DDB and SMP causes Unaligned Reference

[Alan Robinson]

Using the manual escape to debugger with an SMP kernel
causes an unaligned reference while stopping the other
CPU, the system then hangs.

#
#  CPU1 stopping CPUs: 0x00000001...
#  fatal kernel trap (cpu 0):
# 
# trap vector = 0x1e (Unaligned Reference)
# cr.iip      = 0xe0000000009a1320
# cr.ipsr     = 0x1210080a6010 (mfl,ic,i,dt,dfh,rt,cpl=0,it,ri=1,bn)
# cr.isr      = 0x20200000000 (code=0,vector=0,w,ei=1)
# cr.ifa      = 0xe000000000004228
# curthread   = 0xe00000003d9bb080
# pid = 12, comm = idle: cpu0
# CPU0 stopping CPUs: 0x00000002...
# 
(The self built kernel was built using projects/ia64 cvsup'd sources 
which are now a day or two old)

The cr.iip points into swapctx (or savectx) which was called from
from interrupt() as follows....

        } else if (vector == ipi_vector[IPI_STOP]) {
                u_int32_t mybit = PCPU_GET(cpumask);

                CTR1(KTR_SMP, "IPI_STOP, cpuid=%d", PCPU_GET(cpuid));
                savectx(PCPU_GET(pcb));
                stopped_cpus |= mybit;
                while ((started_cpus & mybit) == 0)
                        /* spin */;
                started_cpus &= ~mybit;
                stopped_cpus &= ~mybit;
                if (PCPU_GET(cpuid) == 0 && cpustop_restartfunc != NULL) {
                        void (*f)(void) = cpustop_restartfunc;
                        cpustop_restartfunc = NULL;
                        (*f)();
                }

I cannot find a place where PCU_SET(pcb) is used nor can I find a 
direct setting of pc_pcb :-(

The following 2 line change in sys/ia64/ia64/mp_machdep.c fixes the problem
for me.
 
+static struct pcb ia64_intr_pcb[MAXCPU];
void
cpu_mp_start()
{
        struct pcpu *pc;
        
        ap_spin = 1;
   
        SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
                pc->pc_current_pmap = kernel_pmap;
                pc->pc_other_cpus = all_cpus & ~pc->pc_cpumask;
+               pc->pc_pcb = &(ia64_intr_pcb[pc->pc_cpuid]);
                if (pc->pc_cpuid > 0) {
                        ap_stack = malloc(KSTACK_PAGES * PAGE_SIZE, M_PMAP,
                            M_WAITOK); 
                        ap_pcpu = pc;
                        ap_delay = 2000;
                        ap_awake = 0;


I can now get the following:

# manual escape to debugger
# CPU1 stopping CPUs: 0x00000001... stopped.
# Stopped at      Debugger+0x31:        nop.m 0x0
# db> trace
# Debugger(0xe000000000a10a78, 0xe000000000990bf0, 0x692) at Debugger+0x30
# scgetc(0xe0000000002ae800, 0x2, 0xe0000000011f2cb0) at scgetc+0xbe0
# sckbdevent(0xe000000000b232e8, 0x0, 0xe0000000002ae800) at sckbdevent+0x640
# ukbd_interrupt(0xe000000000b232e8, 0x0) at ukbd_interrupt+0x850
# ukbd_intr(0xe000000001167a00, 0xe000000000b232e8, 0x0) at ukbd_intr+0x80
# usb_transfer_complete(0xe000000001167a00, 0xe000000000ae6110) at usb_transfer_complete+0x400
# ohci_softintr(0xe0000000010ab000) at ohci_softintr+0x240
# usb_schedsoftintr(0xe0000000010ab000, 0xe000000000ae6110, 0xe0000000006728f0, 0x30a) at usb_schedsoftintr+0x50
# ohci_intr1(0xe0000000010ab000) at ohci_intr1+0x440
# ohci_intr(0xe0000000010ab000, 0xe0000000006f97b0, 0x1024) at ohci_intr+0x80
# ithread_loop(0xe0000000011a8200, 0xe000000001118100, 0xe000000000ae6110, 0xe000000000ab5590) at ithread_loop+0x460
# fork_exit(0xe000000000a1d4d8, 0xe0000000011a8200, 0xa000000000019580) at fork_exit+0x1b0
# enter_userland() at enter_userland
# db> c
# CPU1 restarting CPUs: 0x00000001... restarted.
# manual escape to debugger
# CPU0 stopping CPUs: 0x00000002... stopped.
# Stopped at      Debugger+0x31:        nop.m 0x0
# db> c
# CPU0 restarting CPUs: 0x00000002... restarted.
# 

is the double continue normal ?

Alan