Received: from eggs.gnu.org ([2001:4830:134:3::10]:49057) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XPpPZ-0008KM-Cw for qemu-devel@nongnu.org; Fri, 05 Sep 2014 05:02:35 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1XPpPT-0004G6-5v for qemu-devel@nongnu.org; Fri, 05 Sep 2014 05:02:29 -0400 Received: from mx1.redhat.com ([209.132.183.28]:8764) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XPpPS-0004Fy-OR for qemu-devel@nongnu.org; Fri, 05 Sep 2014 05:02:23 -0400 From: Markus Armbruster References: <1408050186-27800-1-git-send-email-jsnow@redhat.com> <20140818131229.GB1273@stefanha-thinkpad.redhat.com> <53F3A366.1010307@redhat.com> <20140820142716.GA3673@stefanha-thinkpad.redhat.com> <5408A76E.1080501@redhat.com> Date: Fri, 05 Sep 2014 11:02:17 +0200 In-Reply-To: <5408A76E.1080501@redhat.com> (John Snow's message of "Thu, 04 Sep 2014 13:54:54 -0400") Message-ID: <87fvg6jv8m.fsf@blackfin.pond.sub.org> MIME-Version: 1.0 Content-Type: text/plain Subject: Re: [Qemu-devel] [PATCH v3] ide: Add resize callback to ide/core List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: John Snow Cc: Kevin Wolf , Stefan Hajnoczi , qemu-devel John Snow writes: > On 09/04/2014 12:13 PM, Stefan Hajnoczi wrote: >> This patch seems to break tests/bios-tables-test.c: >> ERROR:tests/bios-tables-test.c:744:test_acpi_one: assertion failed >> (signature == SIGNATURE): (0x00000000 == 0x0000dead) >> GTester: last random seed: R02S3d881198f35228a485b4c3d116dff3b1 >> >> I have run it many times to make sure the failure is deterministic and >> I used git-bisect(1) to find this commit as the cause. >> >> Not sure why but it seems to break the test. Please take a look. >> >> Dropped from the block branch. >> >> Stefan >> > > I've fixed it in a v4, but before I submit and while I'm at it, you > didn't like the comments I left in the identify functions because they > were "prone to bitrot." would you prefer I excised them entirely? > > I found them helpful because it makes sense to read these functions > alongside the identify data specs, and excising any references to > those word indices in their natural order obfuscates the code > needlessly. > > My inclination is to leave them in. I'd replace them by a pointer to the factored-out code. > Meanwhile, the bios-tables-test problem is such: > We serve the identify request out of the io_buffer, not the > identify_data cache, thus for 2nd and subsequent requests for > identify_data, we get correct size information, but for the 1st > request to ide_identify, we get zeroes. I found that part confusing and stared at it until I believed it works. Looks like I believed too quickly. > I corrected this by making > ide_identify and ide_atapi_identify mimic the flow of > ide_cfata_identify, which is more clear about the nature of the two > buffers. Yup, that one's much better. > Why this causes a failure here and for only the Q35 machine type I am > not certain, but this is the causative bug. My best guess is different firmware behavior.