lpfc warnings on sparc64

Noticed this wgile compiling 2.6.27-rc7+git on sparc64 - maybe it is of interest.

CC [M] drivers/scsi/lpfc/lpfc_sli.o
drivers/scsi/lpfc/lpfc_sli.c: In function 'lpfc_sli_poll_fcp_ring':
drivers/scsi/lpfc/lpfc_sli.c:1365: warning: array subscript is above array bounds drivers/scsi/lpfc/lpfc_sli.c:1365: warning: array subscript is above array bounds drivers/scsi/lpfc/lpfc_sli.c: In function 'lpfc_sli_handle_fast_ring_event':
drivers/scsi/lpfc/lpfc_sli.c:1541: warning: array subscript is above array bounds drivers/scsi/lpfc/lpfc_sli.c:1541: warning: array subscript is above array bounds

Meelis Roos ([email protected])

As best I can tell this is a bogus warning ... it triggers on x86 as well, but I can't find any reason gcc should think there's an array bounds error.

James

Interesting:
Here's the code lines:

     ...
     irsp->un.ulpWord[5],
     *(((uint32_t *)irsp) + 6),
     ...

it's not tripping on the ulpWord[5] line - which is what you would think, but rather the "+6" line. Given that the structure size for irsp is certainly big enough, this has to be some loss of "object" context in the compiler.

E.g: it wouldn't surprise me if there's a base register that
contains the address of irsp (%reg) and that the un.ulpWord[5] results in a displacement on the pointer (e.g. "20(%reg)" ), and that the ptr+6 also results in something similar (e.g. "24(%reg)") - and the range checking code assumed the context un.ulpWord meant ?(%reg), and lost track of the type or object difference between the next instruction that happened to have a ? that was too big for the array.

-- james s