Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
Original-Nachricht Datum: Tue, 13 Jan 2009 16:01:42 +1100 Von: Benjamin Herrenschmidt b...@kernel.crashing.org An: Gerhard Pircher gerhard_pirc...@gmx.net CC: linuxppc-dev@ozlabs.org, grant.lik...@secretlab.ca Betreff: Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards I think I throw away the IDE controller node for now, as libata just reads the PCI register settings (progif) and I guess the IDE subsystem will do the same in the future. Well... if all AmigaOne use a 8259, they probably use the same interrupt routing except for PCI slots. In which case, I would -still- prefer if you had a proper interrupt tree, and at runtime or boot-wrapper time, fixed up the PCI host interrupt-map property to contain the right values for a given board. That would be the ideal solution. But there is no reliable way to identify, if the board is an AmigaOneSE, AmigaOneXE or an uA1. AFAIK the firmware checks for a specifc onboard PCI sound chip to differentiate between an A1XE and the uA1, but this sound chip is also used on various PCI sound cards. I don't really like that approach. I'm not going to include the new platform in .29, it's way too late anyway (it should have been published a couple of weeks before the merge window at least I'd say) so we have some time til .30 to polish things a bit. Okay. Some months more don't matter after all these years. :-) Another area to look at is to cleanup the non-coherent DMA thingy. The config option should just enable a set of non-coherent backend ops, but we should still be able to decide which ones to use (coherent vs. non-coherent) at runtime. What about drivers that need to handle non coherent DMA differently (e.g. because they map DMA memory to userspace)? Will the #ifdef CONFIG_NOT_COHERENT_CACHE be replaced with calls to dma_is_consistent()? (for example in the radeon driver, that you fixed for non coherent DMA platforms in 2.6.25). If I may express a wish: a reworked CPU feature fixup code for PPC32 will avoid bigger problems with the AmigaOne platform in the future. Thanks a lot! Gerhard -- Pt! Schon vom neuen GMX MultiMessenger gehört? Der kann`s mit allen: http://www.gmx.net/de/go/multimessenger ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
Original-Nachricht Datum: Mon, 12 Jan 2009 16:12:18 +1100 Von: Benjamin Herrenschmidt b...@kernel.crashing.org An: Gerhard Pircher gerhard_pirc...@gmx.net CC: Grant Likely grant.lik...@secretlab.ca, linuxppc-dev@ozlabs.org Betreff: Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards The code in the kernel that retreives the interrupt that way is clearly marked as a fishy workaround for bogus firmwares :-) bogus applies a little bit to the PCI init code of the A1 firmware. ;-) But I'm not going to reject things based on that, it will work for simple board using really only legacy interrupts like yours... Thanks! For the flattened device tree, I think we've settled on the convention that every node with an IRQ connection should have both the interrupt-parent and interrupts properties. (ie. don't rely on the parent node's interrupt-parent property.) Even for ISA devices? I disagree with Grant here. Especially in simple ISA cases like that, there's really no point in bloating the device-tree. Okay, so I leave it as it is. Can this PCI device be probed? Typically PCI devices don't get added to the flattened device tree because PCI is a probeable bus. Yes, it can be probed. I thought it would be a good idea to include it, because the IDE controller operates in legacy mode. I planned to specify the two legacy interrupts in this node (as you can see), but the kernel didn't like them. Well, the kernel just didn't make use of them I'd say :-) But that can probably be fixed with the appropriate hacks. I think I throw away the IDE controller node for now, as libata just reads the PCI register settings (progif) and I guess the IDE subsystem will do the same in the future. Gerhard -- Sensationsangebot verlängert: GMX FreeDSL - Telefonanschluss + DSL für nur 16,37 Euro/mtl.!* http://dsl.gmx.de/?ac=OM.AD.PD003K1308T4569a ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
I think I throw away the IDE controller node for now, as libata just reads the PCI register settings (progif) and I guess the IDE subsystem will do the same in the future. Well... if all AmigaOne use a 8259, they probably use the same interrupt routing except for PCI slots. In which case, I would -still- prefer if you had a proper interrupt tree, and at runtime or boot-wrapper time, fixed up the PCI host interrupt-map property to contain the right values for a given board. I'm not going to include the new platform in .29, it's way too late anyway (it should have been published a couple of weeks before the merge window at least I'd say) so we have some time til .30 to polish things a bit. Another area to look at is to cleanup the non-coherent DMA thingy. The config option should just enable a set of non-coherent backend ops, but we should still be able to decide which ones to use (coherent vs. non-coherent) at runtime. Cheers, Ben. ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
On Wed, 2009-01-07 at 09:41 -0700, Grant Likely wrote: For the flattened device tree, I think we've settled on the convention that every node with an IRQ connection should have both the interrupt-parent and interrupts properties. (ie. don't rely on the parent node's interrupt-parent property.) Ah ? I use the trick of relying on the parent node regulary :-) Where did we discuss that ? Ben. ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
Yes, all AmigaOne boards have physical PCI slots (at least 1). The different interrupt routing wasn't a problem so far, as the firmware writes the IRQ number to the interrupt line register of every PCI device. The code in the kernel that retreives the interrupt that way is clearly marked as a fishy workaround for bogus firmwares :-) But I'm not going to reject things based on that, it will work for simple board using really only legacy interrupts like yours... Currently the kernel reads the IRQ number from this register, if there is no interrupt mapping property. I know that it's not a good idea to rely on kernel fallback behavior, but it makes a lot of things easier in this case. Sort-of. As long as it's really 8259 interrupts, I suppose it's acceptable. For the flattened device tree, I think we've settled on the convention that every node with an IRQ connection should have both the interrupt-parent and interrupts properties. (ie. don't rely on the parent node's interrupt-parent property.) Even for ISA devices? I disagree with Grant here. Especially in simple ISA cases like that, there's really no point in bloating the device-tree. Can this PCI device be probed? Typically PCI devices don't get added to the flattened device tree because PCI is a probeable bus. Yes, it can be probed. I thought it would be a good idea to include it, because the IDE controller operates in legacy mode. I planned to specify the two legacy interrupts in this node (as you can see), but the kernel didn't like them. Well, the kernel just didn't make use of them I'd say :-) But that can probably be fixed with the appropriate hacks. Cheers, Ben. ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
On Sun, Jan 11, 2009 at 10:07 PM, Benjamin Herrenschmidt b...@kernel.crashing.org wrote: On Wed, 2009-01-07 at 09:41 -0700, Grant Likely wrote: For the flattened device tree, I think we've settled on the convention that every node with an IRQ connection should have both the interrupt-parent and interrupts properties. (ie. don't rely on the parent node's interrupt-parent property.) Ah ? I use the trick of relying on the parent node regulary :-) Where did we discuss that ? It is also entirely possible that I'm smoking something exotic. g. -- Grant Likely, B.Sc., P.Eng. Secret Lab Technologies Ltd. ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
[PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
This device tree does not provide the correct CPU name, as various CPU models and revisions are used in AmigaOnes. Also the PCI root node does not contain a interrupt mapping property, as all boards have different interrupt routing. However the kernel can do a 1:1 mapping of all PCI interrupts, as only i8259 legacy interrupts are used. Signed-off-by: Gerhard Pircher gerhard_pirc...@gmx.net --- arch/powerpc/boot/dts/amigaone.dts | 233 1 files changed, 233 insertions(+), 0 deletions(-) create mode 100644 arch/powerpc/boot/dts/amigaone.dts diff --git a/arch/powerpc/boot/dts/amigaone.dts b/arch/powerpc/boot/dts/amigaone.dts new file mode 100644 index 000..9794bbc --- /dev/null +++ b/arch/powerpc/boot/dts/amigaone.dts @@ -0,0 +1,233 @@ +/* + * AmigaOne Device Tree Source + * + * Copyright 2008 Gerhard Pircher (gerhard_pirc...@gmx.net) + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License as published by the + * Free Software Foundation; either version 2 of the License, or (at your + * option) any later version. + */ + +/dts-v1/; + +/ { + model = AmigaOne; + compatible = eyetech,amigaone,mai-logic,teron; + coherency-off; + #address-cells = 1; + #size-cells = 1; + + cpus { + #cpus = 1; + #address-cells = 1; + #size-cells = 0; + + c...@0 { + device_type = cpu; + reg = 0; + d-cache-line-size = 32; // 32 bytes + i-cache-line-size = 32; // 32 bytes + d-cache-size = 32768; // L1, 32K + i-cache-size = 32768; // L1, 32K + timebase-frequency = 0; // 33.3 MHz, from U-boot + clock-frequency = 0; // From U-boot + bus-frequency = 0;// From U-boot + }; + }; + + memory { + device_type = memory; + reg = 0 0;// From U-boot + }; + + p...@8000 { + device_type = pci; + compatible = mai-logic,articia-s; + bus-frequency = ; + bus-range = 0 0xff; + ranges = 0x0100 0 0x 0xfe00 0 0x00c0 // PCI I/O + 0x0200 0 0x8000 0x8000 0 0x7d00 // PCI memory + 0x0200 0 0x 0xfd00 0 0x0100; // PCI alias memory (ISA) + 8259-interrupt-acknowledge = 0xfef0; + /* Do not define a interrupt-parent here, if there is no interrupt-map property. */ + #address-cells = 3; + #size-cells = 2; + + h...@0 { + compatible = pciclass,0600; + vendor-id = 0x10cc; + device-id = 0x0660; + revision-id = 0x0001; + class-code = 0x0006; + subsystem-id = 0; + subsystem-vendor-id = 0; + devsel-speed = 0x0001; + 66mhz-capable; + min-grant = 0; + max-latency = 0; + // AGP aperture is unset. +// reg = 0x4210 0 0x 0 0x0040; +// assigned-addresses = 0x4210 0 0x 0 0x0040; + }; + + i...@7 { + device_type = isa; + compatible = pciclass,0601; + vendor-id = 0x1106; + device-id = 0x0686; + revision-id = 0x0010; + class-code = 0x00060100; + subsystem-id = 0; + subsystem-vendor-id = 0; + devsel-speed = 0x0001; + min-grant = 0; + max-latency = 0; + /* First 64k for I/O at 0x0 on PCI mapped to 0x0 on ISA. */ + ranges = 0x0001 0 0x0100 0 0x 0x0001; + interrupt-parent = i8259; + #interrupt-cells = 2; + #address-cells = 2; + #size-cells = 1; + + dma-control...@0 { + device_type = dma-controller; + compatible = pnpPNP,200; + reg = 1 0x 0x0020 + 1 0x0080 0x0010 + 1 0x00c0 0x0020; + /* Channel 4 reserverd, cascade mode, 2x32k transfer/counter
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
On Wed, Jan 7, 2009 at 7:01 AM, Gerhard Pircher gerhard_pirc...@gmx.net wrote: This device tree does not provide the correct CPU name, as various CPU models and revisions are used in AmigaOnes. Also the PCI root node does not contain a interrupt mapping property, as all boards have different interrupt routing. However the kernel can do a 1:1 mapping of all PCI interrupts, as only i8259 legacy interrupts are used. Sounds to me like you need different device tree variants for each of the AmigaOne boards. Do any of the boards have physical PCI slots? If so, then the lack of an interrupt map will break them. Signed-off-by: Gerhard Pircher gerhard_pirc...@gmx.net --- arch/powerpc/boot/dts/amigaone.dts | 233 1 files changed, 233 insertions(+), 0 deletions(-) create mode 100644 arch/powerpc/boot/dts/amigaone.dts diff --git a/arch/powerpc/boot/dts/amigaone.dts b/arch/powerpc/boot/dts/amigaone.dts new file mode 100644 index 000..9794bbc --- /dev/null +++ b/arch/powerpc/boot/dts/amigaone.dts @@ -0,0 +1,233 @@ +/* + * AmigaOne Device Tree Source + * + * Copyright 2008 Gerhard Pircher (gerhard_pirc...@gmx.net) + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License as published by the + * Free Software Foundation; either version 2 of the License, or (at your + * option) any later version. + */ + +/dts-v1/; + +/ { + model = AmigaOne; + compatible = eyetech,amigaone,mai-logic,teron; Experience has shown that trying to claim one board to be compatible with another is too ambiguous. It is better to make the board level compatible property have a single value specifying the exact board model and have the platform support code include a list of supported board models. Otherwise you end up with odd heuristic code to try and differentiate between the models (for bug fixes and such). + coherency-off; + #address-cells = 1; + #size-cells = 1; + + cpus { + #cpus = 1; + #address-cells = 1; + #size-cells = 0; + + c...@0 { + device_type = cpu; + reg = 0; + d-cache-line-size = 32; // 32 bytes + i-cache-line-size = 32; // 32 bytes + d-cache-size = 32768; // L1, 32K + i-cache-size = 32768; // L1, 32K + timebase-frequency = 0; // 33.3 MHz, from U-boot + clock-frequency = 0; // From U-boot + bus-frequency = 0;// From U-boot + }; + }; + + memory { + device_type = memory; + reg = 0 0;// From U-boot + }; + + p...@8000 { + device_type = pci; + compatible = mai-logic,articia-s; + bus-frequency = ; + bus-range = 0 0xff; + ranges = 0x0100 0 0x 0xfe00 0 0x00c0 // PCI I/O + 0x0200 0 0x8000 0x8000 0 0x7d00 // PCI memory + 0x0200 0 0x 0xfd00 0 0x0100; // PCI alias memory (ISA) + 8259-interrupt-acknowledge = 0xfef0; + /* Do not define a interrupt-parent here, if there is no interrupt-map property. */ + #address-cells = 3; + #size-cells = 2; + + h...@0 { + compatible = pciclass,0600; + vendor-id = 0x10cc; + device-id = 0x0660; + revision-id = 0x0001; + class-code = 0x0006; + subsystem-id = 0; + subsystem-vendor-id = 0; + devsel-speed = 0x0001; + 66mhz-capable; + min-grant = 0; + max-latency = 0; + // AGP aperture is unset. +// reg = 0x4210 0 0x 0 0x0040; +// assigned-addresses = 0x4210 0 0x 0 0x0040; + }; What does this node describe? Is it something that isn't probeable? + 8...@60 { + device_type = 8042; + reg = 1 0x0060 0x0001 + 1 0x0064 0x0001; + // IRQ1, IRQ12 (rising edge) + interrupts = 1 3 12 3; For the flattened device tree, I think we've settled on the convention that every node with an IRQ connection should have both the
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
On Wed, Jan 07, 2009 at 09:41:14AM -0700, Grant Likely wrote: + 8...@60 { + device_type = 8042; + reg = 1 0x0060 0x0001 + 1 0x0064 0x0001; + // IRQ1, IRQ12 (rising edge) + interrupts = 1 3 12 3; For the flattened device tree, I think we've settled on the convention that every node with an IRQ connection should have both the interrupt-parent and interrupts properties. (ie. don't rely on the parent node's interrupt-parent property.) Why? -Scott ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
On Wed, Jan 7, 2009 at 3:10 PM, Scott Wood scottw...@freescale.com wrote: On Wed, Jan 07, 2009 at 09:41:14AM -0700, Grant Likely wrote: + 8...@60 { + device_type = 8042; + reg = 1 0x0060 0x0001 + 1 0x0064 0x0001; + // IRQ1, IRQ12 (rising edge) + interrupts = 1 3 12 3; For the flattened device tree, I think we've settled on the convention that every node with an IRQ connection should have both the interrupt-parent and interrupts properties. (ie. don't rely on the parent node's interrupt-parent property.) Why? Defensive programming. To not rely on implicit relationships g. -- Grant Likely, B.Sc., P.Eng. Secret Lab Technologies Ltd. ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
Re: [PATCH 2/5] powerpc: Generic device tree for all AmigaOne boards
On Wed, Jan 7, 2009 at 3:23 PM, Scott Wood scottw...@freescale.com wrote: Grant Likely wrote: On Wed, Jan 7, 2009 at 3:10 PM, Scott Wood scottw...@freescale.com wrote: Why? Defensive programming. To not rely on implicit relationships It doesn't seem any more likely to introduce a fault by specifying the interrupt controller in one place than in many places. We do the same thing with #address-cells and #size-cells for the same reason. You can get away with leaving them out; but being explicit is more clear. g. -- Grant Likely, B.Sc., P.Eng. Secret Lab Technologies Ltd. ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev