aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2008-08-04 19:04:36 -0700
committerLinus Torvalds <torvalds@linux-foundation.org>2008-08-04 19:04:36 -0700
commite9ba9698187ddbc0c5bfcf41de0349a662d23d02 (patch)
treeb11f8658d0bb05fe8bd826de1ead328d44e84005
parent2e1e9212ed8c532c6b324de77d3cafef5d2bc846 (diff)
parentae583885bfd07474789059cdef399289bd66c8d0 (diff)
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6: sparc64: Remove all cpumask_t local variables in xcall dispatch. sparc64: Kill error_mask from hypervisor_xcall_deliver(). sparc64: Build cpu list and mondo block at top-level xcall_deliver(). sparc64: Disable local interrupts around xcall_deliver_impl() invocation. sparc64: Make all xcall_deliver's go through common helper function. sparc64: Always allocate the send mondo blocks, even on non-sun4v. sparc64: Make smp_cross_call_masked() take a cpumask_t pointer. sparc64: Directly call xcall_deliver() in smp_start_sync_tick_client. sparc64: Call xcall_deliver() directly in some cases. sparc64: Use cpumask_t pointers and for_each_cpu_mask_nr() in xcall_deliver. sparc64: Use xcall_deliver() consistently. sparc64: Use function pointer for cross-call sending. arch/sparc64/kernel/signal.c: removed duplicated #include sparc64: Need to disable preemption around smp_tsb_sync().
-rw-r--r--arch/sparc64/kernel/irq.c19
-rw-r--r--arch/sparc64/kernel/signal.c1
-rw-r--r--arch/sparc64/kernel/smp.c292
-rw-r--r--arch/sparc64/mm/tsb.c5
4 files changed, 166 insertions, 151 deletions
diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c
index c481673d249..ba43d85e8dd 100644
--- a/arch/sparc64/kernel/irq.c
+++ b/arch/sparc64/kernel/irq.c
@@ -915,12 +915,18 @@ static void __init sun4v_init_mondo_queues(void)
alloc_one_mondo(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
alloc_one_kbuf(&tb->nonresum_kernel_buf_pa,
tb->nonresum_qmask);
+ }
+}
+
+static void __init init_send_mondo_info(void)
+{
+ int cpu;
+
+ for_each_possible_cpu(cpu) {
+ struct trap_per_cpu *tb = &trap_block[cpu];
init_cpu_send_mondo_info(tb);
}
-
- /* Load up the boot cpu's entries. */
- sun4v_register_mondo_queues(hard_smp_processor_id());
}
static struct irqaction timer_irq_action = {
@@ -949,6 +955,13 @@ void __init init_IRQ(void)
if (tlb_type == hypervisor)
sun4v_init_mondo_queues();
+ init_send_mondo_info();
+
+ if (tlb_type == hypervisor) {
+ /* Load up the boot cpu's entries. */
+ sun4v_register_mondo_queues(hard_smp_processor_id());
+ }
+
/* We need to clear any IRQ's pending in the soft interrupt
* registers, a spurious one could be left around from the
* PROM timer which we just disabled.
diff --git a/arch/sparc64/kernel/signal.c b/arch/sparc64/kernel/signal.c
index ca5a6ae3a6e..ec82d76dc6f 100644
--- a/arch/sparc64/kernel/signal.c
+++ b/arch/sparc64/kernel/signal.c
@@ -23,7 +23,6 @@
#include <linux/tty.h>
#include <linux/binfmts.h>
#include <linux/bitops.h>
-#include <linux/tracehook.h>
#include <asm/uaccess.h>
#include <asm/ptrace.h>
diff --git a/arch/sparc64/kernel/smp.c b/arch/sparc64/kernel/smp.c
index 340842e51ce..27b81775a4d 100644
--- a/arch/sparc64/kernel/smp.c
+++ b/arch/sparc64/kernel/smp.c
@@ -459,27 +459,35 @@ again:
}
}
-static inline void spitfire_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask)
+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));
- for_each_cpu_mask(i, mask)
- spitfire_xcall_helper(data0, data1, data2, pstate, i);
+ 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(u64 data0, u64 data1, u64 data2, cpumask_t mask)
+static void cheetah_xcall_deliver(struct trap_per_cpu *tb, int cnt)
{
- u64 pstate, ver, busy_mask;
int nack_busy_id, is_jbus, need_more;
+ u64 *mondo, pstate, ver, busy_mask;
+ u16 *cpu_list;
- if (cpus_empty(mask))
- return;
+ 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
@@ -502,7 +510,7 @@ retry:
"stxa %2, [%5] %6\n\t"
"membar #Sync\n\t"
: /* no outputs */
- : "r" (data0), "r" (data1), "r" (data2),
+ : "r" (mondo[0]), "r" (mondo[1]), "r" (mondo[2]),
"r" (0x40), "r" (0x50), "r" (0x60),
"i" (ASI_INTR_W));
@@ -511,11 +519,16 @@ retry:
{
int i;
- for_each_cpu_mask(i, mask) {
- u64 target = (i << 14) | 0x70;
+ 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 << (i * 2));
+ busy_mask |= (0x1UL << (nr * 2));
} else {
target |= (nack_busy_id << 24);
busy_mask |= (0x1UL <<
@@ -549,11 +562,13 @@ retry:
__asm__ __volatile__("wrpr %0, 0x0, %%pstate"
: : "r" (pstate));
if (unlikely(need_more)) {
- int i, cnt = 0;
- for_each_cpu_mask(i, mask) {
- cpu_clear(i, mask);
- cnt++;
- if (cnt == 32)
+ 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;
@@ -584,16 +599,20 @@ retry:
/* Clear out the mask bits for cpus which did not
* NACK us.
*/
- for_each_cpu_mask(i, mask) {
- u64 check_mask;
+ 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*i));
+ check_mask = (0x2UL << (2*nr));
else
check_mask = (0x2UL <<
this_busy_nack);
if ((dispatch_stat & check_mask) == 0)
- cpu_clear(i, mask);
+ cpu_list[i] = 0xffff;
this_busy_nack += 2;
if (this_busy_nack == 64)
break;
@@ -605,47 +624,17 @@ retry:
}
/* Multi-cpu list version. */
-static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask)
+static void hypervisor_xcall_deliver(struct trap_per_cpu *tb, int cnt)
{
- struct trap_per_cpu *tb;
+ int retries, this_cpu, prev_sent, i, saw_cpu_error;
+ unsigned long status;
u16 *cpu_list;
- u64 *mondo;
- cpumask_t error_mask;
- unsigned long flags, status;
- int cnt, retries, this_cpu, prev_sent, i;
-
- if (cpus_empty(mask))
- return;
-
- /* We have to do this whole thing with interrupts fully disabled.
- * Otherwise if we send an xcall from interrupt context it will
- * corrupt both our mondo block and cpu list state.
- *
- * One consequence of this is that we cannot use timeout mechanisms
- * that depend upon interrupts being delivered locally. So, for
- * example, we cannot sample jiffies and expect it to advance.
- *
- * Fortunately, udelay() uses %stick/%tick so we can use that.
- */
- local_irq_save(flags);
this_cpu = smp_processor_id();
- tb = &trap_block[this_cpu];
-
- mondo = __va(tb->cpu_mondo_block_pa);
- mondo[0] = data0;
- mondo[1] = data1;
- mondo[2] = data2;
- wmb();
cpu_list = __va(tb->cpu_list_pa);
- /* Setup the initial cpu list. */
- cnt = 0;
- for_each_cpu_mask(i, mask)
- cpu_list[cnt++] = i;
-
- cpus_clear(error_mask);
+ saw_cpu_error = 0;
retries = 0;
prev_sent = 0;
do {
@@ -690,10 +679,9 @@ static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t
continue;
err = sun4v_cpu_state(cpu);
- if (err >= 0 &&
- err == HV_CPU_STATE_ERROR) {
+ if (err == HV_CPU_STATE_ERROR) {
+ saw_cpu_error = (cpu + 1);
cpu_list[i] = 0xffff;
- cpu_set(cpu, error_mask);
}
}
} else if (unlikely(status != HV_EWOULDBLOCK))
@@ -717,32 +705,24 @@ static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t
}
} while (1);
- local_irq_restore(flags);
-
- if (unlikely(!cpus_empty(error_mask)))
+ if (unlikely(saw_cpu_error))
goto fatal_mondo_cpu_error;
return;
fatal_mondo_cpu_error:
printk(KERN_CRIT "CPU[%d]: SUN4V mondo cpu error, some target cpus "
- "were in error state\n",
- this_cpu);
- printk(KERN_CRIT "CPU[%d]: Error mask [ ", this_cpu);
- for_each_cpu_mask(i, error_mask)
- printk("%d ", i);
- printk("]\n");
+ "(including %d) were in error state\n",
+ this_cpu, saw_cpu_error - 1);
return;
fatal_mondo_timeout:
- local_irq_restore(flags);
printk(KERN_CRIT "CPU[%d]: SUN4V mondo timeout, no forward "
" progress after %d retries.\n",
this_cpu, retries);
goto dump_cpu_list_and_out;
fatal_mondo_error:
- local_irq_restore(flags);
printk(KERN_CRIT "CPU[%d]: Unexpected SUN4V mondo error %lu\n",
this_cpu, status);
printk(KERN_CRIT "CPU[%d]: Args were cnt(%d) cpulist_pa(%lx) "
@@ -756,58 +736,93 @@ dump_cpu_list_and_out:
printk("]\n");
}
-/* Send cross call to all processors mentioned in MASK
- * except self.
+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_mask_nr(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, cpumask_t 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));
- int this_cpu = get_cpu();
- cpus_and(mask, mask, cpu_online_map);
- cpu_clear(this_cpu, mask);
-
- if (tlb_type == spitfire)
- spitfire_xcall_deliver(data0, data1, data2, mask);
- else if (tlb_type == cheetah || tlb_type == cheetah_plus)
- cheetah_xcall_deliver(data0, data1, data2, mask);
- else
- hypervisor_xcall_deliver(data0, data1, data2, mask);
- /* NOTE: Caller runs local copy on master. */
+ xcall_deliver(data0, data1, data2, mask);
+}
- put_cpu();
+/* 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)
{
- cpumask_t mask = cpumask_of_cpu(cpu);
-
- smp_cross_call_masked(&xcall_sync_tick,
- 0, 0, 0, mask);
+ xcall_deliver((u64) &xcall_sync_tick, 0, 0,
+ &cpumask_of_cpu(cpu));
}
extern unsigned long xcall_call_function;
void arch_send_call_function_ipi(cpumask_t mask)
{
- smp_cross_call_masked(&xcall_call_function, 0, 0, 0, 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)
{
- cpumask_t mask = cpumask_of_cpu(cpu);
-
- smp_cross_call_masked(&xcall_call_function_single, 0, 0, 0, mask);
+ xcall_deliver((u64) &xcall_call_function_single, 0, 0,
+ &cpumask_of_cpu(cpu));
}
-/* Send cross call to all processors except self. */
-#define smp_cross_call(func, ctx, data1, data2) \
- smp_cross_call_masked(func, ctx, data1, data2, cpu_online_map)
-
void smp_call_function_client(int irq, struct pt_regs *regs)
{
clear_softint(1 << irq);
@@ -877,7 +892,6 @@ static inline void __local_flush_dcache_page(struct page *page)
void smp_flush_dcache_page_impl(struct page *page, int cpu)
{
- cpumask_t mask = cpumask_of_cpu(cpu);
int this_cpu;
if (tlb_type == hypervisor)
@@ -893,29 +907,24 @@ void smp_flush_dcache_page_impl(struct page *page, int cpu)
__local_flush_dcache_page(page);
} else if (cpu_online(cpu)) {
void *pg_addr = page_address(page);
- u64 data0;
+ u64 data0 = 0;
if (tlb_type == spitfire) {
- data0 =
- ((u64)&xcall_flush_dcache_page_spitfire);
+ data0 = ((u64)&xcall_flush_dcache_page_spitfire);
if (page_mapping(page) != NULL)
data0 |= ((u64)1 << 32);
- spitfire_xcall_deliver(data0,
- __pa(pg_addr),
- (u64) pg_addr,
- mask);
} else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
#ifdef DCACHE_ALIASING_POSSIBLE
- data0 =
- ((u64)&xcall_flush_dcache_page_cheetah);
- cheetah_xcall_deliver(data0,
- __pa(pg_addr),
- 0, mask);
+ 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);
+ atomic_inc(&dcpage_flushes_xcall);
#endif
+ }
}
put_cpu();
@@ -923,66 +932,41 @@ void smp_flush_dcache_page_impl(struct page *page, int cpu)
void flush_dcache_page_all(struct mm_struct *mm, struct page *page)
{
- void *pg_addr = page_address(page);
- cpumask_t mask = cpu_online_map;
- u64 data0;
+ void *pg_addr;
int this_cpu;
+ u64 data0;
if (tlb_type == hypervisor)
return;
this_cpu = get_cpu();
- cpu_clear(this_cpu, mask);
-
#ifdef CONFIG_DEBUG_DCFLUSH
atomic_inc(&dcpage_flushes);
#endif
- if (cpus_empty(mask))
- goto flush_self;
+ 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);
- spitfire_xcall_deliver(data0,
- __pa(pg_addr),
- (u64) pg_addr,
- mask);
} else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
#ifdef DCACHE_ALIASING_POSSIBLE
data0 = ((u64)&xcall_flush_dcache_page_cheetah);
- cheetah_xcall_deliver(data0,
- __pa(pg_addr),
- 0, mask);
#endif
}
+ if (data0) {
+ xcall_deliver(data0, __pa(pg_addr),
+ (u64) pg_addr, &cpu_online_map);
#ifdef CONFIG_DEBUG_DCFLUSH
- atomic_inc(&dcpage_flushes_xcall);
+ atomic_inc(&dcpage_flushes_xcall);
#endif
- flush_self:
+ }
__local_flush_dcache_page(page);
put_cpu();
}
-static void __smp_receive_signal_mask(cpumask_t mask)
-{
- smp_cross_call_masked(&xcall_receive_signal, 0, 0, 0, mask);
-}
-
-void smp_receive_signal(int cpu)
-{
- cpumask_t mask = cpumask_of_cpu(cpu);
-
- if (cpu_online(cpu))
- __smp_receive_signal_mask(mask);
-}
-
-void smp_receive_signal_client(int irq, struct pt_regs *regs)
-{
- clear_softint(1 << irq);
-}
-
void smp_new_mmu_context_version_client(int irq, struct pt_regs *regs)
{
struct mm_struct *mm;
@@ -1083,7 +1067,7 @@ void smp_flush_tlb_mm(struct mm_struct *mm)
smp_cross_call_masked(&xcall_flush_tlb_mm,
ctx, 0, 0,
- mm->cpu_vm_mask);
+ &mm->cpu_vm_mask);
local_flush_and_out:
__flush_tlb_mm(ctx, SECONDARY_CONTEXT);
@@ -1101,7 +1085,7 @@ void smp_flush_tlb_pending(struct mm_struct *mm, unsigned long nr, unsigned long
else
smp_cross_call_masked(&xcall_flush_tlb_pending,
ctx, nr, (unsigned long) vaddrs,
- mm->cpu_vm_mask);
+ &mm->cpu_vm_mask);
__flush_tlb_pending(ctx, nr, vaddrs);
@@ -1202,6 +1186,16 @@ 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;
@@ -1370,7 +1364,13 @@ void __init smp_cpus_done(unsigned int max_cpus)
void smp_send_reschedule(int cpu)
{
- smp_receive_signal(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
diff --git a/arch/sparc64/mm/tsb.c b/arch/sparc64/mm/tsb.c
index 3547937b17a..587f8efb2e0 100644
--- a/arch/sparc64/mm/tsb.c
+++ b/arch/sparc64/mm/tsb.c
@@ -1,9 +1,10 @@
/* arch/sparc64/mm/tsb.c
*
- * Copyright (C) 2006 David S. Miller <davem@davemloft.net>
+ * Copyright (C) 2006, 2008 David S. Miller <davem@davemloft.net>
*/
#include <linux/kernel.h>
+#include <linux/preempt.h>
#include <asm/system.h>
#include <asm/page.h>
#include <asm/tlbflush.h>
@@ -415,7 +416,9 @@ retry_tsb_alloc:
tsb_context_switch(mm);
/* Now force other processors to do the same. */
+ preempt_disable();
smp_tsb_sync(mm);
+ preempt_enable();
/* Now it is safe to free the old tsb. */
kmem_cache_free(tsb_caches[old_cache_index], old_tsb);