vmmon: clean up the uptime handling

The do_gettimeofday() helper was removed by commit e4b92b108c6c
("timekeeping: remove obsolete time accessors") in v5.0-rc1 and commit
c766d1472c70 ("y2038: hide timeval/timespec/itimerval/itimerspec types") in
v5.6-rc3 hides struct timeval from kernel code.

Rather than copying what newer 15.5.x versions do, clean up the whole mess
of simultaneous crosschecking between jiffies and wall time. Apparently
this was introduced once to work around the absence of a high resolution
monotonic timer. As kernel provides one at least since version 2.6.17 and
64-bit interface for it since 3.17 and VMware is unlikely to ever release
any update for Workstation/Player 14.1.x or 12.5.x anyway, let's just drop
all of it and use ktime_get_ns() instead.

INSTALL

@@ -150,7 +150,7 @@ test build).
 When using a downloaded tarball, simply run
 
   tar -cf vmmon.tar vmmon-only
-  tar -cf vmnet.tar vmmon-only
+  tar -cf vmnet.tar vmnet-only
 
 to create the tarballs. In this case, there is no protection against
 unwanted local changes.

vmmon-only/Makefile

@@ -107,7 +107,7 @@ prebuild:: ;
 postbuild:: ;
 
 $(DRIVER_KO): prebuild
-	$(MAKE) -C $(BUILD_DIR) SUBDIRS=$$PWD SRCROOT=$$PWD/$(SRCROOT) \
+	$(MAKE) -C $(BUILD_DIR) M=$$PWD SRCROOT=$$PWD/$(SRCROOT) \
 	  MODULEBUILDDIR=$(MODULEBUILDDIR) modules
 	$(MAKE) -C $$PWD SRCROOT=$$PWD/$(SRCROOT) \
 	  MODULEBUILDDIR=$(MODULEBUILDDIR) postbuild

vmmon-only/include/hashFunc.h

@@ -251,6 +251,7 @@ static INLINE ub8 hash2(register const ub8 *k, /* the key */
   {
     /* c is reserved for the length */
   case  2: b+=k[1];
+	   /* fall through */
   case  1: a+=k[0];
     /* case 0: nothing left to add */
   }

vmmon-only/linux/driver.c

@@ -73,6 +73,9 @@ static Bool LinuxDriverCheckPadding(void);
 
 struct VMXLinuxState linuxState;
 
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 17, 0)
+typedef int vm_fault_t;
+#endif
 
 /*
  *----------------------------------------------------------------------
@@ -97,9 +100,9 @@ long LinuxDriver_Ioctl(struct file *filp, u_int iocmd,
 
 static int LinuxDriver_Close(struct inode *inode, struct file *filp);
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
-static int LinuxDriverFault(struct vm_fault *fault);
+static vm_fault_t LinuxDriverFault(struct vm_fault *fault);
 #else
-static int LinuxDriverFault(struct vm_area_struct *vma, struct vm_fault *fault);
+static vm_fault_t LinuxDriverFault(struct vm_area_struct *vma, struct vm_fault *fault);
 #endif
 static int LinuxDriverMmap(struct file *filp, struct vm_area_struct *vma);
 
@@ -594,7 +597,7 @@ LinuxDriver_Close(struct inode *inode, // IN
  *-----------------------------------------------------------------------------
  */
 
-static int
+static vm_fault_t
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
 LinuxDriverFault(struct vm_fault *fault)     //IN/OUT
 #else

vmmon-only/linux/hostif.c

@@ -116,13 +116,7 @@
  */
 #define LOCKED_PAGE_SLACK 10000
 
-static struct {
-   Atomic_uint64     uptimeBase;
-   VersionedAtomic   version;
-   uint64            monotimeBase;
-   unsigned long     jiffiesBase;
-   struct timer_list timer;
-} uptimeState;
+u64 uptime_base;
 
 /*
  * First Page Locking strategy
@@ -181,6 +175,21 @@ static struct {
 uint8 monitorIPIVector;
 uint8 hvIPIVector;
 
+static unsigned long compat_totalram_pages(void)
+{
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 0, 0)
+	return totalram_pages;
+#else
+	return totalram_pages();
+#endif
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 0, 0) && defined(VERIFY_WRITE)
+	#define write_access_ok(addr, size) access_ok(VERIFY_WRITE, addr, size)
+#else
+	#define write_access_ok(addr, size) access_ok(addr, size)
+#endif
+
 /*
  *-----------------------------------------------------------------------------
  *
@@ -1495,14 +1504,7 @@ unsigned int
 HostIF_EstimateLockedPageLimit(const VMDriver* vm,                // IN
                                unsigned int currentlyLockedPages) // IN
 {
-   /*
-    * This variable is available and exported to modules,
-    * since at least 2.6.0.
-    */
-
-   extern unsigned long totalram_pages;
-
-   unsigned int totalPhysicalPages = totalram_pages;
+   unsigned int totalPhysicalPages = compat_totalram_pages();
 
    /*
     * Use the memory information linux exports as of late for a more
@@ -1600,143 +1602,6 @@ HostIF_WaitForFreePages(unsigned int timeoutMs)  // IN:
 }
 
 
-/*
- *----------------------------------------------------------------------
- *
- * HostIFReadUptimeWork --
- *
- *      Reads the current uptime.  The uptime is based on getimeofday,
- *      which provides the needed high resolution.  However, we don't
- *      want uptime to be warped by e.g. calls to settimeofday.  So, we
- *      use a jiffies based monotonic clock to sanity check the uptime.
- *      If the uptime is more than one second from the monotonic time,
- *      we assume that the time of day has been set, and recalculate the
- *      uptime base to get uptime back on track with monotonic time.  On
- *      the other hand, we do expect jiffies based monotonic time and
- *      timeofday to have small drift (due to NTP rate correction, etc).
- *      We handle this by rebasing the jiffies based monotonic clock
- *      every second (see HostIFUptimeResyncMono).
- *
- * Results:
- *      The uptime, in units of UPTIME_FREQ.  Also returns the jiffies
- *      value that was used in the monotonic time calculation.
- *
- * Side effects:
- *      May reset the uptime base in the case gettimeofday warp was
- *      detected.
- *
- *----------------------------------------------------------------------
- */
-
-static uint64
-HostIFReadUptimeWork(unsigned long *j)  // OUT: current jiffies
-{
-   struct timeval tv;
-   uint64 monotime, uptime, upBase, monoBase;
-   int64 diff;
-   uint32 version;
-   unsigned long jifs, jifBase;
-   unsigned int attempts = 0;
-
- retry:
-   do {
-      version  = VersionedAtomic_BeginTryRead(&uptimeState.version);
-      jifs     = jiffies;
-      jifBase  = uptimeState.jiffiesBase;
-      monoBase = uptimeState.monotimeBase;
-   } while (!VersionedAtomic_EndTryRead(&uptimeState.version, version));
-
-   do_gettimeofday(&tv);
-   upBase = Atomic_Read64(&uptimeState.uptimeBase);
-
-   monotime = (uint64)(jifs - jifBase) * (UPTIME_FREQ / HZ);
-   monotime += monoBase;
-
-   uptime = tv.tv_usec * (UPTIME_FREQ / 1000000) + tv.tv_sec * UPTIME_FREQ;
-   uptime += upBase;
-
-   /*
-    * Use the jiffies based monotonic time to sanity check gettimeofday.
-    * If they differ by more than one second, assume the time of day has
-    * been warped, and use the jiffies time to undo (most of) the warp.
-    */
-
-   diff = uptime - monotime;
-   if (UNLIKELY(diff < -UPTIME_FREQ || diff > UPTIME_FREQ)) {
-      /* Compute a new uptimeBase to get uptime back on track. */
-      uint64 newUpBase = monotime - (uptime - upBase);
-
-      attempts++;
-      if (!Atomic_CMPXCHG64(&uptimeState.uptimeBase, upBase, newUpBase) &&
-          attempts < 5) {
-         /* Another thread updated uptimeBase.  Recalculate uptime. */
-         goto retry;
-      }
-      uptime = monotime;
-
-      Log("%s: detected settimeofday: fixed uptimeBase old %"FMT64"u "
-          "new %"FMT64"u attempts %u\n", __func__,
-          upBase, newUpBase, attempts);
-   }
-   *j = jifs;
-
-   return uptime;
-}
-
-
-/*
- *----------------------------------------------------------------------
- *
- * HostIFUptimeResyncMono --
- *
- *      Timer that fires ever second to resynchronize the jiffies based
- *      monotonic time with the uptime.
- *
- * Results:
- *      None
- *
- * Side effects:
- *      Resets the monotonic time bases so that jiffies based monotonic
- *      time does not drift from gettimeofday over the long term.
- *
- *----------------------------------------------------------------------
- */
-
-static void
-HostIFUptimeResyncMono(struct timer_list *timer)  // IN: ignored
-{
-   unsigned long jifs;
-   uintptr_t flags;
-
-   /*
-    * Read the uptime and the corresponding jiffies value.  This will
-    * also correct the uptime (which is based on time of day) if needed
-    * before we rebase monotonic time (which is based on jiffies).
-    */
-
-   uint64 uptime = HostIFReadUptimeWork(&jifs);
-
-   /*
-    * Every second, recalculate monoBase and jiffiesBase to squash small
-    * drift between gettimeofday and jiffies.  Also, this prevents
-    * (jiffies - jiffiesBase) wrap on 32-bits.
-    */
-
-   SAVE_FLAGS(flags);
-   CLEAR_INTERRUPTS();
-   VersionedAtomic_BeginWrite(&uptimeState.version);
-
-   uptimeState.monotimeBase = uptime;
-   uptimeState.jiffiesBase  = jifs;
-
-   VersionedAtomic_EndWrite(&uptimeState.version);
-   RESTORE_FLAGS(flags);
-
-   /* Reschedule this timer to expire in one second. */
-   mod_timer(&uptimeState.timer, jifs + HZ);
-}
-
-
 /*
  *----------------------------------------------------------------------
  *
@@ -1748,8 +1613,7 @@ HostIFUptimeResyncMono(struct timer_list *timer)  // IN: ignored
  *      None
  *
  * Side effects:
- *      Sets the initial values for the uptime state, and schedules
- *      the uptime timer.
+ *      Sets the initial value for the uptime base.
  *
  *----------------------------------------------------------------------
  */
@@ -1757,22 +1621,7 @@ HostIFUptimeResyncMono(struct timer_list *timer)  // IN: ignored
 void
 HostIF_InitUptime(void)
 {
-   struct timeval tv;
-
-   uptimeState.jiffiesBase = jiffies;
-   do_gettimeofday(&tv);
-   Atomic_Write64(&uptimeState.uptimeBase,
-                  -(tv.tv_usec * (UPTIME_FREQ / 1000000) +
-                    tv.tv_sec * UPTIME_FREQ));
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 15, 0) && !defined(timer_setup)
-   init_timer(&uptimeState.timer);
-   uptimeState.timer.function = (void *)HostIFUptimeResyncMono;
-   uptimeState.timer.data = (unsigned long)&uptimeState.timer;
-#else
-   timer_setup(&uptimeState.timer, HostIFUptimeResyncMono, 0);
-#endif
-   mod_timer(&uptimeState.timer, jiffies + HZ);
+   uptime_base = ktime_get_ns();
 }
 
 
@@ -1781,13 +1630,13 @@ HostIF_InitUptime(void)
  *
  * HostIF_CleanupUptime --
  *
- *      Cleanup uptime state, called at module unloading time.
+ *      No-op, left for backward compatibility.
  *
  * Results:
  *      None
  *
  * Side effects:
- *      Deschedule the uptime timer.
+ *      None
  *
  *----------------------------------------------------------------------
  */
@@ -1795,7 +1644,6 @@ HostIF_InitUptime(void)
 void
 HostIF_CleanupUptime(void)
 {
-   del_timer_sync(&uptimeState.timer);
 }
 
 
@@ -1819,9 +1667,10 @@ HostIF_CleanupUptime(void)
 uint64
 HostIF_ReadUptime(void)
 {
-   unsigned long jifs;
+   u64 tm;
 
-   return HostIFReadUptimeWork(&jifs);
+   tm = ktime_get_ns();
+   return (tm - uptime_base) / (NSEC_PER_SEC / UPTIME_FREQ);
 }
 
 
@@ -2165,7 +2014,7 @@ isVAReadable(VA r)  // IN:
    int ret;
 
    old_fs = get_fs();
-   set_fs(get_ds());
+   set_fs(KERNEL_DS);
    r = APICR_TO_ADDR(r, APICR_VERSION);
    ret = HostIF_CopyFromUser(&dummy, r, sizeof dummy);
    set_fs(old_fs);
@@ -2198,7 +2047,7 @@ SetVMAPICAddr(VMDriver *vm, // IN/OUT: driver state
    volatile void *hostapic;
 
    ASSERT_ON_COMPILE(APICR_SIZE <= PAGE_SIZE);
-   hostapic = (volatile void *) ioremap_nocache(ma, PAGE_SIZE);
+   hostapic = (volatile void *) ioremap(ma, PAGE_SIZE);
    if (hostapic) {
       if ((APIC_VERSIONREG(hostapic) & 0xF0) == 0x10) {
          vm->hostAPIC.base = (volatile uint32 (*)[4]) hostapic;
@@ -2366,7 +2215,7 @@ HostIF_SemaphoreWait(VMDriver *vm,   // IN:
    }
 
    old_fs = get_fs();
-   set_fs(get_ds());
+   set_fs(KERNEL_DS);
 
    {
       struct poll_wqueues table;
@@ -2495,7 +2344,7 @@ HostIF_SemaphoreSignal(uint64 *args)  // IN:
    }
 
    old_fs = get_fs();
-   set_fs(get_ds());
+   set_fs(KERNEL_DS);
 
    /*
     * Always write sizeof(uint64) bytes. This works fine for eventfd and
@@ -3155,7 +3004,7 @@ HostIF_SetFastClockRate(unsigned int rate) // IN: Frequency in Hz.
       }
    } else {
       if (linuxState.fastClockThread) {
-         force_sig(SIGKILL, linuxState.fastClockThread);
+         send_sig(SIGKILL, linuxState.fastClockThread, 1);
          kthread_stop(linuxState.fastClockThread);
 
          linuxState.fastClockThread = NULL;
@@ -3201,7 +3050,7 @@ HostIF_MapUserMem(VA addr,                  // IN: User memory virtual address
 
    ASSERT(handle);
 
-   if (!access_ok(VERIFY_WRITE, p, size)) {
+   if (!write_access_ok(p, size)) {
       printk(KERN_ERR "%s: Couldn't verify write to uva 0x%p with size %"
              FMTSZ"u\n", __func__, p, size);
 

vmnet-only/Makefile

@@ -107,7 +107,7 @@ prebuild:: ;
 postbuild:: ;
 
 $(DRIVER_KO): prebuild
-	$(MAKE) -C $(BUILD_DIR) SUBDIRS=$$PWD SRCROOT=$$PWD/$(SRCROOT) \
+	$(MAKE) -C $(BUILD_DIR) M=$$PWD SRCROOT=$$PWD/$(SRCROOT) \
 	  MODULEBUILDDIR=$(MODULEBUILDDIR) modules
 	$(MAKE) -C $$PWD SRCROOT=$$PWD/$(SRCROOT) \
 	  MODULEBUILDDIR=$(MODULEBUILDDIR) postbuild

vmnet-only/bridge.c

@@ -66,7 +66,7 @@
 #endif
 
 #if LOGLEVEL >= 4
-static struct timeval vnetTime;
+static u64 vnetTime;
 #endif
 
 typedef struct VNetBridge VNetBridge;
@@ -693,7 +693,7 @@ VNetBridgeReceiveFromVNet(VNetJack        *this, // IN: jack
 
 	 netif_rx_ni(clone);
 #	 if LOGLEVEL >= 4
-	 do_gettimeofday(&vnetTime);
+	 vnetTime = ktime_get_ns();
 #	 endif
       }
    }
@@ -1490,12 +1490,11 @@ VNetBridgeReceiveFromDev(struct sk_buff *skb,         // IN: packet to receive
 
 #  if LOGLEVEL >= 4
    {
-      struct timeval now;
-      do_gettimeofday(&now);
+      u64 now;
+
+      now = ktime_get_ns();
       LOG(3, (KERN_DEBUG "bridge-%s: time %d\n",
-	      bridge->name,
-	      (int)((now.tv_sec * 1000000 + now.tv_usec)
-                    - (vnetTime.tv_sec * 1000000 + vnetTime.tv_usec))));
+	      bridge->name, (int)((now - vnetTime) / NSEC_PER_USEC)));
    }
 #  endif
 

vmnet-only/procfs.c

@@ -161,6 +161,14 @@ VNetProcOpen(struct inode *inode,   // IN:
 }
 
 /* Our procfs callbacks.  We only need to specialize open. */
+#ifdef DEFINE_PROC_SHOW_ATTRIBUTE
+static const struct proc_ops fops = {
+   .proc_open    = VNetProcOpen,
+   .proc_read    = seq_read,
+   .proc_lseek   = seq_lseek,
+   .proc_release = single_release,
+};
+#else
 static struct file_operations fops = {
    .open    = VNetProcOpen,
    .read    = seq_read,
@@ -168,6 +176,7 @@ static struct file_operations fops = {
    .release = single_release,
 };
 #endif
+#endif
 
 
 /*

vmnet-only/userif.c

@@ -78,11 +78,37 @@ static int  VNetUserIfSetUplinkState(VNetPort *port, uint8 linkUp);
 extern unsigned int  vnet_max_qlen;
 
 #if COMPAT_LINUX_VERSION_CHECK_LT(3, 2, 0)
-#   define compat_kmap(page) kmap(page)
-#   define compat_kunmap(page) kunmap(page)
+#   define compat_kmap_frag(frag) kmap((frag)->page)
+#   define compat_kunmap_frag(page) kunmap((frag)->page)
 #else
-#   define compat_kmap(page) kmap((page).p)
-#   define compat_kunmap(page) kunmap((page).p)
+#   define compat_kmap_frag(frag) kmap(skb_frag_page(frag))
+#   define compat_kunmap_frag(frag) kunmap(skb_frag_page(frag))
+#endif
+
+static unsigned int compat_skb_frag_size(const skb_frag_t *frag)
+{
+#if COMPAT_LINUX_VERSION_CHECK_LT(3, 2, 0)
+	return frag->size;
+#else
+	return skb_frag_size(frag);
+#endif
+}
+
+static unsigned int compat_skb_frag_off(const skb_frag_t *frag)
+{
+#if COMPAT_LINUX_VERSION_CHECK_LT(5, 4, 0) && \
+	!(defined(CONFIG_SUSE_VERSION) && CONFIG_SUSE_VERSION == 15 && \
+	  defined(CONFIG_SUSE_PATCHLEVEL) && CONFIG_SUSE_PATCHLEVEL >= 2)
+	return frag->page_offset;
+#else
+	return skb_frag_off(frag);
+#endif
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 0, 0) && defined(VERIFY_WRITE)
+	#define write_access_ok(addr, size) access_ok(VERIFY_WRITE, addr, size)
+#else
+	#define write_access_ok(addr, size) access_ok(addr, size)
 #endif
 
 /*
@@ -142,7 +168,7 @@ VNetUserIfMapPtr(VA uAddr,        // IN: pointer to user memory
                  struct page **p, // OUT: locked page
                  void **ptr)      // OUT: kernel mapped pointer
 {
-   if (!access_ok(VERIFY_WRITE, (void *)uAddr, size) ||
+   if (!write_access_ok((void *)uAddr, size) ||
        (((uAddr + size - 1) & ~(PAGE_SIZE - 1)) !=
         (uAddr & ~(PAGE_SIZE - 1)))) {
       return -EINVAL;
@@ -559,20 +585,21 @@ VNetCsumCopyDatagram(const struct sk_buff *skb,	// IN: skb to copy
    for (frag = skb_shinfo(skb)->frags;
 	frag != skb_shinfo(skb)->frags + skb_shinfo(skb)->nr_frags;
 	frag++) {
-      if (frag->size > 0) {
+      if (compat_skb_frag_size(frag) > 0) {
 	 unsigned int tmpCsum;
 	 const void *vaddr;
 
-	 vaddr = compat_kmap(frag->page);
-	 tmpCsum = csum_and_copy_to_user(vaddr + frag->page_offset,
-					 curr, frag->size, 0, &err);
-	 compat_kunmap(frag->page);
+	 vaddr = compat_kmap_frag(frag);
+	 tmpCsum = csum_and_copy_to_user(vaddr + compat_skb_frag_off(frag),
+					 curr, compat_skb_frag_size(frag), 0,
+					 &err);
+	 compat_kunmap_frag(frag);
 
 	 if (err) {
 	    return err;
 	 }
 	 csum = csum_block_add(csum, tmpCsum, curr - buf);
-	 curr += frag->size;
+	 curr += compat_skb_frag_size(frag);
       }
    }