DESCRIPTION: ---------------------------------------------------- ## lnxi-patch-8 ## src/mainboard/arima/hdama/mptable.c Add smp_write_processors_inorder() to workaround 2.6.11 bug. dynamically compute ioapic_id vs. hardcode. Add SATA support. Added reboot_if_hotswap() to catch bad board strapping hardware bug.
DIFFSTAT: ----------------------------------------------------
mptable.c | 213 ++++++++++++++++++++++++++++++++++++++++++++++++-------------- 1 files changed, 167 insertions(+), 46 deletions(-)
PATCH: ----------------------------------------------------
Index: mainboard/arima/hdama/mptable.c =================================================================== --- mainboard/arima/hdama/mptable.c (revision 1105) +++ mainboard/arima/hdama/mptable.c (working copy) @@ -3,7 +3,61 @@ #include <device/pci.h> #include <string.h> #include <stdint.h> +#include <cpu/x86/lapic.h> +#include <arch/cpu.h> +#include <arch/io.h>
+#define HT_INIT_CONTROL 0x6c +#define HTIC_BIOSR_Detect (1<<5) + +/* If we assume a symmetric processor configuration we can + * get all of the information we need to write the processor + * entry from the bootstrap processor. + * Plus I don't think linux really even cares. + * Having the proper apicid's in the table so the non-bootstrap + * processors can be woken up should be enough. + */ +void smp_write_processors_inorder(struct mp_config_table *mc) +{ + int boot_apic_id; + int order_id; + unsigned apic_version; + unsigned cpu_features; + unsigned cpu_feature_flags; + struct cpuid_result result; + device_t cpu; + + boot_apic_id = lapicid(); + apic_version = lapic_read(LAPIC_LVR) & 0xff; + result = cpuid(1); + cpu_features = result.eax; + cpu_feature_flags = result.edx; + /* order the output of the cpus to fix a bug in kernel 6 11 */ + for(order_id = 0;order_id <256; order_id++) { + for(cpu = all_devices; cpu; cpu = cpu->next) { + unsigned long cpu_flag; + if ((cpu->path.type != DEVICE_PATH_APIC) || + (cpu->bus->dev->path.type != DEVICE_PATH_APIC_CLUSTER)) + { + continue; + } + if (!cpu->enabled) { + continue; + } + cpu_flag = MPC_CPU_ENABLED; + if (boot_apic_id == cpu->path.u.apic.apic_id) { + cpu_flag = MPC_CPU_ENABLED | MPC_CPU_BOOTPROCESSOR; + } + if(cpu->path.u.apic.apic_id == order_id) { + smp_write_processor(mc, + cpu->path.u.apic.apic_id, apic_version, + cpu_flag, cpu_features, cpu_feature_flags); + break; + } + } + } +} + static unsigned node_link_to_bus(unsigned node, unsigned link) { device_t dev; @@ -38,6 +92,21 @@ return 0; }
+unsigned max_apicid(void) +{ + unsigned max_apicid; + device_t dev; + max_apicid = 0; + for(dev = all_devices; dev; dev = dev->next) { + if (dev->path.type != DEVICE_PATH_APIC) + continue; + if (dev->path.u.apic.apic_id > max_apicid) { + max_apicid = dev->path.u.apic.apic_id; + } + } + return max_apicid; +} + void *smp_write_config_table(void *v) { static const char sig[4] = "PCMP"; @@ -50,6 +119,10 @@ unsigned char bus_8131_1; unsigned char bus_8131_2; unsigned char bus_8111_1; + unsigned apicid_base; + unsigned apicid_8111; + unsigned apicid_8131_1; + unsigned apicid_8131_2;
mc = (void *)(((char *)v) + SMP_FLOATING_TABLE_LEN); memset(mc, 0, sizeof(*mc)); @@ -68,8 +141,12 @@ mc->mpe_checksum = 0; mc->reserved = 0;
- smp_write_processors(mc); + smp_write_processors_inorder(mc);
+ apicid_base = max_apicid() + 1; + apicid_8111 = apicid_base; + apicid_8131_1 = apicid_base + 1; + apicid_8131_2 = apicid_base + 2; { device_t dev;
@@ -124,7 +201,7 @@ smp_write_bus(mc, bus_isa, "ISA ");
/* IOAPIC handling */ - smp_write_ioapic(mc, 2, 0x11, 0xfec00000); + smp_write_ioapic(mc, apicid_8111, 0x11, 0xfec00000); { device_t dev; struct resource *res; @@ -133,7 +210,7 @@ if (dev) { res = find_resource(dev, PCI_BASE_ADDRESS_0); if (res) { - smp_write_ioapic(mc, 0x03, 0x11, res->base); + smp_write_ioapic(mc, apicid_8131_1, 0x11, res->base); } } /* 8131 apic 4 */ @@ -141,44 +218,44 @@ if (dev) { res = find_resource(dev, PCI_BASE_ADDRESS_0); if (res) { - smp_write_ioapic(mc, 0x04, 0x11, res->base); + smp_write_ioapic(mc, apicid_8131_2, 0x11, res->base); } } }
/* ISA backward compatibility interrupts */ smp_write_intsrc(mc, mp_ExtINT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x00, 0x02, 0x00); + bus_isa, 0x00, apicid_8111, 0x00); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x01, 0x02, 0x01); + bus_isa, 0x01, apicid_8111, 0x01); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x00, 0x02, 0x02); + bus_isa, 0x00, apicid_8111, 0x02); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x03, 0x02, 0x03); + bus_isa, 0x03, apicid_8111, 0x03); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x04, 0x02, 0x04); + bus_isa, 0x04, apicid_8111, 0x04); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x05, 0x02, 0x05); + bus_isa, 0x05, apicid_8111, 0x05); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x06, 0x02, 0x06); + bus_isa, 0x06, apicid_8111, 0x06); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x07, 0x02, 0x07); + bus_isa, 0x07, apicid_8111, 0x07); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x08, 0x02, 0x08); + bus_isa, 0x08, apicid_8111, 0x08); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x09, 0x02, 0x09); + bus_isa, 0x09, apicid_8111, 0x09); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x0a, 0x02, 0x0a); + bus_isa, 0x0a, apicid_8111, 0x0a); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x0b, 0x02, 0x0b); + bus_isa, 0x0b, apicid_8111, 0x0b); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x0c, 0x02, 0x0c); + bus_isa, 0x0c, apicid_8111, 0x0c); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x0d, 0x02, 0x0d); + bus_isa, 0x0d, apicid_8111, 0x0d); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x0e, 0x02, 0x0e); + bus_isa, 0x0e, apicid_8111, 0x0e); smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, - bus_isa, 0x0f, 0x02, 0x0f); + bus_isa, 0x0f, apicid_8111, 0x0f);
/* Standard local interrupt assignments */ smp_write_lintsrc(mc, mp_ExtINT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, @@ -188,46 +265,48 @@
/* PCI Ints: Type Trigger Polarity Bus ID PCIDEVNUM|IRQ APIC ID PIN# */ /* On board nics */ - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x03<<2)|0, 0x02, 0x13); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x04<<2)|0, 0x02, 0x13); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x03<<2)|0, apicid_8111, 0x13); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x04<<2)|0, apicid_8111, 0x13); + /* On board SATA */ + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x05<<2)|0, apicid_8111, 0x11);
/* PCI Slot 1 */ - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x01<<2)|0, 0x02, 0x11); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x01<<2)|1, 0x02, 0x12); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x01<<2)|2, 0x02, 0x13); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x01<<2)|3, 0x02, 0x10); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x01<<2)|0, apicid_8111, 0x11); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x01<<2)|1, apicid_8111, 0x12); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x01<<2)|2, apicid_8111, 0x13); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x01<<2)|3, apicid_8111, 0x10);
/* PCI Slot 2 */ - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x02<<2)|0, 0x02, 0x12); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x02<<2)|1, 0x02, 0x13); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x02<<2)|2, 0x02, 0x10); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x02<<2)|3, 0x02, 0x11); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x02<<2)|0, apicid_8111, 0x12); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x02<<2)|1, apicid_8111, 0x13); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x02<<2)|2, apicid_8111, 0x10); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_2, (0x02<<2)|3, apicid_8111, 0x11);
/* PCI Slot 3 */ - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x01<<2)|0, 0x02, 0x11); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x01<<2)|1, 0x02, 0x12); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x01<<2)|2, 0x02, 0x13); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x01<<2)|3, 0x02, 0x10); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x01<<2)|0, apicid_8111, 0x11); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x01<<2)|1, apicid_8111, 0x12); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x01<<2)|2, apicid_8111, 0x13); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x01<<2)|3, apicid_8111, 0x10);
/* PCI Slot 4 */ - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x02<<2)|0, 0x02, 0x12); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x02<<2)|1, 0x02, 0x13); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x02<<2)|2, 0x02, 0x10); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x02<<2)|3, 0x02, 0x11); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x02<<2)|0, apicid_8111, 0x12); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x02<<2)|1, apicid_8111, 0x13); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x02<<2)|2, apicid_8111, 0x10); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8131_1, (0x02<<2)|3, apicid_8111, 0x11);
/* PCI Slot 5 */ #warning "FIXME get the irqs right, it's just hacked to work for now" - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x05<<2)|0, 0x02, 0x11); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x05<<2)|1, 0x02, 0x12); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x05<<2)|2, 0x02, 0x13); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x05<<2)|3, 0x02, 0x10); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x05<<2)|0, apicid_8111, 0x11); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x05<<2)|1, apicid_8111, 0x12); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x05<<2)|2, apicid_8111, 0x13); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x05<<2)|3, apicid_8111, 0x10);
/* PCI Slot 6 */ #warning "FIXME get the irqs right, it's just hacked to work for now" - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x04<<2)|0, 0x02, 0x10); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x04<<2)|1, 0x02, 0x11); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x04<<2)|2, 0x02, 0x12); - smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x04<<2)|3, 0x02, 0x13); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x04<<2)|0, apicid_8111, 0x10); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x04<<2)|1, apicid_8111, 0x11); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x04<<2)|2, apicid_8111, 0x12); + smp_write_intsrc(mc, mp_INT, MP_IRQ_TRIGGER_DEFAULT|MP_IRQ_POLARITY_DEFAULT, bus_8111_1, (0x04<<2)|3, apicid_8111, 0x13);
/* There is no extension information... */
@@ -239,9 +318,51 @@ return smp_next_mpe_entry(mc); }
+void reboot_if_hotswap(void) +{ + /* Hack patch work around for hot swap enable 33mhz problem */ + device_t dev; + uint32_t data; + unsigned long htic; + int reset; + int i; + + reset = 0; + printk_debug("Looking for bad PCIX MHz input\n"); + dev = dev_find_slot(1, PCI_DEVFN(0x02,0)); + data = pci_read_config32(dev, 0xa0); + if(!(((data>>16)&0x03)==0x03)) { + reset=1; + printk_debug("Bad PCIX MHz - Reset\n"); + } + printk_debug("Looking for bad Hot Swap Enable\n"); + dev = dev_find_slot(1, PCI_DEVFN(0x01,0)); + data = pci_read_config32(dev, 0x48); + if(data & 0x0c) { + reset=1; + printk_debug("Bad Hot Swap start - Reset\n"); + } + if(reset) { + /* enable cf9 */ + dev = dev_find_slot(node_link_to_bus(0, 0), PCI_DEVFN(0x04,3)); + pci_write_config8(dev, 0x41, 0xf1); + /* reset */ + dev = dev_find_slot(0, PCI_DEVFN(0x18,0)); + htic = pci_read_config32(dev, HT_INIT_CONTROL); + htic &= ~HTIC_BIOSR_Detect; + pci_write_config32(dev, HT_INIT_CONTROL, htic); + outb(0x0e, 0x0cf9); + } + else { + printk_debug("OK 133MHz & Hot Swap is off\n"); + } +} + unsigned long write_smp_table(unsigned long addr) { void *v; + reboot_if_hotswap(); + v = smp_write_floating_table(addr); return (unsigned long)smp_write_config_table(v); }
D0 ? YH
On 9/2/05, jason schildt jschildt@lnxi.com wrote:
DESCRIPTION:
## lnxi-patch-10 ## src/northbridge/amd/amdk8/cpu_rev.c Testing for ranges of CPUs instead of exact matches.
DIFFSTAT:
cpu_rev.c | 33 ++++++--------------------------- 1 files changed, 6 insertions(+), 27 deletions(-)
PATCH:
Index: northbridge/amd/amdk8/cpu_rev.c
--- northbridge/amd/amdk8/cpu_rev.c (revision 1105) +++ northbridge/amd/amdk8/cpu_rev.c (working copy) @@ -3,44 +3,23 @@ { return (cpuid_eax(1) & 0xfffef) == 0x0f00; } -//AMD_D0_SUPPORT -static int is_cpu_pre_d0(void) -{
- return (cpuid_eax(1) & 0xfff0f) < 0x10f00;
-}
-static int is_cpu_d0(void) +static int is_cpu_pre_b3(void) {
- return (cpuid_eax(1) & 0xfff0f) == 0x10f00;
- return (cpuid_eax(1) & 0xfffef) < 0x0f41;
}
-//AMD_E0_SUPPORT -static int is_cpu_pre_e0(void) -{
- return (cpuid_eax(1) & 0xfff0f) < 0x20f00;
-}
-static int is_cpu_e0(void) -{
- return (cpuid_eax(1) & 0xfff00) == 0x20f00;
-}
static int is_cpu_pre_c0(void) { return (cpuid_eax(1) & 0xfffef) < 0x0f48; }
-static int is_cpu_c0(void) +static int is_cpu_pre_d0(void) {
- return (cpuid_eax(1) & 0xfffef) == 0x0f48;
- return (cpuid_eax(1) & 0xfff0f) < 0x10000;
}
-static int is_cpu_pre_b3(void) +static int is_cpu_pre_e0(void) {
- return (cpuid_eax(1) & 0xfffef) < 0x0f41;
- return (cpuid_eax(1) & 0xfff0f) < 0x20f00;
}
-static int is_cpu_b3(void) -{
- return (cpuid_eax(1) & 0xfffef) == 0x0f41;
-}
-- Jason W. Schildt LinuxBIOS Software Engineer Linux Networx
-- LinuxBIOS mailing list LinuxBIOS@openbios.org http://www.openbios.org/mailman/listinfo/linuxbios
yhlu yinghailu@gmail.com writes:
D0 ?
The thinking here is that cpu versions test should be done against ranges instead of exact matches. So only the is_cpu_pre_xxx versions remain in cpu_rev.c
d0 is likely the one rev where an exact match would work but it is a dangerous habit to get into.
Ranges can be much more easily extended into the future as we see new processor variants.
Eric
again about cpu_bus creation for all core1. I prefer to create enry to that in scan_cpu_bus in northbridge.c 1. you don't to reorder that in write_smp_processors 2. for 8 way or four way system, We need to set CPU_INIT_SERIAL=0, to init ecc for ap the same time, then with amd_sibling_init create entry for core1 in node nodex, could conflit access with cpu_initli....in node0/core0. esp for the last core1 could be missed to init, because before nodex/core0 add the core1 to the cpu bus, node0/core0 already finish the start cpu bus. YH
On 9/2/05, jason schildt jschildt@lnxi.com wrote:
DESCRIPTION:
## lnxi-patch-4 Big Dual Core Changes ## - (8)files
src/include/cpu/amd/dualcore.h Code Cleanup.
src/cpu/amd/model_fxx/model_fxx_init.c Code cleanup: config guards in header. dynamic memory hoisting instead of hard coding. cpu ranges instead of exact cpu matching. naming change (dev -> cpu)
src/cpu/amd/dualcore/dualcore.c Remove chunk with these function calls #if'd out: get_core_num_in_bsp() # Not used. set_apicid_cpuid_lo() # Used in Tyan Opteron boards. real_start_other_core() # Not used. start_other_core() # Not used. get_nodes() # Not used. start_other_cores() # Used in Tyan Opteron Boards.
src/cpu/amd/dualcore/amd_sibling.c Make amd_sibling_init work and generalize get_node_core_id.
src/cpu/amd/dualcore/dualcore_id.c Optimized struct node_core_id so that it fits in 1 register with romcc.
src/northbridge/amd/amdk8/coherent_ht.c Dualcore cleanup work: Added startup_othercores() only look at low nibble to see link freq.
src/northbridge/amd/amdk8/amdk8.h Added e0 register
src/northbridge/amd/amdk8/northbridge.c Added dynamic memory hoisting - hoist_memory() included 2.6.11/12 kernel bug fix Cleaned up cpu_bus_scan()
src/northbridge/amd/amdk8/raminit.c factored out set_dimm_map() from set_dimm_size() added e_step_cpu() This function holds the logic only needed for e-step. removed hardcoded memory hole - set_e0_mem_hole()
DIFFSTAT:
cpu/amd/dualcore/amd_sibling.c | 180 +++++++++++------------------- cpu/amd/dualcore/dualcore.c | 127 ++++++++------------- cpu/amd/dualcore/dualcore_id.c | 4 cpu/amd/model_fxx/model_fxx_init.c | 98 +++++----------- include/cpu/amd/dualcore.h | 9 - northbridge/amd/amdk8/amdk8.h | 1 northbridge/amd/amdk8/coherent_ht.c | 89 +++++++------- northbridge/amd/amdk8/northbridge.c | 215 ++++++++++++------------------------ northbridge/amd/amdk8/raminit.c | 154 ++++++++++++------------- 9 files changed, 346 insertions(+), 531 deletions(-)
PATCH:
Index: src/include/cpu/amd/dualcore.h
--- src/include/cpu/amd/dualcore.h (revision 1105) +++ src/include/cpu/amd/dualcore.h (working copy) @@ -2,18 +2,13 @@ #define CPU_AMD_DUALCORE_H
struct device; -void amd_sibling_init(struct device *cpu);
-int is_e0_later_in_bsp(int nodeid); -unsigned int read_nb_cfg_54(void);
struct node_core_id { unsigned nodeid; unsigned coreid; };
-// it can be used to get unitid and coreid it running only -struct node_core_id get_node_core_id(unsigned int nb_cfg_54); -unsigned get_apicid_base(unsigned ioapic_num); +void amd_sibling_init(struct device *cpu, struct node_core_id id); +struct node_core_id get_node_core_id(void);
#endif /* CPU_AMD_DUALCORE_H */ Index: src/cpu/amd/model_fxx/model_fxx_init.c =================================================================== --- src/cpu/amd/model_fxx/model_fxx_init.c (revision 1105) +++ src/cpu/amd/model_fxx/model_fxx_init.c (working copy) @@ -21,10 +21,7 @@ #include <cpu/x86/cache.h> #include <cpu/x86/mtrr.h> #include <cpu/x86/mem.h>
-#if CONFIG_LOGICAL_CPUS==1 #include <cpu/amd/dualcore.h> -#endif
#include "model_fxx_msr.h"
@@ -152,9 +149,6 @@ static void init_ecc_memory(unsigned node_id) { unsigned long startk, begink, endk; -#if K8_E0_MEM_HOLE_SIZEK != 0
- unsigned long hole_startk = 0, hole_endk = 0;
-#endif unsigned long basek; struct mtrr_state mtrr_state; device_t f1_dev, f2_dev, f3_dev; @@ -199,25 +193,13 @@ startk = (pci_read_config32(f1_dev, 0x40 + (node_id*8)) & 0xffff0000) >> 2; endk = ((pci_read_config32(f1_dev, 0x44 + (node_id*8)) & 0xffff0000)
- 0x4000;
-#if K8_E0_MEM_HOLE_SIZEK != 0
- if (!is_cpu_pre_e0()) {
- uint32_t val;
- val = pci_read_config32(f1_dev, 0xf0);
- if((val & 1)==1) {
- hole_startk = ((val & (0xff<<24)) >> 10);
- hole_endk = ((val & (0xff<<8))<<(16-10)) - startk;
- hole_endk += hole_startk;
- }
- }
-#endif
/* Don't start too early */ begink = startk; if (begink < CONFIG_LB_MEM_TOPK) { begink = CONFIG_LB_MEM_TOPK; }
- printk_debug("Clearing memory %uK - %uK: ", startk, endk);
- printk_debug("Clearing memory %uK - %uK: ", begink, endk);
/* Save the normal state */ save_mtrr_state(&mtrr_state); @@ -234,9 +216,6 @@ unsigned long size; void *addr;
-#if K8_E0_MEM_HOLE_SIZEK != 0
- if ((basek >= hole_startk) && (basek < hole_endk)) continue;
-#endif /* Report every 64M */ if ((basek % (64*1024)) == 0) { /* Restore the normal state */ @@ -340,6 +319,7 @@
/* Erratum 91 prefetch miss is handled in the kernel */
/* Erratum 106 ... */ msr = rdmsr_amd(LS_CFG_MSR); msr.lo |= 1 << 25; @@ -350,7 +330,7 @@ msr.hi |= 1 << (43 - 32); wrmsr_amd(BU_CFG_MSR, msr);
- if(is_cpu_d0()) {
- if (is_cpu_pre_e0() && !is_cpu_pre_d0()) {
/* Erratum 110 ...*/ msr = rdmsr_amd(CPU_ID_HYPER_EXT_FEATURES); msr.hi |=1; @@ -362,26 +342,34 @@ msr = rdmsr_amd(CPU_ID_EXT_FEATURES_MSR); msr.hi |=1; wrmsr_amd(CPU_ID_EXT_FEATURES_MSR, msr);
- /* Erratum 113 ... */
- msr = rdmsr_amd(BU_CFG_MSR);
- msr.hi |= (1 << 16);
- wrmsr_amd(BU_CFG_MSR, msr);
}
/* Erratum 122 */
- msr = rdmsr(HWCR_MSR);
- msr.lo |= 1 << 6;
- wrmsr(HWCR_MSR, msr);
if (!is_cpu_pre_c0()) {
msr = rdmsr(HWCR_MSR);
msr.lo |= 1 << 6;
wrmsr(HWCR_MSR, msr);
}
/* Erratum 123? dual core deadlock? */
/* Erratum 131 */
msr = rdmsr(NB_CFG_MSR);
msr.lo |= 1 << 20;
wrmsr(NB_CFG_MSR, msr);
}
-void model_fxx_init(device_t dev) +void model_fxx_init(device_t cpu) { unsigned long i; msr_t msr; -#if CONFIG_LOGICAL_CPUS struct node_core_id id;
- unsigned siblings;
- id.coreid=0;
-#else
- unsigned nodeid;
-#endif
/* Turn on caching if we haven't already */ x86_enable_cache(); @@ -404,43 +392,18 @@ /* Enable the local cpu apics */ setup_lapic();
-#if CONFIG_LOGICAL_CPUS == 1
- siblings = cpuid_ecx(0x80000008) & 0xff;
- /* Find our node and core */
- id = get_node_core_id();
- id = get_node_core_id(read_nb_cfg_54()); // pre e0 nb_cfg_54 can not be
set
- if(siblings>0) {
- msr = rdmsr_amd(CPU_ID_FEATURES_MSR);
- msr.lo |= 1 << 28;
- wrmsr_amd(CPU_ID_FEATURES_MSR, msr);
- msr = rdmsr_amd(LOGICAL_CPUS_NUM_MSR);
- msr.lo = (siblings+1)<<16;
- wrmsr_amd(LOGICAL_CPUS_NUM_MSR, msr);
- msr = rdmsr_amd(CPU_ID_EXT_FEATURES_MSR);
- msr.hi |= 1<<(33-32);
- wrmsr_amd(CPU_ID_EXT_FEATURES_MSR, msr);
- }
- /* Is this a bad location? In particular can another node prefecth
- /* Is this a bad location? In particular can another node prefetch
- data from this node before we have initialized it?
*/
- if (id.coreid == 0) init_ecc_memory(id.nodeid); // only do it for core 0
-#else
- /* Is this a bad location? In particular can another node prefecth
- data from this node before we have initialized it?
- */
- nodeid = lapicid() & 0xf;
- init_ecc_memory(nodeid);
-#endif
- if (id.coreid == 0) {
- init_ecc_memory(id.nodeid); // only do it for core 0
- }
-#if CONFIG_LOGICAL_CPUS==1
- /* Start up my cpu siblings */
-// if(id.coreid==0) amd_sibling_init(dev); // Don't need core1 is already be put in the CPU BUS in bus_cpu_scan -#endif
- /* Deal with sibling cpus */
- amd_sibling_init(cpu, id);
}
static struct device_operations cpu_dev_ops = { @@ -451,7 +414,7 @@ { X86_VENDOR_AMD, 0xf51 }, /* SH7-B3 */ { X86_VENDOR_AMD, 0xf58 }, /* SH7-C0 */ { X86_VENDOR_AMD, 0xf48 }, -#if 1
{ X86_VENDOR_AMD, 0xf5A }, /* SH7-CG */ { X86_VENDOR_AMD, 0xf4A }, { X86_VENDOR_AMD, 0xf7A }, @@ -483,7 +446,6 @@ { X86_VENDOR_AMD, 0x20fc2 }, { X86_VENDOR_AMD, 0x20f12 }, /* JH-E6 */ { X86_VENDOR_AMD, 0x20f32 }, -#endif
{ 0, 0 }, }; Index: src/cpu/amd/dualcore/dualcore.c =================================================================== --- src/cpu/amd/dualcore/dualcore.c (revision 1105) +++ src/cpu/amd/dualcore/dualcore.c (working copy) @@ -1,99 +1,68 @@ /* 2004.12 yhlu add dual core support */
-#ifndef SET_NB_CFG_54 -#define SET_NB_CFG_54 1 -#endif
#include "cpu/amd/dualcore/dualcore_id.c"
-static inline unsigned get_core_num_in_bsp(unsigned nodeid) +static void do_k8_init_and_stop_secondaries(void) {
- return ((pci_read_config32(PCI_DEV(0, 0x18+nodeid, 3), 0xe8)>>12)
& 3); -}
-static inline -#if SET_NB_CFG_54 == 1
- uint8_t
-#else
- void
-#endif
- set_apicid_cpuid_lo(void) {
-#if SET_NB_CFG_54
- //for pre_e0, even we set nb_cfg_54, but it will still be 0
- //for e0 later you should use get_node_id(read_nb_cfg_54()) even for
single core cpu
- //get siblings via cpuid(0x80000008) ecx[7:0]
- #if CONFIG_MAX_PHYSICAL_CPUS != 8
- if( get_core_num_in_bsp(0) == 0) {
- /*first node only has one core, pre_e0
- all e0 single core installed don't need enable lo too,
- So if mixing e0 single core and dual core,
- don't put single core in first socket */
- return 0;
- }
- #endif
- struct node_core_id id;
- device_t dev;
- unsigned apicid;
- unsigned max_siblings;
- msr_t msr;
- if(read_option(CMOS_VSTART_dual_core, CMOS_VLEN_dual_core, 0) != 0)
{ // disable dual_core
- return 0;
- }
- /* Skip this if there was a built in self test failure */
- // set the NB_CFG[54]=1; why the OS will be happy with that ???
- msr_t msr;
- msr = rdmsr(NB_CFG_MSR);
- msr.hi |= (1<<(54-32)); // InitApicIdCpuIdLo
- wrmsr(NB_CFG_MSR, msr);
- if (is_cpu_pre_e0()) {
- id.nodeid = lapicid() & 0x7;
- id.coreid = 0;
- } else {
- /* Which cpu are we on? */
- id = get_node_core_id_x();
- return 1;
- /* Set NB_CFG_MSR
- Linux expect the core to be in the least signficant bits.
- */
- msr = rdmsr(NB_CFG_MSR);
- msr.hi |= (1<<(54-32)); // InitApicIdCpuIdLo
- wrmsr(NB_CFG_MSR, msr);
- }
-#endif
- /* For now assume all cpus have the same number of siblings */
- max_siblings = (cpuid_ecx(0x80000008) & 0xff) + 1;
-}
- /* Set the lapicid */
- lapic_write(LAPIC_ID,((id.nodeid*max_siblings) + id.coreid) << 24);
-static inline void real_start_other_core(unsigned nodeid) -{
- uint32_t dword;
- // set PCI_DEV(0, 0x18+nodeid, 3), 0x44 bit 27 to redirect all MC4
accesses and error logging to core0
- dword = pci_read_config32(PCI_DEV(0, 0x18+nodeid, 3), 0x44);
- dword |= 1<<27; // NbMcaToMstCpuEn bit
- pci_write_config32(PCI_DEV(0, 0x18+nodeid, 3), 0x44, dword);
- // set PCI_DEV(0, 0x18+nodeid, 0), 0x68 bit 5 to start core1
- dword = pci_read_config32(PCI_DEV(0, 0x18+nodeid, 0), 0x68);
- dword |= 1<<5;
- pci_write_config32(PCI_DEV(0, 0x18+nodeid, 0), 0x68, dword);
- /* Remember the cpuid */
- if (id.coreid == 0) {
- dev = PCI_DEV(0, 0x18 + id.nodeid, 2);
- pci_write_config32(dev, 0x9c, cpuid_eax(1));
- }
- /* Maybe call distinguish_cpu_resets only on the last core? */
- distinguish_cpu_resets(id.nodeid);
- if (!boot_cpu()) {
- stop_this_cpu();
- }
}
-//it is running on core0 of every node -static inline void start_other_core(unsigned nodeid) {
- if(read_option(CMOS_VSTART_dual_core, CMOS_VLEN_dual_core, 0) !=
- { // disable dual_core
- return;
- }
- if( get_core_num() >0) { // defined in dualcore_id.c
- real_start_other_core(nodeid);
- }
-}
-static inline unsigned get_nodes(void) +static void k8_init_and_stop_secondaries(void) {
- return ((pci_read_config32(PCI_DEV(0, 0x18, 0), 0x60)>>4) & 7) + 1;
-}
- /* This doesn't work with Cache As Ram because it messes with
- the MTRR state, which breaks the init detection.
- do_k8_init_and_stop_secondaries should be usable by CAR code.
- */
-//it is running on core0 of node0 -static inline void start_other_cores(void) {
- unsigned nodes;
- unsigned nodeid;
- int init_detected;
- if(read_option(CMOS_VSTART_dual_core, CMOS_VLEN_dual_core, 0) !=
- { // disable dual_core
- return;
- }
- init_detected = early_mtrr_init_detected();
- amd_early_mtrr_init();
- nodes = get_nodes();
- for(nodeid=0; nodeid<nodes; nodeid++) {
- if( get_core_num_in_bsp(nodeid) > 0) {
- real_start_other_core(nodeid);
- }
- enable_lapic();
- init_timer();
- if (init_detected) {
- asm volatile ("jmp __cpu_reset");
}
- do_k8_init_and_stop_secondaries();
} Index: src/cpu/amd/dualcore/amd_sibling.c =================================================================== --- src/cpu/amd/dualcore/amd_sibling.c (revision 1105) +++ src/cpu/amd/dualcore/amd_sibling.c (working copy) @@ -1,4 +1,5 @@ /* 2004.12 yhlu add dual core support */ +/* 24 June 2005 Cleaned up dual core support Eric Biederman */
#include <console/console.h> #include <cpu/cpu.h> @@ -14,59 +15,87 @@
static int first_time = 1; static int disable_siblings = !CONFIG_LOGICAL_CPUS; +void amd_sibling_init(device_t cpu, struct node_core_id id) +{
unsigned long i;
unsigned siblings, max_siblings;
/* On the bootstrap processor see if I want sibling cpus enabled */
if (first_time) {
first_time = 0;
get_option(&disable_siblings, "dual_core");
}
-int is_e0_later_in_bsp(int nodeid) -{
- uint32_t val;
- uint32_t val_old;
- int e0_later;
- if(nodeid==0) { // we don't need to do that for node 0 in core0/node0
- return !is_cpu_pre_e0();
- siblings = cpuid_ecx(0x80000008) & 0xff;
- printk_debug("%d Sibling Cores found\n", siblings);
- /* For now assume all cpus have the same number of siblings */
- max_siblings = siblings + 1;
- /* Wishlist? make dual cores look like hyperthreading */
- /* See if I am a sibling cpu */
- if (disable_siblings && (id.coreid != 0)) {
- cpu->enabled = 0;
}
- // d0 will be treated as e0 with this methods, but the d0 nb_cfg_54
always 0
- device_t dev;
- dev = dev_find_slot(0, PCI_DEVFN(0x18+nodeid,2));
- if(!dev) return 0;
- val_old = pci_read_config32(dev, 0x80);
- val = val_old;
- val |= (1<<3);
- pci_write_config32(dev, 0x80, val);
- val = pci_read_config32(dev, 0x80);
- e0_later = !!(val & (1<<3));
- if(e0_later) { // pre_e0 bit 3 always be 0 and can not be changed
- pci_write_config32(dev, 0x80, val_old); // restore it
- if (id.coreid == 0) {
- /* On the primary cpu find the siblings */
- for (i = 1; i <= siblings; i++) {
- struct device_path cpu_path;
- device_t new;
- /* Build the cpu device path */
- cpu_path.type = DEVICE_PATH_APIC;
- cpu_path.u.apic.apic_id =
- (id.nodeid*max_siblings) + i;
- new = alloc_dev(cpu->bus, &cpu_path);
- if (!new) {
- continue;
- }
- /* Report what I have done */
- printk_debug("CPU: %s %s\n",
- dev_path(new), new->enabled?"enabled":"disabled");
- }
}
- return e0_later;
}
-unsigned int read_nb_cfg_54(void) +struct node_core_id get_node_core_id(void) {
- msr_t msr;
- msr = rdmsr(NB_CFG_MSR);
- return ( ( msr.hi >> (54-32)) & 1);
-}
-struct node_core_id get_node_core_id(unsigned int nb_cfg_54) { struct node_core_id id;
- // get the apicid via cpuid(1) ebx[27:24]
- if(nb_cfg_54) {
- // when NB_CFG[54] is set, nodid = ebx[27:25], coreid =
ebx[24]
- id.coreid = (cpuid_ebx(1) >> 24) & 0xf;
- id.nodeid = (id.coreid>>1);
- id.coreid &= 1;
- } else { // single core should be here too
- unsigned siblings;
- /* Get the apicid at reset */
- id.nodeid = (cpuid_ebx(1) >> 24) & 0xff;
- id.coreid = 0;
- /* Find out how many siblings we have */
- siblings = cpuid_ecx(0x80000008) & 0xff;
- if (siblings) {
- unsigned bits;
- msr_t msr;
- bits = 0;
- while ((1 << bits) <= siblings)
- bits++;
- msr = rdmsr(NB_CFG_MSR);
- if ((msr.hi >> (54-32)) & 1) {
- // when NB_CFG[54] is set, nodeid = ebx[27:25], coreid = ebx[24]
- id.coreid = id.nodeid & ((1 << bits) - 1);
- id.nodeid >>= bits;
- } else {
// when NB_CFG[54] is clear, nodeid = ebx[26:24], coreid = ebx[27]
- id.nodeid = (cpuid_ebx(1) >> 24) & 0xf;
- id.coreid = (id.nodeid>>3);
- id.nodeid &= 7;
- id.coreid = id.nodeid >> 3;
- id.nodeid &= 7;
- }
- } else {
- if (!is_cpu_pre_e0()) {
- id.nodeid >>= 1;
- }
}
- return id;
- return id;
+}
-}
+#if 0 static int get_max_siblings(int nodes) { device_t dev; @@ -169,76 +198,5 @@
return apicid_base; } -#if 0 -void amd_sibling_init(device_t cpu) -{
unsigned i, siblings;
struct cpuid_result result;
unsigned nb_cfg_54;
struct node_core_id id;
/* On the bootstrap processor see if I want sibling cpus enabled */
if (first_time) {
first_time = 0;
get_option(&disable_siblings, "dual_core");
}
result = cpuid(0x80000008);
/* See how many sibling cpus we have */
/* Is dualcore supported */
siblings = (result.ecx & 0xff);
if ( siblings < 1) {
return;
}
-#if 1
- printk_debug("CPU: %u %d siblings\n",
- cpu->path.u.apic.apic_id,
- siblings);
#endif
- nb_cfg_54 = read_nb_cfg_54();
-#if 1
- id = get_node_core_id(nb_cfg_54); // pre e0 nb_cfg_54 can not be set
- /* See if I am a sibling cpu */
- //if ((cpu->path.u.apic.apic_id>>(nb_cfg_54?0:3)) & siblings ) { //
siblings = 1, 3, 7, 15,....
- //if ( ( (cpu->path.u.apic.apic_id>>(nb_cfg_54?0:3)) % (siblings+1) ) !=
0 ) {
- if(id.coreid != 0) {
- if (disable_siblings) {
- cpu->enabled = 0;
- }
- return;
- }
-#endif
- /* I am the primary cpu start up my siblings */
- for(i = 1; i <= siblings; i++) {
- struct device_path cpu_path;
- device_t new;
- /* Build the cpu device path */
- cpu_path.type = DEVICE_PATH_APIC;
- cpu_path.u.apic.apic_id = cpu->path.u.apic.apic_id + i *
(nb_cfg_54?1:8);
- /* See if I can find the cpu */
- new = find_dev_path(cpu->bus, &cpu_path);
- /* Allocate the new cpu device structure */
- if(!new) {
- new = alloc_dev(cpu->bus, &cpu_path);
- new->enabled = 1;
- new->initialized = 0;
- }
-#if 1
- printk_debug("CPU: %u has sibling %u\n",
- cpu->path.u.apic.apic_id,
- new->path.u.apic.apic_id);
-#endif
- /* Start the new cpu */
- if(new->enabled && !new->initialized)
- start_cpu(new);
- }
-} -#endif
Index: src/cpu/amd/dualcore/dualcore_id.c
--- src/cpu/amd/dualcore/dualcore_id.c (revision 1105) +++ src/cpu/amd/dualcore/dualcore_id.c (working copy) @@ -11,8 +11,8 @@ }
struct node_core_id {
- unsigned nodeid;
- unsigned coreid;
- unsigned nodeid:8;
- unsigned coreid:8;
};
static inline struct node_core_id get_node_core_id(unsigned nb_cfg_54) { Index: src/northbridge/amd/amdk8/coherent_ht.c =================================================================== --- src/northbridge/amd/amdk8/coherent_ht.c (revision 1105) +++ src/northbridge/amd/amdk8/coherent_ht.c (working copy) @@ -155,23 +155,6 @@
}
-#ifndef ENABLE_APIC_EXT_ID -#define ENABLE_APIC_EXT_ID 0 -#endif
-static void enable_apic_ext_id(u8 node) -{ -#if ENABLE_APIC_EXT_ID==1 -#warning "FIXME Is the right place to enable apic ext id here?"
- u32 val;
- val = pci_read_config32(NODE_HT(node), 0x68);
- val |= (HTTC_APIC_EXT_SPUR | HTTC_APIC_EXT_ID |
HTTC_APIC_EXT_BRD_CST);
- pci_write_config32(NODE_HT(node), 0x68, val);
-#endif -}
static void enable_routing(u8 node) { u32 val; @@ -292,20 +275,18 @@ return 1; }
-static uint16_t read_freq_cap(device_t dev, uint8_t pos) +static unsigned read_freq_cap(device_t dev, unsigned pos) { /* Handle bugs in valid hypertransport frequency reporting */
- uint16_t freq_cap;
- unsigned freq_cap;
uint32_t id;
freq_cap = pci_read_config16(dev, pos); freq_cap &= ~(1 << HT_FREQ_VENDOR); /* Ignore Vendor HT frequencies */
-#if K8_HT_FREQ_1G_SUPPORT == 1 if (!is_cpu_pre_e0()) { return freq_cap; } -#endif
id = pci_read_config32(dev, 0);
@@ -339,8 +320,10 @@
/* See if I am changing the link freqency */ old_freq = pci_read_config8(node1, link1 + PCI_HT_CAP_HOST_FREQ);
- old_freq &= 0x0f;
needs_reset |= old_freq != freq; old_freq = pci_read_config8(node2, link2 + PCI_HT_CAP_HOST_FREQ);
- old_freq &= 0x0f;
needs_reset |= old_freq != freq;
/* Set the Calulcated link frequency */ @@ -382,7 +365,6 @@
/* Set node2's widths */ pci_write_config8(node2, link2 + PCI_HT_CAP_HOST_WIDTH + 1, width);
return needs_reset; }
@@ -1625,9 +1607,9 @@ } #endif /* CONFIG_MAX_PHYSICAL_CPUS > 1 */
+static unsigned count_cpus(unsigned nodes) +{ #if CONFIG_LOGICAL_CPUS==1 -static unsigned verify_dualcore(unsigned nodes) -{ unsigned node, totalcpus, tmp;
totalcpus = 0; @@ -1637,25 +1619,21 @@ }
return totalcpus; +#else
- return nodes;
+#endif
} -#endif
static void coherent_ht_finalize(unsigned nodes) {
- unsigned total_cpus;
- unsigned cpu_node_count;
unsigned node; int rev_a0; -#if CONFIG_LOGICAL_CPUS==1
- unsigned total_cpus;
- total_cpus = count_cpus(nodes);
- cpu_node_count = ((total_cpus -1)<<16)|((nodes - 1) << 4);
- if(read_option(CMOS_VSTART_dual_core, CMOS_VLEN_dual_core, 0) == 0) { /*
dual_core */
- total_cpus = verify_dualcore(nodes);
- }
- else {
- total_cpus = nodes;
- }
-#endif
/* set up cpu count and node count and enable Limit
- Config Space Range for all available CPUs.
- Also clear non coherent hypertransport bus range
@@ -1672,11 +1650,7 @@ /* Set the Total CPU and Node count in the system */ val = pci_read_config32(dev, 0x60); val &= (~0x000F0070); -#if CONFIG_LOGICAL_CPUS==1
- val |= ((total_cpus-1)<<16)|((nodes-1)<<4);
-#else
- val |= ((nodes-1)<<16)|((nodes-1)<<4);
-#endif
- val |= cpu_node_count;
pci_write_config32(dev, 0x60, val);
/* Only respond to real cpu pci configuration cycles @@ -1786,6 +1760,33 @@ return needs_reset; }
+static void startup_other_cores(unsigned nodes) +{
- unsigned node;
- for(node = 0; node < nodes; node++) {
- device_t dev;
- unsigned siblings;
- dev = NODE_MC(node);
- siblings = (pci_read_config32(dev, 0xe8) >> 12) & 0x3;
- if (siblings) {
- device_t dev_f0;
- unsigned val;
- /* Redirect all MC4 accesses and error logging to core0 */
- val = pci_read_config32(dev, 0x44);
- val |= (1 << 27); //NbMcaToMstCpuEn bit
- pci_write_config32(dev, 0x44, val);
- dev_f0 = NODE_HT(node);
- /* Enable extended apic id's and second core */
- val = pci_read_config32(dev_f0, 0x68);
- val |= (1 << 18) | (1 << 17) | ( 1 << 5);
- pci_write_config32(dev_f0, 0x68, val);
- }
- }
+}
static int setup_coherent_ht_domain(void) { struct setup_smp_result result; @@ -1799,15 +1800,15 @@ enable_bsp_routing();
#if CONFIG_MAX_PHYSICAL_CPUS > 1
- result = setup_smp();
- result.nodes = verify_mp_capabilities(result.nodes);
- clear_dead_routes(result.nodes);
- result = setup_smp();
#endif
- result.nodes = verify_mp_capabilities(result.nodes);
- clear_dead_routes(result.nodes);
if (result.nodes == 1) { setup_uniprocessor(); } coherent_ht_finalize(result.nodes);
- startup_other_cores(result.nodes);
result.needs_reset = apply_cpu_errata_fixes(result.nodes, result.needs_reset); result.needs_reset = optimize_link_read_pointers(result.nodes, result.needs_reset); return result.needs_reset; Index: src/northbridge/amd/amdk8/amdk8.h =================================================================== --- src/northbridge/amd/amdk8/amdk8.h (revision 1105) +++ src/northbridge/amd/amdk8/amdk8.h (working copy) @@ -136,6 +136,7 @@ #define DCL_DisInRcvrs (1<<24) #define DCL_BypMax_SHIFT 25 #define DCL_En2T (1<<28) +#define DCL_UpperCSMap (1<<29) #define DRAM_CONFIG_HIGH 0x94 #define DCH_ASYNC_LAT_SHIFT 0 #define DCH_ASYNC_LAT_MASK 0xf Index: src/northbridge/amd/amdk8/northbridge.c =================================================================== --- src/northbridge/amd/amdk8/northbridge.c (revision 1105) +++ src/northbridge/amd/amdk8/northbridge.c (working copy) @@ -17,9 +17,9 @@ #include <cpu/cpu.h>
#include <cpu/x86/lapic.h> +#include <cpu/amd/dualcore.h>
#if CONFIG_LOGICAL_CPUS==1 -#include <cpu/amd/dualcore.h> #include <pc80/mc146818rtc.h> #endif
@@ -27,11 +27,8 @@ #include "root_complex/chip.h" #include "northbridge.h" #include "amdk8.h" +#include "cpu_rev.c"
-#if K8_E0_MEM_HOLE_SIZEK != 0 -#include "./cpu_rev.c" -#endif
#define FX_DEVS 8 static device_t __f0_dev[FX_DEVS]; static device_t __f1_dev[FX_DEVS]; @@ -640,6 +637,41 @@ return tolm; }
+static uint32_t hoist_memory(unsigned long mmio_basek, int i) +{
- int ii;
- uint32_t carry_over;
- device_t dev;
- uint32_t base, limit;
- uint32_t basek;
- uint32_t hoist;
- carry_over = (4*1024*1024) - mmio_basek;
- for(ii=7;ii>i;ii--) {
- base = f1_read_config32(0x40 + (ii << 3));
- limit = f1_read_config32(0x44 + (ii << 3));
- if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) {
- continue;
- }
- f1_write_config32(0x44 + (ii << 3),limit + (carry_over << 2));
- f1_write_config32(0x40 + (ii << 3),base + (carry_over << 2));
- }
- limit = f1_read_config32(0x44 + (i << 3));
- f1_write_config32(0x44 + (i << 3),limit + (carry_over << 2));
- dev = __f1_dev[i];
- base = pci_read_config32(dev, 0x40 + (i << 3));
- basek = (pci_read_config32(dev, 0x40 + (i << 3)) & 0xffff0000) >> 2;
- hoist = /* hole start address */
- ((mmio_basek << 10) & 0xff000000) +
- /* hole address to memory controller address */
- (((basek + carry_over) >> 6) & 0x0000ff00) +
- /* enable */
- 1;
- pci_write_config32(dev, 0xf0, hoist);
- return carry_over;
+}
static void pci_domain_set_resources(device_t dev) { unsigned long mmio_basek; @@ -648,41 +680,23 @@
pci_tolm = find_pci_tolm(&dev->link[0]);
- /* Work around for NUMA bug in all kernels before 2.6.13.
- If pci memory hole is too small, the kernel memory to NUMA
- node mapping will fail to initialize and system will run in
- non-NUMA mode.
- */
- if(pci_tolm > 0xf8000000) pci_tolm = 0xf8000000;
#warning "FIXME handle interleaved nodes" mmio_basek = pci_tolm >> 10; /* Round mmio_basek to something the processor can support */ mmio_basek &= ~((1 << 6) -1);
-#if 1 -#warning "FIXME improve mtrr.c so we don't use up all of the mtrrs with a 64M MMIO hole"
- /* Round the mmio hold to 64M */
- mmio_basek &= ~((64*1024) - 1);
-#endif
-#if K8_E0_MEM_HOLE_SIZEK != 0
- if (!is_cpu_pre_e0())
- for (i = 0; i < 8; i++) {
- uint32_t base;
- base = f1_read_config32(0x40 + (i << 3));
- if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) {
- continue;
- }
- base = pci_read_config32(__f1_dev[i], 0xf0);
- if((base & 1)==0) continue;
- base &= 0xff<<24;
- base >>= 10;
- if (mmio_basek > base) {
- mmio_basek = base;
- }
- break; // only one hole
- }
-#endif
idx = 10; for(i = 0; i < 8; i++) { uint32_t base, limit; unsigned basek, limitk, sizek;
base = f1_read_config32(0x40 + (i << 3)); limit = f1_read_config32(0x44 + (i << 3)); if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) { @@ -708,6 +722,9 @@ pre_sizek = mmio_basek - basek; ram_resource(dev, idx++, basek, pre_sizek); sizek -= pre_sizek;
- if(! is_cpu_pre_e0() ) {
- sizek += hoist_memory(mmio_basek,i);
- }
basek = mmio_basek; } if ((basek + sizek) <= 4*1024*1024) { @@ -767,55 +784,21 @@ .ops_pci_bus = &pci_cf8_conf1, };
-#define APIC_ID_OFFSET 0x10
static unsigned int cpu_bus_scan(device_t dev, unsigned int max) { struct bus *cpu_bus; device_t dev_mc;
- int bsp_apic_id;
- int apic_id_offset;
- unsigned max_siblings;
int i,j;
- unsigned nb_cfg_54;
- int enable_apic_ext_id;
- unsigned siblings;
-#if CONFIG_LOGICAL_CPUS == 1
- int e0_later_single_core;
- int disable_siblings;
-#endif
- nb_cfg_54 = 0;
- enable_apic_ext_id = 0;
- siblings = 0;
- /* Find the bootstrap processors apicid */
- bsp_apic_id = lapicid();
- /* See if I will enable extended ids' */
- apic_id_offset = bsp_apic_id;
-#if CONFIG_LOGICAL_CPUS == 1
- disable_siblings = !CONFIG_LOGICAL_CPUS;
- get_option(&disable_siblings, "dual_core");
- // for pre_e0, nb_cfg_54 can not be set, ( even set, when you read it
still be 0)
- // How can I get the nb_cfg_54 of every node' nb_cfg_54 in bsp??? and
differ d0 and e0 single core
- nb_cfg_54 = read_nb_cfg_54();
-#endif dev_mc = dev_find_slot(0, PCI_DEVFN(0x18, 0)); if (!dev_mc) { die("0:18.0 not found?"); }
- if (pci_read_config32(dev_mc, 0x68) &
(HTTC_APIC_EXT_ID|HTTC_APIC_EXT_BRD_CST))
- {
- enable_apic_ext_id = 1;
- if (apic_id_offset == 0) {
- /* bsp apic id is not changed */
- apic_id_offset = APIC_ID_OFFSET;
- }
- }
- /* For now assume all cpus have the same number of siblings */
- max_siblings = (cpuid_ecx(0x80000008) & 0xff) + 1;
/* Find which cpus are present */ cpu_bus = &dev->link[0]; for(i = 0; i < 8; i++) { @@ -834,82 +817,36 @@ PCI_DEVFN(0x18 + i, j)); } }
- /* Build the cpu device path */
- cpu_path.type = DEVICE_PATH_APIC;
- cpu_path.u.apic.apic_id = i*max_siblings;
-#if CONFIG_LOGICAL_CPUS == 1
- e0_later_single_core = 0;
- if ((!disable_siblings) && dev && dev->enabled) {
- j = (pci_read_config32(dev, 0xe8) >> 12) & 3; // dev is func 3
- printk_debug(" %s siblings=%d\r\n", dev_path(dev), j);
- /* See if I can find the cpu */
- cpu = find_dev_path(cpu_bus, &cpu_path);
- if(nb_cfg_54) {
- // For e0 single core if nb_cfg_54 is set, apicid will be 0, 2, 4....
- // ----> you can mixed single core e0 and dual core e0 at any
sequence
- // That is the typical case
- if(j == 0 ){
- e0_later_single_core = is_e0_later_in_bsp(i);
// single core
- } else {
- e0_later_single_core = 0;
- }
- if(e0_later_single_core) {
- printk_debug("\tFound e0 single core\r\n");
- j=1;
- }
- if(siblings > j ) {
- //actually we can't be here, because d0 nb_cfg_54 can not be set
- //even worse is_e0_later_in_bsp() can not find out if it is d0 or e0
- die("When NB_CFG_54 is set, if you want to mix e0 (single core and
dual core) and single core(pre e0) CPUs, you need to put all the single core (pre e0) CPUs before all the (e0 single or dual core) CPUs\r\n");
- }
- else {
- siblings = j;
- }
- } else {
- siblings = j;
- }
- }
-#endif -#if CONFIG_LOGICAL_CPUS==1
- for (j = 0; j <= (e0_later_single_core?0:siblings); j++ )
{ -#else
- for (j = 0; j <= siblings; j++ ) {
-#endif
- /* Build the cpu device path */
- cpu_path.type = DEVICE_PATH_APIC;
- cpu_path.u.apic.apic_id = i * (nb_cfg_54?(siblings+1):1) + j *
(nb_cfg_54?1:8);
- /* See if I can find the cpu */
- cpu = find_dev_path(cpu_bus, &cpu_path);
- /* Enable the cpu if I have the processor */
- if (dev && dev->enabled) {
- if (!cpu) {
- cpu = alloc_dev(cpu_bus, &cpu_path);
- }
- if (cpu) {
- cpu->enabled = 1;
- }
- /* Enable the cpu if I have the processor */
- if (dev && dev->enabled) {
- if (!cpu) {
- cpu = alloc_dev(cpu_bus, &cpu_path);
}
- /* Disable the cpu if I don't have the processor */
- if (cpu && (!dev || !dev->enabled)) {
- cpu->enabled = 0;
- }
- /* Report what I have done */
if (cpu) {
- if(enable_apic_ext_id) {
if(cpu->path.u.apic.apic_id<apic_id_offset) { //all add offset except bsp core0
if( (cpu->path.u.apic.apic_id > siblings) || (bsp_apic_id!=0) )
cpu->path.u.apic.apic_id += apic_id_offset;
- }
- }
- printk_debug("CPU: %s %s\n",
- dev_path(cpu), cpu->enabled?"enabled":"disabled");
- cpu->enabled = 1;
}
- } //j
- }
- /* Disable the cpu if I don't have the processor */
- if (cpu && (!dev || !dev->enabled)) {
- cpu->enabled = 0;
- }
- /* Report what I have done */
- if (cpu) {
- printk_debug("CPU: %s %s\n",
- dev_path(cpu), cpu->enabled?"enabled":"disabled");
- }
}
return max; }
Index: src/northbridge/amd/amdk8/raminit.c
--- src/northbridge/amd/amdk8/raminit.c (revision 1105) +++ src/northbridge/amd/amdk8/raminit.c (working copy) @@ -585,6 +585,16 @@
}
+static void e_step_cpu(const struct mem_controller *ctrl) +{
- uint32_t dcl,data32;
- /* set bit 29 (upper cs map) of function 2 offset 0x90 */
- dcl = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW);
- dcl |= DCL_UpperCSMap;
- pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dcl);
+}
static int is_dual_channel(const struct mem_controller *ctrl) { uint32_t dcl; @@ -714,28 +724,14 @@ return sz; }
-static const unsigned cs_map_aa[15] = {
- /* (row=12, col=8)(14, 12) ---> (0, 0) (2, 4) */
- 0, 1, 3, 6, 0,
- 0, 2, 4, 7, 9,
- 0, 0, 5, 8,10,
-};
static void set_dimm_size(const struct mem_controller *ctrl, struct dimm_size sz, unsigned index) {
- uint32_t base0, base1, map;
- uint32_t base0, base1;
uint32_t dch;
if (sz.side1 != sz.side2) { sz.side2 = 0; }
- map = pci_read_config32(ctrl->f2, DRAM_BANK_ADDR_MAP);
- map &= ~(0xf << (index * 4));
-#if K8_4RANK_DIMM_SUPPORT == 1
- if(sz.rank == 4) {
- map &= ~(0xf << ( (index + 2) * 4));
- }
-#endif
/* For each base register.
- Place the dimm size in 32 MB quantities in the bits 31 - 21.
@@ -747,22 +743,6 @@
/* Make certain side1 of the dimm is at least 32MB */ if (sz.side1 >= (25 +3)) {
- if(is_cpu_pre_d0()) {
- map |= (sz.side1 - (25 + 3)) << (index *4);
-#if K8_4RANK_DIMM_SUPPORT == 1
- if(sz.rank == 4) {
- map |= (sz.side1 - (25 + 3)) << ( (index + 2) *
4);
- }
-#endif
- }
- else {
- map |= cs_map_aa[(sz.rows - 12) * 5 + (sz.col - 8) ] << (index*4);
-#if K8_4RANK_DIMM_SUPPORT == 1
- if(sz.rank == 4) {
- map |= cs_map_aa[(sz.rows - 12) * 5 + (sz.col -
- ] << ( (index + 2) * 4);
- }
-#endif
- }
base0 = (1 << ((sz.side1 - (25 + 3)) + 21)) | 1; }
@@ -791,8 +771,6 @@ } #endif
- pci_write_config32(ctrl->f2, DRAM_BANK_ADDR_MAP, map);
/* Enable the memory clocks for this DIMM */ if (base0) { dch = pci_read_config32(ctrl->f2, DRAM_CONFIG_HIGH); @@ -806,6 +784,52 @@ } }
+static void set_dimm_map(const struct mem_controller *ctrl,
- struct dimm_size sz, unsigned index)
+{
- static const unsigned cs_map_aa[15] = {
- /* (row=12, col=8)(14, 12) ---> (0, 0) (2, 4) */
- 0, 1, 3, 6, 0,
- 0, 2, 4, 7, 9,
- 0, 0, 5, 8,10,
- };
- uint32_t map;
- int row,col;
- map = pci_read_config32(ctrl->f2, DRAM_BANK_ADDR_MAP);
- map &= ~(0xf << (index * 4));
+#if K8_4RANK_DIMM_SUPPORT == 1
- if(sz.rank == 4) {
- map &= ~(0xf << ( (index + 2) * 4));
- }
+#endif
- if (is_cpu_pre_d0()) {
- map |= (sz.side1 - (25 + 3)) << (index *4);
+#if K8_4RANK_DIMM_SUPPORT == 1
- if(sz.rank == 4) {
- map |= (sz.side1 - (25 + 3)) << ( (index + 2) *
4);
- }
+#endif
- } else {
- unsigned val;
- val = cs_map_aa[(sz.rows - 12) * 5 + (sz.col - 8) ];
- if(val == 0) {
- print_err("Invalid Column or Row count\r\n");
- val = 7;
- }
- map |= val << (index*4);
+#if K8_4RANK_DIMM_SUPPORT == 1
- if(sz.rank == 4) {
- map |= val << ( (index + 2) * 4);
- }
+#endif
- }
- pci_write_config32(ctrl->f2, DRAM_BANK_ADDR_MAP, map);
+}
static long spd_set_ram_size(const struct mem_controller *ctrl, long dimm_mask) { int i; @@ -820,6 +844,7 @@ return -1; /* Report SPD error */ } set_dimm_size(ctrl, sz, i);
- set_dimm_map(ctrl, sz, i);
} return dimm_mask; } @@ -865,6 +890,13 @@ print_spew_hex32(tom_k); print_spew(" KB\r\n");
+#if 0
- /* Report the amount of memory. */
- print_debug("RAM: 0x");
- print_debug_hex32(tom_k);
- print_debug(" KB\r\n");
+#endif
/* Now set top of memory */ msr_t msr; msr.lo = (tom_k & 0x003fffff) << 10; @@ -971,7 +1003,7 @@ if(is_dual_channel(ctrl)) { /* Also we run out of address mask bits if we try and interleave 8 4GB dimms */ if ((bits == 3) && (common_size == (1 << (32 - 3)))) { -// print_debug("8 4GB chip selects cannot be interleaved\r\n");
- print_spew("8 4GB chip selects cannot be
interleaved\r\n"); return 0; } csbase_inc <<=1; @@ -981,7 +1013,7 @@ csbase_inc = csbase_low_d0[common_cs_mode]; if(is_dual_channel(ctrl)) { if( (bits==3) && (common_cs_mode > 8)) { -// print_debug("8 cs_mode>8 chip selects cannot be interleaved\r\n");
- print_spew("8 cs_mode>8 chip selects cannot be
interleaved\r\n"); return 0; } csbase_inc <<=1; @@ -1100,25 +1132,6 @@ return end_k; }
-#if K8_E0_MEM_HOLE_SIZEK != 0 -#define K8_E0_MEM_HOLE_LIMITK 4*1024*1024 -#define K8_E0_MEM_HOLE_BASEK (K8_E0_MEM_HOLE_LIMITK - K8_E0_MEM_HOLE_SIZEK )
-static void set_e0_mem_hole(const struct mem_controller *ctrl, unsigned base_k) -{
- /* Route the addresses to the controller node */
- unsigned val;
- val = pci_read_config32(ctrl->f1,0xf0);
- val &= 0x00ff00fe;
- val = (K8_E0_MEM_HOLE_BASEK << 10) |
((K8_E0_MEM_HOLE_SIZEK+base_k)>>(16-10)) | 1;
- pci_write_config32(ctrl->f1, 0xf0, val);
-}
-#endif
static void order_dimms(const struct mem_controller *ctrl) { unsigned long tom_k, base_k; @@ -1135,14 +1148,6 @@ /* Compute the memory base address */ base_k = memory_end_k(ctrl, ctrl->node_id); tom_k += base_k; -#if K8_E0_MEM_HOLE_SIZEK != 0
- if(!is_cpu_pre_e0()) {
- /* See if I need to check the range cover hole */
- if ((base_k <= K8_E0_MEM_HOLE_BASEK) && (tom_k >
K8_E0_MEM_HOLE_BASEK)) {
- tom_k += K8_E0_MEM_HOLE_SIZEK;
- }
- }
-#endif route_dram_accesses(ctrl, base_k, tom_k); set_top_mem(tom_k); } @@ -2145,12 +2150,11 @@ struct spd_set_memclk_result result; const struct mem_param *param; long dimm_mask; -#if 1
if (!controller_present(ctrl)) { -// print_debug("No memory controller present\r\n");
- print_debug("No memory controller present\r\n");
return; } -#endif hw_enable_ecc(ctrl); activate_spd_rom(ctrl); dimm_mask = spd_detect_dimms(ctrl); @@ -2176,6 +2180,10 @@ if (dimm_mask < 0) goto hw_spd_err; order_dimms(ctrl);
- if( !is_cpu_pre_e0() ) {
- print_debug("E step CPU\r\n");
- e_step_cpu(ctrl);
- }
return; hw_spd_err: /* Unrecoverable error reading SPD data */ @@ -2280,22 +2288,6 @@ } while(((dcl & DCL_MemClrStatus) == 0) || ((dcl & DCL_DramEnable) == 0) ); }
- // init e0 mem hole here
-#if K8_E0_MEM_HOLE_SIZEK != 0
- if (!is_cpu_pre_e0()) {
- uint32_t base, limit;
- unsigned base_k, limit_k;
- base = pci_read_config32(ctrl->f1, 0x40 + (i <<
3));
- limit = pci_read_config32(ctrl->f1, 0x44 + (i <<
3));
- base_k = (base & 0xffff0000) >> 2;
- limit_k = ((limit + 0x00010000) & 0xffff0000) >>
2;
- if ((base_k <= K8_E0_MEM_HOLE_BASEK) && (limit_k
K8_E0_MEM_HOLE_BASEK)) {
- set_e0_mem_hole(ctrl+i, base_k);
- }
- }
-#endif
print_debug(" done\r\n"); }
-- Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
-- LinuxBIOS mailing list LinuxBIOS@openbios.org http://www.openbios.org/mailman/listinfo/linuxbios
* jason schildt jschildt@lnxi.com [050903 00:03]:
+/* We can reduce the size of code generated by romcc by
- changing all of the fixed size types that live in registers
- into simple unsigned variables. (ie s/uint8_t/unsigned/g)
- */
Why is this? I would consider specifying an 8bit type to be more space-safing than using some generic untyped integer value. If not this should be fixed in romcc..
/* AMD K8 Unsupported 1Ghz? */ if (id == (PCI_VENDOR_ID_AMD | (0x1100 << 16))) {
if (is_cpu_pre_e0()) // CK804 support 1G?
device_t dev_2 = PCI_DEV(0,0x18,2);if(pci_read_config32(dev_2,0x9c) < 0x20f00) {
The function call looks a lot more readable here. How much is the gain of manual inlining here?
Stefan
On Tue, Sep 06, 2005 at 12:08:50PM +0200, Stefan Reinauer wrote:
- jason schildt jschildt@lnxi.com [050903 00:03]:
+/* We can reduce the size of code generated by romcc by
- changing all of the fixed size types that live in registers
- into simple unsigned variables. (ie s/uint8_t/unsigned/g)
- */
Why is this? I would consider specifying an 8bit type to be more space-safing than using some generic untyped integer value. If not this should be fixed in romcc..
You're right in your thinking, but at this time ROMCC converts it to an unsigned anyway and winds up producing more code doing so. So in the end the space saving measure costs more in code size produced by ROMCC.
/* AMD K8 Unsupported 1Ghz? */ if (id == (PCI_VENDOR_ID_AMD | (0x1100 << 16))) {
if (is_cpu_pre_e0()) // CK804 support 1G?
device_t dev_2 = PCI_DEV(0,0x18,2);if(pci_read_config32(dev_2,0x9c) < 0x20f00) {The function call looks a lot more readable here. How much is the gain of manual inlining here?
Stefan
-- LinuxBIOS mailing list LinuxBIOS@openbios.org http://www.openbios.org/mailman/listinfo/linuxbios
Stefan Reinauer stepan@openbios.org writes:
- jason schildt jschildt@lnxi.com [050903 00:03]:
+/* We can reduce the size of code generated by romcc by
- changing all of the fixed size types that live in registers
- into simple unsigned variables. (ie s/uint8_t/unsigned/g)
- */
Why is this? I would consider specifying an 8bit type to be more space-safing than using some generic untyped integer value. If not this should be fixed in romcc..
This is a fundamental limit, especially on 32bit x86 with it's non-symmetric registers. romcc allocates registers and registers are not 8 bits. Therefore it requires an extra operation to mask the register value to be 8 bits, after the operation.
Theoretically it could help by allowing use of registers such as %ah but the problem is that you cannot perform a register to register between register combinations like %esi, %ah and it gets even worse when you include the mmx and sse registers. So %ah is essentially unusable.
Since using smaller values does not increase the numbers of registers you can use and using smaller registers requires an extra mask step. Using smaller values increases the code size.
The comment was added to document this fact so we can revisit this later, if it becomes important. Hopefully this begins dispelling the myth that sub word sized quantities are more efficient to use.
If you are really into register savings bit-fields can help. Especially when you have more than 2 values in a register. You have to pack and unpack the values but if you don't have them all unpacked simultaneously it can help.
A sub word type is only slightly better than a bit-field in that you can use the register directly. But it still requires maintenance work to keep from having anything more than a sign bit in the registers high bits.
/* AMD K8 Unsupported 1Ghz? */ if (id == (PCI_VENDOR_ID_AMD | (0x1100 << 16))) {
if (is_cpu_pre_e0()) // CK804 support 1G?
device_t dev_2 = PCI_DEV(0,0x18,2);if(pci_read_config32(dev_2,0x9c) < 0x20f00) {The function call looks a lot more readable here. How much is the gain of manual inlining here?
100% The call actually works. cpuid requires 4 registers and we don't have that many to spare at this point in the code. What this bit does is read a cached copy of the cpu rev from a scratch register in pci configuration space. Probably the clearest thing to have would be a set of functions that perform this test. Something like is_cached_cpu_pre_e0(). Almost as good was be a good comment.
Eric
-
ebiederman＠lnxi.com -
Jason Schildt -
jason schildt
-
Stefan Reinauer -
yhlu