brintos

brintos / linux-shallow public Read only

0
0
Text · 7.3 KiB · 6f3d16d Raw
179 lines · plain
1This document contains brief definitions of LKMM-related terms.  Like most2glossaries, it is not intended to be read front to back (except perhaps3as a way of confirming a diagnosis of OCD), but rather to be searched4for specific terms.5 6 7Address Dependency:  When the address of a later memory access is computed8	based on the value returned by an earlier load, an "address9	dependency" extends from that load extending to the later access.10	Address dependencies are quite common in RCU read-side critical11	sections:12 13	 1 rcu_read_lock();14	 2 p = rcu_dereference(gp);15	 3 do_something(p->a);16	 4 rcu_read_unlock();17 18	 In this case, because the address of "p->a" on line 3 is computed19	 from the value returned by the rcu_dereference() on line 2, the20	 address dependency extends from that rcu_dereference() to that21	 "p->a".  In rare cases, optimizing compilers can destroy address22	 dependencies.	Please see Documentation/RCU/rcu_dereference.rst23	 for more information.24 25	 See also "Control Dependency" and "Data Dependency".26 27Acquire:  With respect to a lock, acquiring that lock, for example,28	using spin_lock().  With respect to a non-lock shared variable,29	a special operation that includes a load and which orders that30	load before later memory references running on that same CPU.31	An example special acquire operation is smp_load_acquire(),32	but atomic_read_acquire() and atomic_xchg_acquire() also include33	acquire loads.34 35	When an acquire load returns the value stored by a release store36	to that same variable, (in other words, the acquire load "reads37	from" the release store), then all operations preceding that38	store "happen before" any operations following that load acquire.39 40	See also "Happens-Before", "Reads-From", "Relaxed", and "Release".41 42Coherence (co):  When one CPU's store to a given variable overwrites43	either the value from another CPU's store or some later value,44	there is said to be a coherence link from the second CPU to45	the first.46 47	It is also possible to have a coherence link within a CPU, which48	is a "coherence internal" (coi) link.  The term "coherence49	external" (coe) link is used when it is necessary to exclude50	the coi case.51 52	See also "From-reads" and "Reads-from".53 54Control Dependency:  When a later store's execution depends on a test55	of a value computed from a value returned by an earlier load,56	a "control dependency" extends from that load to that store.57	For example:58 59	 1 if (READ_ONCE(x))60	 2   WRITE_ONCE(y, 1);61 62	 Here, the control dependency extends from the READ_ONCE() on63	 line 1 to the WRITE_ONCE() on line 2.	Control dependencies are64	 fragile, and can be easily destroyed by optimizing compilers.65	 Please see control-dependencies.txt for more information.66 67	 See also "Address Dependency" and "Data Dependency".68 69Cycle:	Memory-barrier pairing is restricted to a pair of CPUs, as the70	name suggests.	And in a great many cases, a pair of CPUs is all71	that is required.  In other cases, the notion of pairing must be72	extended to additional CPUs, and the result is called a "cycle".73	In a cycle, each CPU's ordering interacts with that of the next:74 75	CPU 0                CPU 1                CPU 276	WRITE_ONCE(x, 1);    WRITE_ONCE(y, 1);    WRITE_ONCE(z, 1);77	smp_mb();            smp_mb();            smp_mb();78	r0 = READ_ONCE(y);   r1 = READ_ONCE(z);   r2 = READ_ONCE(x);79 80	CPU 0's smp_mb() interacts with that of CPU 1, which interacts81	with that of CPU 2, which in turn interacts with that of CPU 082	to complete the cycle.	Because of the smp_mb() calls between83	each pair of memory accesses, the outcome where r0, r1, and r284	are all equal to zero is forbidden by LKMM.85 86	See also "Pairing".87 88Data Dependency:  When the data written by a later store is computed based89	on the value returned by an earlier load, a "data dependency"90	extends from that load to that later store.  For example:91 92	 1 r1 = READ_ONCE(x);93	 2 WRITE_ONCE(y, r1 + 1);94 95	In this case, the data dependency extends from the READ_ONCE()96	on line 1 to the WRITE_ONCE() on line 2.  Data dependencies are97	fragile and can be easily destroyed by optimizing compilers.98	Because optimizing compilers put a great deal of effort into99	working out what values integer variables might have, this is100	especially true in cases where the dependency is carried through101	an integer.102 103	See also "Address Dependency" and "Control Dependency".104 105From-Reads (fr):  When one CPU's store to a given variable happened106	too late to affect the value returned by another CPU's107	load from that same variable, there is said to be a from-reads108	link from the load to the store.109 110	It is also possible to have a from-reads link within a CPU, which111	is a "from-reads internal" (fri) link.  The term "from-reads112	external" (fre) link is used when it is necessary to exclude113	the fri case.114 115	See also "Coherence" and "Reads-from".116 117Fully Ordered:  An operation such as smp_mb() that orders all of118	its CPU's prior accesses with all of that CPU's subsequent119	accesses, or a marked access such as atomic_add_return()120	that orders all of its CPU's prior accesses, itself, and121	all of its CPU's subsequent accesses.122 123Happens-Before (hb): A relation between two accesses in which LKMM124	guarantees the first access precedes the second.  For more125	detail, please see the "THE HAPPENS-BEFORE RELATION: hb"126	section of explanation.txt.127 128Marked Access:  An access to a variable that uses an special function or129	macro such as "r1 = READ_ONCE(x)" or "smp_store_release(&a, 1)".130 131	See also "Unmarked Access".132 133Pairing: "Memory-barrier pairing" reflects the fact that synchronizing134	data between two CPUs requires that both CPUs their accesses.135	Memory barriers thus tend to come in pairs, one executed by136	one of the CPUs and the other by the other CPU.  Of course,137	pairing also occurs with other types of operations, so that a138	smp_store_release() pairs with an smp_load_acquire() that reads139	the value stored.140 141	See also "Cycle".142 143Reads-From (rf):  When one CPU's load returns the value stored by some other144	CPU, there is said to be a reads-from link from the second145	CPU's store to the first CPU's load.  Reads-from links have the146	nice property that time must advance from the store to the load,147	which means that algorithms using reads-from links can use lighter148	weight ordering and synchronization compared to algorithms using149	coherence and from-reads links.150 151	It is also possible to have a reads-from link within a CPU, which152	is a "reads-from internal" (rfi) link.	The term "reads-from153	external" (rfe) link is used when it is necessary to exclude154	the rfi case.155 156	See also Coherence" and "From-reads".157 158Relaxed:  A marked access that does not imply ordering, for example, a159	READ_ONCE(), WRITE_ONCE(), a non-value-returning read-modify-write160	operation, or a value-returning read-modify-write operation whose161	name ends in "_relaxed".162 163	See also "Acquire" and "Release".164 165Release:  With respect to a lock, releasing that lock, for example,166	using spin_unlock().  With respect to a non-lock shared variable,167	a special operation that includes a store and which orders that168	store after earlier memory references that ran on that same CPU.169	An example special release store is smp_store_release(), but170	atomic_set_release() and atomic_cmpxchg_release() also include171	release stores.172 173	See also "Acquire" and "Relaxed".174 175Unmarked Access:  An access to a variable that uses normal C-language176	syntax, for example, "a = b[2]";177 178	See also "Marked Access".179