> To: Pete Zaitcev <zaitcev@redhat.com>
> Cc: linux-kernel@vger.kernel.org
> From: "Martin Schwidefsky" <schwidefsky@de.ibm.com>
> Date: Mon, 18 Mar 2002 15:03:49 +0100
>[...]
> Yep makes sense as well. I actually made the same fix for the kernel
> 2.5.6. I have it basically going but SMP is still a bit broken. It
> hangs on boot with 5 cpus. Seems like the startup of the migration
> threads doesn't complete because load_balance isn't balancing...
Darn, I re-invented the wheel again, it seems.
Attached is the patch to have O(1) on 2.4.18-RH (I backported
some of your things from 2.4.19-pre3). I am not sure if it works
right at all. I would like to see yours, unless it's already in
Linus' tree.
-- Pete
diff -urN -X dontdiff linux-2.4.17-0.18/arch/s390/kernel/process.c linux-2.4.17-0.18-t1/arch/s390/kernel/process.c
--- linux-2.4.17-0.18/arch/s390/kernel/process.c Mon Feb 18 20:02:26 2002
+++ linux-2.4.17-0.18-t1/arch/s390/kernel/process.c Tue Feb 19 03:37:28 2002
@@ -52,12 +52,8 @@
static psw_t wait_psw;
-int cpu_idle(void *unused)
+int cpu_idle(void)
{
- /* endless idle loop with no priority at all */
- init_idle();
- current->nice = 20;
- current->counter = -100;
wait_psw.mask = _WAIT_PSW_MASK;
wait_psw.addr = (unsigned long) &&idle_wakeup | 0x80000000L;
while(1) {
@@ -82,7 +78,7 @@
{
struct task_struct *tsk = current;
- printk("CPU: %d %s\n", tsk->processor, print_tainted());
+ printk("CPU: %d %s\n", tsk->cpu, print_tainted());
printk("Process %s (pid: %d, task: %08lx, ksp: %08x)\n",
current->comm, current->pid, (unsigned long) tsk,
tsk->thread.ksp);
diff -urN -X dontdiff linux-2.4.17-0.18/arch/s390/kernel/s390_ksyms.c linux-2.4.17-0.18-t1/arch/s390/kernel/s390_ksyms.c
--- linux-2.4.17-0.18/arch/s390/kernel/s390_ksyms.c Mon Feb 18 20:05:29 2002
+++ linux-2.4.17-0.18-t1/arch/s390/kernel/s390_ksyms.c Wed Feb 20 02:05:42 2002
@@ -11,6 +11,7 @@
#if CONFIG_IP_MULTICAST
#include <net/arp.h>
#endif
+#include <linux/sched.h> /* XXX sys_shed_yield - broken */
/*
* memory management
@@ -60,3 +60,4 @@
EXPORT_SYMBOL(console_mode);
EXPORT_SYMBOL(console_device);
EXPORT_SYMBOL_NOVERS(do_call_softirq);
+EXPORT_SYMBOL_NOVERS(sys_sched_yield); /* XXX */
diff -urN -X dontdiff linux-2.4.17-0.18/arch/s390/kernel/smp.c linux-2.4.17-0.18-t1/arch/s390/kernel/smp.c
--- linux-2.4.17-0.18/arch/s390/kernel/smp.c Mon Feb 18 20:02:26 2002
+++ linux-2.4.17-0.18-t1/arch/s390/kernel/smp.c Tue Feb 19 22:02:05 2002
@@ -38,7 +38,7 @@
#include <asm/cpcmd.h>
/* prototypes */
-extern int cpu_idle(void * unused);
+extern int cpu_idle(void);
extern __u16 boot_cpu_addr;
extern volatile int __cpu_logical_map[];
@@ -49,13 +49,13 @@
static int max_cpus = NR_CPUS; /* Setup configured maximum number of CPUs to activate */
int smp_num_cpus;
struct _lowcore *lowcore_ptr[NR_CPUS];
-cycles_t cacheflush_time=0;
int smp_threads_ready=0; /* Set when the idlers are all forked. */
static atomic_t smp_commenced = ATOMIC_INIT(0);
spinlock_t kernel_flag __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
unsigned long cpu_online_map;
+unsigned long cache_decay_ticks;
/*
* Setup routine for controlling SMP activation
@@ -95,6 +95,8 @@
static sigp_ccode smp_ext_bitcall(int, ec_bit_sig);
static void smp_ext_bitcall_others(ec_bit_sig);
+static void do_task_migration(void);
+
/*
* Structure and data for smp_call_function(). This is designed to minimise
* static memory requirements. It also looks cleaner.
@@ -259,6 +261,8 @@
do_machine_power_off();
if (test_bit(ec_call_function, &bits))
do_call_function();
+ if (test_bit(ec_task_migration, &bits))
+ do_task_migration();
}
/*
@@ -361,6 +365,35 @@
local_flush_tlb();
}
+static spinlock_t migration_lock = SPIN_LOCK_UNLOCKED;
+static task_t *new_task;
+
+/*
+ * Task migration callback.
+ */
+static void do_task_migration(void)
+{
+ task_t *p;
+
+ p = new_task;
+ spin_unlock(&migration_lock);
+ sched_task_migrated(p);
+}
+
+/*
+ * This function sends a 'task migration' IPI to another CPU.
+ * Must be called from syscall contexts, with interrupts *enabled*.
+ */
+void smp_migrate_task(int cpu, task_t *p)
+{
+ /*
+ * The target CPU will unlock the migration spinlock:
+ */
+ spin_lock(&migration_lock);
+ new_task = p;
+ smp_ext_bitcall(cpu, ec_task_migration);
+}
+
/*
* this function sends a 'reschedule' IPI to another CPU.
* it goes straight through and wastes no time serializing
@@ -372,6 +405,19 @@
smp_ext_bitcall(cpu, ec_schedule);
}
+#if 0 /* not used in current mingo code */
+/*
+ * this function sends a reschedule IPI to all (other) CPUs.
+ * This should only be used if some 'global' task became runnable,
+ * such as a RT task, that must be handled now. The first CPU
+ * that manages to grab the task will run it.
+ */
+void smp_send_reschedule_all(void)
+{
+ send_IPI_allbutself(RESCHEDULE_VECTOR);
+}
+#endif
+
/*
* parameter area for the set/clear control bit callbacks
*/
@@ -449,7 +495,7 @@
{
int curr_cpu;
- current->processor = 0;
+ current->cpu = 0;
smp_num_cpus = 1;
cpu_online_map = 1;
for (curr_cpu = 0;
@@ -490,7 +536,7 @@
pfault_init();
#endif
/* cpu_idle will call schedule for us */
- return cpu_idle(NULL);
+ return cpu_idle();
}
/*
@@ -528,12 +574,9 @@
idle = init_task.prev_task;
if (!idle)
panic("No idle process for CPU %d",cpu);
- idle->processor = cpu;
- idle->cpus_runnable = 1 << cpu; /* we schedule the first task manually */
+ init_idle(idle, cpu);
- del_from_runqueue(idle);
unhash_process(idle);
- init_tasks[cpu] = idle;
cpu_lowcore=&get_cpu_lowcore(cpu);
cpu_lowcore->save_area[15] = idle->thread.ksp;
@@ -585,7 +628,9 @@
panic("Couldn't request external interrupt 0x1202");
smp_count_cpus();
memset(lowcore_ptr,0,sizeof(lowcore_ptr));
-
+
+ cache_decay_ticks = (200 * HZ) / 1000; /* Is 200ms ok? Robust? XXX */
+
/*
* Initialize the logical to physical CPU number mapping
*/
diff -urN -X dontdiff linux-2.4.17-0.18/arch/s390/kernel/traps.c linux-2.4.17-0.18-t1/arch/s390/kernel/traps.c
--- linux-2.4.17-0.18/arch/s390/kernel/traps.c Mon Feb 18 20:02:26 2002
+++ linux-2.4.17-0.18-t1/arch/s390/kernel/traps.c Tue Feb 19 21:05:15 2002
@@ -135,12 +135,14 @@
void show_trace_task(struct task_struct *tsk)
{
+#if 0 /* Mingo's scheduler kills task_has_cpu, so we bite the bullet. */
/*
* We can't print the backtrace of a running process. It is
* unreliable at best and can cause kernel oopses.
*/
if (task_has_cpu(tsk))
return;
+#endif
show_trace((unsigned long *) tsk->thread.ksp);
}
diff -urN -X dontdiff linux-2.4.17-0.18/arch/s390/mm/fault.c linux-2.4.17-0.18-t1/arch/s390/mm/fault.c
--- linux-2.4.17-0.18/arch/s390/mm/fault.c Fri Nov 9 22:58:02 2001
+++ linux-2.4.17-0.18-t1/arch/s390/mm/fault.c Tue Feb 19 03:09:46 2002
@@ -285,8 +285,7 @@
out_of_memory:
up_read(&mm->mmap_sem);
if (tsk->pid == 1) {
- tsk->policy |= SCHED_YIELD;
- schedule();
+ yield();
down_read(&mm->mmap_sem);
goto survive;
}
diff -urN -X dontdiff linux-2.4.17-0.18/arch/s390x/kernel/process.c linux-2.4.17-0.18-t1/arch/s390x/kernel/process.c
--- linux-2.4.17-0.18/arch/s390x/kernel/process.c Mon Feb 18 20:02:28 2002
+++ linux-2.4.17-0.18-t1/arch/s390x/kernel/process.c Tue Feb 19 03:36:16 2002
@@ -52,12 +52,9 @@
static psw_t wait_psw;
-int cpu_idle(void *unused)
+int cpu_idle(void)
{
/* endless idle loop with no priority at all */
- init_idle();
- current->nice = 20;
- current->counter = -100;
wait_psw.mask = _WAIT_PSW_MASK;
wait_psw.addr = (unsigned long) &&idle_wakeup;
while(1) {
diff -urN -X dontdiff linux-2.4.17-0.18/arch/s390x/kernel/smp.c linux-2.4.17-0.18-t1/arch/s390x/kernel/smp.c
--- linux-2.4.17-0.18/arch/s390x/kernel/smp.c Mon Feb 18 20:02:28 2002
+++ linux-2.4.17-0.18-t1/arch/s390x/kernel/smp.c Tue Feb 19 03:36:43 2002
@@ -38,7 +38,7 @@
#include <asm/cpcmd.h>
/* prototypes */
-extern int cpu_idle(void * unused);
+extern int cpu_idle(void);
extern __u16 boot_cpu_addr;
extern volatile int __cpu_logical_map[];
@@ -468,8 +468,9 @@
/* Enable pfault pseudo page faults on this cpu. */
pfault_init();
#endif
+ init_idle();
/* cpu_idle will call schedule for us */
- return cpu_idle(NULL);
+ return cpu_idle();
}
/*
diff -urN -X dontdiff linux-2.4.17-0.18/arch/s390x/mm/fault.c linux-2.4.17-0.18-t1/arch/s390x/mm/fault.c
--- linux-2.4.17-0.18/arch/s390x/mm/fault.c Fri Nov 9 22:58:02 2001
+++ linux-2.4.17-0.18-t1/arch/s390x/mm/fault.c Tue Feb 19 03:10:28 2002
@@ -286,8 +286,7 @@
out_of_memory:
up_read(&mm->mmap_sem);
if (tsk->pid == 1) {
- tsk->policy |= SCHED_YIELD;
- schedule();
+ yield();
down_read(&mm->mmap_sem);
goto survive;
}
diff -urN -X dontdiff linux-2.4.17-0.18/include/asm-s390/bitops.h linux-2.4.17-0.18-t1/include/asm-s390/bitops.h
--- linux-2.4.17-0.18/include/asm-s390/bitops.h Wed Jul 25 23:12:02 2001
+++ linux-2.4.17-0.18-t1/include/asm-s390/bitops.h Tue Feb 19 21:06:47 2002
@@ -752,6 +752,68 @@
}
/*
+ * Yet Another Bitop for mingo's scheduler. Don't use!
+ * Result is undefined if no bit exists, so code should check against 0 first.
+ *
+ * XXX Measure if this actually needs any optimization.
+ */
+static __inline__ int __ffs(unsigned long word)
+{
+ int num = 0;
+
+ if ((word & 0xffff) == 0) {
+ num += 16;
+ word >>= 16;
+ }
+ if ((word & 0xff) == 0) {
+ num += 8;
+ word >>= 8;
+ }
+ if ((word & 0xf) == 0) {
+ num += 4;
+ word >>= 4;
+ }
+ if ((word & 0x3) == 0) {
+ num += 2;
+ word >>= 2;
+ }
+ if ((word & 0x1) == 0)
+ num += 1;
+ return num;
+}
+
+/*
+ * Scheduler induced bitop, better not to use.
+ * The Ingo's brilliant design puts its find_first()
+ * counterpart into <asm/mmu_context.h>.
+ * Some implementations use void *addr as argument, which is
+ * totally broken, because we must be compatible with set_bit().
+ * The size is likely to be 140, but the map is padded with zeroes.
+ *
+ * find_next_bit - find the first set bit in a memory region
+ * @addr: The address to base the search on
+ * @offset: The bitnumber to start searching at
+ * @size: The maximum size to search
+ */
+static __inline__ int find_next_bit(unsigned long *addr, int size, int offset)
+{
+ unsigned long *p = addr + (offset >> 5);
+ int num = offset & ~0x1f;
+ unsigned long word;
+
+ word = *p++;
+ word &= ~((1 << (offset & 0x1f)) - 1);
+ while (num < size) {
+ if (word != 0) {
+ return __ffs(word) + num;
+ }
+ word = *p++;
+ num += 0x20;
+ }
+ return num;
+}
+
+/*
* hweightN: returns the hamming weight (i.e. the number
* of bits set) of a N-bit word
*/
diff -urN -X dontdiff linux-2.4.17-0.18/include/asm-s390/mmu_context.h linux-2.4.17-0.18-t1/include/asm-s390/mmu_context.h
--- linux-2.4.17-0.18/include/asm-s390/mmu_context.h Tue Feb 13 23:13:44 2001
+++ linux-2.4.17-0.18-t1/include/asm-s390/mmu_context.h Tue Feb 19 06:03:42 2002
@@ -10,6 +10,25 @@
#define __S390_MMU_CONTEXT_H
/*
+ * Every architecture must define this function. It's the fastest
+ * way of searching a 140-bit bitmap where the first 100 bits are
+ * unlikely to be set. It's guaranteed that at least one of the 140
+ * bits is cleared.
+ */
+static inline int sched_find_first_bit(unsigned long *b)
+{
+ if (unlikely(b[0]))
+ return __ffs(b[0]);
+ if (unlikely(b[1]))
+ return __ffs(b[1]) + 32;
+ if (unlikely(b[2]))
+ return __ffs(b[2]) + 64;
+ if (b[3])
+ return __ffs(b[3]) + 96;
+ return __ffs(b[4]) + 128;
+}
+
+/*
* get a new mmu context.. S390 don't know about contexts.
*/
#define init_new_context(tsk,mm) 0
diff -urN -X dontdiff linux-2.4.17-0.18/include/asm-s390/sigp.h linux-2.4.17-0.18-t1/include/asm-s390/sigp.h
--- linux-2.4.17-0.18/include/asm-s390/sigp.h Mon Feb 18 21:27:30 2002
+++ linux-2.4.17-0.18-t1/include/asm-s390/sigp.h Tue Feb 19 21:07:33 2002
@@ -63,6 +63,7 @@
ec_halt,
ec_power_off,
ec_call_function,
+ ec_task_migration,
ec_bit_last
} ec_bit_sig;
diff -urN -X dontdiff linux-2.4.17-0.18/include/asm-s390/smp.h linux-2.4.17-0.18-t1/include/asm-s390/smp.h
--- linux-2.4.17-0.18/include/asm-s390/smp.h Mon Feb 18 21:27:34 2002
+++ linux-2.4.17-0.18-t1/include/asm-s390/smp.h Tue Feb 19 21:07:36 2002
@@ -30,19 +30,7 @@
#define NO_PROC_ID 0xFF /* No processor magic marker */
-/*
- * This magic constant controls our willingness to transfer
- * a process across CPUs. Such a transfer incurs misses on the L1
- * cache, and on a P6 or P5 with multiple L2 caches L2 hits. My
- * gut feeling is this will vary by board in value. For a board
- * with separate L2 cache it probably depends also on the RSS, and
- * for a board with shared L2 cache it ought to decay fast as other
- * processes are run.
- */
-
-#define PROC_CHANGE_PENALTY 20 /* Schedule penalty */
-
-#define smp_processor_id() (current->processor)
+#define smp_processor_id() (current->cpu)
extern __inline__ int cpu_logical_map(int cpu)
{
diff -ur -X dontdiff linux-2.4.18-0.1.s390/include/asm-s390x/smp.h linux-2.4.18-0.1-x.s390/include/asm-s390x/smp.h
--- linux-2.4.18-0.1.s390/include/asm-s390x/smp.h Thu Oct 11 09:43:38 2001
+++ linux-2.4.18-0.1-x.s390/include/asm-s390x/smp.h Wed Mar 6 21:17:30 2002
@@ -30,19 +30,7 @@
#define NO_PROC_ID 0xFF /* No processor magic marker */
-/*
- * This magic constant controls our willingness to transfer
- * a process across CPUs. Such a transfer incurs misses on the L1
- * cache, and on a P6 or P5 with multiple L2 caches L2 hits. My
- * gut feeling is this will vary by board in value. For a board
- * with separate L2 cache it probably depends also on the RSS, and
- * for a board with shared L2 cache it ought to decay fast as other
- * processes are run.
- */
-
-#define PROC_CHANGE_PENALTY 20 /* Schedule penalty */
-
-#define smp_processor_id() (current->processor)
+#define smp_processor_id() (current->cpu)
extern __inline__ int cpu_logical_map(int cpu)
{
diff -ur -X dontdiff linux-2.4.18-0.1.s390/arch/s390/kernel/entry.S linux-2.4.18-0.1-x.s390/arch/s390/kernel/entry.S
--- linux-2.4.18-0.1.s390/arch/s390/kernel/entry.S Mon Feb 25 11:37:56 2002
+++ linux-2.4.18-0.1-x.s390/arch/s390/kernel/entry.S Wed Mar 6 19:18:22 2002
@@ -291,17 +289,16 @@
ret_from_fork:
basr %r13,0
l %r13,.Lentry_base-.(%r13) # setup base pointer to &entry_base
+#ifdef CONFIG_SMP
+ # not saving R14 here because we go to sysc_return ultimately
+ l %r1,BASED(.Lschedtail)
+ basr %r14,%r1 # call schedule_tail (unlock stuff)
+#endif
GET_CURRENT # load pointer to task_struct to R9
stosm 24(%r15),0x03 # reenable interrupts
sr %r0,%r0 # child returns 0
st %r0,SP_R2(%r15) # store return value (change R2 on stack)
-#ifdef CONFIG_SMP
- l %r1,BASED(.Lschedtail)
- la %r14,BASED(sysc_return)
- br %r1 # call schedule_tail, return to sysc_return
-#else
b BASED(sysc_return)
-#endif
#
# clone, fork, vfork, exec and sigreturn need glue,
diff -ur -X dontdiff linux-2.4.18-0.1.s390/arch/s390x/kernel/entry.S linux-2.4.18-0.1-x.s390/arch/s390x/kernel/entry.S
--- linux-2.4.18-0.1.s390/arch/s390x/kernel/entry.S Mon Feb 25 11:37:56 2002
+++ linux-2.4.18-0.1-x.s390/arch/s390x/kernel/entry.S Wed Mar 6 19:41:58 2002
@@ -277,15 +275,13 @@
#
.globl ret_from_fork
ret_from_fork:
+#ifdef CONFIG_SMP
+ brasl %r14,schedule_tail
+#endif
GET_CURRENT # load pointer to task_struct to R9
stosm 48(%r15),0x03 # reenable interrupts
xc SP_R2(8,%r15),SP_R2(%r15) # child returns 0
-#ifdef CONFIG_SMP
- larl %r14,sysc_return
- jg schedule_tail # return to sysc_return
-#else
j sysc_return
-#endif
#
# clone, fork, vfork, exec and sigreturn need glue,
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Mar 23 2002 - 22:00:17 EST