bitkeeper revision 1.1244 (42308dfaWqh7O5QB-sS9WJ1R4LYoEA)

various hacks for batching mode and stats gathering

Signed-off-by: michael.fetterman@cl.cam.ac.uk

.rootkeys

@@ -159,6 +159,7 @@
 40f56238nWMQg7CKbyTy0KJNvCzbtg linux-2.6.10-xen-sparse/arch/xen/i386/kernel/signal.c
 41811cac4lkCB-fHir6CcxuEJ2pGsQ linux-2.6.10-xen-sparse/arch/xen/i386/kernel/smp.c
 41811ca9mbGpqBrZVrUGEiv8CTV3ng linux-2.6.10-xen-sparse/arch/xen/i386/kernel/smpboot.c
+42308df8u332Gs7XX-jX4gsfFU2zOQ linux-2.6.10-xen-sparse/arch/xen/i386/kernel/syscall_stats.c
 40f56238qVGkpO_ycnQA8k03kQzAgA linux-2.6.10-xen-sparse/arch/xen/i386/kernel/time.c
 40f56238NzTgeO63RGoxHrW5NQeO3Q linux-2.6.10-xen-sparse/arch/xen/i386/kernel/timers/Makefile
 40f56238BMqG5PuSHufpjbvp_helBw linux-2.6.10-xen-sparse/arch/xen/i386/kernel/timers/timer_tsc.c
@@ -601,6 +602,7 @@
 40c9c469kT0H9COWzA4XzPBjWK0WsA tools/misc/netfix
 4022a73cEKvrYe_DVZW2JlAxobg9wg tools/misc/nsplitd/Makefile
 4022a73cKms4Oq030x2JBzUB426lAQ tools/misc/nsplitd/nsplitd.c
+42308df9dv_ZuP49nNPIROEMQ3F_LA tools/misc/xc_shadow.c
 3f5ef5a2ir1kVAthS14Dc5QIRCEFWg tools/misc/xen-clone
 3f5ef5a2dTZP0nnsFoeq2jRf3mWDDg tools/misc/xen-clone.README
 405eedf6_nnNhFQ1I85lhCkLK6jFGA tools/misc/xencons

linux-2.6.10-xen-sparse/arch/xen/Kconfig

@@ -115,9 +115,38 @@ config XEN_BLKDEV_TAP
 	  space.  Odds are that you want to say N here.
 
 config XEN_WRITABLE_PAGETABLES
-	bool
+	bool "writable page tables"
 	default y
 
+config XEN_SYSCALL_STATS
+	bool "system call statistics"
+	default n
+
+config XEN_DEBUG_NO_MMU_BATCHING
+	bool "Disables batching on MMU updates"
+	default n
+    help
+      This does a hypercall per PTE update
+      we only use this for benchmarking
+      enable only if you know what you are doing
+
+config XEN_BATCH_MODE1
+	bool "A variant of writable pagetable using the batch interface"
+	default n
+    help
+      default is no batching and minor mods for some batching
+      we only use this for benchmarking
+      enable only if you know what you are doing
+
+config XEN_BATCH_MODE2
+	bool "forward port of 2.4 batching"
+	default n
+    help
+      default is batching + flushes where 2.4 had them
+      we only use this for benchmarking
+      enable only if you know what you are doing
+
+
 config XEN_SCRUB_PAGES
 	bool "Scrub memory before freeing it to Xen"
 	default y

linux-2.6.10-xen-sparse/arch/xen/i386/kernel/Makefile

@@ -43,6 +43,8 @@ c-obj-$(CONFIG_HPET_TIMER) 	+= time_hpet.o
 c-obj-$(CONFIG_EFI) 		+= efi.o efi_stub.o
 c-obj-$(CONFIG_EARLY_PRINTK)	+= early_printk.o
 
+c-obj-$(CONFIG_XEN_SYSCALL_STATS) += syscall_stats.o
+
 EXTRA_AFLAGS   := -traditional
 
 c-obj-$(CONFIG_SCx200)		+= scx200.o

linux-2.6.10-xen-sparse/arch/xen/i386/kernel/entry.S

@@ -280,7 +280,10 @@ sysenter_past_esp:
 	jnz syscall_trace_entry
 	cmpl $(nr_syscalls), %eax
 	jae syscall_badsys
-	call *sys_call_table(,%eax,4)
+#ifdef CONFIG_XEN_SYSCALL_STATS
+    lock incl syscall_stats(,%eax,4)
+#endif
+    call *sys_call_table(,%eax,4)
 	movl %eax,EAX(%esp)
 	cli
 	movl TI_flags(%ebp), %ecx
@@ -305,7 +308,10 @@ ENTRY(system_call)
 	cmpl $(nr_syscalls), %eax
 	jae syscall_badsys
 syscall_call:
-	call *sys_call_table(,%eax,4)
+#ifdef CONFIG_XEN_SYSCALL_STATS
+    lock incl syscall_stats(,%eax,4)
+#endif
+    call *sys_call_table(,%eax,4)
 	movl %eax,EAX(%esp)		# store the return value
 syscall_exit:
 	XEN_BLOCK_EVENTS(%esi)		# make sure we don't miss an interrupt

linux-2.6.10-xen-sparse/arch/xen/i386/kernel/syscall_stats.c

@@ -0,0 +1,91 @@
+/* -*-  Mode:C; c-basic-offset:4; tab-width:4 -*-
+ ****************************************************************************
+ * (C) 2005 - Rolf Neugebauer - Intel Research Cambridge
+ ****************************************************************************
+ *
+ *        File: syscall_stats.c
+ *      Author: Rolf Neugebauer (rolf.neugebauer@intel.com)
+ *        Date: Mar 2005
+ * 
+ * Description: add a proc interface to get per system call stats
+ */
+
+
+#include <linux/config.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include <asm/unistd.h>
+
+unsigned long syscall_stats[NR_syscalls];
+static unsigned char foobar[4];
+
+/* a write just resests the counter */
+static ssize_t syscall_write(struct file *f, const  char *data,
+                             size_t size, loff_t  *pos)
+{
+    printk("resetting syscall stats\n");
+    memset(&syscall_stats, 0, sizeof(syscall_stats));
+    return size;
+}
+
+static int show_syscall(struct seq_file *m, void *v)
+{
+    int i;
+    for ( i=0; i<NR_syscalls; i++ )
+    {
+        seq_printf(m, "%lu ", syscall_stats[i]);
+    }
+    seq_printf(m, "\n");
+    return 0;
+}
+
+static void *c_start(struct seq_file *m, loff_t *pos)
+{
+    return *pos == 0 ? foobar : NULL;
+}
+
+static void *c_next(struct seq_file *m, void *v, loff_t *pos)
+{
+    ++*pos;
+    return c_start(m, pos);
+}
+
+static void c_stop(struct seq_file *m, void *v)
+{
+}
+
+static struct seq_operations syscall_op = {
+    start:  c_start,
+    next:   c_next,
+    stop:   c_stop,
+    show:   show_syscall,
+};
+
+static int syscall_open(struct inode *inode, struct file *file)
+{
+    return seq_open(file, &syscall_op);
+}
+
+static struct file_operations proc_syscall_operations = {
+    open:           syscall_open,
+    read:           seq_read,
+    write:          syscall_write,
+    llseek:         seq_lseek,
+    release:        seq_release,
+};
+
+
+static struct proc_dir_entry *entry;
+
+static int __init syscall_stats_init(void)
+{
+    printk("Initialising syscall stats.\n");
+
+    entry = create_proc_entry("syscalls", 0777, NULL);
+    if (entry)
+        entry->proc_fops = &proc_syscall_operations;
+    else
+        printk("Unable to create /proc/syscalls.\n");
+    return 0;
+}
+subsys_initcall(syscall_stats_init);

linux-2.6.10-xen-sparse/arch/xen/i386/mm/hypervisor.c

@@ -133,6 +133,9 @@ void queue_l1_entry_update(pte_t *ptr, unsigned long val)
     per_cpu(update_queue[idx], cpu).ptr = virt_to_machine(ptr);
     per_cpu(update_queue[idx], cpu).val = val;
     increment_index();
+#ifdef CONFIG_XEN_DEBUG_NO_MMU_BATCHING
+    __flush_page_update_queue();
+#endif
     spin_unlock_irqrestore(&update_lock, flags);
 }
 

linux-2.6.10-xen-sparse/include/asm-xen/asm-i386/pgtable-2level.h

@@ -36,7 +36,16 @@ do { \
 #endif
 #define set_pte_atomic(pteptr, pteval) set_pte(pteptr, pteval)
 #else
+#if defined(CONFIG_XEN_DEBUG_NO_MMU_BATCHING)
+#define set_pte(pteptr, pteval)\
+    set_pte_batched(pteptr, pteval)
+#elif defined(CONFIG_XEN_BATCH_MODE)
+#define set_pte(pteptr, pteval)({\
+    set_pte_batched(pteptr, pteval);\
+    _flush_page_update_queue();})
+#else
 #define set_pte(pteptr, pteval) (*(pteptr) = pteval)
+#endif
 #define set_pte_atomic(pteptr, pteval) set_pte(pteptr,pteval)
 #endif
 /*

linux-2.6.10-xen-sparse/mm/memory.c

@@ -218,6 +218,11 @@ out:
  * dst->page_table_lock is held on entry and exit,
  * but may be dropped within pmd_alloc() and pte_alloc_map().
  */
+#ifdef CONFIG_XEN_BATCH_MODE1
+#undef set_pte
+#define set_pte(pteptr, pteval)\
+    set_pte_batched(pteptr, pteval);
+#endif
 int copy_page_range(struct mm_struct *dst, struct mm_struct *src,
 			struct vm_area_struct *vma)
 {
@@ -354,8 +359,11 @@ cont_copy_pte_range_noset:
 			cond_resched_lock(&dst->page_table_lock);
 cont_copy_pmd_range:
 			src_pmd++;
-			dst_pmd++;
+			dst_pmd++;            
 		} while ((unsigned long)src_pmd & PMD_TABLE_MASK);
+#ifdef CONFIG_XEN_BATCH_MODE1
+        _flush_page_update_queue();
+#endif
 	}
 out_unlock:
 	spin_unlock(&src->page_table_lock);
@@ -445,9 +453,19 @@ static void zap_pte_range(struct mmu_gather *tlb,
 			free_swap_and_cache(pte_to_swp_entry(pte));
 		pte_clear(ptep);
 	}
+#ifdef CONFIG_XEN_BATCH_MODE1
+    _flush_page_update_queue();
+#endif
 	pte_unmap(ptep-1);
 }
 
+#ifdef CONFIG_XEN_BATCH_MODE1
+#undef set_pte
+#define set_pte(pteptr, pteval)\
+    set_pte_batched(pteptr, pteval);\
+    _flush_page_update_queue()
+#endif
+
 static void zap_pmd_range(struct mmu_gather *tlb,
 		pgd_t * dir, unsigned long address,
 		unsigned long size, struct zap_details *details)
@@ -839,6 +857,11 @@ out:
 
 EXPORT_SYMBOL(get_user_pages);
 
+#ifdef CONFIG_XEN_BATCH_MODE1
+#undef set_pte
+#define set_pte(pteptr, pteval)\
+    set_pte_batched(pteptr, pteval);
+#endif
 static void zeromap_pte_range(pte_t * pte, unsigned long address,
                                      unsigned long size, pgprot_t prot)
 {
@@ -855,7 +878,18 @@ static void zeromap_pte_range(pte_t * pte, unsigned long address,
 		address += PAGE_SIZE;
 		pte++;
 	} while (address && (address < end));
+
+#ifdef CONFIG_XEN_BATCH_MODE1
+    _flush_page_update_queue();
+#endif
+
 }
+#ifdef CONFIG_XEN_BATCH_MODE1
+#undef set_pte
+#define set_pte(pteptr, pteval)\
+    set_pte_batched(pteptr, pteval);\
+    _flush_page_update_queue()
+#endif
 
 static inline int zeromap_pmd_range(struct mm_struct *mm, pmd_t * pmd, unsigned long address,
                                     unsigned long size, pgprot_t prot)
@@ -917,6 +951,11 @@ int zeromap_page_range(struct vm_area_struct *vma, unsigned long address, unsign
  * mappings are removed. any references to nonexistent pages results
  * in null mappings (currently treated as "copy-on-access")
  */
+#ifdef CONFIG_XEN_BATCH_MODE1
+#undef set_pte
+#define set_pte(pteptr, pteval)\
+    set_pte_batched(pteptr, pteval);
+#endif
 static inline void remap_pte_range(pte_t * pte, unsigned long address, unsigned long size,
 	unsigned long pfn, pgprot_t prot)
 {
@@ -934,7 +973,16 @@ static inline void remap_pte_range(pte_t * pte, unsigned long address, unsigned
 		pfn++;
 		pte++;
 	} while (address && (address < end));
+#ifdef CONFIG_XEN_BATCH_MODE1
+    _flush_page_update_queue();
+#endif
 }
+#ifdef CONFIG_XEN_BATCH_MODE1
+#undef set_pte
+#define set_pte(pteptr, pteval)\
+    set_pte_batched(pteptr, pteval);\
+    _flush_page_update_queue()
+#endif
 
 static inline int remap_pmd_range(struct mm_struct *mm, pmd_t * pmd, unsigned long address, unsigned long size,
 	unsigned long pfn, pgprot_t prot)

tools/misc/Makefile

@@ -14,7 +14,7 @@ CFLAGS   += $(INCLUDES)
 
 HDRS     = $(wildcard *.h)
 
-TARGETS  = xenperf
+TARGETS  = xenperf xc_shadow
 
 INSTALL_BIN  = $(TARGETS) xencons
 INSTALL_SBIN = netfix xm xend xensv xenperf

tools/misc/cpuperf/cpuperf.c

@@ -246,7 +246,7 @@ int main(int argc, char **argv)
         while((cpu_mask&1)) {
             int i;
             for (i=0x300;i<0x312;i++) {
-                printf("%010llx ",cpus_rdmsr( cpu_mask, i ) );
+                printf("%010llu ",cpus_rdmsr( cpu_mask, i ) );
             }
             printf("\n");
             cpu_mask>>=1;

tools/misc/xc_shadow.c

@@ -0,0 +1,70 @@
+/* -*-  Mode:C; c-basic-offset:4; tab-width:4 -*-
+ ****************************************************************************
+ * (C) 2005 - Rolf Neugebauer - Intel Research Cambridge
+ ****************************************************************************
+ *
+ *        File: xc_shadow.c
+ *      Author: Rolf Neugebauer (rolf.neugebauer@intel.com)
+ *        Date: Mar 2005
+ * 
+ * Description: 
+ */
+
+
+#include <xc.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/mman.h>
+#include <errno.h>
+#include <string.h>
+
+void usage()
+{
+    printf("xc_shadow: -[0|1|2]\n");
+    printf("    set shadow mode\n");
+    exit(0);
+}
+
+int main(int argc, char *argv[])
+{
+    int xc_handle;
+    int mode;
+
+    if ( argc > 1 )
+    {
+        char *p = argv[1];
+        if (*p++ == '-') {
+            if (*p == '1')
+                mode = 1;
+            else if (*p == '2')
+                mode = 2;
+            else if (*p == '0')
+                mode = 0;
+            else
+                usage();
+        } else
+            usage();
+    } 
+    else
+        usage();
+
+    if ( (xc_handle = xc_interface_open()) == -1 )
+    {
+        fprintf(stderr, "Error opening xc interface: %d (%s)\n",
+                errno, strerror(errno));
+        return 1;
+    }
+
+    if ( xc_shadow_control(xc_handle,
+                           0,
+                           mode, 
+                           NULL,
+                           0,
+                           NULL) < 0 )
+    {    
+        fprintf(stderr, "Error reseting performance counters: %d (%s)\n",
+                errno, strerror(errno));
+        return 1;
+    }
+    return 0;
+}