x86: cleanup early per cpu variables/accesses v4
[safe/jmp/linux-2.6] / arch / x86 / kernel / setup_64.c
index 413b8fc..e8df64f 100644 (file)
 #include <linux/crash_dump.h>
 #include <linux/root_dev.h>
 #include <linux/pci.h>
+#include <asm/pci-direct.h>
 #include <linux/efi.h>
 #include <linux/acpi.h>
 #include <linux/kallsyms.h>
 #include <linux/edd.h>
+#include <linux/iscsi_ibft.h>
 #include <linux/mmzone.h>
 #include <linux/kexec.h>
 #include <linux/cpufreq.h>
 #include <linux/dmi.h>
 #include <linux/dma-mapping.h>
 #include <linux/ctype.h>
+#include <linux/sort.h>
 #include <linux/uaccess.h>
 #include <linux/init_ohci1394_dma.h>
+#include <linux/kvm_para.h>
 
 #include <asm/mtrr.h>
 #include <asm/uaccess.h>
@@ -58,7 +62,6 @@
 #include <asm/mmu_context.h>
 #include <asm/proto.h>
 #include <asm/setup.h>
-#include <asm/mach_apic.h>
 #include <asm/numa.h>
 #include <asm/sections.h>
 #include <asm/dmi.h>
 #include <asm/mce.h>
 #include <asm/ds.h>
 #include <asm/topology.h>
+#include <asm/trampoline.h>
+#include <asm/pat.h>
 
+#include <mach_apic.h>
 #ifdef CONFIG_PARAVIRT
 #include <asm/paravirt.h>
 #else
@@ -114,7 +120,7 @@ extern int root_mountflags;
 
 char __initdata command_line[COMMAND_LINE_SIZE];
 
-struct resource standard_io_resources[] = {
+static struct resource standard_io_resources[] = {
        { .name = "dma1", .start = 0x00, .end = 0x1f,
                .flags = IORESOURCE_BUSY | IORESOURCE_IO },
        { .name = "pic1", .start = 0x20, .end = 0x21,
@@ -123,7 +129,9 @@ struct resource standard_io_resources[] = {
                .flags = IORESOURCE_BUSY | IORESOURCE_IO },
        { .name = "timer1", .start = 0x50, .end = 0x53,
                .flags = IORESOURCE_BUSY | IORESOURCE_IO },
-       { .name = "keyboard", .start = 0x60, .end = 0x6f,
+       { .name = "keyboard", .start = 0x60, .end = 0x60,
+               .flags = IORESOURCE_BUSY | IORESOURCE_IO },
+       { .name = "keyboard", .start = 0x64, .end = 0x64,
                .flags = IORESOURCE_BUSY | IORESOURCE_IO },
        { .name = "dma page reg", .start = 0x80, .end = 0x8f,
                .flags = IORESOURCE_BUSY | IORESOURCE_IO },
@@ -188,6 +196,7 @@ contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
        bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
        e820_register_active_regions(0, start_pfn, end_pfn);
        free_bootmem_with_active_regions(0, end_pfn);
+       early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
        reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
 }
 #endif
@@ -262,6 +271,40 @@ void __attribute__((weak)) __init memory_setup(void)
        machine_specific_memory_setup();
 }
 
+static void __init parse_setup_data(void)
+{
+       struct setup_data *data;
+       unsigned long pa_data;
+
+       if (boot_params.hdr.version < 0x0209)
+               return;
+       pa_data = boot_params.hdr.setup_data;
+       while (pa_data) {
+               data = early_ioremap(pa_data, PAGE_SIZE);
+               switch (data->type) {
+               default:
+                       break;
+               }
+#ifndef CONFIG_DEBUG_BOOT_PARAMS
+               free_early(pa_data, pa_data+sizeof(*data)+data->len);
+#endif
+               pa_data = data->next;
+               early_iounmap(data, PAGE_SIZE);
+       }
+}
+
+#ifdef CONFIG_PCI_MMCONFIG
+extern void __cpuinit fam10h_check_enable_mmcfg(void);
+extern void __init check_enable_amd_mmconf_dmi(void);
+#else
+void __cpuinit fam10h_check_enable_mmcfg(void)
+{
+}
+void __init check_enable_amd_mmconf_dmi(void)
+{
+}
+#endif
+
 /*
  * setup_arch - architecture-specific boot-time initializations
  *
@@ -314,6 +357,8 @@ void __init setup_arch(char **cmdline_p)
        strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
        *cmdline_p = command_line;
 
+       parse_setup_data();
+
        parse_early_param();
 
 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
@@ -347,7 +392,7 @@ void __init setup_arch(char **cmdline_p)
 
        check_efer();
 
-       init_memory_mapping(0, (max_pfn_mapped << PAGE_SHIFT));
+       max_pfn_mapped = init_memory_mapping(0, (max_pfn_mapped << PAGE_SHIFT));
        if (efi_enabled)
                efi_init();
 
@@ -357,13 +402,8 @@ void __init setup_arch(char **cmdline_p)
 
        io_delay_init();
 
-#ifdef CONFIG_SMP
-       /* setup to use the early static init tables during kernel startup */
-       x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
-       x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
-#ifdef CONFIG_NUMA
-       x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
-#endif
+#ifdef CONFIG_KVM_CLOCK
+       kvmclock_init();
 #endif
 
 #ifdef CONFIG_ACPI
@@ -395,7 +435,7 @@ void __init setup_arch(char **cmdline_p)
        contig_initmem_init(0, end_pfn);
 #endif
 
-       early_res_to_bootmem();
+       dma32_reserve_bootmem();
 
 #ifdef CONFIG_ACPI_SLEEP
        /*
@@ -419,11 +459,14 @@ void __init setup_arch(char **cmdline_p)
                unsigned long end_of_mem    = end_pfn << PAGE_SHIFT;
 
                if (ramdisk_end <= end_of_mem) {
-                       reserve_bootmem_generic(ramdisk_image, ramdisk_size);
+                       /*
+                        * don't need to reserve again, already reserved early
+                        * in x86_64_start_kernel, and early_res_to_bootmem
+                        * convert that to reserved in bootmem
+                        */
                        initrd_start = ramdisk_image + PAGE_OFFSET;
                        initrd_end = initrd_start+ramdisk_size;
                } else {
-                       /* Assumes everything on node 0 */
                        free_bootmem(ramdisk_image, ramdisk_size);
                        printk(KERN_ERR "initrd extends beyond end of memory "
                               "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
@@ -433,6 +476,9 @@ void __init setup_arch(char **cmdline_p)
        }
 #endif
        reserve_crashkernel();
+
+       reserve_ibft_region();
+
        paging_init();
        map_vsyscall();
 
@@ -455,6 +501,8 @@ void __init setup_arch(char **cmdline_p)
        init_apic_mappings();
        ioapic_init_mappings();
 
+       kvm_guest_init();
+
        /*
         * We trust e820 completely. No explicit ROM probing in memory.
         */
@@ -475,6 +523,9 @@ void __init setup_arch(char **cmdline_p)
        conswitchp = &dummy_con;
 #endif
 #endif
+
+       /* do this before identify_cpu for boot cpu */
+       check_enable_amd_mmconf_dmi();
 }
 
 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
@@ -727,8 +778,24 @@ static void __cpuinit init_amd(struct cpuinfo_x86 *c)
        /* MFENCE stops RDTSC speculation */
        set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
 
+       if (c->x86 == 0x10)
+               fam10h_check_enable_mmcfg();
+
        if (amd_apic_timer_broken())
                disable_apic_timer = 1;
+
+       if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
+               unsigned long long tseg;
+
+               /*
+                * Split up direct mapping around the TSEG SMM area.
+                * Don't do it for gbpages because there seems very little
+                * benefit in doing so.
+                */
+               if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg) &&
+               (tseg >> PMD_SHIFT) < (max_pfn_mapped >> (PMD_SHIFT-PAGE_SHIFT)))
+                       set_memory_4k((unsigned long)__va(tseg), 1);
+       }
 }
 
 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
@@ -872,6 +939,32 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
        srat_detect_node();
 }
 
+static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
+{
+       if (c->x86 == 0x6 && c->x86_model >= 0xf)
+               set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
+}
+
+static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
+{
+       /* Cache sizes */
+       unsigned n;
+
+       n = c->extended_cpuid_level;
+       if (n >= 0x80000008) {
+               unsigned eax = cpuid_eax(0x80000008);
+               c->x86_virt_bits = (eax >> 8) & 0xff;
+               c->x86_phys_bits = eax & 0xff;
+       }
+
+       if (c->x86 == 0x6 && c->x86_model >= 0xf) {
+               c->x86_cache_alignment = c->x86_clflush_size * 2;
+               set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
+               set_cpu_cap(c, X86_FEATURE_REP_GOOD);
+       }
+       set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
+}
+
 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
 {
        char *v = c->x86_vendor_id;
@@ -880,6 +973,8 @@ static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
                c->x86_vendor = X86_VENDOR_AMD;
        else if (!strcmp(v, "GenuineIntel"))
                c->x86_vendor = X86_VENDOR_INTEL;
+       else if (!strcmp(v, "CentaurHauls"))
+               c->x86_vendor = X86_VENDOR_CENTAUR;
        else
                c->x86_vendor = X86_VENDOR_UNKNOWN;
 }
@@ -962,22 +1057,19 @@ static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
        if (c->extended_cpuid_level >= 0x80000007)
                c->x86_power = cpuid_edx(0x80000007);
 
-
-       clear_cpu_cap(c, X86_FEATURE_PAT);
-
        switch (c->x86_vendor) {
        case X86_VENDOR_AMD:
                early_init_amd(c);
-               if (c->x86 >= 0xf && c->x86 <= 0x11)
-                       set_cpu_cap(c, X86_FEATURE_PAT);
                break;
        case X86_VENDOR_INTEL:
                early_init_intel(c);
-               if (c->x86 == 0xF || (c->x86 == 6 && c->x86_model >= 15))
-                       set_cpu_cap(c, X86_FEATURE_PAT);
+               break;
+       case X86_VENDOR_CENTAUR:
+               early_init_centaur(c);
                break;
        }
 
+       validate_pat_support(c);
 }
 
 /*
@@ -1012,6 +1104,10 @@ void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
                init_intel(c);
                break;
 
+       case X86_VENDOR_CENTAUR:
+               init_centaur(c);
+               break;
+
        case X86_VENDOR_UNKNOWN:
        default:
                display_cacheinfo(c);