> Several of us have reported a degenerate mode, that > I term as "tmf timeout", in which a aic94xx based card > becomes inoperable. Alas, the same hardware running another > OS does not exhibit that problem (or at least not as much). > > asd_write_reg_word(asd_ha, LmSEQ_INTEN_SAVE(lseq), > > - (u16) LmM0INTEN_MASK & 0xFFFF0000 >> 16); > > + (u16) ((LmM0INTEN_MASK & 0xFFFF0000) >> 16)); > > asd_write_reg_word(asd_ha, LmSEQ_INTEN_SAVE(lseq) + 2, > > (u16) LmM0INTEN_MASK & 0xFFFF); > > asd_write_reg_byte(asd_ha, LmSEQ_LINK_RST_FRM_LEN(lseq), 0); > BTW Luben was pointing out that the call you patched > and the following call can be combined into a less > trouble prone asd_write_reg_dword() call. In that case there's another bug - we should write upper 16 bits to addr + 2, not the lower ones. IOW, the old code was broken attempt to write upper 16 bits to addr (ends up writing _lower_ 16 bits) writing lower 16 bits to addr + 2 With this patch we get the first call do what it clearly intended to do (unless it's a deliberate obfuscation from hell). _IF_ we really want to write the damn thing little-endian, the order should be reverted on top of that. I.e. asd_write_reg_word(asd_ha, LmSEQ_INTEN_SAVE(lseq), (u16) LmM0INTEN_MASK & 0xFFFF); asd_write_reg_word(asd_ha, LmSEQ_INTEN_SAVE(lseq) + 2, (u16) ((LmM0INTEN_MASK & 0xFFFF0000) >> 16)); or, indeed, asd_write_reg_dword(). - To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
