1==================================
2Cache and TLB Flushing Under Linux
3==================================
4
5:Author: David S. Miller <davem@redhat.com>
6
7This document describes the cache/tlb flushing interfaces called
8by the Linux VM subsystem.  It enumerates over each interface,
9describes its intended purpose, and what side effect is expected
10after the interface is invoked.
11
12The side effects described below are stated for a uniprocessor
13implementation, and what is to happen on that single processor.  The
14SMP cases are a simple extension, in that you just extend the
15definition such that the side effect for a particular interface occurs
16on all processors in the system.  Don't let this scare you into
17thinking SMP cache/tlb flushing must be so inefficient, this is in
18fact an area where many optimizations are possible.  For example,
19if it can be proven that a user address space has never executed
20on a cpu (see mm_cpumask()), one need not perform a flush
21for this address space on that cpu.
22
23First, the TLB flushing interfaces, since they are the simplest.  The
24"TLB" is abstracted under Linux as something the cpu uses to cache
25virtual-->physical address translations obtained from the software
26page tables.  Meaning that if the software page tables change, it is
27possible for stale translations to exist in this "TLB" cache.
28Therefore when software page table changes occur, the kernel will
29invoke one of the following flush methods _after_ the page table
30changes occur:
31
321) ``void flush_tlb_all(void)``
33
34	The most severe flush of all.  After this interface runs,
35	any previous page table modification whatsoever will be
36	visible to the cpu.
37
38	This is usually invoked when the kernel page tables are
39	changed, since such translations are "global" in nature.
40
412) ``void flush_tlb_mm(struct mm_struct *mm)``
42
43	This interface flushes an entire user address space from
44	the TLB.  After running, this interface must make sure that
45	any previous page table modifications for the address space
46	'mm' will be visible to the cpu.  That is, after running,
47	there will be no entries in the TLB for 'mm'.
48
49	This interface is used to handle whole address space
50	page table operations such as what happens during
51	fork, and exec.
52
533) ``void flush_tlb_range(struct vm_area_struct *vma,
54   unsigned long start, unsigned long end)``
55
56	Here we are flushing a specific range of (user) virtual
57	address translations from the TLB.  After running, this
58	interface must make sure that any previous page table
59	modifications for the address space 'vma->vm_mm' in the range
60	'start' to 'end-1' will be visible to the cpu.  That is, after
61	running, there will be no entries in the TLB for 'mm' for
62	virtual addresses in the range 'start' to 'end-1'.
63
64	The "vma" is the backing store being used for the region.
65	Primarily, this is used for munmap() type operations.
66
67	The interface is provided in hopes that the port can find
68	a suitably efficient method for removing multiple page
69	sized translations from the TLB, instead of having the kernel
70	call flush_tlb_page (see below) for each entry which may be
71	modified.
72
734) ``void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)``
74
75	This time we need to remove the PAGE_SIZE sized translation
76	from the TLB.  The 'vma' is the backing structure used by
77	Linux to keep track of mmap'd regions for a process, the
78	address space is available via vma->vm_mm.  Also, one may
79	test (vma->vm_flags & VM_EXEC) to see if this region is
80	executable (and thus could be in the 'instruction TLB' in
81	split-tlb type setups).
82
83	After running, this interface must make sure that any previous
84	page table modification for address space 'vma->vm_mm' for
85	user virtual address 'addr' will be visible to the cpu.  That
86	is, after running, there will be no entries in the TLB for
87	'vma->vm_mm' for virtual address 'addr'.
88
89	This is used primarily during fault processing.
90
915) ``void update_mmu_cache(struct vm_area_struct *vma,
92   unsigned long address, pte_t *ptep)``
93
94	At the end of every page fault, this routine is invoked to
95	tell the architecture specific code that a translation
96	now exists at virtual address "address" for address space
97	"vma->vm_mm", in the software page tables.
98
99	A port may use this information in any way it so chooses.
100	For example, it could use this event to pre-load TLB
101	translations for software managed TLB configurations.
102	The sparc64 port currently does this.
103
104Next, we have the cache flushing interfaces.  In general, when Linux
105is changing an existing virtual-->physical mapping to a new value,
106the sequence will be in one of the following forms::
107
108	1) flush_cache_mm(mm);
109	   change_all_page_tables_of(mm);
110	   flush_tlb_mm(mm);
111
112	2) flush_cache_range(vma, start, end);
113	   change_range_of_page_tables(mm, start, end);
114	   flush_tlb_range(vma, start, end);
115
116	3) flush_cache_page(vma, addr, pfn);
117	   set_pte(pte_pointer, new_pte_val);
118	   flush_tlb_page(vma, addr);
119
120The cache level flush will always be first, because this allows
121us to properly handle systems whose caches are strict and require
122a virtual-->physical translation to exist for a virtual address
123when that virtual address is flushed from the cache.  The HyperSparc
124cpu is one such cpu with this attribute.
125
126The cache flushing routines below need only deal with cache flushing
127to the extent that it is necessary for a particular cpu.  Mostly,
128these routines must be implemented for cpus which have virtually
129indexed caches which must be flushed when virtual-->physical
130translations are changed or removed.  So, for example, the physically
131indexed physically tagged caches of IA32 processors have no need to
132implement these interfaces since the caches are fully synchronized
133and have no dependency on translation information.
134
135Here are the routines, one by one:
136
1371) ``void flush_cache_mm(struct mm_struct *mm)``
138
139	This interface flushes an entire user address space from
140	the caches.  That is, after running, there will be no cache
141	lines associated with 'mm'.
142
143	This interface is used to handle whole address space
144	page table operations such as what happens during exit and exec.
145
1462) ``void flush_cache_dup_mm(struct mm_struct *mm)``
147
148	This interface flushes an entire user address space from
149	the caches.  That is, after running, there will be no cache
150	lines associated with 'mm'.
151
152	This interface is used to handle whole address space
153	page table operations such as what happens during fork.
154
155	This option is separate from flush_cache_mm to allow some
156	optimizations for VIPT caches.
157
1583) ``void flush_cache_range(struct vm_area_struct *vma,
159   unsigned long start, unsigned long end)``
160
161	Here we are flushing a specific range of (user) virtual
162	addresses from the cache.  After running, there will be no
163	entries in the cache for 'vma->vm_mm' for virtual addresses in
164	the range 'start' to 'end-1'.
165
166	The "vma" is the backing store being used for the region.
167	Primarily, this is used for munmap() type operations.
168
169	The interface is provided in hopes that the port can find
170	a suitably efficient method for removing multiple page
171	sized regions from the cache, instead of having the kernel
172	call flush_cache_page (see below) for each entry which may be
173	modified.
174
1754) ``void flush_cache_page(struct vm_area_struct *vma, unsigned long addr, unsigned long pfn)``
176
177	This time we need to remove a PAGE_SIZE sized range
178	from the cache.  The 'vma' is the backing structure used by
179	Linux to keep track of mmap'd regions for a process, the
180	address space is available via vma->vm_mm.  Also, one may
181	test (vma->vm_flags & VM_EXEC) to see if this region is
182	executable (and thus could be in the 'instruction cache' in
183	"Harvard" type cache layouts).
184
185	The 'pfn' indicates the physical page frame (shift this value
186	left by PAGE_SHIFT to get the physical address) that 'addr'
187	translates to.  It is this mapping which should be removed from
188	the cache.
189
190	After running, there will be no entries in the cache for
191	'vma->vm_mm' for virtual address 'addr' which translates
192	to 'pfn'.
193
194	This is used primarily during fault processing.
195
1965) ``void flush_cache_kmaps(void)``
197
198	This routine need only be implemented if the platform utilizes
199	highmem.  It will be called right before all of the kmaps
200	are invalidated.
201
202	After running, there will be no entries in the cache for
203	the kernel virtual address range PKMAP_ADDR(0) to
204	PKMAP_ADDR(LAST_PKMAP).
205
206	This routing should be implemented in asm/highmem.h
207
2086) ``void flush_cache_vmap(unsigned long start, unsigned long end)``
209   ``void flush_cache_vunmap(unsigned long start, unsigned long end)``
210
211	Here in these two interfaces we are flushing a specific range
212	of (kernel) virtual addresses from the cache.  After running,
213	there will be no entries in the cache for the kernel address
214	space for virtual addresses in the range 'start' to 'end-1'.
215
216	The first of these two routines is invoked after vmap_range()
217	has installed the page table entries.  The second is invoked
218	before vunmap_range() deletes the page table entries.
219
220There exists another whole class of cpu cache issues which currently
221require a whole different set of interfaces to handle properly.
222The biggest problem is that of virtual aliasing in the data cache
223of a processor.
224
225Is your port susceptible to virtual aliasing in its D-cache?
226Well, if your D-cache is virtually indexed, is larger in size than
227PAGE_SIZE, and does not prevent multiple cache lines for the same
228physical address from existing at once, you have this problem.
229
230If your D-cache has this problem, first define asm/shmparam.h SHMLBA
231properly, it should essentially be the size of your virtually
232addressed D-cache (or if the size is variable, the largest possible
233size).  This setting will force the SYSv IPC layer to only allow user
234processes to mmap shared memory at address which are a multiple of
235this value.
236
237.. note::
238
239  This does not fix shared mmaps, check out the sparc64 port for
240  one way to solve this (in particular SPARC_FLAG_MMAPSHARED).
241
242Next, you have to solve the D-cache aliasing issue for all
243other cases.  Please keep in mind that fact that, for a given page
244mapped into some user address space, there is always at least one more
245mapping, that of the kernel in its linear mapping starting at
246PAGE_OFFSET.  So immediately, once the first user maps a given
247physical page into its address space, by implication the D-cache
248aliasing problem has the potential to exist since the kernel already
249maps this page at its virtual address.
250
251  ``void copy_user_page(void *to, void *from, unsigned long addr, struct page *page)``
252  ``void clear_user_page(void *to, unsigned long addr, struct page *page)``
253
254	These two routines store data in user anonymous or COW
255	pages.  It allows a port to efficiently avoid D-cache alias
256	issues between userspace and the kernel.
257
258	For example, a port may temporarily map 'from' and 'to' to
259	kernel virtual addresses during the copy.  The virtual address
260	for these two pages is chosen in such a way that the kernel
261	load/store instructions happen to virtual addresses which are
262	of the same "color" as the user mapping of the page.  Sparc64
263	for example, uses this technique.
264
265	The 'addr' parameter tells the virtual address where the
266	user will ultimately have this page mapped, and the 'page'
267	parameter gives a pointer to the struct page of the target.
268
269	If D-cache aliasing is not an issue, these two routines may
270	simply call memcpy/memset directly and do nothing more.
271
272  ``void flush_dcache_page(struct page *page)``
273
274        This routines must be called when:
275
276	  a) the kernel did write to a page that is in the page cache page
277	     and / or in high memory
278	  b) the kernel is about to read from a page cache page and user space
279	     shared/writable mappings of this page potentially exist.  Note
280	     that {get,pin}_user_pages{_fast} already call flush_dcache_page
281	     on any page found in the user address space and thus driver
282	     code rarely needs to take this into account.
283
284	.. note::
285
286	      This routine need only be called for page cache pages
287	      which can potentially ever be mapped into the address
288	      space of a user process.  So for example, VFS layer code
289	      handling vfs symlinks in the page cache need not call
290	      this interface at all.
291
292	The phrase "kernel writes to a page cache page" means, specifically,
293	that the kernel executes store instructions that dirty data in that
294	page at the page->virtual mapping of that page.  It is important to
295	flush here to handle D-cache aliasing, to make sure these kernel stores
296	are visible to user space mappings of that page.
297
298	The corollary case is just as important, if there are users which have
299	shared+writable mappings of this file, we must make sure that kernel
300	reads of these pages will see the most recent stores done by the user.
301
302	If D-cache aliasing is not an issue, this routine may simply be defined
303	as a nop on that architecture.
304
305        There is a bit set aside in page->flags (PG_arch_1) as "architecture
306	private".  The kernel guarantees that, for pagecache pages, it will
307	clear this bit when such a page first enters the pagecache.
308
309	This allows these interfaces to be implemented much more efficiently.
310	It allows one to "defer" (perhaps indefinitely) the actual flush if
311	there are currently no user processes mapping this page.  See sparc64's
312	flush_dcache_page and update_mmu_cache implementations for an example
313	of how to go about doing this.
314
315	The idea is, first at flush_dcache_page() time, if page_file_mapping()
316	returns a mapping, and mapping_mapped on that mapping returns %false,
317	just mark the architecture private page flag bit.  Later, in
318	update_mmu_cache(), a check is made of this flag bit, and if set the
319	flush is done and the flag bit is cleared.
320
321	.. important::
322
323			It is often important, if you defer the flush,
324			that the actual flush occurs on the same CPU
325			as did the cpu stores into the page to make it
326			dirty.  Again, see sparc64 for examples of how
327			to deal with this.
328
329  ``void copy_to_user_page(struct vm_area_struct *vma, struct page *page,
330  unsigned long user_vaddr, void *dst, void *src, int len)``
331  ``void copy_from_user_page(struct vm_area_struct *vma, struct page *page,
332  unsigned long user_vaddr, void *dst, void *src, int len)``
333
334	When the kernel needs to copy arbitrary data in and out
335	of arbitrary user pages (f.e. for ptrace()) it will use
336	these two routines.
337
338	Any necessary cache flushing or other coherency operations
339	that need to occur should happen here.  If the processor's
340	instruction cache does not snoop cpu stores, it is very
341	likely that you will need to flush the instruction cache
342	for copy_to_user_page().
343
344  ``void flush_anon_page(struct vm_area_struct *vma, struct page *page,
345  unsigned long vmaddr)``
346
347  	When the kernel needs to access the contents of an anonymous
348	page, it calls this function (currently only
349	get_user_pages()).  Note: flush_dcache_page() deliberately
350	doesn't work for an anonymous page.  The default
351	implementation is a nop (and should remain so for all coherent
352	architectures).  For incoherent architectures, it should flush
353	the cache of the page at vmaddr.
354
355  ``void flush_icache_range(unsigned long start, unsigned long end)``
356
357  	When the kernel stores into addresses that it will execute
358	out of (eg when loading modules), this function is called.
359
360	If the icache does not snoop stores then this routine will need
361	to flush it.
362
363  ``void flush_icache_page(struct vm_area_struct *vma, struct page *page)``
364
365	All the functionality of flush_icache_page can be implemented in
366	flush_dcache_page and update_mmu_cache. In the future, the hope
367	is to remove this interface completely.
368
369The final category of APIs is for I/O to deliberately aliased address
370ranges inside the kernel.  Such aliases are set up by use of the
371vmap/vmalloc API.  Since kernel I/O goes via physical pages, the I/O
372subsystem assumes that the user mapping and kernel offset mapping are
373the only aliases.  This isn't true for vmap aliases, so anything in
374the kernel trying to do I/O to vmap areas must manually manage
375coherency.  It must do this by flushing the vmap range before doing
376I/O and invalidating it after the I/O returns.
377
378  ``void flush_kernel_vmap_range(void *vaddr, int size)``
379
380       flushes the kernel cache for a given virtual address range in
381       the vmap area.  This is to make sure that any data the kernel
382       modified in the vmap range is made visible to the physical
383       page.  The design is to make this area safe to perform I/O on.
384       Note that this API does *not* also flush the offset map alias
385       of the area.
386
387  ``void invalidate_kernel_vmap_range(void *vaddr, int size) invalidates``
388
389       the cache for a given virtual address range in the vmap area
390       which prevents the processor from making the cache stale by
391       speculatively reading data while the I/O was occurring to the
392       physical pages.  This is only necessary for data reads into the
393       vmap area.
394