1========================= 2Unaligned Memory Accesses 3========================= 4 5:Author: Daniel Drake <dsd@gentoo.org>, 6:Author: Johannes Berg <johannes@sipsolutions.net> 7 8:With help from: Alan Cox, Avuton Olrich, Heikki Orsila, Jan Engelhardt, 9 Kyle McMartin, Kyle Moffett, Randy Dunlap, Robert Hancock, Uli Kunitz, 10 Vadim Lobanov 11 12 13Linux runs on a wide variety of architectures which have varying behaviour 14when it comes to memory access. This document presents some details about 15unaligned accesses, why you need to write code that doesn't cause them, 16and how to write such code! 17 18 19The definition of an unaligned access 20===================================== 21 22Unaligned memory accesses occur when you try to read N bytes of data starting 23from an address that is not evenly divisible by N (i.e. addr % N != 0). 24For example, reading 4 bytes of data from address 0x10004 is fine, but 25reading 4 bytes of data from address 0x10005 would be an unaligned memory 26access. 27 28The above may seem a little vague, as memory access can happen in different 29ways. The context here is at the machine code level: certain instructions read 30or write a number of bytes to or from memory (e.g. movb, movw, movl in x86 31assembly). As will become clear, it is relatively easy to spot C statements 32which will compile to multiple-byte memory access instructions, namely when 33dealing with types such as u16, u32 and u64. 34 35 36Natural alignment 37================= 38 39The rule mentioned above forms what we refer to as natural alignment: 40When accessing N bytes of memory, the base memory address must be evenly 41divisible by N, i.e. addr % N == 0. 42 43When writing code, assume the target architecture has natural alignment 44requirements. 45 46In reality, only a few architectures require natural alignment on all sizes 47of memory access. However, we must consider ALL supported architectures; 48writing code that satisfies natural alignment requirements is the easiest way 49to achieve full portability. 50 51 52Why unaligned access is bad 53=========================== 54 55The effects of performing an unaligned memory access vary from architecture 56to architecture. It would be easy to write a whole document on the differences 57here; a summary of the common scenarios is presented below: 58 59 - Some architectures are able to perform unaligned memory accesses 60 transparently, but there is usually a significant performance cost. 61 - Some architectures raise processor exceptions when unaligned accesses 62 happen. The exception handler is able to correct the unaligned access, 63 at significant cost to performance. 64 - Some architectures raise processor exceptions when unaligned accesses 65 happen, but the exceptions do not contain enough information for the 66 unaligned access to be corrected. 67 - Some architectures are not capable of unaligned memory access, but will 68 silently perform a different memory access to the one that was requested, 69 resulting in a subtle code bug that is hard to detect! 70 71It should be obvious from the above that if your code causes unaligned 72memory accesses to happen, your code will not work correctly on certain 73platforms and will cause performance problems on others. 74 75 76Code that does not cause unaligned access 77========================================= 78 79At first, the concepts above may seem a little hard to relate to actual 80coding practice. After all, you don't have a great deal of control over 81memory addresses of certain variables, etc. 82 83Fortunately things are not too complex, as in most cases, the compiler 84ensures that things will work for you. For example, take the following 85structure:: 86 87 struct foo { 88 u16 field1; 89 u32 field2; 90 u8 field3; 91 }; 92 93Let us assume that an instance of the above structure resides in memory 94starting at address 0x10000. With a basic level of understanding, it would 95not be unreasonable to expect that accessing field2 would cause an unaligned 96access. You'd be expecting field2 to be located at offset 2 bytes into the 97structure, i.e. address 0x10002, but that address is not evenly divisible 98by 4 (remember, we're reading a 4 byte value here). 99 100Fortunately, the compiler understands the alignment constraints, so in the 101above case it would insert 2 bytes of padding in between field1 and field2. 102Therefore, for standard structure types you can always rely on the compiler 103to pad structures so that accesses to fields are suitably aligned (assuming 104you do not cast the field to a type of different length). 105 106Similarly, you can also rely on the compiler to align variables and function 107parameters to a naturally aligned scheme, based on the size of the type of 108the variable. 109 110At this point, it should be clear that accessing a single byte (u8 or char) 111will never cause an unaligned access, because all memory addresses are evenly 112divisible by one. 113 114On a related topic, with the above considerations in mind you may observe 115that you could reorder the fields in the structure in order to place fields 116where padding would otherwise be inserted, and hence reduce the overall 117resident memory size of structure instances. The optimal layout of the 118above example is:: 119 120 struct foo { 121 u32 field2; 122 u16 field1; 123 u8 field3; 124 }; 125 126For a natural alignment scheme, the compiler would only have to add a single 127byte of padding at the end of the structure. This padding is added in order 128to satisfy alignment constraints for arrays of these structures. 129 130Another point worth mentioning is the use of __attribute__((packed)) on a 131structure type. This GCC-specific attribute tells the compiler never to 132insert any padding within structures, useful when you want to use a C struct 133to represent some data that comes in a fixed arrangement 'off the wire'. 134 135You might be inclined to believe that usage of this attribute can easily 136lead to unaligned accesses when accessing fields that do not satisfy 137architectural alignment requirements. However, again, the compiler is aware 138of the alignment constraints and will generate extra instructions to perform 139the memory access in a way that does not cause unaligned access. Of course, 140the extra instructions obviously cause a loss in performance compared to the 141non-packed case, so the packed attribute should only be used when avoiding 142structure padding is of importance. 143 144 145Code that causes unaligned access 146================================= 147 148With the above in mind, let's move onto a real life example of a function 149that can cause an unaligned memory access. The following function taken 150from include/linux/etherdevice.h is an optimized routine to compare two 151ethernet MAC addresses for equality:: 152 153 bool ether_addr_equal(const u8 *addr1, const u8 *addr2) 154 { 155 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS 156 u32 fold = ((*(const u32 *)addr1) ^ (*(const u32 *)addr2)) | 157 ((*(const u16 *)(addr1 + 4)) ^ (*(const u16 *)(addr2 + 4))); 158 159 return fold == 0; 160 #else 161 const u16 *a = (const u16 *)addr1; 162 const u16 *b = (const u16 *)addr2; 163 return ((a[0] ^ b[0]) | (a[1] ^ b[1]) | (a[2] ^ b[2])) == 0; 164 #endif 165 } 166 167In the above function, when the hardware has efficient unaligned access 168capability, there is no issue with this code. But when the hardware isn't 169able to access memory on arbitrary boundaries, the reference to a[0] causes 1702 bytes (16 bits) to be read from memory starting at address addr1. 171 172Think about what would happen if addr1 was an odd address such as 0x10003. 173(Hint: it'd be an unaligned access.) 174 175Despite the potential unaligned access problems with the above function, it 176is included in the kernel anyway but is understood to only work normally on 17716-bit-aligned addresses. It is up to the caller to ensure this alignment or 178not use this function at all. This alignment-unsafe function is still useful 179as it is a decent optimization for the cases when you can ensure alignment, 180which is true almost all of the time in ethernet networking context. 181 182 183Here is another example of some code that could cause unaligned accesses:: 184 185 void myfunc(u8 *data, u32 value) 186 { 187 [...] 188 *((u32 *) data) = cpu_to_le32(value); 189 [...] 190 } 191 192This code will cause unaligned accesses every time the data parameter points 193to an address that is not evenly divisible by 4. 194 195In summary, the 2 main scenarios where you may run into unaligned access 196problems involve: 197 198 1. Casting variables to types of different lengths 199 2. Pointer arithmetic followed by access to at least 2 bytes of data 200 201 202Avoiding unaligned accesses 203=========================== 204 205The easiest way to avoid unaligned access is to use the get_unaligned() and 206put_unaligned() macros provided by the <asm/unaligned.h> header file. 207 208Going back to an earlier example of code that potentially causes unaligned 209access:: 210 211 void myfunc(u8 *data, u32 value) 212 { 213 [...] 214 *((u32 *) data) = cpu_to_le32(value); 215 [...] 216 } 217 218To avoid the unaligned memory access, you would rewrite it as follows:: 219 220 void myfunc(u8 *data, u32 value) 221 { 222 [...] 223 value = cpu_to_le32(value); 224 put_unaligned(value, (u32 *) data); 225 [...] 226 } 227 228The get_unaligned() macro works similarly. Assuming 'data' is a pointer to 229memory and you wish to avoid unaligned access, its usage is as follows:: 230 231 u32 value = get_unaligned((u32 *) data); 232 233These macros work for memory accesses of any length (not just 32 bits as 234in the examples above). Be aware that when compared to standard access of 235aligned memory, using these macros to access unaligned memory can be costly in 236terms of performance. 237 238If use of such macros is not convenient, another option is to use memcpy(), 239where the source or destination (or both) are of type u8* or unsigned char*. 240Due to the byte-wise nature of this operation, unaligned accesses are avoided. 241 242 243Alignment vs. Networking 244======================== 245 246On architectures that require aligned loads, networking requires that the IP 247header is aligned on a four-byte boundary to optimise the IP stack. For 248regular ethernet hardware, the constant NET_IP_ALIGN is used. On most 249architectures this constant has the value 2 because the normal ethernet 250header is 14 bytes long, so in order to get proper alignment one needs to 251DMA to an address which can be expressed as 4*n + 2. One notable exception 252here is powerpc which defines NET_IP_ALIGN to 0 because DMA to unaligned 253addresses can be very expensive and dwarf the cost of unaligned loads. 254 255For some ethernet hardware that cannot DMA to unaligned addresses like 2564*n+2 or non-ethernet hardware, this can be a problem, and it is then 257required to copy the incoming frame into an aligned buffer. Because this is 258unnecessary on architectures that can do unaligned accesses, the code can be 259made dependent on CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS like so:: 260 261 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS 262 skb = original skb 263 #else 264 skb = copy skb 265 #endif 266