qemu/util/cacheflush.c
Philippe Mathieu-Daudé bb6af0fa51 util/cacheflush: Avoid possible redundant dcache flush on Darwin
<libkern/OSCacheControl.h> describes sys_icache_invalidate() as
"equivalent to sys_cache_control(kCacheFunctionPrepareForExecution)",
having kCacheFunctionPrepareForExecution defined as:

  /* Prepare memory for execution.  This should be called
   * after writing machine instructions to memory, before
   * executing them.  It syncs the dcache and icache. [...]
   */

Since the dcache is also sync'd, we can avoid the sys_dcache_flush()
call when both rx/rw pointers are equal.

Suggested-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-Id: <20230605195911.96033-1-philmd@linaro.org>
2023-06-13 11:28:58 +02:00

377 lines
9.9 KiB
C

/*
* Info about, and flushing the host cpu caches.
*
* This work is licensed under the terms of the GNU GPL, version 2 or later.
* See the COPYING file in the top-level directory.
*/
#include "qemu/osdep.h"
#include "qemu/cacheflush.h"
#include "qemu/cacheinfo.h"
#include "qemu/bitops.h"
#include "qemu/host-utils.h"
#include "qemu/atomic.h"
int qemu_icache_linesize = 0;
int qemu_icache_linesize_log;
int qemu_dcache_linesize = 0;
int qemu_dcache_linesize_log;
/*
* Operating system specific cache detection mechanisms.
*/
#if defined(_WIN32)
static void sys_cache_info(int *isize, int *dsize)
{
SYSTEM_LOGICAL_PROCESSOR_INFORMATION *buf;
DWORD size = 0;
BOOL success;
size_t i, n;
/*
* Check for the required buffer size first. Note that if the zero
* size we use for the probe results in success, then there is no
* data available; fail in that case.
*/
success = GetLogicalProcessorInformation(0, &size);
if (success || GetLastError() != ERROR_INSUFFICIENT_BUFFER) {
return;
}
n = size / sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION);
size = n * sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION);
buf = g_new0(SYSTEM_LOGICAL_PROCESSOR_INFORMATION, n);
if (!GetLogicalProcessorInformation(buf, &size)) {
goto fail;
}
for (i = 0; i < n; i++) {
if (buf[i].Relationship == RelationCache
&& buf[i].Cache.Level == 1) {
switch (buf[i].Cache.Type) {
case CacheUnified:
*isize = *dsize = buf[i].Cache.LineSize;
break;
case CacheInstruction:
*isize = buf[i].Cache.LineSize;
break;
case CacheData:
*dsize = buf[i].Cache.LineSize;
break;
default:
break;
}
}
}
fail:
g_free(buf);
}
#elif defined(CONFIG_DARWIN)
# include <sys/sysctl.h>
static void sys_cache_info(int *isize, int *dsize)
{
/* There's only a single sysctl for both I/D cache line sizes. */
long size;
size_t len = sizeof(size);
if (!sysctlbyname("hw.cachelinesize", &size, &len, NULL, 0)) {
*isize = *dsize = size;
}
}
#elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
# include <sys/sysctl.h>
static void sys_cache_info(int *isize, int *dsize)
{
/* There's only a single sysctl for both I/D cache line sizes. */
int size;
size_t len = sizeof(size);
if (!sysctlbyname("machdep.cacheline_size", &size, &len, NULL, 0)) {
*isize = *dsize = size;
}
}
#else
/* POSIX */
static void sys_cache_info(int *isize, int *dsize)
{
# ifdef _SC_LEVEL1_ICACHE_LINESIZE
int tmp_isize = (int) sysconf(_SC_LEVEL1_ICACHE_LINESIZE);
if (tmp_isize > 0) {
*isize = tmp_isize;
}
# endif
# ifdef _SC_LEVEL1_DCACHE_LINESIZE
int tmp_dsize = (int) sysconf(_SC_LEVEL1_DCACHE_LINESIZE);
if (tmp_dsize > 0) {
*dsize = tmp_dsize;
}
# endif
}
#endif /* sys_cache_info */
/*
* Architecture (+ OS) specific cache detection mechanisms.
*/
#if defined(__powerpc__)
static bool have_coherent_icache;
#endif
#if defined(__aarch64__) && !defined(CONFIG_DARWIN) && !defined(CONFIG_WIN32)
/*
* Apple does not expose CTR_EL0, so we must use system interfaces.
* Windows neither, but we use a generic implementation of flush_idcache_range
* in this case.
*/
static uint64_t save_ctr_el0;
static void arch_cache_info(int *isize, int *dsize)
{
uint64_t ctr;
/*
* The real cache geometry is in CCSIDR_EL1/CLIDR_EL1/CSSELR_EL1,
* but (at least under Linux) these are marked protected by the
* kernel. However, CTR_EL0 contains the minimum linesize in the
* entire hierarchy, and is used by userspace cache flushing.
*
* We will also use this value in flush_idcache_range.
*/
asm volatile("mrs\t%0, ctr_el0" : "=r"(ctr));
save_ctr_el0 = ctr;
if (*isize == 0 || *dsize == 0) {
if (*isize == 0) {
*isize = 4 << (ctr & 0xf);
}
if (*dsize == 0) {
*dsize = 4 << ((ctr >> 16) & 0xf);
}
}
}
#elif defined(_ARCH_PPC) && defined(__linux__)
# include "elf.h"
static void arch_cache_info(int *isize, int *dsize)
{
if (*isize == 0) {
*isize = qemu_getauxval(AT_ICACHEBSIZE);
}
if (*dsize == 0) {
*dsize = qemu_getauxval(AT_DCACHEBSIZE);
}
have_coherent_icache = qemu_getauxval(AT_HWCAP) & PPC_FEATURE_ICACHE_SNOOP;
}
#else
static void arch_cache_info(int *isize, int *dsize) { }
#endif /* arch_cache_info */
/*
* ... and if all else fails ...
*/
static void fallback_cache_info(int *isize, int *dsize)
{
/* If we can only find one of the two, assume they're the same. */
if (*isize) {
if (*dsize) {
/* Success! */
} else {
*dsize = *isize;
}
} else if (*dsize) {
*isize = *dsize;
} else {
#if defined(_ARCH_PPC)
/*
* For PPC, we're going to use the cache sizes computed for
* flush_idcache_range. Which means that we must use the
* architecture minimum.
*/
*isize = *dsize = 16;
#else
/* Otherwise, 64 bytes is not uncommon. */
*isize = *dsize = 64;
#endif
}
}
static void __attribute__((constructor)) init_cache_info(void)
{
int isize = 0, dsize = 0;
sys_cache_info(&isize, &dsize);
arch_cache_info(&isize, &dsize);
fallback_cache_info(&isize, &dsize);
assert((isize & (isize - 1)) == 0);
assert((dsize & (dsize - 1)) == 0);
qemu_icache_linesize = isize;
qemu_icache_linesize_log = ctz32(isize);
qemu_dcache_linesize = dsize;
qemu_dcache_linesize_log = ctz32(dsize);
qatomic64_init();
}
/*
* Architecture (+ OS) specific cache flushing mechanisms.
*/
#if defined(__i386__) || defined(__x86_64__) || defined(__s390__)
/* Caches are coherent and do not require flushing; symbol inline. */
#elif defined(__aarch64__) && !defined(CONFIG_WIN32)
/*
* For Windows, we use generic implementation of flush_idcache_range, that
* performs a call to FlushInstructionCache, through __builtin___clear_cache.
*/
#ifdef CONFIG_DARWIN
/* Apple does not expose CTR_EL0, so we must use system interfaces. */
#include <libkern/OSCacheControl.h>
void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
{
if (rx == rw) {
/*
* sys_icache_invalidate() syncs the dcache and icache,
* so no need to call sys_dcache_flush().
*/
} else {
sys_dcache_flush((void *)rw, len);
}
sys_icache_invalidate((void *)rx, len);
}
#else
/*
* This is a copy of gcc's __aarch64_sync_cache_range, modified
* to fit this three-operand interface.
*/
void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
{
const unsigned CTR_IDC = 1u << 28;
const unsigned CTR_DIC = 1u << 29;
const uint64_t ctr_el0 = save_ctr_el0;
const uintptr_t icache_lsize = qemu_icache_linesize;
const uintptr_t dcache_lsize = qemu_dcache_linesize;
uintptr_t p;
/*
* If CTR_EL0.IDC is enabled, Data cache clean to the Point of Unification
* is not required for instruction to data coherence.
*/
if (!(ctr_el0 & CTR_IDC)) {
/*
* Loop over the address range, clearing one cache line at once.
* Data cache must be flushed to unification first to make sure
* the instruction cache fetches the updated data.
*/
for (p = rw & -dcache_lsize; p < rw + len; p += dcache_lsize) {
asm volatile("dc\tcvau, %0" : : "r" (p) : "memory");
}
asm volatile("dsb\tish" : : : "memory");
}
/*
* If CTR_EL0.DIC is enabled, Instruction cache cleaning to the Point
* of Unification is not required for instruction to data coherence.
*/
if (!(ctr_el0 & CTR_DIC)) {
for (p = rx & -icache_lsize; p < rx + len; p += icache_lsize) {
asm volatile("ic\tivau, %0" : : "r"(p) : "memory");
}
asm volatile ("dsb\tish" : : : "memory");
}
asm volatile("isb" : : : "memory");
}
#endif /* CONFIG_DARWIN */
#elif defined(__mips__)
#ifdef __OpenBSD__
#include <machine/sysarch.h>
#else
#include <sys/cachectl.h>
#endif
void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
{
if (rx != rw) {
cacheflush((void *)rw, len, DCACHE);
}
cacheflush((void *)rx, len, ICACHE);
}
#elif defined(__powerpc__)
void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
{
uintptr_t p, b, e;
size_t dsize, isize;
/*
* Some processors have coherent caches and support a simplified
* flushing procedure. See
* POWER9 UM, 4.6.2.2 Instruction Cache Block Invalidate (icbi)
* https://ibm.ent.box.com/s/tmklq90ze7aj8f4n32er1mu3sy9u8k3k
*/
if (have_coherent_icache) {
asm volatile ("sync\n\t"
"icbi 0,%0\n\t"
"isync"
: : "r"(rx) : "memory");
return;
}
dsize = qemu_dcache_linesize;
isize = qemu_icache_linesize;
b = rw & ~(dsize - 1);
e = (rw + len + dsize - 1) & ~(dsize - 1);
for (p = b; p < e; p += dsize) {
asm volatile ("dcbst 0,%0" : : "r"(p) : "memory");
}
asm volatile ("sync" : : : "memory");
b = rx & ~(isize - 1);
e = (rx + len + isize - 1) & ~(isize - 1);
for (p = b; p < e; p += isize) {
asm volatile ("icbi 0,%0" : : "r"(p) : "memory");
}
asm volatile ("sync" : : : "memory");
asm volatile ("isync" : : : "memory");
}
#elif defined(__sparc__)
void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
{
/* No additional data flush to the RW virtual address required. */
uintptr_t p, end = (rx + len + 7) & -8;
for (p = rx & -8; p < end; p += 8) {
__asm__ __volatile__("flush\t%0" : : "r" (p));
}
}
#else
void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
{
if (rw != rx) {
__builtin___clear_cache((char *)rw, (char *)rw + len);
}
__builtin___clear_cache((char *)rx, (char *)rx + len);
}
#endif