Integrator devicetree support
cover Integrator devicetree support
This is a patch set for devicetree on the Integrator/AP and Integrator/CP I've been cooking. I did some stepwise development in recent kernels moving to sparse IRQs, per-system clock source then common clk. Now we can do single zImage and also devicetree on this platform, so this is a first patch set.
In case you're wondering why, take down your Torvalds autobiography and read the title. It was also a bit of learning exercise, plus it's fun to run devicetree on this ARMv4 system.
This patch set moves all the non-DT platform code into #ifndef CONFIG_OF sections for clarity. The plan is to delete them after deprecation.
Remaining roadblocks to deleting non-DT code altogether.
- Uncertain about how to do the fpga-irq DT binding so please review closely.
- The Integrator/AP ARM core tiles have add-on logic modules called Integrator/IM-PD1 with a load of PrimeCells and stuff. These spawn off a special bus called "lm" in arch/arm/mach-integrator/lm.c, and the core module support is in impd1.c. It'd probably be pretty easy to create device tree bindings for these, but I have no such module and no clue on where to get one. Is anyone actually using it? Russell?
- The special PCI adapter in the Integrator has to be DT:ed, but AFAICT there is still a problem with how to represent these things in DT. Any hints/examples, Arnd, Rob? arch/arm/mach-integrator/pci_v3.c
Yours,
Linus Walleij
Linus Walleij (6):
ARM: integrator: call common init function from machine
ARM: integrator: check PL010 device name rather than base address
ARM: plat-versatile: add DT support to FPGA IRQ controller
ARM: integrator: initial device tree support
ARM: integrator: convert AMBA devices to device tree
ARM: integrator: convert platform devices to Device Tree
Documentation/devicetree/bindings/arm/arm-boards | 12 + .../devicetree/bindings/arm/versatile-fpga-irq.txt | 35 +++ arch/arm/boot/dts/integratorap.dts | 78 ++++++ arch/arm/boot/dts/integratorcp.dts | 122 +++++++++ arch/arm/mach-integrator/common.h | 3 + arch/arm/mach-integrator/core.c | 14 +- arch/arm/mach-integrator/integrator_ap.c | 214 ++++++++++----- arch/arm/mach-integrator/integrator_cp.c | 296 +++++++++++++-------- arch/arm/plat-versatile/fpga-irq.c | 41 +++ arch/arm/plat-versatile/include/plat/fpga-irq.h | 2 + 10 files changed, 640 insertions(+), 177 deletions(-) create mode 100644 Documentation/devicetree/bindings/arm/versatile-fpga-irq.txt create mode 100644 arch/arm/boot/dts/integratorap.dts create mode 100644 arch/arm/boot/dts/integratorcp.dts
Signature · 1 line
1.7.11.2
cover Integrator devicetree support
Quoted 3 lines · Linus Walleij
This patch set moves all the non-DT platform code into #ifndef
CONFIG_OFsections for clarity. The plan is to delete them after deprecation.
I'm fine with your playing around with this in any way you like, but I think in general we should have a new CONFIG_ATAG symbol, so you can build a kernel that supports booting both with and without DT for the same platform, depending on what the boot loader supports.
Then you can enclose all the non-DT sections in #ifdef CONFIG_ATAG and we can remove them at some later stage.
Quoted 4 lines
- The special PCI adapter in the Integrator has to be DT:ed, but AFAICT there is still a problem with how to represent these things in DT. Any hints/examples, Arnd, Rob?
arch/arm/mach-integrator/pci_v3.c
PCI bindings are a bit tricky but they are well-defined, see www.openfirmware.org/ofwg/bindings/pci/pci2_1.pdf
The main thing to understand is the use of 3-cell address spaces and the interrupt map. Each PCI host must have
#interrupt-cells = <1>;
#size-cells = <2>;
#address-cells = <3>;The "ranges" property is used to transform the PCI address space into the host address space.
The first cell identifies the device and address space using a bit-masked notation. The important part to know here is
0x00000000 config space0x01000000 I/O space0x02000000 memory space0x03000000 64 bit memory space0x42000000 memory space, prefetchable0x43000000 64 bit memory space, prefetchable
Each entry in the ranges property consists of 3 cells for the PCI address start, one cell for a 32 bit parent but start address where this is mapped, and two cells for the size.
For your bus, the ranges then would be defined as
ranges = <0x00000000 0x00000000 0x00000000 /* configuration space: */
0x61000000 0x00000000 0x00100000 /* 16 MB at 61000000 */ 0x01000000 0x00000000 0x00000000 /* I/O space: */ 0x60000000 0x00000000 0x00010000 /* 64KB@0x60000000 */ 0x02000000 0x00000000 0x40000000 /* non-prefetchable mem */ 0x40000000 0x00000000 0x10000000 /* identity-mapped */ 0x42000000 0x00000000 0x50000000 /* prefetchable mem */ 0x50000000 0x00000000 0x10000000> /* identity-mapped */
I'm not quite sure if I understood the comments about LB_BASE0 etc correctly, so it might actually be different.
You also need an interrupt-map, which maps the slots to the connected irq lines in a board-specific way:
interrupt-map = <
/* IDSEL 9 */ 0x4800 0 0 1 &pic 14 /* INT A on slot 9 is irq 14 */ 0x4800 0 0 2 &pic 15 /* INT B on slot 9 is irq 15 */ 0x4800 0 0 3 &pic 16 /* INT C on slot 9 is irq 16 */ 0x4800 0 0 4 &pic 17 /* INT D on slot 9 is irq 17 */ /* IDSEL 10 */ 0x5000 0 0 1 &pic 15 /* INT A on slot 10 is irq 15 */ 0x5000 0 0 2 &pic 16 /* INT B on slot 10 is irq 16 */ 0x5000 0 0 3 &pic 17 /* INT C on slot 10 is irq 17 */ 0x5000 0 0 4 &pic 14 /* INT D on slot 10 is irq 18 */ /* IDSEL 11 */ 0x5800 0 0 1 &pic 16 /* INT A on slot 11 is irq 16 */ 0x5800 0 0 2 &pic 17 /* INT B on slot 11 is irq 17 */ 0x5800 0 0 3 &pic 14 /* INT C on slot 11 is irq 18 */ 0x5800 0 0 4 &pic 15 /* INT D on slot 11 is irq 14 */ /* IDSEL 12 */ 0x6000 0 0 1 &pic 17 /* INT A on slot 12 is irq 17 */ 0x6000 0 0 2 &pic 14 /* INT B on slot 12 is irq 14 */ 0x6000 0 0 3 &pic 15 /* INT C on slot 12 is irq 15 */ 0x6000 0 0 4 &pic 16 /* INT D on slot 12 is irq 16 */ >; Arnd
cover Integrator devicetree support
Quoted 5 lines · Arnd Bergmann
Quoted 3 lines · Linus Walleij
This patch set moves all the non-DT platform code into #ifndef
CONFIG_OFsections for clarity. The plan is to delete them after deprecation.
Quoted 8 lines
I'm fine with your playing around with this in any way you like, but I think in general we should have a new
CONFIG_ATAGsymbol, so you can build a kernel that supports booting both with and without DT for the same platform, depending on what the boot loader supports.Then you can enclose all the non-DT sections in #ifdef
CONFIG_ATAGand we can remove them at some later stage.
Since we already have CONFIG_USE_OF I think CONFIG_USE_ATAGS is more apropriate, but I get the idea. So it begins like this:
arch/arm/Kconfig +7 -0
@@ -1907,6 +1907,12 @@ endmenu menu "Boot options" +config USE_ATAGS + bool "ATAGS support" + default y + help + Include support for ATAGS-based machine descriptions. + config USE_OF bool "Flattened Device Tree support" select OF
@@ -2110,6 +2116,7 @@ config KEXEC config ATAGS_PROC bool "Export atags in procfs" + depends on USE_ATAGS depends on KEXEC default y help
But then we hit #ifdeffing and compiling out a lot of dangerous stuff in arch/arm/kernel and then I get the creeps. I can try, but I think this is typical Nico territory for example...
Yours,
Linus Walleij
cover Integrator devicetree support
Quoted 3 lines · Linus Walleij
Since we already have
CONFIG_USE_OFI thinkCONFIG_USE_ATAGSis more apropriate, but I get the idea.
There is both CONFIG_USE_OF and CONFIG_OF. The former is what a platform selects and it turns on a few other options as well, while the latter is the one that you typically check in #ifdef.
We might need both of them for ATAG as well, but I think if we have just one, CONFIG_ATAGS or CONFIG_ATAG would be nicer because it's slightly shorter.
Quoted 28 lines
So it begins like this:
arch/arm/Kconfig +7
-0@@ -1907,6 +1907,12 @@ endmenu menu "Boot options" +config USE_ATAGS + bool "ATAGS support" + default y + help + Include support for ATAGS-based machine descriptions. + config USE_OF bool "Flattened Device Tree support" select OF@@ -2110,6 +2116,7 @@ config KEXEC config ATAGS_PROC bool "Export atags in procfs" + depends on USE_ATAGS depends on KEXEC default y help
Quoted 3 lines
But then we hit #ifdeffing and compiling out a lot of dangerous stuff in
arch/arm/kerneland then I get the creeps. I can try, but I think this is typical Nico territory for example...
We don't have to enclose every single usage of ATAGS with this, but I think it's helpful in platforms that you want to migrate away from ATAG to DT-only. I think we should also use this method in mach-ux500, for instance.
Putting MACHINE_START and DT_MACHINE_START inside of CONFIG_ATAG and CONFIG_OF respectively would help ensure that we don't accidentally build support for a board that cannot work.
Arnd
cover Integrator devicetree support
In reply to Linus WalleijQuoted 3 lines · Linus Walleij
But then we hit #ifdeffing and compiling out a lot of dangerous stuff in
arch/arm/kerneland then I get the creeps. I can try, but I think this is typical Nico territory for example...
I did post a patch moving all the ATAG stuff to a separate source file a while ago, exactly for that purpose. I should try to dig it out and refresh it.
Nicolas
cover Integrator devicetree support
Quoted 9 lines · Nicolas Pitre
Quoted 3 lines · Linus Walleij
But then we hit #ifdeffing and compiling out a lot of dangerous stuff in
arch/arm/kerneland then I get the creeps. I can try, but I think this is typical Nico territory for example...I did post a patch moving all the ATAG stuff to a separate source file a while ago, exactly for that purpose. I should try to dig it out and refresh it.
OK... In this case "a while ago" = 2 years!
Please see the 2 patches to follow.
Nicolas