From mboxrd@z Thu Jan 1 00:00:00 1970 From: Pat LaVarre Subject: Re: [PATCH] atapi request sense work Date: 21 May 2004 11:59:08 -0600 Sender: linux-ide-owner@vger.kernel.org Message-ID: <1085162347.4791.24.camel@patibmrh9> References: <1084717146.3576.3.camel@patibmrh9> <40AD7FB2.10506@pobox.com><1 085153750.6103.33.camel@patibmrh9><200405211746.35015.bzolnier@elka.pw.edu. pl> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: Received: from email-out1.iomega.com ([147.178.1.82]:42422 "EHLO email.iomega.com") by vger.kernel.org with ESMTP id S265938AbUEUR7T (ORCPT ); Fri, 21 May 2004 13:59:19 -0400 Received: from royntex01.iomegacorp.com (edison [10.1.1.82]) by email.iomega.com (Postfix) with ESMTP id 263201B44 for ; Fri, 21 May 2004 11:59:19 -0600 (MDT) In-Reply-To: <200405211746.35015.bzolnier@elka.pw.edu.pl> List-Id: linux-ide@vger.kernel.org To: linux-ide@vger.kernel.org > Keep posting... seconded. Hi! Below inline is a trivial -bk8 patch that ducks ever trying PATAPI auto sense. In a kernel shipping with ATA_ENABLE_ATAPI defined, this would be a very bad thing, because we thus falsely report success when in fact we have not copied read data in or write data out. But at my desk I find this alternative more usable than vanilla 2.6.6-bk8, because this way I don't have to issue unsolicited op x03 "REQUEST SENSE" to clear out the always-initially-present unit attentions. Instead, I just ignore the first few garbage results I get. For example: $ plscsi /dev/sg0 -X time 5 0 -i 8 -x "25 00 00:00:00:00 00 00:00 00" -v x 00000000 25 00 00:00:00:00 00 00:00 00 .. .. .. .. .. .. "%@@@@@@@@@" x 00000000 00:00:00:00 00:00:00:00 .. .. .. .. .. .. .. .. "@@@@@@@@" // 0 = plscsi.main exit int $ $ plscsi /dev/sg0 -X time 5 0 -i 8 -x "25 00 00:00:00:00 00 00:00 00" -v x 00000000 25 00 00:00:00:00 00 00:00 00 .. .. .. .. .. .. "%@@@@@@@@@" x 00000000 01:04:C9:3F 00:00:08:00 .. .. .. .. .. .. .. .. "ADI?@@H@" // 0 = plscsi.main exit int $ See that? The first result is zeroed nonsense, courtesy the x 6 29 Reset unit attention we know was present. Only the second result is actually meaningful. I can now proceed to test how well our DMA no-data/ data-in/ data-out protocols work apart from timeout, auto sense, and the rest of error recovery. After that, I can return to inserting a PIO unsolicited op x03 "REQUEST SENSE" after every command, and then beyond that I can make the auto sense occur no more often than it should. Pat LaVarre diff -Nurp linux-2.6.6-bk8/drivers/scsi/libata-core.c linux-2.6.6-bk8-pel/drivers/scsi/libata-core.c --- linux-2.6.6-bk8/drivers/scsi/libata-core.c 2004-05-21 09:44:22.000000000 -0600 +++ linux-2.6.6-bk8-pel/drivers/scsi/libata-core.c 2004-05-21 11:29:01.571933824 -0600 @@ -2303,6 +2303,9 @@ void ata_qc_complete(struct ata_queued_c ata_sg_clean(qc); if (cmd) { +#if 1 /* pretend good */ + cmd->result = SAM_STAT_GOOD; +#else /* FIXME: admit lost write data, stale read data, etc. */ if (unlikely(drv_stat & (ATA_ERR | ATA_BUSY | ATA_DRQ))) { if (is_atapi_taskfile(&qc->tf)) cmd->result = SAM_STAT_CHECK_CONDITION; @@ -2311,6 +2314,7 @@ void ata_qc_complete(struct ata_queued_c } else { cmd->result = SAM_STAT_GOOD; } +#endif qc->scsidone(cmd); } diff -Nurp linux-2.6.6-bk8/drivers/scsi/libata-scsi.c linux-2.6.6-bk8-pel/drivers/scsi/libata-scsi.c