aboutsummaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
authorArnd Bergmann <arnd@arndb.de>2006-01-05 09:05:29 -0500
committerPaul Mackerras <paulus@samba.org>2006-01-08 23:44:57 -0500
commit2fb9d2063626374dd8a2514b3a730facac8235d8 (patch)
treeb410dcdbc5aee656c37951be36951130450549e7 /arch
parentaeb013772a2cc85a8d0baffd64977d2888bc781d (diff)
[PATCH] spufs: set irq affinity for running threads
For far, all SPU triggered interrupts always end up on the first SMT thread, which is a bad solution. This patch implements setting the affinity to the CPU that was running last when entering execution on an SPU. This should result in a significant reduction in IPI calls and better cache locality for SPE thread specific data. Signed-off-by: Arnd Bergmann <arndb@de.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
Diffstat (limited to 'arch')
-rw-r--r--arch/powerpc/platforms/cell/interrupt.c42
-rw-r--r--arch/powerpc/platforms/cell/interrupt.h1
-rw-r--r--arch/powerpc/platforms/cell/spu_base.c8
-rw-r--r--arch/powerpc/platforms/cell/spufs/sched.c5
4 files changed, 40 insertions, 16 deletions
diff --git a/arch/powerpc/platforms/cell/interrupt.c b/arch/powerpc/platforms/cell/interrupt.c
index 7fbe78a9327d..63aa52acf441 100644
--- a/arch/powerpc/platforms/cell/interrupt.c
+++ b/arch/powerpc/platforms/cell/interrupt.c
@@ -23,6 +23,7 @@
23#include <linux/config.h> 23#include <linux/config.h>
24#include <linux/interrupt.h> 24#include <linux/interrupt.h>
25#include <linux/irq.h> 25#include <linux/irq.h>
26#include <linux/module.h>
26#include <linux/percpu.h> 27#include <linux/percpu.h>
27#include <linux/types.h> 28#include <linux/types.h>
28 29
@@ -55,6 +56,7 @@ struct iic_regs {
55 56
56struct iic { 57struct iic {
57 struct iic_regs __iomem *regs; 58 struct iic_regs __iomem *regs;
59 u8 target_id;
58}; 60};
59 61
60static DEFINE_PER_CPU(struct iic, iic); 62static DEFINE_PER_CPU(struct iic, iic);
@@ -172,12 +174,11 @@ int iic_get_irq(struct pt_regs *regs)
172 return irq; 174 return irq;
173} 175}
174 176
175static struct iic_regs __iomem *find_iic(int cpu) 177static int setup_iic(int cpu, struct iic *iic)
176{ 178{
177 struct device_node *np; 179 struct device_node *np;
178 int nodeid = cpu / 2; 180 int nodeid = cpu / 2;
179 unsigned long regs; 181 unsigned long regs;
180 struct iic_regs __iomem *iic_regs;
181 182
182 for (np = of_find_node_by_type(NULL, "cpu"); 183 for (np = of_find_node_by_type(NULL, "cpu");
183 np; 184 np;
@@ -188,20 +189,23 @@ static struct iic_regs __iomem *find_iic(int cpu)
188 189
189 if (!np) { 190 if (!np) {
190 printk(KERN_WARNING "IIC: CPU %d not found\n", cpu); 191 printk(KERN_WARNING "IIC: CPU %d not found\n", cpu);
191 iic_regs = NULL; 192 iic->regs = NULL;
192 } else { 193 iic->target_id = 0xff;
193 regs = *(long *)get_property(np, "iic", NULL); 194 return -ENODEV;
194
195 /* hack until we have decided on the devtree info */
196 regs += 0x400;
197 if (cpu & 1)
198 regs += 0x20;
199
200 printk(KERN_DEBUG "IIC for CPU %d at %lx\n", cpu, regs);
201 iic_regs = __ioremap(regs, sizeof(struct iic_regs),
202 _PAGE_NO_CACHE);
203 } 195 }
204 return iic_regs; 196
197 regs = *(long *)get_property(np, "iic", NULL);
198
199 /* hack until we have decided on the devtree info */
200 regs += 0x400;
201 if (cpu & 1)
202 regs += 0x20;
203
204 printk(KERN_DEBUG "IIC for CPU %d at %lx\n", cpu, regs);
205 iic->regs = __ioremap(regs, sizeof(struct iic_regs),
206 _PAGE_NO_CACHE);
207 iic->target_id = (nodeid << 4) + ((cpu & 1) ? 0xf : 0xe);
208 return 0;
205} 209}
206 210
207#ifdef CONFIG_SMP 211#ifdef CONFIG_SMP
@@ -227,6 +231,12 @@ void iic_cause_IPI(int cpu, int mesg)
227 out_be64(&per_cpu(iic, cpu).regs->generate, (IIC_NUM_IPIS - 1 - mesg) << 4); 231 out_be64(&per_cpu(iic, cpu).regs->generate, (IIC_NUM_IPIS - 1 - mesg) << 4);
228} 232}
229 233
234u8 iic_get_target_id(int cpu)
235{
236 return per_cpu(iic, cpu).target_id;
237}
238EXPORT_SYMBOL_GPL(iic_get_target_id);
239
230static irqreturn_t iic_ipi_action(int irq, void *dev_id, struct pt_regs *regs) 240static irqreturn_t iic_ipi_action(int irq, void *dev_id, struct pt_regs *regs)
231{ 241{
232 smp_message_recv(iic_irq_to_ipi(irq), regs); 242 smp_message_recv(iic_irq_to_ipi(irq), regs);
@@ -276,7 +286,7 @@ void iic_init_IRQ(void)
276 irq_offset = 0; 286 irq_offset = 0;
277 for_each_cpu(cpu) { 287 for_each_cpu(cpu) {
278 iic = &per_cpu(iic, cpu); 288 iic = &per_cpu(iic, cpu);
279 iic->regs = find_iic(cpu); 289 setup_iic(cpu, iic);
280 if (iic->regs) 290 if (iic->regs)
281 out_be64(&iic->regs->prio, 0xff); 291 out_be64(&iic->regs->prio, 0xff);
282 } 292 }
diff --git a/arch/powerpc/platforms/cell/interrupt.h b/arch/powerpc/platforms/cell/interrupt.h
index 37d58e6fd0c6..a14bd38791c0 100644
--- a/arch/powerpc/platforms/cell/interrupt.h
+++ b/arch/powerpc/platforms/cell/interrupt.h
@@ -54,6 +54,7 @@ extern void iic_setup_cpu(void);
54extern void iic_local_enable(void); 54extern void iic_local_enable(void);
55extern void iic_local_disable(void); 55extern void iic_local_disable(void);
56 56
57extern u8 iic_get_target_id(int cpu);
57 58
58extern void spider_init_IRQ(void); 59extern void spider_init_IRQ(void);
59extern int spider_get_irq(unsigned long int_pending); 60extern int spider_get_irq(unsigned long int_pending);
diff --git a/arch/powerpc/platforms/cell/spu_base.c b/arch/powerpc/platforms/cell/spu_base.c
index 7fe3fa3da0e9..d75ae03df686 100644
--- a/arch/powerpc/platforms/cell/spu_base.c
+++ b/arch/powerpc/platforms/cell/spu_base.c
@@ -507,6 +507,14 @@ int spu_irq_class_1_bottom(struct spu *spu)
507 return ret; 507 return ret;
508} 508}
509 509
510void spu_irq_setaffinity(struct spu *spu, int cpu)
511{
512 u64 target = iic_get_target_id(cpu);
513 u64 route = target << 48 | target << 32 | target << 16;
514 spu_int_route_set(spu, route);
515}
516EXPORT_SYMBOL_GPL(spu_irq_setaffinity);
517
510static void __iomem * __init map_spe_prop(struct device_node *n, 518static void __iomem * __init map_spe_prop(struct device_node *n,
511 const char *name) 519 const char *name)
512{ 520{
diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
index c34198c29159..963182fbd1aa 100644
--- a/arch/powerpc/platforms/cell/spufs/sched.c
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -357,6 +357,11 @@ int spu_activate(struct spu_context *ctx, u64 flags)
357 if (!spu) 357 if (!spu)
358 return (signal_pending(current)) ? -ERESTARTSYS : -EAGAIN; 358 return (signal_pending(current)) ? -ERESTARTSYS : -EAGAIN;
359 bind_context(spu, ctx); 359 bind_context(spu, ctx);
360 /*
361 * We're likely to wait for interrupts on the same
362 * CPU that we are now on, so send them here.
363 */
364 spu_irq_setaffinity(spu, raw_smp_processor_id());
360 put_active_spu(spu); 365 put_active_spu(spu);
361 return 0; 366 return 0;
362} 367}