Hello Dan, I've run into a problem with the h/w accelerated RAID-5 driver (on the ppc440spe-based board). After some investigations I've come to conclusion that the issue is with the async_tx_submit() implementation in ASYNC_TX. Let's we have the following 3 dependent operations: OP1, OP2, OP3. The OP1 operation has to be run on one, DMA0, channel, OP2 and OP3 operations have to be run on another, DMA1, channel. OP2 has to be run after completion of OP1, OP3 - after completion of OP2. I. e. : OP1 (DMA0) <--- OP2 (DMA1) <--- OP3 (DMA1). With the current implementation of async_tx_submit() we have the following picture when we sequentially call async_tx_submit() to submit these 3 operations: - the OP2 operation will *not be* submitted to the h/w channel, but, because of the (OP1->chan != OP2->chan) condition, will be put to depend_list of OP1; and it's RIGHT; - the OP3 operation will *be* submitted to the h/w channel, because OP2 & OP3 do not conform the (OP2->chan != OP3->chan) condition. Thus OP3 will be put to the h/w channel *before* its parent, OP2; and it's WRONG. It looks more logical for me to check in async_tx_submit() - does the dep_tx operation (on which the current tx operation is dependent on) depend on some other operation itself. If dep_tx depends on some other operation then, regardless of the channel on which tx has to be run, we should not put tx to the h/w but bind it to depend_list of dep_tx instead. Why there are no such algorithms in async_tx_submit() ? As far as the RAID-5 h/w accelerated driver is concerned this means that with the current implementation of async_tx_submit(), for example, the (PREXOR + BIODRAIN) combination works incorrectly for the cases when BIODRAIN is split into a sequence of MEMCPY operations and there are two different channels for XOR and COPY operations: only one, the first, MEMCPY will wait until completion of PREXOR, but all the others MEMCPYs of this BIODRAIN will be put into the h/w channel right away. Is it correct ? I think - not. Regards, Yuri -- Yuri Tikhonov, Senior Software Engineer Emcraft Systems, www.emcraft.com - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
