1 PORTING FREEBSD DRIVERS TO DRAGONFLY 2 3* Copy the driver code to the appropriate DragonFly directory. For example, 4 a disk driver /usr/src/sys/dev/blah in FreeBSD would likely be 5 /usr/src/sys/dev/disk/blah in DragonFly. 6 7* Keep all the SVN IDs in the files as a future reference point. dports' SVN 8 will do that by default. When using the FreeBSD git repo, please note the 9 files' IDs manually, either in the files themselves or in the commit message. 10 The general idea is that it must not get lost. 11 12* Driver local #include's probably use a <dev/blah/blah.h> path. These 13 need to be changed to "blah.h". '.' is not included in the #include 14 path in FreeBSD builds, but it is in DragonFly builds. 15 16* Other #include's may reference things in <dev/...> which in DragonFly 17 reside in <bus/...>. In particular, dev/pccard becomes bus/pccard. 18 Note that defines in FreeBSD's pccard_cis.h reside in DragonFly's 19 pccardreg.h . 20 21* The following kernel functions have been renamed in DragonFly: 22 23 malloc(), free() etc. -> kmalloc(), kfree() etc. 24 printf() etc. -> kprintf() etc. 25 psignal() -> ksignal() 26 random() -> krandom() 27 28* MUTEX conversion - mutexes are generally replaced by spinlocks. However, 29 DragonFly spinlocks are more restrictive than FreeBSD mutexes so a 30 direct replacement is not necessarily appropriate in all cases. A lockmgr 31 lock should be used when a direct replacement is not appropriate. 32 In particular, DragonFly does not allow recursive exclusive spinlocks 33 and does not allow multiple exclusive spinlocks to be held by any given 34 thread. 35 36 Instances of <sys/mutex.h> should be replaced with <sys/spinlock.h>. 37 38 When replacing mutexes with spinlocks it is a good idea to rename 39 the structural field (typically 'mtx') to something else (typically 'spin'). 40 41 The &Giant mutex is typically converted to get_mplock() and rel_mplock(). 42 However, there are places where FreeBSD unlocks giant around some code and 43 then relocks giant... those should simply be removed. 44 45 FreeBSD has weird callout + mutex functions. DragonFly does not integrate 46 the two. Instead, the driver in DragonFly must obtain the spinlocks 47 in question in the callback routine. 48 49 As a rule of thumb, MTX_DEF mutexes should be replaced with exclusive, 50 recursive lockmgr locks. 51 52 So, suppose the original code is using 53 struct mtx my_mtx; 54 you'd normally rename it to 55 struct lock my_lock; 56 57 and change the initialization from something like 58 mtx_init(&my_mtx, "mymtx", "whatever", MTX_DEF); 59 to 60 lockinit(&my_lock, "mylock", 0, LK_CANRECURSE); 61 62 Destroying it is trivial, 63 mtx_destroy(&my_mtx); 64 becomes 65 lockuninit(&my_lock); 66 67 You use the same function for locking and unlocking a lockmgr lock, 68 so exchange 69 mtx_lock(&my_mtx); 70 with 71 lockmgr(&my_lock, LK_EXCLUSIVE); 72 and 73 mtx_unlock(&my_mtx); 74 with 75 lockmgr(&my_lock, LK_RELEASE); 76 77 For testing the lock status, one would use 78 lockstatus(&my_lock, curthread); 79 in place of 80 mtx_owned(&my_mtx); 81 82 An 83 mtx_trylock(&my_mtx); 84 call is replaced with 85 lockmgr(&my_lock, LK_EXCLUSIVE|LK_NOWAIT); 86 87 As for mtx_assert() calls, translate them like this: 88 89 mtx_assert(&my_mtx, MA_OWNED) -> KKASSERT(lockstatus(&my_lock, curthread) != 0) 90 mtx_assert(&my_mtx, MA_NOTOWNED) -> KKASSERT(lockstatus(&my_lock, curthread) == 0) 91 92 In DragonFly, lockstatus() does not return information about whether there have been 93 recursive lock acquisitions, so there is no generic way to emulate the 94 95 mtx_assert(&my_mtx, MA_OWNED|MA_RECURSED); 96 mtx_assert(&my_mtx, MA_OWNED|MA_NOTRECURSED); 97 98 calls. 99 100* rwlock conversion: Use lockmgr locks 101 102* UMA conversion - generally speaking UMA should be converted to a standard 103 kmalloc. 104 105 Note however that in FreeBSD M_NOWAIT is often used in cases where, in fact, 106 the kmalloc cannot fail without blowing something up or causing a fatal 107 (and very unexpected) I/O error. M_INTWAIT should be used for these cases. 108 109* CDEVSW conversion - see other devices. Generally speaking a major number 110 is needed and a function map needs to be specified more explicitly. 111 112 Most calls passing struct cdev pointers are dev_t's in DragonFly. 113 114 All device vectors in DragonFly pass a dev_<name>_args structure pointer 115 instead of explicit arguments. 116 117 Strategy calls - we pass BIO's and a lot of BUF fields are in the BIO 118 in FreeBSD, but left in the BUF in DragonFly. FreeBSD for some reason 119 names its struct bio pointers 'bp', its a good idea to rename them to 'bio' 120 to avoid confusion and have a struct buf *bp = bio->bio_buf; pointer to 121 access the buf. 122 123* MSLEEP/TSLEEP conversion. The DragonFly msleep/tsleep do not have 'PRI' 124 priorities. 0 should be used. 125 126* BUS_* FUNCTIONS 127 128 bus_setup_intr() - replace INTR_TYPE_* flags with 0. There is an extra 129 argument for an interrupt interlock using the sys/serializer.h interface. 130 This can either be left NULL or you can convert the spinlock(s) for 131 the driver into serializer locks and integrate the interrupt service 132 routine with a serializer. 133 134* CAM CODE - cam_simq* code refcounts, so shared device queues (raid and 135 multi-channel devices) are not freed before all references have gone 136 away. 137 138* callout_drain() should be replaced by callout_stop_sync() 139 140* UNRHDR functions - DragonFly uses a more generic idr(9) subsystem 141 compatible with the Linux API of the same name 142 143 This LWN article describes it in details: http://lwn.net/Articles/103209/ 144 145 A typical conversion looks like this: 146 147 #include <sys/idr.h> 148 149 free_unr() has to be replaced by idr_remove() 150 151 alloc_unr() has to be replaced by a code sequence using idr_pre_get and 152 idr_get_new such as this one: 153 154 retry: 155 if (idr_pre_get(xxx) ==0) { 156 kprintf("Memory allocation error\n"); 157 return error; 158 } 159 spin_lock(xxx); 160 ret = idr_get_new(xxx); 161 spin_unlock(xxx); 162 if (ret == EAGAIN) 163 goto retry; 164 165* MPASS macro - Replace it with KKASSERT 166 167 168* PROC_LOCK / PROC_UNLOCK: to be determined on a case-by-case basis 169 170 Some of the time these macros can be removed entirely 171 172 In some cases, some locking must be done; lwkt_gettoken(&proc_token) 173 and the corresponding lwkt_reltoken() call should be good replacements 174 175 It is not a good idea to blindly implement these macros globally, some 176 particular proc subsystem locking semantics differ enough between FreeBSD 177 and DragonFly that this would cause problems 178 179 180* In DragonFly 3.3 format specifier %D was removed from kprintf. As a 181 replacement functions kether_ntoa() and hexncpy() were added. 182 183 - Ethernet address (MAC) to its hexadecimal form: 184 185 char ethstr[ETHER_ADDRSTRLEN + 1]; 186 u_char hwaddr[6]; 187 188 kprintf("MAC address %s\n", kether_ntoa(hwaddr, ethstr) 189 190 - Generic conversion (block of bytes to hexadecimal form): 191 192 char hexstr[18]; 193 u_char mydata[6] = {1, 2, 3, 4, 5 ,6}; 194 195 /* 196 * Below statement would print: 197 * 198 * 01-02-03-04-05-06 199 */ 200 kprintf("%s\n", hexncpy(mydata, 6, hexstr, HEX_NCPYLEN(6), "-")); 201 202* TAILQ_XXX_SAFE 203 204 Use TAILQ_XXX_MUTABLE; the macros have the same effect, only the name is 205 different 206 207* kern_yield() 208 209 Replace by lwkt_yield() 210 211* vm_page_lock() and vm_page_unlock() 212 213 Not needed on DragonFly, remove these calls 214 215* vm_pager_get_pages() 216 217 Removed, use vm_pager_get_page() instead 218 219* VPO_BUSY 220 221 Replace by PG_BUSY 222 223* kern_psignal() 224 225 Replace by ksignal() 226