OpenE2K
/
qemu-e2k
13
4
Fork 0
Code Issues 4 Pull Requests Projects 2 Releases Activity

Clean up code_gen_buffer allocation. 

Add tcg_remove_ops_after.
 Fix tcg_constant_* documentation.
 Improve TB chaining documentation.
 Fix float32_exp2.
 Fix arm tcg_out_op function signature.
 -----BEGIN PGP SIGNATURE-----
 
 iQFRBAABCgA7FiEEekgeeIaLTbaoWgXAZN846K9+IV8FAmDGrQMdHHJpY2hhcmQu
 aGVuZGVyc29uQGxpbmFyby5vcmcACgkQZN846K9+IV9kZgf+LSKbkimQKNGJNpFp
 xdMG2B0681tHyO7aiDHZqZf9Izeey7x9vGZRZzPfxomdN8qYT2PiklNx2yIIxNRt
 WdJ3e7+l7cYjAGY6HdrTqJ6ZiqOOftMzMuHWiXiHD0rMYTIjXgAfsf+H+lVRwMCR
 BbQBB1ttUJzDSLkM5B2rNuWEjup4shAMgijkipkqkaWrzZIAvfIkcfScZyqWFguG
 GoWnQxIHq7XMbveUX1Tu1JcdVZlXmuMl0LMQ8Qj5sbep1gjyYixuV6lWupp6SPX9
 quRBsyVSmAa4frK67huJ4WVq4gZ2VylNvpiwjwoChYgJ8TOU73n7KGZOAl6i0iq2
 ytR6Pw==
 =ft63
 -----END PGP SIGNATURE-----

Merge remote-tracking branch 'remotes/rth-gitlab/tags/pull-tcg-20210613' into staging

Clean up code_gen_buffer allocation.
Add tcg_remove_ops_after.
Fix tcg_constant_* documentation.
Improve TB chaining documentation.
Fix float32_exp2.
Fix arm tcg_out_op function signature.

# gpg: Signature made Mon 14 Jun 2021 02:12:35 BST
# gpg:                using RSA key 7A481E78868B4DB6A85A05C064DF38E8AF7E215F
# gpg:                issuer "richard.henderson@linaro.org"
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>" [full]
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth-gitlab/tags/pull-tcg-20210613: (34 commits)
  docs/devel: Explain in more detail the TB chaining mechanisms
  softfloat: Fix tp init in float32_exp2
  tcg/arm: Fix tcg_out_op function signature
  tcg: Fix documentation for tcg_constant_* vs tcg_temp_free_*
  tcg: Introduce tcg_remove_ops_after
  tcg: Move tcg_init_ctx and tcg_ctx from accel/tcg/
  tcg: When allocating for !splitwx, begin with PROT_NONE
  tcg: Merge buffer protection and guard page protection
  tcg: Round the tb_size default from qemu_get_host_physmem
  util/osdep: Add qemu_mprotect_rw
  tcg: Sink qemu_madvise call to common code
  tcg: Return the map protection from alloc_code_gen_buffer
  tcg: Allocate code_gen_buffer into struct tcg_region_state
  tcg: Move in_code_gen_buffer and tests to region.c
  tcg: Tidy split_cross_256mb
  tcg: Tidy tcg_n_regions
  tcg: Rename region.start to region.after_prologue
  tcg: Replace region.end with region.total_size
  tcg: Move MAX_CODE_GEN_BUFFER_SIZE to tcg-target.h
  tcg: Introduce tcg_max_ctxs
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This commit is contained in:
Peter Maydell 2021-06-14 10:25:37 +01:00
parent 894fc4fd67 a5a8b84772
commit fbe7919ece
27 changed files with 1266 additions and 1090 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
accel/tcg/internal.h
View File

@ -16,5 +16,7 @@ TranslationBlock *tb_gen_code(CPUState *cpu, target_ulong pc,
int cflags);
void QEMU_NORETURN cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
void page_init(void);
void tb_htable_init(void);
#endif /* ACCEL_TCG_INTERNAL_H */

32
accel/tcg/tcg-all.c
View File

@ -32,6 +32,11 @@
#include "qemu/error-report.h"
#include "qemu/accel.h"
#include "qapi/qapi-builtin-visit.h"
#include "qemu/units.h"
#if !defined(CONFIG_USER_ONLY)
#include "hw/boards.h"
#endif
#include "internal.h"
struct TCGState {
AccelState parent_obj;
@ -105,22 +110,29 @@ static void tcg_accel_instance_init(Object *obj)
bool mttcg_enabled;
static int tcg_init(MachineState *ms)
static int tcg_init_machine(MachineState *ms)
{
TCGState *s = TCG_STATE(current_accel());
#ifdef CONFIG_USER_ONLY
unsigned max_cpus = 1;
#else
unsigned max_cpus = ms->smp.max_cpus;
#endif
tcg_exec_init(s->tb_size * 1024 * 1024, s->splitwx_enabled);
tcg_allowed = true;
mttcg_enabled = s->mttcg_enabled;
page_init();
tb_htable_init();
tcg_init(s->tb_size * MiB, s->splitwx_enabled, max_cpus);
#if defined(CONFIG_SOFTMMU)
/*
* Initialize TCG regions only for softmmu.
*
* This needs to be done later for user mode, because the prologue
* generation needs to be delayed so that GUEST_BASE is already set.
* There's no guest base to take into account, so go ahead and
* initialize the prologue now.
*/
#ifndef CONFIG_USER_ONLY
tcg_region_init();
#endif /* !CONFIG_USER_ONLY */
tcg_prologue_init(tcg_ctx);
#endif
return 0;
}
@ -200,7 +212,7 @@ static void tcg_accel_class_init(ObjectClass *oc, void *data)
{
AccelClass *ac = ACCEL_CLASS(oc);
ac->name = "tcg";
ac->init_machine = tcg_init;
ac->init_machine = tcg_init_machine;
ac->allowed = &tcg_allowed;
object_class_property_add_str(oc, "thread",

439
accel/tcg/translate-all.c
View File

@ -18,7 +18,6 @@
*/
#include "qemu/osdep.h"
#include "qemu/units.h"
#include "qemu-common.h"
#define NO_CPU_IO_DEFS
@ -49,7 +48,6 @@
#include "exec/cputlb.h"
#include "exec/translate-all.h"
#include "qemu/bitmap.h"
#include "qemu/error-report.h"
#include "qemu/qemu-print.h"
#include "qemu/timer.h"
#include "qemu/main-loop.h"
@ -220,9 +218,6 @@ static int v_l2_levels;
static void *l1_map[V_L1_MAX_SIZE];
/* code generation context */
TCGContext tcg_init_ctx;
__thread TCGContext *tcg_ctx;
TBContext tb_ctx;
static void page_table_config_init(void)
@ -245,11 +240,6 @@ static void page_table_config_init(void)
assert(v_l2_levels >= 0);
}
static void cpu_gen_init(void)
{
tcg_context_init(&tcg_init_ctx);
}
/* Encode VAL as a signed leb128 sequence at P.
Return P incremented past the encoded value. */
static uint8_t *encode_sleb128(uint8_t *p, target_long val)
@ -415,7 +405,7 @@ bool cpu_restore_state(CPUState *cpu, uintptr_t host_pc, bool will_exit)
return false;
}
static void page_init(void)
void page_init(void)
{
page_size_init();
page_table_config_init();
@ -900,408 +890,6 @@ static void page_lock_pair(PageDesc **ret_p1, tb_page_addr_t phys1,
}
}
/* Minimum size of the code gen buffer. This number is randomly chosen,
but not so small that we can't have a fair number of TB's live. */
#define MIN_CODE_GEN_BUFFER_SIZE (1 * MiB)
/* Maximum size of the code gen buffer we'd like to use. Unless otherwise
indicated, this is constrained by the range of direct branches on the
host cpu, as used by the TCG implementation of goto_tb. */
#if defined(__x86_64__)
# define MAX_CODE_GEN_BUFFER_SIZE (2 * GiB)
#elif defined(__sparc__)
# define MAX_CODE_GEN_BUFFER_SIZE (2 * GiB)
#elif defined(__powerpc64__)
# define MAX_CODE_GEN_BUFFER_SIZE (2 * GiB)
#elif defined(__powerpc__)
# define MAX_CODE_GEN_BUFFER_SIZE (32 * MiB)
#elif defined(__aarch64__)
# define MAX_CODE_GEN_BUFFER_SIZE (2 * GiB)
#elif defined(__s390x__)
/* We have a +- 4GB range on the branches; leave some slop. */
# define MAX_CODE_GEN_BUFFER_SIZE (3 * GiB)
#elif defined(__mips__)
/* We have a 256MB branch region, but leave room to make sure the
main executable is also within that region. */
# define MAX_CODE_GEN_BUFFER_SIZE (128 * MiB)
#else
# define MAX_CODE_GEN_BUFFER_SIZE ((size_t)-1)
#endif
#if TCG_TARGET_REG_BITS == 32
#define DEFAULT_CODE_GEN_BUFFER_SIZE_1 (32 * MiB)
#ifdef CONFIG_USER_ONLY
/*
* For user mode on smaller 32 bit systems we may run into trouble
* allocating big chunks of data in the right place. On these systems
* we utilise a static code generation buffer directly in the binary.
*/
#define USE_STATIC_CODE_GEN_BUFFER
#endif
#else /* TCG_TARGET_REG_BITS == 64 */
#ifdef CONFIG_USER_ONLY
/*
* As user-mode emulation typically means running multiple instances
* of the translator don't go too nuts with our default code gen
* buffer lest we make things too hard for the OS.
*/
#define DEFAULT_CODE_GEN_BUFFER_SIZE_1 (128 * MiB)
#else
/*
* We expect most system emulation to run one or two guests per host.
* Users running large scale system emulation may want to tweak their
* runtime setup via the tb-size control on the command line.
*/
#define DEFAULT_CODE_GEN_BUFFER_SIZE_1 (1 * GiB)
#endif
#endif
#define DEFAULT_CODE_GEN_BUFFER_SIZE \
(DEFAULT_CODE_GEN_BUFFER_SIZE_1 < MAX_CODE_GEN_BUFFER_SIZE \
? DEFAULT_CODE_GEN_BUFFER_SIZE_1 : MAX_CODE_GEN_BUFFER_SIZE)
static size_t size_code_gen_buffer(size_t tb_size)
{
/* Size the buffer. */
if (tb_size == 0) {
size_t phys_mem = qemu_get_host_physmem();
if (phys_mem == 0) {
tb_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
} else {
tb_size = MIN(DEFAULT_CODE_GEN_BUFFER_SIZE, phys_mem / 8);
}
}
if (tb_size < MIN_CODE_GEN_BUFFER_SIZE) {
tb_size = MIN_CODE_GEN_BUFFER_SIZE;
}
if (tb_size > MAX_CODE_GEN_BUFFER_SIZE) {
tb_size = MAX_CODE_GEN_BUFFER_SIZE;
}
return tb_size;
}
#ifdef __mips__
/* In order to use J and JAL within the code_gen_buffer, we require
that the buffer not cross a 256MB boundary. */
static inline bool cross_256mb(void *addr, size_t size)
{
return ((uintptr_t)addr ^ ((uintptr_t)addr + size)) & ~0x0ffffffful;
}
/* We weren't able to allocate a buffer without crossing that boundary,
so make do with the larger portion of the buffer that doesn't cross.
Returns the new base of the buffer, and adjusts code_gen_buffer_size. */
static inline void *split_cross_256mb(void *buf1, size_t size1)
{
void *buf2 = (void *)(((uintptr_t)buf1 + size1) & ~0x0ffffffful);
size_t size2 = buf1 + size1 - buf2;
size1 = buf2 - buf1;
if (size1 < size2) {
size1 = size2;
buf1 = buf2;
}
tcg_ctx->code_gen_buffer_size = size1;
return buf1;
}
#endif
#ifdef USE_STATIC_CODE_GEN_BUFFER
static uint8_t static_code_gen_buffer[DEFAULT_CODE_GEN_BUFFER_SIZE]
__attribute__((aligned(CODE_GEN_ALIGN)));
static bool alloc_code_gen_buffer(size_t tb_size, int splitwx, Error **errp)
{
void *buf, *end;
size_t size;
if (splitwx > 0) {
error_setg(errp, "jit split-wx not supported");
return false;
}
/* page-align the beginning and end of the buffer */
buf = static_code_gen_buffer;
end = static_code_gen_buffer + sizeof(static_code_gen_buffer);
buf = QEMU_ALIGN_PTR_UP(buf, qemu_real_host_page_size);
end = QEMU_ALIGN_PTR_DOWN(end, qemu_real_host_page_size);
size = end - buf;
/* Honor a command-line option limiting the size of the buffer. */
if (size > tb_size) {
size = QEMU_ALIGN_DOWN(tb_size, qemu_real_host_page_size);
}
tcg_ctx->code_gen_buffer_size = size;
#ifdef __mips__
if (cross_256mb(buf, size)) {
buf = split_cross_256mb(buf, size);
size = tcg_ctx->code_gen_buffer_size;
}
#endif
if (qemu_mprotect_rwx(buf, size)) {
error_setg_errno(errp, errno, "mprotect of jit buffer");
return false;
}
qemu_madvise(buf, size, QEMU_MADV_HUGEPAGE);
tcg_ctx->code_gen_buffer = buf;
return true;
}
#elif defined(_WIN32)
static bool alloc_code_gen_buffer(size_t size, int splitwx, Error **errp)
{
void *buf;
if (splitwx > 0) {
error_setg(errp, "jit split-wx not supported");
return false;
}
buf = VirtualAlloc(NULL, size, MEM_RESERVE | MEM_COMMIT,
PAGE_EXECUTE_READWRITE);
if (buf == NULL) {
error_setg_win32(errp, GetLastError(),
"allocate %zu bytes for jit buffer", size);
return false;
}
tcg_ctx->code_gen_buffer = buf;
tcg_ctx->code_gen_buffer_size = size;
return true;
}
#else
static bool alloc_code_gen_buffer_anon(size_t size, int prot,
int flags, Error **errp)
{
void *buf;
buf = mmap(NULL, size, prot, flags, -1, 0);
if (buf == MAP_FAILED) {
error_setg_errno(errp, errno,
"allocate %zu bytes for jit buffer", size);
return false;
}
tcg_ctx->code_gen_buffer_size = size;
#ifdef __mips__
if (cross_256mb(buf, size)) {
/*
* Try again, with the original still mapped, to avoid re-acquiring
* the same 256mb crossing.
*/
size_t size2;
void *buf2 = mmap(NULL, size, prot, flags, -1, 0);
switch ((int)(buf2 != MAP_FAILED)) {
case 1:
if (!cross_256mb(buf2, size)) {
/* Success! Use the new buffer. */
munmap(buf, size);
break;
}
/* Failure. Work with what we had. */
munmap(buf2, size);
/* fallthru */
default:
/* Split the original buffer. Free the smaller half. */
buf2 = split_cross_256mb(buf, size);
size2 = tcg_ctx->code_gen_buffer_size;
if (buf == buf2) {
munmap(buf + size2, size - size2);
} else {
munmap(buf, size - size2);
}
size = size2;
break;
}
buf = buf2;
}
#endif
/* Request large pages for the buffer. */
qemu_madvise(buf, size, QEMU_MADV_HUGEPAGE);
tcg_ctx->code_gen_buffer = buf;
return true;
}
#ifndef CONFIG_TCG_INTERPRETER
#ifdef CONFIG_POSIX
#include "qemu/memfd.h"
static bool alloc_code_gen_buffer_splitwx_memfd(size_t size, Error **errp)
{
void *buf_rw = NULL, *buf_rx = MAP_FAILED;
int fd = -1;
#ifdef __mips__
/* Find space for the RX mapping, vs the 256MiB regions. */
if (!alloc_code_gen_buffer_anon(size, PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS |
MAP_NORESERVE, errp)) {
return false;
}
/* The size of the mapping may have been adjusted. */
size = tcg_ctx->code_gen_buffer_size;
buf_rx = tcg_ctx->code_gen_buffer;
#endif
buf_rw = qemu_memfd_alloc("tcg-jit", size, 0, &fd, errp);
if (buf_rw == NULL) {
goto fail;
}
#ifdef __mips__
void *tmp = mmap(buf_rx, size, PROT_READ | PROT_EXEC,
MAP_SHARED | MAP_FIXED, fd, 0);
if (tmp != buf_rx) {
goto fail_rx;
}
#else
buf_rx = mmap(NULL, size, PROT_READ | PROT_EXEC, MAP_SHARED, fd, 0);
if (buf_rx == MAP_FAILED) {
goto fail_rx;
}
#endif
close(fd);
tcg_ctx->code_gen_buffer = buf_rw;
tcg_ctx->code_gen_buffer_size = size;
tcg_splitwx_diff = buf_rx - buf_rw;
/* Request large pages for the buffer and the splitwx. */
qemu_madvise(buf_rw, size, QEMU_MADV_HUGEPAGE);
qemu_madvise(buf_rx, size, QEMU_MADV_HUGEPAGE);
return true;
fail_rx:
error_setg_errno(errp, errno, "failed to map shared memory for execute");
fail:
if (buf_rx != MAP_FAILED) {
munmap(buf_rx, size);
}
if (buf_rw) {
munmap(buf_rw, size);
}
if (fd >= 0) {
close(fd);
}
return false;
}
#endif /* CONFIG_POSIX */
#ifdef CONFIG_DARWIN
#include <mach/mach.h>
extern kern_return_t mach_vm_remap(vm_map_t target_task,
mach_vm_address_t *target_address,
mach_vm_size_t size,
mach_vm_offset_t mask,
int flags,
vm_map_t src_task,
mach_vm_address_t src_address,
boolean_t copy,
vm_prot_t *cur_protection,
vm_prot_t *max_protection,
vm_inherit_t inheritance);
static bool alloc_code_gen_buffer_splitwx_vmremap(size_t size, Error **errp)
{
kern_return_t ret;
mach_vm_address_t buf_rw, buf_rx;
vm_prot_t cur_prot, max_prot;
/* Map the read-write portion via normal anon memory. */
if (!alloc_code_gen_buffer_anon(size, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, errp)) {
return false;
}
buf_rw = (mach_vm_address_t)tcg_ctx->code_gen_buffer;
buf_rx = 0;
ret = mach_vm_remap(mach_task_self(),
&buf_rx,
size,
0,
VM_FLAGS_ANYWHERE,
mach_task_self(),
buf_rw,
false,
&cur_prot,
&max_prot,
VM_INHERIT_NONE);
if (ret != KERN_SUCCESS) {
/* TODO: Convert "ret" to a human readable error message. */
error_setg(errp, "vm_remap for jit splitwx failed");
munmap((void *)buf_rw, size);
return false;
}
if (mprotect((void *)buf_rx, size, PROT_READ | PROT_EXEC) != 0) {
error_setg_errno(errp, errno, "mprotect for jit splitwx");
munmap((void *)buf_rx, size);
munmap((void *)buf_rw, size);
return false;
}
tcg_splitwx_diff = buf_rx - buf_rw;
return true;
}
#endif /* CONFIG_DARWIN */
#endif /* CONFIG_TCG_INTERPRETER */
static bool alloc_code_gen_buffer_splitwx(size_t size, Error **errp)
{
#ifndef CONFIG_TCG_INTERPRETER
# ifdef CONFIG_DARWIN
return alloc_code_gen_buffer_splitwx_vmremap(size, errp);
# endif
# ifdef CONFIG_POSIX
return alloc_code_gen_buffer_splitwx_memfd(size, errp);
# endif
#endif
error_setg(errp, "jit split-wx not supported");
return false;
}
static bool alloc_code_gen_buffer(size_t size, int splitwx, Error **errp)
{
ERRP_GUARD();
int prot, flags;
if (splitwx) {
if (alloc_code_gen_buffer_splitwx(size, errp)) {
return true;
}
/*
* If splitwx force-on (1), fail;
* if splitwx default-on (-1), fall through to splitwx off.
*/
if (splitwx > 0) {
return false;
}
error_free_or_abort(errp);
}
prot = PROT_READ | PROT_WRITE | PROT_EXEC;
flags = MAP_PRIVATE | MAP_ANONYMOUS;
#ifdef CONFIG_TCG_INTERPRETER
/* The tcg interpreter does not need execute permission. */
prot = PROT_READ | PROT_WRITE;
#elif defined(CONFIG_DARWIN)
/* Applicable to both iOS and macOS (Apple Silicon). */
if (!splitwx) {
flags |= MAP_JIT;
}
#endif
return alloc_code_gen_buffer_anon(size, prot, flags, errp);
}
#endif /* USE_STATIC_CODE_GEN_BUFFER, WIN32, POSIX */
static bool tb_cmp(const void *ap, const void *bp)
{
const TranslationBlock *a = ap;
@ -1316,36 +904,13 @@ static bool tb_cmp(const void *ap, const void *bp)
a->page_addr[1] == b->page_addr[1];
}
static void tb_htable_init(void)
void tb_htable_init(void)
{
unsigned int mode = QHT_MODE_AUTO_RESIZE;
qht_init(&tb_ctx.htable, tb_cmp, CODE_GEN_HTABLE_SIZE, mode);
}
/* Must be called before using the QEMU cpus. 'tb_size' is the size
(in bytes) allocated to the translation buffer. Zero means default
size. */
void tcg_exec_init(unsigned long tb_size, int splitwx)
{
bool ok;
tcg_allowed = true;
cpu_gen_init();
page_init();
tb_htable_init();
ok = alloc_code_gen_buffer(size_code_gen_buffer(tb_size),
splitwx, &error_fatal);
assert(ok);
#if defined(CONFIG_SOFTMMU)
/* There's no guest base to take into account, so go ahead and
initialize the prologue now. */
tcg_prologue_init(tcg_ctx);
#endif
}
/* call with @p->lock held */
static inline void invalidate_page_bitmap(PageDesc *p)
{

3
bsd-user/main.c
View File

@ -813,7 +813,7 @@ int main(int argc, char **argv)
envlist_free(envlist);
/*
* Now that page sizes are configured in tcg_exec_init() we can do
* Now that page sizes are configured we can do
* proper page alignment for guest_base.
*/
guest_base = HOST_PAGE_ALIGN(guest_base);
@ -879,7 +879,6 @@ int main(int argc, char **argv)
* the real value of GUEST_BASE into account.
*/
tcg_prologue_init(tcg_ctx);
tcg_region_init();
/* build Task State */
memset(ts, 0, sizeof(TaskState));

101
docs/devel/tcg.rst
View File

@ -11,13 +11,14 @@ performances.
QEMU's dynamic translation backend is called TCG, for "Tiny Code
Generator". For more information, please take a look at ``tcg/README``.
Some notable features of QEMU's dynamic translator are:
The following sections outline some notable features and implementation
details of QEMU's dynamic translator.
CPU state optimisations
-----------------------
The target CPUs have many internal states which change the way it
evaluates instructions. In order to achieve a good speed, the
The target CPUs have many internal states which change the way they
evaluate instructions. In order to achieve a good speed, the
translation phase considers that some state information of the virtual
CPU cannot change in it. The state is recorded in the Translation
Block (TB). If the state changes (e.g. privilege level), a new TB will
@ -31,17 +32,95 @@ Direct block chaining
---------------------
After each translated basic block is executed, QEMU uses the simulated
Program Counter (PC) and other cpu state information (such as the CS
Program Counter (PC) and other CPU state information (such as the CS
segment base value) to find the next basic block.
In order to accelerate the most common cases where the new simulated PC
is known, QEMU can patch a basic block so that it jumps directly to the
next one.
In its simplest, less optimized form, this is done by exiting from the
current TB, going through the TB epilogue, and then back to the
main loop. That’s where QEMU looks for the next TB to execute,
translating it from the guest architecture if it isn’t already available
in memory. Then QEMU proceeds to execute this next TB, starting at the
prologue and then moving on to the translated instructions.
The most portable code uses an indirect jump. An indirect jump makes
it easier to make the jump target modification atomic. On some host
architectures (such as x86 or PowerPC), the ``JUMP`` opcode is
directly patched so that the block chaining has no overhead.
Exiting from the TB this way will cause the ``cpu_exec_interrupt()``
callback to be re-evaluated before executing additional instructions.
It is mandatory to exit this way after any CPU state changes that may
unmask interrupts.
In order to accelerate the cases where the TB for the new
simulated PC is already available, QEMU has mechanisms that allow
multiple TBs to be chained directly, without having to go back to the
main loop as described above. These mechanisms are:
``lookup_and_goto_ptr``
^^^^^^^^^^^^^^^^^^^^^^^
Calling ``tcg_gen_lookup_and_goto_ptr()`` will emit a call to
``helper_lookup_tb_ptr``. This helper will look for an existing TB that
matches the current CPU state. If the destination TB is available its
code address is returned, otherwise the address of the JIT epilogue is
returned. The call to the helper is always followed by the tcg ``goto_ptr``
opcode, which branches to the returned address. In this way, we either
branch to the next TB or return to the main loop.
``goto_tb + exit_tb``
^^^^^^^^^^^^^^^^^^^^^
The translation code usually implements branching by performing the
following steps:
1. Call ``tcg_gen_goto_tb()`` passing a jump slot index (either 0 or 1)
as a parameter.
2. Emit TCG instructions to update the CPU state with any information
that has been assumed constant and is required by the main loop to
correctly locate and execute the next TB. For most guests, this is
just the PC of the branch destination, but others may store additional
data. The information updated in this step must be inferable from both
``cpu_get_tb_cpu_state()`` and ``cpu_restore_state()``.
3. Call ``tcg_gen_exit_tb()`` passing the address of the current TB and
the jump slot index again.
Step 1, ``tcg_gen_goto_tb()``, will emit a ``goto_tb`` TCG
instruction that later on gets translated to a jump to an address
associated with the specified jump slot. Initially, this is the address
of step 2's instructions, which update the CPU state information. Step 3,
``tcg_gen_exit_tb()``, exits from the current TB returning a tagged
pointer composed of the last executed TB’s address and the jump slot
index.
The first time this whole sequence is executed, step 1 simply jumps
to step 2. Then the CPU state information gets updated and we exit from
the current TB. As a result, the behavior is very similar to the less
optimized form described earlier in this section.
Next, the main loop looks for the next TB to execute using the
current CPU state information (creating the TB if it wasn’t already
available) and, before starting to execute the new TB’s instructions,
patches the previously executed TB by associating one of its jump
slots (the one specified in the call to ``tcg_gen_exit_tb()``) with the
address of the new TB.
The next time this previous TB is executed and we get to that same
``goto_tb`` step, it will already be patched (assuming the destination TB
is still in memory) and will jump directly to the first instruction of
the destination TB, without going back to the main loop.
For the ``goto_tb + exit_tb`` mechanism to be used, the following
conditions need to be satisfied:
* The change in CPU state must be constant, e.g., a direct branch and
not an indirect branch.
* The direct branch cannot cross a page boundary. Memory mappings
may change, causing the code at the destination address to change.
Note that, on step 3 (``tcg_gen_exit_tb()``), in addition to the
jump slot index, the address of the TB just executed is also returned.
This address corresponds to the TB that will be patched; it may be
different than the one that was directly executed from the main loop
if the latter had already been chained to other TBs.
Self-modifying code and translated code invalidation
----------------------------------------------------

1
fpu/meson.build Normal file
View File

@ -0,0 +1 @@
specific_ss.add(when: 'CONFIG_TCG', if_true: files('softfloat.c'))

2
fpu/softfloat.c
View File

@ -4818,7 +4818,7 @@ float32 float32_exp2(float32 a, float_status *status)
float_raise(float_flag_inexact, status);
float64_unpack_canonical(&xnp, float64_ln2, status);
float64_unpack_canonical(&tp, float64_ln2, status);
xp = *parts_mul(&xp, &tp, status);
xnp = xp;

1
include/qemu/osdep.h
View File

@ -512,6 +512,7 @@ void sigaction_invoke(struct sigaction *action,
#endif
int qemu_madvise(void *addr, size_t len, int advice);
int qemu_mprotect_rw(void *addr, size_t size);
int qemu_mprotect_rwx(void *addr, size_t size);
int qemu_mprotect_none(void *addr, size_t size);

2
include/sysemu/tcg.h
View File

@ -8,8 +8,6 @@
#ifndef SYSEMU_TCG_H
#define SYSEMU_TCG_H
void tcg_exec_init(unsigned long tb_size, int splitwx);
#ifdef CONFIG_TCG
extern bool tcg_allowed;
#define tcg_enabled() (tcg_allowed)

28
include/tcg/tcg.h
View File

@ -689,22 +689,12 @@ static inline bool temp_readonly(TCGTemp *ts)
return ts->kind >= TEMP_FIXED;
}
extern TCGContext tcg_init_ctx;
extern __thread TCGContext *tcg_ctx;
extern const void *tcg_code_gen_epilogue;
extern uintptr_t tcg_splitwx_diff;
extern TCGv_env cpu_env;
static inline bool in_code_gen_buffer(const void *p)
{
const TCGContext *s = &tcg_init_ctx;
/*
* Much like it is valid to have a pointer to the byte past the
* end of an array (so long as you don't dereference it), allow
* a pointer to the byte past the end of the code gen buffer.
*/
return (size_t)(p - s->code_gen_buffer) <= s->code_gen_buffer_size;
}
bool in_code_gen_buffer(const void *p);
#ifdef CONFIG_DEBUG_TCG
const void *tcg_splitwx_to_rx(void *rw);
@ -873,7 +863,6 @@ void *tcg_malloc_internal(TCGContext *s, int size);
void tcg_pool_reset(TCGContext *s);
TranslationBlock *tcg_tb_alloc(TCGContext *s);
void tcg_region_init(void);
void tb_destroy(TranslationBlock *tb);
void tcg_region_reset_all(void);
@ -906,7 +895,7 @@ static inline void *tcg_malloc(int size)
}
}
void tcg_context_init(TCGContext *s);
void tcg_init(size_t tb_size, int splitwx, unsigned max_cpus);
void tcg_register_thread(void);
void tcg_prologue_init(TCGContext *s);
void tcg_func_start(TCGContext *s);
@ -1082,6 +1071,16 @@ void tcg_op_remove(TCGContext *s, TCGOp *op);
TCGOp *tcg_op_insert_before(TCGContext *s, TCGOp *op, TCGOpcode opc);
TCGOp *tcg_op_insert_after(TCGContext *s, TCGOp *op, TCGOpcode opc);
/**
* tcg_remove_ops_after:
* @op: target operation
*
* Discard any opcodes emitted since @op. Expected usage is to save
* a starting point with tcg_last_op(), speculatively emit opcodes,
* then decide whether or not to keep those opcodes after the fact.
*/
void tcg_remove_ops_after(TCGOp *op);
void tcg_optimize(TCGContext *s);
/* Allocate a new temporary and initialize it with a constant. */
@ -1096,7 +1095,8 @@ TCGv_vec tcg_const_ones_vec_matching(TCGv_vec);
/*
* Locate or create a read-only temporary that is a constant.
* This kind of temporary need not and should not be freed.
* This kind of temporary need not be freed, but for convenience
* will be silently ignored by tcg_temp_free_*.
*/
TCGTemp *tcg_constant_internal(TCGType type, int64_t val);

1
linux-user/main.c
View File

@ -868,7 +868,6 @@ int main(int argc, char **argv, char **envp)
generating the prologue until now so that the prologue can take
the real value of GUEST_BASE into account. */
tcg_prologue_init(tcg_ctx);
tcg_region_init();
target_cpu_copy_regs(env, regs);

12
meson.build
View File

@ -1968,16 +1968,6 @@ subdir('softmmu')
common_ss.add(capstone)
specific_ss.add(files('cpu.c', 'disas.c', 'gdbstub.c'), capstone)
specific_ss.add(when: 'CONFIG_TCG', if_true: files(
'fpu/softfloat.c',
'tcg/optimize.c',
'tcg/tcg-common.c',
'tcg/tcg-op-gvec.c',
'tcg/tcg-op-vec.c',
'tcg/tcg-op.c',
'tcg/tcg.c',
))
specific_ss.add(when: 'CONFIG_TCG_INTERPRETER', if_true: files('tcg/tci.c'))
# Work around a gcc bug/misfeature wherein constant propagation looks
# through an alias:
@ -2007,6 +1997,8 @@ subdir('net')
subdir('replay')
subdir('semihosting')
subdir('hw')
subdir('tcg')
subdir('fpu')
subdir('accel')
subdir('plugins')
subdir('bsd-user')

1
tcg/aarch64/tcg-target.h
View File

@ -15,6 +15,7 @@
#define TCG_TARGET_INSN_UNIT_SIZE 4
#define TCG_TARGET_TLB_DISPLACEMENT_BITS 24
#define MAX_CODE_GEN_BUFFER_SIZE (2 * GiB)
#undef TCG_TARGET_STACK_GROWSUP
typedef enum {

3
tcg/arm/tcg-target.c.inc
View File

@ -1984,7 +1984,8 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is64)
static void tcg_out_epilogue(TCGContext *s);
static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
const TCGArg *args, const int *const_args)
const TCGArg args[TCG_MAX_OP_ARGS],
const int const_args[TCG_MAX_OP_ARGS])
{
TCGArg a0, a1, a2, a3, a4, a5;
int c;

1
tcg/arm/tcg-target.h
View File

@ -60,6 +60,7 @@ extern int arm_arch;
#undef TCG_TARGET_STACK_GROWSUP
#define TCG_TARGET_INSN_UNIT_SIZE 4
#define TCG_TARGET_TLB_DISPLACEMENT_BITS 16
#define MAX_CODE_GEN_BUFFER_SIZE UINT32_MAX
typedef enum {
TCG_REG_R0 = 0,

2
tcg/i386/tcg-target.h
View File

@ -31,9 +31,11 @@
#ifdef __x86_64__
# define TCG_TARGET_REG_BITS 64
# define TCG_TARGET_NB_REGS 32
# define MAX_CODE_GEN_BUFFER_SIZE (2 * GiB)
#else
# define TCG_TARGET_REG_BITS 32
# define TCG_TARGET_NB_REGS 24
# define MAX_CODE_GEN_BUFFER_SIZE UINT32_MAX
#endif
typedef enum {

14
tcg/meson.build Normal file
View File

@ -0,0 +1,14 @@
tcg_ss = ss.source_set()
tcg_ss.add(files(
'optimize.c',
'region.c',
'tcg.c',
'tcg-common.c',
'tcg-op.c',
'tcg-op-gvec.c',
'tcg-op-vec.c',
))
tcg_ss.add(when: 'CONFIG_TCG_INTERPRETER', if_true: files('tci.c'))
specific_ss.add_all(when: 'CONFIG_TCG', if_true: tcg_ss)

6
tcg/mips/tcg-target.h
View File

@ -39,6 +39,12 @@
#define TCG_TARGET_TLB_DISPLACEMENT_BITS 16
#define TCG_TARGET_NB_REGS 32
/*
* We have a 256MB branch region, but leave room to make sure the
* main executable is also within that region.
*/
#define MAX_CODE_GEN_BUFFER_SIZE (128 * MiB)
typedef enum {
TCG_REG_ZERO = 0,
TCG_REG_AT,

2
tcg/ppc/tcg-target.h
View File

@ -27,8 +27,10 @@
#ifdef _ARCH_PPC64
# define TCG_TARGET_REG_BITS 64
# define MAX_CODE_GEN_BUFFER_SIZE (2 * GiB)
#else
# define TCG_TARGET_REG_BITS 32
# define MAX_CODE_GEN_BUFFER_SIZE (32 * MiB)
#endif
#define TCG_TARGET_NB_REGS 64

999
tcg/region.c Normal file
View File

@ -0,0 +1,999 @@
/*
* Memory region management for Tiny Code Generator for QEMU
*
* Copyright (c) 2008 Fabrice Bellard
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include "qemu/osdep.h"
#include "qemu/units.h"
#include "qapi/error.h"
#include "exec/exec-all.h"
#include "tcg/tcg.h"
#include "tcg-internal.h"
struct tcg_region_tree {
QemuMutex lock;
GTree *tree;
/* padding to avoid false sharing is computed at run-time */
};
/*
* We divide code_gen_buffer into equally-sized "regions" that TCG threads
* dynamically allocate from as demand dictates. Given appropriate region
* sizing, this minimizes flushes even when some TCG threads generate a lot
* more code than others.
*/
struct tcg_region_state {
QemuMutex lock;
/* fields set at init time */
void *start_aligned;
void *after_prologue;
size_t n;
size_t size; /* size of one region */
size_t stride; /* .size + guard size */
size_t total_size; /* size of entire buffer, >= n * stride */
/* fields protected by the lock */
size_t current; /* current region index */
size_t agg_size_full; /* aggregate size of full regions */
};
static struct tcg_region_state region;
/*
* This is an array of struct tcg_region_tree's, with padding.
* We use void * to simplify the computation of region_trees[i]; each
* struct is found every tree_size bytes.
*/
static void *region_trees;
static size_t tree_size;
bool in_code_gen_buffer(const void *p)
{
/*
* Much like it is valid to have a pointer to the byte past the
* end of an array (so long as you don't dereference it), allow
* a pointer to the byte past the end of the code gen buffer.
*/
return (size_t)(p - region.start_aligned) <= region.total_size;
}
#ifdef CONFIG_DEBUG_TCG
const void *tcg_splitwx_to_rx(void *rw)
{
/* Pass NULL pointers unchanged. */
if (rw) {
g_assert(in_code_gen_buffer(rw));
rw += tcg_splitwx_diff;
}
return rw;
}
void *tcg_splitwx_to_rw(const void *rx)
{
/* Pass NULL pointers unchanged. */
if (rx) {
rx -= tcg_splitwx_diff;
/* Assert that we end with a pointer in the rw region. */
g_assert(in_code_gen_buffer(rx));
}
return (void *)rx;
}
#endif /* CONFIG_DEBUG_TCG */
/* compare a pointer @ptr and a tb_tc @s */
static int ptr_cmp_tb_tc(const void *ptr, const struct tb_tc *s)
{
if (ptr >= s->ptr + s->size) {
return 1;
} else if (ptr < s->ptr) {
return -1;
}
return 0;
}
static gint tb_tc_cmp(gconstpointer ap, gconstpointer bp)
{
const struct tb_tc *a = ap;
const struct tb_tc *b = bp;
/*
* When both sizes are set, we know this isn't a lookup.
* This is the most likely case: every TB must be inserted; lookups
* are a lot less frequent.
*/
if (likely(a->size && b->size)) {
if (a->ptr > b->ptr) {
return 1;
} else if (a->ptr < b->ptr) {
return -1;
}
/* a->ptr == b->ptr should happen only on deletions */
g_assert(a->size == b->size);
return 0;
}
/*
* All lookups have either .size field set to 0.
* From the glib sources we see that @ap is always the lookup key. However
* the docs provide no guarantee, so we just mark this case as likely.
*/
if (likely(a->size == 0)) {
return ptr_cmp_tb_tc(a->ptr, b);
}
return ptr_cmp_tb_tc(b->ptr, a);
}
static void tcg_region_trees_init(void)
{
size_t i;
tree_size = ROUND_UP(sizeof(struct tcg_region_tree), qemu_dcache_linesize);
region_trees = qemu_memalign(qemu_dcache_linesize, region.n * tree_size);
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
qemu_mutex_init(&rt->lock);
rt->tree = g_tree_new(tb_tc_cmp);
}
}
static struct tcg_region_tree *tc_ptr_to_region_tree(const void *p)
{
size_t region_idx;
/*
* Like tcg_splitwx_to_rw, with no assert. The pc may come from
* a signal handler over which the caller has no control.
*/
if (!in_code_gen_buffer(p)) {
p -= tcg_splitwx_diff;
if (!in_code_gen_buffer(p)) {
return NULL;
}
}
if (p < region.start_aligned) {
region_idx = 0;
} else {
ptrdiff_t offset = p - region.start_aligned;
if (offset > region.stride * (region.n - 1)) {
region_idx = region.n - 1;
} else {
region_idx = offset / region.stride;
}
}
return region_trees + region_idx * tree_size;
}
void tcg_tb_insert(TranslationBlock *tb)
{
struct tcg_region_tree *rt = tc_ptr_to_region_tree(tb->tc.ptr);
g_assert(rt != NULL);
qemu_mutex_lock(&rt->lock);
g_tree_insert(rt->tree, &tb->tc, tb);
qemu_mutex_unlock(&rt->lock);
}
void tcg_tb_remove(TranslationBlock *tb)
{
struct tcg_region_tree *rt = tc_ptr_to_region_tree(tb->tc.ptr);
g_assert(rt != NULL);
qemu_mutex_lock(&rt->lock);
g_tree_remove(rt->tree, &tb->tc);
qemu_mutex_unlock(&rt->lock);
}
/*
* Find the TB 'tb' such that
* tb->tc.ptr <= tc_ptr < tb->tc.ptr + tb->tc.size
* Return NULL if not found.
*/
TranslationBlock *tcg_tb_lookup(uintptr_t tc_ptr)
{
struct tcg_region_tree *rt = tc_ptr_to_region_tree((void *)tc_ptr);
TranslationBlock *tb;
struct tb_tc s = { .ptr = (void *)tc_ptr };
if (rt == NULL) {
return NULL;
}
qemu_mutex_lock(&rt->lock);
tb = g_tree_lookup(rt->tree, &s);
qemu_mutex_unlock(&rt->lock);
return tb;
}
static void tcg_region_tree_lock_all(void)
{
size_t i;
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
qemu_mutex_lock(&rt->lock);
}
}
static void tcg_region_tree_unlock_all(void)
{
size_t i;
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
qemu_mutex_unlock(&rt->lock);
}
}
void tcg_tb_foreach(GTraverseFunc func, gpointer user_data)
{
size_t i;
tcg_region_tree_lock_all();
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
g_tree_foreach(rt->tree, func, user_data);
}
tcg_region_tree_unlock_all();
}
size_t tcg_nb_tbs(void)
{
size_t nb_tbs = 0;
size_t i;
tcg_region_tree_lock_all();
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
nb_tbs += g_tree_nnodes(rt->tree);
}
tcg_region_tree_unlock_all();
return nb_tbs;
}
static gboolean tcg_region_tree_traverse(gpointer k, gpointer v, gpointer data)
{
TranslationBlock *tb = v;
tb_destroy(tb);
return FALSE;
}
static void tcg_region_tree_reset_all(void)
{
size_t i;
tcg_region_tree_lock_all();
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
g_tree_foreach(rt->tree, tcg_region_tree_traverse, NULL);
/* Increment the refcount first so that destroy acts as a reset */
g_tree_ref(rt->tree);
g_tree_destroy(rt->tree);
}
tcg_region_tree_unlock_all();
}
static void tcg_region_bounds(size_t curr_region, void **pstart, void **pend)
{
void *start, *end;
start = region.start_aligned + curr_region * region.stride;
end = start + region.size;
if (curr_region == 0) {
start = region.after_prologue;
}
/* The final region may have a few extra pages due to earlier rounding. */
if (curr_region == region.n - 1) {
end = region.start_aligned + region.total_size;
}
*pstart = start;
*pend = end;
}
static void tcg_region_assign(TCGContext *s, size_t curr_region)
{
void *start, *end;
tcg_region_bounds(curr_region, &start, &end);
s->code_gen_buffer = start;
s->code_gen_ptr = start;
s->code_gen_buffer_size = end - start;
s->code_gen_highwater = end - TCG_HIGHWATER;
}
static bool tcg_region_alloc__locked(TCGContext *s)
{
if (region.current == region.n) {
return true;
}
tcg_region_assign(s, region.current);
region.current++;
return false;
}
/*
* Request a new region once the one in use has filled up.
* Returns true on error.
*/
bool tcg_region_alloc(TCGContext *s)
{
bool err;
/* read the region size now; alloc__locked will overwrite it on success */
size_t size_full = s->code_gen_buffer_size;
qemu_mutex_lock(&region.lock);
err = tcg_region_alloc__locked(s);
if (!err) {
region.agg_size_full += size_full - TCG_HIGHWATER;
}
qemu_mutex_unlock(&region.lock);
return err;
}
/*
* Perform a context's first region allocation.
* This function does _not_ increment region.agg_size_full.
*/
static void tcg_region_initial_alloc__locked(TCGContext *s)
{
bool err = tcg_region_alloc__locked(s);
g_assert(!err);
}
void tcg_region_initial_alloc(TCGContext *s)
{
qemu_mutex_lock(&region.lock);
tcg_region_initial_alloc__locked(s);
qemu_mutex_unlock(&region.lock);
}
/* Call from a safe-work context */
void tcg_region_reset_all(void)
{
unsigned int n_ctxs = qatomic_read(&tcg_cur_ctxs);
unsigned int i;
qemu_mutex_lock(&region.lock);
region.current = 0;
region.agg_size_full = 0;
for (i = 0; i < n_ctxs; i++) {
TCGContext *s = qatomic_read(&tcg_ctxs[i]);
tcg_region_initial_alloc__locked(s);
}
qemu_mutex_unlock(&region.lock);
tcg_region_tree_reset_all();
}
static size_t tcg_n_regions(size_t tb_size, unsigned max_cpus)
{
#ifdef CONFIG_USER_ONLY
return 1;
#else
size_t n_regions;
/*
* It is likely that some vCPUs will translate more code than others,
* so we first try to set more regions than max_cpus, with those regions
* being of reasonable size. If that's not possible we make do by evenly
* dividing the code_gen_buffer among the vCPUs.
*/
/* Use a single region if all we have is one vCPU thread */
if (max_cpus == 1 || !qemu_tcg_mttcg_enabled()) {
return 1;
}
/*
* Try to have more regions than max_cpus, with each region being >= 2 MB.
* If we can't, then just allocate one region per vCPU thread.
*/
n_regions = tb_size / (2 * MiB);
if (n_regions <= max_cpus) {
return max_cpus;
}
return MIN(n_regions, max_cpus * 8);
#endif
}
/*
* Minimum size of the code gen buffer. This number is randomly chosen,
* but not so small that we can't have a fair number of TB's live.
*
* Maximum size, MAX_CODE_GEN_BUFFER_SIZE, is defined in tcg-target.h.
* Unless otherwise indicated, this is constrained by the range of
* direct branches on the host cpu, as used by the TCG implementation
* of goto_tb.
*/
#define MIN_CODE_GEN_BUFFER_SIZE (1 * MiB)
#if TCG_TARGET_REG_BITS == 32
#define DEFAULT_CODE_GEN_BUFFER_SIZE_1 (32 * MiB)
#ifdef CONFIG_USER_ONLY
/*
* For user mode on smaller 32 bit systems we may run into trouble
* allocating big chunks of data in the right place. On these systems
* we utilise a static code generation buffer directly in the binary.
*/
#define USE_STATIC_CODE_GEN_BUFFER
#endif
#else /* TCG_TARGET_REG_BITS == 64 */
#ifdef CONFIG_USER_ONLY
/*
* As user-mode emulation typically means running multiple instances
* of the translator don't go too nuts with our default code gen
* buffer lest we make things too hard for the OS.
*/
#define DEFAULT_CODE_GEN_BUFFER_SIZE_1 (128 * MiB)
#else
/*
* We expect most system emulation to run one or two guests per host.
* Users running large scale system emulation may want to tweak their
* runtime setup via the tb-size control on the command line.
*/
#define DEFAULT_CODE_GEN_BUFFER_SIZE_1 (1 * GiB)
#endif
#endif
#define DEFAULT_CODE_GEN_BUFFER_SIZE \
(DEFAULT_CODE_GEN_BUFFER_SIZE_1 < MAX_CODE_GEN_BUFFER_SIZE \
? DEFAULT_CODE_GEN_BUFFER_SIZE_1 : MAX_CODE_GEN_BUFFER_SIZE)
#ifdef __mips__
/*
* In order to use J and JAL within the code_gen_buffer, we require
* that the buffer not cross a 256MB boundary.
*/
static inline bool cross_256mb(void *addr, size_t size)
{
return ((uintptr_t)addr ^ ((uintptr_t)addr + size)) & ~0x0ffffffful;
}
/*
* We weren't able to allocate a buffer without crossing that boundary,
* so make do with the larger portion of the buffer that doesn't cross.
* Returns the new base and size of the buffer in *obuf and *osize.
*/
static inline void split_cross_256mb(void **obuf, size_t *osize,
void *buf1, size_t size1)
{
void *buf2 = (void *)(((uintptr_t)buf1 + size1) & ~0x0ffffffful);
size_t size2 = buf1 + size1 - buf2;
size1 = buf2 - buf1;
if (size1 < size2) {
size1 = size2;
buf1 = buf2;
}
*obuf = buf1;
*osize = size1;
}
#endif
#ifdef USE_STATIC_CODE_GEN_BUFFER
static uint8_t static_code_gen_buffer[DEFAULT_CODE_GEN_BUFFER_SIZE]
__attribute__((aligned(CODE_GEN_ALIGN)));
static int alloc_code_gen_buffer(size_t tb_size, int splitwx, Error **errp)
{
void *buf, *end;
size_t size;
if (splitwx > 0) {
error_setg(errp, "jit split-wx not supported");
return -1;
}
/* page-align the beginning and end of the buffer */
buf = static_code_gen_buffer;
end = static_code_gen_buffer + sizeof(static_code_gen_buffer);
buf = QEMU_ALIGN_PTR_UP(buf, qemu_real_host_page_size);
end = QEMU_ALIGN_PTR_DOWN(end, qemu_real_host_page_size);
size = end - buf;
/* Honor a command-line option limiting the size of the buffer. */
if (size > tb_size) {
size = QEMU_ALIGN_DOWN(tb_size, qemu_real_host_page_size);
}
#ifdef __mips__
if (cross_256mb(buf, size)) {
split_cross_256mb(&buf, &size, buf, size);
}
#endif
region.start_aligned = buf;
region.total_size = size;
return PROT_READ | PROT_WRITE;
}
#elif defined(_WIN32)
static int alloc_code_gen_buffer(size_t size, int splitwx, Error **errp)
{
void *buf;
if (splitwx > 0) {
error_setg(errp, "jit split-wx not supported");
return -1;
}
buf = VirtualAlloc(NULL, size, MEM_RESERVE | MEM_COMMIT,
PAGE_EXECUTE_READWRITE);
if (buf == NULL) {
error_setg_win32(errp, GetLastError(),
"allocate %zu bytes for jit buffer", size);
return false;
}
region.start_aligned = buf;
region.total_size = size;
return PAGE_READ | PAGE_WRITE | PAGE_EXEC;
}
#else
static int alloc_code_gen_buffer_anon(size_t size, int prot,
int flags, Error **errp)
{
void *buf;
buf = mmap(NULL, size, prot, flags, -1, 0);
if (buf == MAP_FAILED) {
error_setg_errno(errp, errno,
"allocate %zu bytes for jit buffer", size);
return -1;
}
#ifdef __mips__
if (cross_256mb(buf, size)) {
/*
* Try again, with the original still mapped, to avoid re-acquiring
* the same 256mb crossing.
*/
size_t size2;
void *buf2 = mmap(NULL, size, prot, flags, -1, 0);
switch ((int)(buf2 != MAP_FAILED)) {
case 1:
if (!cross_256mb(buf2, size)) {
/* Success! Use the new buffer. */
munmap(buf, size);
break;
}
/* Failure. Work with what we had. */
munmap(buf2, size);
/* fallthru */
default:
/* Split the original buffer. Free the smaller half. */
split_cross_256mb(&buf2, &size2, buf, size);
if (buf == buf2) {
munmap(buf + size2, size - size2);
} else {
munmap(buf, size - size2);
}
size = size2;
break;
}
buf = buf2;
}
#endif
region.start_aligned = buf;
region.total_size = size;
return prot;
}
#ifndef CONFIG_TCG_INTERPRETER
#ifdef CONFIG_POSIX
#include "qemu/memfd.h"
static bool alloc_code_gen_buffer_splitwx_memfd(size_t size, Error **errp)
{
void *buf_rw = NULL, *buf_rx = MAP_FAILED;
int fd = -1;
#ifdef __mips__
/* Find space for the RX mapping, vs the 256MiB regions. */
if (alloc_code_gen_buffer_anon(size, PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS |
MAP_NORESERVE, errp) < 0) {
return false;
}
/* The size of the mapping may have been adjusted. */
buf_rx = region.start_aligned;
size = region.total_size;
#endif
buf_rw = qemu_memfd_alloc("tcg-jit", size, 0, &fd, errp);
if (buf_rw == NULL) {
goto fail;
}
#ifdef __mips__
void *tmp = mmap(buf_rx, size, PROT_READ | PROT_EXEC,
MAP_SHARED | MAP_FIXED, fd, 0);
if (tmp != buf_rx) {
goto fail_rx;
}
#else
buf_rx = mmap(NULL, size, PROT_READ | PROT_EXEC, MAP_SHARED, fd, 0);
if (buf_rx == MAP_FAILED) {
goto fail_rx;
}
#endif
close(fd);
region.start_aligned = buf_rw;
region.total_size = size;
tcg_splitwx_diff = buf_rx - buf_rw;
return PROT_READ | PROT_WRITE;
fail_rx:
error_setg_errno(errp, errno, "failed to map shared memory for execute");
fail:
if (buf_rx != MAP_FAILED) {
munmap(buf_rx, size);
}
if (buf_rw) {
munmap(buf_rw, size);
}
if (fd >= 0) {
close(fd);
}
return -1;
}
#endif /* CONFIG_POSIX */
#ifdef CONFIG_DARWIN
#include <mach/mach.h>
extern kern_return_t mach_vm_remap(vm_map_t target_task,
mach_vm_address_t *target_address,
mach_vm_size_t size,
mach_vm_offset_t mask,
int flags,
vm_map_t src_task,
mach_vm_address_t src_address,
boolean_t copy,
vm_prot_t *cur_protection,
vm_prot_t *max_protection,
vm_inherit_t inheritance);
static int alloc_code_gen_buffer_splitwx_vmremap(size_t size, Error **errp)
{
kern_return_t ret;
mach_vm_address_t buf_rw, buf_rx;
vm_prot_t cur_prot, max_prot;
/* Map the read-write portion via normal anon memory. */
if (!alloc_code_gen_buffer_anon(size, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, errp)) {
return -1;
}
buf_rw = (mach_vm_address_t)region.start_aligned;
buf_rx = 0;
ret = mach_vm_remap(mach_task_self(),
&buf_rx,
size,
0,
VM_FLAGS_ANYWHERE,
mach_task_self(),
buf_rw,
false,
&cur_prot,
&max_prot,
VM_INHERIT_NONE);
if (ret != KERN_SUCCESS) {
/* TODO: Convert "ret" to a human readable error message. */
error_setg(errp, "vm_remap for jit splitwx failed");
munmap((void *)buf_rw, size);
return -1;
}
if (mprotect((void *)buf_rx, size, PROT_READ | PROT_EXEC) != 0) {
error_setg_errno(errp, errno, "mprotect for jit splitwx");
munmap((void *)buf_rx, size);
munmap((void *)buf_rw, size);
return -1;
}
tcg_splitwx_diff = buf_rx - buf_rw;
return PROT_READ | PROT_WRITE;
}
#endif /* CONFIG_DARWIN */
#endif /* CONFIG_TCG_INTERPRETER */
static int alloc_code_gen_buffer_splitwx(size_t size, Error **errp)
{
#ifndef CONFIG_TCG_INTERPRETER
# ifdef CONFIG_DARWIN
return alloc_code_gen_buffer_splitwx_vmremap(size, errp);
# endif
# ifdef CONFIG_POSIX
return alloc_code_gen_buffer_splitwx_memfd(size, errp);
# endif
#endif
error_setg(errp, "jit split-wx not supported");
return -1;
}
static int alloc_code_gen_buffer(size_t size, int splitwx, Error **errp)
{
ERRP_GUARD();
int prot, flags;
if (splitwx) {
prot = alloc_code_gen_buffer_splitwx(size, errp);
if (prot >= 0) {
return prot;
}
/*
* If splitwx force-on (1), fail;
* if splitwx default-on (-1), fall through to splitwx off.
*/
if (splitwx > 0) {
return -1;
}
error_free_or_abort(errp);
}
/*
* macOS 11.2 has a bug (Apple Feedback FB8994773) in which mprotect
* rejects a permission change from RWX -> NONE when reserving the
* guard pages later. We can go the other way with the same number
* of syscalls, so always begin with PROT_NONE.
*/
prot = PROT_NONE;
flags = MAP_PRIVATE | MAP_ANONYMOUS;
#ifdef CONFIG_DARWIN
/* Applicable to both iOS and macOS (Apple Silicon). */
if (!splitwx) {
flags |= MAP_JIT;
}
#endif
return alloc_code_gen_buffer_anon(size, prot, flags, errp);
}
#endif /* USE_STATIC_CODE_GEN_BUFFER, WIN32, POSIX */
/*
* Initializes region partitioning.
*
* Called at init time from the parent thread (i.e. the one calling
* tcg_context_init), after the target's TCG globals have been set.
*
* Region partitioning works by splitting code_gen_buffer into separate regions,
* and then assigning regions to TCG threads so that the threads can translate
* code in parallel without synchronization.
*
* In softmmu the number of TCG threads is bounded by max_cpus, so we use at
* least max_cpus regions in MTTCG. In !MTTCG we use a single region.
* Note that the TCG options from the command-line (i.e. -accel accel=tcg,[...])
* must have been parsed before calling this function, since it calls
* qemu_tcg_mttcg_enabled().
*
* In user-mode we use a single region. Having multiple regions in user-mode
* is not supported, because the number of vCPU threads (recall that each thread
* spawned by the guest corresponds to a vCPU thread) is only bounded by the
* OS, and usually this number is huge (tens of thousands is not uncommon).
* Thus, given this large bound on the number of vCPU threads and the fact
* that code_gen_buffer is allocated at compile-time, we cannot guarantee
* that the availability of at least one region per vCPU thread.
*
* However, this user-mode limitation is unlikely to be a significant problem
* in practice. Multi-threaded guests share most if not all of their translated
* code, which makes parallel code generation less appealing than in softmmu.
*/
void tcg_region_init(size_t tb_size, int splitwx, unsigned max_cpus)
{
const size_t page_size = qemu_real_host_page_size;
size_t region_size;
int have_prot, need_prot;
/* Size the buffer. */
if (tb_size == 0) {
size_t phys_mem = qemu_get_host_physmem();
if (phys_mem == 0) {
tb_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
} else {
tb_size = QEMU_ALIGN_DOWN(phys_mem / 8, page_size);
tb_size = MIN(DEFAULT_CODE_GEN_BUFFER_SIZE, tb_size);
}
}
if (tb_size < MIN_CODE_GEN_BUFFER_SIZE) {
tb_size = MIN_CODE_GEN_BUFFER_SIZE;
}
if (tb_size > MAX_CODE_GEN_BUFFER_SIZE) {
tb_size = MAX_CODE_GEN_BUFFER_SIZE;
}
have_prot = alloc_code_gen_buffer(tb_size, splitwx, &error_fatal);
assert(have_prot >= 0);
/* Request large pages for the buffer and the splitwx. */
qemu_madvise(region.start_aligned, region.total_size, QEMU_MADV_HUGEPAGE);
if (tcg_splitwx_diff) {
qemu_madvise(region.start_aligned + tcg_splitwx_diff,
region.total_size, QEMU_MADV_HUGEPAGE);
}
/*
* Make region_size a multiple of page_size, using aligned as the start.
* As a result of this we might end up with a few extra pages at the end of
* the buffer; we will assign those to the last region.
*/
region.n = tcg_n_regions(tb_size, max_cpus);
region_size = tb_size / region.n;
region_size = QEMU_ALIGN_DOWN(region_size, page_size);
/* A region must have at least 2 pages; one code, one guard */
g_assert(region_size >= 2 * page_size);
region.stride = region_size;
/* Reserve space for guard pages. */
region.size = region_size - page_size;
region.total_size -= page_size;
/*
* The first region will be smaller than the others, via the prologue,
* which has yet to be allocated. For now, the first region begins at
* the page boundary.
*/
region.after_prologue = region.start_aligned;
/* init the region struct */
qemu_mutex_init(&region.lock);
/*
* Set guard pages in the rw buffer, as that's the one into which
* buffer overruns could occur. Do not set guard pages in the rx
* buffer -- let that one use hugepages throughout.
* Work with the page protections set up with the initial mapping.
*/
need_prot = PAGE_READ | PAGE_WRITE;
#ifndef CONFIG_TCG_INTERPRETER
if (tcg_splitwx_diff == 0) {
need_prot |= PAGE_EXEC;
}
#endif
for (size_t i = 0, n = region.n; i < n; i++) {
void *start, *end;
tcg_region_bounds(i, &start, &end);
if (have_prot != need_prot) {
int rc;
if (need_prot == (PAGE_READ | PAGE_WRITE | PAGE_EXEC)) {
rc = qemu_mprotect_rwx(start, end - start);
} else if (need_prot == (PAGE_READ | PAGE_WRITE)) {
rc = qemu_mprotect_rw(start, end - start);
} else {
g_assert_not_reached();
}
if (rc) {
error_setg_errno(&error_fatal, errno,
"mprotect of jit buffer");
}
}
if (have_prot != 0) {
/* Guard pages are nice for bug detection but are not essential. */
(void)qemu_mprotect_none(end, page_size);
}
}
tcg_region_trees_init();
/*
* Leave the initial context initialized to the first region.
* This will be the context into which we generate the prologue.
* It is also the only context for CONFIG_USER_ONLY.
*/
tcg_region_initial_alloc__locked(&tcg_init_ctx);
}
void tcg_region_prologue_set(TCGContext *s)
{
/* Deduct the prologue from the first region. */
g_assert(region.start_aligned == s->code_gen_buffer);
region.after_prologue = s->code_ptr;
/* Recompute boundaries of the first region. */
tcg_region_assign(s, 0);
/* Register the balance of the buffer with gdb. */
tcg_register_jit(tcg_splitwx_to_rx(region.after_prologue),
region.start_aligned + region.total_size -
region.after_prologue);
}
/*
* Returns the size (in bytes) of all translated code (i.e. from all regions)
* currently in the cache.
* See also: tcg_code_capacity()
* Do not confuse with tcg_current_code_size(); that one applies to a single
* TCG context.
*/
size_t tcg_code_size(void)
{
unsigned int n_ctxs = qatomic_read(&tcg_cur_ctxs);
unsigned int i;
size_t total;
qemu_mutex_lock(&region.lock);
total = region.agg_size_full;
for (i = 0; i < n_ctxs; i++) {
const TCGContext *s = qatomic_read(&tcg_ctxs[i]);
size_t size;
size = qatomic_read(&s->code_gen_ptr) - s->code_gen_buffer;
g_assert(size <= s->code_gen_buffer_size);
total += size;
}
qemu_mutex_unlock(&region.lock);
return total;
}
/*
* Returns the code capacity (in bytes) of the entire cache, i.e. including all
* regions.
* See also: tcg_code_size()
*/
size_t tcg_code_capacity(void)
{
size_t guard_size, capacity;
/* no need for synchronization; these variables are set at init time */
guard_size = region.stride - region.size;
capacity = region.total_size;
capacity -= (region.n - 1) * guard_size;
capacity -= region.n * TCG_HIGHWATER;
return capacity;
}
size_t tcg_tb_phys_invalidate_count(void)
{
unsigned int n_ctxs = qatomic_read(&tcg_cur_ctxs);
unsigned int i;
size_t total = 0;
for (i = 0; i < n_ctxs; i++) {
const TCGContext *s = qatomic_read(&tcg_ctxs[i]);
total += qatomic_read(&s->tb_phys_invalidate_count);
}
return total;
}

1
tcg/riscv/tcg-target.h
View File

@ -34,6 +34,7 @@
#define TCG_TARGET_INSN_UNIT_SIZE 4
#define TCG_TARGET_TLB_DISPLACEMENT_BITS 20
#define TCG_TARGET_NB_REGS 32
#define MAX_CODE_GEN_BUFFER_SIZE ((size_t)-1)
typedef enum {
TCG_REG_ZERO,

3
tcg/s390/tcg-target.h
View File

@ -28,6 +28,9 @@
#define TCG_TARGET_INSN_UNIT_SIZE 2
#define TCG_TARGET_TLB_DISPLACEMENT_BITS 19
/* We have a +- 4GB range on the branches; leave some slop. */
#define MAX_CODE_GEN_BUFFER_SIZE (3 * GiB)
typedef enum TCGReg {
TCG_REG_R0 = 0,
TCG_REG_R1,

1
tcg/sparc/tcg-target.h
View File

@ -30,6 +30,7 @@
#define TCG_TARGET_INSN_UNIT_SIZE 4
#define TCG_TARGET_TLB_DISPLACEMENT_BITS 32
#define TCG_TARGET_NB_REGS 32
#define MAX_CODE_GEN_BUFFER_SIZE (2 * GiB)
typedef enum {
TCG_REG_G0 = 0,

40
tcg/tcg-internal.h Normal file
View File

@ -0,0 +1,40 @@
/*
* Internal declarations for Tiny Code Generator for QEMU
*
* Copyright (c) 2008 Fabrice Bellard
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#ifndef TCG_INTERNAL_H
#define TCG_INTERNAL_H 1
#define TCG_HIGHWATER 1024
extern TCGContext tcg_init_ctx;
extern TCGContext **tcg_ctxs;
extern unsigned int tcg_cur_ctxs;
extern unsigned int tcg_max_ctxs;
void tcg_region_init(size_t tb_size, int splitwx, unsigned max_cpus);
bool tcg_region_alloc(TCGContext *s);
void tcg_region_initial_alloc(TCGContext *s);
void tcg_region_prologue_set(TCGContext *s);
#endif /* TCG_INTERNAL_H */

649
tcg/tcg.c
View File

@ -43,11 +43,6 @@
#define NO_CPU_IO_DEFS
#include "exec/exec-all.h"
#if !defined(CONFIG_USER_ONLY)
#include "hw/boards.h"
#endif
#include "tcg/tcg-op.h"
#if UINTPTR_MAX == UINT32_MAX
@ -63,6 +58,7 @@
#include "elf.h"
#include "exec/log.h"
#include "tcg-internal.h"
/* Forward declarations for functions declared in tcg-target.c.inc and
used here. */
@ -153,10 +149,12 @@ static bool tcg_target_const_match(int64_t val, TCGType type, int ct);
static int tcg_out_ldst_finalize(TCGContext *s);
#endif
#define TCG_HIGHWATER 1024
TCGContext tcg_init_ctx;
__thread TCGContext *tcg_ctx;
static TCGContext **tcg_ctxs;
static unsigned int n_tcg_ctxs;
TCGContext **tcg_ctxs;
unsigned int tcg_cur_ctxs;
unsigned int tcg_max_ctxs;
TCGv_env cpu_env = 0;
const void *tcg_code_gen_epilogue;
uintptr_t tcg_splitwx_diff;
@ -165,42 +163,6 @@ uintptr_t tcg_splitwx_diff;
tcg_prologue_fn *tcg_qemu_tb_exec;
#endif
struct tcg_region_tree {
QemuMutex lock;
GTree *tree;
/* padding to avoid false sharing is computed at run-time */
};
/*
* We divide code_gen_buffer into equally-sized "regions" that TCG threads
* dynamically allocate from as demand dictates. Given appropriate region
* sizing, this minimizes flushes even when some TCG threads generate a lot
* more code than others.
*/
struct tcg_region_state {
QemuMutex lock;
/* fields set at init time */
void *start;
void *start_aligned;
void *end;
size_t n;
size_t size; /* size of one region */
size_t stride; /* .size + guard size */
/* fields protected by the lock */
size_t current; /* current region index */
size_t agg_size_full; /* aggregate size of full regions */
};
static struct tcg_region_state region;
/*
* This is an array of struct tcg_region_tree's, with padding.
* We use void * to simplify the computation of region_trees[i]; each
* struct is found every tree_size bytes.
*/
static void *region_trees;
static size_t tree_size;
static TCGRegSet tcg_target_available_regs[TCG_TYPE_COUNT];
static TCGRegSet tcg_target_call_clobber_regs;
@ -457,456 +419,6 @@ static const TCGTargetOpDef constraint_sets[] = {
#include "tcg-target.c.inc"
/* compare a pointer @ptr and a tb_tc @s */
static int ptr_cmp_tb_tc(const void *ptr, const struct tb_tc *s)
{
if (ptr >= s->ptr + s->size) {
return 1;
} else if (ptr < s->ptr) {
return -1;
}
return 0;
}
static gint tb_tc_cmp(gconstpointer ap, gconstpointer bp)
{
const struct tb_tc *a = ap;
const struct tb_tc *b = bp;
/*
* When both sizes are set, we know this isn't a lookup.
* This is the most likely case: every TB must be inserted; lookups
* are a lot less frequent.
*/
if (likely(a->size && b->size)) {
if (a->ptr > b->ptr) {
return 1;
} else if (a->ptr < b->ptr) {
return -1;
}
/* a->ptr == b->ptr should happen only on deletions */
g_assert(a->size == b->size);
return 0;
}
/*
* All lookups have either .size field set to 0.
* From the glib sources we see that @ap is always the lookup key. However
* the docs provide no guarantee, so we just mark this case as likely.
*/
if (likely(a->size == 0)) {
return ptr_cmp_tb_tc(a->ptr, b);
}
return ptr_cmp_tb_tc(b->ptr, a);
}
static void tcg_region_trees_init(void)
{
size_t i;
tree_size = ROUND_UP(sizeof(struct tcg_region_tree), qemu_dcache_linesize);
region_trees = qemu_memalign(qemu_dcache_linesize, region.n * tree_size);
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
qemu_mutex_init(&rt->lock);
rt->tree = g_tree_new(tb_tc_cmp);
}
}
static struct tcg_region_tree *tc_ptr_to_region_tree(const void *p)
{
size_t region_idx;
/*
* Like tcg_splitwx_to_rw, with no assert. The pc may come from
* a signal handler over which the caller has no control.
*/
if (!in_code_gen_buffer(p)) {
p -= tcg_splitwx_diff;
if (!in_code_gen_buffer(p)) {
return NULL;
}
}
if (p < region.start_aligned) {
region_idx = 0;
} else {
ptrdiff_t offset = p - region.start_aligned;
if (offset > region.stride * (region.n - 1)) {
region_idx = region.n - 1;
} else {
region_idx = offset / region.stride;
}
}
return region_trees + region_idx * tree_size;
}
void tcg_tb_insert(TranslationBlock *tb)
{
struct tcg_region_tree *rt = tc_ptr_to_region_tree(tb->tc.ptr);
g_assert(rt != NULL);
qemu_mutex_lock(&rt->lock);
g_tree_insert(rt->tree, &tb->tc, tb);
qemu_mutex_unlock(&rt->lock);
}
void tcg_tb_remove(TranslationBlock *tb)
{
struct tcg_region_tree *rt = tc_ptr_to_region_tree(tb->tc.ptr);
g_assert(rt != NULL);
qemu_mutex_lock(&rt->lock);
g_tree_remove(rt->tree, &tb->tc);
qemu_mutex_unlock(&rt->lock);
}
/*
* Find the TB 'tb' such that
* tb->tc.ptr <= tc_ptr < tb->tc.ptr + tb->tc.size
* Return NULL if not found.
*/
TranslationBlock *tcg_tb_lookup(uintptr_t tc_ptr)
{
struct tcg_region_tree *rt = tc_ptr_to_region_tree((void *)tc_ptr);
TranslationBlock *tb;
struct tb_tc s = { .ptr = (void *)tc_ptr };
if (rt == NULL) {
return NULL;
}
qemu_mutex_lock(&rt->lock);
tb = g_tree_lookup(rt->tree, &s);
qemu_mutex_unlock(&rt->lock);
return tb;
}
static void tcg_region_tree_lock_all(void)
{
size_t i;
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
qemu_mutex_lock(&rt->lock);
}
}
static void tcg_region_tree_unlock_all(void)
{
size_t i;
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
qemu_mutex_unlock(&rt->lock);
}
}
void tcg_tb_foreach(GTraverseFunc func, gpointer user_data)
{
size_t i;
tcg_region_tree_lock_all();
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
g_tree_foreach(rt->tree, func, user_data);
}
tcg_region_tree_unlock_all();
}
size_t tcg_nb_tbs(void)
{
size_t nb_tbs = 0;
size_t i;
tcg_region_tree_lock_all();
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
nb_tbs += g_tree_nnodes(rt->tree);
}
tcg_region_tree_unlock_all();
return nb_tbs;
}
static gboolean tcg_region_tree_traverse(gpointer k, gpointer v, gpointer data)
{
TranslationBlock *tb = v;
tb_destroy(tb);
return FALSE;
}
static void tcg_region_tree_reset_all(void)
{
size_t i;
tcg_region_tree_lock_all();
for (i = 0; i < region.n; i++) {
struct tcg_region_tree *rt = region_trees + i * tree_size;
g_tree_foreach(rt->tree, tcg_region_tree_traverse, NULL);
/* Increment the refcount first so that destroy acts as a reset */
g_tree_ref(rt->tree);
g_tree_destroy(rt->tree);
}
tcg_region_tree_unlock_all();
}
static void tcg_region_bounds(size_t curr_region, void **pstart, void **pend)
{
void *start, *end;
start = region.start_aligned + curr_region * region.stride;
end = start + region.size;
if (curr_region == 0) {
start = region.start;
}
if (curr_region == region.n - 1) {
end = region.end;
}
*pstart = start;
*pend = end;
}
static void tcg_region_assign(TCGContext *s, size_t curr_region)
{
void *start, *end;
tcg_region_bounds(curr_region, &start, &end);
s->code_gen_buffer = start;
s->code_gen_ptr = start;
s->code_gen_buffer_size = end - start;
s->code_gen_highwater = end - TCG_HIGHWATER;
}
static bool tcg_region_alloc__locked(TCGContext *s)
{
if (region.current == region.n) {
return true;
}
tcg_region_assign(s, region.current);
region.current++;
return false;
}
/*
* Request a new region once the one in use has filled up.
* Returns true on error.
*/
static bool tcg_region_alloc(TCGContext *s)
{
bool err;
/* read the region size now; alloc__locked will overwrite it on success */
size_t size_full = s->code_gen_buffer_size;
qemu_mutex_lock(&region.lock);
err = tcg_region_alloc__locked(s);
if (!err) {
region.agg_size_full += size_full - TCG_HIGHWATER;
}
qemu_mutex_unlock(&region.lock);
return err;
}
/*
* Perform a context's first region allocation.
* This function does _not_ increment region.agg_size_full.
*/
static inline bool tcg_region_initial_alloc__locked(TCGContext *s)
{
return tcg_region_alloc__locked(s);
}
/* Call from a safe-work context */
void tcg_region_reset_all(void)
{
unsigned int n_ctxs = qatomic_read(&n_tcg_ctxs);
unsigned int i;
qemu_mutex_lock(&region.lock);
region.current = 0;
region.agg_size_full = 0;
for (i = 0; i < n_ctxs; i++) {
TCGContext *s = qatomic_read(&tcg_ctxs[i]);
bool err = tcg_region_initial_alloc__locked(s);
g_assert(!err);
}
qemu_mutex_unlock(&region.lock);
tcg_region_tree_reset_all();
}
#ifdef CONFIG_USER_ONLY
static size_t tcg_n_regions(void)
{
return 1;
}
#else
/*
* It is likely that some vCPUs will translate more code than others, so we
* first try to set more regions than max_cpus, with those regions being of
* reasonable size. If that's not possible we make do by evenly dividing
* the code_gen_buffer among the vCPUs.
*/
static size_t tcg_n_regions(void)
{
size_t i;
/* Use a single region if all we have is one vCPU thread */
#if !defined(CONFIG_USER_ONLY)
MachineState *ms = MACHINE(qdev_get_machine());
unsigned int max_cpus = ms->smp.max_cpus;
#endif
if (max_cpus == 1 || !qemu_tcg_mttcg_enabled()) {
return 1;
}
/* Try to have more regions than max_cpus, with each region being >= 2 MB */
for (i = 8; i > 0; i--) {
size_t regions_per_thread = i;
size_t region_size;
region_size = tcg_init_ctx.code_gen_buffer_size;
region_size /= max_cpus * regions_per_thread;
if (region_size >= 2 * 1024u * 1024) {
return max_cpus * regions_per_thread;
}
}
/* If we can't, then just allocate one region per vCPU thread */
return max_cpus;
}
#endif
/*
* Initializes region partitioning.
*
* Called at init time from the parent thread (i.e. the one calling
* tcg_context_init), after the target's TCG globals have been set.
*
* Region partitioning works by splitting code_gen_buffer into separate regions,
* and then assigning regions to TCG threads so that the threads can translate
* code in parallel without synchronization.
*
* In softmmu the number of TCG threads is bounded by max_cpus, so we use at
* least max_cpus regions in MTTCG. In !MTTCG we use a single region.
* Note that the TCG options from the command-line (i.e. -accel accel=tcg,[...])
* must have been parsed before calling this function, since it calls
* qemu_tcg_mttcg_enabled().
*
* In user-mode we use a single region. Having multiple regions in user-mode
* is not supported, because the number of vCPU threads (recall that each thread
* spawned by the guest corresponds to a vCPU thread) is only bounded by the
* OS, and usually this number is huge (tens of thousands is not uncommon).
* Thus, given this large bound on the number of vCPU threads and the fact
* that code_gen_buffer is allocated at compile-time, we cannot guarantee
* that the availability of at least one region per vCPU thread.
*
* However, this user-mode limitation is unlikely to be a significant problem
* in practice. Multi-threaded guests share most if not all of their translated
* code, which makes parallel code generation less appealing than in softmmu.
*/
void tcg_region_init(void)
{
void *buf = tcg_init_ctx.code_gen_buffer;
void *aligned;
size_t size = tcg_init_ctx.code_gen_buffer_size;
size_t page_size = qemu_real_host_page_size;
size_t region_size;
size_t n_regions;
size_t i;
n_regions = tcg_n_regions();
/* The first region will be 'aligned - buf' bytes larger than the others */
aligned = QEMU_ALIGN_PTR_UP(buf, page_size);
g_assert(aligned < tcg_init_ctx.code_gen_buffer + size);
/*
* Make region_size a multiple of page_size, using aligned as the start.
* As a result of this we might end up with a few extra pages at the end of
* the buffer; we will assign those to the last region.
*/
region_size = (size - (aligned - buf)) / n_regions;
region_size = QEMU_ALIGN_DOWN(region_size, page_size);
/* A region must have at least 2 pages; one code, one guard */
g_assert(region_size >= 2 * page_size);
/* init the region struct */
qemu_mutex_init(&region.lock);
region.n = n_regions;
region.size = region_size - page_size;
region.stride = region_size;
region.start = buf;
region.start_aligned = aligned;
/* page-align the end, since its last page will be a guard page */
region.end = QEMU_ALIGN_PTR_DOWN(buf + size, page_size);
/* account for that last guard page */
region.end -= page_size;
/*
* Set guard pages in the rw buffer, as that's the one into which
* buffer overruns could occur. Do not set guard pages in the rx
* buffer -- let that one use hugepages throughout.
*/
for (i = 0; i < region.n; i++) {
void *start, *end;
tcg_region_bounds(i, &start, &end);
/*
* macOS 11.2 has a bug (Apple Feedback FB8994773) in which mprotect
* rejects a permission change from RWX -> NONE. Guard pages are
* nice for bug detection but are not essential; ignore any failure.
*/
(void)qemu_mprotect_none(end, page_size);
}
tcg_region_trees_init();
/* In user-mode we support only one ctx, so do the initial allocation now */
#ifdef CONFIG_USER_ONLY
{
bool err = tcg_region_initial_alloc__locked(tcg_ctx);
g_assert(!err);
}
#endif
}
#ifdef CONFIG_DEBUG_TCG
const void *tcg_splitwx_to_rx(void *rw)
{
/* Pass NULL pointers unchanged. */
if (rw) {
g_assert(in_code_gen_buffer(rw));
rw += tcg_splitwx_diff;
}
return rw;
}
void *tcg_splitwx_to_rw(const void *rx)
{
/* Pass NULL pointers unchanged. */
if (rx) {
rx -= tcg_splitwx_diff;
/* Assert that we end with a pointer in the rw region. */
g_assert(in_code_gen_buffer(rx));
}
return (void *)rx;
}
#endif /* CONFIG_DEBUG_TCG */
static void alloc_tcg_plugin_context(TCGContext *s)
{
#ifdef CONFIG_PLUGIN
@ -939,10 +451,8 @@ void tcg_register_thread(void)
#else
void tcg_register_thread(void)
{
MachineState *ms = MACHINE(qdev_get_machine());
TCGContext *s = g_malloc(sizeof(*s));
unsigned int i, n;
bool err;
*s = tcg_init_ctx;
@ -956,79 +466,19 @@ void tcg_register_thread(void)
}
/* Claim an entry in tcg_ctxs */
n = qatomic_fetch_inc(&n_tcg_ctxs);
g_assert(n < ms->smp.max_cpus);
n = qatomic_fetch_inc(&tcg_cur_ctxs);
g_assert(n < tcg_max_ctxs);
qatomic_set(&tcg_ctxs[n], s);
if (n > 0) {
alloc_tcg_plugin_context(s);
tcg_region_initial_alloc(s);
}
tcg_ctx = s;
qemu_mutex_lock(&region.lock);
err = tcg_region_initial_alloc__locked(tcg_ctx);
g_assert(!err);
qemu_mutex_unlock(&region.lock);
}
#endif /* !CONFIG_USER_ONLY */
/*
* Returns the size (in bytes) of all translated code (i.e. from all regions)
* currently in the cache.
* See also: tcg_code_capacity()
* Do not confuse with tcg_current_code_size(); that one applies to a single
* TCG context.
*/
size_t tcg_code_size(void)
{
unsigned int n_ctxs = qatomic_read(&n_tcg_ctxs);
unsigned int i;
size_t total;
qemu_mutex_lock(&region.lock);
total = region.agg_size_full;
for (i = 0; i < n_ctxs; i++) {
const TCGContext *s = qatomic_read(&tcg_ctxs[i]);
size_t size;
size = qatomic_read(&s->code_gen_ptr) - s->code_gen_buffer;
g_assert(size <= s->code_gen_buffer_size);
total += size;
}
qemu_mutex_unlock(&region.lock);
return total;
}
/*
* Returns the code capacity (in bytes) of the entire cache, i.e. including all
* regions.
* See also: tcg_code_size()
*/
size_t tcg_code_capacity(void)
{
size_t guard_size, capacity;
/* no need for synchronization; these variables are set at init time */
guard_size = region.stride - region.size;
capacity = region.end + guard_size - region.start;
capacity -= region.n * (guard_size + TCG_HIGHWATER);
return capacity;
}
size_t tcg_tb_phys_invalidate_count(void)
{
unsigned int n_ctxs = qatomic_read(&n_tcg_ctxs);
unsigned int i;
size_t total = 0;
for (i = 0; i < n_ctxs; i++) {
const TCGContext *s = qatomic_read(&tcg_ctxs[i]);
total += qatomic_read(&s->tb_phys_invalidate_count);
}
return total;
}
/* pool based memory allocation */
void *tcg_malloc_internal(TCGContext *s, int size)
{
@ -1101,8 +551,9 @@ static void process_op_defs(TCGContext *s);
static TCGTemp *tcg_global_reg_new_internal(TCGContext *s, TCGType type,
TCGReg reg, const char *name);
void tcg_context_init(TCGContext *s)
static void tcg_context_init(unsigned max_cpus)
{
TCGContext *s = &tcg_init_ctx;
int op, total_args, n, i;
TCGOpDef *def;
TCGArgConstraint *args_ct;
@ -1167,11 +618,11 @@ void tcg_context_init(TCGContext *s)
*/
#ifdef CONFIG_USER_ONLY
tcg_ctxs = &tcg_ctx;
n_tcg_ctxs = 1;
tcg_cur_ctxs = 1;
tcg_max_ctxs = 1;
#else
MachineState *ms = MACHINE(qdev_get_machine());
unsigned int max_cpus = ms->smp.max_cpus;
tcg_ctxs = g_new(TCGContext *, max_cpus);
tcg_max_ctxs = max_cpus;
tcg_ctxs = g_new0(TCGContext *, max_cpus);
#endif
tcg_debug_assert(!tcg_regset_test_reg(s->reserved_regs, TCG_AREG0));
@ -1179,6 +630,12 @@ void tcg_context_init(TCGContext *s)
cpu_env = temp_tcgv_ptr(ts);
}
void tcg_init(size_t tb_size, int splitwx, unsigned max_cpus)
{
tcg_context_init(max_cpus);
tcg_region_init(tb_size, splitwx, max_cpus);
}
/*
* Allocate TBs right before their corresponding translated code, making
* sure that TBs and code are on different cache lines.
@ -1206,32 +663,16 @@ TranslationBlock *tcg_tb_alloc(TCGContext *s)
void tcg_prologue_init(TCGContext *s)
{
size_t prologue_size, total_size;
void *buf0, *buf1;
size_t prologue_size;
/* Put the prologue at the beginning of code_gen_buffer. */
buf0 = s->code_gen_buffer;
total_size = s->code_gen_buffer_size;
s->code_ptr = buf0;
s->code_buf = buf0;
s->code_ptr = s->code_gen_ptr;
s->code_buf = s->code_gen_ptr;
s->data_gen_ptr = NULL;
/*
* The region trees are not yet configured, but tcg_splitwx_to_rx
* needs the bounds for an assert.
*/
region.start = buf0;
region.end = buf0 + total_size;
#ifndef CONFIG_TCG_INTERPRETER
tcg_qemu_tb_exec = (tcg_prologue_fn *)tcg_splitwx_to_rx(buf0);
tcg_qemu_tb_exec = (tcg_prologue_fn *)tcg_splitwx_to_rx(s->code_ptr);
#endif
/* Compute a high-water mark, at which we voluntarily flush the buffer
and start over. The size here is arbitrary, significantly larger
than we expect the code generation for any one opcode to require. */
s->code_gen_highwater = s->code_gen_buffer + (total_size - TCG_HIGHWATER);
#ifdef TCG_TARGET_NEED_POOL_LABELS
s->pool_labels = NULL;
#endif
@ -1248,32 +689,25 @@ void tcg_prologue_init(TCGContext *s)
}
#endif
buf1 = s->code_ptr;
prologue_size = tcg_current_code_size(s);
#ifndef CONFIG_TCG_INTERPRETER
flush_idcache_range((uintptr_t)tcg_splitwx_to_rx(buf0), (uintptr_t)buf0,
tcg_ptr_byte_diff(buf1, buf0));
flush_idcache_range((uintptr_t)tcg_splitwx_to_rx(s->code_buf),
(uintptr_t)s->code_buf, prologue_size);
#endif
/* Deduct the prologue from the buffer. */
prologue_size = tcg_current_code_size(s);
s->code_gen_ptr = buf1;
s->code_gen_buffer = buf1;
s->code_buf = buf1;
total_size -= prologue_size;
s->code_gen_buffer_size = total_size;
tcg_register_jit(tcg_splitwx_to_rx(s->code_gen_buffer), total_size);
tcg_region_prologue_set(s);
#ifdef DEBUG_DISAS
if (qemu_loglevel_mask(CPU_LOG_TB_OUT_ASM)) {
FILE *logfile = qemu_log_lock();
qemu_log("PROLOGUE: [size=%zu]\n", prologue_size);
if (s->data_gen_ptr) {
size_t code_size = s->data_gen_ptr - buf0;
size_t code_size = s->data_gen_ptr - s->code_gen_ptr;
size_t data_size = prologue_size - code_size;
size_t i;
log_disas(buf0, code_size);
log_disas(s->code_gen_ptr, code_size);
for (i = 0; i < data_size; i += sizeof(tcg_target_ulong)) {
if (sizeof(tcg_target_ulong) == 8) {
@ -1287,7 +721,7 @@ void tcg_prologue_init(TCGContext *s)
}
}
} else {
log_disas(buf0, prologue_size);
log_disas(s->code_gen_ptr, prologue_size);
}
qemu_log("\n");
qemu_log_flush();
@ -2649,6 +2083,19 @@ void tcg_op_remove(TCGContext *s, TCGOp *op)
#endif
}
void tcg_remove_ops_after(TCGOp *op)
{
TCGContext *s = tcg_ctx;
while (true) {
TCGOp *last = tcg_last_op();
if (last == op) {
return;
}
tcg_op_remove(s, last);
}
}
static TCGOp *tcg_op_alloc(TCGOpcode opc)
{
TCGContext *s = tcg_ctx;
@ -4480,7 +3927,7 @@ static void tcg_reg_alloc_call(TCGContext *s, TCGOp *op)
static inline
void tcg_profile_snapshot(TCGProfile *prof, bool counters, bool table)
{
unsigned int n_ctxs = qatomic_read(&n_tcg_ctxs);
unsigned int n_ctxs = qatomic_read(&tcg_cur_ctxs);
unsigned int i;
for (i = 0; i < n_ctxs; i++) {
@ -4543,7 +3990,7 @@ void tcg_dump_op_count(void)
int64_t tcg_cpu_exec_time(void)
{
unsigned int n_ctxs = qatomic_read(&n_tcg_ctxs);
unsigned int n_ctxs = qatomic_read(&tcg_cur_ctxs);
unsigned int i;
int64_t ret = 0;

1
tcg/tci/tcg-target.h
View File

@ -43,6 +43,7 @@
#define TCG_TARGET_INTERPRETER 1
#define TCG_TARGET_INSN_UNIT_SIZE 1
#define TCG_TARGET_TLB_DISPLACEMENT_BITS 32
#define MAX_CODE_GEN_BUFFER_SIZE ((size_t)-1)
#if UINTPTR_MAX == UINT32_MAX
# define TCG_TARGET_REG_BITS 32

9
util/osdep.c
View File

@ -97,6 +97,15 @@ static int qemu_mprotect__osdep(void *addr, size_t size, int prot)
#endif
}
int qemu_mprotect_rw(void *addr, size_t size)
{
#ifdef _WIN32
return qemu_mprotect__osdep(addr, size, PAGE_READWRITE);
#else
return qemu_mprotect__osdep(addr, size, PROT_READ | PROT_WRITE);
#endif
}
int qemu_mprotect_rwx(void *addr, size_t size)
{
#ifdef _WIN32