path: root/vmmon.patch

--- a/vmmon/Makefile
+++ b/vmmon/Makefile
@@ -43,7 +43,11 @@
 endif
 
 
+ifdef KVERSION
+VM_UNAME = $(KVERSION)
+else
 VM_UNAME = $(shell uname -r)
+endif
 
 # Header directory for the running kernel
 ifdef LINUXINCLUDE
@@ -100,6 +104,13 @@ auto-build: $(DRIVER_KO)
 $(DRIVER): $(DRIVER_KO)
 	if [ $< -nt $@ ] || [ ! -e $@ ] ; then cp -f $< $@; fi
 
+# Use SUBDIRS on 2.x, 3.x, 4.x.  Use M on newer kernels.
+ifeq ($(filter-out 2 3 4,$(firstword $(subst ., ,$(VM_UNAME)))),)
+DIRVAR := SUBDIRS
+else
+DIRVAR := M
+endif
+
 # Pass gcc version down the chain, so we can detect if kernel attempts to use unapproved compiler
 VM_CCVER := $(VMCCVER)
 export VM_CCVER
@@ -117,7 +128,7 @@ prebuild:: ;
 postbuild:: ;
 
 $(DRIVER_KO): prebuild
-	$(MAKE) -C $(BUILD_DIR) SUBDIRS=$$PWD SRCROOT=$$PWD/$(SRCROOT) \
+	$(MAKE) -C $(BUILD_DIR) $(DIRVAR)=$$PWD SRCROOT=$$PWD/$(SRCROOT) \
 	  MODULEBUILDDIR=$(MODULEBUILDDIR) modules
 	$(MAKE) -C $$PWD SRCROOT=$$PWD/$(SRCROOT) \
 	  MODULEBUILDDIR=$(MODULEBUILDDIR) postbuild
--- a/vmmon/include/pgtbl.h
+++ b/vmmon/include/pgtbl.h
@@ -91,7 +91,11 @@
          } else {
             pte_t *pte;
 
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6,5,0)
+            pte = pte_offset_kernel(pmd, addr);
+#else
             pte = pte_offset_map(pmd, addr);
+#endif
             if (pte_present(*pte) == 0) {
                pte_unmap(pte);
                return INVALID_MPN;
--- a/vmmon/linux/driver.c
+++ b/vmmon/linux/driver.c
@@ -104,7 +104,9 @@ long LinuxDriver_Ioctl(struct file *filp
 
 static int LinuxDriver_Close(struct inode *inode, struct file *filp);
 static unsigned int LinuxDriverPoll(struct file *file, poll_table *wait);
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 17, 0)
+static vm_fault_t LinuxDriverFault(struct vm_fault *fault);
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
 static int LinuxDriverFault(struct vm_fault *fault);
 #elif defined(VMW_NOPAGE_2624)
 static int LinuxDriverFault(struct vm_area_struct *vma, struct vm_fault *fault);
@@ -115,7 +117,7 @@ static struct page *LinuxDriverNoPage(st
 #endif
 static int LinuxDriverMmap(struct file *filp, struct vm_area_struct *vma);
 
-static void LinuxDriverPollTimeout(unsigned long clientData);
+static void LinuxDriverPollTimeout(struct timer_list *clientData);
 static unsigned int LinuxDriverEstimateTSCkHz(void);
 
 static struct vm_operations_struct vmuser_mops = {
@@ -227,7 +229,7 @@ LinuxDriverEstimateTSCkHz(void)
  *----------------------------------------------------------------------
  */
 static void
-LinuxDriverEstimateTSCkHzDeferred(unsigned long data)
+LinuxDriverEstimateTSCkHzDeferred(struct timer_list *data)
 {
    LinuxDriverEstimateTSCkHz();
 }
@@ -265,9 +267,7 @@ LinuxDriverInitTSCkHz(void)
    }
 
    Vmx86_ReadTSCAndUptime(&tsckHzStartTime);
-   tscTimer.function = LinuxDriverEstimateTSCkHzDeferred;
    tscTimer.expires  = jiffies + 4 * HZ;
-   tscTimer.data     = 0;
    add_timer(&tscTimer);
 }
 
@@ -275,7 +275,7 @@
 /*
  *----------------------------------------------------------------------
  *
- * init_module --
+ * LinuxDriverInit --
  *
  *      linux module entry point. Called by /sbin/insmod command
  *
@@ -288,7 +288,7 @@
  */
 
 int
-init_module(void)
+LinuxDriverInit(void)
 {
    int retval;
 
@@ -309,9 +309,13 @@
     */
 
    init_waitqueue_head(&linuxState.pollQueue);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 15, 0) && !defined(timer_setup)
    init_timer(&linuxState.pollTimer);
    linuxState.pollTimer.data = 0;
-   linuxState.pollTimer.function = LinuxDriverPollTimeout;
+   linuxState.pollTimer.function = (void *)LinuxDriverPollTimeout;
+#else
+   timer_setup(&linuxState.pollTimer, LinuxDriverPollTimeout, 0);
+#endif
 
    linuxState.fastClockThread = NULL;
    linuxState.fastClockFile = NULL;
@@ -360,7 +364,13 @@
        linuxState.deviceName, linuxState.major, linuxState.minor);
 
    HostIF_InitUptime();
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 15, 0) && !defined(timer_setup)
    init_timer(&tscTimer);
+   tscTimer.function = (void *)LinuxDriverEstimateTSCkHzDeferred;
+   tscTimer.data = (unsigned long)&tscTimer;
+#else
+   timer_setup(&tscTimer, LinuxDriverEstimateTSCkHzDeferred, 0);
+#endif
    LinuxDriverInitTSCkHz();
    Vmx86_InitIDList();
 
@@ -372,7 +382,7 @@
 /*
  *----------------------------------------------------------------------
  *
- * cleanup_module --
+ * LinuxDriverExit --
  *
  *      Called by /sbin/rmmod
  *
@@ -381,7 +391,7 @@
  */
 
 void
-cleanup_module(void)
+LinuxDriverExit(void)
 {
    /*
     * XXX smp race?
@@ -741,14 +751,12 @@ void
 LinuxDriverWakeUp(Bool selective)  // IN:
 {
    if (selective && linuxState.pollList != NULL) {
-      struct timeval tv;
       VmTimeType now;
       VMLinux *p;
       VMLinux *next;
 
       HostIF_PollListLock(1);
-      do_gettimeofday(&tv);
-      now = tv.tv_sec * 1000000ULL + tv.tv_usec;
+      now = ktime_get_ns() / NSEC_PER_USEC;
 
       for (p = linuxState.pollList; p != NULL; p = next) {
          next = p->pollForw;
@@ -815,12 +823,10 @@ LinuxDriverPoll(struct file *filp,  // I
       }
    } else {
       if (linuxState.fastClockThread && vmLinux->pollTimeoutPtr != NULL) {
-         struct timeval tv;
+         u64 now = ktime_get_ns() / NSEC_PER_USEC;
 
-         do_gettimeofday(&tv);
          poll_wait(filp, &vmLinux->pollQueue, wait);
-         vmLinux->pollTime = *vmLinux->pollTimeoutPtr +
-                                       tv.tv_sec * 1000000ULL + tv.tv_usec;
+         vmLinux->pollTime = *vmLinux->pollTimeoutPtr + now;
          if (vmLinux->pollBack == NULL) {
             HostIF_PollListLock(2);
             if (vmLinux->pollBack == NULL) {
@@ -858,7 +864,7 @@ LinuxDriverPoll(struct file *filp,  // I
  */
 
 static void
-LinuxDriverPollTimeout(unsigned long clientData)  // IN:
+LinuxDriverPollTimeout(struct timer_list *clientData)  // IN:
 {
    LinuxDriverWakeUp(FALSE);
 }
@@ -884,7 +890,12 @@ LinuxDriverPollTimeout(unsigned long cli
  */
 
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
-static int
+static
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 17, 0)
+vm_fault_t
+#else
+int
+#endif
 LinuxDriverFault(struct vm_fault *fault)     //IN/OUT
 #elif defined(VMW_NOPAGE_2624)
 static int LinuxDriverFault(struct vm_area_struct *vma, //IN
@@ -1067,7 +1078,11 @@ LinuxDriverMmap(struct file *filp,
       return err;
    }
    /* Clear VM_IO, otherwise SuSE's kernels refuse to do get_user_pages */
+#if COMPAT_LINUX_VERSION_CHECK_LT(6, 3, 0)
    vma->vm_flags &= ~VM_IO;
+#else
+   vm_flags_clear(vma, VM_IO);
+#endif
 
    return 0;
 }
@@ -2106,3 +2121,5 @@
  * by default (i.e., neither mkinitrd nor modprobe will accept it).
  */
 MODULE_INFO(supported, "external");
+module_init(LinuxDriverInit);
+module_exit(LinuxDriverExit);
--- a/vmmon/linux/hostif.c
+++ b/vmmon/linux/hostif.c
@@ -68,6 +68,8 @@
 #endif
 
 #include <asm/io.h>
+#include <asm/tlbflush.h>
+#include <asm/irq_vectors.h>
 #include <asm/uaccess.h>
 #include <linux/mc146818rtc.h>
 #include <linux/capability.h>
@@ -77,6 +79,7 @@
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
 #include <linux/taskstats_kern.h> // For linux/sched/signal.h without version check
 #endif
+#include <linux/eventfd.h>
 
 #include "vmware.h"
 #include "x86apic.h"
@@ -118,6 +121,13 @@
 #   define close_rtc(filp, files) filp_close(filp, files)
 #endif
 
+/* task's state is read-once rather than volatile from 5.14-rc2. */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 15, 0) || defined(get_current_state)
+#define get_task_state(task) READ_ONCE((task)->__state)
+#else
+#define get_task_state(task) ((task)->state)
+#endif
+
 #define UPTIME_FREQ CONST64(1000000)
 
 /*
@@ -477,7 +487,7 @@
    while ((vcpuid = VCPUSet_FindFirst(&req)) != VCPUID_INVALID) {
       struct task_struct *t = vm->vmhost->vcpuSemaTask[vcpuid];
       VCPUSet_Remove(&req, vcpuid);
-      if (t && (t->state & TASK_INTERRUPTIBLE)) {
+      if (t && (get_task_state(t) & TASK_INTERRUPTIBLE)) {
          wake_up_process(t);
       }
    }
@@ -629,6 +632,15 @@ HostIF_FastClockUnlock(int callerID) //
    MutexUnlock(&fastClockMutex, callerID);
 }
 
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0)
+static int crosspage_set_exec(pte_t *pte, unsigned long addr, void *data)
+{
+	struct page *p = data;
+
+	set_pte(pte, mk_pte(p, VM_PAGE_KERNEL_EXEC));
+	return 0;
+}
+#endif
 
 /*
  *-----------------------------------------------------------------------------
@@ -695,7 +707,29 @@ HostIF_PollListUnlock(int callerID) // I
 static void *
 MapCrossPage(struct page *p)  // IN:
 {
+#if COMPAT_LINUX_VERSION_CHECK_LT(5, 8, 0)
    return vmap(&p, 1, VM_MAP, VM_PAGE_KERNEL_EXEC);
+#else
+   void *addr;
+
+   addr = vmap(&p, 1, VM_MAP, VM_PAGE_KERNEL_EXEC);
+   if (!addr)
+	   return NULL;
+
+   /* Starting with 5.8, vmap() always sets the NX bit, but the cross
+    * page needs to be executable. */
+   if (apply_to_page_range(current->mm, (unsigned long)addr, PAGE_SIZE,
+			   crosspage_set_exec, p)) {
+      vunmap(addr);
+      return NULL;
+   }
+
+   preempt_disable();
+   __flush_tlb_all();
+   preempt_enable();
+
+   return addr;
+#endif
 }
 
 
@@ -1164,16 +1198,7 @@ HostIFGetUserPages(void *uvAddr,
 {
    int retval;
 
-   down_read(&current->mm->mmap_sem);
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 9, 0)
-   retval = get_user_pages((unsigned long)uvAddr, numPages, 0, ppages, NULL);
-#elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
-   retval = get_user_pages((unsigned long)uvAddr, numPages, 0, 0, ppages, NULL);
-#else
-   retval = get_user_pages(current, current->mm, (unsigned long)uvAddr,
-                           numPages, 0, 0, ppages, NULL);
-#endif
-   up_read(&current->mm->mmap_sem);
+   retval = get_user_pages_fast((unsigned long)uvAddr, numPages, 0, ppages);
 
    return retval != numPages;
 }
@@ -1574,9 +1599,13 @@ HostIF_EstimateLockedPageLimit(const VMD
     * since at least 2.6.0.
     */
 
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 0, 0)
    extern unsigned long totalram_pages;
 
    unsigned int totalPhysicalPages = totalram_pages;
+#else
+   unsigned int totalPhysicalPages = totalram_pages();
+#endif
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 28)
    return MemDefaults_CalcMaxLockedPages(totalPhysicalPages);
@@ -1594,17 +1623,36 @@ HostIF_EstimateLockedPageLimit(const VMD
    unsigned int reservedPages = MEMDEFAULTS_MIN_HOST_PAGES;
    unsigned int hugePages = (vm == NULL) ? 0 :
       BYTES_2_PAGES(vm->memInfo.hugePageBytes);
-   unsigned int lockedPages = global_page_state(NR_PAGETABLE) +
-                              global_page_state(NR_SLAB_UNRECLAIMABLE) +
-                              global_page_state(NR_UNEVICTABLE) +
-                              hugePages + reservedPages;
-   unsigned int anonPages =
+   unsigned int lockedPages = hugePages + reservedPages;
+   unsigned int anonPages;
+   unsigned int swapPages = BYTES_2_PAGES(linuxState.swapSize);
+
+   /* global_page_state is global_zone_page_state in 4.14. */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 14, 0)
+   lockedPages += global_zone_page_state(NR_PAGETABLE);
+#else
+   lockedPages += global_page_state(NR_PAGETABLE);
+#endif
+   /* NR_SLAB_* moved from zone to node in 4.13. */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 9, 0)
+   lockedPages += global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B);
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 13, 0)
+   lockedPages += global_node_page_state(NR_SLAB_UNRECLAIMABLE);
+#else
+   lockedPages += global_page_state(NR_SLAB_UNRECLAIMABLE);
+#endif
+   /* NR_UNEVICTABLE moved from global to node in 4.8. */
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 8, 0)
-      global_page_state(NR_ANON_MAPPED);
+   lockedPages += global_node_page_state(NR_UNEVICTABLE);
 #else
-      global_page_state(NR_ANON_PAGES);
+   lockedPages += global_page_state(NR_UNEVICTABLE);
+#endif
+   /* NR_ANON_MAPPED moved & changed name in 4.8. */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 8, 0)
+   anonPages = global_node_page_state(NR_ANON_MAPPED);
+#else
+   anonPages = global_page_state(NR_ANON_PAGES);
 #endif
-   unsigned int swapPages = BYTES_2_PAGES(linuxState.swapSize);
 
    if (anonPages > swapPages) {
       lockedPages += anonPages - swapPages; 
@@ -1664,6 +1712,49 @@ HostIF_WaitForFreePages(unsigned int tim
 /*
  *----------------------------------------------------------------------
  *
+ * HostIFGetTime --
+ *
+ *      Reads the current time in UPTIME_FREQ units.
+ *
+ * Results:
+ *      The uptime, in units of UPTIME_FREQ.
+ *
+ * Side effects:
+ *      None.
+ *
+ *----------------------------------------------------------------------
+ */
+
+static uint64
+HostIFGetTime(void)
+{
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 0, 0)
+   struct timeval tv;
+
+   do_gettimeofday(&tv);
+   return tv.tv_usec * (UPTIME_FREQ / 1000000) + tv.tv_sec * UPTIME_FREQ;
+#else
+   struct timespec64 now;
+
+   /*
+    * Use raw time used by Posix timers.  This time is not affected by
+    * NTP adjustments, so it may drift from real time and monotonic time,
+    * but it will stay in sync with other timers.
+    */
+   ktime_get_raw_ts64(&now);
+   /*
+    * UPTIME_FREQ resolution is lower than tv_nsec,
+    * so we have to do division...
+    */
+   ASSERT_ON_COMPILE(1000000000 % UPTIME_FREQ == 0);
+   return now.tv_nsec / (1000000000 / UPTIME_FREQ) + now.tv_sec * UPTIME_FREQ;
+#endif
+}
+
+
+/*
+ *----------------------------------------------------------------------
+ *
  * HostIFReadUptimeWork --
  *
  *      Reads the current uptime.  The uptime is based on getimeofday,
@@ -1692,16 +1783,12 @@ HostIF_WaitForFreePages(unsigned int tim
 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;
 
-   /* Assert that HostIF_InitUptime has been called. */
-   ASSERT(uptimeState.timer.function);
-
  retry:
    do {
       version  = VersionedAtomic_BeginTryRead(&uptimeState.version);
@@ -1710,13 +1797,12 @@ HostIFReadUptimeWork(unsigned long *j)
       monoBase = uptimeState.monotimeBase;
    } while (!VersionedAtomic_EndTryRead(&uptimeState.version, version));
 
-   do_gettimeofday(&tv);
+   uptime = HostIFGetTime();
    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;
    
    /* 
@@ -1767,7 +1853,7 @@ HostIFReadUptimeWork(unsigned long *j)
  */
 
 static void
-HostIFUptimeResyncMono(unsigned long data)  // IN: ignored
+HostIFUptimeResyncMono(struct timer_list *timer)  // IN: ignored
 {
    unsigned long jifs;
    uintptr_t flags;
@@ -1821,16 +1907,19 @@ HostIFUptimeResyncMono(unsigned long dat
 void
 HostIF_InitUptime(void)
 {
-   struct timeval tv;
+   uint64 tm;
 
    uptimeState.jiffiesBase = jiffies;
-   do_gettimeofday(&tv);
-   Atomic_Write64(&uptimeState.uptimeBase, 
-                  -(tv.tv_usec * (UPTIME_FREQ / 1000000) + 
-                    tv.tv_sec * UPTIME_FREQ));
+   tm = HostIFGetTime();
+   Atomic_Write64(&uptimeState.uptimeBase, -tm);
 
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 15, 0) && !defined(timer_setup)
    init_timer(&uptimeState.timer);
-   uptimeState.timer.function = HostIFUptimeResyncMono;
+   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);
 }
 
@@ -2246,16 +2335,26 @@ HostIF_VMLockIsHeld(VMDriver *vm) // IN
 static Bool
 isVAReadable(VA r)  // IN:
 {
-   mm_segment_t old_fs;
    uint32 dummy;
    int ret;
    
+#if defined(HAVE_GET_KERNEL_NOFAULT) || LINUX_VERSION_CODE >= KERNEL_VERSION(5, 17, 0)
+   /*
+    * Exists from 5.10, first indicated by HAVE_GET_KERNEL_NOFAULT,
+    * and from post-5.17 just existing everywhere.
+    */
+   ret = get_kernel_nofault(dummy, (void *)r);
+#else
+   {
+   mm_segment_t old_fs;
+ 
    old_fs = get_fs();
-   set_fs(get_ds());
+   set_fs(KERNEL_DS);
    r = APICR_TO_ADDR(r, APICR_VERSION);
    ret = HostIF_CopyFromUser(&dummy, (void*)r, sizeof(dummy));
    set_fs(old_fs);
-
+   }
+#endif
    return ret == 0;
 }
 
@@ -2284,7 +2379,7 @@ SetVMAPICAddr(VMDriver *vm, // IN/OUT: d
    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;
@@ -2440,7 +2535,6 @@ HostIF_SemaphoreWait(VMDriver *vm,   //
                      uint64 *args)   // IN:
 {
    struct file *file;
-   mm_segment_t old_fs;
    int res;
    int waitFD = args[0];
    int timeoutms = args[2];
@@ -2451,22 +2545,19 @@ HostIF_SemaphoreWait(VMDriver *vm,   //
       return MX_WAITERROR;
    }
 
-   old_fs = get_fs();
-   set_fs(get_ds());
-
    {
       struct poll_wqueues table;
       unsigned int mask;
       
       poll_initwait(&table);
-      current->state = TASK_INTERRUPTIBLE;
+      __set_current_state(TASK_INTERRUPTIBLE);
       mask = file->f_op->poll(file, &table.pt);
       if (!(mask & (POLLIN | POLLERR | POLLHUP))) {
          vm->vmhost->vcpuSemaTask[vcpuid] = current;
          schedule_timeout(timeoutms * HZ / 1000);  // convert to Hz
          vm->vmhost->vcpuSemaTask[vcpuid] = NULL;
       }
-      current->state = TASK_RUNNING;
+      __set_current_state(TASK_RUNNING);
       poll_freewait(&table);
    }
 
@@ -2475,9 +2566,11 @@ HostIF_SemaphoreWait(VMDriver *vm,   //
     * the code to happily deal with a pipe or an eventfd. We only care about
     * reading no bytes (EAGAIN - non blocking fd) or sizeof(uint64).
     */
-
-   res = file->f_op->read(file, (char *) &value, sizeof value, &file->f_pos);
-
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 14, 0)
+   res = kernel_read(file, file->f_pos, (char *)&value, sizeof value);
+#else
+   res = kernel_read(file, &value, sizeof value, &file->f_pos);
+#endif
    if (res == sizeof value) {
       res = MX_WAITNORMAL;
    } else {
@@ -2486,7 +2579,6 @@ HostIF_SemaphoreWait(VMDriver *vm,   //
       }
    }
 
-   set_fs(old_fs);
    fput(file);
 
    /*
@@ -2539,7 +2638,7 @@
    FOR_EACH_VCPU_IN_SET(vcs, vcpuid) {
       struct task_struct *t = vm->vmhost->vcpuSemaTask[vcpuid];
       vm->vmhost->vcpuSemaTask[vcpuid] = NULL;
-      if (t && (t->state & TASK_INTERRUPTIBLE)) {
+      if (t && (get_task_state(t) & TASK_INTERRUPTIBLE)) {
          wake_up_process(t);
       }
    } ROF_EACH_VCPU_IN_SET();
@@ -2569,8 +2668,8 @@ HostIF_SemaphoreForceWakeup(VMDriver *vm
 int
 HostIF_SemaphoreSignal(uint64 *args)  // IN:
 {
+   struct eventfd_ctx *eventfd;
    struct file *file;
-   mm_segment_t old_fs;
    int res;
    int signalFD = args[1];
    uint64 value = 1;  // make an eventfd happy should it be there
@@ -2580,22 +2672,36 @@ HostIF_SemaphoreSignal(uint64 *args)  //
       return MX_WAITERROR;
    }
 
-   old_fs = get_fs();
-   set_fs(get_ds());
+   /*
+    * If it's eventfd, use specific eventfd interface as kernel writes
+    * to eventfd may not be allowed in kernel 5.10 and later.
+    */
+   eventfd = eventfd_ctx_fileget(file);
+   if (!IS_ERR(eventfd)) {
+#if COMPAT_LINUX_VERSION_CHECK_LT(6, 8, 0)
+      eventfd_signal(eventfd, 1);
+#else
+      eventfd_signal(eventfd);
+#endif
+      fput(file);
+      return MX_WAITNORMAL;
+   }
 
    /*
     * Always write sizeof(uint64) bytes. This works fine for eventfd and
     * pipes. The data written is formatted to make an eventfd happy should
     * it be present.
     */
-
-   res = file->f_op->write(file, (char *) &value, sizeof value, &file->f_pos);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 14, 0)
+   res = kernel_write(file, (char *)&value, sizeof value, file->f_pos);
+#else
+   res = kernel_write(file, &value, sizeof value, &file->f_pos);
+#endif
 
    if (res == sizeof value) {
       res = MX_WAITNORMAL;
    }
 
-   set_fs(old_fs);
    fput(file);
 
    /*
@@ -3364,12 +3466,9 @@ HostIFFastClockThread(void *data)  // IN
 {
    struct file *filp = (struct file *) data;
    int res;
-   mm_segment_t oldFS;
    unsigned int rate = 0;
    unsigned int prevRate = 0;
 
-   oldFS = get_fs();
-   set_fs(KERNEL_DS);
    allow_signal(SIGKILL);
    set_user_nice(current, linuxState.fastClockPriority);
 
@@ -3403,8 +3502,6 @@ HostIFFastClockThread(void *data)  // IN
 
  out:
    LinuxDriverWakeUp(TRUE);
-   set_fs(oldFS);
-
    /*
     * Do not exit thread until we are told to do so.
     */
@@ -3524,7 +3621,6 @@ HostIF_SetFastClockRate(unsigned int rat
       }
    } else {
       if (linuxState.fastClockThread) {
-         force_sig(SIGKILL, linuxState.fastClockThread);
          kthread_stop(linuxState.fastClockThread);
 	 close_rtc(linuxState.fastClockFile, current->files);
 
@@ -3572,7 +3668,12 @@ HostIF_MapUserMem(VA addr,
 
    ASSERT(handle);
 
-   if (!access_ok(VERIFY_WRITE, p, size)) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 0, 0)
+   if (!access_ok(VERIFY_WRITE, p, size))
+#else
+   if (!access_ok(p, size))
+#endif
+   {
       printk(KERN_ERR "%s: Couldn't verify write to uva 0x%p with size %"
              FMTSZ"u\n", __func__, p, size);
 
From 4c2a103fd2d71f2084f1fe7ceacb816b9832ffa2 Mon Sep 17 00:00:00 2001
From: Michal Kubecek <mkubecek@suse.cz>
Date: Sun, 22 Oct 2023 23:24:05 +0200
Subject: [PATCH] vmmon: use get_user_pages to get page PFN

As a side effect of mainline commit 0d940a9b270b ("mm/pgtable: allow
pte_offset_map[_lock]() to fail") in 6.5-rc1, __pte_offset_map(), called by
pte_offset_map(), is no longer exported. WMware developers decided to hack
around this by replacing pte_offset_map() by pte_offset_kernel() which does
not seem to be a good idea and apparently may trigger warn checks in RCU
code on some systems as mentioned in the discussion on issue #223.
Therefore let's use the same solution as we had for 17.0.2 and older
versions as it does not show these problems.

Based on an upstream IRC discussion and the hva_to_pfn_*() family of
functions in KVM code, what PgtblVa2MPNLocked() does seems to be an
incomplete and partial open coded logic of get_user_pages() and as it is
only used to get PFN from a virtual address, it can be easily implemented
using get_user_pages() family.

Without knowledge what exactly are the PFNs used for in VMware, it is hard
to guess the right flags, these seem to work and have been tested by
multiple users over last few weeks.

We could likely use get_user_pages() also on older kernels and it might be
actually cleaner and more reliable as existing open coded implementation
does not seem to handle some corner cases but without knowledge of VMware
internals, it will be safer to stick to existing code where possible.
---
 vmmon-only/include/pgtbl.h | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/vmmon-only/include/pgtbl.h b/vmmon-only/include/pgtbl.h
index 3f43c62..7eaa49a 100644
--- a/vmmon-only/include/pgtbl.h
+++ b/vmmon-only/include/pgtbl.h
@@ -25,6 +25,7 @@
 #include "compat_pgtable.h"
 #include "compat_spinlock.h"
 #include "compat_page.h"
+#include "compat_version.h"
 
 
 /*
@@ -45,6 +46,7 @@
  *-----------------------------------------------------------------------------
  */
 
+#if COMPAT_LINUX_VERSION_CHECK_LT(6, 5, 0) // only used by PgtblVa2MPN() below
 static INLINE MPN
 PgtblVa2MPNLocked(struct mm_struct *mm, // IN: Mm structure of a process
                   VA addr)              // IN: Address in the virtual address
@@ -110,6 +112,7 @@ PgtblVa2MPNLocked(struct mm_struct *mm, // IN: Mm structure of a process
    }
    return mpn;
 }
+#endif
 
 
 /*
@@ -129,6 +132,8 @@ PgtblVa2MPNLocked(struct mm_struct *mm, // IN: Mm structure of a process
  *-----------------------------------------------------------------------------
  */
 
+#if COMPAT_LINUX_VERSION_CHECK_LT(6, 5, 0)
+
 static INLINE MPN
 PgtblVa2MPN(VA addr)  // IN
 {
@@ -143,4 +148,24 @@ PgtblVa2MPN(VA addr)  // IN
    return mpn;
 }
 
+#else /* COMPAT_LINUX_VERSION_CHECK_LT(6, 5, 0) */
+
+static INLINE MPN
+PgtblVa2MPN(VA addr)  // IN
+{
+   struct page *page;
+   int npages;
+   MPN mpn;
+
+   npages = get_user_pages_unlocked(addr, 1, &page, FOLL_HWPOISON);
+   if (npages != 1)
+	   return INVALID_MPN;
+   mpn = page_to_pfn(page);
+   put_page(page);
+
+   return mpn;
+}
+
+#endif /* COMPAT_LINUX_VERSION_CHECK_LT(6, 5, 0) */
+
 #endif /* __PGTBL_H__ */
From 218fec600d0af1c5e9f4af819b216fb9c69bb838 Mon Sep 17 00:00:00 2001
From: Michal Kubecek <mkubecek@suse.cz>
Date: Fri, 12 Jan 2024 08:30:33 +0100
Subject: [PATCH] modules: fix build with -Wmissing-prototypes

Mainline commit 0fcb70851fbf ("Makefile.extrawarn: turn on
missing-prototypes globally") in 6.8-rc1 enables -Wmissing-prototypes
globally, revealing a lot of unclean code and also some actual problems.
This is also the case in vmmon and vmnet modules.

Most of them are addressed by making functions used only within one file
static. The missing prototype of random_get_entropy_fallback() is handled
by including <linux/timex.h> rather than <asm/timex.h>.

Finally, there are four functions in vmnet module which are actually used
in multiple files but instead of proper declarations, their prototype is
duplicated in vmnet-only/driver.c, risking that the two copies won't match
(which actually happened in one case). The cleanest solution would be
creating separate header files for them (bridge.h, netif.h, userif.h and
vnetUserListener.h) and including them in the respective source file and
driver.c. As the developers already handle similar cases by simply putting
the declarations into vnetInt.h, let us do the same to keep things simple.
---
 vmmon-only/common/vmx86.c |  2 +-
 vmmon-only/linux/driver.c |  4 ++--
 vmmon-only/linux/hostif.c |  6 +++---
 vmnet-only/bridge.c       |  2 +-
 vmnet-only/driver.c       | 16 ++--------------
 vmnet-only/vnetInt.h      |  7 +++++++
 6 files changed, 16 insertions(+), 21 deletions(-)

diff --git a/vmmon-only/common/vmx86.c b/vmmon-only/common/vmx86.c
index 156e94a9..c5ed3e25 100644
--- a/vmmon-only/common/vmx86.c
+++ b/vmmon-only/common/vmx86.c
@@ -58,7 +58,7 @@
 #include "x86svm.h"
 #include "x86cpuid_asm.h"
 #if defined(linux)
-#include <asm/timex.h>
+#include <linux/timex.h>
 #endif
 #include "x86perfctr.h"
 
diff --git a/vmmon-only/linux/driver.c b/vmmon-only/linux/driver.c
index 9c855869..32e9c5ea 100644
--- a/vmmon-only/linux/driver.c
+++ b/vmmon-only/linux/driver.c
@@ -287,7 +287,7 @@ LinuxDriverInitTSCkHz(void)
  *----------------------------------------------------------------------
  */
 
-int
+static int
 LinuxDriverInit(void)
 {
    int retval;
@@ -390,7 +390,7 @@ LinuxDriverInit(void)
  *----------------------------------------------------------------------
  */
 
-void
+static void
 LinuxDriverExit(void)
 {
    /*
diff --git a/vmmon-only/linux/hostif.c b/vmmon-only/linux/hostif.c
index 3bce32c3..b83e21d3 100644
--- a/vmmon-only/linux/hostif.c
+++ b/vmmon-only/linux/hostif.c
@@ -2951,7 +2951,7 @@ HostIF_CallOnEachCPU(void (*func)(void*), // IN: function to call
  *-----------------------------------------------------------------------------
  */
 
-Bool
+static Bool
 HostIFCheckTrackedMPN(VMDriver *vm, // IN: The VM instance
                       MPN mpn)      // IN: The MPN
 {
@@ -3067,7 +3067,7 @@ HostIF_ReadPage(VMDriver *vm,        // IN: The VM instance
  *----------------------------------------------------------------------
  */
 
-int
+static int
 HostIFWritePageWork(MPN mpn,              // MPN of the page
                     VA64 addr,            // data to write to the page
                     Bool kernelBuffer)    // is the buffer in kernel space?
@@ -3373,7 +3373,7 @@ HostIFDoIoctl(struct file *filp,
  *----------------------------------------------------------------------
  */
 
-int
+static int
 HostIFStartTimer(Bool rateChanged,  //IN: Did rate change? 
 		 unsigned int rate, //IN: current clock rate
                  struct file *filp) //IN: /dev/rtc descriptor