Author: ranma Date: Wed Dec 8 22:40:12 2010 New Revision: 6153 URL: https://tracker.coreboot.org/trac/coreboot/changeset/6153
Log: Tobias Diedrich wrote:
Definitively a iasl problem, it can't even disassemble it's own output back to something equivalent to the input file. It seems to be generating Bytecode for the Add where it shouldn't.
Here is a solution using the SSDT.
Unfortunately iasl does not resolve simple arithmetic at compile time, so we can not use Add(DEFAULT_PMBASE, PCNTRL) in the Processor statement. This patch instead dynamically generates the processor statement. I can't use the speedstep generate_cpu_entries() directly since the cpu doesn't support speedstep. For now the code is in the southbridge directory, but maybe it should go into cpu/intel/ somewhere. IIRC notebook cpus of the era can already have speedstep, so it would probably be possible to pair the i82371eb with a speedstep-capable cpu... Also, I don't know if multiprocessor boards (abit bp6?) would need to be handled differently.
Abuild-tested.
Signed-off-by: Tobias Diedrich ranma+coreboot@tdiedrich.de Acked-by: Peter Stuge peter@stuge.se
Added: trunk/src/cpu/intel/car/Makefile.inc trunk/src/cpu/intel/car/disable_cache_as_ram.c trunk/src/cpu/intel/car/post_cache_as_ram.c Modified: trunk/src/mainboard/asus/p2b/dsdt.asl trunk/src/southbridge/intel/i82371eb/acpi_tables.c
Added: trunk/src/cpu/intel/car/Makefile.inc ==============================================================================
Added: trunk/src/cpu/intel/car/disable_cache_as_ram.c ==============================================================================
Added: trunk/src/cpu/intel/car/post_cache_as_ram.c ==============================================================================
Modified: trunk/src/mainboard/asus/p2b/dsdt.asl ============================================================================== --- trunk/src/mainboard/asus/p2b/dsdt.asl Wed Dec 8 20:58:30 2010 (r6152) +++ trunk/src/mainboard/asus/p2b/dsdt.asl Wed Dec 8 22:40:12 2010 (r6153) @@ -21,15 +21,6 @@
DefinitionBlock ("DSDT.aml", "DSDT", 2, "CORE ", "COREBOOT", 1) { - /* Define the main processor.*/ - Scope (_PR) - { - /* Looks like the P_CNT field can't be a name or method (except - * builtins like Add()) and has to be hardcoded or generated - * into SSDT */ - Processor (CPU0, 0x01, Add(DEFAULT_PMBASE, PCNTRL), 0x06) {} - } - /* For now only define 2 power states: * - S0 which is fully on * - S5 which is soft off
Modified: trunk/src/southbridge/intel/i82371eb/acpi_tables.c ============================================================================== --- trunk/src/southbridge/intel/i82371eb/acpi_tables.c Wed Dec 8 20:58:30 2010 (r6152) +++ trunk/src/southbridge/intel/i82371eb/acpi_tables.c Wed Dec 8 22:40:12 2010 (r6153) @@ -26,9 +26,46 @@ #include <arch/smp/mpspec.h> #include <device/device.h> #include <device/pci_ids.h> +#include "i82371eb.h"
extern const unsigned char AmlCode[];
+static int determine_total_number_of_cores(void) +{ + device_t cpu; + int count = 0; + for(cpu = all_devices; cpu; cpu = cpu->next) { + if ((cpu->path.type != DEVICE_PATH_APIC) || + (cpu->bus->dev->path.type != DEVICE_PATH_APIC_CLUSTER)) { + continue; + } + if (!cpu->enabled) { + continue; + } + count++; + } + return count; +} + +void generate_cpu_entries(void) +{ + int len; + int len_pr; + int cpu, pcontrol_blk=DEFAULT_PMBASE+PCNTRL, plen=6; + int numcpus = determine_total_number_of_cores(); + printk(BIOS_DEBUG, "Found %d CPU(s).\n", numcpus); + + /* without the outer scope, furhter ssdt addition will end up + * within the processor statement */ + len = acpigen_write_scope("\_PR"); + for (cpu=0; cpu < numcpus; cpu++) { + len_pr = acpigen_write_processor(cpu, pcontrol_blk, plen); + acpigen_patch_len(len_pr - 1); + len += len_pr; + } + acpigen_patch_len(len - 1); +} + unsigned long __attribute__((weak)) acpi_fill_slit(unsigned long current) { // Not implemented @@ -57,6 +94,10 @@ const char *oem_table_id) { acpigen_write_mainboard_resources("\_SB.PCI0.MBRS", "_CRS"); + /* generate_cpu_entries() generates weird bytecode and has to come + * last or else the following entries will end up inside the + * processor scope */ + generate_cpu_entries(); return (unsigned long) acpigen_get_current(); }
* repository service svn@coreboot.org [101208 22:40]:
Author: ranma Date: Wed Dec 8 22:40:12 2010 New Revision: 6153 URL: https://tracker.coreboot.org/trac/coreboot/changeset/6153
Log: This patch instead dynamically generates the processor statement. I can't use the speedstep generate_cpu_entries() directly since the cpu doesn't support speedstep.
Any chance to move the code to some subdirectory of cpu/intel? The southbridge seems to be the wrong place for CPU ACPI code generation.
Stefan
Stefan Reinauer wrote:
- repository service svn@coreboot.org [101208 22:40]:
Author: ranma Date: Wed Dec 8 22:40:12 2010 New Revision: 6153 URL: https://tracker.coreboot.org/trac/coreboot/changeset/6153
Log: This patch instead dynamically generates the processor statement. I can't use the speedstep generate_cpu_entries() directly since the cpu doesn't support speedstep.
Any chance to move the code to some subdirectory of cpu/intel? The southbridge seems to be the wrong place for CPU ACPI code generation.
Any suggestion for a directory name? 'non-speedstep' maybe? I first wanted to add it to the speedstep ssdt generator, but since it removes most of the code and is for pre-speedstep CPUs I put it into the chipset directory for now.
* repository service svn@coreboot.org [101208 22:40]:
Author: ranma Date: Wed Dec 8 22:40:12 2010 New Revision: 6153 URL: https://tracker.coreboot.org/trac/coreboot/changeset/6153
Signed-off-by: Tobias Diedrich ranma+coreboot@tdiedrich.de Acked-by: Peter Stuge peter@stuge.se
Added: trunk/src/cpu/intel/car/Makefile.inc trunk/src/cpu/intel/car/disable_cache_as_ram.c trunk/src/cpu/intel/car/post_cache_as_ram.c
Added: trunk/src/cpu/intel/car/Makefile.inc
Added: trunk/src/cpu/intel/car/disable_cache_as_ram.c
Added: trunk/src/cpu/intel/car/post_cache_as_ram.c
WTH is this?