TODO.modules revision 1.18
1/* $NetBSD: TODO.modules,v 1.18 2018/12/28 21:51:49 pgoyette Exp $ */ 2 3Some notes on the limitations of our current (as of 7.99.35) module 4subsystem. This list was triggered by an Email exchange between 5christos and pgoyette. 6 7 1. Builtin drivers can't depend on modularized drivers (the modularized 8 drivers are attempted to load as builtins). 9 10 The assumption is that dependencies are loaded before those 11 modules which depend on them. At load time, a module's 12 undefined global symbols are resolved; if any symbols can't 13 be resolved, the load fails. Similarly, if a module is 14 included in (built-into) the kernel, all of its symbols must 15 be resolvable by the linker, otherwise the link fails. 16 17 There are ways around this (such as, having the parent 18 module's initialization command recursively call the module 19 load code), but they're often gross hacks. 20 21 Another alternative (which is used by ppp) is to provide a 22 "registration" mechanism for the "child" modules, and then when 23 the need for a specific child module is encountered, use 24 module_autoload() to load the child module. Of course, this 25 requires that the parent module know about all potentially 26 loadable children. 27 28 2. Currently, config(1) has no way to "no define" drivers 29 XXX: I don't think this is true anymore. I think we can 30 undefine drivers now, see MODULAR in amd64, which does 31 no ath* and no select sppp* 32 33 3. It is not always obvious by their names which drivers/options 34 correspond to which modules. 35 36 4. Right now critical drivers that would need to be pre-loaded (ffs, 37 exec_elf64) are still built-in so that we don't need to alter the boot 38 blocks to boot. 39 40 This was a conscious decision by core@ some years ago. It is 41 not a requirement that ffs or exec_* be built-in. The only 42 requirement is that the root file-system's module must be 43 available when the module subsystem is initialized, in order 44 to load other modules. This can be accomplished by having the 45 boot loader "push" the module at boot time. (It used to do 46 this in all cases; currently the "push" only occurs if the 47 booted filesystem is not ffs.) 48 49 5. Not all parent bus drivers are capable of rescan, so some drivers 50 just have to be built-in. 51 52 6. Many (most?) drivers are not yet modularized 53 54 7. There's currently no provisions for autoconfig to figure out which 55 modules are needed, and thus to load the required modules. 56 57 In the "normal" built-in world, autoconfigure can only ask 58 existing drivers if they're willing to manage (ie, attach) a 59 device. Removing the built-in drivers tends to limit the 60 availability of possible managers. There's currently no 61 mechanism for identifying and loading drivers based on what 62 devices might be found. 63 64 8. Even for existing modules, there are "surprise" dependencies with 65 code that has not yet been modularized. 66 67 For example, even though the bpf code has been modularized, 68 there is some shared code in bpf_filter.c which is needed by 69 both ipfilter and ppp. ipf is already modularized, but ppp 70 is not. Thus, even though bpf_filter is modular, it MUST be 71 included as a built-in module if you also have ppp in your 72 configuration. 73 74 Another example is sysmon_taskq module. It is required by 75 other parts of the sysmon subsystem, including the 76 "sysmon_power" module. Unfortunately, even though the 77 sysmon_power code is modularized, it is referenced by the 78 acpi code which has not been modularized. Therefore, if your 79 configuration has acpi, then you must include the "sysmon_power" 80 module built-in the kernel. And therefore you also need to 81 have "sysmon_taskq" and "sysmon" built-in since "sysmon_power" 82 rerefences them. 83 84 9. As a corollary to #8 above, having dependencies on modules from code 85 which has not been modularized makes it extremely difficult to test 86 the module code adequately. Testing of module code should include 87 both testing-as-a-built-in module and testing-as-a-loaded-module, and 88 all dependencies need to be identified. 89 9010. The current /stand/$ARCH/$VERSION/modules/ hierarchy won't scale as 91 we get more and more modules. There are hundreds of potential device 92 driver modules. 93 9411. There currently isn't any good way to handle attachment-specific 95 modules. The build infrastructure (ie, sys/modules/Makefile) doesn't 96 readily lend itself to bus-specific modules irrespective of $ARCH, 97 and maintaining distrib/sets/lists/modules/* is awkward at best. 98 99 Furthermore, devices such as ld(4), which can attach to a large set 100 of parent devices, need to be modified. The parent devices need to 101 provide a common attribute (for example, ld_bus), and the ld driver 102 should attach to that attribute rather than to each parent. But 103 currently, config(1) doesn't handle this - it doesn't allow an 104 attribute to be used as the device tree's pseudo-root. The current 105 directory structure where driver foo is split between ic/foo.c 106 and bus1/foo_bus1.c ... busn/foo_busn.c is annoying. It would be 107 better to switch to the FreeBSD model which puts all the driver 108 files in one directory. 109 11012. Item #11 gets even murkier when a particular parent can provide more 111 than one attribute. 112 11313. It seems that we might want some additional sets-lists "attributes" 114 to control contents of distributions. As an example, many of our 115 architectures have PCI bus capabilities, but not all. It is rather 116 painful to need to maintain individual architectures' modules/md_* 117 sets lists, especially when we already have to conditionalize the 118 build of the modules based on architecture. If we had a single 119 "attribute" for PCI-bus-capable, the same attribute could be used to 120 select which modules to build and which modules from modules/mi to 121 include in the release. (This is not limited to PCI; recently we 122 encounter similar issues with spkr aka spkr_synth module.) 123 12414. As has been pointed out more than once, the current method of storing 125 modules in a version-specific subdirectory of /stand is sub-optimal 126 and leads to much difficulty and/or confusion. A better mechanism of 127 associating a kernel and its modules needs to be developed. Some 128 have suggested having a top-level directory (say, /netbsd) with a 129 kernel and its modules at /netbsd/kernel and /netbsd/modules/... 130 Whatever new mechanism we arrive at will probably require changes to 131 installation procedures and bootstrap code, and will need to handle 132 both the new and old mechanisms for compatability. 133 134 One additional option mentioned is to be able to specify, at boot 135 loader time, an alternate value for the os-release portion of the 136 default module path, i.e. /stand/$MACHINE/$ALT-RELEASE/modules/ 137 138 The following statement regarding this issue was previously issued 139 by the "core" group: 140 141 Date: Fri, 27 Jul 2012 08:02:56 +0200 142 From: <redacted> 143 To: <redacted> 144 Subject: Core statement on directory naming for kernel modules 145 146 The core group would also like to see the following changes in 147 the near future: 148 149 Implementation of the scheme described by Luke Mewburn in 150 <http://mail-index.NetBSD.org/current-users/2009/05/10/msg009372.html> 151 to allow a kernel and its modules to be kept together. 152 Changes to config(1) to extend the existing notion of whether or not 153 an option is built-in to the kernel, to three states: built-in, not 154 built-in but loadable as a module, entirely excluded and not even 155 loadable as a module. 156 157 15815. The existing config(5) framework provides an excellent mechanism 159 for managing the content of kernels. Unfortunately, this mechanism 160 does not apply for modules, and instead we need to manually manage 161 a list of files to include in the module, the set of compiler 162 definitions with which to build those files, and also the set of 163 other modules on which a module depends. We really need a common 164 mechanism to define and build modules, whether they are included as 165 "built-in" modules or as separately-loadable modules. 166 167 (From John Nemeth) Some sort of mechanism for a (driver) module 168 to declare the list of vendor/product/other tuples that it can 169 handle would be nice. Perhaps this would go in the module's .plist 170 file? (See #17 below.) Then drivers that scan for children might 171 be able to search the modules directory for an "appropriate" module 172 for each child, and auto-load. 173 17416. PR kern/52821 exposes another limitation of config(1) WRT modules. 175 Here, an explicit device attachment is required, because we cannot 176 rely on all kernel configs to contain the attribute at which the 177 modular driver wants to attach. Unfortunately, the explicit 178 attachment causes conflicts with built-in drivers. (See the PR for 179 more details.) 180 18117. (From John Nemeth) It would be potentially useful if a "push" from 182 the bootloader could also load-and-push a module's .plist (if it 183 exists. 184 18518. (From John Nemeth) Some sort of schema for a module to declare the 186 options (or other things?) that the module understands. This could 187 result in a module-options editor to manipulate the .plist 188 18919. (From John Nemeth) Currently, the order of module initialization is 190 based on module classes and declared dependencies. It might be 191 useful to have additional classes (or sub-classes) with additional 192 invocations of module_class_init(), and it might be useful to have a 193 non-dependency mechanism to provide "IF module-A and module-B are 194 BOTH present, module-A needs to be initialized before module-B". 195 19620. (Long-ago memory rises to the surface) Note that currently there is 197 nothing that requires a module's name to correspond in any way with 198 the name of file from which the module is loaded. Thus, it is 199 possible to attempt to access device /dev/x, discover that there is 200 no such device so we autoload /stand/.../x/x.kmod and initialize 201 the module loaded, even if the loaded module is for some other 202 device entirely! 203 20421. We currently do not support "weak" symbols in the in-kernel linker. 205 It would take some serious thought to get such support right. For 206 example, consider module A with a weak reference to symbol S which 207 is defined in module B. If module B is loaded first, and then 208 module A, the symbol gets resolved. But if module A is loaded first, 209 the symbol won't be resolved. If we subsequently load module B, we 210 would have to "go back" and re-run the linker for module A. 211 212 Additional difficulties arise when the module which defines the 213 weak symbol gets unloaded. Then, you would need to re-run the 214 linker and _unresolve_ the weak symbol which is no longer defined. 215