Merge branch 'scale-amd64' of ssh://amsterdam.csail.mit.edu/home/am0/6.828/xv6 into scale-amd64

cdaa456b · Frans Kaashoek · 6fb270fb · ed844a14 · cdaa456b · cdaa456b
--- a/e1000.c
+++ b/e1000.c
@@ -279,9 +279,10 @@ e1000attach(struct pci_func *pcif)
  r = eeprom_read((u16*)e1000.hwaddr, EEPROM_OFF_MACADDR, 3);
  if (r < 0)
    return 0;
-  cprintf("%x:%x:%x:%x:%x:%x\n",
+  if (VERBOSE)
-          e1000.hwaddr[0], e1000.hwaddr[1], e1000.hwaddr[2],
+      cprintf("%x:%x:%x:%x:%x:%x\n",
-          e1000.hwaddr[3], e1000.hwaddr[4], e1000.hwaddr[5]);
+              e1000.hwaddr[0], e1000.hwaddr[1], e1000.hwaddr[2],
+              e1000.hwaddr[3], e1000.hwaddr[4], e1000.hwaddr[5]);
  u32 ralow = ((u32) e1000.hwaddr[0]) | ((u32) e1000.hwaddr[1] << 8) | 
      ((u32) e1000.hwaddr[2] << 16) | ((u32) e1000.hwaddr[3] << 24);

--- a/file.h
+++ b/file.h
@@ -50,3 +50,4 @@ extern struct devsw devsw[];
 #define CONSOLE 1
 #define NETIF   2
+#define SAMPLER 3
--- a/init.c
+++ b/init.c
@@ -21,9 +21,10 @@ main(void)
  if (mknod("netif", 2, 1) < 0)
      printf(2, "init: mknod netif failed\n");
+  if (mknod("sampler", 3, 1) < 0)
+      printf(2, "init: mknod sampler failed\n");
  for(;;){
-    printf(1, "init: starting sh\n");
    pid = fork(0);
    if(pid < 0){
      printf(1, "init: fork failed\n");

--- a/kalloc.c
+++ b/kalloc.c
@@ -271,7 +271,8 @@ initkalloc(u64 mbaddr)
    }
  }
-  cprintf("%lu mbytes\n", membytes / (1<<20));
+  if (VERBOSE)
+    cprintf("%lu mbytes\n", membytes / (1<<20));
  n = membytes / NCPU;
  if (n & (PGSIZE-1))
    n = PGROUNDDOWN(n);

--- a/main.c
+++ b/main.c
@@ -109,7 +109,8 @@ cmain(u64 mbmagic, u64 mbaddr)
  initpci();
  initnet();
-  cprintf("ncpu %d %lu MHz\n", ncpu, cpuhz / 1000000);
+  if (VERBOSE)
+    cprintf("ncpu %d %lu MHz\n", ncpu, cpuhz / 1000000);
  inituser();      // first user process
  bootothers();    // start other processors

--- a/param.h
+++ b/param.h
@@ -15,6 +15,7 @@
 #define QUANTUM      10  // scheduling time quantum and tick length (in msec)
 #define WQSHIFT       4  // 2^WORKSHIFT work queue slots
 #define VICTIMAGE 1000000 // cycles a proc executes before an eligible victim
+#define VERBOSE       0  // print kernel diagnostics
 #if defined(HW_josmp)
 #define NCPU         16  // maximum number of CPUs
 #define MTRACE       0

--- a/pci.c
+++ b/pci.c
@@ -246,10 +246,11 @@ pci_func_enable(struct pci_func *f)
              PCI_VENDOR(f->dev_id), PCI_PRODUCT(f->dev_id),
              regnum, base, size);
  }
-  cprintf("PCI function %x:%x.%d (%x:%x) enabled\n",
+  if (VERBOSE)
-          f->bus->busno, f->dev, f->func,
+    cprintf("PCI function %x:%x.%d (%x:%x) enabled\n",
-          PCI_VENDOR(f->dev_id), PCI_PRODUCT(f->dev_id));
+            f->bus->busno, f->dev, f->func,
+            PCI_VENDOR(f->dev_id), PCI_PRODUCT(f->dev_id));
 }
 void

--- a/sampler.c
+++ b/sampler.c
@@ -8,6 +8,7 @@
 #include "bits.h"
 #include "amd64.h"
 #include "cpu.h"
+#include "sampler.h"
 static const u64 debug_sel = 
  0UL << 32 |
@@ -27,14 +28,9 @@ struct pmu {
 };
 struct pmu pmu;
-struct pmuevent {
-  u64 rip;
-};
 struct pmulog {
-  u64 head;
+  u64 count;
-  u64 tail;
+  u64 capacity;
-  u64 size;
  struct pmuevent *event;
  __padout__;
 } __mpalign__;
@@ -79,9 +75,9 @@ sampdump(void)
 {
  for (int c = 0; c < NCPU; c++) {
    struct pmulog *l = &pmulog[c];    
-    cprintf("%u samples %lu\n", c, l->head - l->tail);
+    cprintf("%u samples %lu\n", c, l->count);
-    for (u64 i = l->tail; i < l->tail+4 && i < l->head; i++)
+    for (u64 i = 0; i < 4 && i < l->count; i++)
-      cprintf(" %lx\n", l->event[i % l->size].rip);
+      cprintf(" %lx\n", l->event[i].rip);
  }
 }
@@ -111,11 +107,11 @@ samplog(struct trapframe *tf)
  struct pmulog *l;
  l = &pmulog[cpunum()];
-  if ((l->head - l->tail) == l->size)
+  if (l->count == l->capacity)
    return 0;
-  l->event[l->head % l->size].rip = tf->rip;
+  l->event[l->count].rip = tf->rip;
-  l->head++;
+  l->count++;
  return 1;
 }
@@ -139,6 +135,75 @@ sampintr(struct trapframe *tf)
  return 1;
 }
+static int
+readlog(char *dst, u32 off, u32 n)
+{
+  struct pmulog *q = &pmulog[NCPU];
+  struct pmulog *p;
+  int ret = 0;
+  u64 cur = 0;
+  for (p = &pmulog[0]; p != q && n != 0; p++) {
+    u64 len = p->count * sizeof(struct pmuevent);
+    char *buf = (char*)p->event;
+    if (cur <= off && off < cur+len) {
+      u64 boff = off-cur;
+      u64 cc = MIN(len-boff, n);
+      memmove(dst, buf+boff, cc);
+      n -= cc;
+      ret += cc;
+      off += cc;
+      dst += cc;
+    }
+    cur += len;
+  }
+  return ret;
+}
+static int
+sampread(struct inode *ip, char *dst, u32 off, u32 n)
+{
+  struct pmulog *q = &pmulog[NCPU];
+  struct pmulog *p;
+  struct logheader *hdr;
+  u64 hdrlen;
+  int ret;
+  int i;
+  ret = 0;
+  hdrlen = sizeof(*hdr) + sizeof(hdr->cpu[0])*NCPU;
+  if (off < hdrlen) {
+    u64 len = hdrlen;
+    u64 cc;
+    hdr = kmalloc(len);
+    if (hdr == NULL)
+      return -1;
+    hdr->ncpus = NCPU;
+    for (p = &pmulog[0]; p != q; p++) {
+      u64 sz = p->count * sizeof(struct pmuevent);
+      hdr->cpu[i].offset = len;
+      hdr->cpu[i].size = sz;
+      len += sz;
+    }
+    cc = MIN(hdrlen-off, n);
+    memmove(dst, (void*)hdr + off, cc);
+    kmfree(hdr);
+    n -= cc;
+    ret += cc;
+    off += cc;
+    dst += cc;
+  }
+  if (off > hdrlen)
+    ret += readlog(dst, off-hdrlen, n);
+  return ret;
+}
 void
 initsamp(void)
 {
@@ -148,7 +213,8 @@ initsamp(void)
    name[3] = 0;
    cpuid(0, 0, &name[0], &name[2], &name[1]);
-    cprintf("%s\n", s);
+    if (VERBOSE)
+      cprintf("%s\n", s);
    if (!strcmp(s, "AuthenticAMD"))
      pmu = amdpmu;
    else if (!strcmp(s, "GenuineIntel"))
@@ -165,5 +231,8 @@ initsamp(void)
  if (p == NULL)
    panic("initprof: ksalloc");
  pmulog[cpunum()].event = p;
-  pmulog[cpunum()].size = PERFSIZE / sizeof(struct pmuevent);
+  pmulog[cpunum()].capacity = PERFSIZE / sizeof(struct pmuevent);
+  devsw[SAMPLER].write = NULL;
+  devsw[SAMPLER].read = sampread;
 }
--- a/sampler.h
+++ b/sampler.h
+struct pmuevent {
+  u64 rip;
+};
+struct logheader {
+  u32 ncpus;
+  struct {
+    u64 offset;
+    u64 size;
+  } cpu[];
+};