motorhead/qemu
1
0
Fork 0
mirror of https://github.com/Motorhead1991/qemu.git synced 2025-08-01 23:03:54 -06:00
Code Issues Projects Releases Packages Wiki Activity Actions

cputlb: Merge and move memory_notdirty_write_{prepare,complete}

Browse source 
Since 9458a9a1df, all readers of the dirty bitmaps wait
for the rcu lock, which means that they wait until the end
of any executing TranslationBlock.

As a consequence, there is no need for the actual access
to happen in between the _prepare and _complete.  Therefore,
we can improve things by merging the two functions into
notdirty_write and dropping the NotDirtyInfo structure.

In addition, the only users of notdirty_write are in cputlb.c,
so move the merged function there.  Pass in the CPUIOTLBEntry
from which the ram_addr_t may be computed.

Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This commit is contained in:
Richard Henderson 2019-09-21 18:47:59 -07:00
parent 8f5db64153
commit 707526ad86
3 changed files with 42 additions and 143 deletions
Download patch file Download diff file Expand all files Collapse all files

65
include/exec/memory-internal.h
View file

@ -49,70 +49,5 @@ void address_space_dispatch_free(AddressSpaceDispatch *d);
void mtree_print_dispatch(struct AddressSpaceDispatch *d,
MemoryRegion *root);
struct page_collection;
/* Opaque struct for passing info from memory_notdirty_write_prepare()
* to memory_notdirty_write_complete(). Callers should treat all fields
* as private, with the exception of @active.
*
* @active is a field which is not touched by either the prepare or
* complete functions, but which the caller can use if it wishes to
* track whether it has called prepare for this struct and so needs
* to later call the complete function.
*/
typedef struct {
CPUState *cpu;
struct page_collection *pages;
ram_addr_t ram_addr;
vaddr mem_vaddr;
unsigned size;
bool active;
} NotDirtyInfo;
/**
* memory_notdirty_write_prepare: call before writing to non-dirty memory
* @ndi: pointer to opaque NotDirtyInfo struct
* @cpu: CPU doing the write
* @mem_vaddr: virtual address of write
* @ram_addr: the ram address of the write
* @size: size of write in bytes
*
* Any code which writes to the host memory corresponding to
* guest RAM which has been marked as NOTDIRTY must wrap those
* writes in calls to memory_notdirty_write_prepare() and
* memory_notdirty_write_complete():
*
* NotDirtyInfo ndi;
* memory_notdirty_write_prepare(&ndi, ....);
* ... perform write here ...
* memory_notdirty_write_complete(&ndi);
*
* These calls will ensure that we flush any TCG translated code for
* the memory being written, update the dirty bits and (if possible)
* remove the slowpath callback for writing to the memory.
*
* This must only be called if we are using TCG; it will assert otherwise.
*
* We may take locks in the prepare call, so callers must ensure that
* they don't exit (via longjump or otherwise) without calling complete.
*
* This call must only be made inside an RCU critical section.
* (Note that while we're executing a TCG TB we're always in an
* RCU critical section, which is likely to be the case for callers
* of these functions.)
*/
void memory_notdirty_write_prepare(NotDirtyInfo *ndi,
CPUState *cpu,
vaddr mem_vaddr,
ram_addr_t ram_addr,
unsigned size);
/**
* memory_notdirty_write_complete: finish write to non-dirty memory
* @ndi: pointer to the opaque NotDirtyInfo struct which was initialized
* by memory_not_dirty_write_prepare().
*/
void memory_notdirty_write_complete(NotDirtyInfo *ndi);
#endif
#endif