Wolfgang Denk wrote:
In message c166aa9f05091309552efac7d@mail.gmail.com you wrote:
Not on MIPS (unless someone was daft enough to map it with the TLB). The code causes a TLB store miss exception. Since this is in 'common' code, I think it's a bug.
OK.
But all you need to do is #define CFG_MEMTEST_SCRATCH in your board config file, right?
With a static define of the scratch address I can't test one of the memories correctly becuase the scratch address is pointing to the other memory, so the scratch writes don't serve their purpose.
OK, but this is a special case, isn't it?
Would it be acceptable for you if we just move the initialization
dummy = (vu_long*)CFG_MEMTEST_SCRATCH;
below the computation of "end", so that you can use
#define CFG_MEMTEST_SCRATCH (--end)
in your board config file?
Best regards,
Wolfgang Denk
I'm not sure you want that, exactly. You are affecting the variable "end" as a side effect in a macro which is generally a bad idea. A no side effect version would be: #define CFG_MEMTEST_SCRATCH (end - 1)
You are also making assumptions on the type of end (the variable, that is). I'm assuming this is a 32 bit wide memory test, sorry for not verifying that. You may want to coerce "end" to make sure the "- 1" subtracts the proper number of bytes: #define CFG_MEMTEST_SCRATCH ((void *)end - 1)
gvb