release: 1.8.0: MAP_LOCKED and pre-populate PMIx shared memory PFNs

Change-Id: I171c87f0f49cf2f791693e397a1d94b1bc2d0440

.gitmodules

@@ -1,12 +1,12 @@
 [submodule "ihk"]
 	path = ihk
-	url = https://github.com/RIKEN-SysSoft/ihk.git
+	url = https://github.com/ihkmckernel/ihk.git
 [submodule "executer/user/lib/libdwarf/libdwarf"]
 	path = executer/user/lib/libdwarf/libdwarf
 	url = https://github.com/bgerofi/libdwarf.git
 [submodule "executer/user/lib/syscall_intercept"]
 	path = executer/user/lib/syscall_intercept
-	url = https://github.com/RIKEN-SysSoft/syscall_intercept.git
+	url = https://github.com/ihkmckernel/syscall_intercept.git
 [submodule "executer/user/lib/uti"]
 	path = executer/user/lib/uti
-	url = https://github.com/RIKEN-SysSoft/uti.git
+	url = https://github.com/ihkmckernel/uti.git

CMakeLists.txt

@@ -7,7 +7,7 @@ endif (NOT CMAKE_BUILD_TYPE)
 enable_language(C ASM)
 
 project(mckernel C ASM)
-set(MCKERNEL_VERSION "1.7.6")
+set(MCKERNEL_VERSION "1.8.0")
 
 # See "Fedora Packaging Guidelines -- Versioning"
 set(MCKERNEL_RELEASE "")
@@ -105,10 +105,17 @@ execute_process(COMMAND bash -c "rpm -qi FJSVpxkrm-plugin-mckernel | awk '$1 ==
   OUTPUT_VARIABLE KRM_VERSION OUTPUT_STRIP_TRAILING_WHITESPACE)
 message("KRM_VERSION: ${KRM_VERSION}")
 
-if(NOT "${KRM_VERSION}" STREQUAL "" AND "${KRM_VERSION}" VERSION_LESS_EQUAL 4.0.1)
-	option(ENABLE_KRM_WORKAROUND "krm workaround" ON)
-else()
+execute_process(COMMAND bash -c "rpm -qi FJSVpxkrm-plugin-mckernel | awk '$1 == \"Release\" && $2 == \":\" { print $3 }'"
+  OUTPUT_VARIABLE KRM_RELEASE OUTPUT_STRIP_TRAILING_WHITESPACE)
+message("KRM_RELEASE: ${KRM_RELEASE}")
+
+if("${KRM_VERSION}" STREQUAL "")
 	option(ENABLE_KRM_WORKAROUND "krm workaround" OFF)
+elseif("${KRM_VERSION}" VERSION_GREATER_EQUAL 4.0.2 OR
+    ("${KRM_VERSION}" VERSION_EQUAL 4.0.1 AND "${KRM_RELEASE}" VERSION_GREATER_EQUAL 25.13.1.0))
+	option(ENABLE_KRM_WORKAROUND "krm workaround" OFF)
+else()
+	option(ENABLE_KRM_WORKAROUND "krm workaround" ON)
 endif()
 
 if(ENABLE_KRM_WORKAROUND)
@@ -124,6 +131,14 @@ if(ENABLE_FUGAKU_DEBUG)
 	set(KBUILD_C_FLAGS "${KBUILD_C_FLAGS} -DENABLE_FUGAKU_DEBUG")
 endif()
 
+# redirect kernel messages to Linux's /dev/kmsg
+option(ENABLE_KMSG_REDIRECT "Redirect kernel message to Linux's /dev/kmsg" OFF)
+
+if(ENABLE_KMSG_REDIRECT)
+	add_definitions(-DENABLE_KMSG_REDIRECT)
+	set(KBUILD_C_FLAGS "${KBUILD_C_FLAGS} -DENABLE_KMSG_REDIRECT")
+endif()
+
 option(PROFILE_ENABLE "System call profile" ON)
 if(PROFILE_ENABLE)
 	add_definitions(-DPROFILE_ENABLE)
@@ -340,4 +355,5 @@ message("ENABLE_WERROR: ${ENABLE_WERROR}")
 message("ENABLE_UBSAN: ${ENABLE_UBSAN}")
 message("ENABLE_LINUX_WORK_IRQ_FOR_IKC: ${ENABLE_LINUX_WORK_IRQ_FOR_IKC}")
 message("ENABLE_PER_CPU_ALLOC_CACHE: ${ENABLE_PER_CPU_ALLOC_CACHE}")
+message("ENABLE_KMSG_REDIRECT: ${ENABLE_KMSG_REDIRECT}")
 message("-------------------------------")

arch/arm64/kernel/cpu.c

@@ -912,7 +912,6 @@ unsigned long cpu_enable_interrupt_save(void)
 	return flags;
 }
 
-#ifdef ENABLE_FUGAKU_HACKS
 int cpu_interrupt_disabled(void)
 {
 	unsigned long flags;
@@ -925,7 +924,6 @@ int cpu_interrupt_disabled(void)
 		: "memory");
 	return (flags == masked);
 }
-#endif
 
 #else /* defined(CONFIG_HAS_NMI) */
 
@@ -989,6 +987,18 @@ unsigned long cpu_enable_interrupt_save(void)
 		: "memory");
 	return flags;
 }
+
+int cpu_interrupt_disabled(void)
+{
+	unsigned long flags;
+
+	asm volatile(
+		"mrs    %0, daif	// arch_local_irq_save\n"
+		: "=r" (flags)
+		:
+		: "memory");
+	return !!(flags & 0x2);
+}
 #endif /* defined(CONFIG_HAS_NMI) */
 
 /* we not have "pause" instruction, instead "yield" instruction */

arch/x86_64/kernel/cpu.c

@@ -1273,6 +1273,15 @@ unsigned long cpu_enable_interrupt_save(void)
 	return flags;
 }
 
+int cpu_interrupt_disabled(void)
+{
+	unsigned long flags;
+
+	asm volatile("pushf; pop %0" : "=r"(flags) : : "memory", "cc");
+
+	return !(flags & 0x200);
+}
+
 /*@
   @ behavior valid_vector:
   @   assumes 32 <= vector <= 255;

docs/NEWS.rst

@@ -1,3 +1,130 @@
+=============================================
+Version 1.8.0 (Mar 23, 2021)
+=============================================
+
+----------------------
+IHK major updates
+----------------------
+N/A
+
+------------------------
+IHK major bug fixes
+------------------------
+N/A
+
+----------------------
+McKernel major updates
+----------------------
+N/A
+
+------------------------
+McKernel major bug fixes
+------------------------
+#. profile: fix infinite recursion for allocation miss event
+#. Fugaku: MAP_LOCKED and pre-populate PMIx shared memory PFNs
+
+=============================================
+Version 1.7.10 (Mar 18, 2021)
+=============================================
+
+----------------------
+IHK major updates
+----------------------
+N/A
+
+------------------------
+IHK major bug fixes
+------------------------
+#. __ihk_device_detect_hungup: detect hungup via device-ioctl
+
+----------------------
+McKernel major updates
+----------------------
+N/A
+
+------------------------
+McKernel major bug fixes
+------------------------
+N/A
+
+=============================================
+Version 1.7.9 (Mar 17, 2021)
+=============================================
+
+----------------------
+IHK major updates
+----------------------
+N/A
+
+------------------------
+IHK major bug fixes
+------------------------
+#. ihklib: ihk_reserve_mem_conf*: fix default values
+#. smp_ihk_os_shutdown: fix memory leak
+#. smp_ihk_os_shutdown: prevent double free
+#. __ihk_os_shutdown: fix smp_ihk_os_shutdown()-related double free
+#. smp_ihk_os_panic_notifier: exclude memory from Linux dump with default setting
+#. smp_ihk_os_panic_notifier: exclude memory from Linux dump while booting, on timeout
+
+----------------------
+McKernel major updates
+----------------------
+N/A
+
+------------------------
+McKernel major bug fixes
+------------------------
+#. mcctrl_wakeup_desc: refcount and fix timeouts
+
+=============================================
+Version 1.7.8 (Mar 12, 2021)
+=============================================
+
+----------------------
+IHK major updates
+----------------------
+N/A
+
+------------------------
+IHK major bug fixes
+------------------------
+#. ihklib: ihk_reserve_cpu: fix job cpu check when using krm
+
+----------------------
+McKernel major updates
+----------------------
+N/A
+
+------------------------
+McKernel major bug fixes
+------------------------
+N/A
+
+=============================================
+Version 1.7.7 (Mar 11, 2021)
+=============================================
+
+----------------------
+IHK major updates
+----------------------
+N/A
+
+------------------------
+IHK major bug fixes
+------------------------
+N/A
+
+----------------------
+McKernel major updates
+----------------------
+N/A
+
+------------------------
+McKernel major bug fixes
+------------------------
+#. mcexec: fput executable just after its contents is transferred
+#. spec: cmake-config cmake parameters
+
 =============================================
 Version 1.7.6 (Mar 11, 2021)
 =============================================

docs/contact.rst

@@ -1,11 +1,5 @@
 Contact
 =======
 
-Please give your feedback to us via one of the following mailing lists.
-Subscription via
-`www.pccluster.org <http://www.pccluster.org/mailman/listinfo/mckernel-users>`__
-is needed.
-
--  English: mckernel-users@pccluster.org
--  Japanese: mckernel-users-jp@pccluster.org
+Please give your feedback to us via the following mailing list: ihkmckernel@googlegroups.com
 

docs/install.rst

@@ -97,7 +97,7 @@ Clone the source code:
 
    mkdir -p ~/src/ihk+mckernel/
    cd ~/src/ihk+mckernel/
-   git clone --recursive -b development https://github.com/RIKEN-SysSoft/mckernel.git
+   git clone --recursive -b development https://github.com/ihkmckernel/mckernel.git
 
 (Optional) Checkout to the specific branch or version:
 
@@ -166,22 +166,6 @@ Create the tarball and the spec file:
    make dist
    cp mckernel-<version>.tar.gz <rpmbuild>/SOURCES
 
-(optional) Edit the following line in ``scripts/mckernel.spec`` to change
-cmake options. For example:
-
-::
-
-   %cmake -DCMAKE_BUILD_TYPE=Release \
-	-DUNAME_R=%{kernel_version} \
-	-DKERNEL_DIR=%{kernel_dir} \
-	%{?cmake_libdir:-DCMAKE_INSTALL_LIBDIR=%{cmake_libdir}} \
-	%{?build_target:-DBUILD_TARGET=%{build_target}} \
-	%{?toolchain_file:-DCMAKE_TOOLCHAIN_FILE=%{toolchain_file}} \
-	-DENABLE_TOFU=ON -DENABLE_FUGAKU_HACKS=ON \
-	-DENABLE_KRM_WORKAROUND=OFF -DWITH_KRM=ON \
-	-DENABLE_FUGAKU_DEBUG=OFF -DENABLE_UTI=ON \
-	.
-
 Create the rpm package:
 
 When not cross-compiling:

docs/spec/ihk/ihk.tex

@@ -666,7 +666,7 @@ IHKはLinuxに以下の機能を提供する。
 このパラメタの目的は、Linuxによる空き領域の分断化が激しい状況においてメモリ予約処理時間を抑えることである。上記の状況で予約処理時間が長くなるのは、小さいサイズでの物理連続領域が大量に存在するので、小さいサイズでの要求回数が非常に大きくなるためである。
 
 \subsubsection*{\texttt{IHK\_RESERVE\_MEM\_MAX\_SIZE\_RATIO\_ALL}}
-\verb|ihk_reserve_mem()|でサイズに-1を指定した場合と\verb|IHK_RESERVE_MEM_BALANCED_ENABLE|に非ゼロを指定した場合に用いられる予約サイズを、予約時点で測定した空き容量に指定した値を乗じたものにする。なお、ゼロ以下の値または98より大きい値を設定しようとすると\verb:-EINVAL:を返す。また、デフォルト設定は98\%である。
+\verb|ihk_reserve_mem()|でサイズに-1を指定した場合と\verb|IHK_RESERVE_MEM_BALANCED_ENABLE|に非ゼロを指定した場合に用いられる予約サイズを、予約時点で測定した空き容量に指定した値を乗じたものにする。なお、ゼロ以下の値を設定しようとしたり、また富岳では95、その他のシステムでは98より大きい値を設定しようとしたりすると\verb:-EINVAL:を返す。また、デフォルト設定は富岳では95\%、その他のシステムでは98\%である。
 
 目的は、Linuxによる空き領域の分断化が激しい状況においてメモリ予約処理時間を抑えること、また予約時にLinuxのプロセスのメモリ要求が満たされない状況にならないようにすることである。
 

docs/users.rst

@@ -87,14 +87,14 @@ executable:
 ``<processes-per-node>`` is the number of the processes per node and
 calculated by (number of MPI processes) / (number of nodes).
 
-For example, ``<processes-per-node>`` equals to 4 (=32/8) when
+For example, ``<processes-per-node>`` equals to 4 (=8/2) when
 specifying the number of processes and nodes as follows with
-Fujitsu Technical Computing Suite.
+MPICH.
 
 .. code-block:: none
 
-   #PJM --mpi "proc=32"
-   #PJM -L "node=8"
+   mpirun -n 8 -hosts host1,host2 ./cpi
+
 
 (Advanced) When using Utility Thread offloading Interface (UTI)
 ---------------------------------------------------------------
@@ -112,11 +112,11 @@ Add ``--enable-uti`` option to ``mcexec``:
 Limitations
 ===========
 
-1.  Pseudo devices such as /dev/mem and /dev/zero are not mmap()ed
+#.  Pseudo devices such as /dev/mem and /dev/zero are not mmap()ed
     correctly even if the mmap() returns a success. An access of their
     mapping receives the SIGSEGV signal.
 
-2.  clone() supports only the following flags. All the other flags cause
+#.  clone() supports only the following flags. All the other flags cause
     clone() to return error or are simply ignored.
 
     -  CLONE_CHILD_CLEARTID
@@ -126,32 +126,32 @@ Limitations
     -  CLONE_SIGHAND
     -  CLONE_VM
 
-3.  PAPI has the following restriction.
+#.  PAPI has the following restriction.
 
     -  Number of counters a user can use at the same time is up to the
        number of the physical counters in the processor.
 
-4.  msync writes back only the modified pages mapped by the calling
+#.  msync writes back only the modified pages mapped by the calling
     process.
 
-5.  The following syscalls always return the ENOSYS error.
+#.  The following syscalls always return the ENOSYS error.
 
     -  migrate_pages()
     -  move_pages()
     -  set_robust_list()
 
-6.  The following syscalls always return the EOPNOTSUPP error.
+#.  The following syscalls always return the EOPNOTSUPP error.
 
     -  arch_prctl(ARCH_SET_GS)
     -  signalfd()
 
-7.  signalfd4() returns a fd, but signal is not notified through the fd.
+#.  signalfd4() returns a fd, but signal is not notified through the fd.
 
-8.  set_rlimit sets the limit values but they are not enforced.
+#.  set_rlimit sets the limit values but they are not enforced.
 
-9.  Address randomization is not supported.
+#.  Address randomization is not supported.
 
-10. brk() extends the heap more than requestd when -h (–extend-heap-by=)
+#.  brk() extends the heap more than requestd when -h (–extend-heap-by=)
     option of mcexec is used with the value larger than 4 KiB.
     syscall_pwrite02 of LTP would fail for this reason. This is because
     the test expects that the end of the heap is set to the same address
@@ -161,91 +161,86 @@ Limitations
     than the requested. Therefore, the expected segmentation violation
     doesn’t occur.
 
-11. setpriority()/getpriority() won’t work. They might set/get the
-    priority of a random mcexec thread. This is because there’s no fixed
-    correspondence between a McKernel thread which issues the system
-    call and a mcexec thread which handles the offload request.
+#. setpriority()/getpriority() won’t work. They might set/get the
+   priority of a random mcexec thread. This is because there’s no fixed
+   correspondence between a McKernel thread which issues the system
+   call and a mcexec thread which handles the offload request.
 
-12. mbind() can set the policy but it is not used when allocating
-    physical pages.
+#. mbind() can set the policy but it is not used when allocating
+   physical pages.
 
-13. MPOL_F_RELATIVE_NODES and MPOL_INTERLEAVE flags for
-    set_mempolicy()/mbind() are not supported.
+#. MPOL_F_RELATIVE_NODES and MPOL_INTERLEAVE flags for
+   set_mempolicy()/mbind() are not supported.
 
-14. The MPOL_BIND policy for set_mempolicy()/mbind() works as the same
-    as the MPOL_PREFERRED policy. That is, the physical page allocator
-    doesn’t give up the allocation when the specified nodes are running
-    out of pages but continues to search pages in the other nodes.
+#. The MPOL_BIND policy for set_mempolicy()/mbind() works as the same
+   as the MPOL_PREFERRED policy. That is, the physical page allocator
+   doesn’t give up the allocation when the specified nodes are running
+   out of pages but continues to search pages in the other nodes.
 
-15. Kernel dump on Linux panic requires Linux kernel CentOS-7.4 and
-    later. In addition, crash_kexec_post_notifiers kernel argument must
-    be given to Linux kernel.
+#. Kernel dump on Linux panic requires Linux kernel CentOS-7.4 and
+   later. In addition, crash_kexec_post_notifiers kernel argument must
+   be given to Linux kernel.
 
-16. setfsuid()/setfsgid() cannot change the id of the calling thread.
-    Instead, it changes that of the mcexec worker thread which takes the
-    system-call offload request.
+#. setfsuid()/setfsgid() cannot change the id of the calling thread.
+   Instead, it changes that of the mcexec worker thread which takes the
+   system-call offload request.
 
-17. mmap (hugeTLBfs): The physical pages corresponding to a map are
-    released when no McKernel process exist. The next map gets fresh
-    physical pages.
+#. mmap (hugeTLBfs): The physical pages corresponding to a map are
+   released when no McKernel process exist. The next map gets fresh
+   physical pages.
 
-18. Sticky bit on executable file has no effect.
+#. Sticky bit on executable file has no effect.
 
-19. Linux (RHEL-7 for x86_64) could hang when offlining CPUs in the
-    process of booting McKernel due to the Linux bug, found in
-    Linux-3.10 and fixed in the later version. One way to circumvent
-    this is to always assign the same CPU set to McKernel.
+#. Linux (RHEL-7 for x86_64) could hang when offlining CPUs in the
+   process of booting McKernel due to the Linux bug, found in
+   Linux-3.10 and fixed in the later version. One way to circumvent
+   this is to always assign the same CPU set to McKernel.
 
-20. madvise:
+#. madvise:
 
     -  MADV_HWPOISON and MADV_SOFT_OFFLINE always returns -EPERM.
     -  MADV_MERGEABLE and MADV_UNMERGEABLE always returns -EINVAL.
     -  MADV_HUGEPAGE and MADV_NOHUGEPAGE on file map returns -EINVAL
        except on RHEL-8 for aarch64.
 
-21. brk() and mmap() doesn’t report out-of-memory through its return
-    value. Instead, page-fault reports the error.
+#. brk() and mmap() doesn’t report out-of-memory through its return
+   value. Instead, page-fault reports the error.
 
-22. Anonymous mmap pre-maps requested number of pages when contiguous
-    pages are available. Demand paging is used when not available.
+#. Anonymous mmap pre-maps requested number of pages when contiguous
+   pages are available. Demand paging is used when not available.
 
-23. Mixing page sizes in anonymous shared mapping is not allowed. mmap
-    creates vm_range with one page size. And munmap or mremap that needs
-    the reduced page size changes the sizes of all the pages of the
-    vm_range.
+#. ihk_os_getperfevent() could time-out when invoked from Fujitsu TCS
+   (job-scheduler).
 
-24. ihk_os_getperfevent() could time-out when invoked from Fujitsu TCS
-    (job-scheduler).
+#. The behaviors of madvise and mbind are changed to do nothing and
+   report success as a workaround for Fugaku.
 
-25. The behaviors of madvise and mbind are changed to do nothing and
-    report success as a workaround for Fugaku.
+#. mmap() allows unlimited overcommit. Note that it corresponds to
+   setting sysctl ``vm.overcommit_memory`` to 1.
 
-26. mmap() allows unlimited overcommit. Note that it corresponds to
-    setting sysctl ``vm.overcommit_memory`` to 1.
+#. mlockall() is not supported and returns -EPERM.
 
-27. mlockall() is not supported and returns -EPERM.
+#. munlockall() is not supported and returns zero.
 
-28. munlockall() is not supported and returns zero.
+#. (Fujitsu TCS-only) A job following the one in which __mcctrl_os_read_write_cpu_register() returns ``-ETIME`` fails because xos_hwb related CPU state isn't finalized. You can tell if the function returned ``-ETIME`` by checking if the following line appeared in the Linux kernel message:
 
-29. scheduling behavior is not Linux compatible. For example, sometimes one of the two processes on the same CPU continues to run after yielding.
+   ::
 
-30. (Fujitsu TCS-only) A job following the one in which __mcctrl_os_read_write_cpu_register() returns ``-ETIME`` fails because xos_hwb related CPU state isn't finalized. You can tell if the function returned ``-ETIME`` by checking if the following line appeared in the Linux kernel message:
+      __mcctrl_os_read_write_cpu_register: ERROR sending IKC msg: -62
 
-    ::
+   You can re-initialize xos_hwb related CPU state by the following command:
 
-       __mcctrl_os_read_write_cpu_register: ERROR sending IKC msg: -62
+   ::
 
-    You can re-initialize xos_hwb related CPU state by the following command:
+      sudo systemctl restart xos_hwb
 
-    ::
+#. System calls can write the mcexec VMAs with PROT_WRITE flag not
+   set. This is because we never turn off PROT_WRITE of the mcexec
+   VMAs to circumvent the issue "set_host_vma(): do NOT read protect
+   Linux VMA".
 
-       sudo systemctl restart xos_hwb
+#. procfs entry creation done by Linux work queue could starve when
+   Linux CPUs are flooded with system call offloads. LTP-2019
+   sendmsg02 causes this issue.
 
-31. System calls can write the mcexec VMAs with PROT_WRITE flag not
-    set. This is because we never turn off PROT_WRITE of the mcexec
-    VMAs to circumvent the issue "set_host_vma(): do NOT read protect
-    Linux VMA".
-
-32. procfs entry creation done by Linux work queue could starve when
-    Linux CPUs are flooded with system call offloads. LTP-2019
-    sendmsg02 causes this issue.
+#. Linux kernel dump file doesn't include the memory allocated to McKernel. This is because of the issues in the implementation of the panic notifier handler.

executer/kernel/mcctrl/arch/arm64/archdeps.c

@@ -916,7 +916,7 @@ int __mcctrl_tof_utofu_release_handler(struct inode *inode, struct file *filp,
 	isp.arg = f2pfd->fd;
 
 	ret = mcctrl_ikc_send_wait(f2pfd->os, ppd->ikc_target_cpu,
-			&isp, -20, NULL, NULL, 0);
+			&isp, -1000, NULL, NULL, 0);
 	if (ret != 0) {
 		pr_err("%s: WARNING: IKC req for PID: %d, fd: %d failed\n",
 				__func__, f2pfd->pid, f2pfd->fd);

executer/kernel/mcctrl/control.c

@@ -414,7 +414,7 @@ static void release_handler(ihk_os_t os, void *param)
 	dprintk("%s: SCD_MSG_CLEANUP_PROCESS, info: %p, cpu: %d\n",
 			__FUNCTION__, info, info->cpu);
 	ret = mcctrl_ikc_send_wait(os, info->cpu,
-			&isp, -20, NULL, NULL, 0);
+			&isp, -5000, NULL, NULL, 0);
 	if (ret != 0) {
 		printk("%s: WARNING: failed to send IKC msg: %d\n",
 				__func__, ret);
@@ -513,8 +513,6 @@ static DECLARE_WAIT_QUEUE_HEAD(signalq);
 
 struct mcctrl_signal_desc {
 	struct mcctrl_signal msig;
-	struct mcctrl_wakeup_desc wakeup;
-	void *addrs[1];
 };
 
 static long mcexec_send_signal(ihk_os_t os, struct signal_desc *sigparam)
@@ -554,7 +552,7 @@ static long mcexec_send_signal(ihk_os_t os, struct signal_desc *sigparam)
 	isp.pid = sig.pid;
 	isp.arg = virt_to_phys(msigp);
 
-	rc = mcctrl_ikc_send_wait(os, sig.cpu, &isp, 0, &desc->wakeup,
+	rc = mcctrl_ikc_send_wait(os, sig.cpu, &isp, -1000, NULL,
 				  &do_free, 1, desc);
 	if (rc < 0) {
 		printk("mcexec_send_signal: mcctrl_ikc_send ret=%d\n", rc);
@@ -2010,7 +2008,7 @@ int mcexec_open_exec(ihk_os_t os, char * __user filename)
 	fullpath = d_path(&file->f_path, pathbuf, PATH_MAX);
 	if (IS_ERR(fullpath)) {
 		retval = PTR_ERR(fullpath);
-		goto out_free;
+		goto out_put_file;
 	}
 
 	mcef = kmalloc(sizeof(*mcef), GFP_KERNEL);
@@ -2243,8 +2241,6 @@ long mcctrl_perf_num(ihk_os_t os, unsigned long arg)
 
 struct mcctrl_perf_ctrl_desc {
 	struct perf_ctrl_desc desc;
-	struct mcctrl_wakeup_desc wakeup;
-	void *addrs[1];
 };
 #define wakeup_desc_of_perf_desc(_desc) \
 	(&container_of((_desc), struct mcctrl_perf_ctrl_desc, desc)->wakeup)
@@ -2310,9 +2306,7 @@ long mcctrl_perf_set(ihk_os_t os, struct ihk_perf_event_attr *__user arg)
 		isp.arg = virt_to_phys(perf_desc);
 
 		for (j = 0; j < info->n_cpus; j++) {
-			ret = mcctrl_ikc_send_wait(os, j, &isp,
-					msecs_to_jiffies(10000),
-					wakeup_desc_of_perf_desc(perf_desc),
+			ret = mcctrl_ikc_send_wait(os, j, &isp, 10000, NULL,
 					&need_free, 1, perf_desc);
 			if (ret < 0) {
 				pr_warn("%s: mcctrl_ikc_send_wait ret=%d\n",
@@ -2382,9 +2376,7 @@ long mcctrl_perf_get(ihk_os_t os, unsigned long *__user arg)
 		isp.arg = virt_to_phys(perf_desc);
 
 		for (j = 0; j < info->n_cpus; j++) {
-			ret = mcctrl_ikc_send_wait(os, j, &isp,
-					msecs_to_jiffies(10000),
-					wakeup_desc_of_perf_desc(perf_desc),
+			ret = mcctrl_ikc_send_wait(os, j, &isp, 10000, NULL,
 					&need_free, 1, perf_desc);
 			if (ret < 0) {
 				pr_warn("%s: mcctrl_ikc_send_wait ret=%d\n",
@@ -2454,9 +2446,8 @@ long mcctrl_perf_enable(ihk_os_t os)
 		return -EINVAL;
 	}
 	for (j = 0; j < info->n_cpus; j++) {
-		ret = mcctrl_ikc_send_wait(os, j, &isp, 0,
-					   wakeup_desc_of_perf_desc(perf_desc),
-					   &need_free, 1, perf_desc);
+		ret = mcctrl_ikc_send_wait(os, j, &isp, 0, NULL,
+				&need_free, 1, perf_desc);
 
 		if (ret < 0) {
 			pr_warn("%s: mcctrl_ikc_send_wait ret=%d\n",
@@ -2522,8 +2513,7 @@ long mcctrl_perf_disable(ihk_os_t os)
 		return -EINVAL;
 	}
 	for (j = 0; j < info->n_cpus; j++) {
-		ret = mcctrl_ikc_send_wait(os, j, &isp, 0,
-				wakeup_desc_of_perf_desc(perf_desc),
+		ret = mcctrl_ikc_send_wait(os, j, &isp, 0, NULL,
 				&need_free, 1, perf_desc);
 		if (ret < 0) {
 			pr_warn("%s: mcctrl_ikc_send_wait ret=%d\n",

executer/kernel/mcctrl/futex.c

@@ -182,8 +182,6 @@ static int uti_remote_page_fault(struct mcctrl_usrdata *usrdata,
 			struct mcctrl_per_proc_data *ppd, int tid, int cpu)
 {
 	int error;
-	struct mcctrl_wakeup_desc *desc;
-	int do_frees = 1;
 	struct ikc_scd_packet packet;
 
 	/* Request page fault */
@@ -192,20 +190,9 @@ static int uti_remote_page_fault(struct mcctrl_usrdata *usrdata,
 	packet.fault_reason = reason;
 	packet.fault_tid = tid;
 
-	/* we need to alloc desc ourselves because GFP_ATOMIC */
-retry_alloc:
-	desc = kmalloc(sizeof(*desc), GFP_ATOMIC);
-	if (!desc) {
-		pr_warn("WARNING: coudln't alloc remote page fault wait desc, retrying..\n");
-		goto retry_alloc;
-	}
-
 	/* packet->target_cpu was set in rus_vm_fault if a thread was found */
 	error = mcctrl_ikc_send_wait(usrdata->os, cpu, &packet,
-				     0, desc, &do_frees, 0);
-	if (do_frees) {
-		kfree(desc);
-	}
+				0, NULL, NULL, 0);
 	if (error < 0) {
 		pr_warn("%s: WARNING: failed to request uti remote page fault :%d\n",
 			__func__, error);

executer/kernel/mcctrl/ikc.c

@@ -58,23 +58,41 @@ void mcctrl_os_read_write_cpu_response(ihk_os_t os,
 		struct ikc_scd_packet *pisp);
 void mcctrl_eventfd(ihk_os_t os, struct ikc_scd_packet *pisp);
 
-/* Assumes usrdata->wakeup_descs_lock taken */
-static void mcctrl_wakeup_desc_cleanup(ihk_os_t os,
-		struct mcctrl_wakeup_desc *desc)
+static void mcctrl_wakeup_desc_put(struct mcctrl_wakeup_desc *desc,
+		struct mcctrl_usrdata *usrdata, int free_addrs)
 {
+	unsigned long irqflags;
 	int i;
 
-	list_del(&desc->chain);
-
-	for (i = 0; i < desc->free_addrs_count; i++) {
-		kfree(desc->free_addrs[i]);
+	if (!refcount_dec_and_test(&desc->count)) {
+		return;
 	}
+
+	spin_lock_irqsave(&usrdata->wakeup_descs_lock, irqflags);
+	list_del(&desc->chain);
+	spin_unlock_irqrestore(&usrdata->wakeup_descs_lock, irqflags);
+
+	if (free_addrs) {
+		for (i = 0; i < desc->free_addrs_count; i++) {
+			kfree(desc->free_addrs[i]);
+		}
+	}
+
+	if (desc->free_at_put)
+		kfree(desc);
 }
 
 static void mcctrl_wakeup_cb(ihk_os_t os, struct ikc_scd_packet *packet)
 {
 	struct mcctrl_wakeup_desc *desc = packet->reply;
+	struct mcctrl_usrdata *usrdata = ihk_host_os_get_usrdata(os);
 
+	/* destroy_ikc_channels must have cleaned up descs */
+	if (!usrdata) {
+		pr_err("%s: error: mcctrl_usrdata not found\n",
+				__func__);
+		return;
+	}
 
 	WRITE_ONCE(desc->err, packet->err);
 
@@ -85,29 +103,25 @@ static void mcctrl_wakeup_cb(ihk_os_t os, struct ikc_scd_packet *packet)
 	 * wake up opportunistically between this set and the wake_up call.
 	 *
 	 * If the other side is no longer waiting, free the memory that was
-	 * left for us.
+	 * left for us. The caller has been notified not to free.
 	 */
 	if (cmpxchg(&desc->status, 0, 1)) {
-		struct mcctrl_usrdata *usrdata = ihk_host_os_get_usrdata(os);
-		unsigned long flags;
-
-		/* destroy_ikc_channels must have cleaned up descs */
-		if (!usrdata) {
-			pr_err("%s: error: mcctrl_usrdata not found\n",
-				__func__);
-			return;
-		}
-
-		spin_lock_irqsave(&usrdata->wakeup_descs_lock, flags);
-		mcctrl_wakeup_desc_cleanup(os, desc);
-		spin_unlock_irqrestore(&usrdata->wakeup_descs_lock, flags);
+		mcctrl_wakeup_desc_put(desc, usrdata, 1);
 		return;
 	}
 
+	/*
+	 * Notify waiter before dropping reference to make sure
+	 * wait queue is still valid.
+	 */
 	wake_up_interruptible(&desc->wq);
+	mcctrl_wakeup_desc_put(desc, usrdata, 0);
 }
 
-/* do_frees: 1 when caller should free free_addrs[], 0 otherwise */
+/*
+ * do_frees: 1 when caller should free free_addrs[], 0 otherwise
+ * timeout: timeout in milliseconds
+ */
 int mcctrl_ikc_send_wait(ihk_os_t os, int cpu, struct ikc_scd_packet *pisp,
 		long int timeout, struct mcctrl_wakeup_desc *desc,
 		int *do_frees, int free_addrs_count, ...)
@@ -115,35 +129,60 @@ int mcctrl_ikc_send_wait(ihk_os_t os, int cpu, struct ikc_scd_packet *pisp,
 	int ret, i;
 	int alloc_desc = (desc == NULL);
 	va_list ap;
+	unsigned long flags;
+	struct mcctrl_usrdata *usrdata = ihk_host_os_get_usrdata(os);
+
+	if (!usrdata) {
+		pr_err("%s: error: mcctrl_usrdata not found\n",
+				__func__);
+		return -EINVAL;
+	}
 
 	if (free_addrs_count)
 		*do_frees = 1;
+
 	if (alloc_desc)
 		desc = kmalloc(sizeof(struct mcctrl_wakeup_desc) +
 			       (free_addrs_count + 1) * sizeof(void *),
-			       GFP_KERNEL);
+			       GFP_ATOMIC);
 	if (!desc) {
 		pr_warn("%s: Could not allocate wakeup descriptor", __func__);
 		return -ENOMEM;
 	}
+
 	pisp->reply = desc;
 	va_start(ap, free_addrs_count);
 	for (i = 0; i < free_addrs_count; i++) {
 		desc->free_addrs[i] = va_arg(ap, void*);
 	}
 	va_end(ap);
-	if (alloc_desc)
-		desc->free_addrs[free_addrs_count++] = desc;
 	desc->free_addrs_count = free_addrs_count;
+
+	/* Only free at put time if allocated internally */
+	desc->free_at_put = 0;
+	if (alloc_desc)
+		desc->free_at_put = 1;
+
 	init_waitqueue_head(&desc->wq);
+
+	/* One for the caller and one for the call-back */
+	refcount_set(&desc->count, 2);
+
+	/* XXX: make this a hash-table? */
+	spin_lock_irqsave(&usrdata->wakeup_descs_lock, flags);
+	list_add(&desc->chain, &usrdata->wakeup_descs_list);
+	spin_unlock_irqrestore(&usrdata->wakeup_descs_lock, flags);
+
 	WRITE_ONCE(desc->err, 0);
 	WRITE_ONCE(desc->status, 0);
 
 	ret = mcctrl_ikc_send(os, cpu, pisp);
 	if (ret < 0) {
 		pr_warn("%s: mcctrl_ikc_send failed: %d\n", __func__, ret);
-		if (alloc_desc)
-			kfree(desc);
+		/* Failed to send msg, put twice */
+		mcctrl_wakeup_desc_put(desc, usrdata, 0);
+		mcctrl_wakeup_desc_put(desc, usrdata, 0);
+
 		return ret;
 	}
 
@@ -180,28 +219,16 @@ int mcctrl_ikc_send_wait(ihk_os_t os, int cpu, struct ikc_scd_packet *pisp,
 	 * the callback it will need to free things for us
 	 */
 	if (!cmpxchg(&desc->status, 0, 1)) {
-		struct mcctrl_usrdata *usrdata = ihk_host_os_get_usrdata(os);
-		unsigned long flags;
+		mcctrl_wakeup_desc_put(desc, usrdata, 0);
 
-		if (!usrdata) {
-			pr_err("%s: error: mcctrl_usrdata not found\n",
-				__func__);
-			ret = ret < 0 ? ret : -EINVAL;
-			goto out;
-		}
-
-		spin_lock_irqsave(&usrdata->wakeup_descs_lock, flags);
-		list_add(&desc->chain, &usrdata->wakeup_descs_list);
-		spin_unlock_irqrestore(&usrdata->wakeup_descs_lock, flags);
 		if (do_frees)
 			*do_frees = 0;
 		return ret < 0 ? ret : -ETIME;
 	}
 
 	ret = READ_ONCE(desc->err);
-out:
-	if (alloc_desc)
-		kfree(desc);
+
+	mcctrl_wakeup_desc_put(desc, usrdata, 0);
 	return ret;
 }
 
@@ -605,10 +632,15 @@ void destroy_ikc_channels(ihk_os_t os)
 			ihk_ikc_destroy_channel(usrdata->ikc2linux[i]);
 		}
 	}
+
 	spin_lock_irqsave(&usrdata->wakeup_descs_lock, flags);
 	list_for_each_entry_safe(mwd_entry, mwd_next,
-			&usrdata->wakeup_descs_list, chain) {
-		mcctrl_wakeup_desc_cleanup(os, mwd_entry);
+				&usrdata->wakeup_descs_list, chain) {
+		list_del(&mwd_entry->chain);
+
+		for (i = 0; i < mwd_entry->free_addrs_count; i++) {
+			kfree(mwd_entry->free_addrs[i]);
+		}
 	}
 	spin_unlock_irqrestore(&usrdata->wakeup_descs_lock, flags);
 

executer/kernel/mcctrl/include/refcount.h

@@ -0,0 +1,105 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _LINUX_REFCOUNT_H
+#define _LINUX_REFCOUNT_H
+
+#include <linux/atomic.h>
+#include <linux/compiler.h>
+#include <linux/spinlock_types.h>
+
+struct mutex;
+
+/**
+ * struct refcount_t - variant of atomic_t specialized for reference counts
+ * @refs: atomic_t counter field
+ *
+ * The counter saturates at UINT_MAX and will not move once
+ * there. This avoids wrapping the counter and causing 'spurious'
+ * use-after-free bugs.
+ */
+typedef struct refcount_struct {
+	atomic_t refs;
+} refcount_t;
+
+#define REFCOUNT_INIT(n)	{ .refs = ATOMIC_INIT(n), }
+
+/**
+ * refcount_set - set a refcount's value
+ * @r: the refcount
+ * @n: value to which the refcount will be set
+ */
+static inline void refcount_set(refcount_t *r, unsigned int n)
+{
+	atomic_set(&r->refs, n);
+}
+
+/**
+ * refcount_read - get a refcount's value
+ * @r: the refcount
+ *
+ * Return: the refcount's value
+ */
+static inline unsigned int refcount_read(const refcount_t *r)
+{
+	return atomic_read(&r->refs);
+}
+
+#ifdef CONFIG_REFCOUNT_FULL
+extern __must_check bool refcount_add_not_zero(unsigned int i, refcount_t *r);
+extern void refcount_add(unsigned int i, refcount_t *r);
+
+extern __must_check bool refcount_inc_not_zero(refcount_t *r);
+extern void refcount_inc(refcount_t *r);
+
+extern __must_check bool refcount_sub_and_test(unsigned int i, refcount_t *r);
+
+extern __must_check bool refcount_dec_and_test(refcount_t *r);
+extern void refcount_dec(refcount_t *r);
+#else
+# ifdef CONFIG_ARCH_HAS_REFCOUNT
+#  include <asm/refcount.h>
+# else
+static inline __must_check bool refcount_add_not_zero(unsigned int i, refcount_t *r)
+{
+	return atomic_add_unless(&r->refs, i, 0);
+}
+
+static inline void refcount_add(unsigned int i, refcount_t *r)
+{
+	atomic_add(i, &r->refs);
+}
+
+static inline __must_check bool refcount_inc_not_zero(refcount_t *r)
+{
+	return atomic_add_unless(&r->refs, 1, 0);
+}
+
+static inline void refcount_inc(refcount_t *r)
+{
+	atomic_inc(&r->refs);
+}
+
+static inline __must_check bool refcount_sub_and_test(unsigned int i, refcount_t *r)
+{
+	return atomic_sub_and_test(i, &r->refs);
+}
+
+static inline __must_check bool refcount_dec_and_test(refcount_t *r)
+{
+	return atomic_dec_and_test(&r->refs);
+}
+
+static inline void refcount_dec(refcount_t *r)
+{
+	atomic_dec(&r->refs);
+}
+# endif /* !CONFIG_ARCH_HAS_REFCOUNT */
+#endif /* CONFIG_REFCOUNT_FULL */
+
+extern __must_check bool refcount_dec_if_one(refcount_t *r);
+extern __must_check bool refcount_dec_not_one(refcount_t *r);
+extern __must_check bool refcount_dec_and_mutex_lock(refcount_t *r, struct mutex *lock);
+extern __must_check bool refcount_dec_and_lock(refcount_t *r, spinlock_t *lock);
+extern __must_check bool refcount_dec_and_lock_irqsave(refcount_t *r,
+						       spinlock_t *lock,
+						       unsigned long *flags);
+#endif /* _LINUX_REFCOUNT_H */

executer/kernel/mcctrl/mcctrl.h

@@ -44,6 +44,10 @@
 #include <linux/semaphore.h>
 #include <linux/rwlock.h>
 #include <linux/threads.h>
+#include <linux/version.h>
+#if KERNEL_VERSION(4, 11, 0) > LINUX_VERSION_CODE
+#include <refcount.h>
+#endif
 #include "sysfs.h"
 
 #define SCD_MSG_PREPARE_PROCESS         0x1
@@ -401,6 +405,8 @@ int mcctrl_ikc_is_valid_thread(ihk_os_t os, int cpu);
 struct mcctrl_wakeup_desc {
 	int status;
 	int err;
+	refcount_t count;
+	int free_at_put;
 	wait_queue_head_t wq;
 	struct list_head chain;
 	int free_addrs_count;

executer/kernel/mcctrl/procfs.c

@@ -611,7 +611,7 @@ static ssize_t __mckernel_procfs_read_write(
 
 		ret = mcctrl_ikc_send_wait(osnum_to_os(e->osnum),
 					   (pid > 0) ? ppd->ikc_target_cpu : 0,
-					   &isp, HZ, NULL, &do_free, 1, r);
+					   &isp, 5000, NULL, &do_free, 1, r);
 
 		if (!do_free && ret >= 0) {
 			ret = -EIO;
@@ -879,7 +879,7 @@ static int mckernel_procfs_buff_release(struct inode *inode, struct file *file)
 
 		rc = -EIO;
 		ret = mcctrl_ikc_send_wait(info->os, 0,
-					   &isp, 5 * HZ, NULL, &do_free, 1, r);
+					   &isp, 5000, NULL, &do_free, 1, r);
 
 		if (!do_free && ret >= 0) {
 			ret = -EIO;
@@ -977,7 +977,7 @@ static ssize_t mckernel_procfs_buff_read(struct file *file, char __user *ubuf,
 		done = 1;
 		ret = mcctrl_ikc_send_wait(os,
 					   (pid > 0) ? ppd->ikc_target_cpu : 0,
-					   &isp, 5 * HZ, NULL, &do_free, 1, r);
+					   &isp, 5000, NULL, &do_free, 1, r);
 
 		if (!do_free && ret >= 0) {
 			ret = -EIO;

executer/kernel/mcctrl/syscall.c

@@ -495,8 +495,6 @@ int remote_page_fault(struct mcctrl_usrdata *usrdata, void *fault_addr,
 		      struct ikc_scd_packet *packet)
 {
 	int error;
-	struct mcctrl_wakeup_desc *desc;
-	int do_frees = 1;
 	
 	dprintk("%s: tid: %d, fault_addr: %p, reason: %lu\n",
 			__FUNCTION__, task_pid_vnr(current), fault_addr, (unsigned long)reason);
@@ -506,19 +504,9 @@ int remote_page_fault(struct mcctrl_usrdata *usrdata, void *fault_addr,
 	packet->fault_address = (unsigned long)fault_addr;
 	packet->fault_reason = reason;
 
-	/* we need to alloc desc ourselves because GFP_ATOMIC */
-retry_alloc:
-	desc = kmalloc(sizeof(*desc), GFP_ATOMIC);
-	if (!desc) {
-		pr_warn("WARNING: coudln't alloc remote page fault wait desc, retrying..\n");
-		goto retry_alloc;
-	}
-
 	/* packet->target_cpu was set in rus_vm_fault if a thread was found */
 	error = mcctrl_ikc_send_wait(usrdata->os, packet->target_cpu, packet,
-				     0, desc, &do_frees, 0);
-	if (do_frees)
-		kfree(desc);
+				     0, NULL, NULL, 0);
 	if (error < 0) {
 		pr_warn("%s: WARNING: failed to request remote page fault PID %d: %d\n",
 			__func__, packet->pid, error);

executer/user/mcexec.c

@@ -2419,6 +2419,7 @@ int main(int argc, char **argv)
 	}
 #endif // MCEXEC_BIND_MOUNT
 
+	/* fget executable as well */
 	if ((ret = load_elf_desc_shebang(argv[optind], &desc,
 					 &shebang_argv, 1 /* execvp */))) {
 		fprintf(stderr, "%s: could not load program: %s\n",
@@ -2860,6 +2861,14 @@ int main(int argc, char **argv)
 		fprintf(stderr, "error: transferring image\n");
 		return -1;
 	}
+
+	/* fput executable */
+	if ((ret = ioctl(fd, MCEXEC_UP_CLOSE_EXEC)) != 0) {
+		fprintf(stderr, "error: MCEXEC_UP_CLOSE_EXEC failed with %d\n",
+			ret);
+		return 1;
+	}
+
 	fflush(stdout);
 	fflush(stderr);
 	
@@ -3489,7 +3498,7 @@ checkexist_resolvelinks:
 			if (buf[0] == '/') {
 				/* cannot snprintf from same source and dest */
 				n = snprintf(tmpbuf2, PATH_MAX, "%s/%s", buf,
-					     linkpath);
+					     linkpath + 1);
 				if (n >= PATH_MAX)
 					return in;
 				strcpy(tmpbuf, tmpbuf2);
@@ -4111,11 +4120,6 @@ int main_loop(struct thread_data_s *my_thread)
 			   It is done by not calling do_syscall_return(fd, cpu, 0, 0, 0, 0, 0);
 			   here and making McKernel side wait until release_handler() is called. */
 
-			/* Drop executable file */
-			if ((ret = ioctl(fd, MCEXEC_UP_CLOSE_EXEC)) != 0) {
-				fprintf(stderr, "WARNING: close_exec() couldn't find exec file?\n");
-			}
-
 			__dprintf("__NR_exit/__NR_exit_group: %ld (cpu_id: %d)\n",
 					w.sr.args[0], cpu);
 			if(w.sr.number == __NR_exit_group){
@@ -4308,15 +4312,6 @@ gettid_out:
 				__dprintf("pid(%d): signals and syscall threads OK\n", 
 						getpid());
 
-				/* Hold executable also in the child process */
-				if ((ret = ioctl(fd, MCEXEC_UP_OPEN_EXEC, exec_path)) 
-					!= 0) {
-					fprintf(stderr, "Error: open_exec() fails for %s: %d (fd: %d)\n", 
-							exec_path, ret, fd);
-					fs->status = -errno;
-					goto fork_child_sync_pipe;
-				}
-
 				/* Check if we need to limit number of threads in the pool */
 				if ((ret = ioctl(fd, MCEXEC_UP_GET_NUM_POOL_THREADS)) < 0) {
 					fprintf(stderr, "Error: obtaining thread pool count\n");
@@ -4472,6 +4467,7 @@ fork_err:
 					}
 					filename = pathbuf;
 
+					/* fget executable as well */
 					if ((ret = load_elf_desc_shebang(filename, &desc,
 									 &shebang_argv, 0)) != 0) {
 						goto return_execve1;
@@ -4569,6 +4565,13 @@ return_execve1:
 					}
 					__dprintf("%s", "execve(): image transferred\n");
 
+					/* fput executable */
+					if ((ret = ioctl(fd, MCEXEC_UP_CLOSE_EXEC)) != 0) {
+						fprintf(stderr, "error: MCEXEC_UP_CLOSE_EXEC failed with %d\n",
+							ret);
+						return 1;
+					}
+
 					if (close_cloexec_fds(fd) < 0) {
 						ret = EINVAL;
 						goto return_execve2;

kernel/debug.c

@@ -87,7 +87,7 @@ void kputs(char *buf)
 	debug_spin_unlock_irqrestore(&kmsg_buf->lock, flags_inner);
 	kprintf_unlock(flags_outer);
 
-	if (irqflags_can_interrupt(flags_outer) &&
+	if (!cpu_interrupt_disabled() &&
 			DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
 		eventfd(IHK_OS_EVENTFD_TYPE_KMSG);
 		ihk_mc_delay_us(IHK_KMSG_NOTIFY_DELAY);
@@ -128,7 +128,7 @@ int __kprintf(const char *format, ...)
 	}
 
 	debug_spin_unlock_irqrestore(&kmsg_buf->lock, flags_inner);
-	if (irqflags_can_interrupt(flags_inner) &&
+	if (!cpu_interrupt_disabled() &&
 			DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
 		eventfd(IHK_OS_EVENTFD_TYPE_KMSG);
 		ihk_mc_delay_us(IHK_KMSG_NOTIFY_DELAY);
@@ -171,7 +171,7 @@ int kprintf(const char *format, ...)
 	debug_spin_unlock_irqrestore(&kmsg_buf->lock, flags_inner);
 	kprintf_unlock(flags_outer);
 
-	if (irqflags_can_interrupt(flags_outer) &&
+	if (!cpu_interrupt_disabled() &&
 			DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
 		eventfd(IHK_OS_EVENTFD_TYPE_KMSG);
 		ihk_mc_delay_us(IHK_KMSG_NOTIFY_DELAY);

kernel/devobj.c

@@ -136,6 +136,27 @@ int devobj_create(int fd, size_t len, off_t off, struct memobj **objp, int *maxp
 	error = 0;
 	*objp = to_memobj(obj);
 	*maxprotp = result.maxprot;
+
+#ifdef ENABLE_FUGAKU_HACKS
+	/* Pre-populate device file PFNs for PMIx shared mem */
+	if (!strncmp(obj->memobj.path,
+				"/var/opt/FJSVtcs/ple/daemonif", 29)) {
+		off_t offset;
+		uintptr_t phys;
+		unsigned long flag;
+
+		for (offset = 0; offset < obj->memobj.size; offset += PAGE_SIZE) {
+			if (devobj_get_page(&obj->memobj, offset, PAGE_P2ALIGN,
+						&phys, &flag, 0) < 0) {
+				kprintf("%s: WARNING: failed to populate offset %lu in %s\n",
+						__func__, offset, obj->memobj.path);
+			}
+		}
+		dkprintf("%s: pre-populated PFNs for %s, len: %lu\n",
+			__func__, obj->memobj.path, obj->memobj.size);
+	}
+#endif
+
 	obj = NULL;
 
 out:
@@ -200,6 +221,10 @@ static int devobj_get_page(struct memobj *memobj, off_t off, int p2align, uintpt
 	uintptr_t attr;
 	ihk_mc_user_context_t ctx;
 	int ix;
+	unsigned long irqstate;
+#ifdef ENABLE_FUGAKU_HACKS
+	int page_fault_attempts = 5;
+#endif
 
 	dkprintf("devobj_get_page(%p %lx,%lx,%d)\n", memobj, obj->handle, off, p2align);
 
@@ -214,8 +239,15 @@ static int devobj_get_page(struct memobj *memobj, off_t off, int p2align, uintpt
 #ifdef PROFILE_ENABLE
 	profile_event_add(PROFILE_page_fault_dev_file, PAGE_SIZE);
 #endif // PROFILE_ENABLE
+
+	irqstate = ihk_mc_spinlock_lock(&obj->pfn_table_lock);
 	pfn = obj->pfn_table[ix];
+	ihk_mc_spinlock_unlock(&obj->pfn_table_lock, irqstate);
+
 	if (!(pfn & PFN_VALID)) {
+#ifdef ENABLE_FUGAKU_HACKS
+pf_retry:
+#endif
 		ihk_mc_syscall_arg0(&ctx) = PAGER_REQ_PFN;
 		ihk_mc_syscall_arg1(&ctx) = obj->handle;
 		ihk_mc_syscall_arg2(&ctx) = off & ~(PAGE_SIZE - 1);
@@ -241,8 +273,24 @@ static int devobj_get_page(struct memobj *memobj, off_t off, int p2align, uintpt
 			pfn |= attr;
 			dkprintf("devobj_get_page(%p %lx,%lx,%d):PFN_PRESENT after %#lx\n", memobj, obj->handle, off, p2align, pfn);
 		}
+#ifdef ENABLE_FUGAKU_HACKS
+		else if (page_fault_attempts > 0) {
+			kprintf("%s(): va: 0x%lx !PFN_PRESENT for offset %lu in %s, "
+					"page_fault_attempts: %d\n",
+					__func__, virt_addr, off,
+					memobj->path ? memobj->path : "<unknown>",
+					page_fault_attempts);
+			--page_fault_attempts;
+			goto pf_retry;
+		}
+#endif
 
-		obj->pfn_table[ix] = pfn;
+		/* Update atomically if unset */
+		irqstate = ihk_mc_spinlock_lock(&obj->pfn_table_lock);
+		if (obj->pfn_table[ix] == 0) {
+			obj->pfn_table[ix] = pfn;
+		}
+		ihk_mc_spinlock_unlock(&obj->pfn_table_lock, irqstate);
 		// Don't call memory_stat_rss_add() because devobj related pages don't reside in main memory
 	}
 

kernel/profile.c

@@ -100,8 +100,13 @@ void profile_event_add(enum profile_event_type type, uint64_t tsc)
 		return;
 
 	if (!cpu_local_var(current)->profile_events) {
-		if (profile_alloc_events(cpu_local_var(current)) < 0)
+		if (type == PROFILE_mpol_alloc_missed) {
 			return;
+		}
+
+		if (profile_alloc_events(cpu_local_var(current)) < 0) {
+			return;
+		}
 	}
 
 	if (type < PROFILE_EVENT_MAX) {

kernel/syscall.c

@@ -246,6 +246,12 @@ long do_syscall(struct syscall_request *req, int cpu)
 			unsigned long flags;
 			DECLARE_WAITQ_ENTRY(scd_wq_entry, cpu_local_var(current));
 
+#ifdef ENABLE_FUGAKU_HACKS
+			if (req->number == __NR_epoll_wait ||
+					req->number == __NR_epoll_pwait)
+				goto schedule;
+#endif
+
 			if (thread->rpf_backlog) {
 				void (*func)(void *) = thread->rpf_backlog;
 				void *arg = thread->rpf_arg;
@@ -287,6 +293,9 @@ long do_syscall(struct syscall_request *req, int cpu)
 				continue;
 			}
 
+#ifdef ENABLE_FUGAKU_HACKS
+schedule:
+#endif
 			flags = cpu_disable_interrupt_save();
 
 			/* Try to sleep until notified */
@@ -2210,6 +2219,8 @@ straight_out:
 		}
 #endif // PROFILE_ENABLE
 		if (error == -ESRCH) {
+			int populate_flags = 0;
+
 			dkprintf("do_mmap:hit non VREG\n");
 			/*
 			 * XXX: temporary:
@@ -2221,8 +2232,21 @@ straight_out:
 				vrflags &= ~VR_MEMTYPE_MASK;
 				vrflags |= VR_MEMTYPE_UC;
 			}
+
+#ifdef ENABLE_FUGAKU_HACKS
+#ifdef ENABLE_TOFU
+			if (!strncmp("/var/opt/FJSVtcs/ple/daemonif/",
+						thread->proc->fd_path[fd], 30)) {
+				dkprintf("%s: MAP_POPULATE | MAP_LOCKED for %s\n",
+					__func__, thread->proc->fd_path[fd]);
+				populate_flags = (MAP_POPULATE | MAP_LOCKED);
+			}
+#endif
+#endif
+
 			error = devobj_create(fd, len, off, &memobj, &maxprot, 
-					prot, (flags & (MAP_POPULATE | MAP_LOCKED)));
+					prot,
+					populate_flags | (flags & (MAP_POPULATE | MAP_LOCKED)));
 
 			if (!error) {
 #ifdef PROFILE_ENABLE
@@ -11149,7 +11173,16 @@ long syscall(int num, ihk_mc_user_context_t *ctx)
 	}
 #endif // PROFILE_ENABLE
 
-	if (smp_load_acquire(&v->flags) & CPU_FLAG_NEED_RESCHED) {
+#ifdef ENABLE_FUGAKU_HACKS
+	/* Do not deschedule when returning from an event (e.g., MPI) */
+	if (!(num == __NR_epoll_wait ||
+				num == __NR_epoll_pwait ||
+				num == __NR_ppoll) &&
+			smp_load_acquire(&v->flags) & CPU_FLAG_NEED_RESCHED)
+#else
+	if (smp_load_acquire(&v->flags) & CPU_FLAG_NEED_RESCHED)
+#endif
+	{
 		check_need_resched();
 	}
 

lib/include/ihk/cpu.h

@@ -23,9 +23,7 @@ extern int num_processors;
 
 void cpu_enable_interrupt(void);
 void cpu_disable_interrupt(void);
-#ifdef ENABLE_FUGAKU_HACKS
 int cpu_interrupt_disabled(void);
-#endif
 void cpu_halt(void);
 #ifdef ENABLE_FUGAKU_HACKS
 void cpu_halt_panic(void);

scripts/mckernel.spec.in

@@ -15,6 +15,11 @@
 %{!?kernel_dir: %global kernel_dir /usr/src/kernels/%{kernel_version}}
 %define krequires %(echo %{kernel_version} | sed "s/.%{_target_cpu}$//")
 %define ktag %(echo %{krequires} | tr '-' '_' | sed -e 's/\.el[0-9_]*$//' | sed -e 's/\.\([a-zA-Z]\)/_\1/')
+%if "@ENABLE_UTI@" == "ON"
+%define enable_uti 1
+%else
+%define enable_uti 0
+%endif
 
 Name: mckernel
 Version: @MCKERNEL_VERSION@
@@ -77,9 +82,10 @@ pushd build
 	%{?cmake_libdir:-DCMAKE_INSTALL_LIBDIR=%{cmake_libdir}} \
 	%{?build_target:-DBUILD_TARGET=%{build_target}} \
 	%{?toolchain_file:-DCMAKE_TOOLCHAIN_FILE=%{toolchain_file}} \
-	-DENABLE_TOFU=ON -DENABLE_FUGAKU_HACKS=ON \
-	-DENABLE_KRM_WORKAROUND=OFF -DWITH_KRM=ON \
-	-DENABLE_FUGAKU_DEBUG=OFF -DENABLE_UTI=ON \
+	-DENABLE_TOFU=@ENABLE_TOFU@ -DENABLE_FUGAKU_HACKS=@ENABLE_FUGAKU_HACKS@ \
+	-DENABLE_KRM_WORKAROUND=@ENABLE_KRM_WORKAROUND@ -DWITH_KRM=@WITH_KRM@ \
+	-DENABLE_FUGAKU_DEBUG=@ENABLE_FUGAKU_DEBUG@ -DENABLE_UTI=@ENABLE_UTI@ \
+	-DENABLE_FJMPI_WORKAROUND=@ENABLE_FJMPI_WORKAROUND@ \
 	..
 %make_build
 popd
@@ -112,6 +118,7 @@ popd
 %{_libdir}/libsched_yield.so.1.0.0
 %{_libdir}/libsched_yield.so
 %{_libdir}/libldump2mcdump.so
+%if 0%{?enable_uti}
 %{_libdir}/libmck_syscall_intercept.so
 %{_libdir}/libsyscall_intercept.so.0.1.0
 %{_libdir}/libsyscall_intercept.so.0
@@ -119,6 +126,7 @@ popd
 %{_libdir}/mck/libuti.so.1.0.0
 %{_libdir}/mck/libuti.so.1
 %{_libdir}/mck/libuti.so
+%endif
 %{_sysconfdir}/irqbalance_mck.in
 %{_mandir}/man1/mcreboot.1.gz
 %{_mandir}/man1/ihkconfig.1.gz

scripts/mcreboot-smp.sh.in

@@ -119,7 +119,7 @@ if [ "${pid}" != "" ]; then
     ${SUDO} kill -9 ${pid} > /dev/null 2> /dev/null
 fi
 if [ "${redirect_kmsg}" != "0" -o "${mon_interval}" != "-1" ]; then
-    ${SBINDIR}/ihkmond -f ${facility} -k ${redirect_kmsg} -i ${mon_interval}
+    ${SUDO} ${SBINDIR}/ihkmond -f ${facility} -k ${redirect_kmsg} -i ${mon_interval}
 fi
 
 disable_irqbalance_mck() {

test/issues/1463/C1463.sh

@@ -0,0 +1,58 @@
+#/bin/sh
+
+USELTP=0
+USEOSTEST=0
+
+. ../../common.sh
+
+issue="1463"
+tid=01
+
+TEST_DIR="/tmp/test"
+ABS_PATH="${TEST_DIR}"
+REL_PATH="./test"
+ABS_LN="${TEST_DIR}_1463_abs_ln"
+REL_LN="${TEST_DIR}_1463_rel_ln"
+
+mkdir -p ${TEST_DIR}
+touch ${TEST_DIR}/L.dir
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+ln -fns ${ABS_PATH} ${ABS_LN}
+mcexec readlink ${ABS_LN}/L.dir | tee ./${tname}.txt
+cnt=`grep "a.dir" ./${tname}.txt | wc -l`
+if [ ${cnt} -eq 1 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+ln -fns ${REL_PATH} ${REL_LN}
+mcexec readlink ${REL_LN}/L.dir | tee ./${tname}.txt
+cnt=`grep "a.dir" ./${tname}.txt | wc -l`
+if [ ${cnt} -eq 1 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+
+tname=`printf "C${issue}T%02d" ${tid}`
+echo "*** ${tname} start *******************************"
+mcexec cat /sys/devices/system/cpu/offline | tee ./${tname}.txt
+echo "** (expected blank output)"
+lines=`grep -e "[0-9]" ./${tname}.txt | wc -l`
+if [ ${lines} -eq 0 ]; then
+	echo "*** ${tname} PASSED ******************************"
+else
+	echo "*** ${tname} FAILED ******************************"
+fi
+let tid++
+echo ""
+

test/issues/1463/Makefile

@@ -0,0 +1,12 @@
+CFLAGS=
+LDFLAGS=
+
+TARGET=
+
+all: $(TARGET)
+
+test: all
+	sh ./C1463.sh
+clean:
+	rm -f $(TARGET) *.o *.txt
+

test/issues/1463/README

@@ -0,0 +1,49 @@
+【Issue#1463 動作確認】
+□ テスト内容
+Issue#1463の修正は、mcexec.c: mcoverlay_path() 内で行われる/sys/ 配下への
+リンク解決処理に関する修正である。
+上記のリンク解決処理を/tmp/ 配下にも行うようにするテストパッチを適用した上で
+/tmp/ 配下へのmcoverlay_path()を実行して動作を確認する。
+
+なお、リンク解決処理の動作確認には、McKernelによって作成される、
+/sys/devices/virtual/mcos/mcos0/sys/test/L.dir を利用する。
+このL.dirは、同ディレクトリのa.dir へのシンボリックリンクとなっている。
+
+C1463T01:
+  以下の流れで、リンク解決処理対象パスの途中に絶対パスのシンボリックリンクが
+ 存在している場合にも、/sys/devices/virtual/mcos/mcos0/sys/ 下に
+ 誘導されることを確認する。
+  a. /tmp/test/L.dir に空のファイルを作成
+  b. /tmp/test への絶対パスのシンボリックリンクとして、/tmp/test_1463_abs_ln を作成
+  c. mcexec readlink /tmp/test_1463_abs_ln/L.dir を実行し、a.dir が出力されることを確認
+
+C1463T02:
+  以下の流れで、リンク解決処理対象パスの途中に相対パスのシンボリックリンクが
+ 存在している場合にも、/sys/devices/virtual/mcos/mcos0/sys/ 下に
+ 誘導されることを確認する。
+  a. /tmp/test/L.dir に空のファイルを作成
+  b. /tmp/test への相対パスのシンボリックリンクとして、/tmp/test_1463_rel_ln を作成
+  c. mcexec readlink /tmp/test_1463_rel_ln/L.dir を実行し、a.dir が出力されることを確認
+
+C1463T03:
+  以下の流れで、/sys/ 配下へのアクセスが/sys/devices/virtual/mcos/mcos0/sys/ 下に
+ 誘導されることを確認する。
+  a. mcexecで確認した場合の /sys/devices/system/cpu/offline が空であることを確認
+   ※通常、mckernelではofflineのCPUが存在しないため
+
+□ 実行手順
+・下記の手順でテストを実行する
+$ cd <mckernel>
+$ patch -p0 < test/issues/1463/tmp_overlay_path.patch
+(build mckernel)
+$ cd test/issues/1463
+$ make test
+
+McKernelのインストール先や、OSTEST, LTPの配置場所は、
+$HOME/.mck_test_config を参照している
+.mck_test_config は、McKernelをビルドした際に生成されるmck_test_config.sample ファイルを
+$HOMEにコピーし、適宜編集する
+
+□ 実行結果
+x86_64_result.log aarch64_result.log 参照。
+すべての項目をPASSしていることを確認。

test/issues/1463/aarch64_result.log

@@ -0,0 +1,15 @@
+sh ./C1463.sh
+mcstop+release.sh ... done
+mcreboot.sh -c 37-43,49-55 -m 2G@2,2G@3 -r 37-43:36+49-55:48 -O ... done
+*** C1463T01 start *******************************
+a.dir
+*** C1463T01 PASSED ******************************
+
+*** C1463T02 start *******************************
+a.dir
+*** C1463T02 PASSED ******************************
+
+*** C1463T03 start *******************************
+
+** (expected blank output)
+*** C1463T03 PASSED ******************************

test/issues/1463/tmp_overlay_path.patch

@@ -0,0 +1,17 @@
+diff --git executer/user/mcexec.c executer/user/mcexec.c
+index acae1f8..d220dd9 100644
+--- executer/user/mcexec.c
++++ executer/user/mcexec.c
+@@ -3458,6 +3458,12 @@ overlay_path(int dirfd, const char *in, char *buf, int *resolvelinks)
+ 		goto checkexist_resolvelinks;
+ 	}
+ 
++	/* for #1463's test */
++	if (!strncmp(path, "/tmp", 4) &&
++	    (path[4] == '/' || path[4] == '\0')) {
++		goto checkexist_resolvelinks;
++	}
++
+ 	return in;
+ 
+ checkexist_resolvelinks:

test/issues/1463/x86_64_result.log

@@ -0,0 +1,15 @@
+sh ./C1463.sh
+mcstop+release.sh ... done
+mcreboot.sh -c 1-7,9-15,17-23,25-31 -m 10G@0,10G@1 -r 1-7:0+9-15:8+17-23:16+25-31:24 -O ... done
+*** C1463T01 start *******************************
+a.dir
+*** C1463T01 PASSED ******************************
+
+*** C1463T02 start *******************************
+a.dir
+*** C1463T02 PASSED ******************************
+
+*** C1463T03 start *******************************
+
+** (expected blank output)
+*** C1463T03 PASSED ******************************