2. • Staged for 3.15 (in linux-next):
• Odds and ends: APEI cleanups, white space, minor bugs in ACPI driver
• Reduced Hardware mode: config option, strict enforcement ruled out [0]
• Preliminary patches for moving around some arch-dependent code (simplifies ACPI for non-x86 machines)
• Probable for 3.16:
• Core functionality: load and recognize tables, enable ACPI for AArch64 driver use
• SMP/GIC initialization based on MADT
• Arch timer initialization based on ACPI table GTDT
• Enable APEI for AArch64
• This has a dependency on early_ioremap and PCI from 3.15
• Pending acceptance of Core [1]:
• Driver changes enabling boot of FVP with ACPI only
• Better CPU topology initialization
• Better GIC initialization
• Further non-FVP drivers
• Further Coordination with ASWG proposals
[0] Strict enforcement of reduced hardware mode in Linux will not be possible due to existing x86 platforms that violate the specification; will need to fix any problems this causes for ARMv8 (there
are none yet).
[1] Maybe 3.16, maybe 3.17, depending on what upstream decides to do
ACPI Upstreaming
3. • Grand Unified Firmware Interface
• Why: make it easy to use either FDT or ACPI in the same driver
• What: shim layer in kernel to call right API under the right conditions,
providing a simple API merging FDT and ACPI calls
• How: prototype one or more non-trivial drivers
• When: now -- need to complete prototypes, then RFC
• Where: ideally, new drivers only, as needed for ARMv8 servers
• More Info: Prototype GUFI Card (https://cards.linaro.org/browse/CARD-1059)
A Modest Proposal: GUFI
4. • Introduce drivers/gufi to hold code for the new interface
• drivers/gufi/core.c GUFI initialisation code
int gufi_register_protocol(struct gufi_protocol *prot)
• Called to register ACPI/FDT/... with gufi this tells the
core about the custom mappings to GUFI.
void gufi_unregister_protocol(struct gufi_protocol *prot)
• The reverse of the above.
A Modest Proposal: GUFI Structure
6. • Currently 2 protocols available
• OF: drivers/gufi/of_protocol.c
• ACPI: drivers/gufi/acpi_protocol.c
• This is a prototype so has now been expanded to full list
of datatypes which are available with OF/ACPI yet.
A Modest Proposal: GUFI Structure
8. • drivers/net/ethernet/smsc/smc91x.c
-#if IS_BUILTIN(CONFIG_OF)
- match = of_match_device(of_match_ptr(smc91x_match), &pdev->dev);
- if (match) {
- struct device_node *np = pdev->dev.of_node;
+ match = gufi_match_device(smc91x_gufi_match, &pdev->dev);
+ if (gufi_test_match(match)) {
+ struct gufi_device_node *gdn = gufi_look_for_node(
+ pdev->dev.of_node, ACPI_COMPANION(&pdev->dev));
u32 val;
/* Combination of IO widths supported, default to 16-bit */
- if (!of_property_read_u32(np, "reg-io-width", &val)) {
+ if (!gufi_property_read_u32(gdn, "reg-io-width", &val)) {
• Instead of protocol specific property functions use
generic GUFI.
A Modest Proposal: Converting Driver
9. • drivers/net/ethernet/smsc/smc91x.c
- if (!pd && !match) {
+ if (!pd && !gufi_test_match(match)) {
lp->cfg.flags |= (SMC_CAN_USE_8BIT) ? SMC91X_USE_8BIT : 0;
lp->cfg.flags |= (SMC_CAN_USE_16BIT) ? SMC91X_USE_16BIT : 0;
lp->cfg.flags |= (SMC_CAN_USE_32BIT) ? SMC91X_USE_32BIT : 0;
• Instead of protocol specific device matching use generic
GUFI.
A Modest Proposal: Converting Driver
10. 1. Spin-table enabled SMP initialization is walking away
on ARM64? there is no PSCI support in ACPI now, how
can we initialize SMP in ACPI way?
2. Move ACPI based GIC initialization code into irq-gic.c?
how to find a scalable way to support GICv3 in the
future?
3. Single kernel image for both ACPI and FDT?
4. GIC ID and UID, who should be MPIDR? http://www.
kernelhub.org/?msg=417247&p=2
Issues we got for current patches
11. • ACPI Power Management 2:55pm Loulan 4103-4104
Further Sessions
12. More about Linaro Connect: http://connect.linaro.org
More about Linaro: http://www.linaro.org/about/
More about Linaro engineering: http://www.linaro.org/engineering/
Linaro members: www.linaro.org/members