Rusty's Linux Kernel Page

Module Utilities for 2.6

Here is the source for "module-init-tools" which replaces modutils for modern kernels. Your distribution will probably have it in a convenient package by now.

Kernel Patches

This is a page for many of my kernel patches: an increasing number of them apply on top of the others, so I decided to split them up and host them somewhere appropriate (thanks HPA!).

Key

Increasing levels reflect increased amounts of testing...

UNKNOWN	APPLIES	ALMOST	BROKEN	COMPILES	NOCOMPILE	RUNS	NORUN
Untested.	Patches cleanly.	Almost patches: lkpatch is OK	Doesn't patch.	Compiled OK.	Did not compile.	Boots OK.	Did not boot.

Kmod Section

These are my kmod enhancements, mainly concerned with removing the messy code which is required for the 2.4 kernels to do autoloading.

Module Section

These is my module rewrite, which has the following goals:

Allow more of the kernel (eg. IPv4 networking) to be modular without unacceptable penalty if not modular.
Remove any requirement for #ifdef MODULE in common code (eg. boot parameters vs. insmod parameters).
Solve any remaining races/warts in the module code (eg. cleanup() functions which wait, registrations in the wrong order, etc).
Remove and ugly code I find along the way.

Hotplug CPU Section

This is the Hot Plug CPU section: see also the SourceForge Page.

The idea is to use Read Copy Update to safely and unintrusively support addition and removal of CPUs. Of course, once CPUs can appear and disappear, the attempts to keep CPU numbers linear become meaningless, so this patch also removes the distinction between "logical" and "physical" CPU numbers (which were already the same for some ports).

Proc Section

This is my proc rewrite, which has the following goals:

Supply the simplest possible interface for exporting information to/from userspace (ie. simplicity for kernel hackers).
Provide the infrastructure for a logical, predictable user interface for kernel information (ie. simplicity for admins).
Migrate away from uncontrolled, randomly-formatted proc entries.
Allow backward compatibility so we don't break userspace.

Initializer Revision Section

The old form of designated initializers used in the kernel (".foo =") are obsolete: we need to replace them with the ISO C forms before 2.6, otherwise a future gcc change will bury us in warnings. Gcc has always supported both forms anyway, we just chose the wrong one.

This started as a trivial patch effort by Art Haas, but people complained about his patches messing up whitespace, so I wrote my own script to start submitting them.

Misc Section

This section is for patches which don't fit anywhere else.

Rusty Russell. (Yes, you can reach me via kernel org. No spam please.).

Patch Compatibility Chart
Patch name	Status	v2.6.10-rc1-bk12	v2.6.10-rc1-mm2	vppc64-2.5
EarlyParam section:
Handle __early_param and __setup Collision	Booted on 2.6.7-rc2-bk7	BROKEN	BROKEN	BROKEN
Hotcpu section:
Don't Sleep After We're Out Of Task List	Booted on 2.6.8.1-mm2	BROKEN	BROKEN	BROKEN
Debug Checks for CPU Hotplug and stop_machine	Booted on 2.6.7-bk8	APPLIES	APPLIES	BROKEN
Centralize Calls to set_task_cpu() on Fork	Booted on 2.6.6-rc2-mm2	BROKEN	BROKEN	BROKEN
Don't Sleep After We're Out Of Task List	Booted on 2.6.8.1-mm1	BROKEN	BROKEN	BROKEN
Neaten migrate_all_tasks	Tested on 2.6.8.1-mm4	BROKEN	BROKEN	BROKEN
Simplified Kthread Tests	Booted on 2.6.1-rc2-bk1	BROKEN	BROKEN	BROKEN
Create sysfs Nodes For IA64 CPUs		BROKEN	BROKEN	BROKEN
Extra Debugging Checks for Hotplug CPU Needs: Spinlock debugging for PPC64	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Hotplug CPU vs TASK_ZOMBIEs: The Sequel to Hotplug CPU vs TASK_DEAD Needs: Neaten migrate_all_tasks	Tested on 2.6.8.1-mm4	UNKNOWN	UNKNOWN	UNKNOWN
Add basic support to enable CPU hotplug for IA64 Needs: Create sysfs Nodes For IA64 CPUs		UNKNOWN	UNKNOWN	UNKNOWN
x86 code Needs: Hotplug CPU vs TASK_ZOMBIEs: The Sequel to Hotplug CPU vs TASK_DEAD	Booted on 2.6.6-rc3-bk7	UNKNOWN	UNKNOWN	UNKNOWN
Suspend And Resume APIC for x86 Needs: x86 code	Untested	UNKNOWN	UNKNOWN	UNKNOWN
Hotmem section:
Prototype Absolute Memory Indirection Layer for PPC64	Untested	BROKEN	BROKEN	BROKEN
Kmod section:
Enhanced kmod macros	Simple	BROKEN	BROKEN	BROKEN
include kmod.h	Simple	BROKEN	BROKEN	BROKEN
CONFIG_KMOD cleanup Needs: Enhanced kmod macros	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Misc section:
XBUG macro	Tested in 2.5.17	BROKEN	BROKEN	BROKEN
Test Single-Threaded Workqueues	Booted on 2.6.6-rc1-bk3	BROKEN	BROKEN	APPLIES
typecheck.h Addition	Simple	APPLIES	APPLIES	APPLIES
try_dev_hold	Experimental	BROKEN	BROKEN	BROKEN
Simple Dynamic Allocation of groups array	Experimental	APPLIES	APPLIES	APPLIES
task_t removal patch	Trivial	BROKEN	BROKEN	BROKEN
streq implementation	Trivial	APPLIES	APPLIES	APPLIES
Fix stop_machine() For Nick Sched	Tested on 2.6.8.1-mm4	APPLIES	APPLIES	BROKEN
Remove Strange start_cpu_timer Code	Booted on 2.6.5-rc2-bk1	BROKEN	BROKEN	BROKEN
Debug Too-Early Use of softirqs	Untested	APPLIES	APPLIES	APPLIES
Serialize Network Packets for PPPoE	Experimental	BROKEN	BROKEN	BROKEN
Split Scheduler Structure into Per-CPU and Per-Runqueue Parts	Experimental	BROKEN	BROKEN	BROKEN
Revert Sched Domain Lockup Fix	Untested	BROKEN	BROKEN	BROKEN
Don't Check for Dead Tasks on Every Schedule	Booted on 2.6.7-rc2-bk3	BROKEN	BROKEN	BROKEN
Remove rwlocks	Experimental	BROKEN	BROKEN	BROKEN
RTAS System Call	Experimental	BROKEN	BROKEN	BROKEN
Remove CONFIG_PROC_FS Where Compiler Will Discard Anyway	Experimental	APPLIES	APPLIES	APPLIES
Removed inter_module functions	Untested	BROKEN	BROKEN	BROKEN
RCU For Softirqs and Hard IRQs Needs: Use __get_per_cpu instead of per_cpu in RCU.	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
QEMU CONFIG Patch	Booted on 2.6.4-rc1-bk3	APPLIES	APPLIES	APPLIES
put_user/get_user extra checks (x86 only)	Trivial	BROKEN	BROKEN	BROKEN
Merge Duplicate Names in prom.c Device Tree	Tested on Ameslab 19-May-2004	BROKEN	BROKEN	BROKEN
Per-CPU Memory On Right Nodes	Untested	APPLIES	APPLIES	APPLIES
Spinlock debugging for PPC64	Experimental	BROKEN	BROKEN	BROKEN
Module initcall depends	Tested on 2.5.34	BROKEN	BROKEN	BROKEN
__keep and __optional attributes	Trivial	BROKEN	BROKEN	BROKEN
IDE Oops Dumper for x86	Tested on 2.5.34 i386 SMP	BROKEN	BROKEN	BROKEN
on_one_cpu function	Booted on 2.6.5	APPLIES	BROKEN	APPLIES
notifier_unregister should return void	Tested on 2.5.66-bk2	BROKEN	BROKEN	BROKEN
Directed Migration: Don't Change cpumask in sched_balance_exec()	Experimental	BROKEN	BROKEN	BROKEN
Simple Dynamic Allocation of groups array	Experimental	BROKEN	BROKEN	BROKEN
Automagic Proc Entry Creation Needs: Allow Arbitrary Number of Init and Exit Functions	Tested on 2.5.71-bk2	UNKNOWN	UNKNOWN	UNKNOWN
mandocs target	Experimental	BROKEN	BROKEN	BROKEN
Interpret ibm,lrdr-capacity Property Correctly	Trivial	BROKEN	BROKEN	BROKEN
Add Al Viro's Recursive Lock Warning to Locking Guide	Trivial	APPLIES	APPLIES	APPLIES
Spinlock Data Debugging	Untested	BROKEN	BROKEN	BROKEN
kstrdup	Trivial	BROKEN	BROKEN	BROKEN
Kprobes for i386	Tested on 2.5.44 SMP	BROKEN	BROKEN	BROKEN
Kobject Whitespace Style Patch	Trivial	BROKEN	BROKEN	BROKEN
Make kallsyms filter out all non-C symbols	Booted on 2.6.7-bk6	BROKEN	BROKEN	BROKEN
Sort Kallsyms for Stem Compression	Booted on 2.6.6	BROKEN	BROKEN	BROKEN
in_sched_functions()	Booted on 2.6.5-mm4	BROKEN	INCLUDED	BROKEN
Poison __init regions	Experimental	BROKEN	BROKEN	BROKEN
Remove quotes around environment variables	Booted on 2.6-bk	INCLUDED	INCLUDED	BROKEN
i386 Constant Cleanup	Experimental	BROKEN	BROKEN	BROKEN
Hyperthread Scheduler Modifications	Experimental	BROKEN	BROKEN	BROKEN
Update Hacking Guide For 2.6	Experimental	BROKEN	BROKEN	BROKEN
Group Reference Counting Debug Patch	Tested on 21 May 2004	APPLIES	APPLIES	APPLIES
gfp_t	Tested against 2.5.43	BROKEN	BROKEN	BROKEN
generated file removal	Trivial	BROKEN	BROKEN	BROKEN
Clean up futex task state setting	Trivial	BROKEN	BROKEN	BROKEN
Resolve Race Between Requeue and Wakeup Futex Requeue Race	Experimental Experimental	BROKEN	BROKEN	BROKEN
Futexes Hold Reference Counts on Page or MM	Experimental	BROKEN	BROKEN	BROKEN
Minor Futex Tweaks	Trivial	BROKEN	BROKEN	BROKEN
Who's Spuriously Waking Futexes?	Tested on 2.6.3-bk1	BROKEN	BROKEN	APPLIES
Fix cmdline parsing crash	Untested	BROKEN	BROKEN	BROKEN
File Logging Patch	Untested	APPLIES	APPLIES	APPLIES
Fake Schedule for Another CPU	Works for Me	BROKEN	BROKEN	BROKEN
Event Logging Macros II	Experimental	BROKEN	BROKEN	BROKEN
Eliminate Unused Functions	Tested on 2.5.70-bk16	BROKEN	BROKEN	BROKEN
Merge __early_param() with __setup code	Tested on 2.6.5-rc3	BROKEN	BROKEN	BROKEN
Die OSS, die	Relatively Trivial	BROKEN	BROKEN	BROKEN
Detect POSIX Lock Surprises	Booted on 2.6.7-rc3-bk7	BROKEN	BROKEN	BROKEN
Put __per_cpu_offset in the thread struct, remove cpu	Tested on 2.5.69-bk15, dual x86	BROKEN	BROKEN	BROKEN
Deprecate cli/sti/restore_flags etc.	Tested on 2.5.54	BROKEN	BROKEN	BROKEN
Make current an attribute const function	Experimental	BROKEN	BROKEN	BROKEN
Simple Cleanup of Uses of cpumask_t	Booted on 2.6.5	BROKEN	BROKEN	BROKEN
CPU mask patch	Tested on 2.5.44 SMP	BROKEN	BROKEN	BROKEN
Remove Useless cpumask Initializers	Untested	BROKEN	BROKEN	BROKEN
CPU iterator patch	Trivial	BROKEN	BROKEN	BROKEN
copy_device_tree to use Open Firmware claim call	Untested	BROKEN	BROKEN	BROKEN
Check Calls to smp_processor_id()	Experimental	BROKEN	BROKEN	BROKEN
Bigrefs Implementation using __alloc_percpu Needs: Dynamic per-cpu allocation using static per-cpu mechanism (old version)	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Bigrefs Implementation	Tested on 2.5.38	BROKEN	BROKEN	BROKEN
Cache Align Cleanup Patch	Relatively Trivial	BROKEN	BROKEN	BROKEN
2.4 module_param Forward Compatibility Macros	Tested on 2.5.24-pre6	BROKEN	BROKEN	BROKEN
Core initcall depends Needs: Module initcall depends	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Neater returns for i386/x86_64 cpuid/msr on Offline CPUs Needs: on_one_cpu function	Untested	BROKEN	UNKNOWN	BROKEN
Kobject Devpath Sanity Patch Needs: Kobject Whitespace Style Patch	Tested in userspace.	UNKNOWN	UNKNOWN	UNKNOWN
Rusty's Hyperthread Scheduler Modifications Needs: Hyperthread Scheduler Modifications	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
FUTEX_REQUEUE simplification Needs: Minor Futex Tweaks	Booted on 2.6.0-test4-bk9	UNKNOWN	UNKNOWN	UNKNOWN
cpumask_t Removal Needs: Simple Cleanup of Uses of cpumask_t	Untested	UNKNOWN	UNKNOWN	UNKNOWN
Expose kobject_hotplug function. Needs: Kobject Devpath Sanity Patch	Trivial.	UNKNOWN	UNKNOWN	UNKNOWN
Module section:
		APPLIES	INCLUDED	APPLIES
		APPLIES	BROKEN	BROKEN
Figure Out Who Is Inserting Bogus Modules	Booted on 2.6.9-rc3-mm3	APPLIES	INCLUDED	APPLIES
Centralize token pasting and generation of unique IDs	Tested on 2.5.70-bk13	BROKEN	BROKEN	BROKEN
try_then_request_module() in MTD code	Trivial	APPLIES	APPLIES	APPLIES
Remove inter_module functions	Compiled on 2.6.10-rc1-bk10	BROKEN	BROKEN	BROKEN
Remove inter_module functions from mtd drivers.	Compiled on 2.6.10-rc1-bk10	APPLIES	APPLIES	APPLIES
Removed inter_module functions from DRM	Compiles	APPLIES	BROKEN	APPLIES
Check Permission Are Sane on module_param Usage	Tested on 2.6.8-rc6-bk1	APPLIES	APPLIES	APPLIES
Eliminate Obsolete init_module and cleanup_module Uses	Compiled on linux-2.6.9-bk6	BROKEN	APPLIES	BROKEN
More Module Notifiers	Trivial	APPLIES	APPLIES	BROKEN
Remove alias references	Trivial	APPLIES	BROKEN	APPLIES
Module Signature Core Code	Booted on 2.6.9-rc3	APPLIES	APPLIES	APPLIES
Alias support for input_device_id tables	Experimental	APPLIES	APPLIES	BROKEN
Allow Arbitrary Number of Init and Exit Functions Needs:	Tested on 2.5.73-bk1	UNKNOWN	UNKNOWN	UNKNOWN
Fix Module Removal: Don't Ignore try_stop_module Return	Trivial	APPLIES	APPLIES	APPLIES
Remove MODULE_PARM from allyesconfig Needs:	Compiled on 2.6.9-bk6	UNKNOWN	UNKNOWN	UNKNOWN
Module init race fix Needs:	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Eliminate init_module and cleanup_module Magic Needs: Eliminate Obsolete init_module and cleanup_module Uses	Compiled on linux-2.6.9-bk6	UNKNOWN	APPLIES	UNKNOWN
Module Signature Signing Script Needs: Module Signature Core Code	Booted on 2.6.9-rc3	APPLIES	APPLIES	APPLIES
Test Arbitrary Number of Init and Exit Functions Needs: Allow Arbitrary Number of Init and Exit Functions	Tested on 2.5.73-bk1	UNKNOWN	UNKNOWN	UNKNOWN
Mark broken MODULE_PARM uses Needs: Remove MODULE_PARM from allyesconfig	Trivial	UNKNOWN	UNKNOWN	UNKNOWN
Netfilter section:
skb_walk Function	Untested	APPLIES	APPLIES	APPLIES
skb_iter Functions	Tested on 2.6.8-rc2-bk12	BROKEN	BROKEN	BROKEN
Change MASQUERADE to Use Device Address Directly	Untested	INCLUDED	INCLUDED	APPLIES
Remove list macro usage from ip_conntrack.	Experimental	BROKEN	BROKEN	BROKEN
ip_tables Stack Leakage	Trivial	APPLIES	APPLIES	APPLIES
CONFIG_IP_NF_CROSSOVER	Experimental	APPLIES	APPLIES	APPLIES
Hardware Loopback Module	Tested on 2.5.74-bk5	APPLIES	APPLIES	APPLIES
ip_nat_ftp - manip at the right place	Untested	APPLIES	APPLIES	BROKEN
Fix find_appropriate_src() To Actually Work	Tested under nfsim	APPLIES	APPLIES	BROKEN
Fix Early Drop Code	Experimental	BROKEN	BROKEN	BROKEN
Call NAT Helper Modules Directly from Conntrack Modules Needs: More Module Notifiers	Experimental	BROKEN	BROKEN	UNKNOWN
Connection Tracking Protocol Lock optimization	Trivial	BROKEN	BROKEN	BROKEN
Conntrack Hash Allocation using __get_free_pages	Experimental	BROKEN	BROKEN	BROKEN
Conntrack Hashing Patch Rewrite	Experimental	BROKEN	BROKEN	BROKEN
Removes ip_conntrack_expect_tuple_lock	Experimental	BROKEN	BROKEN	BROKEN
Change iptables Targets and Matches to Take Whole Struct	Experimental	BROKEN	BROKEN	BROKEN
ip_conntrack_alter_reply thinko fix	Booted on 2.6.0-test5-bk1	APPLIES	APPLIES	APPLIES
Avoid warning on CONNTRACK_STAT_INC in destroy_conntrack()	Trivial	APPLIES	INCLUDED	BROKEN
Avoid warning on CONNTRACK_STAT_INC in death_by_timeout()	Trivial	INCLUDED	INCLUDED	BROKEN
Fragment ID Wrap Workaround Needs: skb_iter Functions	Untested	UNKNOWN	UNKNOWN	UNKNOWN
Remove NAT to multiple ranges Needs: Change MASQUERADE to Use Device Address Directly Needs: Fix find_appropriate_src() To Actually Work	Untested	BROKEN	BROKEN	UNKNOWN
ip_conntrack_expect_alloc Needs: Removes ip_conntrack_expect_tuple_lock	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Connection Tracking Helper Lock optimization Needs: Connection Tracking Protocol Lock optimization Needs: ip_conntrack_alter_reply thinko fix Needs: Removes ip_conntrack_expect_tuple_lock	Trivial	UNKNOWN	UNKNOWN	UNKNOWN
Remove Double NAT on LOCAL_OUT Needs: Remove NAT to multiple ranges	Untested	UNKNOWN	UNKNOWN	UNKNOWN
Make Expectations Timeouts Compulsory Needs: ip_conntrack_expect_alloc	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Speed up locking for connection tracking Needs: Connection Tracking Helper Lock optimization	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Connection Tracking Expect Fix Needs: Removes ip_conntrack_expect_tuple_lock Needs: ip_conntrack_expect_alloc Needs: Make Expectations Timeouts Compulsory	Experimental	UNKNOWN	UNKNOWN	UNKNOWN
Percpu section:
Remove gratuitous cpu arg in kernel/timer.c	Trivial	BROKEN	BROKEN	BROKEN
Softirq callback parameter removal	Experimental	BROKEN	BROKEN	BROKEN
Make PPC64 Use r14 For Per-Cpu Data	Untested	BROKEN	BROKEN	BROKEN
Make PPC64 Use r14 For Per-Cpu Data	Untested	BROKEN	BROKEN	APPLIES
Unification of per-cpu headers for non-SMP	Trivial	BROKEN	BROKEN	BROKEN
Align per-cpu Section Correctly	Booted on 2.6.9-rc4-bk2	APPLIES	BROKEN	APPLIES
Make scsi user per-cpu variables	Trivial	BROKEN	BROKEN	BROKEN
Use __get_per_cpu instead of per_cpu in RCU.	Experimental	BROKEN	BROKEN	BROKEN
Make cpu_tlbstate a per-cpu variable on i386.	Experimental	BROKEN	BROKEN	BROKEN
Use local_t for linux_mib network counters	Experimental	BROKEN	BROKEN	BROKEN
Use per-cpu-allocated module references	Experimental	BROKEN	BROKEN	BROKEN
local_t test code. Needs: QEMU CONFIG Patch	Booted on 2.6.0-test1	BROKEN	BROKEN	BROKEN
local_t ia64 support	Tested on 2.6.0-test1	BROKEN	BROKEN	BROKEN
Remove softirq_pending.	Tested on 2.5.74	BROKEN	BROKEN	BROKEN
Dynamic per-cpu allocation using static per-cpu mechanism (old version)	Tested on 2.5.69-bk13	BROKEN	BROKEN	BROKEN
Dynamic per-cpu allocation using static per-cpu mechanism	Experimental	BROKEN	BROKEN	BROKEN
Unification of per-cpu headers for SMP Needs: Unification of per-cpu headers for non-SMP	Trivial	UNKNOWN	UNKNOWN	UNKNOWN
Make Generic irq_stat Structure Use per-cpu Needs: Remove softirq_pending.	Tested on 2.5.74	UNKNOWN	UNKNOWN	UNKNOWN
kmalloc testing patch Needs: Dynamic per-cpu allocation using static per-cpu mechanism	Tested on 2.5.69-bk13	UNKNOWN	UNKNOWN	UNKNOWN
Make Softirq Structure Use per-cpu Needs: Make Generic irq_stat Structure Use per-cpu	Tested on 2.5.74-bk1	UNKNOWN	UNKNOWN	UNKNOWN
Gather Stats on per-cpu Usage Needs: Make Generic irq_stat Structure Use per-cpu Needs: Use __get_per_cpu instead of per_cpu in RCU. Needs: Make Softirq Structure Use per-cpu	Experimental	UNKNOWN	UNKNOWN	UNKNOWN