From b830ab665ad96c6b20d51a89b35cbc09ab5a2c29 Mon Sep 17 00:00:00 2001 From: "David S. Miller" Date: Tue, 28 Feb 2006 15:10:26 -0800 Subject: [PATCH] [SPARC64]: Fix bugs in SUN4V cpu mondo dispatch. There were several bugs in the SUN4V cpu mondo dispatch code. In fact, if we ever got a EWOULDBLOCK or other error from the hypervisor call, we'd potentially send a cpu mondo multiple times to the same cpu and even worse we could loop until the timeout resending the same mondo over and over to such cpus. So let's bulletproof this thing as follows: 1) Implement cpu_mondo_send() and cpu_state() hypervisor calls in arch/sparc64/kernel/entry.S, add prototypes to asm/hypervisor.h 2) Don't build and update the cpulist using inline functions, this was causing the cpu mask to not get updated in the caller. 3) Disable interrupts during the entire mondo send, otherwise our cpu list and/or mondo block could get overwritten if we take an interrupt and do a cpu mondo send on the current cpu. 4) Check for all possible error return types from the cpu_mondo_send() hypervisor call. In particular: HV_EOK) Our work is done, all cpus have received the mondo. HV_CPUERROR) One or more of the cpus in the cpu list we passed to the hypervisor are in error state. Use cpu_state() calls over the entries in the cpu list to see which ones. Record them in "error_mask" and report this after we are done sending the mondo to cpus which are not in error state. HV_EWOULDBLOCK) We need to keep trying. Any other error we consider fatal, we report the event and exit immediately. 5) We only timeout if forward progress is not made. Forward progress is defined as having at least one cpu get the mondo successfully in a given cpu_mondo_send() call. Otherwise we bump a counter and delay a little. If the counter hits a limit, we signal an error and report the event. Also, smp_call_function_mask() error handling reports the number of cpus incorrectly. Signed-off-by: David S. Miller --- arch/sparc64/kernel/entry.S | 28 +++++ arch/sparc64/kernel/smp.c | 180 +++++++++++++++++++++---------- include/asm-sparc64/hypervisor.h | 10 ++ 3 files changed, 161 insertions(+), 57 deletions(-) diff --git a/arch/sparc64/kernel/entry.S b/arch/sparc64/kernel/entry.S index 9f3048e64e..6d0b3ed77a 100644 --- a/arch/sparc64/kernel/entry.S +++ b/arch/sparc64/kernel/entry.S @@ -1795,3 +1795,31 @@ sun4v_cpu_yield: ta HV_FAST_TRAP retl nop + + /* %o0: num cpus in cpu list + * %o1: cpu list paddr + * %o2: mondo block paddr + * + * returns %o0: status + */ + .globl sun4v_cpu_mondo_send +sun4v_cpu_mondo_send: + mov HV_FAST_CPU_MONDO_SEND, %o5 + ta HV_FAST_TRAP + retl + nop + + /* %o0: CPU ID + * + * returns %o0: -status if status non-zero, else + * %o0: cpu state as HV_CPU_STATE_* + */ + .globl sun4v_cpu_state +sun4v_cpu_state: + mov HV_FAST_CPU_STATE, %o5 + ta HV_FAST_TRAP + brnz,pn %o0, 1f + sub %g0, %o0, %o0 + mov %o1, %o0 +1: retl + nop diff --git a/arch/sparc64/kernel/smp.c b/arch/sparc64/kernel/smp.c index eb7c0f855b..6bc7fd47e4 100644 --- a/arch/sparc64/kernel/smp.c +++ b/arch/sparc64/kernel/smp.c @@ -556,77 +556,144 @@ retry: } /* Multi-cpu list version. */ -static int init_cpu_list(u16 *list, cpumask_t mask) -{ - int i, cnt; - - cnt = 0; - for_each_cpu_mask(i, mask) - list[cnt++] = i; - - return cnt; -} - -static int update_cpu_list(u16 *list, int orig_cnt, cpumask_t mask) -{ - int i; - - for (i = 0; i < orig_cnt; i++) { - if (list[i] == 0xffff) - cpu_clear(i, mask); - } - - return init_cpu_list(list, mask); -} - static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask) { - int this_cpu = get_cpu(); - struct trap_per_cpu *tb = &trap_block[this_cpu]; - u64 *mondo = __va(tb->cpu_mondo_block_pa); - u16 *cpu_list = __va(tb->cpu_list_pa); - int cnt, retries; + struct trap_per_cpu *tb; + u16 *cpu_list; + u64 *mondo; + cpumask_t error_mask; + unsigned long flags, status; + int cnt, retries, this_cpu, i; + + /* We have to do this whole thing with interrupts fully disabled. + * Otherwise if we send an xcall from interrupt context it will + * corrupt both our mondo block and cpu list state. + * + * One consequence of this is that we cannot use timeout mechanisms + * that depend upon interrupts being delivered locally. So, for + * example, we cannot sample jiffies and expect it to advance. + * + * Fortunately, udelay() uses %stick/%tick so we can use that. + */ + local_irq_save(flags); + + this_cpu = smp_processor_id(); + tb = &trap_block[this_cpu]; + mondo = __va(tb->cpu_mondo_block_pa); mondo[0] = data0; mondo[1] = data1; mondo[2] = data2; wmb(); + cpu_list = __va(tb->cpu_list_pa); + + /* Setup the initial cpu list. */ + cnt = 0; + for_each_cpu_mask(i, mask) + cpu_list[cnt++] = i; + + cpus_clear(error_mask); retries = 0; - cnt = init_cpu_list(cpu_list, mask); do { - register unsigned long func __asm__("%o5"); - register unsigned long arg0 __asm__("%o0"); - register unsigned long arg1 __asm__("%o1"); - register unsigned long arg2 __asm__("%o2"); - - func = HV_FAST_CPU_MONDO_SEND; - arg0 = cnt; - arg1 = tb->cpu_list_pa; - arg2 = tb->cpu_mondo_block_pa; - - __asm__ __volatile__("ta %8" - : "=&r" (func), "=&r" (arg0), - "=&r" (arg1), "=&r" (arg2) - : "0" (func), "1" (arg0), - "2" (arg1), "3" (arg2), - "i" (HV_FAST_TRAP) - : "memory"); - if (likely(arg0 == HV_EOK)) - break; + int forward_progress; + + status = sun4v_cpu_mondo_send(cnt, + tb->cpu_list_pa, + tb->cpu_mondo_block_pa); - if (unlikely(++retries > 100)) { - printk("CPU[%d]: sun4v mondo error %lu\n", - this_cpu, arg0); + /* HV_EOK means all cpus received the xcall, we're done. */ + if (likely(status == HV_EOK)) break; + + /* First, clear out all the cpus in the mask that were + * successfully sent to. The hypervisor indicates this + * by setting the cpu list entry of such cpus to 0xffff. + */ + forward_progress = 0; + for (i = 0; i < cnt; i++) { + if (cpu_list[i] == 0xffff) { + cpu_clear(i, mask); + forward_progress = 1; + } } - cnt = update_cpu_list(cpu_list, cnt, mask); + /* If we get a HV_ECPUERROR, then one or more of the cpus + * in the list are in error state. Use the cpu_state() + * hypervisor call to find out which cpus are in error state. + */ + if (unlikely(status == HV_ECPUERROR)) { + for (i = 0; i < cnt; i++) { + long err; + u16 cpu; + + cpu = cpu_list[i]; + if (cpu == 0xffff) + continue; + + err = sun4v_cpu_state(cpu); + if (err >= 0 && + err == HV_CPU_STATE_ERROR) { + cpu_clear(cpu, mask); + cpu_set(cpu, error_mask); + } + } + } else if (unlikely(status != HV_EWOULDBLOCK)) + goto fatal_mondo_error; + + /* Rebuild the cpu_list[] array and try again. */ + cnt = 0; + for_each_cpu_mask(i, mask) + cpu_list[cnt++] = i; - udelay(2 * cnt); + if (unlikely(!forward_progress)) { + if (unlikely(++retries > 10000)) + goto fatal_mondo_timeout; + + /* Delay a little bit to let other cpus catch up + * on their cpu mondo queue work. + */ + udelay(2 * cnt); + } } while (1); - put_cpu(); + local_irq_restore(flags); + + if (unlikely(!cpus_empty(error_mask))) + goto fatal_mondo_cpu_error; + + return; + +fatal_mondo_cpu_error: + printk(KERN_CRIT "CPU[%d]: SUN4V mondo cpu error, some target cpus " + "were in error state\n", + this_cpu); + printk(KERN_CRIT "CPU[%d]: Error mask [ ", this_cpu); + for_each_cpu_mask(i, error_mask) + printk("%d ", i); + printk("]\n"); + return; + +fatal_mondo_timeout: + local_irq_restore(flags); + printk(KERN_CRIT "CPU[%d]: SUN4V mondo timeout, no forward " + " progress after %d retries.\n", + this_cpu, retries); + goto dump_cpu_list_and_out; + +fatal_mondo_error: + local_irq_restore(flags); + printk(KERN_CRIT "CPU[%d]: Unexpected SUN4V mondo error %lu\n", + this_cpu, status); + printk(KERN_CRIT "CPU[%d]: Args were cnt(%d) cpulist_pa(%lx) " + "mondo_block_pa(%lx)\n", + this_cpu, cnt, tb->cpu_list_pa, tb->cpu_mondo_block_pa); + +dump_cpu_list_and_out: + printk(KERN_CRIT "CPU[%d]: CPU list [ ", this_cpu); + for (i = 0; i < cnt; i++) + printk("%u ", cpu_list[i]); + printk("]\n"); } /* Send cross call to all processors mentioned in MASK @@ -723,9 +790,8 @@ static int smp_call_function_mask(void (*func)(void *info), void *info, out_timeout: spin_unlock(&call_lock); - printk("XCALL: Remote cpus not responding, ncpus=%ld finished=%ld\n", - (long) num_online_cpus() - 1L, - (long) atomic_read(&data.finished)); + printk("XCALL: Remote cpus not responding, ncpus=%d finished=%d\n", + cpus, atomic_read(&data.finished)); return 0; } diff --git a/include/asm-sparc64/hypervisor.h b/include/asm-sparc64/hypervisor.h index 726e2ea03c..612bf31975 100644 --- a/include/asm-sparc64/hypervisor.h +++ b/include/asm-sparc64/hypervisor.h @@ -342,6 +342,8 @@ extern unsigned long sun4v_cpu_qconf(unsigned long type, * ENOCPU Invalid cpu in CPU list * EWOULDBLOCK Some or all of the listed CPUs did not receive * the mondo + * ECPUERROR One or more of the listed CPUs are in error + * state, use HV_FAST_CPU_STATE to see which ones * EINVAL CPU list includes caller's CPU ID * * Send a mondo interrupt to the CPUs in the given CPU list with the @@ -355,6 +357,10 @@ extern unsigned long sun4v_cpu_qconf(unsigned long type, */ #define HV_FAST_CPU_MONDO_SEND 0x42 +#ifndef __ASSEMBLY__ +extern unsigned long sun4v_cpu_mondo_send(unsigned long cpu_count, unsigned long cpu_list_pa, unsigned long mondo_block_pa); +#endif + /* cpu_myid() * TRAP: HV_FAST_TRAP * FUNCTION: HV_FAST_CPU_MYID @@ -382,6 +388,10 @@ extern unsigned long sun4v_cpu_qconf(unsigned long type, #define HV_CPU_STATE_RUNNING 0x02 #define HV_CPU_STATE_ERROR 0x03 +#ifndef __ASSEMBLY__ +extern long sun4v_cpu_state(unsigned long cpuid); +#endif + /* cpu_set_rtba() * TRAP: HV_FAST_TRAP * FUNCTION: HV_FAST_CPU_SET_RTBA -- 2.39.5