The user's test case assigns UINT8_MAX to an object of type uint8_t:

uint8_t tmp = UINT8_MAX; /* gets MISRA diagnostic */
tmp = 0xFFu; /* no diagnostic */

The compiler emits a MISRA-C:2004 diagnostic on the first line, but not the second:

% armcl --check_misra --strict_ansi --verbose_diagnostics try1.c
"try1.c", line 9: warning: (MISRA-C:2004 10.1/R) The value of an expression of
integer type shall not be implicitly converted to a different
underlying type if it is not a conversion to a wider integer type of
the same signedness
uint8_t tmp = UINT8_MAX;

This diagnostic is correct according to the rules for MISRA-C:2004.

Somewhat confusingly, the type of UINT8_MAX is not uint8_t.

C99 7.18. "Limits of specified-width integer types" says: "this expression shall have the same type as would an expression that is an object of the corresponding type converted according to the integer promotions." Therefore this macro has a value of type "signed int," and one gets a MISRA-2004 warning for trying to assign it directly to a variable of type uint8_t.

On the other hand, MISRA-2004 has a special case for constants. MISRA-2004 specifically allows the second case "tmp = 0xffu", because it is a literal constant with a value that fits into x without changing the value. MISRA-2004 does not make an exception for macros that have a value that fits, nor does it recognize UINT8_MAX as a special case, as it is a C99 feature.