x86, smp: remove mach_ipi.h
[safe/jmp/linux-2.6] / arch / x86 / kernel / numaq_32.c
index 27b9082..83bb055 100644 (file)
 #include <asm/numaq.h>
 #include <asm/topology.h>
 #include <asm/processor.h>
-#include <asm/mpspec.h>
+#include <asm/genapic.h>
 #include <asm/e820.h>
+#include <asm/setup.h>
 
 #define        MB_TO_PAGES(addr) ((addr) << (20 - PAGE_SHIFT))
 
-int found_numaq;
-
 /*
  * Function: smp_dump_qct()
  *
@@ -73,6 +72,201 @@ static void __init smp_dump_qct(void)
        }
 }
 
+
+void __cpuinit numaq_tsc_disable(void)
+{
+       if (!found_numaq)
+               return;
+
+       if (num_online_nodes() > 1) {
+               printk(KERN_DEBUG "NUMAQ: disabling TSC\n");
+               setup_clear_cpu_cap(X86_FEATURE_TSC);
+       }
+}
+
+static int __init numaq_pre_time_init(void)
+{
+       numaq_tsc_disable();
+       return 0;
+}
+
+int found_numaq;
+/*
+ * Have to match translation table entries to main table entries by counter
+ * hence the mpc_record variable .... can't see a less disgusting way of
+ * doing this ....
+ */
+struct mpc_config_translation {
+       unsigned char mpc_type;
+       unsigned char trans_len;
+       unsigned char trans_type;
+       unsigned char trans_quad;
+       unsigned char trans_global;
+       unsigned char trans_local;
+       unsigned short trans_reserved;
+};
+
+/* x86_quirks member */
+static int mpc_record;
+static struct mpc_config_translation *translation_table[MAX_MPC_ENTRY]
+    __cpuinitdata;
+
+static inline int generate_logical_apicid(int quad, int phys_apicid)
+{
+       return (quad << 4) + (phys_apicid ? phys_apicid << 1 : 1);
+}
+
+/* x86_quirks member */
+static int mpc_apic_id(struct mpc_cpu *m)
+{
+       int quad = translation_table[mpc_record]->trans_quad;
+       int logical_apicid = generate_logical_apicid(quad, m->apicid);
+
+       printk(KERN_DEBUG "Processor #%d %u:%u APIC version %d (quad %d, apic %d)\n",
+              m->apicid, (m->cpufeature & CPU_FAMILY_MASK) >> 8,
+              (m->cpufeature & CPU_MODEL_MASK) >> 4,
+              m->apicver, quad, logical_apicid);
+       return logical_apicid;
+}
+
+int mp_bus_id_to_node[MAX_MP_BUSSES];
+
+int mp_bus_id_to_local[MAX_MP_BUSSES];
+
+/* x86_quirks member */
+static void mpc_oem_bus_info(struct mpc_bus *m, char *name)
+{
+       int quad = translation_table[mpc_record]->trans_quad;
+       int local = translation_table[mpc_record]->trans_local;
+
+       mp_bus_id_to_node[m->busid] = quad;
+       mp_bus_id_to_local[m->busid] = local;
+       printk(KERN_INFO "Bus #%d is %s (node %d)\n",
+              m->busid, name, quad);
+}
+
+int quad_local_to_mp_bus_id [NR_CPUS/4][4];
+
+/* x86_quirks member */
+static void mpc_oem_pci_bus(struct mpc_bus *m)
+{
+       int quad = translation_table[mpc_record]->trans_quad;
+       int local = translation_table[mpc_record]->trans_local;
+
+       quad_local_to_mp_bus_id[quad][local] = m->busid;
+}
+
+static void __init MP_translation_info(struct mpc_config_translation *m)
+{
+       printk(KERN_INFO
+              "Translation: record %d, type %d, quad %d, global %d, local %d\n",
+              mpc_record, m->trans_type, m->trans_quad, m->trans_global,
+              m->trans_local);
+
+       if (mpc_record >= MAX_MPC_ENTRY)
+               printk(KERN_ERR "MAX_MPC_ENTRY exceeded!\n");
+       else
+               translation_table[mpc_record] = m;      /* stash this for later */
+       if (m->trans_quad < MAX_NUMNODES && !node_online(m->trans_quad))
+               node_set_online(m->trans_quad);
+}
+
+static int __init mpf_checksum(unsigned char *mp, int len)
+{
+       int sum = 0;
+
+       while (len--)
+               sum += *mp++;
+
+       return sum & 0xFF;
+}
+
+/*
+ * Read/parse the MPC oem tables
+ */
+
+static void __init smp_read_mpc_oem(struct mpc_oemtable *oemtable,
+                                   unsigned short oemsize)
+{
+       int count = sizeof(*oemtable);  /* the header size */
+       unsigned char *oemptr = ((unsigned char *)oemtable) + count;
+
+       mpc_record = 0;
+       printk(KERN_INFO "Found an OEM MPC table at %8p - parsing it ... \n",
+              oemtable);
+       if (memcmp(oemtable->signature, MPC_OEM_SIGNATURE, 4)) {
+               printk(KERN_WARNING
+                      "SMP mpc oemtable: bad signature [%c%c%c%c]!\n",
+                      oemtable->signature[0], oemtable->signature[1],
+                      oemtable->signature[2], oemtable->signature[3]);
+               return;
+       }
+       if (mpf_checksum((unsigned char *)oemtable, oemtable->length)) {
+               printk(KERN_WARNING "SMP oem mptable: checksum error!\n");
+               return;
+       }
+       while (count < oemtable->length) {
+               switch (*oemptr) {
+               case MP_TRANSLATION:
+                       {
+                               struct mpc_config_translation *m =
+                                   (struct mpc_config_translation *)oemptr;
+                               MP_translation_info(m);
+                               oemptr += sizeof(*m);
+                               count += sizeof(*m);
+                               ++mpc_record;
+                               break;
+                       }
+               default:
+                       {
+                               printk(KERN_WARNING
+                                      "Unrecognised OEM table entry type! - %d\n",
+                                      (int)*oemptr);
+                               return;
+                       }
+               }
+       }
+}
+
+static int __init numaq_setup_ioapic_ids(void)
+{
+       /* so can skip it */
+       return 1;
+}
+
+static int __init numaq_update_genapic(void)
+{
+       apic->wakeup_cpu = wakeup_secondary_cpu_via_nmi;
+
+       return 0;
+}
+
+static struct x86_quirks numaq_x86_quirks __initdata = {
+       .arch_pre_time_init     = numaq_pre_time_init,
+       .arch_time_init         = NULL,
+       .arch_pre_intr_init     = NULL,
+       .arch_memory_setup      = NULL,
+       .arch_intr_init         = NULL,
+       .arch_trap_init         = NULL,
+       .mach_get_smp_config    = NULL,
+       .mach_find_smp_config   = NULL,
+       .mpc_record             = &mpc_record,
+       .mpc_apic_id            = mpc_apic_id,
+       .mpc_oem_bus_info       = mpc_oem_bus_info,
+       .mpc_oem_pci_bus        = mpc_oem_pci_bus,
+       .smp_read_mpc_oem       = smp_read_mpc_oem,
+       .setup_ioapic_ids       = numaq_setup_ioapic_ids,
+       .update_genapic         = numaq_update_genapic,
+};
+
+void numaq_mps_oem_check(struct mpc_table *mpc, char *oem, char *productid)
+{
+       if (strncmp(oem, "IBM NUMA", 8))
+               printk("Warning!  Not a NUMA-Q system!\n");
+       else
+               found_numaq = 1;
+}
+
 static __init void early_check_numaq(void)
 {
        /*
@@ -84,6 +278,9 @@ static __init void early_check_numaq(void)
         */
        if (smp_found_config)
                early_get_smp_config();
+
+       if (found_numaq)
+               x86_quirks = &numaq_x86_quirks;
 }
 
 int __init get_memcfg_numaq(void)
@@ -95,12 +292,280 @@ int __init get_memcfg_numaq(void)
        return 1;
 }
 
-static int __init numaq_tsc_disable(void)
+/*
+ * APIC driver for the IBM NUMAQ chipset.
+ */
+#define APIC_DEFINITION 1
+#include <linux/threads.h>
+#include <linux/cpumask.h>
+#include <asm/mpspec.h>
+#include <asm/genapic.h>
+#include <asm/fixmap.h>
+#include <asm/apicdef.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/init.h>
+#include <linux/numa.h>
+#include <linux/smp.h>
+#include <asm/numaq.h>
+#include <asm/io.h>
+#include <linux/mmzone.h>
+#include <linux/nodemask.h>
+
+#define NUMAQ_APIC_DFR_VALUE   (APIC_DFR_CLUSTER)
+
+static inline unsigned int numaq_get_apic_id(unsigned long x)
 {
-       if (num_online_nodes() > 1) {
-               printk(KERN_DEBUG "NUMAQ: disabling TSC\n");
-               setup_clear_cpu_cap(X86_FEATURE_TSC);
-       }
-       return 0;
+       return (x >> 24) & 0x0F;
+}
+
+void default_send_IPI_mask_sequence(const struct cpumask *mask, int vector);
+void default_send_IPI_mask_allbutself(const struct cpumask *mask, int vector);
+
+static inline void numaq_send_IPI_mask(const struct cpumask *mask, int vector)
+{
+       default_send_IPI_mask_sequence(mask, vector);
+}
+
+static inline void numaq_send_IPI_allbutself(int vector)
+{
+       default_send_IPI_mask_allbutself(cpu_online_mask, vector);
+}
+
+static inline void numaq_send_IPI_all(int vector)
+{
+       numaq_send_IPI_mask(cpu_online_mask, vector);
 }
-arch_initcall(numaq_tsc_disable);
+
+extern void numaq_mps_oem_check(struct mpc_table *, char *, char *);
+
+#define NUMAQ_TRAMPOLINE_PHYS_LOW (0x8)
+#define NUMAQ_TRAMPOLINE_PHYS_HIGH (0xa)
+
+/*
+ * Because we use NMIs rather than the INIT-STARTUP sequence to
+ * bootstrap the CPUs, the APIC may be in a weird state. Kick it:
+ */
+static inline void numaq_smp_callin_clear_local_apic(void)
+{
+       clear_local_APIC();
+}
+
+static inline void
+numaq_store_NMI_vector(unsigned short *high, unsigned short *low)
+{
+       printk("Storing NMI vector\n");
+       *high =
+         *((volatile unsigned short *)phys_to_virt(NUMAQ_TRAMPOLINE_PHYS_HIGH));
+       *low =
+         *((volatile unsigned short *)phys_to_virt(NUMAQ_TRAMPOLINE_PHYS_LOW));
+}
+
+static inline const cpumask_t *numaq_target_cpus(void)
+{
+       return &CPU_MASK_ALL;
+}
+
+static inline unsigned long
+numaq_check_apicid_used(physid_mask_t bitmap, int apicid)
+{
+       return physid_isset(apicid, bitmap);
+}
+
+static inline unsigned long numaq_check_apicid_present(int bit)
+{
+       return physid_isset(bit, phys_cpu_present_map);
+}
+
+#define apicid_cluster(apicid) (apicid & 0xF0)
+
+static inline int numaq_apic_id_registered(void)
+{
+       return 1;
+}
+
+static inline void numaq_init_apic_ldr(void)
+{
+       /* Already done in NUMA-Q firmware */
+}
+
+static inline void numaq_setup_apic_routing(void)
+{
+       printk("Enabling APIC mode:  %s.  Using %d I/O APICs\n",
+               "NUMA-Q", nr_ioapics);
+}
+
+/*
+ * Skip adding the timer int on secondary nodes, which causes
+ * a small but painful rift in the time-space continuum.
+ */
+static inline int numaq_multi_timer_check(int apic, int irq)
+{
+       return apic != 0 && irq == 0;
+}
+
+static inline physid_mask_t numaq_ioapic_phys_id_map(physid_mask_t phys_map)
+{
+       /* We don't have a good way to do this yet - hack */
+       return physids_promote(0xFUL);
+}
+
+/* Mapping from cpu number to logical apicid */
+extern u8 cpu_2_logical_apicid[];
+
+static inline int numaq_cpu_to_logical_apicid(int cpu)
+{
+       if (cpu >= nr_cpu_ids)
+               return BAD_APICID;
+       return (int)cpu_2_logical_apicid[cpu];
+}
+
+/*
+ * Supporting over 60 cpus on NUMA-Q requires a locality-dependent
+ * cpu to APIC ID relation to properly interact with the intelligent
+ * mode of the cluster controller.
+ */
+static inline int numaq_cpu_present_to_apicid(int mps_cpu)
+{
+       if (mps_cpu < 60)
+               return ((mps_cpu >> 2) << 4) | (1 << (mps_cpu & 0x3));
+       else
+               return BAD_APICID;
+}
+
+static inline int numaq_apicid_to_node(int logical_apicid) 
+{
+       return logical_apicid >> 4;
+}
+
+static inline physid_mask_t numaq_apicid_to_cpu_present(int logical_apicid)
+{
+       int node = numaq_apicid_to_node(logical_apicid);
+       int cpu = __ffs(logical_apicid & 0xf);
+
+       return physid_mask_of_physid(cpu + 4*node);
+}
+
+extern void *xquad_portio;
+
+static inline int numaq_check_phys_apicid_present(int boot_cpu_physical_apicid)
+{
+       return 1;
+}
+
+/*
+ * We use physical apicids here, not logical, so just return the default
+ * physical broadcast to stop people from breaking us
+ */
+static inline unsigned int numaq_cpu_mask_to_apicid(const cpumask_t *cpumask)
+{
+       return 0x0F;
+}
+
+static inline unsigned int
+numaq_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
+                            const struct cpumask *andmask)
+{
+       return 0x0F;
+}
+
+/* No NUMA-Q box has a HT CPU, but it can't hurt to use the default code. */
+static inline int numaq_phys_pkg_id(int cpuid_apic, int index_msb)
+{
+       return cpuid_apic >> index_msb;
+}
+static int __numaq_mps_oem_check(struct mpc_table *mpc, char *oem, char *productid)
+{
+       numaq_mps_oem_check(mpc, oem, productid);
+       return found_numaq;
+}
+
+static int probe_numaq(void)
+{
+       /* already know from get_memcfg_numaq() */
+       return found_numaq;
+}
+
+static void numaq_vector_allocation_domain(int cpu, cpumask_t *retmask)
+{
+       /* Careful. Some cpus do not strictly honor the set of cpus
+        * specified in the interrupt destination when using lowest
+        * priority interrupt delivery mode.
+        *
+        * In particular there was a hyperthreading cpu observed to
+        * deliver interrupts to the wrong hyperthread when only one
+        * hyperthread was specified in the interrupt desitination.
+        */
+       *retmask = (cpumask_t){ { [0] = APIC_ALL_CPUS, } };
+}
+
+static void numaq_setup_portio_remap(void)
+{
+       int num_quads = num_online_nodes();
+
+       if (num_quads <= 1)
+                       return;
+
+       printk("Remapping cross-quad port I/O for %d quads\n", num_quads);
+       xquad_portio = ioremap(XQUAD_PORTIO_BASE, num_quads*XQUAD_PORTIO_QUAD);
+       printk("xquad_portio vaddr 0x%08lx, len %08lx\n",
+               (u_long) xquad_portio, (u_long) num_quads*XQUAD_PORTIO_QUAD);
+}
+
+struct genapic apic_numaq = {
+
+       .name                           = "NUMAQ",
+       .probe                          = probe_numaq,
+       .acpi_madt_oem_check            = NULL,
+       .apic_id_registered             = numaq_apic_id_registered,
+
+       .irq_delivery_mode              = dest_LowestPrio,
+       /* physical delivery on LOCAL quad: */
+       .irq_dest_mode                  = 0,
+
+       .target_cpus                    = numaq_target_cpus,
+       .disable_esr                    = 1,
+       .dest_logical                   = APIC_DEST_LOGICAL,
+       .check_apicid_used              = numaq_check_apicid_used,
+       .check_apicid_present           = numaq_check_apicid_present,
+
+       .vector_allocation_domain       = numaq_vector_allocation_domain,
+       .init_apic_ldr                  = numaq_init_apic_ldr,
+
+       .ioapic_phys_id_map             = numaq_ioapic_phys_id_map,
+       .setup_apic_routing             = numaq_setup_apic_routing,
+       .multi_timer_check              = numaq_multi_timer_check,
+       .apicid_to_node                 = numaq_apicid_to_node,
+       .cpu_to_logical_apicid          = numaq_cpu_to_logical_apicid,
+       .cpu_present_to_apicid          = numaq_cpu_present_to_apicid,
+       .apicid_to_cpu_present          = numaq_apicid_to_cpu_present,
+       .setup_portio_remap             = numaq_setup_portio_remap,
+       .check_phys_apicid_present      = numaq_check_phys_apicid_present,
+       .enable_apic_mode               = NULL,
+       .phys_pkg_id                    = numaq_phys_pkg_id,
+       .mps_oem_check                  = __numaq_mps_oem_check,
+
+       .get_apic_id                    = numaq_get_apic_id,
+       .set_apic_id                    = NULL,
+       .apic_id_mask                   = 0x0F << 24,
+
+       .cpu_mask_to_apicid             = numaq_cpu_mask_to_apicid,
+       .cpu_mask_to_apicid_and         = numaq_cpu_mask_to_apicid_and,
+
+       .send_IPI_mask                  = numaq_send_IPI_mask,
+       .send_IPI_mask_allbutself       = NULL,
+       .send_IPI_allbutself            = numaq_send_IPI_allbutself,
+       .send_IPI_all                   = numaq_send_IPI_all,
+       .send_IPI_self                  = NULL,
+
+       .wakeup_cpu                     = NULL,
+       .trampoline_phys_low            = NUMAQ_TRAMPOLINE_PHYS_LOW,
+       .trampoline_phys_high           = NUMAQ_TRAMPOLINE_PHYS_HIGH,
+
+       /* We don't do anything here because we use NMI's to boot instead */
+       .wait_for_init_deassert         = NULL,
+
+       .smp_callin_clear_local_apic    = numaq_smp_callin_clear_local_apic,
+       .store_NMI_vector               = numaq_store_NMI_vector,
+       .inquire_remote_apic            = NULL,
+};