Use mmap_lock in user-mode to protect TCG state and the page descriptors.
In !user-mode, each vCPU has its own TCG state, so no locks needed.
Per-page locks are used to protect the page descriptors.
Per-TB locks are used in both modes to protect TB jumps.
Some notes:
- tb_lock is removed from notdirty_mem_write by passing a
locked page_collection to tb_invalidate_phys_page_fast.
- tcg_tb_lookup/remove/insert/etc have their own internal lock(s),
so there is no need to further serialize access to them.
- do_tb_flush is run in a safe async context, meaning no other
vCPU threads are running. Therefore acquiring mmap_lock there
is just to please tools such as thread sanitizer.
- Not visible in the diff, but tb_invalidate_phys_page already
has an assert_memory_lock.
- cpu_io_recompile is !user-only, so no mmap_lock there.
- Added mmap_unlock()'s before all siglongjmp's that could
be called in user-mode while mmap_lock is held.
+ Added an assert for !have_mmap_lock() after returning from
the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic.
Performance numbers before/after:
Host: AMD Opteron(tm) Processor 6376
ubuntu 17.04 ppc64 bootup+shutdown time
700 +-+--+----+------+------------+-----------+------------*--+-+
| + + + + + *B |
| before ***B*** ** * |
|tb lock removal ###D### *** |
600 +-+ *** +-+
| ** # |
| *B* #D |
| *** * ## |
500 +-+ *** ### +-+
| * *** ### |
| *B* # ## |
| ** * #D# |
400 +-+ ** ## +-+
| ** ### |
| ** ## |
| ** # ## |
300 +-+ * B* #D# +-+
| B *** ### |
| * ** #### |
| * *** ### |
200 +-+ B *B #D# +-+
| #B* * ## # |
| #* ## |
| + D##D# + + + + |
100 +-+--+----+------+------------+-----------+------------+--+-+
1 8 16 Guest CPUs 48 64
png: https://imgur.com/HwmBHXe
debian jessie aarch64 bootup+shutdown time
90 +-+--+-----+-----+------------+------------+------------+--+-+
| + + + + + + |
| before ***B*** B |
80 +tb lock removal ###D### **D +-+
| **### |
| **## |
70 +-+ ** # +-+
| ** ## |
| ** # |
60 +-+ *B ## +-+
| ** ## |
| *** #D |
50 +-+ *** ## +-+
| * ** ### |
| **B* ### |
40 +-+ **** # ## +-+
| **** #D# |
| ***B** ### |
30 +-+ B***B** #### +-+
| B * * # ### |
| B ###D# |
20 +-+ D ##D## +-+
| D# |
| + + + + + + |
10 +-+--+-----+-----+------------+------------+------------+--+-+
1 8 16 Guest CPUs 48 64
png: https://imgur.com/iGpGFtv
The gains are high for 4-8 CPUs. Beyond that point, however, unrelated
lock contention significantly hurts scalability.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
As part of plumbing MemTxAttrs down to the IOMMU translate method,
add MemTxAttrs as an argument to memory_region_access_valid().
Its callers either have an attrs value to hand, or don't care
and can use MEMTXATTRS_UNSPECIFIED.
The callsite in flatview_access_valid() is part of a recursive
loop flatview_access_valid() -> memory_region_access_valid() ->
subpage_accepts() -> flatview_access_valid(); we make it pass
MEMTXATTRS_UNSPECIFIED for now, until the next several commits
have plumbed an attrs parameter through the rest of the loop
and we can add an attrs parameter to flatview_access_valid().
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20180521140402.23318-8-peter.maydell@linaro.org
MemoryRegionCache was reverted to "normal" address_space_* operations
for 2.9, due to lack of support for IOMMUs. Reinstate the
optimizations, caching only the IOMMU translation at address_cache_init
but not the IOMMU lookup and target AddressSpace translation are not
cached; now that MemoryRegionCache supports IOMMUs, it becomes more widely
applicable too.
The inlined fast path is defined in memory_ldst_cached.inc.h, while the
slow path uses memory_ldst.inc.c as before. The smaller fast path causes
a little code size reduction in MemoryRegionCache users:
hw/virtio/virtio.o text size before: 32373
hw/virtio/virtio.o text size after: 31941
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These accessors are called from inlined functions, and the call sequence
is much more expensive than just inlining the access. Move the
struct declaration to memory-internal.h so that exec.c and memory.c
can both use an inline function.
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The memory-internal.h header claims that it is for "obsolete
exec.c functions" which "will be removed soon". This statement
was added in 2011, six years ago, but the header is still here.
(Admittedly none of the prototypes added in commit 67d95c153b
are still in the header.)
It's convenient to have a place to put prototypes for functions
which are used internally to the various .c files of the memory
system or by the accel/tcg code, which is inevitably fairly
closely coupled. So keep the header but update the comments to
reflect what we're actually using it for.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <1511276888-17834-1-git-send-email-peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The function notdirty_mem_write() has a sequence of actions
it has to do before and after the actual business of writing
data to host RAM to ensure that dirty flags are correctly
updated and we flush any TCG translations for the region.
We need to do this also in other places that write directly
to host RAM, most notably the TCG atomic helper functions.
Pull out the before and after pieces into their own functions.
We use an API where the prepare function stashes the various
bits of information about the write into a struct for the
complete function to use, because in the calls for the atomic
helpers the place where the complete function will be called
doesn't have the information to hand.
Cc: qemu-stable@nongnu.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1511201308-23580-2-git-send-email-peter.maydell@linaro.org
This adds a new "-d" switch to "info mtree" to print dispatch tree
internals.
This changes the way "-f" is handled - it prints now flat views and
associated address spaces.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20170921085110.25598-15-aik@ozlabs.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This renames some helpers to reflect better what they do.
This should cause no behavioural change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20170921085110.25598-9-aik@ozlabs.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
As we are going to share FlatView's between AddressSpace's,
and AddressSpaceDispatch is a structure to perform quick lookup
in FlatView, this moves ASD to FlatView.
After previosly open coded ASD rendering, we can also remove
as->next_dispatch as the new FlatView pointer is stored
on a stack and set to an AS atomically.
flatview_destroy() is executed under RCU instead of
address_space_dispatch_free() now.
This makes mem_begin/mem_commit to work with ASD and mem_add with FV
as later on mem_add will be taking FV as an argument anyway.
This should cause no behavioural change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20170921085110.25598-5-aik@ozlabs.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We are going to share FlatView's between AddressSpace's and per-AS
memory listeners won't suit the purpose anymore so open code
the dispatch tree rendering.
Since there is a good chance that dispatch_listener was the only
listener, this avoids address_space_update_topology_pass() if there is
no registered listeners; this should improve starting time.
This should cause no behavioural change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20170921085110.25598-3-aik@ozlabs.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
address_space_destroy_dispatch is called from an RCU callback and hence
outside the iothread mutex (BQL). However, after address_space_destroy
no new accesses can hit the destroyed AddressSpace so it is not necessary
to observe changes to the memory map. Move the memory_listener_unregister
call earlier, to make it thread-safe again.
Reported-by: Alex Williamson <alex.williamson@redhat.com>
Fixes: 374f2981d1
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
All the functions that use ram_addr_t should be here.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
We have an end parameter in all the callers, and this make it coherent
with the rest of cpu_physical_memory_* functions, that also take a
length parameter.
Once here, move the start/end calculation to
tlb_reset_dirty_range_all() as we don't need it here anymore.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
All uses except one really want the other meaning.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
We were clearing a range of bits, so use bitmap_clear().
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
We were setting a range of bits, so use bitmap_set().
Note: xen has always been wrong, and should have used start instead
of addr from the beginning.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
This operation is way faster than doing it bit by bit.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
Now all functions use the same wording that bitops/bitmap operations
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
And make cpu_physical_memory_get_dirty_flag() to use it. It used to
be the other way around.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
After all the previous patches, spliting the bitmap gets direct.
Note: For some reason, I have to move DIRTY_MEMORY_* definitions to
the beginning of memory.h to make compilation work.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
Document it
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
So remove the flag argument and do it directly. After this change,
there is nothing else using cpu_physical_memory_set_dirty_flags() so
remove it.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
So cpu_physical_memory_get_dirty_flags is not needed anymore
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
So return void.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
The current ioport dispatcher is a complex beast, mostly due to the
need to deal with old portio interface users. But we can overcome it
without converting all portio users by embedding the required base
address of a MemoryRegionPortio access into that data structure. That
removes the need to have the additional MemoryRegionIORange structure
in the loop on every access.
To handle old portio memory ops, we simply install dispatching handlers
for portio memory regions when registering them with the memory core.
This removes the need for the old_portio field.
We can drop the additional aliasing of ioport regions and also the
special address space listener. cpu_in and cpu_out now simply call
address_space_read/write. And we can concentrate portio handling in a
single source file.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We'll use it to implement address_space_access_valid.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It is a private interface between exec.c and memory.c.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Many of these should be cleaned up with proper qdev-/QOM-ification.
Right now there are many catch-all headers in include/hw/ARCH depending
on cpu.h, and this makes it necessary to compile these files per-target.
However, fixing this does not belong in these patches.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
