qsnet-rhel4-2.6.patch

非常经典的一个分布式系统

Index: linux-269-5502/fs/open.c
===================================================================
--- linux-269-5502.orig/fs/open.c
+++ linux-269-5502/fs/open.c
@@ -1029,6 +1029,8 @@ out_error:
 	goto out;
 }
 
+EXPORT_SYMBOL(sys_open);
+
 #ifndef __alpha__
 
 /*
Index: linux-269-5502/fs/read_write.c
===================================================================
--- linux-269-5502.orig/fs/read_write.c
+++ linux-269-5502/fs/read_write.c
@@ -145,6 +145,7 @@ asmlinkage off_t sys_lseek(unsigned int 
 bad:
 	return retval;
 }
+EXPORT_SYMBOL(sys_lseek);
 
 #ifdef __ARCH_WANT_SYS_LLSEEK
 asmlinkage long sys_llseek(unsigned int fd, unsigned long offset_high,
Index: linux-269-5502/fs/select.c
===================================================================
--- linux-269-5502.orig/fs/select.c
+++ linux-269-5502/fs/select.c
@@ -539,3 +539,4 @@ out_fds:
 	poll_freewait(&table);
 	return err;
 }
+EXPORT_SYMBOL_GPL(sys_poll);
Index: linux-269-5502/fs/exec.c
===================================================================
--- linux-269-5502.orig/fs/exec.c
+++ linux-269-5502/fs/exec.c
@@ -56,6 +56,8 @@
 #include <linux/kmod.h>
 #endif
 
+#include <linux/ptrack.h>
+
 int core_uses_pid;
 char core_pattern[65] = "core";
 int suid_dumpable = 0;
@@ -1214,6 +1216,9 @@ int do_execve(char * filename,
 	if (retval < 0)
 		goto out;
 
+	/* notify any ptrack callbacks of the process exec */
+	ptrack_call_callbacks(PTRACK_PHASE_EXEC, NULL);
+
 	retval = search_binary_handler(bprm,regs);
 	if (retval >= 0) {
 		free_arg_pages(bprm);
Index: linux-269-5502/arch/i386/Kconfig
===================================================================
--- linux-269-5502.orig/arch/i386/Kconfig
+++ linux-269-5502/arch/i386/Kconfig
@@ -960,6 +960,9 @@ config REGPARM
 	generate incorrect output with certain kernel constructs when
 	-mregparm=3 is used.
 
+source "mm/Kconfig"
+source "kernel/Kconfig"
+	
 endmenu
 
 
Index: linux-269-5502/arch/i386/defconfig
===================================================================
--- linux-269-5502.orig/arch/i386/defconfig
+++ linux-269-5502/arch/i386/defconfig
@@ -119,6 +119,8 @@ CONFIG_MTRR=y
 CONFIG_IRQBALANCE=y
 CONFIG_HAVE_DEC_LOCK=y
 # CONFIG_REGPARM is not set
+CONFIG_IOPROC=y
+CONFIG_PTRACK=y
 
 #
 # Power management options (ACPI, APM)
Index: linux-269-5502/arch/ia64/Kconfig
===================================================================
--- linux-269-5502.orig/arch/ia64/Kconfig
+++ linux-269-5502/arch/ia64/Kconfig
@@ -316,6 +316,9 @@ config IA64_PALINFO
 	  To use this option, you have to ensure that the "/proc file system
 	  support" (CONFIG_PROC_FS) is enabled, too.
 
+source "mm/Kconfig"
+source "kernel/Kconfig"
+
 source "drivers/firmware/Kconfig"
 
 source "fs/Kconfig.binfmt"
Index: linux-269-5502/arch/ia64/defconfig
===================================================================
--- linux-269-5502.orig/arch/ia64/defconfig
+++ linux-269-5502/arch/ia64/defconfig
@@ -83,6 +83,8 @@ CONFIG_IA32_SUPPORT=y
 CONFIG_COMPAT=y
 CONFIG_PERFMON=y
 CONFIG_IA64_PALINFO=y
+CONFIG_IOPROC=y
+CONFIG_PTRACK=y
 
 #
 # Firmware Drivers
Index: linux-269-5502/arch/x86_64/Kconfig
===================================================================
--- linux-269-5502.orig/arch/x86_64/Kconfig
+++ linux-269-5502/arch/x86_64/Kconfig
@@ -401,6 +401,9 @@ config X86_MCE_AMD
  	   Additional support for AMD specific MCE features such as
  	   the DRAM Error Threshold.
 
+source "mm/Kconfig"
+source "kernel/Kconfig"
+
 endmenu
 
 
Index: linux-269-5502/arch/x86_64/defconfig
===================================================================
--- linux-269-5502.orig/arch/x86_64/defconfig
+++ linux-269-5502/arch/x86_64/defconfig
@@ -87,6 +87,8 @@ CONFIG_NR_CPUS=8
 CONFIG_GART_IOMMU=y
 CONFIG_SWIOTLB=y
 CONFIG_X86_MCE=y
+CONFIG_IOPROC=y
+CONFIG_PTRACK=y
 
 #
 # Power management options
Index: linux-269-5502/kernel/ptrack.c
===================================================================
--- /dev/null
+++ linux-269-5502/kernel/ptrack.c
@@ -0,0 +1,145 @@
+/*
+ *    Copyright (C) 2000  Regents of the University of California
+ *
+ *    This program is free software; you can redistribute it and/or modify
+ *    it under the terms of the GNU General Public License as published by
+ *    the Free Software Foundation; either version 2 of the License, or
+ *    (at your option) any later version.
+ *
+ *    This program is distributed in the hope that it will be useful,
+ *    but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *    GNU General Public License for more details.
+ *
+ *    You should have received a copy of the GNU General Public License
+ *    along with this program; if not, write to the Free Software
+ *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ * Derived from exit_actn.c by
+ *    Copyright (C) 2003 Quadrics Ltd.
+ */
+
+
+#include <linux/module.h>
+#include <linux/spinlock.h>
+#include <linux/sched.h>
+#include <linux/ptrack.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+
+#include <asm/errno.h>
+
+int
+ptrack_register (ptrack_callback_t callback, void *arg)
+{
+       struct ptrack_desc *desc = kmalloc (sizeof (struct ptrack_desc), GFP_KERNEL);
+       
+       if (desc == NULL)
+               return -ENOMEM;
+
+	desc->callback = callback;
+	desc->arg      = arg;
+       
+       list_add_tail (&desc->link, &current->ptrack_list);
+       
+       return 0;
+}
+
+void
+ptrack_deregister (ptrack_callback_t callback, void *arg)
+{      
+       struct list_head *el, *nel;
+       
+       list_for_each_safe (el, nel, &current->ptrack_list) {
+               struct ptrack_desc *desc = list_entry (el, struct ptrack_desc, link);
+               
+               if (desc->callback == callback && desc->arg == arg) {
+                       list_del (&desc->link);
+                       kfree (desc);
+               }
+       }
+}
+
+int
+ptrack_registered (ptrack_callback_t callback, void *arg)
+{
+       struct list_head *el;
+       
+       list_for_each (el, &current->ptrack_list) {
+               struct ptrack_desc *desc = list_entry (el, struct ptrack_desc, link);
+               
+               if (desc->callback == callback && desc->arg == arg)
+                       return 1;
+       }
+       return 0;
+}      
+        
+int
+ptrack_call_callbacks (int phase, struct task_struct *child)
+{
+       struct list_head *el, *nel;
+       struct ptrack_desc *new;
+	int res;
+
+       if (phase == PTRACK_PHASE_CLONE)
+               INIT_LIST_HEAD (&child->ptrack_list);
+
+       list_for_each_safe (el, nel, &current->ptrack_list) {
+               struct ptrack_desc *desc = list_entry (el, struct ptrack_desc, link);
+               
+	       res = desc->callback (desc->arg, phase, child);
+               
+               switch (phase)
+               {
+               case PTRACK_PHASE_EXIT:
+                       list_del (&desc->link);
+                       kfree (desc);
+                       break;
+                       
+               case PTRACK_PHASE_CLONE:
+		       switch (res)
+		       {
+		       case PTRACK_FINISHED:
+			       break;
+
+		       case PTRACK_INNHERIT:
+			       if ((new = kmalloc (sizeof (struct ptrack_desc), GFP_ATOMIC)) == NULL)
+			       {
+				       /* allocation failed - notify that this process is not going
+					* to be started by signalling clone failure.
+					*/
+				       desc->callback (desc->arg, PTRACK_PHASE_CLONE_FAIL, child);
+				       
+				       goto failed;
+			       }
+
+			       	new->callback = desc->callback;
+				new->arg      = desc->arg;
+                               
+                               list_add_tail (&new->link, &child->ptrack_list);
+			       break;
+
+		       case PTRACK_DENIED:
+			       goto failed;
+                       }
+		       break;
+               }
+       }
+
+       return 0;
+
+ failed:
+       while (! list_empty (&child->ptrack_list))
+       {
+	       struct ptrack_desc *desc = list_entry (child->ptrack_list.next, struct ptrack_desc, link);
+	       
+	       desc->callback (desc->arg, PTRACK_PHASE_CLONE_FAIL, child);
+
+	       list_del (&desc->link);
+	       kfree (desc);
+       }
+       return 1;
+}
+EXPORT_SYMBOL(ptrack_register);
+EXPORT_SYMBOL(ptrack_deregister);
+EXPORT_SYMBOL(ptrack_registered);
Index: linux-269-5502/kernel/signal.c
===================================================================
--- linux-269-5502.orig/kernel/signal.c
+++ linux-269-5502/kernel/signal.c
@@ -2329,6 +2329,7 @@ sys_kill(int pid, int sig)
 
 	return kill_something_info(sig, &info, pid);
 }
+EXPORT_SYMBOL_GPL(sys_kill);
 
 /**
  *  sys_tgkill - send signal to one specific thread
Index: linux-269-5502/kernel/Kconfig
===================================================================
--- /dev/null
+++ linux-269-5502/kernel/Kconfig
@@ -0,0 +1,14 @@
+#
+# Kernel subsystem specific config
+# 
+
+# Support for Process Tracking callbacks
+#
+config PTRACK
+	bool "Enable PTRACK process tracking hooks"
+	default y
+	help
+	This option enables hooks to be called when processes are
+	created and destoryed in order for a resource management 
+	system to know which processes are a member of a "job" and 
+	to be able to clean up when the job is terminated.
Index: linux-269-5502/kernel/Makefile
===================================================================
--- linux-269-5502.orig/kernel/Makefile
+++ linux-269-5502/kernel/Makefile
@@ -26,6 +26,7 @@ obj-$(CONFIG_AUDIT) += audit.o
 obj-$(CONFIG_AUDITSYSCALL) += auditsc.o
 obj-$(CONFIG_AUDITFILESYSTEM) += auditfs.o
 obj-$(CONFIG_KPROBES) += kprobes.o
+obj-$(CONFIG_PTRACK) += ptrack.o
 
 ifneq ($(CONFIG_IA64),y)
 # According to Alan Modra <[e-mail]>, the -fno-omit-frame-pointer is
Index: linux-269-5502/kernel/exit.c
===================================================================
--- linux-269-5502.orig/kernel/exit.c
+++ linux-269-5502/kernel/exit.c
@@ -32,6 +32,8 @@
 #include <asm/pgtable.h>
 #include <asm/mmu_context.h>
 
+#include <linux/ptrack.h>
+
 extern void sem_exit (void);
 extern struct task_struct *child_reaper;
 
@@ -825,6 +827,9 @@ asmlinkage NORET_TYPE void do_exit(long 
 		current->tux_exit();
 	}
 
+	/* Notify any ptrack callbacks of the process exit */
+	ptrack_call_callbacks(PTRACK_PHASE_EXIT, NULL);
+
 	if (unlikely(tsk->audit_context))
 		audit_free(tsk);
 	__exit_mm(tsk);
Index: linux-269-5502/kernel/fork.c
===================================================================
--- linux-269-5502.orig/kernel/fork.c
+++ linux-269-5502/kernel/fork.c
@@ -14,6 +14,7 @@
 #include <linux/config.h>
 #include <linux/slab.h>
 #include <linux/init.h>
+#include <linux/ptrack.h>
 #include <linux/unistd.h>
 #include <linux/smp_lock.h>
 #include <linux/module.h>
@@ -443,6 +444,9 @@ static struct mm_struct * mm_init(struct
 	mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 	mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 	mm->ioctx_list = NULL;
+#ifdef CONFIG_IOPROC
+	mm->ioproc_ops = NULL;
+#endif
 	mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 	mm->free_area_cache = TASK_UNMAPPED_BASE;
 
@@ -1312,6 +1316,11 @@ long do_fork(unsigned long clone_flags,
 			set_tsk_thread_flag(p, TIF_SIGPENDING);
 		}
 
+		if (ptrack_call_callbacks(PTRACK_PHASE_CLONE, p)) {
+			sigaddset(&p->pending.signal, SIGKILL);
+			set_tsk_thread_flag(p, TIF_SIGPENDING);
+		}
+
 		if (!(clone_flags & CLONE_STOPPED))
 			wake_up_new_task(p, clone_flags);
 		else
Index: linux-269-5502/Makefile
===================================================================
--- linux-269-5502.orig/Makefile
+++ linux-269-5502/Makefile
@@ -1,7 +1,7 @@
 VERSION = 2
 PATCHLEVEL = 6
 SUBLEVEL = 9
-EXTRAVERSION = -prep
+EXTRAVERSION = -prep.qp3.5.34.4qsnet
 RHEL_VERSION = 4
 RHEL_UPDATE = 5
 NAME=AC 1
Index: linux-269-5502/Documentation/vm/ioproc.txt
===================================================================
--- /dev/null
+++ linux-269-5502/Documentation/vm/ioproc.txt
@@ -0,0 +1,467 @@
+Linux IOPROC patch overview
+===========================
+
+The network interface for an HPC network differs significantly from
+network interfaces for traditional IP networks. HPC networks tend to
+be used directly from user processes and perform large RDMA transfers
+between theses processes address space. They also have a requirement
+for low latency communication, and typically achieve this by OS bypass
+techniques.  This then requires a different model to traditional
+interconnects, in that a process may need to expose a large amount of
+it's address space to the network RDMA.
+
+Locking down of memory has been a common mechanism for performing
+this, together with a pin-down cache implemented in user
+libraries. The disadvantage of this method is that large portions of
+the physical memory can be locked down for a single process, even if
+it's working set changes over the different phases of it's
+execution. This leads to inefficient memory utilisation - akin to the
+disadvantage of swapping compared to paging.
+
+This model also has problems where memory is being dynamically
+allocated and freed, since the pin down cache is unaware that memory
+may have been released by a call to munmap() and so it will still be
+locking down the now unused pages.
+
+Some modern HPC network interfaces implement their own MMU and are
+able to handle a translation fault during a network access. The
+Quadrics (http://www.quadrics.com) devices (Elan3 and Elan4) have done
+this for some time and we expect others to follow the same route in
+the relatively near future. These NICs are able to operate in an