v2: - Various updates called out in individual patches based on review from Andy Shevchenko. Still hacks, but nicer ones now. >From the school of dirty hacks we do to prove something works, enable work to proceed elsewhere: MCTP over I2C from ACPI emulated hosts (both x86 and ARM64). The first 4 patches might be suitable for upstream inclusion, the last two not - though I hope we can move to Niyas' work on ACPI clock management once that is ready. Why do this crazy thing? Ultimately we want a standards based way to use the CXL Fabric Management API FM-API. In real systems that is likely to be driven from a separate 'host' such as a BMC, but for test purposes it is convenient to be able to do that from an QEMU emulated machine that is also capable of using the CXL kernel stack. That CXL kernel stack is currently ACPI only (and people care about x86 for some reason). One of the defined interfaces over which FM-API commands can be issued is MCTP. The kernel MCTP stack has upstream drivers for MCTP over I2C. Upstream QEMU emulates the Aspeed I2C controller with the necessary two way support. There are patches on list adding the MCTP parts https://lore.kernel.org/qemu-devel/20230425063540.46143-2-its@xxxxxxxxxxxxx/ and I've ported an earlier CXL FMAPI EP emulator over to that. ACPI has a 'magic' HID of PRP0001 which allows the use of a device tree binding (mostly) with an ACPI DSDT entry. A suitable chunk is something like (dumped from a working x86 setup) Device (MCTP) { Name (_HID, "PRP0001") // _HID: Hardware ID Name (_DSD, Package (0x02) // _DSD: Device-Specific Data { ToUUID ("daffd814-6eba-4d8c-8a91-bc9bbf4aa301") /* Device Properties for _DSD */, Package (0x03) { Package (0x02) { "compatible", "aspeed,ast2600-i2c-bus" }, Package (0x02) { "bus-frequency", 0x00061A80 }, Package (0x02) { "mctp-controller", One } } }) Name (_CRS, ResourceTemplate () // _CRS: Current Resource Settings { QWordMemory (ResourceProducer, PosDecode, MinFixed, MaxFixed, NonCacheable, ReadWrite, 0x0000000000000000, // Granularity 0x00000004800FC080, // Range Minimum 0x00000004800FC0FF, // Range Maximum 0x0000000000000000, // Translation Offset 0x0000000000000080, // Length ,, , AddressRangeMemory, TypeStatic) Interrupt (ResourceConsumer, Level, ActiveHigh, Shared, ,, ) { 0x00000007, } }) } Device (MCTS) { Name (_HID, "PRP0001") // _HID: Hardware ID Name (_CRS, ResourceTemplate () // _CRS: Current Resource Settings { I2cSerialBusV2 (0x0050, DeviceInitiated, 0x000186A0, AddressingMode7Bit, "\\_SB.MCTP", 0x00, ResourceProducer, , Exclusive, ) }) Name (_DSD, Package (0x02) // _DSD: Device-Specific Data { ToUUID ("daffd814-6eba-4d8c-8a91-bc9bbf4aa301") /* Device Properties for _DSD */, Package (0x01) { Package (0x02) { "compatible", "mctp-i2c-controller" } } }) } QEMU patches will follow soon and will include documentation on how to actually poke this to do something useful. I'll post a reply to this with the link when posted. Cc: Niyas Sait <niyas.sait@xxxxxxxxxx> Cc: Klaus Jensen <its@xxxxxxxxxxxxx> Cc: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> Cc: Jeremy Kerr <jk@xxxxxxxxxxxxxxxxxxxx> Cc: Matt Johnston <matt@xxxxxxxxxxxxxxxxxxxx> Cc: Shesha Bhushan Sreenivasamurthy <sheshas@xxxxxxxxxxx> Jonathan Cameron (6): i2c: acpi: set slave mode flag i2c: aspeed: Use platform_get_irq() instead of opencoding i2c: aspeed: Don't report error when optional dt bus-frequency not supplied i2c: aspeed: switch to generic fw properties. i2c: aspeed: Set the fwnode for the adap->dev HACK: i2c: aspeed: Comment clock out and make reset optional drivers/i2c/busses/i2c-aspeed.c | 43 +++++++++++++++------------------ drivers/i2c/i2c-core-acpi.c | 3 +++ 2 files changed, 22 insertions(+), 24 deletions(-) -- 2.39.2
