diff --git a/arch/ia64/kernel/mca.c b/arch/ia64/kernel/mca.c
index 8d48420..6dc726a 100644
--- a/arch/ia64/kernel/mca.c
+++ b/arch/ia64/kernel/mca.c
@@ -48,6 +48,9 @@
* Delete dead variables and functions.
* Reorder to remove the need for forward declarations and to consolidate
* related code.
+ *
+ * Convert MCA/INIT handlers to use per event stacks and SAL/OS state.
*/
#include <linux/config.h>
#include <linux/types.h>
@@ -77,6 +80,8 @@
#include <asm/irq.h>
#include <asm/hw_irq.h>
+#include "entry.h"
+
#if defined(IA64_MCA_DEBUG_INFO)
# define IA64_MCA_DEBUG(fmt...) printk(fmt)
#else
@@ -84,9 +89,7 @@
#endif
/* Used by mca_asm.S */
-ia64_mca_sal_to_os_state_t ia64_sal_to_os_handoff_state;
-ia64_mca_os_to_sal_state_t ia64_os_to_sal_handoff_state;
-u64 ia64_mca_serialize;
+u32 ia64_mca_serialize;
DEFINE_PER_CPU(u64, ia64_mca_data); /* == __per_cpu_mca[smp_processor_id()] */
DEFINE_PER_CPU(u64, ia64_mca_per_cpu_pte); /* PTE to map per-CPU area */
DEFINE_PER_CPU(u64, ia64_mca_pal_pte); /* PTE to map PAL code */
@@ -95,8 +98,10 @@ DEFINE_PER_CPU(u64, ia64_mca_pal_base); /* vaddr PAL code granule */
unsigned long __per_cpu_mca[NR_CPUS];
/* In mca_asm.S */
-extern void ia64_monarch_init_handler (void);
-extern void ia64_slave_init_handler (void);
+extern void ia64_os_init_dispatch_monarch (void);
+extern void ia64_os_init_dispatch_slave (void);
+
+static int monarch_cpu = -1;
static ia64_mc_info_t ia64_mc_info;
@@ -234,7 +239,8 @@ ia64_log_get(int sal_info_type, u8 **buffer, int irq_safe)
* This function retrieves a specified error record type from SAL
* and wakes up any processes waiting for error records.
*
- * Inputs : sal_info_type (Type of error record MCA/CMC/CPE/INIT)
+ * Inputs : sal_info_type (Type of error record MCA/CMC/CPE)
+ * FIXME: remove MCA and irq_safe.
*/
static void
ia64_mca_log_sal_error_record(int sal_info_type)
@@ -242,7 +248,7 @@ ia64_mca_log_sal_error_record(int sal_info_type)
u8 *buffer;
sal_log_record_header_t *rh;
u64 size;
- int irq_safe = sal_info_type != SAL_INFO_TYPE_MCA && sal_info_type != SAL_INFO_TYPE_INIT;
+ int irq_safe = sal_info_type != SAL_INFO_TYPE_MCA;
#ifdef IA64_MCA_DEBUG_INFO
static const char * const rec_name[] = { "MCA", "INIT", "CMC", "CPE" };
#endif
@@ -330,182 +336,6 @@ ia64_mca_cpe_int_handler (int cpe_irq, void *arg, struct pt_regs *ptregs)
#endif /* CONFIG_ACPI */
-static void
-show_min_state (pal_min_state_area_t *minstate)
-{
- u64 iip = minstate->pmsa_iip + ((struct ia64_psr *)(&minstate->pmsa_ipsr))->ri;
- u64 xip = minstate->pmsa_xip + ((struct ia64_psr *)(&minstate->pmsa_xpsr))->ri;
-
- printk("NaT bits\t%016lx\n", minstate->pmsa_nat_bits);
- printk("pr\t\t%016lx\n", minstate->pmsa_pr);
- printk("b0\t\t%016lx ", minstate->pmsa_br0); print_symbol("%s\n", minstate->pmsa_br0);
- printk("ar.rsc\t\t%016lx\n", minstate->pmsa_rsc);
- printk("cr.iip\t\t%016lx ", iip); print_symbol("%s\n", iip);
- printk("cr.ipsr\t\t%016lx\n", minstate->pmsa_ipsr);
- printk("cr.ifs\t\t%016lx\n", minstate->pmsa_ifs);
- printk("xip\t\t%016lx ", xip); print_symbol("%s\n", xip);
- printk("xpsr\t\t%016lx\n", minstate->pmsa_xpsr);
- printk("xfs\t\t%016lx\n", minstate->pmsa_xfs);
- printk("b1\t\t%016lx ", minstate->pmsa_br1);
- print_symbol("%s\n", minstate->pmsa_br1);
-
- printk("\nstatic registers r0-r15:\n");
- printk(" r0- 3 %016lx %016lx %016lx %016lx\n",
- 0UL, minstate->pmsa_gr[0], minstate->pmsa_gr[1], minstate->pmsa_gr[2]);
- printk(" r4- 7 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_gr[3], minstate->pmsa_gr[4],
- minstate->pmsa_gr[5], minstate->pmsa_gr[6]);
- printk(" r8-11 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_gr[7], minstate->pmsa_gr[8],
- minstate->pmsa_gr[9], minstate->pmsa_gr[10]);
- printk("r12-15 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_gr[11], minstate->pmsa_gr[12],
- minstate->pmsa_gr[13], minstate->pmsa_gr[14]);
-
- printk("\nbank 0:\n");
- printk("r16-19 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_bank0_gr[0], minstate->pmsa_bank0_gr[1],
- minstate->pmsa_bank0_gr[2], minstate->pmsa_bank0_gr[3]);
- printk("r20-23 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_bank0_gr[4], minstate->pmsa_bank0_gr[5],
- minstate->pmsa_bank0_gr[6], minstate->pmsa_bank0_gr[7]);
- printk("r24-27 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_bank0_gr[8], minstate->pmsa_bank0_gr[9],
- minstate->pmsa_bank0_gr[10], minstate->pmsa_bank0_gr[11]);
- printk("r28-31 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_bank0_gr[12], minstate->pmsa_bank0_gr[13],
- minstate->pmsa_bank0_gr[14], minstate->pmsa_bank0_gr[15]);
-
- printk("\nbank 1:\n");
- printk("r16-19 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_bank1_gr[0], minstate->pmsa_bank1_gr[1],
- minstate->pmsa_bank1_gr[2], minstate->pmsa_bank1_gr[3]);
- printk("r20-23 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_bank1_gr[4], minstate->pmsa_bank1_gr[5],
- minstate->pmsa_bank1_gr[6], minstate->pmsa_bank1_gr[7]);
- printk("r24-27 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_bank1_gr[8], minstate->pmsa_bank1_gr[9],
- minstate->pmsa_bank1_gr[10], minstate->pmsa_bank1_gr[11]);
- printk("r28-31 %016lx %016lx %016lx %016lx\n",
- minstate->pmsa_bank1_gr[12], minstate->pmsa_bank1_gr[13],
- minstate->pmsa_bank1_gr[14], minstate->pmsa_bank1_gr[15]);
-}
-
-static void
-fetch_min_state (pal_min_state_area_t *ms, struct pt_regs *pt, struct switch_stack *sw)
-{
- u64 *dst_banked, *src_banked, bit, shift, nat_bits;
- int i;
-
- /*
- * First, update the pt-regs and switch-stack structures with the contents stored
- * in the min-state area:
- */
- if (((struct ia64_psr *) &ms->pmsa_ipsr)->ic == 0) {
- pt->cr_ipsr = ms->pmsa_xpsr;
- pt->cr_iip = ms->pmsa_xip;
- pt->cr_ifs = ms->pmsa_xfs;
- } else {
- pt->cr_ipsr = ms->pmsa_ipsr;
- pt->cr_iip = ms->pmsa_iip;
- pt->cr_ifs = ms->pmsa_ifs;
- }
- pt->ar_rsc = ms->pmsa_rsc;
- pt->pr = ms->pmsa_pr;
- pt->r1 = ms->pmsa_gr[0];
- pt->r2 = ms->pmsa_gr[1];
- pt->r3 = ms->pmsa_gr[2];
- sw->r4 = ms->pmsa_gr[3];
- sw->r5 = ms->pmsa_gr[4];
- sw->r6 = ms->pmsa_gr[5];
- sw->r7 = ms->pmsa_gr[6];
- pt->r8 = ms->pmsa_gr[7];
- pt->r9 = ms->pmsa_gr[8];
- pt->r10 = ms->pmsa_gr[9];
- pt->r11 = ms->pmsa_gr[10];
- pt->r12 = ms->pmsa_gr[11];
- pt->r13 = ms->pmsa_gr[12];
- pt->r14 = ms->pmsa_gr[13];
- pt->r15 = ms->pmsa_gr[14];
- dst_banked = &pt->r16; /* r16-r31 are contiguous in struct pt_regs */
- src_banked = ms->pmsa_bank1_gr;
- for (i = 0; i < 16; ++i)
- dst_banked[i] = src_banked[i];
- pt->b0 = ms->pmsa_br0;
- sw->b1 = ms->pmsa_br1;
-
- /* construct the NaT bits for the pt-regs structure: */
-# define PUT_NAT_BIT(dst, addr) \
- do { \
- bit = nat_bits & 1; nat_bits >>= 1; \
- shift = ((unsigned long) addr >> 3) & 0x3f; \
- dst = ((dst) & ~(1UL << shift)) | (bit << shift); \
- } while (0)
-
- /* Rotate the saved NaT bits such that bit 0 corresponds to pmsa_gr[0]: */
- shift = ((unsigned long) &ms->pmsa_gr[0] >> 3) & 0x3f;
- nat_bits = (ms->pmsa_nat_bits >> shift) | (ms->pmsa_nat_bits << (64 - shift));
-
- PUT_NAT_BIT(sw->caller_unat, &pt->r1);
- PUT_NAT_BIT(sw->caller_unat, &pt->r2);
- PUT_NAT_BIT(sw->caller_unat, &pt->r3);
- PUT_NAT_BIT(sw->ar_unat, &sw->r4);
- PUT_NAT_BIT(sw->ar_unat, &sw->r5);
- PUT_NAT_BIT(sw->ar_unat, &sw->r6);
- PUT_NAT_BIT(sw->ar_unat, &sw->r7);
- PUT_NAT_BIT(sw->caller_unat, &pt->r8); PUT_NAT_BIT(sw->caller_unat, &pt->r9);
- PUT_NAT_BIT(sw->caller_unat, &pt->r10); PUT_NAT_BIT(sw->caller_unat, &pt->r11);
- PUT_NAT_BIT(sw->caller_unat, &pt->r12); PUT_NAT_BIT(sw->caller_unat, &pt->r13);
- PUT_NAT_BIT(sw->caller_unat, &pt->r14); PUT_NAT_BIT(sw->caller_unat, &pt->r15);
- nat_bits >>= 16; /* skip over bank0 NaT bits */
- PUT_NAT_BIT(sw->caller_unat, &pt->r16); PUT_NAT_BIT(sw->caller_unat, &pt->r17);
- PUT_NAT_BIT(sw->caller_unat, &pt->r18); PUT_NAT_BIT(sw->caller_unat, &pt->r19);
- PUT_NAT_BIT(sw->caller_unat, &pt->r20); PUT_NAT_BIT(sw->caller_unat, &pt->r21);
- PUT_NAT_BIT(sw->caller_unat, &pt->r22); PUT_NAT_BIT(sw->caller_unat, &pt->r23);
- PUT_NAT_BIT(sw->caller_unat, &pt->r24); PUT_NAT_BIT(sw->caller_unat, &pt->r25);
- PUT_NAT_BIT(sw->caller_unat, &pt->r26); PUT_NAT_BIT(sw->caller_unat, &pt->r27);
- PUT_NAT_BIT(sw->caller_unat, &pt->r28); PUT_NAT_BIT(sw->caller_unat, &pt->r29);
- PUT_NAT_BIT(sw->caller_unat, &pt->r30); PUT_NAT_BIT(sw->caller_unat, &pt->r31);
-}
-
-static void
-init_handler_platform (pal_min_state_area_t *ms,
- struct pt_regs *pt, struct switch_stack *sw)
-{
- struct unw_frame_info info;
-
- /* if a kernel debugger is available call it here else just dump the registers */
-
- /*
- * Wait for a bit. On some machines (e.g., HP's zx2000 and zx6000, INIT can be
- * generated via the BMC's command-line interface, but since the console is on the
- * same serial line, the user will need some time to switch out of the BMC before
- * the dump begins.
- */
- printk("Delaying for 5 seconds...\n");
- udelay(5*1000000);
- show_min_state(ms);
-
- printk("Backtrace of current task (pid %d, %s)\n", current->pid, current->comm);
- fetch_min_state(ms, pt, sw);
- unw_init_from_interruption(&info, current, pt, sw);
- ia64_do_show_stack(&info, NULL);
-
- if (read_trylock(&tasklist_lock)) {
- struct task_struct *g, *t;
- do_each_thread (g, t) {
- if (t == current)
- continue;
-
- printk("\nBacktrace of pid %d (%s)\n", t->pid, t->comm);
- show_stack(t, NULL);
- } while_each_thread (g, t);
- }
-
- printk("\nINIT dump complete. Please reboot now.\n");
- while (1); /* hang city if no debugger */
-}
-
#ifdef CONFIG_ACPI
/*
* ia64_mca_register_cpev
@@ -648,42 +478,6 @@ ia64_mca_cmc_vector_enable_keventd(void *unused)
}
/*
- * ia64_mca_wakeup_ipi_wait
- *
- * Wait for the inter-cpu interrupt to be sent by the
- * monarch processor once it is done with handling the
- * MCA.
- *
- * Inputs : None
- * Outputs : None
- */
-static void
-ia64_mca_wakeup_ipi_wait(void)
-{
- int irr_num = (IA64_MCA_WAKEUP_VECTOR >> 6);
- int irr_bit = (IA64_MCA_WAKEUP_VECTOR & 0x3f);
- u64 irr = 0;
-
- do {
- switch(irr_num) {
- case 0:
- irr = ia64_getreg(_IA64_REG_CR_IRR0);
- break;
- case 1:
- irr = ia64_getreg(_IA64_REG_CR_IRR1);
- break;
- case 2:
- irr = ia64_getreg(_IA64_REG_CR_IRR2);
- break;
- case 3:
- irr = ia64_getreg(_IA64_REG_CR_IRR3);
- break;
- }
- cpu_relax();
- } while (!(irr & (1UL << irr_bit))) ;
-}
-
-/*
* ia64_mca_wakeup
*
* Send an inter-cpu interrupt to wake-up a particular cpu
@@ -748,11 +542,9 @@ ia64_mca_rendez_int_handler(int rendez_irq, void *arg, struct pt_regs *ptregs)
*/
ia64_sal_mc_rendez();
- /* Wait for the wakeup IPI from the monarch
- * This waiting is done by polling on the wakeup-interrupt
- * vector bit in the processor's IRRs
- */
- ia64_mca_wakeup_ipi_wait();
+ /* Wait for the monarch cpu to exit. */
+ while (monarch_cpu != -1)
+ cpu_relax(); /* spin until monarch leaves */
/* Enable all interrupts */
local_irq_restore(flags);
@@ -780,53 +572,13 @@ ia64_mca_wakeup_int_handler(int wakeup_irq, void *arg, struct pt_regs *ptregs)
return IRQ_HANDLED;
}
-/*
- * ia64_return_to_sal_check
- *
- * This is function called before going back from the OS_MCA handler
- * to the OS_MCA dispatch code which finally takes the control back
- * to the SAL.
- * The main purpose of this routine is to setup the OS_MCA to SAL
- * return state which can be used by the OS_MCA dispatch code
- * just before going back to SAL.
- *
- * Inputs : None
- * Outputs : None
- */
-
-static void
-ia64_return_to_sal_check(int recover)
-{
-
- /* Copy over some relevant stuff from the sal_to_os_mca_handoff
- * so that it can be used at the time of os_mca_to_sal_handoff
- */
- ia64_os_to_sal_handoff_state.imots_sal_gp =
- ia64_sal_to_os_handoff_state.imsto_sal_gp;
-
- ia64_os_to_sal_handoff_state.imots_sal_check_ra =
- ia64_sal_to_os_handoff_state.imsto_sal_check_ra;
-
- if (recover)
- ia64_os_to_sal_handoff_state.imots_os_status = IA64_MCA_CORRECTED;
- else
- ia64_os_to_sal_handoff_state.imots_os_status = IA64_MCA_COLD_BOOT;
-
- /* Default = tell SAL to return to same context */
- ia64_os_to_sal_handoff_state.imots_context = IA64_MCA_SAME_CONTEXT;
-
- ia64_os_to_sal_handoff_state.imots_new_min_state =
- (u64 *)ia64_sal_to_os_handoff_state.pal_min_state;
-
-}
-
/* Function pointer for extra MCA recovery */
int (*ia64_mca_ucmc_extension)
- (void*,ia64_mca_sal_to_os_state_t*,ia64_mca_os_to_sal_state_t*)
+ (void*,struct ia64_sal_os_state*)
= NULL;
int
-ia64_reg_MCA_extension(void *fn)
+ia64_reg_MCA_extension(int (*fn)(void *, struct ia64_sal_os_state *))
{
if (ia64_mca_ucmc_extension)
return 1;
@@ -845,8 +597,321 @@ ia64_unreg_MCA_extension(void)
EXPORT_SYMBOL(ia64_reg_MCA_extension);
EXPORT_SYMBOL(ia64_unreg_MCA_extension);
+
+static inline void
+copy_reg(const u64 *fr, u64 fnat, u64 *tr, u64 *tnat)
+{
+ u64 fslot, tslot, nat;
+ *tr = *fr;
+ fslot = ((unsigned long)fr >> 3) & 63;
+ tslot = ((unsigned long)tr >> 3) & 63;
+ *tnat &= ~(1UL << tslot);
+ nat = (fnat >> fslot) & 1;
+ *tnat |= (nat << tslot);
+}
+
+/* On entry to this routine, we are running on the per cpu stack, see
+ * mca_asm.h. The original stack has not been touched by this event. Some of
+ * the original stack's registers will be in the RBS on this stack. This stack
+ * also contains a partial pt_regs and switch_stack, the rest of the data is in
+ * PAL minstate.
+ *
+ * The first thing to do is modify the original stack to look like a blocked
+ * task so we can run backtrace on the original task. Also mark the per cpu
+ * stack as current to ensure that we use the correct task state, it also means
+ * that we can do backtrace on the MCA/INIT handler code itself.
+ */
+
+static task_t *
+ia64_mca_modify_original_stack(struct pt_regs *regs,
+ const struct switch_stack *sw,
+ struct ia64_sal_os_state *sos,
+ const char *type)
+{
+ char *p, comm[sizeof(current->comm)];
+ ia64_va va;
+ extern char ia64_leave_kernel[]; /* Need asm address, not function descriptor */
+ const pal_min_state_area_t *ms = sos->pal_min_state;
+ task_t *previous_current;
+ struct pt_regs *old_regs;
+ struct switch_stack *old_sw;
+ unsigned size = sizeof(struct pt_regs) +
+ sizeof(struct switch_stack) + 16;
+ u64 *old_bspstore, *old_bsp;
+ u64 *new_bspstore, *new_bsp;
+ u64 old_unat, old_rnat, new_rnat, nat;
+ u64 slots, loadrs = regs->loadrs;
+ u64 r12 = ms->pmsa_gr[12-1], r13 = ms->pmsa_gr[13-1];
+ u64 ar_bspstore = regs->ar_bspstore;
+ u64 ar_bsp = regs->ar_bspstore + (loadrs >> 16);
+ const u64 *bank;
+ const char *msg;
+ int cpu = smp_processor_id();
+
+ previous_current = curr_task(cpu);
+ set_curr_task(cpu, current);
+ if ((p = strchr(current->comm, ' ')))
+ *p = '\0';
+
+ /* Best effort attempt to cope with MCA/INIT delivered while in
+ * physical mode.
+ */
+ regs->cr_ipsr = ms->pmsa_ipsr;
+ if (ia64_psr(regs)->dt == 0) {
+ va.l = r12;
+ if (va.f.reg == 0) {
+ va.f.reg = 7;
+ r12 = va.l;
+ }
+ va.l = r13;
+ if (va.f.reg == 0) {
+ va.f.reg = 7;
+ r13 = va.l;
+ }
+ }
+ if (ia64_psr(regs)->rt == 0) {
+ va.l = ar_bspstore;
+ if (va.f.reg == 0) {
+ va.f.reg = 7;
+ ar_bspstore = va.l;
+ }