osdep.h will also define the available Windows API version for QEMU.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20181122110039.15972-2-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Michael Clark <mjc@sifive.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <c445175310fa836b61fd862a55628907f0093194.1545246859.git.alistair.francis@wdc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This is essentially redundant with tlb_c.dirty.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Especially for guests with large numbers of tlbs, like ARM or PPC,
we may well not use all of them in between flush operations.
Remember which tlbs have been used since the last flush, and
avoid any useless flushing.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Our only statistic so far was "full" tlb flushes, where all mmu_idx
are flushed at the same time.
Now count "partial" tlb flushes where sets of mmu_idx are flushed,
but the set is not maximal. Account one per mmu_idx flushed, as
that is the unit of work performed.
We don't actually count elided flushes yet, but go ahead and change
the interface presented to the monitor all at once.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The difference between the two sets of APIs is now miniscule.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The difference between the two sets of APIs is now miniscule.
This allows tlb_flush, tlb_flush_all_cpus, and tlb_flush_all_cpus_synced
to be merged with their corresponding by_mmuidx functions as well. For
accounting, consider mmu_idx_bitmask = ALL_MMUIDX_BITS to be a full flush.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The rest of the tlb victim cache is per-tlb,
the next use index should be as well.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The set of large pages in the kernel is probably not the same
as the set of large pages in the application. Forcing one
range to cover both will flush more often than necessary.
This allows tlb_flush_page_async_work to flush just the one
mmu_idx implicated, which in turn allows us to remove
tlb_check_page_and_flush_by_mmuidx_async_work.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Protect it with the tlb_lock instead of using atomics.
The move puts it in or near the same cacheline as the lock;
using the lock means we don't need a second atomic operation
in order to perform the update. Which makes it cheap to also
update pending_flush in tlb_flush_by_mmuidx_async_work.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The bugs this was working around were fixed with commits
022d6378c7 target/unicore32: remove tlb_flush from uc32_init_fn
6e11beecfd target/alpha: remove tlb_flush from alpha_cpu_initfn
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This is the first of several moves to reduce the size of the
CPU_COMMON_TLB macro and improve some locality of refernce.
Tested-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
GCC7+ will no longer advertise support for 16-byte __atomic operations
if only cmpxchg is supported, as for x86_64. Fortunately, x86_64 still
has support for __sync_compare_and_swap_16 and we can make use of that.
AArch64 does not have, nor ever has had such support, so open-code it.
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Isolate the computation of an index from an address into a
helper before we change that function.
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
[ cota: convert tlb_vaddr_to_host; use atomic_read on addr_write ]
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <20181009175129.17888-2-cota@braap.org>
Currently we rely on atomic operations for cross-CPU invalidations.
There are two cases that these atomics miss: cross-CPU invalidations
can race with either (1) vCPU threads flushing their TLB, which
happens via memset, or (2) vCPUs calling tlb_reset_dirty on their TLB,
which updates .addr_write with a regular store. This results in
undefined behaviour, since we're mixing regular and atomic ops
on concurrent accesses.
Fix it by using tlb_lock, a per-vCPU lock. All updaters of tlb_table
and the corresponding victim cache now hold the lock.
The readers that do not hold tlb_lock must use atomic reads when
reading .addr_write, since this field can be updated by other threads;
the conversion to atomic reads is done in the next patch.
Note that an alternative fix would be to expand the use of atomic ops.
However, in the case of TLB flushes this would have a huge performance
impact, since (1) TLB flushes can happen very frequently and (2) we
currently use a full memory barrier to flush each TLB entry, and a TLB
has many entries. Instead, acquiring the lock is barely slower than a
full memory barrier since it is uncontended, and with a single lock
acquisition we can flush the entire TLB.
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <20181009174557.16125-6-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
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>
Message-Id: <20181009174557.16125-5-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Paves the way for the addition of a per-TLB lock.
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <20181009174557.16125-4-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Consistently access u16.high with atomics to avoid
undefined behaviour in MTTCG.
Note that icount_decr.u16.low is only used in icount mode,
so regular accesses to it are OK.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <20181010144853.13005-2-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Rather than test NOCHAIN before linking, do not emit the
goto_tb opcode at all. We already do this for goto_ptr.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The global cpu_single_env variable has been removed more than 5 years
ago, so apparently nobody used this dead debug code in that timeframe
anymore. Thus let's remove it completely now.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1537204134-15905-1-git-send-email-thuth@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
QEMU cannot pass through the breakpoints when 'si' command is used
in remote gdb. This patch disables inserting the breakpoints
when we are already single stepping though the gdb remote protocol.
This patch also fixes icount calculation for the blocks that include
breakpoints - instruction with breakpoint is not executed and shouldn't
be used in icount calculation.
Signed-off-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
Message-Id: <20180912081910.3228.8523.stgit@pasha-VirtualBox>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Accessing the HT from an iterator results almost always
in a deadlock. Given that only one qht-internal function
uses this argument, drop it from the interface.
Suggested-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>
We set up TLB entries in tlb_set_page_with_attrs(), where we have
some logic for determining whether the TLB entry is considered
to be RAM-backed, and thus has a valid addend field. When we
look at the TLB entry in get_page_addr_code(), we use different
logic for determining whether to treat the page as RAM-backed
and use the addend field. This is confusing, and in fact buggy,
because the code in tlb_set_page_with_attrs() correctly decides
that rom_device memory regions not in romd mode are not RAM-backed,
but the code in get_page_addr_code() thinks they are RAM-backed.
This typically results in "Bad ram pointer" assertion if the
guest tries to execute from such a memory region.
Fix this by making get_page_addr_code() just look at the
TLB_MMIO bit in the code_address field of the TLB, which
tlb_set_page_with_attrs() sets if and only if the addend
field is not valid for code execution.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180713150945.12348-1-peter.maydell@linaro.org
Now that all the callers can handle get_page_addr_code() returning -1,
remove all the code which tries to handle execution from MMIO regions
or small-MMU-region RAM areas. This will mean that we can correctly
execute from these areas, rather than ending up either aborting QEMU
or delivering an incorrect guest exception.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Tested-by: Cédric Le Goater <clg@kaod.org>
Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180710160013.26559-6-peter.maydell@linaro.org
If get_page_addr_code() returns -1, this indicates that there is no RAM
page we can read a full TB from. Instead we must create a TB which
contains a single instruction and which we do not cache, so it is
executed only once.
Since this means we can now have TBs which are not in any page list,
we also need to make tb_phys_invalidate() handle them (by not trying
to remove them from a nonexistent page list).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Tested-by: Cédric Le Goater <clg@kaod.org>
Message-id: 20180710160013.26559-5-peter.maydell@linaro.org
When we support execution from non-RAM MMIO regions, get_page_addr_code()
will return -1 to indicate that there is no RAM at the requested address.
Handle this in tb_check_watchpoint() -- if the exception happened for a
PC which doesn't correspond to RAM then there is no need to invalidate
any TBs, because the one-instruction TB will not have been cached.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Cédric Le Goater <clg@kaod.org>
Message-id: 20180710160013.26559-4-peter.maydell@linaro.org
When we support execution from non-RAM MMIO regions, get_page_addr_code()
will return -1 to indicate that there is no RAM at the requested address.
Handle this in the cpu-exec TB hashtable lookup code, treating it as
"no match found".
Note that the call to get_page_addr_code() in tb_lookup_cmp() needs
no changes -- a return of -1 will already correctly result in the
function returning false.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Tested-by: Cédric Le Goater <clg@kaod.org>
Message-id: 20180710160013.26559-3-peter.maydell@linaro.org
The io_readx() function needs to know whether the load it is
doing is an MMU_DATA_LOAD or an MMU_INST_FETCH, so that it
can pass the right value to the cpu_transaction_failed()
function. Plumb this information through from the softmmu
code.
This is currently not often going to give the wrong answer,
because usually instruction fetches go via get_page_addr_code().
However once we switch over to handling execution from non-RAM by
creating single-insn TBs, the path for an insn fetch to generate
a bus error will be through cpu_ld*_code() and io_readx(),
so without this change we will generate a d-side fault when we
should generate an i-side fault.
We also have to pass the access type via a CPU struct global
down to unassigned_mem_read(), for the benefit of the targets
which still use the cpu_unassigned_access() hook (m68k, mips,
sparc, xtensa).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Tested-by: Cédric Le Goater <clg@kaod.org>
Message-id: 20180710160013.26559-2-peter.maydell@linaro.org
In commit 4b1a3e1e34 we added a check for whether the TLB entry
we had following a tlb_fill had the INVALID bit set. This could
happen in some circumstances because a stale or wrong TLB entry was
pulled out of the victim cache. However, after commit
68fea03855 (which prevents stale entries being in the victim
cache) and the previous commit (which ensures we don't incorrectly
hit in the victim cache)) this should never be possible.
Drop the check on TLB_INVALID_MASK from the "is this a TLB_RECHECK?"
condition, and instead assert that the tlb fill procedure has given
us a valid TLB entry (or longjumped out with a guest exception).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20180713141636.18665-3-peter.maydell@linaro.org
In get_page_addr_code(), we were incorrectly looking in the victim
TLB for an entry which matched the target address for reads, not
for code accesses. This meant that we could hit on a victim TLB
entry that indicated that the address was readable but not
executable, and incorrectly bypass the call to tlb_fill() which
should generate the guest MMU exception. Fix this bug.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20180713141636.18665-2-peter.maydell@linaro.org
The typo was found by codespell.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Message-Id: <20180712194454.26765-1-sw@weilnetz.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This fixes a record-replay regression introduced by 95590e2
("translate-all: discard TB when tb_link_page returns an existing
matching TB", 2018-06-15). The problem is that code using CF_NOCACHE
assumes that the TB returned from tb_gen_code is always a
newly-generated one. This assumption, however, was broken in
the aforementioned commit.
Fix it by honouring CF_NOCACHE, so that tb_gen_code always
returns a newly-generated TB when CF_NOCACHE is passed to it.
Do this by avoiding the TB hash table if CF_NOCACHE is set.
Reported-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
Tested-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Alistair Francis <alistair.francis@wdc.com>
Message-id: 1530806837-5416-1-git-send-email-cota@braap.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
When installing a TLB entry, remove any cached version of the
same page in the VTLB. If the existing TLB entry matches, do
not copy into the VTLB, but overwrite it.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
In get_page_addr_code() when we check whether the TLB entry
is marked as TLB_RECHECK, we should not go down that code
path if the TLB entry is not valid at all (ie the TLB_INVALID
bit is set).
Tested-by: Laurent Vivier <laurent@vivier.eu>
Reported-by: Laurent Vivier <laurent@vivier.eu>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20180629161731.16239-1-peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
In commit 71b9a45330 we changed the condition we use
to determine whether we need to refill the TLB in
get_page_addr_code() to
if (unlikely(env->tlb_table[mmu_idx][index].addr_code !=
(addr & (TARGET_PAGE_MASK | TLB_INVALID_MASK)))) {
This isn't the right check (it will falsely fail if the
input addr happens to have the low bit corresponding to
TLB_INVALID_MASK set, for instance). Replace it with a
use of the new tlb_hit() function, which is the correct test.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20180629162122.19376-3-peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The condition to check whether an address has hit against a particular
TLB entry is not completely trivial. We do this in various places, and
in fact in one place (get_page_addr_code()) we have got the condition
wrong. Abstract it out into new tlb_hit() and tlb_hit_page() inline
functions (one for a known-page-aligned address and one for an
arbitrary address), and use them in all the places where we had the
condition correct.
This is a no-behaviour-change patch; we leave fixing the buggy
code in get_page_addr_code() to a subsequent patch.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20180629162122.19376-2-peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Commit 0b5c91f ("translate-all: use per-page locking in !user-mode",
2018-06-15) introduced per-page locking. It assumed that the physical
pages corresponding to a TB (at most two pages) are always distinct,
which is wrong. For instance, an xtensa test provided by Max Filippov
is broken by the commit, since the test maps two virtual pages
to the same physical page:
virt1: 7fff, virt2: 8000
phys1 6000fff, phys2 6000000
Fix it by removing the assumption from page_lock_pair.
If the two physical page addresses are equal, we only lock
the PageDesc once. Note that the two callers of page_lock_pair,
namely page_unlock_tb and tb_link_page, are also updated so that
we do not try to unlock the same PageDesc twice.
Fixes: 0b5c91f74f
Reported-by: Max Filippov <jcmvbkbc@gmail.com>
Tested-by: Max Filippov <jcmvbkbc@gmail.com>
Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <1529944302-14186-1-git-send-email-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
When guest CPU PM is enabled, and with -cpu host, expose the host CPU
MWAIT leaf in the CPUID so guest can make good PM decisions.
Note: the result is 100% CPU utilization reported by host as host
no longer knows that the CPU is halted.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20180622192148.178309-3-mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Place them in exec.c, exec-all.h and ram_addr.h. This removes
knowledge of translate-all.h (which is an internal header) from
several files outside accel/tcg and removes knowledge of
AddressSpace from translate-all.c (as it only operates on ram_addr_t).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Determining the size of a field is useful when you don't have a struct
variable handy. Open-coding this is ugly.
This patch adds the sizeof_field() macro, which is similar to
typeof_field(). Existing instances are updated to use the macro.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20180614164431.29305-1-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
We do not trace guest atomic accesses. Fix it.
Tested with a modified atomic_add-bench so that it executes
a deterministic number of instructions, i.e. fixed seeding,
no threading and fixed number of loop iterations instead
of running for a certain time.
Before:
- With parallel_cpus = false (no clone syscall so it is never set to true):
220070 memory accesses
- With parallel_cpus = true (hard-coded):
212105 memory accesses <-- we're not tracing the atomics!
After:
220070 memory accesses regardless of parallel_cpus.
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-id: 1527028012-21888-6-git-send-email-cota@braap.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Add support for MMU protection regions that are smaller than
TARGET_PAGE_SIZE. We do this by marking the TLB entry for those
pages with a flag TLB_RECHECK. This flag causes us to always
take the slow-path for accesses. In the slow path we can then
special case them to always call tlb_fill() again, so we have
the correct information for the exact address being accessed.
This change allows us to handle reading and writing from small
regions; we cannot deal with execution from the small region.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20180620130619.11362-2-peter.maydell@linaro.org
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>
tb_lock was needed when the function did retranslation. However,
since fca8a500d5 ("tcg: Save insn data and use it in
cpu_restore_state_from_tb") we don't do retranslation.
Get rid of the comment.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The acquisition of tb_lock was added when the async tlb_flush
was introduced in e3b9ca810 ("cputlb: introduce tlb_flush_* async work.")
tb_lock was there to allow us to do memset() on the tb_jmp_cache's.
However, since f3ced3c592 ("tcg: consistently access cpu->tb_jmp_cache
atomically") all accesses to tb_jmp_cache are atomic, so tb_lock
is not needed here. Get rid of it.
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This applies to both user-mode and !user-mode emulation.
Instead of relying on a global lock, protect the list of incoming
jumps with tb->jmp_lock. This lock also protects tb->cflags,
so update all tb->cflags readers outside tb->jmp_lock to use
atomic reads via tb_cflags().
In order to find the destination TB (and therefore its jmp_lock)
from the origin TB, we introduce tb->jmp_dest[].
I considered not using a linked list of jumps, which simplifies
code and makes the struct smaller. However, it unnecessarily increases
memory usage, which results in a performance decrease. See for
instance these numbers booting+shutting down debian-arm:
Time (s) Rel. err (%) Abs. err (s) Rel. slowdown (%)
------------------------------------------------------------------------------
before 20.88 0.74 0.154512 0.
after 20.81 0.38 0.079078 -0.33524904
GTree 21.02 0.28 0.058856 0.67049808
GHashTable + xxhash 21.63 1.08 0.233604 3.5919540
Using a hash table or a binary tree to keep track of the jumps
doesn't really pay off, not only due to the increased memory usage,
but also because most TBs have only 0 or 1 jumps to them. The maximum
number of jumps when booting debian-arm that I measured is 35, but
as we can see in the histogram below a TB with that many incoming jumps
is extremely rare; the average TB has 0.80 incoming jumps.
n_jumps: 379208; avg jumps/tb: 0.801099
dist: [0.0,1.0)|▄█▁▁▁▁▁▁▁▁▁▁▁ ▁▁▁▁▁▁ ▁▁▁ ▁▁▁ ▁|[34.0,35.0]
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
