Merge ../torvalds-2.6/
authorGreg KH <gregkh@suse.de>
Mon, 12 Sep 2005 19:10:59 +0000 (12:10 -0700)
committerGreg Kroah-Hartman <gregkh@suse.de>
Mon, 12 Sep 2005 19:10:59 +0000 (12:10 -0700)
1122 files changed:
COPYING
Documentation/00-INDEX
Documentation/CodingStyle
Documentation/DMA-API.txt
Documentation/DMA-ISA-LPC.txt [new file with mode: 0644]
Documentation/DocBook/journal-api.tmpl
Documentation/DocBook/kernel-hacking.tmpl
Documentation/DocBook/usb.tmpl
Documentation/MSI-HOWTO.txt
Documentation/RCU/RTFP.txt
Documentation/RCU/UP.txt
Documentation/RCU/checklist.txt
Documentation/RCU/rcu.txt
Documentation/RCU/rcuref.txt [new file with mode: 0644]
Documentation/RCU/whatisRCU.txt [new file with mode: 0644]
Documentation/applying-patches.txt [new file with mode: 0644]
Documentation/cpu-freq/cpufreq-stats.txt
Documentation/cpusets.txt
Documentation/crypto/descore-readme.txt
Documentation/dvb/bt8xx.txt
Documentation/dvb/ci.txt
Documentation/fb/cyblafb/bugs [new file with mode: 0644]
Documentation/fb/cyblafb/credits [new file with mode: 0644]
Documentation/fb/cyblafb/documentation [new file with mode: 0644]
Documentation/fb/cyblafb/fb.modes [new file with mode: 0644]
Documentation/fb/cyblafb/performance [new file with mode: 0644]
Documentation/fb/cyblafb/todo [new file with mode: 0644]
Documentation/fb/cyblafb/usage [new file with mode: 0644]
Documentation/fb/cyblafb/whycyblafb [new file with mode: 0644]
Documentation/fb/intel810.txt
Documentation/fb/modedb.txt
Documentation/feature-removal-schedule.txt
Documentation/filesystems/files.txt [new file with mode: 0644]
Documentation/filesystems/fuse.txt [new file with mode: 0644]
Documentation/filesystems/proc.txt
Documentation/filesystems/v9fs.txt [new file with mode: 0644]
Documentation/filesystems/vfs.txt
Documentation/ioctl/cdrom.txt
Documentation/kbuild/makefiles.txt
Documentation/kdump/kdump.txt
Documentation/kernel-parameters.txt
Documentation/mono.txt
Documentation/networking/bonding.txt
Documentation/networking/wan-router.txt
Documentation/pci.txt
Documentation/powerpc/eeh-pci-error-recovery.txt
Documentation/s390/s390dbf.txt
Documentation/scsi/ibmmca.txt
Documentation/sound/alsa/ALSA-Configuration.txt
Documentation/sparse.txt
Documentation/sysrq.txt
Documentation/uml/UserModeLinux-HOWTO.txt
Documentation/usb/gadget_serial.txt
Documentation/video4linux/CARDLIST.bttv
Documentation/video4linux/CARDLIST.saa7134
Documentation/video4linux/CARDLIST.tuner
Documentation/video4linux/Zoran
Documentation/x86_64/boot-options.txt
Kbuild [new file with mode: 0644]
MAINTAINERS
Makefile
REPORTING-BUGS
arch/alpha/Makefile
arch/alpha/kernel/alpha_ksyms.c
arch/alpha/kernel/entry.S
arch/alpha/kernel/head.S
arch/alpha/kernel/module.c
arch/alpha/kernel/osf_sys.c
arch/alpha/kernel/smp.c
arch/alpha/lib/dbg_stackcheck.S
arch/alpha/lib/dbg_stackkill.S
arch/arm/Kconfig
arch/arm/Kconfig.debug
arch/arm/Makefile
arch/arm/common/scoop.c
arch/arm/kernel/calls.S
arch/arm/kernel/entry-common.S
arch/arm/kernel/entry-header.S
arch/arm/kernel/head.S
arch/arm/kernel/iwmmxt.S
arch/arm/lib/copy_page.S
arch/arm/lib/csumpartialcopyuser.S
arch/arm/lib/getuser.S
arch/arm/lib/putuser.S
arch/arm/mach-pxa/corgi.c
arch/arm/mach-pxa/corgi_ssp.c
arch/arm/mach-s3c2410/devs.c
arch/arm/mach-s3c2410/mach-h1940.c
arch/arm/mm/copypage-v3.S
arch/arm/mm/copypage-v4wb.S
arch/arm/mm/copypage-v4wt.S
arch/arm/mm/proc-arm1020.S
arch/arm/mm/proc-arm1020e.S
arch/arm/mm/proc-arm1022.S
arch/arm/mm/proc-arm1026.S
arch/arm/mm/proc-arm6_7.S
arch/arm/mm/proc-arm720.S
arch/arm/mm/proc-macros.S
arch/arm/mm/proc-sa110.S
arch/arm/mm/proc-sa1100.S
arch/arm/mm/proc-v6.S
arch/arm/mm/tlb-v3.S
arch/arm/mm/tlb-v4.S
arch/arm/mm/tlb-v4wb.S
arch/arm/mm/tlb-v4wbi.S
arch/arm/mm/tlb-v6.S
arch/arm/nwfpe/entry26.S
arch/arm/vfp/entry.S
arch/arm26/Makefile
arch/arm26/kernel/entry.S
arch/arm26/lib/copy_page.S
arch/arm26/lib/csumpartialcopyuser.S
arch/arm26/lib/getuser.S
arch/arm26/lib/putuser.S
arch/arm26/mm/proc-funcs.S
arch/arm26/nwfpe/entry.S
arch/cris/Makefile
arch/cris/arch-v10/kernel/entry.S
arch/cris/arch-v32/kernel/entry.S
arch/frv/kernel/asm-offsets.c [new file with mode: 0644]
arch/h8300/Makefile
arch/i386/Makefile
arch/i386/boot/video.S
arch/i386/kernel/acpi/earlyquirk.c
arch/i386/kernel/acpi/wakeup.S
arch/i386/kernel/cpu/common.c
arch/i386/kernel/entry.S
arch/i386/kernel/head.S
arch/i386/kernel/io_apic.c
arch/i386/kernel/mpparse.c
arch/i386/kernel/ptrace.c
arch/i386/kernel/setup.c
arch/i386/kernel/smpboot.c
arch/i386/kernel/srat.c
arch/i386/kernel/time.c
arch/i386/kernel/vmlinux.lds.S
arch/i386/kernel/vsyscall-sigreturn.S
arch/i386/kernel/vsyscall.lds.S
arch/i386/pci/acpi.c
arch/i386/pci/mmconfig.c
arch/i386/power/swsusp.S
arch/ia64/Makefile
arch/ia64/hp/sim/boot/boot_head.S
arch/ia64/ia32/ia32_entry.S
arch/ia64/ia32/sys_ia32.c
arch/ia64/kernel/asm-offsets.c
arch/ia64/kernel/entry.S
arch/ia64/kernel/fsys.S
arch/ia64/kernel/gate.S
arch/ia64/kernel/head.S
arch/ia64/kernel/ivt.S
arch/ia64/kernel/mca.c
arch/ia64/kernel/mca_asm.S
arch/ia64/kernel/mca_drv.c
arch/ia64/kernel/minstate.h
arch/ia64/kernel/palinfo.c
arch/ia64/kernel/perfmon.c
arch/ia64/kernel/salinfo.c
arch/ia64/kernel/unwind.c
arch/ia64/mm/init.c
arch/ia64/sn/kernel/setup.c
arch/ia64/sn/kernel/xpnet.c
arch/m32r/Kconfig
arch/m32r/kernel/asm-offsets.c [new file with mode: 0644]
arch/m32r/kernel/smp.c
arch/m68k/Makefile
arch/m68k/amiga/amisound.c
arch/m68k/fpsp040/skeleton.S
arch/m68k/ifpsp060/iskeleton.S
arch/m68k/kernel/entry.S
arch/m68k/kernel/head.S
arch/m68k/mac/macboing.c
arch/m68k/math-emu/fp_emu.h
arch/m68knommu/Makefile
arch/m68knommu/platform/68360/head-ram.S [new file with mode: 0644]
arch/m68knommu/platform/68360/head-rom.S [new file with mode: 0644]
arch/mips/Kconfig
arch/mips/Makefile
arch/mips/configs/tb0287_defconfig [new file with mode: 0644]
arch/mips/kernel/asm-offsets.c [moved from arch/mips/kernel/offset.c with 100% similarity]
arch/mips/kernel/genrtc.c
arch/mips/kernel/i8259.c
arch/mips/kernel/irixioctl.c
arch/mips/kernel/irixsig.c
arch/mips/kernel/r2300_fpu.S
arch/mips/kernel/r2300_switch.S
arch/mips/kernel/r4k_fpu.S
arch/mips/kernel/r4k_switch.S
arch/mips/kernel/r6000_fpu.S
arch/mips/kernel/scall32-o32.S
arch/mips/kernel/scall64-64.S
arch/mips/kernel/syscall.c
arch/mips/kernel/sysirix.c
arch/mips/lib-32/memset.S
arch/mips/lib-64/memset.S
arch/mips/lib/dec_and_lock.c
arch/mips/lib/memcpy.S
arch/mips/lib/strlen_user.S
arch/mips/lib/strncpy_user.S
arch/mips/lib/strnlen_user.S
arch/mips/pci/Makefile
arch/mips/pci/fixup-tb0287.c [new file with mode: 0644]
arch/parisc/Makefile
arch/parisc/hpux/gate.S
arch/parisc/hpux/wrappers.S
arch/parisc/kernel/entry.S
arch/parisc/kernel/head.S
arch/parisc/kernel/process.c
arch/parisc/kernel/ptrace.c
arch/parisc/kernel/signal.c
arch/parisc/kernel/syscall.S
arch/parisc/lib/Makefile
arch/parisc/lib/bitops.c
arch/parisc/lib/debuglocks.c [deleted file]
arch/parisc/lib/fixup.S
arch/ppc/8xx_io/cs4218_tdm.c
arch/ppc/Kconfig
arch/ppc/Makefile
arch/ppc/boot/common/ns16550.c
arch/ppc/boot/common/util.S
arch/ppc/kernel/Makefile
arch/ppc/kernel/cpu_setup_6xx.S
arch/ppc/kernel/cpu_setup_power4.S
arch/ppc/kernel/dma-mapping.c
arch/ppc/kernel/entry.S
arch/ppc/kernel/fpu.S
arch/ppc/kernel/head.S
arch/ppc/kernel/head_44x.S
arch/ppc/kernel/head_4xx.S
arch/ppc/kernel/head_8xx.S
arch/ppc/kernel/head_fsl_booke.S
arch/ppc/kernel/idle.c
arch/ppc/kernel/idle_6xx.S
arch/ppc/kernel/idle_power4.S
arch/ppc/kernel/misc.S
arch/ppc/kernel/smp.c
arch/ppc/kernel/swsusp.S
arch/ppc/kernel/traps.c
arch/ppc/lib/Makefile
arch/ppc/lib/dec_and_lock.c
arch/ppc/mm/fault.c
arch/ppc/mm/hashtable.S
arch/ppc/platforms/4xx/ebony.c
arch/ppc/platforms/hdpu.c
arch/ppc/platforms/pmac_sleep.S
arch/ppc/platforms/pmac_smp.c
arch/ppc/syslib/cpc700_pic.c
arch/ppc/syslib/i8259.c
arch/ppc/syslib/ibm440gx_common.c
arch/ppc/syslib/mv64x60.c
arch/ppc/syslib/open_pic2.c
arch/ppc/syslib/ppc403_pic.c
arch/ppc/syslib/qspan_pci.c
arch/ppc/syslib/xilinx_pic.c
arch/ppc64/Makefile
arch/ppc64/kernel/cpu_setup_power4.S
arch/ppc64/kernel/entry.S
arch/ppc64/kernel/head.S
arch/ppc64/kernel/iSeries_pci.c
arch/ppc64/kernel/idle_power4.S
arch/ppc64/kernel/maple_pci.c
arch/ppc64/kernel/misc.S
arch/ppc64/kernel/pSeries_setup.c
arch/ppc64/kernel/pSeries_smp.c
arch/ppc64/kernel/pci.c
arch/ppc64/kernel/pmac_pci.c
arch/ppc64/kernel/pmac_setup.c
arch/ppc64/kernel/pmc.c
arch/ppc64/kernel/process.c
arch/ppc64/kernel/ptrace.c
arch/ppc64/kernel/ptrace32.c
arch/ppc64/kernel/ras.c
arch/ppc64/kernel/setup.c
arch/ppc64/kernel/signal.c
arch/ppc64/kernel/signal32.c
arch/ppc64/kernel/vdso32/cacheflush.S
arch/ppc64/kernel/vdso32/datapage.S
arch/ppc64/kernel/vdso32/gettimeofday.S
arch/ppc64/kernel/vdso64/cacheflush.S
arch/ppc64/kernel/vdso64/datapage.S
arch/ppc64/kernel/vdso64/gettimeofday.S
arch/ppc64/kernel/xics.c
arch/ppc64/lib/dec_and_lock.c
arch/ppc64/lib/locks.c
arch/ppc64/mm/fault.c
arch/ppc64/mm/hash_low.S
arch/ppc64/mm/slb_low.S
arch/ppc64/xmon/privinst.h
arch/ppc64/xmon/xmon.c
arch/s390/Makefile
arch/s390/kernel/compat_linux.c
arch/s390/kernel/entry.S
arch/s390/kernel/entry64.S
arch/s390/kernel/head.S
arch/s390/kernel/head64.S
arch/s390/lib/spinlock.c
arch/s390/lib/uaccess.S
arch/s390/lib/uaccess64.S
arch/sh/Makefile
arch/sh/boards/adx/irq_maskreg.c
arch/sh/boards/bigsur/io.c
arch/sh/boards/bigsur/irq.c
arch/sh/boards/cqreek/irq.c
arch/sh/boards/harp/irq.c
arch/sh/boards/overdrive/irq.c
arch/sh/boards/renesas/hs7751rvoip/irq.c
arch/sh/boards/renesas/rts7751r2d/irq.c
arch/sh/boards/renesas/systemh/irq.c
arch/sh/boards/superh/microdev/irq.c
arch/sh/cchips/hd6446x/hd64465/io.c
arch/sh/cchips/voyagergx/irq.c
arch/sh/kernel/cpu/irq_imask.c
arch/sh/kernel/cpu/irq_ipr.c
arch/sh/kernel/cpu/sh4/irq_intc2.c
arch/sh64/Makefile
arch/sh64/kernel/irq_intc.c
arch/sparc/Makefile
arch/sparc/kernel/entry.S
arch/sparc/kernel/sclow.S
arch/sparc/kernel/sparc_ksyms.c
arch/sparc/lib/Makefile
arch/sparc/lib/atomic32.c
arch/sparc/lib/debuglocks.c [deleted file]
arch/sparc/mm/hypersparc.S
arch/sparc/mm/swift.S
arch/sparc/mm/tsunami.S
arch/sparc/mm/viking.S
arch/sparc64/kernel/asm-offsets.c [new file with mode: 0644]
arch/sparc64/kernel/process.c
arch/sparc64/kernel/sparc64_ksyms.c
arch/sparc64/lib/Makefile
arch/sparc64/lib/debuglocks.c [deleted file]
arch/sparc64/solaris/ioctl.c
arch/sparc64/solaris/timod.c
arch/um/Makefile
arch/um/include/mem.h
arch/um/kernel/asm-offsets.c [new file with mode: 0644]
arch/um/kernel/dyn.lds.S
arch/um/kernel/ksyms.c
arch/um/kernel/physmem.c
arch/um/kernel/trap_kern.c
arch/um/kernel/uml.lds.S
arch/v850/Makefile
arch/v850/kernel/asm-offsets.c [moved from arch/v850/kernel/asm-consts.c with 100% similarity]
arch/v850/kernel/entry.S
arch/v850/kernel/irq.c
arch/v850/kernel/setup.c
arch/v850/kernel/sim.c
arch/x86_64/Makefile
arch/x86_64/boot/Makefile
arch/x86_64/boot/compressed/misc.c
arch/x86_64/defconfig
arch/x86_64/ia32/ia32_ioctl.c
arch/x86_64/ia32/ia32entry.S
arch/x86_64/ia32/sys_ia32.c
arch/x86_64/ia32/vsyscall-syscall.S
arch/x86_64/ia32/vsyscall-sysenter.S
arch/x86_64/kernel/Makefile
arch/x86_64/kernel/acpi/sleep.c
arch/x86_64/kernel/aperture.c
arch/x86_64/kernel/apic.c
arch/x86_64/kernel/asm-offsets.c
arch/x86_64/kernel/crash.c
arch/x86_64/kernel/e820.c
arch/x86_64/kernel/early_printk.c
arch/x86_64/kernel/entry.S
arch/x86_64/kernel/genapic.c
arch/x86_64/kernel/genapic_cluster.c
arch/x86_64/kernel/genapic_flat.c
arch/x86_64/kernel/head.S
arch/x86_64/kernel/i8259.c
arch/x86_64/kernel/init_task.c
arch/x86_64/kernel/io_apic.c
arch/x86_64/kernel/irq.c
arch/x86_64/kernel/mce.c
arch/x86_64/kernel/mpparse.c
arch/x86_64/kernel/msr.c [deleted file]
arch/x86_64/kernel/nmi.c
arch/x86_64/kernel/pci-gart.c
arch/x86_64/kernel/process.c
arch/x86_64/kernel/setup.c
arch/x86_64/kernel/setup64.c
arch/x86_64/kernel/smp.c
arch/x86_64/kernel/smpboot.c
arch/x86_64/kernel/suspend.c
arch/x86_64/kernel/suspend_asm.S
arch/x86_64/kernel/time.c
arch/x86_64/kernel/traps.c
arch/x86_64/kernel/vmlinux.lds.S
arch/x86_64/kernel/vsyscall.c
arch/x86_64/lib/copy_user.S
arch/x86_64/lib/getuser.S
arch/x86_64/lib/putuser.S
arch/x86_64/mm/fault.c
arch/x86_64/mm/init.c
arch/x86_64/mm/k8topology.c
arch/x86_64/mm/numa.c
arch/x86_64/mm/srat.c
arch/x86_64/pci/k8-bus.c
arch/x86_64/pci/mmconfig.c
arch/xtensa/Makefile
arch/xtensa/kernel/align.S
arch/xtensa/kernel/entry.S
arch/xtensa/kernel/process.c
arch/xtensa/kernel/vectors.S
drivers/acorn/block/fd1772.c
drivers/acpi/sleep/main.c
drivers/acpi/sleep/poweroff.c
drivers/acpi/sleep/proc.c
drivers/atm/idt77105.c
drivers/atm/iphase.c
drivers/base/dmapool.c
drivers/block/acsi.c
drivers/block/acsi_slm.c
drivers/block/ataflop.c
drivers/block/cciss.c
drivers/block/cfq-iosched.c
drivers/block/deadline-iosched.c
drivers/block/floppy.c
drivers/block/paride/pcd.c
drivers/block/paride/pf.c
drivers/block/paride/pg.c
drivers/block/paride/pt.c
drivers/block/ps2esdi.c
drivers/block/swim3.c
drivers/block/swim_iop.c
drivers/block/umem.c
drivers/block/xd.c
drivers/block/z2ram.c
drivers/cdrom/aztcd.c
drivers/cdrom/gscd.c
drivers/cdrom/optcd.c
drivers/cdrom/sbpcd.c
drivers/cdrom/sjcd.c
drivers/cdrom/sonycd535.c
drivers/char/agp/backend.c
drivers/char/applicom.c
drivers/char/cyclades.c
drivers/char/ftape/lowlevel/fdc-io.c
drivers/char/hangcheck-timer.c
drivers/char/hpet.c
drivers/char/hw_random.c
drivers/char/ip2/i2lib.c
drivers/char/ip2main.c
drivers/char/ipmi/ipmi_devintf.c
drivers/char/ipmi/ipmi_si_intf.c
drivers/char/ipmi/ipmi_watchdog.c
drivers/char/istallion.c
drivers/char/keyboard.c
drivers/char/lcd.c
drivers/char/lp.c
drivers/char/mxser.c
drivers/char/n_tty.c
drivers/char/pcmcia/synclink_cs.c
drivers/char/pty.c
drivers/char/synclink.c
drivers/char/synclinkmp.c
drivers/char/tty_io.c
drivers/char/vt.c
drivers/char/watchdog/mixcomwd.c
drivers/ide/ide-io.c
drivers/ide/ide-tape.c
drivers/ide/ide-timing.h
drivers/ide/legacy/ide-cs.c
drivers/ide/pci/hpt366.c
drivers/ieee1394/video1394.c
drivers/infiniband/Kconfig
drivers/infiniband/core/Makefile
drivers/infiniband/core/cm.c
drivers/infiniband/core/mad_rmpp.c
drivers/infiniband/core/sa_query.c
drivers/infiniband/core/ucm.c
drivers/infiniband/core/ucm.h
drivers/infiniband/core/uverbs.h
drivers/infiniband/core/uverbs_cmd.c
drivers/infiniband/core/uverbs_main.c
drivers/infiniband/hw/mthca/mthca_qp.c
drivers/infiniband/ulp/ipoib/ipoib_main.c
drivers/input/evdev.c
drivers/input/joystick/iforce/iforce-packets.c
drivers/input/joystick/iforce/iforce-usb.c
drivers/input/keyboard/atkbd.c
drivers/input/keyboard/sunkbd.c
drivers/input/mouse/Makefile
drivers/input/mouse/alps.c
drivers/input/mouse/logips2pp.c
drivers/input/mouse/psmouse-base.c
drivers/input/mouse/psmouse.h
drivers/input/mouse/trackpoint.c [new file with mode: 0644]
drivers/input/mouse/trackpoint.h [new file with mode: 0644]
drivers/input/serio/i8042-io.h
drivers/input/serio/i8042-ip22io.h
drivers/input/serio/i8042-jazzio.h
drivers/input/serio/i8042-sparcio.h
drivers/input/serio/i8042-x86ia64io.h
drivers/input/serio/i8042.c
drivers/isdn/i4l/isdn_bsdcomp.c
drivers/isdn/i4l/isdn_common.c
drivers/md/bitmap.c
drivers/md/dm-exception-store.c
drivers/md/dm-raid1.c
drivers/md/linear.c
drivers/md/md.c
drivers/md/multipath.c
drivers/md/raid0.c
drivers/md/raid1.c
drivers/md/raid10.c
drivers/md/raid5.c
drivers/md/raid6main.c
drivers/media/Makefile
drivers/media/common/ir-common.c
drivers/media/common/saa7146_core.c
drivers/media/common/saa7146_fops.c
drivers/media/common/saa7146_i2c.c
drivers/media/dvb/b2c2/flexcop-fe-tuner.c
drivers/media/dvb/bt8xx/bt878.c
drivers/media/dvb/bt8xx/bt878.h
drivers/media/dvb/bt8xx/dst.c
drivers/media/dvb/bt8xx/dst_ca.c
drivers/media/dvb/bt8xx/dst_common.h
drivers/media/dvb/bt8xx/dvb-bt8xx.c
drivers/media/dvb/bt8xx/dvb-bt8xx.h
drivers/media/dvb/cinergyT2/Kconfig
drivers/media/dvb/cinergyT2/cinergyT2.c
drivers/media/dvb/dvb-core/demux.h
drivers/media/dvb/dvb-core/dmxdev.c
drivers/media/dvb/dvb-core/dvb_ca_en50221.c
drivers/media/dvb/dvb-core/dvb_demux.c
drivers/media/dvb/dvb-core/dvb_demux.h
drivers/media/dvb/dvb-core/dvb_net.c
drivers/media/dvb/dvb-usb/Kconfig
drivers/media/dvb/dvb-usb/Makefile
drivers/media/dvb/dvb-usb/a800.c
drivers/media/dvb/dvb-usb/cxusb.c
drivers/media/dvb/dvb-usb/cxusb.h
drivers/media/dvb/dvb-usb/dibusb-mb.c
drivers/media/dvb/dvb-usb/dibusb-mc.c
drivers/media/dvb/dvb-usb/digitv.c
drivers/media/dvb/dvb-usb/dtt200u-fe.c
drivers/media/dvb/dvb-usb/dtt200u.c
drivers/media/dvb/dvb-usb/dvb-usb-ids.h
drivers/media/dvb/dvb-usb/dvb-usb-init.c
drivers/media/dvb/dvb-usb/dvb-usb.h
drivers/media/dvb/dvb-usb/nova-t-usb2.c
drivers/media/dvb/dvb-usb/umt-010.c
drivers/media/dvb/dvb-usb/vp702x-fe.c [new file with mode: 0644]
drivers/media/dvb/dvb-usb/vp702x.c [new file with mode: 0644]
drivers/media/dvb/dvb-usb/vp702x.h [new file with mode: 0644]
drivers/media/dvb/dvb-usb/vp7045.c
drivers/media/dvb/frontends/cx24110.c
drivers/media/dvb/frontends/cx24110.h
drivers/media/dvb/frontends/dib3000mb.c
drivers/media/dvb/frontends/dib3000mc.c
drivers/media/dvb/frontends/mt352.c
drivers/media/dvb/frontends/nxt6000.c
drivers/media/dvb/frontends/or51132.c
drivers/media/dvb/frontends/s5h1420.c
drivers/media/dvb/frontends/s5h1420.h
drivers/media/dvb/frontends/stv0297.c
drivers/media/dvb/frontends/stv0297.h
drivers/media/dvb/frontends/stv0299.c
drivers/media/dvb/frontends/stv0299.h
drivers/media/dvb/frontends/tda1004x.c
drivers/media/dvb/frontends/ves1820.c
drivers/media/dvb/ttpci/av7110.c
drivers/media/dvb/ttpci/av7110.h
drivers/media/dvb/ttpci/av7110_hw.c
drivers/media/dvb/ttpci/av7110_ir.c
drivers/media/dvb/ttpci/av7110_v4l.c
drivers/media/dvb/ttpci/budget-av.c
drivers/media/dvb/ttpci/budget-ci.c
drivers/media/dvb/ttpci/budget-patch.c
drivers/media/dvb/ttpci/budget.c
drivers/media/dvb/ttusb-budget/dvb-ttusb-budget.c
drivers/media/dvb/ttusb-dec/ttusb_dec.c
drivers/media/video/Kconfig
drivers/media/video/Makefile
drivers/media/video/btcx-risc.c
drivers/media/video/btcx-risc.h
drivers/media/video/bttv-cards.c
drivers/media/video/bttv-driver.c
drivers/media/video/bttv-gpio.c
drivers/media/video/bttv-i2c.c
drivers/media/video/bttv-if.c
drivers/media/video/bttv-risc.c
drivers/media/video/bttv-vbi.c
drivers/media/video/bttv.h
drivers/media/video/bttvp.h
drivers/media/video/cpia_usb.c
drivers/media/video/cx88/cx88-blackbird.c
drivers/media/video/cx88/cx88-cards.c
drivers/media/video/cx88/cx88-core.c
drivers/media/video/cx88/cx88-dvb.c
drivers/media/video/cx88/cx88-i2c.c
drivers/media/video/cx88/cx88-input.c
drivers/media/video/cx88/cx88-mpeg.c
drivers/media/video/cx88/cx88-reg.h
drivers/media/video/cx88/cx88-tvaudio.c
drivers/media/video/cx88/cx88-vbi.c
drivers/media/video/cx88/cx88-video.c
drivers/media/video/cx88/cx88.h
drivers/media/video/ir-kbd-gpio.c
drivers/media/video/ir-kbd-i2c.c
drivers/media/video/msp3400.c
drivers/media/video/msp3400.h
drivers/media/video/mt20xx.c
drivers/media/video/rds.h [new file with mode: 0644]
drivers/media/video/saa6588.c [new file with mode: 0644]
drivers/media/video/saa7134/saa7134-cards.c
drivers/media/video/saa7134/saa7134-core.c
drivers/media/video/saa7134/saa7134-dvb.c
drivers/media/video/saa7134/saa7134-empress.c
drivers/media/video/saa7134/saa7134-i2c.c
drivers/media/video/saa7134/saa7134-input.c
drivers/media/video/saa7134/saa7134-oss.c
drivers/media/video/saa7134/saa7134-reg.h
drivers/media/video/saa7134/saa7134-ts.c
drivers/media/video/saa7134/saa7134-tvaudio.c
drivers/media/video/saa7134/saa7134-vbi.c
drivers/media/video/saa7134/saa7134-video.c
drivers/media/video/saa7134/saa7134.h
drivers/media/video/stradis.c
drivers/media/video/tda8290.c
drivers/media/video/tda9887.c
drivers/media/video/tea5767.c
drivers/media/video/tuner-core.c
drivers/media/video/tuner-simple.c
drivers/media/video/tvaudio.c
drivers/media/video/tveeprom.c
drivers/media/video/tvmixer.c
drivers/media/video/v4l1-compat.c
drivers/media/video/v4l2-common.c
drivers/media/video/video-buf-dvb.c
drivers/media/video/video-buf.c
drivers/media/video/zoran_driver.c
drivers/media/video/zr36120.c
drivers/mfd/Kconfig
drivers/mfd/Makefile
drivers/mfd/ucb1x00-assabet.c [new file with mode: 0644]
drivers/mfd/ucb1x00-core.c [new file with mode: 0644]
drivers/mfd/ucb1x00-ts.c [new file with mode: 0644]
drivers/mfd/ucb1x00.h [new file with mode: 0644]
drivers/misc/hdpuftrs/hdpu_cpustate.c
drivers/mmc/wbsd.c
drivers/mtd/devices/mtdram.c
drivers/mtd/ftl.c
drivers/mtd/maps/uclinux.c
drivers/net/Kconfig
drivers/net/atari_bionet.c
drivers/net/atari_pamsnet.c
drivers/net/bsd_comp.c
drivers/net/cris/eth_v10.c
drivers/net/cs89x0.c
drivers/net/fec.c
drivers/net/fec.h
drivers/net/hamradio/yam.c
drivers/net/mv643xx_eth.c
drivers/net/ppp_generic.c
drivers/net/sungem.c
drivers/net/sunhme.c
drivers/net/tulip/de4x5.c
drivers/parisc/iosapic.c
drivers/parisc/lasi.c
drivers/parport/ieee1284.c
drivers/parport/ieee1284_ops.c
drivers/parport/parport_pc.c
drivers/pci/hotplug.c
drivers/pci/hotplug/pciehprm_acpi.c
drivers/pci/pci-sysfs.c
drivers/pci/pci.h
drivers/pci/probe.c
drivers/pci/quirks.c
drivers/pci/setup-bus.c
drivers/pcmcia/Kconfig
drivers/pcmcia/Makefile
drivers/pcmcia/cs.c
drivers/pcmcia/cs_internal.h
drivers/pcmcia/ds.c
drivers/pcmcia/omap_cf.c [new file with mode: 0644]
drivers/pcmcia/pcmcia_resource.c
drivers/pcmcia/yenta_socket.c
drivers/sbus/char/aurora.c
drivers/sbus/char/bbc_envctrl.c
drivers/sbus/char/envctrl.c
drivers/scsi/53c7xx.c
drivers/scsi/ch.c
drivers/scsi/cpqfcTSinit.c
drivers/scsi/ibmmca.c
drivers/scsi/osst.c
drivers/scsi/pluto.c
drivers/scsi/qla2xxx/qla_dbg.c
drivers/serial/8250.c
drivers/serial/mcfserial.c
drivers/serial/serial_txx9.c
drivers/telephony/ixj.c
drivers/usb/host/hc_crisv10.c
drivers/usb/input/hid-core.c
drivers/usb/input/hid-debug.h
drivers/usb/input/hid-input.c
drivers/usb/input/hid.h
drivers/usb/input/hiddev.c
drivers/usb/media/stv680.c
drivers/video/Kconfig
drivers/video/Makefile
drivers/video/aty/aty128fb.c
drivers/video/aty/atyfb_base.c
drivers/video/aty/radeon_base.c
drivers/video/console/bitblit.c
drivers/video/console/fbcon.c
drivers/video/console/fbcon.h
drivers/video/console/vgacon.c
drivers/video/cyblafb.c [new file with mode: 0644]
drivers/video/fbcvt.c [new file with mode: 0644]
drivers/video/fbmem.c
drivers/video/fbmon.c
drivers/video/geode/Kconfig
drivers/video/geode/display_gx1.c
drivers/video/geode/geodefb.h
drivers/video/geode/gx1fb_core.c
drivers/video/geode/video_cs5530.c
drivers/video/i810/Makefile
drivers/video/i810/i810-i2c.c [new file with mode: 0644]
drivers/video/i810/i810.h
drivers/video/i810/i810_main.c
drivers/video/i810/i810_main.h
drivers/video/intelfb/intelfb.h
drivers/video/intelfb/intelfbdrv.c
drivers/video/intelfb/intelfbhw.c
drivers/video/matrox/matroxfb_misc.c
drivers/video/modedb.c
drivers/video/nvidia/nv_i2c.c
drivers/video/nvidia/nv_local.h
drivers/video/nvidia/nv_of.c
drivers/video/nvidia/nv_proto.h
drivers/video/nvidia/nv_setup.c
drivers/video/nvidia/nvidia.c
drivers/video/pxafb.c
drivers/video/pxafb.h
drivers/video/s3c2410fb.c [new file with mode: 0644]
drivers/video/s3c2410fb.h [new file with mode: 0644]
drivers/video/savage/savagefb-i2c.c
drivers/video/savage/savagefb.h
drivers/video/savage/savagefb_driver.c
drivers/video/sis/300vtbl.h
drivers/video/sis/310vtbl.h
drivers/video/sis/Makefile
drivers/video/sis/init.c
drivers/video/sis/init.h
drivers/video/sis/init301.c
drivers/video/sis/init301.h
drivers/video/sis/initdef.h
drivers/video/sis/initextlfb.c [new file with mode: 0644]
drivers/video/sis/oem300.h
drivers/video/sis/oem310.h
drivers/video/sis/osdef.h
drivers/video/sis/sis.h
drivers/video/sis/sis_accel.c
drivers/video/sis/sis_accel.h
drivers/video/sis/sis_main.c
drivers/video/sis/sis_main.h
drivers/video/sis/vgatypes.h
drivers/video/sis/vstruct.h
drivers/video/tridentfb.c
drivers/video/vesafb.c
drivers/video/vgastate.c
drivers/w1/w1_ds2433.c
fs/9p/9p.c [new file with mode: 0644]
fs/9p/9p.h [new file with mode: 0644]
fs/9p/Makefile [new file with mode: 0644]
fs/9p/conv.c [new file with mode: 0644]
fs/9p/conv.h [new file with mode: 0644]
fs/9p/debug.h [new file with mode: 0644]
fs/9p/error.c [new file with mode: 0644]
fs/9p/error.h [new file with mode: 0644]
fs/9p/fid.c [new file with mode: 0644]
fs/9p/fid.h [new file with mode: 0644]
fs/9p/mux.c [new file with mode: 0644]
fs/9p/mux.h [new file with mode: 0644]
fs/9p/trans_fd.c [new file with mode: 0644]
fs/9p/trans_sock.c [new file with mode: 0644]
fs/9p/transport.h [new file with mode: 0644]
fs/9p/v9fs.c [new file with mode: 0644]
fs/9p/v9fs.h [new file with mode: 0644]
fs/9p/v9fs_vfs.h [new file with mode: 0644]
fs/9p/vfs_dentry.c [new file with mode: 0644]
fs/9p/vfs_dir.c [new file with mode: 0644]
fs/9p/vfs_file.c [new file with mode: 0644]
fs/9p/vfs_inode.c [new file with mode: 0644]
fs/9p/vfs_super.c [new file with mode: 0644]
fs/Kconfig
fs/Makefile
fs/affs/inode.c
fs/aio.c
fs/autofs/autofs_i.h
fs/autofs/dirhash.c
fs/autofs/inode.c
fs/bfs/bfs.h
fs/bfs/dir.c
fs/bfs/file.c
fs/bfs/inode.c
fs/buffer.c
fs/cifs/connect.c
fs/compat.c
fs/compat_ioctl.c
fs/cramfs/uncompress.c
fs/dcache.c
fs/exec.c
fs/ext2/ialloc.c
fs/ext2/inode.c
fs/ext2/xattr.h
fs/ext2/xattr_security.c
fs/ext3/ialloc.c
fs/ext3/inode.c
fs/ext3/xattr.h
fs/ext3/xattr_security.c
fs/fat/inode.c
fs/fcntl.c
fs/file.c
fs/file_table.c
fs/fuse/Makefile [new file with mode: 0644]
fs/fuse/dev.c [new file with mode: 0644]
fs/fuse/dir.c [new file with mode: 0644]
fs/fuse/file.c [new file with mode: 0644]
fs/fuse/fuse_i.h [new file with mode: 0644]
fs/fuse/inode.c [new file with mode: 0644]
fs/hostfs/hostfs_kern.c
fs/hpfs/inode.c
fs/inode.c
fs/jbd/transaction.c
fs/jffs/inode-v23.c
fs/jffs/intrep.c
fs/jfs/acl.c
fs/jfs/inode.c
fs/jfs/jfs_acl.h
fs/jfs/jfs_xattr.h
fs/jfs/namei.c
fs/jfs/xattr.c
fs/lockd/clntproc.c
fs/locks.c
fs/minix/inode.c
fs/namei.c
fs/namespace.c
fs/ncpfs/inode.c
fs/nfs/inode.c
fs/nfs/nfs3proc.c
fs/nfs/nfs4proc.c
fs/ntfs/aops.c
fs/open.c
fs/pipe.c
fs/proc/array.c
fs/proc/base.c
fs/proc/inode.c
fs/qnx4/inode.c
fs/reiserfs/inode.c
fs/reiserfs/journal.c
fs/reiserfs/super.c
fs/select.c
fs/smbfs/inode.c
fs/smbfs/proc.c
fs/sysv/inode.c
fs/udf/inode.c
fs/ufs/inode.c
fs/xfs/linux-2.6/time.h
fs/xfs/linux-2.6/xfs_buf.c
fs/xfs/linux-2.6/xfs_super.c
fs/xfs/support/ktrace.c
include/asm-alpha/spinlock.h
include/asm-alpha/spinlock_types.h [new file with mode: 0644]
include/asm-arm/arch-pxa/pxafb.h
include/asm-arm/arch-s3c2410/fb.h [new file with mode: 0644]
include/asm-arm/arch-s3c2410/regs-lcd.h
include/asm-arm/spinlock.h
include/asm-arm/spinlock_types.h [new file with mode: 0644]
include/asm-arm/unistd.h
include/asm-arm26/hardirq.h
include/asm-generic/tlb.h
include/asm-generic/vmlinux.lds.h
include/asm-i386/apic.h
include/asm-i386/div64.h
include/asm-i386/mach-default/mach_reboot.h
include/asm-i386/mmzone.h
include/asm-i386/numa.h [new file with mode: 0644]
include/asm-i386/processor.h
include/asm-i386/spinlock.h
include/asm-i386/spinlock_types.h [new file with mode: 0644]
include/asm-i386/thread_info.h
include/asm-i386/topology.h
include/asm-i386/unistd.h
include/asm-ia64/mca.h
include/asm-ia64/mca_asm.h
include/asm-ia64/ptrace.h
include/asm-ia64/sn/sn_feature_sets.h [new file with mode: 0644]
include/asm-ia64/sn/sn_sal.h
include/asm-ia64/spinlock.h
include/asm-ia64/spinlock_types.h [new file with mode: 0644]
include/asm-ia64/system.h
include/asm-ia64/thread_info.h
include/asm-ia64/unwind.h
include/asm-m32r/spinlock.h
include/asm-m32r/spinlock_types.h [new file with mode: 0644]
include/asm-m68knommu/bitops.h
include/asm-m68knommu/checksum.h
include/asm-m68knommu/m527xsim.h
include/asm-m68knommu/m528xsim.h
include/asm-m68knommu/mcfcache.h
include/asm-m68knommu/mcfdma.h
include/asm-mips/asmmacro-32.h
include/asm-mips/asmmacro-64.h
include/asm-mips/irq.h
include/asm-mips/sim.h
include/asm-mips/spinlock.h
include/asm-mips/spinlock_types.h [new file with mode: 0644]
include/asm-mips/stackframe.h
include/asm-mips/vr41xx/tb0287.h [new file with mode: 0644]
include/asm-parisc/assembly.h
include/asm-parisc/atomic.h
include/asm-parisc/bitops.h
include/asm-parisc/cacheflush.h
include/asm-parisc/processor.h
include/asm-parisc/spinlock.h
include/asm-parisc/spinlock_types.h [new file with mode: 0644]
include/asm-parisc/system.h
include/asm-powerpc/siginfo.h
include/asm-ppc/irq.h
include/asm-ppc/ptrace.h
include/asm-ppc/reg.h
include/asm-ppc/smp.h
include/asm-ppc/spinlock.h
include/asm-ppc/spinlock_types.h [new file with mode: 0644]
include/asm-ppc/system.h
include/asm-ppc64/hvcall.h
include/asm-ppc64/machdep.h
include/asm-ppc64/pci-bridge.h
include/asm-ppc64/plpar_wrappers.h
include/asm-ppc64/processor.h
include/asm-ppc64/ptrace-common.h
include/asm-ppc64/ptrace.h
include/asm-ppc64/spinlock.h
include/asm-ppc64/spinlock_types.h [new file with mode: 0644]
include/asm-ppc64/system.h
include/asm-s390/spinlock.h
include/asm-s390/spinlock_types.h [new file with mode: 0644]
include/asm-sh/irq.h
include/asm-sh/spinlock.h
include/asm-sh/spinlock_types.h [new file with mode: 0644]
include/asm-sparc/ptrace.h
include/asm-sparc/spinlock.h
include/asm-sparc/spinlock_types.h [new file with mode: 0644]
include/asm-sparc64/spinlock.h
include/asm-sparc64/spinlock_types.h [new file with mode: 0644]
include/asm-um/page.h
include/asm-um/pgtable.h
include/asm-um/spinlock_types.h [new file with mode: 0644]
include/asm-x86_64/apic.h
include/asm-x86_64/apicdef.h
include/asm-x86_64/bug.h
include/asm-x86_64/calling.h
include/asm-x86_64/current.h
include/asm-x86_64/desc.h
include/asm-x86_64/dma-mapping.h
include/asm-x86_64/dwarf2.h
include/asm-x86_64/fixmap.h
include/asm-x86_64/hardirq.h
include/asm-x86_64/hw_irq.h
include/asm-x86_64/io.h
include/asm-x86_64/ipi.h
include/asm-x86_64/irq.h
include/asm-x86_64/kdebug.h
include/asm-x86_64/local.h
include/asm-x86_64/mmzone.h
include/asm-x86_64/msr.h
include/asm-x86_64/numa.h
include/asm-x86_64/page.h
include/asm-x86_64/pci.h
include/asm-x86_64/pda.h
include/asm-x86_64/pgalloc.h
include/asm-x86_64/pgtable.h
include/asm-x86_64/processor.h
include/asm-x86_64/proto.h
include/asm-x86_64/signal.h
include/asm-x86_64/smp.h
include/asm-x86_64/spinlock.h
include/asm-x86_64/spinlock_types.h [new file with mode: 0644]
include/asm-x86_64/system.h
include/asm-x86_64/timex.h
include/asm-x86_64/tlbflush.h
include/asm-x86_64/topology.h
include/asm-x86_64/vsyscall.h
include/asm-xtensa/ptrace.h
include/asm-xtensa/uaccess.h
include/linux/bfs_fs.h
include/linux/bio.h
include/linux/bit_spinlock.h [new file with mode: 0644]
include/linux/blkdev.h
include/linux/chio.h
include/linux/crc16.h
include/linux/dccp.h
include/linux/dmapool.h
include/linux/dmi.h
include/linux/fb.h
include/linux/file.h
include/linux/fs.h
include/linux/fuse.h [new file with mode: 0644]
include/linux/in6.h
include/linux/init_task.h
include/linux/input.h
include/linux/interrupt.h
include/linux/ipv6.h
include/linux/jbd.h
include/linux/jiffies.h
include/linux/pci.h
include/linux/radix-tree.h
include/linux/raid/bitmap.h
include/linux/raid/linear.h
include/linux/raid/md_k.h
include/linux/raid/md_p.h
include/linux/raid/raid1.h
include/linux/raid/raid5.h
include/linux/rcupdate.h
include/linux/rcuref.h [new file with mode: 0644]
include/linux/reiserfs_fs.h
include/linux/sched.h
include/linux/security.h
include/linux/slab.h
include/linux/spinlock.h
include/linux/spinlock_api_smp.h [new file with mode: 0644]
include/linux/spinlock_api_up.h [new file with mode: 0644]
include/linux/spinlock_types.h [new file with mode: 0644]
include/linux/spinlock_types_up.h [new file with mode: 0644]
include/linux/spinlock_up.h [new file with mode: 0644]
include/linux/time.h
include/linux/timer.h
include/linux/tty.h
include/linux/videodev.h
include/linux/videodev2.h
include/linux/writeback.h
include/media/audiochip.h
include/media/id.h
include/media/ir-common.h
include/media/saa7146.h
include/media/tuner.h
include/media/tveeprom.h
include/media/video-buf.h
include/pcmcia/ds.h
include/rdma/ib_cm.h
include/rdma/ib_mad.h
include/rdma/ib_sa.h
include/rdma/ib_user_cm.h
include/rdma/ib_user_verbs.h
include/sound/pcm.h
include/sound/tea575x-tuner.h
include/video/cyblafb.h [new file with mode: 0644]
include/video/sisfb.h
ipc/mqueue.c
kernel/Makefile
kernel/acct.c
kernel/compat.c
kernel/cpuset.c
kernel/exit.c
kernel/fork.c
kernel/rcupdate.c
kernel/sched.c
kernel/signal.c
kernel/softirq.c
kernel/spinlock.c
kernel/timer.c
lib/Kconfig.debug
lib/Makefile
lib/dec_and_lock.c
lib/kernel_lock.c
lib/radix-tree.c
lib/sort.c
lib/spinlock_debug.c [new file with mode: 0644]
mm/bootmem.c
mm/filemap.c
mm/memory.c
mm/nommu.c
mm/oom_kill.c
mm/page-writeback.c
mm/page_alloc.c
mm/shmem.c
mm/slab.c
mm/swap_state.c
mm/swapfile.c
mm/vmalloc.c
net/atm/mpc.c
net/core/dst.c
net/core/netpoll.c
net/core/pktgen.c
net/dccp/ccids/ccid3.c
net/dccp/ccids/ccid3.h
net/dccp/ccids/lib/packet_history.h
net/dccp/dccp.h
net/dccp/input.c
net/dccp/ipv4.c
net/dccp/minisocks.c
net/dccp/options.c
net/dccp/output.c
net/decnet/dn_route.c
net/ipv4/af_inet.c
net/ipv4/fib_trie.c
net/ipv4/inetpeer.c
net/ipv4/netfilter/ipt_owner.c
net/ipv4/tcp_output.c
net/ipv6/addrconf.c
net/ipv6/exthdrs.c
net/ipv6/ip6_fib.c
net/ipv6/ip6_flowlabel.c
net/ipv6/netfilter/ip6t_owner.c
net/ipv6/netfilter/ip6t_rt.c
net/netrom/nr_loopback.c
net/sched/sch_api.c
scripts/Kbuild.include
scripts/reference_discarded.pl
security/dummy.c
security/selinux/hooks.c
sound/isa/sb/sb16_csp.c
sound/oss/midibuf.c
sound/oss/skeleton.c [deleted file]
sound/oss/soundcard.c
sound/oss/sys_timer.c
sound/oss/uart6850.c

diff --git a/COPYING b/COPYING
index 2a7e338ec2fc6aac461a11fe8049799e65639166..ca442d313d86dc67e0a2e5d584b465bd382cbf5c 100644 (file)
--- a/COPYING
+++ b/COPYING
@@ -18,7 +18,7 @@
                       Version 2, June 1991
 
  Copyright (C) 1989, 1991 Free Software Foundation, Inc.
-                       59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+                       51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  Everyone is permitted to copy and distribute verbatim copies
  of this license document, but changing it is not allowed.
 
@@ -321,7 +321,7 @@ the "copyright" line and a pointer to where the full notice is found.
 
     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
+    Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 
 
 Also add information on how to contact you by electronic and paper mail.
index f28a24e0279b4c01aa8596e3f296ec50ef0ad70a..433cf5e9ae04bd17a04d50c320db9f100c68da63 100644 (file)
@@ -46,6 +46,8 @@ SubmittingPatches
        - procedure to get a source patch included into the kernel tree.
 VGA-softcursor.txt
        - how to change your VGA cursor from a blinking underscore.
+applying-patches.txt
+       - description of various trees and how to apply their patches.
 arm/
        - directory with info about Linux on the ARM architecture.
 basic_profiling.txt
@@ -275,7 +277,7 @@ tty.txt
 unicode.txt
        - info on the Unicode character/font mapping used in Linux.
 uml/
-       - directory with infomation about User Mode Linux.
+       - directory with information about User Mode Linux.
 usb/
        - directory with info regarding the Universal Serial Bus.
 video4linux/
index f25b3953f51398a023afd8b633aa5168313ccc4a..22e5f9036f3c193f88d14611b2e5fe669b2c9b9e 100644 (file)
@@ -236,6 +236,9 @@ ugly), but try to avoid excess.  Instead, put the comments at the head
 of the function, telling people what it does, and possibly WHY it does
 it.
 
+When commenting the kernel API functions, please use the kerneldoc format.
+See the files Documentation/kernel-doc-nano-HOWTO.txt and scripts/kernel-doc
+for details.
 
                Chapter 8: You've made a mess of it
 
index 6ee3cd6134dfad024bccff633ac2bece79fc0db6..1af0f2d5022066f1a205ad5265771e3ab112c018 100644 (file)
@@ -121,7 +121,7 @@ pool's device.
                        dma_addr_t addr);
 
 This puts memory back into the pool.  The pool is what was passed to
-the the pool allocation routine; the cpu and dma addresses are what
+the pool allocation routine; the cpu and dma addresses are what
 were returned when that routine allocated the memory being freed.
 
 
diff --git a/Documentation/DMA-ISA-LPC.txt b/Documentation/DMA-ISA-LPC.txt
new file mode 100644 (file)
index 0000000..705f6be
--- /dev/null
@@ -0,0 +1,151 @@
+                        DMA with ISA and LPC devices
+                        ============================
+
+                      Pierre Ossman <drzeus@drzeus.cx>
+
+This document describes how to do DMA transfers using the old ISA DMA
+controller. Even though ISA is more or less dead today the LPC bus
+uses the same DMA system so it will be around for quite some time.
+
+Part I - Headers and dependencies
+---------------------------------
+
+To do ISA style DMA you need to include two headers:
+
+#include <linux/dma-mapping.h>
+#include <asm/dma.h>
+
+The first is the generic DMA API used to convert virtual addresses to
+physical addresses (see Documentation/DMA-API.txt for details).
+
+The second contains the routines specific to ISA DMA transfers. Since
+this is not present on all platforms make sure you construct your
+Kconfig to be dependent on ISA_DMA_API (not ISA) so that nobody tries
+to build your driver on unsupported platforms.
+
+Part II - Buffer allocation
+---------------------------
+
+The ISA DMA controller has some very strict requirements on which
+memory it can access so extra care must be taken when allocating
+buffers.
+
+(You usually need a special buffer for DMA transfers instead of
+transferring directly to and from your normal data structures.)
+
+The DMA-able address space is the lowest 16 MB of _physical_ memory.
+Also the transfer block may not cross page boundaries (which are 64
+or 128 KiB depending on which channel you use).
+
+In order to allocate a piece of memory that satisfies all these
+requirements you pass the flag GFP_DMA to kmalloc.
+
+Unfortunately the memory available for ISA DMA is scarce so unless you
+allocate the memory during boot-up it's a good idea to also pass
+__GFP_REPEAT and __GFP_NOWARN to make the allocater try a bit harder.
+
+(This scarcity also means that you should allocate the buffer as
+early as possible and not release it until the driver is unloaded.)
+
+Part III - Address translation
+------------------------------
+
+To translate the virtual address to a physical use the normal DMA
+API. Do _not_ use isa_virt_to_phys() even though it does the same
+thing. The reason for this is that the function isa_virt_to_phys()
+will require a Kconfig dependency to ISA, not just ISA_DMA_API which
+is really all you need. Remember that even though the DMA controller
+has its origins in ISA it is used elsewhere.
+
+Note: x86_64 had a broken DMA API when it came to ISA but has since
+been fixed. If your arch has problems then fix the DMA API instead of
+reverting to the ISA functions.
+
+Part IV - Channels
+------------------
+
+A normal ISA DMA controller has 8 channels. The lower four are for
+8-bit transfers and the upper four are for 16-bit transfers.
+
+(Actually the DMA controller is really two separate controllers where
+channel 4 is used to give DMA access for the second controller (0-3).
+This means that of the four 16-bits channels only three are usable.)
+
+You allocate these in a similar fashion as all basic resources:
+
+extern int request_dma(unsigned int dmanr, const char * device_id);
+extern void free_dma(unsigned int dmanr);
+
+The ability to use 16-bit or 8-bit transfers is _not_ up to you as a
+driver author but depends on what the hardware supports. Check your
+specs or test different channels.
+
+Part V - Transfer data
+----------------------
+
+Now for the good stuff, the actual DMA transfer. :)
+
+Before you use any ISA DMA routines you need to claim the DMA lock
+using claim_dma_lock(). The reason is that some DMA operations are
+not atomic so only one driver may fiddle with the registers at a
+time.
+
+The first time you use the DMA controller you should call
+clear_dma_ff(). This clears an internal register in the DMA
+controller that is used for the non-atomic operations. As long as you
+(and everyone else) uses the locking functions then you only need to
+reset this once.
+
+Next, you tell the controller in which direction you intend to do the
+transfer using set_dma_mode(). Currently you have the options
+DMA_MODE_READ and DMA_MODE_WRITE.
+
+Set the address from where the transfer should start (this needs to
+be 16-bit aligned for 16-bit transfers) and how many bytes to
+transfer. Note that it's _bytes_. The DMA routines will do all the
+required translation to values that the DMA controller understands.
+
+The final step is enabling the DMA channel and releasing the DMA
+lock.
+
+Once the DMA transfer is finished (or timed out) you should disable
+the channel again. You should also check get_dma_residue() to make
+sure that all data has been transfered.
+
+Example:
+
+int flags, residue;
+
+flags = claim_dma_lock();
+
+clear_dma_ff();
+
+set_dma_mode(channel, DMA_MODE_WRITE);
+set_dma_addr(channel, phys_addr);
+set_dma_count(channel, num_bytes);
+
+dma_enable(channel);
+
+release_dma_lock(flags);
+
+while (!device_done());
+
+flags = claim_dma_lock();
+
+dma_disable(channel);
+
+residue = dma_get_residue(channel);
+if (residue != 0)
+       printk(KERN_ERR "driver: Incomplete DMA transfer!"
+               " %d bytes left!\n", residue);
+
+release_dma_lock(flags);
+
+Part VI - Suspend/resume
+------------------------
+
+It is the driver's responsibility to make sure that the machine isn't
+suspended while a DMA transfer is in progress. Also, all DMA settings
+are lost when the system suspends so if your driver relies on the DMA
+controller being in a certain state then you have to restore these
+registers upon resume.
index 1ef6f43c6d8f6a2d3f1be39ac6fcf4781ddb746d..341aaa4ce481a5a9af1f85d17be35e83e2b8b075 100644 (file)
@@ -116,7 +116,7 @@ filesystem. Almost.
 
 You still need to actually journal your filesystem changes, this
 is done by wrapping them into transactions. Additionally you
-also need to wrap the modification of each of the the buffers
+also need to wrap the modification of each of the buffers
 with calls to the journal layer, so it knows what the modifications
 you are actually making are. To do this use  journal_start() which
 returns a transaction handle.
@@ -128,7 +128,7 @@ and its counterpart journal_stop(), which indicates the end of a transaction
 are nestable calls, so you can reenter a transaction if necessary,
 but remember you must call journal_stop() the same number of times as
 journal_start() before the transaction is completed (or more accurately
-leaves the the update phase). Ext3/VFS makes use of this feature to simplify 
+leaves the update phase). Ext3/VFS makes use of this feature to simplify
 quota support.
 </para>
 
index 49a9ef82d575fe99c60fe1ad4b382cf221a2ce86..6367bba32d22256dae461b2649c3f51367af93bc 100644 (file)
@@ -8,8 +8,7 @@
   
   <authorgroup>
    <author>
-    <firstname>Paul</firstname>
-    <othername>Rusty</othername>
+    <firstname>Rusty</firstname>
     <surname>Russell</surname>
     <affiliation>
      <address>
@@ -20,7 +19,7 @@
   </authorgroup>
 
   <copyright>
-   <year>2001</year>
+   <year>2005</year>
    <holder>Rusty Russell</holder>
   </copyright>
 
@@ -64,7 +63,7 @@
  <chapter id="introduction">
   <title>Introduction</title>
   <para>
-   Welcome, gentle reader, to Rusty's Unreliable Guide to Linux
+   Welcome, gentle reader, to Rusty's Remarkably Unreliable Guide to Linux
    Kernel Hacking.  This document describes the common routines and
    general requirements for kernel code: its goal is to serve as a
    primer for Linux kernel development for experienced C
 
    <listitem>
     <para>
-     not associated with any process, serving a softirq, tasklet or bh;
+     not associated with any process, serving a softirq or tasklet;
     </para>
    </listitem>
 
    <listitem>
     <para>
-     running in kernel space, associated with a process;
+     running in kernel space, associated with a process (user context);
     </para>
    </listitem>
 
   </itemizedlist>
 
   <para>
-   There is a strict ordering between these: other than the last
-   category (userspace) each can only be pre-empted by those above.
-   For example, while a softirq is running on a CPU, no other
-   softirq will pre-empt it, but a hardware interrupt can.  However,
-   any other CPUs in the system execute independently.
+   There is an ordering between these.  The bottom two can preempt
+   each other, but above that is a strict hierarchy: each can only be
+   preempted by the ones above it.  For example, while a softirq is
+   running on a CPU, no other softirq will preempt it, but a hardware
+   interrupt can.  However, any other CPUs in the system execute
+   independently.
   </para>
 
   <para>
    <title>User Context</title>
 
    <para>
-    User context is when you are coming in from a system call or
-    other trap: you can sleep, and you own the CPU (except for
-    interrupts) until you call <function>schedule()</function>.  
-    In other words, user context (unlike userspace) is not pre-emptable.
+    User context is when you are coming in from a system call or other
+    trap: like userspace, you can be preempted by more important tasks
+    and by interrupts.  You can sleep, by calling
+    <function>schedule()</function>.
    </para>
 
    <note>
 
    <caution>
     <para>
-     Beware that if you have interrupts or bottom halves disabled 
+     Beware that if you have preemption or softirqs disabled
      (see below), <function>in_interrupt()</function> will return a 
      false positive.
     </para>
     <hardware>keyboard</hardware> are examples of real
     hardware which produce interrupts at any time.  The kernel runs
     interrupt handlers, which services the hardware.  The kernel
-    guarantees that this handler is never re-entered: if another
+    guarantees that this handler is never re-entered: if the same
     interrupt arrives, it is queued (or dropped).  Because it
     disables interrupts, this handler has to be fast: frequently it
-    simply acknowledges the interrupt, marks a `software interrupt'
+    simply acknowledges the interrupt, marks a 'software interrupt'
     for execution and exits.
    </para>
 
   </sect1>
 
   <sect1 id="basics-softirqs">
-   <title>Software Interrupt Context: Bottom Halves, Tasklets, softirqs</title>
+   <title>Software Interrupt Context: Softirqs and Tasklets</title>
 
    <para>
     Whenever a system call is about to return to userspace, or a
-    hardware interrupt handler exits, any `software interrupts'
+    hardware interrupt handler exits, any 'software interrupts'
     which are marked pending (usually by hardware interrupts) are
     run (<filename>kernel/softirq.c</filename>).
    </para>
 
    <para>
     Much of the real interrupt handling work is done here.  Early in
-    the transition to <acronym>SMP</acronym>, there were only `bottom 
+    the transition to <acronym>SMP</acronym>, there were only 'bottom
     halves' (BHs), which didn't take advantage of multiple CPUs.  Shortly 
     after we switched from wind-up computers made of match-sticks and snot,
-    we abandoned this limitation.
+    we abandoned this limitation and switched to 'softirqs'.
    </para>
 
    <para>
     <filename class="headerfile">include/linux/interrupt.h</filename> lists the
-    different BH's.  No matter how many CPUs you have, no two BHs will run at 
-    the same time. This made the transition to SMP simpler, but sucks hard for
-    scalable performance.  A very important bottom half is the timer
-    BH (<filename class="headerfile">include/linux/timer.h</filename>): you
-    can register to have it call functions for you in a given length of time.
+    different softirqs.  A very important softirq is the
+    timer softirq (<filename
+    class="headerfile">include/linux/timer.h</filename>): you can
+    register to have it call functions for you in a given length of
+    time.
    </para>
 
    <para>
-    2.3.43 introduced softirqs, and re-implemented the (now
-    deprecated) BHs underneath them.  Softirqs are fully-SMP
-    versions of BHs: they can run on as many CPUs at once as
-    required.  This means they need to deal with any races in shared
-    data using their own locks.  A bitmask is used to keep track of
-    which are enabled, so the 32 available softirqs should not be
-    used up lightly.  (<emphasis>Yes</emphasis>, people will
-    notice).
-   </para>
-
-   <para>
-    tasklets (<filename class="headerfile">include/linux/interrupt.h</filename>)
-    are like softirqs, except they are dynamically-registrable (meaning you 
-    can have as many as you want), and they also guarantee that any tasklet 
-    will only run on one CPU at any time, although different tasklets can 
-    run simultaneously (unlike different BHs).  
+    Softirqs are often a pain to deal with, since the same softirq
+    will run simultaneously on more than one CPU.  For this reason,
+    tasklets (<filename
+    class="headerfile">include/linux/interrupt.h</filename>) are more
+    often used: they are dynamically-registrable (meaning you can have
+    as many as you want), and they also guarantee that any tasklet
+    will only run on one CPU at any time, although different tasklets
+    can run simultaneously.
    </para>
    <caution>
     <para>
-     The name `tasklet' is misleading: they have nothing to do with `tasks', 
+     The name 'tasklet' is misleading: they have nothing to do with 'tasks',
      and probably more to do with some bad vodka Alexey Kuznetsov had at the 
      time.
     </para>
    </caution>
 
    <para>
-    You can tell you are in a softirq (or bottom half, or tasklet)
+    You can tell you are in a softirq (or tasklet)
     using the <function>in_softirq()</function> macro 
     (<filename class="headerfile">include/linux/interrupt.h</filename>).
    </para>
     <term>A rigid stack limit</term>
     <listitem>
      <para>
-      The kernel stack is about 6K in 2.2 (for most
-      architectures: it's about 14K on the Alpha), and shared
-      with interrupts so you can't use it all.  Avoid deep
-      recursion and huge local arrays on the stack (allocate
-      them dynamically instead).
+      Depending on configuration options the kernel stack is about 3K to 6K for most 32-bit architectures: it's
+      about 14K on most 64-bit archs, and often shared with interrupts
+      so you can't use it all.  Avoid deep recursion and huge local
+      arrays on the stack (allocate them dynamically instead).
      </para>
     </listitem>
    </varlistentry>
@@ -339,7 +330,7 @@ asmlinkage long sys_mycall(int arg)
 
   <para>
    If all your routine does is read or write some parameter, consider
-   implementing a <function>sysctl</function> interface instead.
+   implementing a <function>sysfs</function> interface instead.
   </para>
 
   <para>
@@ -417,7 +408,10 @@ cond_resched(); /* Will sleep */
   </para>
 
   <para>
-   You will eventually lock up your box if you break these rules.  
+   You should always compile your kernel
+   <symbol>CONFIG_DEBUG_SPINLOCK_SLEEP</symbol> on, and it will warn
+   you if you break these rules.  If you <emphasis>do</emphasis> break
+   the rules, you will eventually lock up your box.
   </para>
 
   <para>
@@ -515,8 +509,7 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
       success).
      </para>
     </caution>
-    [Yes, this moronic interface makes me cringe.  Please submit a
-    patch and become my hero --RR.]
+    [Yes, this moronic interface makes me cringe.  The flamewar comes up every year or so. --RR.]
    </para>
    <para>
     The functions may sleep implicitly. This should never be called
@@ -587,10 +580,11 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
    </variablelist>
 
    <para>
-    If you see a <errorname>kmem_grow: Called nonatomically from int
-    </errorname> warning message you called a memory allocation function
-    from interrupt context without <constant>GFP_ATOMIC</constant>.
-    You should really fix that.  Run, don't walk.
+    If you see a <errorname>sleeping function called from invalid
+    context</errorname> warning message, then maybe you called a
+    sleeping allocation function from interrupt context without
+    <constant>GFP_ATOMIC</constant>.  You should really fix that.
+    Run, don't walk.
    </para>
 
    <para>
@@ -639,16 +633,16 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
   </sect1>
 
   <sect1 id="routines-udelay">
-   <title><function>udelay()</function>/<function>mdelay()</function>
+   <title><function>mdelay()</function>/<function>udelay()</function>
      <filename class="headerfile">include/asm/delay.h</filename>
      <filename class="headerfile">include/linux/delay.h</filename>
    </title>
 
    <para>
-    The <function>udelay()</function> function can be used for small pauses.
-    Do not use large values with <function>udelay()</function> as you risk
+    The <function>udelay()</function> and <function>ndelay()</function> functions can be used for small pauses.
+    Do not use large values with them as you risk
     overflow - the helper function <function>mdelay()</function> is useful
-    here, or even consider <function>schedule_timeout()</function>.
+    here, or consider <function>msleep()</function>.
    </para> 
   </sect1>
  
@@ -698,8 +692,8 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
     These routines disable soft interrupts on the local CPU, and
     restore them.  They are reentrant; if soft interrupts were
     disabled before, they will still be disabled after this pair
-    of functions has been called.  They prevent softirqs, tasklets
-    and bottom halves from running on the current CPU.
+    of functions has been called.  They prevent softirqs and tasklets
+    from running on the current CPU.
    </para>
   </sect1>
 
@@ -708,10 +702,16 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
     <filename class="headerfile">include/asm/smp.h</filename></title>
    
    <para>
-    <function>smp_processor_id()</function> returns the current
-    processor number, between 0 and <symbol>NR_CPUS</symbol> (the
-    maximum number of CPUs supported by Linux, currently 32).  These
-    values are not necessarily continuous.
+    <function>get_cpu()</function> disables preemption (so you won't
+    suddenly get moved to another CPU) and returns the current
+    processor number, between 0 and <symbol>NR_CPUS</symbol>.  Note
+    that the CPU numbers are not necessarily continuous.  You return
+    it again with <function>put_cpu()</function> when you are done.
+   </para>
+   <para>
+    If you know you cannot be preempted by another task (ie. you are
+    in interrupt context, or have preemption disabled) you can use
+    smp_processor_id().
    </para>
   </sect1>
 
@@ -722,19 +722,14 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
    <para>
     After boot, the kernel frees up a special section; functions
     marked with <type>__init</type> and data structures marked with
-    <type>__initdata</type> are dropped after boot is complete (within
-    modules this directive is currently ignored).  <type>__exit</type>
+    <type>__initdata</type> are dropped after boot is complete: similarly
+    modules discard this memory after initialization.  <type>__exit</type>
     is used to declare a function which is only required on exit: the
     function will be dropped if this file is not compiled as a module.
     See the header file for use. Note that it makes no sense for a function
     marked with <type>__init</type> to be exported to modules with 
     <function>EXPORT_SYMBOL()</function> - this will break.
    </para>
-   <para>
-   Static data structures marked as <type>__initdata</type> must be initialised
-   (as opposed to ordinary static data which is zeroed BSS) and cannot be 
-   <type>const</type>.
-   </para> 
 
   </sect1>
 
@@ -762,9 +757,8 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
    <para>
     The function can return a negative error number to cause
     module loading to fail (unfortunately, this has no effect if
-    the module is compiled into the kernel).  For modules, this is
-    called in user context, with interrupts enabled, and the
-    kernel lock held, so it can sleep.
+    the module is compiled into the kernel).  This function is
+    called in user context with interrupts enabled, so it can sleep.
    </para>
   </sect1>
   
@@ -779,6 +773,34 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
     reached zero.  This function can also sleep, but cannot fail:
     everything must be cleaned up by the time it returns.
    </para>
+
+   <para>
+    Note that this macro is optional: if it is not present, your
+    module will not be removable (except for 'rmmod -f').
+   </para>
+  </sect1>
+
+  <sect1 id="routines-module-use-counters">
+   <title> <function>try_module_get()</function>/<function>module_put()</function>
+    <filename class="headerfile">include/linux/module.h</filename></title>
+
+   <para>
+    These manipulate the module usage count, to protect against
+    removal (a module also can't be removed if another module uses one
+    of its exported symbols: see below).  Before calling into module
+    code, you should call <function>try_module_get()</function> on
+    that module: if it fails, then the module is being removed and you
+    should act as if it wasn't there.  Otherwise, you can safely enter
+    the module, and call <function>module_put()</function> when you're
+    finished.
+   </para>
+
+   <para>
+   Most registerable structures have an
+   <structfield>owner</structfield> field, such as in the
+   <structname>file_operations</structname> structure. Set this field
+   to the macro <symbol>THIS_MODULE</symbol>.
+   </para>
   </sect1>
 
  <!-- add info on new-style module refcounting here -->
@@ -821,7 +843,7 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
     There is a macro to do this:
     <function>wait_event_interruptible()</function>
 
-    <filename class="headerfile">include/linux/sched.h</filename> The
+    <filename class="headerfile">include/linux/wait.h</filename> The
     first argument is the wait queue head, and the second is an
     expression which is evaluated; the macro returns
     <returnvalue>0</returnvalue> when this expression is true, or
@@ -847,10 +869,11 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
    <para>
     Call <function>wake_up()</function>
 
-    <filename class="headerfile">include/linux/sched.h</filename>;,
+    <filename class="headerfile">include/linux/wait.h</filename>;,
     which will wake up every process in the queue.  The exception is
     if one has <constant>TASK_EXCLUSIVE</constant> set, in which case
-    the remainder of the queue will not be woken.
+    the remainder of the queue will not be woken.  There are other variants
+    of this basic function available in the same header.
    </para>
   </sect1>
  </chapter>
@@ -863,7 +886,7 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
    first class of operations work on <type>atomic_t</type>
 
    <filename class="headerfile">include/asm/atomic.h</filename>; this
-   contains a signed integer (at least 24 bits long), and you must use
+   contains a signed integer (at least 32 bits long), and you must use
    these functions to manipulate or read atomic_t variables.
    <function>atomic_read()</function> and
    <function>atomic_set()</function> get and set the counter,
@@ -882,13 +905,12 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
 
   <para>
    Note that these functions are slower than normal arithmetic, and
-   so should not be used unnecessarily.  On some platforms they
-   are much slower, like 32-bit Sparc where they use a spinlock.
+   so should not be used unnecessarily.
   </para>
 
   <para>
-   The second class of atomic operations is atomic bit operations on a
-   <type>long</type>, defined in
+   The second class of atomic operations is atomic bit operations on an
+   <type>unsigned long</type>, defined in
 
    <filename class="headerfile">include/linux/bitops.h</filename>.  These
    operations generally take a pointer to the bit pattern, and a bit
@@ -899,7 +921,7 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
    <function>test_and_clear_bit()</function> and
    <function>test_and_change_bit()</function> do the same thing,
    except return true if the bit was previously set; these are
-   particularly useful for very simple locking.
+   particularly useful for atomically setting flags.
   </para>
   
   <para>
@@ -907,12 +929,6 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
    than BITS_PER_LONG.  The resulting behavior is strange on big-endian
    platforms though so it is a good idea not to do this.
   </para>
-
-  <para>
-   Note that the order of bits depends on the architecture, and in
-   particular, the bitfield passed to these operations must be at
-   least as large as a <type>long</type>.
-  </para>
  </chapter>
 
  <chapter id="symbols">
@@ -932,11 +948,8 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
     <filename class="headerfile">include/linux/module.h</filename></title>
 
    <para>
-    This is the classic method of exporting a symbol, and it works
-    for both modules and non-modules.  In the kernel all these
-    declarations are often bundled into a single file to help
-    genksyms (which searches source files for these declarations).
-    See the comment on genksyms and Makefiles below.
+    This is the classic method of exporting a symbol: dynamically
+    loaded modules will be able to use the symbol as normal.
    </para>
   </sect1>
 
@@ -949,7 +962,8 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
     symbols exported by <function>EXPORT_SYMBOL_GPL()</function> can
     only be seen by modules with a
     <function>MODULE_LICENSE()</function> that specifies a GPL
-    compatible license.
+    compatible license.  It implies that the function is considered
+    an internal implementation issue, and not really an interface.
    </para>
   </sect1>
  </chapter>
@@ -962,12 +976,13 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
     <filename class="headerfile">include/linux/list.h</filename></title>
 
    <para>
-    There are three sets of linked-list routines in the kernel
-    headers, but this one seems to be winning out (and Linus has
-    used it).  If you don't have some particular pressing need for
-    a single list, it's a good choice.  In fact, I don't care
-    whether it's a good choice or not, just use it so we can get
-    rid of the others.
+    There used to be three sets of linked-list routines in the kernel
+    headers, but this one is the winner.  If you don't have some
+    particular pressing need for a single list, it's a good choice.
+   </para>
+
+   <para>
+    In particular, <function>list_for_each_entry</function> is useful.
    </para>
   </sect1>
 
@@ -979,14 +994,13 @@ printk(KERN_INFO "my ip: %d.%d.%d.%d\n", NIPQUAD(ipaddress));
     convention, and return <returnvalue>0</returnvalue> for success,
     and a negative error number
     (eg. <returnvalue>-EFAULT</returnvalue>) for failure.  This can be
-    unintuitive at first, but it's fairly widespread in the networking
-    code, for example.
+    unintuitive at first, but it's fairly widespread in the kernel.
    </para>
 
    <para>
-    The filesystem code uses <function>ERR_PTR()</function>
+    Using <function>ERR_PTR()</function>
 
-    <filename class="headerfile">include/linux/fs.h</filename>; to
+    <filename class="headerfile">include/linux/err.h</filename>; to
     encode a negative error number into a pointer, and
     <function>IS_ERR()</function> and <function>PTR_ERR()</function>
     to get it back out again: avoids a separate pointer parameter for
@@ -1040,7 +1054,7 @@ static struct block_device_operations opt_fops = {
     supported, due to lack of general use, but the following are
     considered standard (see the GCC info page section "C
     Extensions" for more details - Yes, really the info page, the
-    man page is only a short summary of the stuff in info):
+    man page is only a short summary of the stuff in info).
    </para>
    <itemizedlist>
     <listitem>
@@ -1091,7 +1105,7 @@ static struct block_device_operations opt_fops = {
     </listitem>
     <listitem>
      <para>
-      Function names as strings (__FUNCTION__)
+      Function names as strings (__func__).
      </para>
     </listitem>
     <listitem>
@@ -1164,63 +1178,35 @@ static struct block_device_operations opt_fops = {
    <listitem>
     <para>
      Usually you want a configuration option for your kernel hack.
-     Edit <filename>Config.in</filename> in the appropriate directory
-     (but under <filename>arch/</filename> it's called
-     <filename>config.in</filename>).  The Config Language used is not
-     bash, even though it looks like bash; the safe way is to use only
-     the constructs that you already see in
-     <filename>Config.in</filename> files (see
-     <filename>Documentation/kbuild/kconfig-language.txt</filename>).
-     It's good to run "make xconfig" at least once to test (because
-     it's the only one with a static parser).
-    </para>
-
-    <para>
-     Variables which can be Y or N use <type>bool</type> followed by a
-     tagline and the config define name (which must start with
-     CONFIG_).  The <type>tristate</type> function is the same, but
-     allows the answer M (which defines
-     <symbol>CONFIG_foo_MODULE</symbol> in your source, instead of
-     <symbol>CONFIG_FOO</symbol>) if <symbol>CONFIG_MODULES</symbol>
-     is enabled.
+     Edit <filename>Kconfig</filename> in the appropriate directory.
+     The Config language is simple to use by cut and paste, and there's
+     complete documentation in
+     <filename>Documentation/kbuild/kconfig-language.txt</filename>.
     </para>
 
     <para>
      You may well want to make your CONFIG option only visible if
      <symbol>CONFIG_EXPERIMENTAL</symbol> is enabled: this serves as a
      warning to users.  There many other fancy things you can do: see
-     the various <filename>Config.in</filename> files for ideas.
+     the various <filename>Kconfig</filename> files for ideas.
     </para>
-   </listitem>
 
-   <listitem>
     <para>
-     Edit the <filename>Makefile</filename>: the CONFIG variables are
-     exported here so you can conditionalize compilation with `ifeq'.
-     If your file exports symbols then add the names to
-     <varname>export-objs</varname> so that genksyms will find them.
-     <caution>
-      <para>
-       There is a restriction on the kernel build system that objects
-       which export symbols must have globally unique names.
-       If your object does not have a globally unique name then the
-       standard fix is to move the
-       <function>EXPORT_SYMBOL()</function> statements to their own
-       object with a unique name.
-       This is why several systems have separate exporting objects,
-       usually suffixed with ksyms.
-      </para>
-     </caution>
+     In your description of the option, make sure you address both the
+     expert user and the user who knows nothing about your feature.  Mention
+     incompatibilities and issues here.  <emphasis> Definitely
+     </emphasis> end your description with <quote> if in doubt, say N
+     </quote> (or, occasionally, `Y'); this is for people who have no
+     idea what you are talking about.
     </para>
    </listitem>
 
    <listitem>
     <para>
-     Document your option in Documentation/Configure.help.  Mention
-     incompatibilities and issues here.  <emphasis> Definitely
-     </emphasis> end your description with <quote> if in doubt, say N
-     </quote> (or, occasionally, `Y'); this is for people who have no
-     idea what you are talking about.
+     Edit the <filename>Makefile</filename>: the CONFIG variables are
+     exported here so you can usually just add a "obj-$(CONFIG_xxx) +=
+     xxx.o" line.  The syntax is documented in
+     <filename>Documentation/kbuild/makefiles.txt</filename>.
     </para>
    </listitem>
 
@@ -1253,20 +1239,12 @@ static struct block_device_operations opt_fops = {
   </para>
 
   <para>
-   <filename>include/linux/brlock.h:</filename>
+   <filename>include/asm-i386/delay.h:</filename>
   </para>
   <programlisting>
-extern inline void br_read_lock (enum brlock_indices idx)
-{
-        /*
-         * This causes a link-time bug message if an
-         * invalid index is used:
-         */
-        if (idx >= __BR_END)
-                __br_lock_usage_bug();
-
-        read_lock(&amp;__brlock_array[smp_processor_id()][idx]);
-}
+#define ndelay(n) (__builtin_constant_p(n) ? \
+        ((n) > 20000 ? __bad_ndelay() : __const_udelay((n) * 5ul)) : \
+        __ndelay(n))
   </programlisting>
 
   <para>
index f3ef0bf435e9ebde6ba5174f43ef7168059058fa..705c442c7bf4cfcd687ea84aa92e6713463f6fcb 100644 (file)
@@ -841,7 +841,7 @@ usbdev_ioctl (int fd, int ifno, unsigned request, void *param)
                    File modification time is not updated by this request.
                    </para><para>
                    Those struct members are from some interface descriptor
-                   applying to the the current configuration.
+                   applying to the current configuration.
                    The interface number is the bInterfaceNumber value, and
                    the altsetting number is the bAlternateSetting value.
                    (This resets each endpoint in the interface.)
index d5032eb480aa364593d73bd6f211ec23bc2b990e..63edc5f847c45a4b1771c4e35872317ca2bbcbc7 100644 (file)
@@ -430,7 +430,7 @@ which may result in system hang. The software driver of specific
 MSI-capable hardware is responsible for whether calling
 pci_enable_msi or not. A return of zero indicates the kernel
 successfully initializes the MSI/MSI-X capability structure of the
-device funtion. The device function is now running on MSI/MSI-X mode.
+device function. The device function is now running on MSI/MSI-X mode.
 
 5.6 How to tell whether MSI/MSI-X is enabled on device function
 
index 9c6d450138ead56f9f1ef4c2e55794b2b30ff77c..fcbcbc35b122fda0223e22c267f7bca240e48b5b 100644 (file)
@@ -2,7 +2,8 @@ Read the F-ing Papers!
 
 
 This document describes RCU-related publications, and is followed by
-the corresponding bibtex entries.
+the corresponding bibtex entries.  A number of the publications may
+be found at http://www.rdrop.com/users/paulmck/RCU/.
 
 The first thing resembling RCU was published in 1980, when Kung and Lehman
 [Kung80] recommended use of a garbage collector to defer destruction
@@ -113,6 +114,10 @@ describing how to make RCU safe for soft-realtime applications [Sarma04c],
 and a paper describing SELinux performance with RCU [JamesMorris04b].
 
 
+2005 has seen further adaptation of RCU to realtime use, permitting
+preemption of RCU realtime critical sections [PaulMcKenney05a,
+PaulMcKenney05b].
+
 Bibtex Entries
 
 @article{Kung80
@@ -410,3 +415,32 @@ Oregon Health and Sciences University"
 \url{http://www.livejournal.com/users/james_morris/2153.html}
 [Viewed December 10, 2004]"
 }
+
+@unpublished{PaulMcKenney05a
+,Author="Paul E. McKenney"
+,Title="{[RFC]} {RCU} and {CONFIG\_PREEMPT\_RT} progress"
+,month="May"
+,year="2005"
+,note="Available:
+\url{http://lkml.org/lkml/2005/5/9/185}
+[Viewed May 13, 2005]"
+,annotation="
+       First publication of working lock-based deferred free patches
+       for the CONFIG_PREEMPT_RT environment.
+"
+}
+
+@conference{PaulMcKenney05b
+,Author="Paul E. McKenney and Dipankar Sarma"
+,Title="Towards Hard Realtime Response from the Linux Kernel on SMP Hardware"
+,Booktitle="linux.conf.au 2005"
+,month="April"
+,year="2005"
+,address="Canberra, Australia"
+,note="Available:
+\url{http://www.rdrop.com/users/paulmck/RCU/realtimeRCU.2005.04.23a.pdf}
+[Viewed May 13, 2005]"
+,annotation="
+       Realtime turns into making RCU yet more realtime friendly.
+"
+}
index 3bfb84b3b7dbc5120bbcbf14c52ed95fb2a72507..aab4a9ec3931520554a76b2e33e5096b5a0ca632 100644 (file)
@@ -8,7 +8,7 @@ is that since there is only one CPU, it should not be necessary to
 wait for anything else to get done, since there are no other CPUs for
 anything else to be happening on.  Although this approach will -sort- -of-
 work a surprising amount of the time, it is a very bad idea in general.
-This document presents two examples that demonstrate exactly how bad an
+This document presents three examples that demonstrate exactly how bad an
 idea this is.
 
 
@@ -26,6 +26,9 @@ from softirq, the list scan would find itself referencing a newly freed
 element B.  This situation can greatly decrease the life expectancy of
 your kernel.
 
+This same problem can occur if call_rcu() is invoked from a hardware
+interrupt handler.
+
 
 Example 2: Function-Call Fatality
 
@@ -44,8 +47,37 @@ its arguments would cause it to fail to make the fundamental guarantee
 underlying RCU, namely that call_rcu() defers invoking its arguments until
 all RCU read-side critical sections currently executing have completed.
 
-Quick Quiz: why is it -not- legal to invoke synchronize_rcu() in
-this case?
+Quick Quiz #1: why is it -not- legal to invoke synchronize_rcu() in
+       this case?
+
+
+Example 3: Death by Deadlock
+
+Suppose that call_rcu() is invoked while holding a lock, and that the
+callback function must acquire this same lock.  In this case, if
+call_rcu() were to directly invoke the callback, the result would
+be self-deadlock.
+
+In some cases, it would possible to restructure to code so that
+the call_rcu() is delayed until after the lock is released.  However,
+there are cases where this can be quite ugly:
+
+1.     If a number of items need to be passed to call_rcu() within
+       the same critical section, then the code would need to create
+       a list of them, then traverse the list once the lock was
+       released.
+
+2.     In some cases, the lock will be held across some kernel API,
+       so that delaying the call_rcu() until the lock is released
+       requires that the data item be passed up via a common API.
+       It is far better to guarantee that callbacks are invoked
+       with no locks held than to have to modify such APIs to allow
+       arbitrary data items to be passed back up through them.
+
+If call_rcu() directly invokes the callback, painful locking restrictions
+or API changes would be required.
+
+Quick Quiz #2: What locking restriction must RCU callbacks respect?
 
 
 Summary
@@ -53,12 +85,35 @@ Summary
 Permitting call_rcu() to immediately invoke its arguments or permitting
 synchronize_rcu() to immediately return breaks RCU, even on a UP system.
 So do not do it!  Even on a UP system, the RCU infrastructure -must-
-respect grace periods.
-
-
-Answer to Quick Quiz
-
-The calling function is scanning an RCU-protected linked list, and
-is therefore within an RCU read-side critical section.  Therefore,
-the called function has been invoked within an RCU read-side critical
-section, and is not permitted to block.
+respect grace periods, and -must- invoke callbacks from a known environment
+in which no locks are held.
+
+
+Answer to Quick Quiz #1:
+       Why is it -not- legal to invoke synchronize_rcu() in this case?
+
+       Because the calling function is scanning an RCU-protected linked
+       list, and is therefore within an RCU read-side critical section.
+       Therefore, the called function has been invoked within an RCU
+       read-side critical section, and is not permitted to block.
+
+Answer to Quick Quiz #2:
+       What locking restriction must RCU callbacks respect?
+
+       Any lock that is acquired within an RCU callback must be
+       acquired elsewhere using an _irq variant of the spinlock
+       primitive.  For example, if "mylock" is acquired by an
+       RCU callback, then a process-context acquisition of this
+       lock must use something like spin_lock_irqsave() to
+       acquire the lock.
+
+       If the process-context code were to simply use spin_lock(),
+       then, since RCU callbacks can be invoked from softirq context,
+       the callback might be called from a softirq that interrupted
+       the process-context critical section.  This would result in
+       self-deadlock.
+
+       This restriction might seem gratuitous, since very few RCU
+       callbacks acquire locks directly.  However, a great many RCU
+       callbacks do acquire locks -indirectly-, for example, via
+       the kfree() primitive.
index 8f3fb77c9cd32f2732351c2a78701110d2fc2454..e118a7c1a0928d9aadc5f4cd34d4e110a930a1ee 100644 (file)
@@ -43,6 +43,10 @@ over a rather long period of time, but improvements are always welcome!
        rcu_read_lock_bh()) in the read-side critical sections,
        and are also an excellent aid to readability.
 
+       As a rough rule of thumb, any dereference of an RCU-protected
+       pointer must be covered by rcu_read_lock() or rcu_read_lock_bh()
+       or by the appropriate update-side lock.
+
 3.     Does the update code tolerate concurrent accesses?
 
        The whole point of RCU is to permit readers to run without
@@ -90,7 +94,11 @@ over a rather long period of time, but improvements are always welcome!
 
                The rcu_dereference() primitive is used by the various
                "_rcu()" list-traversal primitives, such as the
-               list_for_each_entry_rcu().
+               list_for_each_entry_rcu().  Note that it is perfectly
+               legal (if redundant) for update-side code to use
+               rcu_dereference() and the "_rcu()" list-traversal
+               primitives.  This is particularly useful in code
+               that is common to readers and updaters.
 
        b.      If the list macros are being used, the list_add_tail_rcu()
                and list_add_rcu() primitives must be used in order
@@ -150,16 +158,9 @@ over a rather long period of time, but improvements are always welcome!
 
        Use of the _rcu() list-traversal primitives outside of an
        RCU read-side critical section causes no harm other than
-       a slight performance degradation on Alpha CPUs and some
-       confusion on the part of people trying to read the code.
-
-       Another way of thinking of this is "If you are holding the
-       lock that prevents the data structure from changing, why do
-       you also need RCU-based protection?"  That said, there may
-       well be situations where use of the _rcu() list-traversal
-       primitives while the update-side lock is held results in
-       simpler and more maintainable code.  The jury is still out
-       on this question.
+       a slight performance degradation on Alpha CPUs.  It can
+       also be quite helpful in reducing code bloat when common
+       code is shared between readers and updaters.
 
 10.    Conversely, if you are in an RCU read-side critical section,
        you -must- use the "_rcu()" variants of the list macros.
index eb444006683e28140cc0c1eb6d02f74f6f344503..6fa092251586e14930c6d51d8bde1d4566512138 100644 (file)
@@ -64,6 +64,54 @@ o    I hear that RCU is patented?  What is with that?
        Of these, one was allowed to lapse by the assignee, and the
        others have been contributed to the Linux kernel under GPL.
 
+o      I hear that RCU needs work in order to support realtime kernels?
+
+       Yes, work in progress.
+
 o      Where can I find more information on RCU?
 
        See the RTFP.txt file in this directory.
+       Or point your browser at http://www.rdrop.com/users/paulmck/RCU/.
+
+o      What are all these files in this directory?
+
+
+       NMI-RCU.txt
+
+               Describes how to use RCU to implement dynamic
+               NMI handlers, which can be revectored on the fly,
+               without rebooting.
+
+       RTFP.txt
+
+               List of RCU-related publications and web sites.
+
+       UP.txt
+
+               Discussion of RCU usage in UP kernels.
+
+       arrayRCU.txt
+
+               Describes how to use RCU to protect arrays, with
+               resizeable arrays whose elements reference other
+               data structures being of the most interest.
+
+       checklist.txt
+
+               Lists things to check for when inspecting code that
+               uses RCU.
+
+       listRCU.txt
+
+               Describes how to use RCU to protect linked lists.
+               This is the simplest and most common use of RCU
+               in the Linux kernel.
+
+       rcu.txt
+
+               You are reading it!
+
+       whatisRCU.txt
+
+               Overview of how the RCU implementation works.  Along
+               the way, presents a conceptual view of RCU.
diff --git a/Documentation/RCU/rcuref.txt b/Documentation/RCU/rcuref.txt
new file mode 100644 (file)
index 0000000..a23fee6
--- /dev/null
@@ -0,0 +1,74 @@
+Refcounter framework for elements of lists/arrays protected by
+RCU.
+
+Refcounting on elements of  lists which are protected by traditional
+reader/writer spinlocks or semaphores are straight forward as in:
+
+1.                                     2.
+add()                                  search_and_reference()
+{                                      {
+       alloc_object                            read_lock(&list_lock);
+       ...                                     search_for_element
+       atomic_set(&el->rc, 1);                 atomic_inc(&el->rc);
+       write_lock(&list_lock);                 ...
+       add_element                             read_unlock(&list_lock);
+       ...                                     ...
+       write_unlock(&list_lock);       }
+}
+
+3.                                     4.
+release_referenced()                   delete()
+{                                      {
+       ...                             write_lock(&list_lock);
+       atomic_dec(&el->rc, relfunc)    ...
+       ...                             delete_element
+}                                      write_unlock(&list_lock);
+                                       ...
+                                       if (atomic_dec_and_test(&el->rc))
+                                               kfree(el);
+                                       ...
+                                       }
+
+If this list/array is made lock free using rcu as in changing the
+write_lock in add() and delete() to spin_lock and changing read_lock
+in search_and_reference to rcu_read_lock(), the rcuref_get in
+search_and_reference could potentially hold reference to an element which
+has already been deleted from the list/array.  rcuref_lf_get_rcu takes
+care of this scenario. search_and_reference should look as;
+
+1.                                     2.
+add()                                  search_and_reference()
+{                                      {
+       alloc_object                            rcu_read_lock();
+       ...                                     search_for_element
+       atomic_set(&el->rc, 1);                 if (rcuref_inc_lf(&el->rc)) {
+       write_lock(&list_lock);                         rcu_read_unlock();
+                                                       return FAIL;
+       add_element                             }
+       ...                                     ...
+       write_unlock(&list_lock);               rcu_read_unlock();
+}                                      }
+3.                                     4.
+release_referenced()                   delete()
+{                                      {
+       ...                             write_lock(&list_lock);
+       rcuref_dec(&el->rc, relfunc)    ...
+       ...                             delete_element
+}                                      write_unlock(&list_lock);
+                                       ...
+                                       if (rcuref_dec_and_test(&el->rc))
+                                               call_rcu(&el->head, el_free);
+                                       ...
+                                       }
+
+Sometimes, reference to the element need to be obtained in the
+update (write) stream.  In such cases, rcuref_inc_lf might be an overkill
+since the spinlock serialising list updates are held. rcuref_inc
+is to be used in such cases.
+For arches which do not have cmpxchg rcuref_inc_lf
+api uses a hashed spinlock implementation and the same hashed spinlock
+is acquired in all rcuref_xxx primitives to preserve atomicity.
+Note: Use rcuref_inc api only if you need to use rcuref_inc_lf on the
+refcounter atleast at one place.  Mixing rcuref_inc and atomic_xxx api
+might lead to races. rcuref_inc_lf() must be used in lockfree
+RCU critical sections only.
diff --git a/Documentation/RCU/whatisRCU.txt b/Documentation/RCU/whatisRCU.txt
new file mode 100644 (file)
index 0000000..354d89c
--- /dev/null
@@ -0,0 +1,902 @@
+What is RCU?
+
+RCU is a synchronization mechanism that was added to the Linux kernel
+during the 2.5 development effort that is optimized for read-mostly
+situations.  Although RCU is actually quite simple once you understand it,
+getting there can sometimes be a challenge.  Part of the problem is that
+most of the past descriptions of RCU have been written with the mistaken
+assumption that there is "one true way" to describe RCU.  Instead,
+the experience has been that different people must take different paths
+to arrive at an understanding of RCU.  This document provides several
+different paths, as follows:
+
+1.     RCU OVERVIEW
+2.     WHAT IS RCU'S CORE API?
+3.     WHAT ARE SOME EXAMPLE USES OF CORE RCU API?
+4.     WHAT IF MY UPDATING THREAD CANNOT BLOCK?
+5.     WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RCU?
+6.     ANALOGY WITH READER-WRITER LOCKING
+7.     FULL LIST OF RCU APIs
+8.     ANSWERS TO QUICK QUIZZES
+
+People who prefer starting with a conceptual overview should focus on
+Section 1, though most readers will profit by reading this section at
+some point.  People who prefer to start with an API that they can then
+experiment with should focus on Section 2.  People who prefer to start
+with example uses should focus on Sections 3 and 4.  People who need to
+understand the RCU implementation should focus on Section 5, then dive
+into the kernel source code.  People who reason best by analogy should
+focus on Section 6.  Section 7 serves as an index to the docbook API
+documentation, and Section 8 is the traditional answer key.
+
+So, start with the section that makes the most sense to you and your
+preferred method of learning.  If you need to know everything about
+everything, feel free to read the whole thing -- but if you are really
+that type of person, you have perused the source code and will therefore
+never need this document anyway.  ;-)
+
+
+1.  RCU OVERVIEW
+
+The basic idea behind RCU is to split updates into "removal" and
+"reclamation" phases.  The removal phase removes references to data items
+within a data structure (possibly by replacing them with references to
+new versions of these data items), and can run concurrently with readers.
+The reason that it is safe to run the removal phase concurrently with
+readers is the semantics of modern CPUs guarantee that readers will see
+either the old or the new version of the data structure rather than a
+partially updated reference.  The reclamation phase does the work of reclaiming
+(e.g., freeing) the data items removed from the data structure during the
+removal phase.  Because reclaiming data items can disrupt any readers
+concurrently referencing those data items, the reclamation phase must
+not start until readers no longer hold references to those data items.
+
+Splitting the update into removal and reclamation phases permits the
+updater to perform the removal phase immediately, and to defer the
+reclamation phase until all readers active during the removal phase have
+completed, either by blocking until they finish or by registering a
+callback that is invoked after they finish.  Only readers that are active
+during the removal phase need be considered, because any reader starting
+after the removal phase will be unable to gain a reference to the removed
+data items, and therefore cannot be disrupted by the reclamation phase.
+
+So the typical RCU update sequence goes something like the following:
+
+a.     Remove pointers to a data structure, so that subsequent
+       readers cannot gain a reference to it.
+
+b.     Wait for all previous readers to complete their RCU read-side
+       critical sections.
+
+c.     At this point, there cannot be any readers who hold references
+       to the data structure, so it now may safely be reclaimed
+       (e.g., kfree()d).
+
+Step (b) above is the key idea underlying RCU's deferred destruction.