kernel-ark/arch/sparc/kernel/smp_64.c

1547 lines
36 KiB
C
Raw Normal View History

/* smp.c: Sparc64 SMP support.
*
* Copyright (C) 1997, 2007, 2008 David S. Miller (davem@davemloft.net)
*/
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/mm.h>
#include <linux/pagemap.h>
#include <linux/threads.h>
#include <linux/smp.h>
#include <linux/interrupt.h>
#include <linux/kernel_stat.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/fs.h>
#include <linux/seq_file.h>
#include <linux/cache.h>
#include <linux/jiffies.h>
#include <linux/profile.h>
#include <linux/bootmem.h>
#include <linux/vmalloc.h>
#include <linux/cpu.h>
#include <asm/head.h>
#include <asm/ptrace.h>
#include <asm/atomic.h>
#include <asm/tlbflush.h>
#include <asm/mmu_context.h>
#include <asm/cpudata.h>
#include <asm/hvtramp.h>
#include <asm/io.h>
#include <asm/timer.h>
#include <asm/irq.h>
#include <asm/irq_regs.h>
#include <asm/page.h>
#include <asm/pgtable.h>
#include <asm/oplib.h>
#include <asm/uaccess.h>
#include <asm/starfire.h>
#include <asm/tlb.h>
[SPARC64]: Elminate all usage of hard-coded trap globals. UltraSPARC has special sets of global registers which are switched to for certain trap types. There is one set for MMU related traps, one set of Interrupt Vector processing, and another set (called the Alternate globals) for all other trap types. For what seems like forever we've hard coded the values in some of these trap registers. Some examples include: 1) Interrupt Vector global %g6 holds current processors interrupt work struct where received interrupts are managed for IRQ handler dispatch. 2) MMU global %g7 holds the base of the page tables of the currently active address space. 3) Alternate global %g6 held the current_thread_info() value. Such hardcoding has resulted in some serious issues in many areas. There are some code sequences where having another register available would help clean up the implementation. Taking traps such as cross-calls from the OBP firmware requires some trick code sequences wherein we have to save away and restore all of the special sets of global registers when we enter/exit OBP. We were also using the IMMU TSB register on SMP to hold the per-cpu area base address, which doesn't work any longer now that we actually use the TSB facility of the cpu. The implementation is pretty straight forward. One tricky bit is getting the current processor ID as that is different on different cpu variants. We use a stub with a fancy calling convention which we patch at boot time. The calling convention is that the stub is branched to and the (PC - 4) to return to is in register %g1. The cpu number is left in %g6. This stub can be invoked by using the __GET_CPUID macro. We use an array of per-cpu trap state to store the current thread and physical address of the current address space's page tables. The TRAP_LOAD_THREAD_REG loads %g6 with the current thread from this table, it uses __GET_CPUID and also clobbers %g1. TRAP_LOAD_IRQ_WORK is used by the interrupt vector processing to load the current processor's IRQ software state into %g6. It also uses __GET_CPUID and clobbers %g1. Finally, TRAP_LOAD_PGD_PHYS loads the physical address base of the current address space's page tables into %g7, it clobbers %g1 and uses __GET_CPUID. Many refinements are possible, as well as some tuning, with this stuff in place. Signed-off-by: David S. Miller <davem@davemloft.net>
2006-02-27 07:24:22 +00:00
#include <asm/sections.h>
#include <asm/prom.h>
#include <asm/mdesc.h>
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
#include <asm/ldc.h>
#include <asm/hypervisor.h>
sparc64: fix and optimize irq distribution irq_choose_cpu() should compare the affinity mask against cpu_online_map rather than CPU_MASK_ALL, since irq_select_affinity() sets the interrupt's affinity mask to cpu_online_map "and" CPU_MASK_ALL (which ends up being just cpu_online_map). The mask comparison in irq_choose_cpu() will always fail since the two masks are not the same. So the CPU chosen is the first CPU in the intersection of cpu_online_map and CPU_MASK_ALL, which is always CPU0. That means all interrupts are reassigned to CPU0... Distributing interrupts to CPUs in a linearly increasing round robin fashion is not optimal for the UltraSPARC T1/T2. Also, the irq_rover in irq_choose_cpu() causes an interrupt to be assigned to a different processor each time the interrupt is allocated and released. This may lead to an unbalanced distribution over time. A static mapping of interrupts to processors is done to optimize and balance interrupt distribution. For the T1/T2, interrupts are spread to different cores first, and then to strands within a core. The following is some benchmarks showing the effects of interrupt distribution on a T2. The test was done with iperf using a pair of T5220 boxes, each with a 10GBe NIU (XAUI) connected back to back. TCP | Stock Linear RR IRQ Optimized IRQ Streams | 2.6.30-rc5 Distribution Distribution | GBits/sec GBits/sec GBits/sec --------+----------------------------------------- 1 0.839 0.862 0.868 8 1.16 4.96 5.88 16 1.15 6.40 8.04 100 1.09 7.28 8.68 Signed-off-by: Hong H. Pham <hong.pham@windriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-06-04 09:10:11 +00:00
#include "cpumap.h"
int sparc64_multi_core __read_mostly;
DEFINE_PER_CPU(cpumask_t, cpu_sibling_map) = CPU_MASK_NONE;
cpumask_t cpu_core_map[NR_CPUS] __read_mostly =
{ [0 ... NR_CPUS-1] = CPU_MASK_NONE };
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
EXPORT_SYMBOL(cpu_core_map);
static cpumask_t smp_commenced_mask;
void smp_info(struct seq_file *m)
{
int i;
seq_printf(m, "State:\n");
for_each_online_cpu(i)
seq_printf(m, "CPU%d:\t\tonline\n", i);
}
void smp_bogo(struct seq_file *m)
{
int i;
for_each_online_cpu(i)
seq_printf(m,
"Cpu%dClkTck\t: %016lx\n",
i, cpu_data(i).clock_tick);
}
extern void setup_sparc64_timer(void);
static volatile unsigned long callin_flag = 0;
void __cpuinit smp_callin(void)
{
int cpuid = hard_smp_processor_id();
[SPARC64]: Elminate all usage of hard-coded trap globals. UltraSPARC has special sets of global registers which are switched to for certain trap types. There is one set for MMU related traps, one set of Interrupt Vector processing, and another set (called the Alternate globals) for all other trap types. For what seems like forever we've hard coded the values in some of these trap registers. Some examples include: 1) Interrupt Vector global %g6 holds current processors interrupt work struct where received interrupts are managed for IRQ handler dispatch. 2) MMU global %g7 holds the base of the page tables of the currently active address space. 3) Alternate global %g6 held the current_thread_info() value. Such hardcoding has resulted in some serious issues in many areas. There are some code sequences where having another register available would help clean up the implementation. Taking traps such as cross-calls from the OBP firmware requires some trick code sequences wherein we have to save away and restore all of the special sets of global registers when we enter/exit OBP. We were also using the IMMU TSB register on SMP to hold the per-cpu area base address, which doesn't work any longer now that we actually use the TSB facility of the cpu. The implementation is pretty straight forward. One tricky bit is getting the current processor ID as that is different on different cpu variants. We use a stub with a fancy calling convention which we patch at boot time. The calling convention is that the stub is branched to and the (PC - 4) to return to is in register %g1. The cpu number is left in %g6. This stub can be invoked by using the __GET_CPUID macro. We use an array of per-cpu trap state to store the current thread and physical address of the current address space's page tables. The TRAP_LOAD_THREAD_REG loads %g6 with the current thread from this table, it uses __GET_CPUID and also clobbers %g1. TRAP_LOAD_IRQ_WORK is used by the interrupt vector processing to load the current processor's IRQ software state into %g6. It also uses __GET_CPUID and clobbers %g1. Finally, TRAP_LOAD_PGD_PHYS loads the physical address base of the current address space's page tables into %g7, it clobbers %g1 and uses __GET_CPUID. Many refinements are possible, as well as some tuning, with this stuff in place. Signed-off-by: David S. Miller <davem@davemloft.net>
2006-02-27 07:24:22 +00:00
__local_per_cpu_offset = __per_cpu_offset(cpuid);
if (tlb_type == hypervisor)
sun4v_ktsb_register();
[SPARC64]: Elminate all usage of hard-coded trap globals. UltraSPARC has special sets of global registers which are switched to for certain trap types. There is one set for MMU related traps, one set of Interrupt Vector processing, and another set (called the Alternate globals) for all other trap types. For what seems like forever we've hard coded the values in some of these trap registers. Some examples include: 1) Interrupt Vector global %g6 holds current processors interrupt work struct where received interrupts are managed for IRQ handler dispatch. 2) MMU global %g7 holds the base of the page tables of the currently active address space. 3) Alternate global %g6 held the current_thread_info() value. Such hardcoding has resulted in some serious issues in many areas. There are some code sequences where having another register available would help clean up the implementation. Taking traps such as cross-calls from the OBP firmware requires some trick code sequences wherein we have to save away and restore all of the special sets of global registers when we enter/exit OBP. We were also using the IMMU TSB register on SMP to hold the per-cpu area base address, which doesn't work any longer now that we actually use the TSB facility of the cpu. The implementation is pretty straight forward. One tricky bit is getting the current processor ID as that is different on different cpu variants. We use a stub with a fancy calling convention which we patch at boot time. The calling convention is that the stub is branched to and the (PC - 4) to return to is in register %g1. The cpu number is left in %g6. This stub can be invoked by using the __GET_CPUID macro. We use an array of per-cpu trap state to store the current thread and physical address of the current address space's page tables. The TRAP_LOAD_THREAD_REG loads %g6 with the current thread from this table, it uses __GET_CPUID and also clobbers %g1. TRAP_LOAD_IRQ_WORK is used by the interrupt vector processing to load the current processor's IRQ software state into %g6. It also uses __GET_CPUID and clobbers %g1. Finally, TRAP_LOAD_PGD_PHYS loads the physical address base of the current address space's page tables into %g7, it clobbers %g1 and uses __GET_CPUID. Many refinements are possible, as well as some tuning, with this stuff in place. Signed-off-by: David S. Miller <davem@davemloft.net>
2006-02-27 07:24:22 +00:00
__flush_tlb_all();
setup_sparc64_timer();
if (cheetah_pcache_forced_on)
cheetah_enable_pcache();
local_irq_enable();
callin_flag = 1;
__asm__ __volatile__("membar #Sync\n\t"
"flush %%g6" : : : "memory");
/* Clear this or we will die instantly when we
* schedule back to this idler...
*/
current_thread_info()->new_child = 0;
/* Attach to the address space of init_task. */
atomic_inc(&init_mm.mm_count);
current->active_mm = &init_mm;
/* inform the notifiers about the new cpu */
notify_cpu_starting(cpuid);
while (!cpu_isset(cpuid, smp_commenced_mask))
rmb();
ipi_call_lock_irq();
cpu_set(cpuid, cpu_online_map);
ipi_call_unlock_irq();
/* idle thread is expected to have preempt disabled */
preempt_disable();
}
void cpu_panic(void)
{
printk("CPU[%d]: Returns from cpu_idle!\n", smp_processor_id());
panic("SMP bolixed\n");
}
/* This tick register synchronization scheme is taken entirely from
* the ia64 port, see arch/ia64/kernel/smpboot.c for details and credit.
*
* The only change I've made is to rework it so that the master
* initiates the synchonization instead of the slave. -DaveM
*/
#define MASTER 0
#define SLAVE (SMP_CACHE_BYTES/sizeof(unsigned long))
#define NUM_ROUNDS 64 /* magic value */
#define NUM_ITERS 5 /* likewise */
static DEFINE_SPINLOCK(itc_sync_lock);
static unsigned long go[SLAVE + 1];
#define DEBUG_TICK_SYNC 0
static inline long get_delta (long *rt, long *master)
{
unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
unsigned long tcenter, t0, t1, tm;
unsigned long i;
for (i = 0; i < NUM_ITERS; i++) {
t0 = tick_ops->get_tick();
go[MASTER] = 1;
membar_safe("#StoreLoad");
while (!(tm = go[SLAVE]))
rmb();
go[SLAVE] = 0;
wmb();
t1 = tick_ops->get_tick();
if (t1 - t0 < best_t1 - best_t0)
best_t0 = t0, best_t1 = t1, best_tm = tm;
}
*rt = best_t1 - best_t0;
*master = best_tm - best_t0;
/* average best_t0 and best_t1 without overflow: */
tcenter = (best_t0/2 + best_t1/2);
if (best_t0 % 2 + best_t1 % 2 == 2)
tcenter++;
return tcenter - best_tm;
}
void smp_synchronize_tick_client(void)
{
long i, delta, adj, adjust_latency = 0, done = 0;
unsigned long flags, rt, master_time_stamp, bound;
#if DEBUG_TICK_SYNC
struct {
long rt; /* roundtrip time */
long master; /* master's timestamp */
long diff; /* difference between midpoint and master's timestamp */
long lat; /* estimate of itc adjustment latency */
} t[NUM_ROUNDS];
#endif
go[MASTER] = 1;
while (go[MASTER])
rmb();
local_irq_save(flags);
{
for (i = 0; i < NUM_ROUNDS; i++) {
delta = get_delta(&rt, &master_time_stamp);
if (delta == 0) {
done = 1; /* let's lock on to this... */
bound = rt;
}
if (!done) {
if (i > 0) {
adjust_latency += -delta;
adj = -delta + adjust_latency/4;
} else
adj = -delta;
tick_ops->add_tick(adj);
}
#if DEBUG_TICK_SYNC
t[i].rt = rt;
t[i].master = master_time_stamp;
t[i].diff = delta;
t[i].lat = adjust_latency/4;
#endif
}
}
local_irq_restore(flags);
#if DEBUG_TICK_SYNC
for (i = 0; i < NUM_ROUNDS; i++)
printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
t[i].rt, t[i].master, t[i].diff, t[i].lat);
#endif
printk(KERN_INFO "CPU %d: synchronized TICK with master CPU "
"(last diff %ld cycles, maxerr %lu cycles)\n",
smp_processor_id(), delta, rt);
}
static void smp_start_sync_tick_client(int cpu);
static void smp_synchronize_one_tick(int cpu)
{
unsigned long flags, i;
go[MASTER] = 0;
smp_start_sync_tick_client(cpu);
/* wait for client to be ready */
while (!go[MASTER])
rmb();
/* now let the client proceed into his loop */
go[MASTER] = 0;
membar_safe("#StoreLoad");
spin_lock_irqsave(&itc_sync_lock, flags);
{
for (i = 0; i < NUM_ROUNDS*NUM_ITERS; i++) {
while (!go[MASTER])
rmb();
go[MASTER] = 0;
wmb();
go[SLAVE] = tick_ops->get_tick();
membar_safe("#StoreLoad");
}
}
spin_unlock_irqrestore(&itc_sync_lock, flags);
}
#if defined(CONFIG_SUN_LDOMS) && defined(CONFIG_HOTPLUG_CPU)
/* XXX Put this in some common place. XXX */
static unsigned long kimage_addr_to_ra(void *p)
{
unsigned long val = (unsigned long) p;
return kern_base + (val - KERNBASE);
}
static void __cpuinit ldom_startcpu_cpuid(unsigned int cpu, unsigned long thread_reg, void **descrp)
{
extern unsigned long sparc64_ttable_tl0;
extern unsigned long kern_locked_tte_data;
struct hvtramp_descr *hdesc;
unsigned long trampoline_ra;
struct trap_per_cpu *tb;
u64 tte_vaddr, tte_data;
unsigned long hv_err;
int i;
hdesc = kzalloc(sizeof(*hdesc) +
(sizeof(struct hvtramp_mapping) *
num_kernel_image_mappings - 1),
GFP_KERNEL);
if (!hdesc) {
printk(KERN_ERR "ldom_startcpu_cpuid: Cannot allocate "
"hvtramp_descr.\n");
return;
}
*descrp = hdesc;
hdesc->cpu = cpu;
hdesc->num_mappings = num_kernel_image_mappings;
tb = &trap_block[cpu];
hdesc->fault_info_va = (unsigned long) &tb->fault_info;
hdesc->fault_info_pa = kimage_addr_to_ra(&tb->fault_info);
hdesc->thread_reg = thread_reg;
tte_vaddr = (unsigned long) KERNBASE;
tte_data = kern_locked_tte_data;
for (i = 0; i < hdesc->num_mappings; i++) {
hdesc->maps[i].vaddr = tte_vaddr;
hdesc->maps[i].tte = tte_data;
tte_vaddr += 0x400000;
tte_data += 0x400000;
}
trampoline_ra = kimage_addr_to_ra(hv_cpu_startup);
hv_err = sun4v_cpu_start(cpu, trampoline_ra,
kimage_addr_to_ra(&sparc64_ttable_tl0),
__pa(hdesc));
if (hv_err)
printk(KERN_ERR "ldom_startcpu_cpuid: sun4v_cpu_start() "
"gives error %lu\n", hv_err);
}
#endif
extern unsigned long sparc64_cpu_startup;
/* The OBP cpu startup callback truncates the 3rd arg cookie to
* 32-bits (I think) so to be safe we have it read the pointer
* contained here so we work on >4GB machines. -DaveM
*/
static struct thread_info *cpu_new_thread = NULL;
static int __cpuinit smp_boot_one_cpu(unsigned int cpu)
{
unsigned long entry =
(unsigned long)(&sparc64_cpu_startup);
unsigned long cookie =
(unsigned long)(&cpu_new_thread);
struct task_struct *p;
void *descr = NULL;
int timeout, ret;
p = fork_idle(cpu);
if (IS_ERR(p))
return PTR_ERR(p);
callin_flag = 0;
cpu_new_thread = task_thread_info(p);
if (tlb_type == hypervisor) {
#if defined(CONFIG_SUN_LDOMS) && defined(CONFIG_HOTPLUG_CPU)
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
if (ldom_domaining_enabled)
ldom_startcpu_cpuid(cpu,
(unsigned long) cpu_new_thread,
&descr);
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
else
#endif
prom_startcpu_cpuid(cpu, entry, cookie);
} else {
struct device_node *dp = of_find_node_by_cpuid(cpu);
prom_startcpu(dp->node, entry, cookie);
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
for (timeout = 0; timeout < 50000; timeout++) {
if (callin_flag)
break;
udelay(100);
}
[SPARC64]: Get SUN4V SMP working. The sibling cpu bringup is extremely fragile. We can only perform the most basic calls until we take over the trap table from the firmware/hypervisor on the new cpu. This means no accesses to %g4, %g5, %g6 since those can't be TLB translated without our trap handlers. In order to achieve this: 1) Change sun4v_init_mondo_queues() so that it can operate in several modes. It can allocate the queues, or install them in the current processor, or both. The boot cpu does both in it's call early on. Later, the boot cpu allocates the sibling cpu queue, starts the sibling cpu, then the sibling cpu loads them in. 2) init_cur_cpu_trap() is changed to take the current_thread_info() as an argument instead of reading %g6 directly on the current cpu. 3) Create a trampoline stack for the sibling cpus. We do our basic kernel calls using this stack, which is locked into the kernel image, then go to our proper thread stack after taking over the trap table. 4) While we are in this delicate startup state, we put 0xdeadbeef into %g4/%g5/%g6 in order to catch accidental accesses. 5) On the final prom_set_trap_table*() call, we put &init_thread_union into %g6. This is a hack to make prom_world(0) work. All that wants to do is restore the %asi register using get_thread_current_ds(). Longer term we should just do the OBP calls to set the trap table by hand just like we do for everything else. This would avoid that silly prom_world(0) issue, then we can remove the init_thread_union hack. Signed-off-by: David S. Miller <davem@davemloft.net>
2006-02-17 09:29:17 +00:00
if (callin_flag) {
ret = 0;
} else {
printk("Processor %d is stuck.\n", cpu);
ret = -ENODEV;
}
cpu_new_thread = NULL;
kfree(descr);
return ret;
}
static void spitfire_xcall_helper(u64 data0, u64 data1, u64 data2, u64 pstate, unsigned long cpu)
{
u64 result, target;
int stuck, tmp;
if (this_is_starfire) {
/* map to real upaid */
cpu = (((cpu & 0x3c) << 1) |
((cpu & 0x40) >> 4) |
(cpu & 0x3));
}
target = (cpu << 14) | 0x70;
again:
/* Ok, this is the real Spitfire Errata #54.
* One must read back from a UDB internal register
* after writes to the UDB interrupt dispatch, but
* before the membar Sync for that write.
* So we use the high UDB control register (ASI 0x7f,
* ADDR 0x20) for the dummy read. -DaveM
*/
tmp = 0x40;
__asm__ __volatile__(
"wrpr %1, %2, %%pstate\n\t"
"stxa %4, [%0] %3\n\t"
"stxa %5, [%0+%8] %3\n\t"
"add %0, %8, %0\n\t"
"stxa %6, [%0+%8] %3\n\t"
"membar #Sync\n\t"
"stxa %%g0, [%7] %3\n\t"
"membar #Sync\n\t"
"mov 0x20, %%g1\n\t"
"ldxa [%%g1] 0x7f, %%g0\n\t"
"membar #Sync"
: "=r" (tmp)
: "r" (pstate), "i" (PSTATE_IE), "i" (ASI_INTR_W),
"r" (data0), "r" (data1), "r" (data2), "r" (target),
"r" (0x10), "0" (tmp)
: "g1");
/* NOTE: PSTATE_IE is still clear. */
stuck = 100000;
do {
__asm__ __volatile__("ldxa [%%g0] %1, %0"
: "=r" (result)
: "i" (ASI_INTR_DISPATCH_STAT));
if (result == 0) {
__asm__ __volatile__("wrpr %0, 0x0, %%pstate"
: : "r" (pstate));
return;
}
stuck -= 1;
if (stuck == 0)
break;
} while (result & 0x1);
__asm__ __volatile__("wrpr %0, 0x0, %%pstate"
: : "r" (pstate));
if (stuck == 0) {
printk("CPU[%d]: mondo stuckage result[%016llx]\n",
smp_processor_id(), result);
} else {
udelay(2);
goto again;
}
}
static void spitfire_xcall_deliver(struct trap_per_cpu *tb, int cnt)
{
u64 *mondo, data0, data1, data2;
u16 *cpu_list;
u64 pstate;
int i;
__asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
cpu_list = __va(tb->cpu_list_pa);
mondo = __va(tb->cpu_mondo_block_pa);
data0 = mondo[0];
data1 = mondo[1];
data2 = mondo[2];
for (i = 0; i < cnt; i++)
spitfire_xcall_helper(data0, data1, data2, pstate, cpu_list[i]);
}
/* Cheetah now allows to send the whole 64-bytes of data in the interrupt
* packet, but we have no use for that. However we do take advantage of
* the new pipelining feature (ie. dispatch to multiple cpus simultaneously).
*/
static void cheetah_xcall_deliver(struct trap_per_cpu *tb, int cnt)
{
int nack_busy_id, is_jbus, need_more;
u64 *mondo, pstate, ver, busy_mask;
u16 *cpu_list;
cpu_list = __va(tb->cpu_list_pa);
mondo = __va(tb->cpu_mondo_block_pa);
/* Unfortunately, someone at Sun had the brilliant idea to make the
* busy/nack fields hard-coded by ITID number for this Ultra-III
* derivative processor.
*/
__asm__ ("rdpr %%ver, %0" : "=r" (ver));
is_jbus = ((ver >> 32) == __JALAPENO_ID ||
(ver >> 32) == __SERRANO_ID);
__asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
retry:
need_more = 0;
__asm__ __volatile__("wrpr %0, %1, %%pstate\n\t"
: : "r" (pstate), "i" (PSTATE_IE));
/* Setup the dispatch data registers. */
__asm__ __volatile__("stxa %0, [%3] %6\n\t"
"stxa %1, [%4] %6\n\t"
"stxa %2, [%5] %6\n\t"
"membar #Sync\n\t"
: /* no outputs */
: "r" (mondo[0]), "r" (mondo[1]), "r" (mondo[2]),
"r" (0x40), "r" (0x50), "r" (0x60),
"i" (ASI_INTR_W));
nack_busy_id = 0;
busy_mask = 0;
{
int i;
for (i = 0; i < cnt; i++) {
u64 target, nr;
nr = cpu_list[i];
if (nr == 0xffff)
continue;
target = (nr << 14) | 0x70;
if (is_jbus) {
busy_mask |= (0x1UL << (nr * 2));
} else {
target |= (nack_busy_id << 24);
busy_mask |= (0x1UL <<
(nack_busy_id * 2));
}
__asm__ __volatile__(
"stxa %%g0, [%0] %1\n\t"
"membar #Sync\n\t"
: /* no outputs */
: "r" (target), "i" (ASI_INTR_W));
nack_busy_id++;
if (nack_busy_id == 32) {
need_more = 1;
break;
}
}
}
/* Now, poll for completion. */
{
u64 dispatch_stat, nack_mask;
long stuck;
stuck = 100000 * nack_busy_id;
nack_mask = busy_mask << 1;
do {
__asm__ __volatile__("ldxa [%%g0] %1, %0"
: "=r" (dispatch_stat)
: "i" (ASI_INTR_DISPATCH_STAT));
if (!(dispatch_stat & (busy_mask | nack_mask))) {
__asm__ __volatile__("wrpr %0, 0x0, %%pstate"
: : "r" (pstate));
if (unlikely(need_more)) {
int i, this_cnt = 0;
for (i = 0; i < cnt; i++) {
if (cpu_list[i] == 0xffff)
continue;
cpu_list[i] = 0xffff;
this_cnt++;
if (this_cnt == 32)
break;
}
goto retry;
}
return;
}
if (!--stuck)
break;
} while (dispatch_stat & busy_mask);
__asm__ __volatile__("wrpr %0, 0x0, %%pstate"
: : "r" (pstate));
if (dispatch_stat & busy_mask) {
/* Busy bits will not clear, continue instead
* of freezing up on this cpu.
*/
printk("CPU[%d]: mondo stuckage result[%016llx]\n",
smp_processor_id(), dispatch_stat);
} else {
int i, this_busy_nack = 0;
/* Delay some random time with interrupts enabled
* to prevent deadlock.
*/
udelay(2 * nack_busy_id);
/* Clear out the mask bits for cpus which did not
* NACK us.
*/
for (i = 0; i < cnt; i++) {
u64 check_mask, nr;
nr = cpu_list[i];
if (nr == 0xffff)
continue;
if (is_jbus)
check_mask = (0x2UL << (2*nr));
else
check_mask = (0x2UL <<
this_busy_nack);
if ((dispatch_stat & check_mask) == 0)
cpu_list[i] = 0xffff;
this_busy_nack += 2;
if (this_busy_nack == 64)
break;
}
goto retry;
}
}
}
/* Multi-cpu list version. */
static void hypervisor_xcall_deliver(struct trap_per_cpu *tb, int cnt)
{
int retries, this_cpu, prev_sent, i, saw_cpu_error;
unsigned long status;
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
u16 *cpu_list;
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
this_cpu = smp_processor_id();
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
cpu_list = __va(tb->cpu_list_pa);
saw_cpu_error = 0;
retries = 0;
prev_sent = 0;
do {
int forward_progress, n_sent;
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
status = sun4v_cpu_mondo_send(cnt,
tb->cpu_list_pa,
tb->cpu_mondo_block_pa);
/* HV_EOK means all cpus received the xcall, we're done. */
if (likely(status == HV_EOK))
break;
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
/* First, see if we made any forward progress.
*
* The hypervisor indicates successful sends by setting
* cpu list entries to the value 0xffff.
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
*/
n_sent = 0;
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
for (i = 0; i < cnt; i++) {
if (likely(cpu_list[i] == 0xffff))
n_sent++;
}
forward_progress = 0;
if (n_sent > prev_sent)
forward_progress = 1;
prev_sent = n_sent;
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
/* 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 == HV_CPU_STATE_ERROR) {
saw_cpu_error = (cpu + 1);
cpu_list[i] = 0xffff;
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
}
}
} else if (unlikely(status != HV_EWOULDBLOCK))
goto fatal_mondo_error;
/* Don't bother rewriting the CPU list, just leave the
* 0xffff and non-0xffff entries in there and the
* hypervisor will do the right thing.
*
* Only advance timeout state if we didn't make any
* forward progress.
*/
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
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);
if (unlikely(saw_cpu_error))
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
goto fatal_mondo_cpu_error;
return;
fatal_mondo_cpu_error:
printk(KERN_CRIT "CPU[%d]: SUN4V mondo cpu error, some target cpus "
"(including %d) were in error state\n",
this_cpu, saw_cpu_error - 1);
[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 <davem@davemloft.net>
2006-02-28 23:10:26 +00:00
return;
fatal_mondo_timeout:
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:
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");
}
static void (*xcall_deliver_impl)(struct trap_per_cpu *, int);
static void xcall_deliver(u64 data0, u64 data1, u64 data2, const cpumask_t *mask)
{
struct trap_per_cpu *tb;
int this_cpu, i, cnt;
unsigned long flags;
u16 *cpu_list;
u64 *mondo;
/* 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(i, mask) {
if (i == this_cpu || !cpu_online(i))
continue;
cpu_list[cnt++] = i;
}
if (cnt)
xcall_deliver_impl(tb, cnt);
local_irq_restore(flags);
}
/* Send cross call to all processors mentioned in MASK_P
* except self. Really, there are only two cases currently,
* "&cpu_online_map" and "&mm->cpu_vm_mask".
*/
static void smp_cross_call_masked(unsigned long *func, u32 ctx, u64 data1, u64 data2, const cpumask_t *mask)
{
u64 data0 = (((u64)ctx)<<32 | (((u64)func) & 0xffffffff));
xcall_deliver(data0, data1, data2, mask);
}
/* Send cross call to all processors except self. */
static void smp_cross_call(unsigned long *func, u32 ctx, u64 data1, u64 data2)
{
smp_cross_call_masked(func, ctx, data1, data2, &cpu_online_map);
}
extern unsigned long xcall_sync_tick;
static void smp_start_sync_tick_client(int cpu)
{
xcall_deliver((u64) &xcall_sync_tick, 0, 0,
&cpumask_of_cpu(cpu));
}
extern unsigned long xcall_call_function;
void arch_send_call_function_ipi_mask(const struct cpumask *mask)
{
xcall_deliver((u64) &xcall_call_function, 0, 0, mask);
}
extern unsigned long xcall_call_function_single;
void arch_send_call_function_single_ipi(int cpu)
{
xcall_deliver((u64) &xcall_call_function_single, 0, 0,
&cpumask_of_cpu(cpu));
}
void smp_call_function_client(int irq, struct pt_regs *regs)
{
clear_softint(1 << irq);
generic_smp_call_function_interrupt();
}
void smp_call_function_single_client(int irq, struct pt_regs *regs)
{
clear_softint(1 << irq);
generic_smp_call_function_single_interrupt();
}
static void tsb_sync(void *info)
{
struct trap_per_cpu *tp = &trap_block[raw_smp_processor_id()];
struct mm_struct *mm = info;
/* It is not valid to test "currrent->active_mm == mm" here.
*
* The value of "current" is not changed atomically with
* switch_mm(). But that's OK, we just need to check the
* current cpu's trap block PGD physical address.
*/
if (tp->pgd_paddr == __pa(mm->pgd))
tsb_context_switch(mm);
}
void smp_tsb_sync(struct mm_struct *mm)
{
smp_call_function_many(mm_cpumask(mm), tsb_sync, mm, 1);
}
extern unsigned long xcall_flush_tlb_mm;
extern unsigned long xcall_flush_tlb_pending;
extern unsigned long xcall_flush_tlb_kernel_range;
extern unsigned long xcall_fetch_glob_regs;
extern unsigned long xcall_receive_signal;
extern unsigned long xcall_new_mmu_context_version;
#ifdef CONFIG_KGDB
extern unsigned long xcall_kgdb_capture;
#endif
#ifdef DCACHE_ALIASING_POSSIBLE
extern unsigned long xcall_flush_dcache_page_cheetah;
#endif
extern unsigned long xcall_flush_dcache_page_spitfire;
#ifdef CONFIG_DEBUG_DCFLUSH
extern atomic_t dcpage_flushes;
extern atomic_t dcpage_flushes_xcall;
#endif
static inline void __local_flush_dcache_page(struct page *page)
{
#ifdef DCACHE_ALIASING_POSSIBLE
__flush_dcache_page(page_address(page),
((tlb_type == spitfire) &&
page_mapping(page) != NULL));
#else
if (page_mapping(page) != NULL &&
tlb_type == spitfire)
__flush_icache_page(__pa(page_address(page)));
#endif
}
void smp_flush_dcache_page_impl(struct page *page, int cpu)
{
int this_cpu;
if (tlb_type == hypervisor)
return;
#ifdef CONFIG_DEBUG_DCFLUSH
atomic_inc(&dcpage_flushes);
#endif
this_cpu = get_cpu();
if (cpu == this_cpu) {
__local_flush_dcache_page(page);
} else if (cpu_online(cpu)) {
void *pg_addr = page_address(page);
u64 data0 = 0;
if (tlb_type == spitfire) {
data0 = ((u64)&xcall_flush_dcache_page_spitfire);
if (page_mapping(page) != NULL)
data0 |= ((u64)1 << 32);
} else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
#ifdef DCACHE_ALIASING_POSSIBLE
data0 = ((u64)&xcall_flush_dcache_page_cheetah);
#endif
}
if (data0) {
xcall_deliver(data0, __pa(pg_addr),
(u64) pg_addr, &cpumask_of_cpu(cpu));
#ifdef CONFIG_DEBUG_DCFLUSH
atomic_inc(&dcpage_flushes_xcall);
#endif
}
}
put_cpu();
}
void flush_dcache_page_all(struct mm_struct *mm, struct page *page)
{
void *pg_addr;
int this_cpu;
u64 data0;
if (tlb_type == hypervisor)
return;
this_cpu = get_cpu();
#ifdef CONFIG_DEBUG_DCFLUSH
atomic_inc(&dcpage_flushes);
#endif
data0 = 0;
pg_addr = page_address(page);
if (tlb_type == spitfire) {
data0 = ((u64)&xcall_flush_dcache_page_spitfire);
if (page_mapping(page) != NULL)
data0 |= ((u64)1 << 32);
} else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
#ifdef DCACHE_ALIASING_POSSIBLE
data0 = ((u64)&xcall_flush_dcache_page_cheetah);
#endif
}
if (data0) {
xcall_deliver(data0, __pa(pg_addr),
(u64) pg_addr, &cpu_online_map);
#ifdef CONFIG_DEBUG_DCFLUSH
atomic_inc(&dcpage_flushes_xcall);
#endif
}
__local_flush_dcache_page(page);
put_cpu();
}
void smp_new_mmu_context_version_client(int irq, struct pt_regs *regs)
{
struct mm_struct *mm;
unsigned long flags;
clear_softint(1 << irq);
/* See if we need to allocate a new TLB context because
* the version of the one we are using is now out of date.
*/
mm = current->active_mm;
if (unlikely(!mm || (mm == &init_mm)))
return;
spin_lock_irqsave(&mm->context.lock, flags);
if (unlikely(!CTX_VALID(mm->context)))
get_new_mmu_context(mm);
spin_unlock_irqrestore(&mm->context.lock, flags);
load_secondary_context(mm);
__flush_tlb_mm(CTX_HWBITS(mm->context),
SECONDARY_CONTEXT);
}
void smp_new_mmu_context_version(void)
{
smp_cross_call(&xcall_new_mmu_context_version, 0, 0, 0);
}
#ifdef CONFIG_KGDB
void kgdb_roundup_cpus(unsigned long flags)
{
smp_cross_call(&xcall_kgdb_capture, 0, 0, 0);
}
#endif
void smp_fetch_global_regs(void)
{
smp_cross_call(&xcall_fetch_glob_regs, 0, 0, 0);
}
/* We know that the window frames of the user have been flushed
* to the stack before we get here because all callers of us
* are flush_tlb_*() routines, and these run after flush_cache_*()
* which performs the flushw.
*
* The SMP TLB coherency scheme we use works as follows:
*
* 1) mm->cpu_vm_mask is a bit mask of which cpus an address
* space has (potentially) executed on, this is the heuristic
* we use to avoid doing cross calls.
*
* Also, for flushing from kswapd and also for clones, we
* use cpu_vm_mask as the list of cpus to make run the TLB.
*
* 2) TLB context numbers are shared globally across all processors
* in the system, this allows us to play several games to avoid
* cross calls.
*
* One invariant is that when a cpu switches to a process, and
* that processes tsk->active_mm->cpu_vm_mask does not have the
* current cpu's bit set, that tlb context is flushed locally.
*
* If the address space is non-shared (ie. mm->count == 1) we avoid
* cross calls when we want to flush the currently running process's
* tlb state. This is done by clearing all cpu bits except the current
sparc64: Fix MM refcount check in smp_flush_tlb_pending(). As explained by Benjamin Herrenschmidt: > CPU 0 is running the context, task->mm == task->active_mm == your > context. The CPU is in userspace happily churning things. > > CPU 1 used to run it, not anymore, it's now running fancyfsd which > is a kernel thread, but current->active_mm still points to that > same context. > > Because there's only one "real" user, mm_users is 1 (but mm_count is > elevated, it's just that the presence on CPU 1 as active_mm has no > effect on mm_count(). > > At this point, fancyfsd decides to invalidate a mapping currently mapped > by that context, for example because a networked file has changed > remotely or something like that, using unmap_mapping_ranges(). > > So CPU 1 goes into the zapping code, which eventually ends up calling > flush_tlb_pending(). Your test will succeed, as current->active_mm is > indeed the target mm for the flush, and mm_users is indeed 1. So you > will -not- send an IPI to the other CPU, and CPU 0 will continue happily > accessing the pages that should have been unmapped. To fix this problem, check ->mm instead of ->active_mm, and this means: > So if you test current->mm, you effectively account for mm_users == 1, > so the only way the mm can be active on another processor is as a lazy > mm for a kernel thread. So your test should work properly as long > as you don't have a HW that will do speculative TLB reloads into the > TLB on that other CPU (and even if you do, you flush-on-switch-in should > get rid of any crap here). And therefore we should be OK. Signed-off-by: David S. Miller <davem@davemloft.net>
2009-03-27 08:09:17 +00:00
* processor's in current->mm->cpu_vm_mask and performing the
* flush locally only. This will force any subsequent cpus which run
* this task to flush the context from the local tlb if the process
* migrates to another cpu (again).
*
* 3) For shared address spaces (threads) and swapping we bite the
* bullet for most cases and perform the cross call (but only to
* the cpus listed in cpu_vm_mask).
*
* The performance gain from "optimizing" away the cross call for threads is
* questionable (in theory the big win for threads is the massive sharing of
* address space state across processors).
*/
/* This currently is only used by the hugetlb arch pre-fault
* hook on UltraSPARC-III+ and later when changing the pagesize
* bits of the context register for an address space.
*/
void smp_flush_tlb_mm(struct mm_struct *mm)
{
u32 ctx = CTX_HWBITS(mm->context);
int cpu = get_cpu();
if (atomic_read(&mm->mm_users) == 1) {
cpumask_copy(mm_cpumask(mm), cpumask_of(cpu));
goto local_flush_and_out;
}
smp_cross_call_masked(&xcall_flush_tlb_mm,
ctx, 0, 0,
mm_cpumask(mm));
local_flush_and_out:
__flush_tlb_mm(ctx, SECONDARY_CONTEXT);
put_cpu();
}
void smp_flush_tlb_pending(struct mm_struct *mm, unsigned long nr, unsigned long *vaddrs)
{
u32 ctx = CTX_HWBITS(mm->context);
int cpu = get_cpu();
sparc64: Fix MM refcount check in smp_flush_tlb_pending(). As explained by Benjamin Herrenschmidt: > CPU 0 is running the context, task->mm == task->active_mm == your > context. The CPU is in userspace happily churning things. > > CPU 1 used to run it, not anymore, it's now running fancyfsd which > is a kernel thread, but current->active_mm still points to that > same context. > > Because there's only one "real" user, mm_users is 1 (but mm_count is > elevated, it's just that the presence on CPU 1 as active_mm has no > effect on mm_count(). > > At this point, fancyfsd decides to invalidate a mapping currently mapped > by that context, for example because a networked file has changed > remotely or something like that, using unmap_mapping_ranges(). > > So CPU 1 goes into the zapping code, which eventually ends up calling > flush_tlb_pending(). Your test will succeed, as current->active_mm is > indeed the target mm for the flush, and mm_users is indeed 1. So you > will -not- send an IPI to the other CPU, and CPU 0 will continue happily > accessing the pages that should have been unmapped. To fix this problem, check ->mm instead of ->active_mm, and this means: > So if you test current->mm, you effectively account for mm_users == 1, > so the only way the mm can be active on another processor is as a lazy > mm for a kernel thread. So your test should work properly as long > as you don't have a HW that will do speculative TLB reloads into the > TLB on that other CPU (and even if you do, you flush-on-switch-in should > get rid of any crap here). And therefore we should be OK. Signed-off-by: David S. Miller <davem@davemloft.net>
2009-03-27 08:09:17 +00:00
if (mm == current->mm && atomic_read(&mm->mm_users) == 1)
cpumask_copy(mm_cpumask(mm), cpumask_of(cpu));
else
smp_cross_call_masked(&xcall_flush_tlb_pending,
ctx, nr, (unsigned long) vaddrs,
mm_cpumask(mm));
__flush_tlb_pending(ctx, nr, vaddrs);
put_cpu();
}
void smp_flush_tlb_kernel_range(unsigned long start, unsigned long end)
{
start &= PAGE_MASK;
end = PAGE_ALIGN(end);
if (start != end) {
smp_cross_call(&xcall_flush_tlb_kernel_range,
0, start, end);
__flush_tlb_kernel_range(start, end);
}
}
/* CPU capture. */
/* #define CAPTURE_DEBUG */
extern unsigned long xcall_capture;
static atomic_t smp_capture_depth = ATOMIC_INIT(0);
static atomic_t smp_capture_registry = ATOMIC_INIT(0);
static unsigned long penguins_are_doing_time;
void smp_capture(void)
{
int result = atomic_add_ret(1, &smp_capture_depth);
if (result == 1) {
int ncpus = num_online_cpus();
#ifdef CAPTURE_DEBUG
printk("CPU[%d]: Sending penguins to jail...",
smp_processor_id());
#endif
penguins_are_doing_time = 1;
atomic_inc(&smp_capture_registry);
smp_cross_call(&xcall_capture, 0, 0, 0);
while (atomic_read(&smp_capture_registry) != ncpus)
rmb();
#ifdef CAPTURE_DEBUG
printk("done\n");
#endif
}
}
void smp_release(void)
{
if (atomic_dec_and_test(&smp_capture_depth)) {
#ifdef CAPTURE_DEBUG
printk("CPU[%d]: Giving pardon to "
"imprisoned penguins\n",
smp_processor_id());
#endif
penguins_are_doing_time = 0;
membar_safe("#StoreLoad");
atomic_dec(&smp_capture_registry);
}
}
/* Imprisoned penguins run with %pil == PIL_NORMAL_MAX, but PSTATE_IE
* set, so they can service tlb flush xcalls...
*/
extern void prom_world(int);
void smp_penguin_jailcell(int irq, struct pt_regs *regs)
{
clear_softint(1 << irq);
preempt_disable();
__asm__ __volatile__("flushw");
prom_world(1);
atomic_inc(&smp_capture_registry);
membar_safe("#StoreLoad");
while (penguins_are_doing_time)
rmb();
atomic_dec(&smp_capture_registry);
prom_world(0);
preempt_enable();
}
/* /proc/profile writes can call this, don't __init it please. */
int setup_profiling_timer(unsigned int multiplier)
{
return -EINVAL;
}
void __init smp_prepare_cpus(unsigned int max_cpus)
{
}
void __devinit smp_prepare_boot_cpu(void)
{
}
void __init smp_setup_processor_id(void)
{
if (tlb_type == spitfire)
xcall_deliver_impl = spitfire_xcall_deliver;
else if (tlb_type == cheetah || tlb_type == cheetah_plus)
xcall_deliver_impl = cheetah_xcall_deliver;
else
xcall_deliver_impl = hypervisor_xcall_deliver;
}
void __devinit smp_fill_in_sib_core_maps(void)
{
unsigned int i;
for_each_present_cpu(i) {
unsigned int j;
cpus_clear(cpu_core_map[i]);
if (cpu_data(i).core_id == 0) {
cpu_set(i, cpu_core_map[i]);
continue;
}
for_each_present_cpu(j) {
if (cpu_data(i).core_id ==
cpu_data(j).core_id)
cpu_set(j, cpu_core_map[i]);
}
}
for_each_present_cpu(i) {
unsigned int j;
cpus_clear(per_cpu(cpu_sibling_map, i));
if (cpu_data(i).proc_id == -1) {
cpu_set(i, per_cpu(cpu_sibling_map, i));
continue;
}
for_each_present_cpu(j) {
if (cpu_data(i).proc_id ==
cpu_data(j).proc_id)
cpu_set(j, per_cpu(cpu_sibling_map, i));
}
}
}
[PATCH] Change cpu_up and co from __devinit to __cpuinit Compiling the kernel with CONFIG_HOTPLUG = y and CONFIG_HOTPLUG_CPU = n with CONFIG_RELOCATABLE = y generates the following modpost warnings WARNING: vmlinux - Section mismatch: reference to .init.data: from .text between '_cpu_up' (at offset 0xc0141b7d) and 'cpu_up' WARNING: vmlinux - Section mismatch: reference to .init.data: from .text between '_cpu_up' (at offset 0xc0141b9c) and 'cpu_up' WARNING: vmlinux - Section mismatch: reference to .init.text:__cpu_up from .text between '_cpu_up' (at offset 0xc0141bd8) and 'cpu_up' WARNING: vmlinux - Section mismatch: reference to .init.data: from .text between '_cpu_up' (at offset 0xc0141c05) and 'cpu_up' WARNING: vmlinux - Section mismatch: reference to .init.data: from .text between '_cpu_up' (at offset 0xc0141c26) and 'cpu_up' WARNING: vmlinux - Section mismatch: reference to .init.data: from .text between '_cpu_up' (at offset 0xc0141c37) and 'cpu_up' This is because cpu_up, _cpu_up and __cpu_up (in some architectures) are defined as __devinit AND __cpu_up calls some __cpuinit functions. Since __cpuinit would map to __init with this kind of a configuration, we get a .text refering .init.data warning. This patch solves the problem by converting all of __cpu_up, _cpu_up and cpu_up from __devinit to __cpuinit. The approach is justified since the callers of cpu_up are either dependent on CONFIG_HOTPLUG_CPU or are of __init type. Thus when CONFIG_HOTPLUG_CPU=y, all these cpu up functions would land up in .text section, and when CONFIG_HOTPLUG_CPU=n, all these functions would land up in .init section. Tested on a i386 SMP machine running linux-2.6.20-rc3-mm1. Signed-off-by: Gautham R Shenoy <ego@in.ibm.com> Cc: Vivek Goyal <vgoyal@in.ibm.com> Cc: Mikael Starvik <starvik@axis.com> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: "David S. Miller" <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2007-01-11 07:15:34 +00:00
int __cpuinit __cpu_up(unsigned int cpu)
{
int ret = smp_boot_one_cpu(cpu);
if (!ret) {
cpu_set(cpu, smp_commenced_mask);
while (!cpu_isset(cpu, cpu_online_map))
mb();
if (!cpu_isset(cpu, cpu_online_map)) {
ret = -ENODEV;
} else {
/* On SUN4V, writes to %tick and %stick are
* not allowed.
*/
if (tlb_type != hypervisor)
smp_synchronize_one_tick(cpu);
}
}
return ret;
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
#ifdef CONFIG_HOTPLUG_CPU
void cpu_play_dead(void)
{
int cpu = smp_processor_id();
unsigned long pstate;
idle_task_exit();
if (tlb_type == hypervisor) {
struct trap_per_cpu *tb = &trap_block[cpu];
sun4v_cpu_qconf(HV_CPU_QUEUE_CPU_MONDO,
tb->cpu_mondo_pa, 0);
sun4v_cpu_qconf(HV_CPU_QUEUE_DEVICE_MONDO,
tb->dev_mondo_pa, 0);
sun4v_cpu_qconf(HV_CPU_QUEUE_RES_ERROR,
tb->resum_mondo_pa, 0);
sun4v_cpu_qconf(HV_CPU_QUEUE_NONRES_ERROR,
tb->nonresum_mondo_pa, 0);
}
cpu_clear(cpu, smp_commenced_mask);
membar_safe("#Sync");
local_irq_disable();
__asm__ __volatile__(
"rdpr %%pstate, %0\n\t"
"wrpr %0, %1, %%pstate"
: "=r" (pstate)
: "i" (PSTATE_IE));
while (1)
barrier();
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
int __cpu_disable(void)
{
int cpu = smp_processor_id();
cpuinfo_sparc *c;
int i;
for_each_cpu_mask(i, cpu_core_map[cpu])
cpu_clear(cpu, cpu_core_map[i]);
cpus_clear(cpu_core_map[cpu]);
for_each_cpu_mask(i, per_cpu(cpu_sibling_map, cpu))
cpu_clear(cpu, per_cpu(cpu_sibling_map, i));
cpus_clear(per_cpu(cpu_sibling_map, cpu));
c = &cpu_data(cpu);
c->core_id = 0;
c->proc_id = -1;
smp_wmb();
/* Make sure no interrupts point to this cpu. */
fixup_irqs();
local_irq_enable();
mdelay(1);
local_irq_disable();
ipi_call_lock();
cpu_clear(cpu, cpu_online_map);
ipi_call_unlock();
sparc64: fix and optimize irq distribution irq_choose_cpu() should compare the affinity mask against cpu_online_map rather than CPU_MASK_ALL, since irq_select_affinity() sets the interrupt's affinity mask to cpu_online_map "and" CPU_MASK_ALL (which ends up being just cpu_online_map). The mask comparison in irq_choose_cpu() will always fail since the two masks are not the same. So the CPU chosen is the first CPU in the intersection of cpu_online_map and CPU_MASK_ALL, which is always CPU0. That means all interrupts are reassigned to CPU0... Distributing interrupts to CPUs in a linearly increasing round robin fashion is not optimal for the UltraSPARC T1/T2. Also, the irq_rover in irq_choose_cpu() causes an interrupt to be assigned to a different processor each time the interrupt is allocated and released. This may lead to an unbalanced distribution over time. A static mapping of interrupts to processors is done to optimize and balance interrupt distribution. For the T1/T2, interrupts are spread to different cores first, and then to strands within a core. The following is some benchmarks showing the effects of interrupt distribution on a T2. The test was done with iperf using a pair of T5220 boxes, each with a 10GBe NIU (XAUI) connected back to back. TCP | Stock Linear RR IRQ Optimized IRQ Streams | 2.6.30-rc5 Distribution Distribution | GBits/sec GBits/sec GBits/sec --------+----------------------------------------- 1 0.839 0.862 0.868 8 1.16 4.96 5.88 16 1.15 6.40 8.04 100 1.09 7.28 8.68 Signed-off-by: Hong H. Pham <hong.pham@windriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-06-04 09:10:11 +00:00
cpu_map_rebuild();
return 0;
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
}
void __cpu_die(unsigned int cpu)
{
int i;
for (i = 0; i < 100; i++) {
smp_rmb();
if (!cpu_isset(cpu, smp_commenced_mask))
break;
msleep(100);
}
if (cpu_isset(cpu, smp_commenced_mask)) {
printk(KERN_ERR "CPU %u didn't die...\n", cpu);
} else {
#if defined(CONFIG_SUN_LDOMS)
unsigned long hv_err;
int limit = 100;
do {
hv_err = sun4v_cpu_stop(cpu);
if (hv_err == HV_EOK) {
cpu_clear(cpu, cpu_present_map);
break;
}
} while (--limit > 0);
if (limit <= 0) {
printk(KERN_ERR "sun4v_cpu_stop() fails err=%lu\n",
hv_err);
}
#endif
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-13 23:03:42 +00:00
}
#endif
void __init smp_cpus_done(unsigned int max_cpus)
{
}
void smp_send_reschedule(int cpu)
{
xcall_deliver((u64) &xcall_receive_signal, 0, 0,
&cpumask_of_cpu(cpu));
}
void smp_receive_signal_client(int irq, struct pt_regs *regs)
{
clear_softint(1 << irq);
}
/* This is a nop because we capture all other cpus
* anyways when making the PROM active.
*/
void smp_send_stop(void)
{
}
/**
* pcpu_alloc_bootmem - NUMA friendly alloc_bootmem wrapper for percpu
* @cpu: cpu to allocate for
* @size: size allocation in bytes
* @align: alignment
*
* Allocate @size bytes aligned at @align for cpu @cpu. This wrapper
* does the right thing for NUMA regardless of the current
* configuration.
*
* RETURNS:
* Pointer to the allocated area on success, NULL on failure.
*/
static void * __init pcpu_alloc_bootmem(unsigned int cpu, unsigned long size,
unsigned long align)
{
const unsigned long goal = __pa(MAX_DMA_ADDRESS);
#ifdef CONFIG_NEED_MULTIPLE_NODES
int node = cpu_to_node(cpu);
void *ptr;
if (!node_online(node) || !NODE_DATA(node)) {
ptr = __alloc_bootmem(size, align, goal);
pr_info("cpu %d has no node %d or node-local memory\n",
cpu, node);
pr_debug("per cpu data for cpu%d %lu bytes at %016lx\n",
cpu, size, __pa(ptr));
} else {
ptr = __alloc_bootmem_node(NODE_DATA(node),
size, align, goal);
pr_debug("per cpu data for cpu%d %lu bytes on node%d at "
"%016lx\n", cpu, size, node, __pa(ptr));
}
return ptr;
#else
return __alloc_bootmem(size, align, goal);
#endif
}
static size_t pcpur_size __initdata;
static void **pcpur_ptrs __initdata;
static struct page * __init pcpur_get_page(unsigned int cpu, int pageno)
{
size_t off = (size_t)pageno << PAGE_SHIFT;
if (off >= pcpur_size)
return NULL;
return virt_to_page(pcpur_ptrs[cpu] + off);
}
#define PCPU_CHUNK_SIZE (4UL * 1024UL * 1024UL)
static void __init pcpu_map_range(unsigned long start, unsigned long end,
struct page *page)
{
unsigned long pfn = page_to_pfn(page);
unsigned long pte_base;
BUG_ON((pfn<<PAGE_SHIFT)&(PCPU_CHUNK_SIZE - 1UL));
pte_base = (_PAGE_VALID | _PAGE_SZ4MB_4U |
_PAGE_CP_4U | _PAGE_CV_4U |
_PAGE_P_4U | _PAGE_W_4U);
if (tlb_type == hypervisor)
pte_base = (_PAGE_VALID | _PAGE_SZ4MB_4V |
_PAGE_CP_4V | _PAGE_CV_4V |
_PAGE_P_4V | _PAGE_W_4V);
while (start < end) {
pgd_t *pgd = pgd_offset_k(start);
unsigned long this_end;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
pud = pud_offset(pgd, start);
if (pud_none(*pud)) {
pmd_t *new;
new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
pud_populate(&init_mm, pud, new);
}
pmd = pmd_offset(pud, start);
if (!pmd_present(*pmd)) {
pte_t *new;
new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
pmd_populate_kernel(&init_mm, pmd, new);
}
pte = pte_offset_kernel(pmd, start);
this_end = (start + PMD_SIZE) & PMD_MASK;
if (this_end > end)
this_end = end;
while (start < this_end) {
unsigned long paddr = pfn << PAGE_SHIFT;
pte_val(*pte) = (paddr | pte_base);
start += PAGE_SIZE;
pte++;
pfn++;
}
}
}
void __init setup_per_cpu_areas(void)
{
size_t dyn_size, static_size = __per_cpu_end - __per_cpu_start;
static struct vm_struct vm;
unsigned long delta, cpu;
size_t pcpu_unit_size;
size_t ptrs_size;
pcpur_size = PFN_ALIGN(static_size + PERCPU_MODULE_RESERVE +
PERCPU_DYNAMIC_RESERVE);
dyn_size = pcpur_size - static_size - PERCPU_MODULE_RESERVE;
ptrs_size = PFN_ALIGN(num_possible_cpus() * sizeof(pcpur_ptrs[0]));
pcpur_ptrs = alloc_bootmem(ptrs_size);
for_each_possible_cpu(cpu) {
pcpur_ptrs[cpu] = pcpu_alloc_bootmem(cpu, PCPU_CHUNK_SIZE,
PCPU_CHUNK_SIZE);
free_bootmem(__pa(pcpur_ptrs[cpu] + pcpur_size),
PCPU_CHUNK_SIZE - pcpur_size);
memcpy(pcpur_ptrs[cpu], __per_cpu_load, static_size);
}
/* allocate address and map */
vm.flags = VM_ALLOC;
vm.size = num_possible_cpus() * PCPU_CHUNK_SIZE;
vm_area_register_early(&vm, PCPU_CHUNK_SIZE);
for_each_possible_cpu(cpu) {
unsigned long start = (unsigned long) vm.addr;
unsigned long end;
start += cpu * PCPU_CHUNK_SIZE;
end = start + PCPU_CHUNK_SIZE;
pcpu_map_range(start, end, virt_to_page(pcpur_ptrs[cpu]));
}
pcpu_unit_size = pcpu_setup_first_chunk(pcpur_get_page, static_size,
PERCPU_MODULE_RESERVE, dyn_size,
PCPU_CHUNK_SIZE, vm.addr, NULL);
free_bootmem(__pa(pcpur_ptrs), ptrs_size);
delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
for_each_possible_cpu(cpu) {
__per_cpu_offset(cpu) = delta + cpu * pcpu_unit_size;
}
/* Setup %g5 for the boot cpu. */
__local_per_cpu_offset = __per_cpu_offset(smp_processor_id());
of_fill_in_cpu_data();
if (tlb_type == hypervisor)
mdesc_fill_in_cpu_data(CPU_MASK_ALL_PTR);
}