diff mbox series

util/cacheflush: Avoid flushing dcache twice when not necessary

Message ID 20230605195911.96033-1-philmd@linaro.org
State New
Headers show
Series util/cacheflush: Avoid flushing dcache twice when not necessary | expand

Commit Message

Philippe Mathieu-Daudé June 5, 2023, 7:59 p.m. UTC
<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>
---
Based-on: <20230605175647.88395-2-philmd@linaro.org>
---
 util/cacheflush.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

Comments

Philippe Mathieu-Daudé June 5, 2023, 8 p.m. UTC | #1
Subject: "Avoid flushing dcache twice [on Darwin] when not necessary"

On 5/6/23 21:59, Philippe Mathieu-Daudé wrote:
> <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>
> ---
> Based-on: <20230605175647.88395-2-philmd@linaro.org>
> ---
>   util/cacheflush.c | 9 ++++++++-
>   1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/util/cacheflush.c b/util/cacheflush.c
> index de35616718..a08906155a 100644
> --- a/util/cacheflush.c
> +++ b/util/cacheflush.c
> @@ -241,7 +241,14 @@ static void __attribute__((constructor)) init_cache_info(void)
>   
>   void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
>   {
> -    sys_dcache_flush((void *)rw, 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
Richard Henderson June 5, 2023, 8:03 p.m. UTC | #2
On 6/5/23 12:59, Philippe Mathieu-Daudé wrote:
> <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>
> ---
> Based-on:<20230605175647.88395-2-philmd@linaro.org>
> ---
>   util/cacheflush.c | 9 ++++++++-
>   1 file changed, 8 insertions(+), 1 deletion(-)

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>


r~
BALATON Zoltan June 5, 2023, 9:56 p.m. UTC | #3
On Mon, 5 Jun 2023, Philippe Mathieu-Daudé wrote:
> <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>
> ---
> Based-on: <20230605175647.88395-2-philmd@linaro.org>
> ---
> util/cacheflush.c | 9 ++++++++-
> 1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/util/cacheflush.c b/util/cacheflush.c
> index de35616718..a08906155a 100644
> --- a/util/cacheflush.c
> +++ b/util/cacheflush.c
> @@ -241,7 +241,14 @@ static void __attribute__((constructor)) init_cache_info(void)
>
> void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
> {
> -    sys_dcache_flush((void *)rw, len);
> +    if (rx == rw) {

Isn't it more straight forward to use rx != rw and drop the else branch 
than having an empty if branch? You can still keep the comment above the 
if to explain it if needed.

Regards,
BALATON Zoltan

> +        /*
> +         * 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
>
Philippe Mathieu-Daudé June 5, 2023, 10:33 p.m. UTC | #4
On 5/6/23 23:56, BALATON Zoltan wrote:
> On Mon, 5 Jun 2023, Philippe Mathieu-Daudé wrote:
>> <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>
>> ---
>> Based-on: <20230605175647.88395-2-philmd@linaro.org>
>> ---
>> util/cacheflush.c | 9 ++++++++-
>> 1 file changed, 8 insertions(+), 1 deletion(-)
>>
>> diff --git a/util/cacheflush.c b/util/cacheflush.c
>> index de35616718..a08906155a 100644
>> --- a/util/cacheflush.c
>> +++ b/util/cacheflush.c
>> @@ -241,7 +241,14 @@ static void __attribute__((constructor)) 
>> init_cache_info(void)
>>
>> void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
>> {
>> -    sys_dcache_flush((void *)rw, len);
>> +    if (rx == rw) {
> 
> Isn't it more straight forward to use rx != rw and drop the else branch 
> than having an empty if branch? You can still keep the comment above the 
> if to explain it if needed.

I tried that first but found it was not obvious, so chose this
form because it seemed clearer to me.
diff mbox series

Patch

diff --git a/util/cacheflush.c b/util/cacheflush.c
index de35616718..a08906155a 100644
--- a/util/cacheflush.c
+++ b/util/cacheflush.c
@@ -241,7 +241,14 @@  static void __attribute__((constructor)) init_cache_info(void)
 
 void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
 {
-    sys_dcache_flush((void *)rw, 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