dts/omap5-l4.dtsi| 2 +- > > arch/arm/boot/dts/stih407-family.dtsi | 2 +- > > arch/arm64/boot/dts/allwinner/sun50i-h6.dtsi | 2 +- > > arch/arm64/boot/dts/exynos/exynos5433.dtsi | 4 ++-- > > arch/arm64/boot/dts/exynos/exynos7.dtsi| 2
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Wed, Oct 14, 2020 at 05:09:37PM +0300, Felipe Balbi wrote: > > Hi Serge, > > Serge Semin writes: > > In accordance with the DWC USB3 bindings the corresponding node name is > > suppose to comply with Generic USB HCD DT schema, which requires the USB > > DWC3 is not a simple HDC, though. Yeah, strictly speaking it is equipped with a lot of vendor-specific stuff, which are tuned by the DWC USB3 driver in the kernel. But after that the controller is registered as xhci-hcd device so it's serviced by the xHCI driver, which then registers the HCD device so the corresponding DT node is supposed to be compatible with the next bindings: usb/usb-hcd.yaml, usb/usb-xhci.yaml and usb/snps,dwc3,yaml. I've created the later one so to validate the denoted compatibility. > > > nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a lot > > of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have > > name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will cause > > the dtbs_check procedure failure. Let's fix the nodes naming to be > > compatible with the DWC USB3 DT schema to make dtbs_check happy. > > > > Note we don't change the DWC USB3-compatible nodes names of > > arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the > > in-source comment says that the nodes name need to be preserved as > > "^dwusb@.*" for some backward compatibility. > > interesting, compatibility with what? Some debugfs files, perhaps? :-) Don't really know.) In my experience the worst type of such compatibility is connected with some bootloader magic, which may add/remove/modify properties to nodes with pre-defined names. -Sergey > > In any case, I don't have any problems with this, so I'll let other > folks comment. > > -- > balbi
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
Hi Serge, Serge Semin writes: > In accordance with the DWC USB3 bindings the corresponding node name is > suppose to comply with Generic USB HCD DT schema, which requires the USB DWC3 is not a simple HDC, though. > nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a lot > of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have > name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will cause > the dtbs_check procedure failure. Let's fix the nodes naming to be > compatible with the DWC USB3 DT schema to make dtbs_check happy. > > Note we don't change the DWC USB3-compatible nodes names of > arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the > in-source comment says that the nodes name need to be preserved as > "^dwusb@.*" for some backward compatibility. interesting, compatibility with what? Some debugfs files, perhaps? :-) In any case, I don't have any problems with this, so I'll let other folks comment. -- balbi signature.asc Description: PGP signature
Re: [PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
On Wed, 14 Oct 2020 at 12:23, Serge Semin wrote: > > In accordance with the DWC USB3 bindings the corresponding node name is > suppose to comply with Generic USB HCD DT schema, which requires the USB > nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a lot > of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have > name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will cause > the dtbs_check procedure failure. Let's fix the nodes naming to be > compatible with the DWC USB3 DT schema to make dtbs_check happy. > > Note we don't change the DWC USB3-compatible nodes names of > arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the > in-source comment says that the nodes name need to be preserved as > "^dwusb@.*" for some backward compatibility. > > Signed-off-by: Serge Semin > > --- > > Please, test the patch out to make sure it doesn't brake the dependent DTS > files. I did only a manual grepping of the possible nodes dependencies. 1. It is you who should compare the decompiled DTS, not us. For example: $ for i in dts-old/*/*dtb dts-old/*/*/*dtb; do echo $i; crosc64 scripts/dtc/dtx_diff ${i} dts-new/${i#dts-old/} ; done $ for i in dts-old/*/*dtb dts-old/*/*/*dtb; do echo $i; crosc64 fdtdump ${i} > ${i}.fdt ; crosc64 fdtdump dts-new/${i#dts-old/} > dts-new/${i#dts-old/}.fdt ; diff -ubB ${i}.fdt dts-new/${i#dts-old/}.fdt ; done 2. Split it per arm architectures (and proper subject prefix - not "arch") and subarchitectures so maintainers can pick it up. 3. The subject title could be more accurate - there is no fix here because there was no errors in the first place. Requirement of DWC node names comes recently, so it is more alignment with dtschema. Otherwise automatic-pickup-stable-bot might want to pick up... and it should not go to stable. Best regards, Krzysztof > arch/arm/boot/dts/armada-375.dtsi | 2 +- > arch/arm/boot/dts/exynos5250.dtsi | 2 +- > arch/arm/boot/dts/exynos54xx.dtsi | 4 ++-- > arch/arm/boot/dts/keystone-k2e.dtsi| 4 ++-- > arch/arm/boot/dts/keystone.dtsi| 2 +- > arch/arm/boot/dts/ls1021a.dtsi | 2 +- > arch/arm/boot/dts/omap5-l4.dtsi| 2 +- > arch/arm/boot/dts/stih407-family.dtsi | 2 +- > arch/arm64/boot/dts/allwinner/sun50i-h6.dtsi | 2 +- > arch/arm64/boot/dts/exynos/exynos5433.dtsi | 4 ++-- > arch/arm64/boot/dts/exynos/exynos7.dtsi| 2 +- > arch/arm64/boot/dts/freescale/fsl-ls1012a.dtsi | 4 ++-- > arch/arm64/boot/dts/freescale/fsl-ls1043a.dtsi | 6 +++--- > arch/arm64/boot/dts/freescale/fsl-ls1088a.dtsi | 4 ++-- > arch/arm64/boot/dts/freescale/fsl-ls208xa.dtsi | 4 ++-- > arch/arm64/boot/dts/hisilicon/hi3660.dtsi | 2 +- > arch/arm64/boot/dts/qcom/apq8096-db820c.dtsi | 4 ++-- > arch/arm64/boot/dts/qcom/ipq8074.dtsi | 4 ++-- > arch/arm64/boot/dts/qcom/msm8996.dtsi | 4 ++-- > arch/arm64/boot/dts/qcom/msm8998.dtsi | 2 +- > arch/arm64/boot/dts/qcom/qcs404-evb.dtsi | 2 +- > arch/arm64/boot/dts/qcom/qcs404.dtsi | 4 ++-- > arch/arm64/boot/dts/qcom/sc7180.dtsi | 2 +- > arch/arm64/boot/dts/qcom/sdm845.dtsi | 4 ++-- > arch/arm64/boot/dts/qcom/sm8150.dtsi | 2 +- > 25 files changed, 38 insertions(+), 38 deletions(-) >
[PATCH 20/20] arch: dts: Fix DWC USB3 DT nodes name
In accordance with the DWC USB3 bindings the corresponding node name is suppose to comply with Generic USB HCD DT schema, which requires the USB nodes to have the name acceptable by the regexp: "^usb(@.*)?" . But a lot of the DWC USB3-compatible nodes defined in the ARM/ARM64 DTS files have name as "^dwc3@.*" or "^usb[1-3]@.*" or even "^dwusb@.*", which will cause the dtbs_check procedure failure. Let's fix the nodes naming to be compatible with the DWC USB3 DT schema to make dtbs_check happy. Note we don't change the DWC USB3-compatible nodes names of arch/arm64/boot/dts/apm/{apm-storm.dtsi,apm-shadowcat.dtsi} since the in-source comment says that the nodes name need to be preserved as "^dwusb@.*" for some backward compatibility. Signed-off-by: Serge Semin --- Please, test the patch out to make sure it doesn't brake the dependent DTS files. I did only a manual grepping of the possible nodes dependencies. --- arch/arm/boot/dts/armada-375.dtsi | 2 +- arch/arm/boot/dts/exynos5250.dtsi | 2 +- arch/arm/boot/dts/exynos54xx.dtsi | 4 ++-- arch/arm/boot/dts/keystone-k2e.dtsi| 4 ++-- arch/arm/boot/dts/keystone.dtsi| 2 +- arch/arm/boot/dts/ls1021a.dtsi | 2 +- arch/arm/boot/dts/omap5-l4.dtsi| 2 +- arch/arm/boot/dts/stih407-family.dtsi | 2 +- arch/arm64/boot/dts/allwinner/sun50i-h6.dtsi | 2 +- arch/arm64/boot/dts/exynos/exynos5433.dtsi | 4 ++-- arch/arm64/boot/dts/exynos/exynos7.dtsi| 2 +- arch/arm64/boot/dts/freescale/fsl-ls1012a.dtsi | 4 ++-- arch/arm64/boot/dts/freescale/fsl-ls1043a.dtsi | 6 +++--- arch/arm64/boot/dts/freescale/fsl-ls1088a.dtsi | 4 ++-- arch/arm64/boot/dts/freescale/fsl-ls208xa.dtsi | 4 ++-- arch/arm64/boot/dts/hisilicon/hi3660.dtsi | 2 +- arch/arm64/boot/dts/qcom/apq8096-db820c.dtsi | 4 ++-- arch/arm64/boot/dts/qcom/ipq8074.dtsi | 4 ++-- arch/arm64/boot/dts/qcom/msm8996.dtsi | 4 ++-- arch/arm64/boot/dts/qcom/msm8998.dtsi | 2 +- arch/arm64/boot/dts/qcom/qcs404-evb.dtsi | 2 +- arch/arm64/boot/dts/qcom/qcs404.dtsi | 4 ++-- arch/arm64/boot/dts/qcom/sc7180.dtsi | 2 +- arch/arm64/boot/dts/qcom/sdm845.dtsi | 4 ++-- arch/arm64/boot/dts/qcom/sm8150.dtsi | 2 +- 25 files changed, 38 insertions(+), 38 deletions(-) diff --git a/arch/arm/boot/dts/armada-375.dtsi b/arch/arm/boot/dts/armada-375.dtsi index 9805e507c695..7f2f24a29e6c 100644 --- a/arch/arm/boot/dts/armada-375.dtsi +++ b/arch/arm/boot/dts/armada-375.dtsi @@ -426,7 +426,7 @@ usb1: usb@54000 { status = "disabled"; }; - usb2: usb3@58000 { + usb2: usb@58000 { compatible = "marvell,armada-375-xhci"; reg = <0x58000 0x2>,<0x5b880 0x80>; interrupts = ; diff --git a/arch/arm/boot/dts/exynos5250.dtsi b/arch/arm/boot/dts/exynos5250.dtsi index e3dbe4166836..ebcf8b40ea81 100644 --- a/arch/arm/boot/dts/exynos5250.dtsi +++ b/arch/arm/boot/dts/exynos5250.dtsi @@ -629,7 +629,7 @@ usb_dwc3 { #size-cells = <1>; ranges; - usbdrd_dwc3: dwc3@1200 { + usbdrd_dwc3: usb@1200 { compatible = "synopsys,dwc3"; reg = <0x1200 0x1>; interrupts = ; diff --git a/arch/arm/boot/dts/exynos54xx.dtsi b/arch/arm/boot/dts/exynos54xx.dtsi index 8aa5117e58ce..339243d19a80 100644 --- a/arch/arm/boot/dts/exynos54xx.dtsi +++ b/arch/arm/boot/dts/exynos54xx.dtsi @@ -148,7 +148,7 @@ usbdrd3_0: usb3-0 { #size-cells = <1>; ranges; - usbdrd_dwc3_0: dwc3@1200 { + usbdrd_dwc3_0: usb@1200 { compatible = "snps,dwc3"; reg = <0x1200 0x1>; interrupts = ; @@ -170,7 +170,7 @@ usbdrd3_1: usb3-1 { #size-cells = <1>; ranges; - usbdrd_dwc3_1: dwc3@1240 { + usbdrd_dwc3_1: usb@1240 { compatible = "snps,dwc3"; reg = <0x1240 0x1>; phys = <&usbdrd_phy1 0>, <&usbdrd_phy1 1>; diff --git a/arch/arm/boot/dts/keystone-k2e.dtsi b/arch/arm/boot/dts/keystone-k2e.dtsi index 2d94faf31fab..d625ad10cfad 100644 --- a/arch/arm/boot/dts/keystone-k2e.dtsi +++ b/arch/arm/boot/dts/keystone-k2e.dtsi @@ -52,7 +52,7 @@ &soc0 { usb: usb@268 { interrupts = ; - dwc3@269 { + usb@269 {