On Wednesday 30 January 2008, Rafal Jaworowski wrote:

Hi Mike,

...

It isn't generally save to execute applications outside of U-Boot with caches enabled due to the way the Blackfin processor handles caches (requires software assistance). This patch disables caches before booting an ELF or just booting raw code. The previous discussion on the patch was that we wanted to use weaks instead, but that proved to not be feasible when multiple symbols are involved, which puts us back at the ifdef solution. I've minimized the ugliness by moving the setup step outside of the main function.

Signed-off-by: Mike Frysinger vapier@gentoo.org

diff --git a/common/cmd_boot.c b/common/cmd_boot.c index e68f16f..9d4f026 100644 --- a/common/cmd_boot.c +++ b/common/cmd_boot.c @@ -32,6 +32,23 @@ DECLARE_GLOBAL_DATA_PTR; #endif

+static inline void go_setup(int argc, char *argv[]) +{ +#if defined(CONFIG_I386)

• /*

• * x86 does not use a dedicated register to pass the pointer
  

• * to the global_data
  

• */
  

• argv[0] = (char *)gd;

+#elif defined(CONFIG_BLACKFIN)

• if (dcache_status ())

• dcache_disable ();
  

• if (icache_status ())

• icache_disable ();
  

+#endif +}

int do_go (cmd_tbl_t *cmdtp, int flag, int argc, char *argv[]) { ulong addr, rc; @@ -46,25 +63,20 @@ int do_go (cmd_tbl_t *cmdtp, int flag, int argc, char *argv[])

printf ("## Starting application at 0x%08lX ...\n", addr);

• go_setup(argc, argv);

+#if defined(CONFIG_NIOS) /*

• * pass address parameter as argv[0] (aka command name),
  

• * and all remaining args
  

• */
  

-#if defined(CONFIG_I386)

• /*

• * x86 does not use a dedicated register to pass the pointer
  

• * to the global_data
  

• * Nios function pointers are address >> 1
  

  */

• argv[0] = (char *)gd;

• addr >>= 1;

#endif -#if !defined(CONFIG_NIOS)

• rc = ((ulong (*)(int, char *[]))addr) (--argc, &argv[1]);

-#else

• /*

• * Nios function pointers are address >> 1
  

• * pass address parameter as argv[0] (aka command name),
  

• * and all remaining args
  

  */

• rc = ((ulong (*)(int, char *[]))(addr>>1)) (--argc, &argv[1]);

-#endif

• rc = ((ulong (*)(int, char *[]))addr) (--argc, &argv[1]); if (rc != 0) rcode = 1;

  printf ("## Application terminated, rc = 0x%lX\n", rc);

diff --git a/common/cmd_elf.c b/common/cmd_elf.c index 2eb7453..4683554 100644 --- a/common/cmd_elf.c +++ b/common/cmd_elf.c @@ -27,6 +27,21 @@ DECLARE_GLOBAL_DATA_PTR; #define MAX(a,b) ((a) > (b) ? (a) : (b)) #endif

+static inline void bootelf_setup(int argc, char *argv[]) +{

• /*

• * QNX images require the data cache is disabled.
  

• * Data cache is already flushed, so just turn it off.
  

• */
  

• if (dcache_status ())

• dcache_disable ();
  

Please re-formulate the comment, so it doesn't only refer to QNX, as per our recent discussion disabling d-cache before running an external app is considered a generic U-Boot operation, not platform specific, see this thread:

i dont know anything about the topic. feel free to submit a follow up patch. i merely relocated the existing comment.

Also, I'm wondering if we shouldn't re-enable the cache after returning from an external app (as they are allowed to do so)? Without doing it we end up with degraded performance without any notice, require explicit reset to return to previous state etc.

i did see that deficiency in the code ... that'd require the code to stay inline in the function :/ -mike