Date: Mon, 26 Oct 2015 05:24:44 -0700 From: Adrian Chadd <adrian.chadd@gmail.com> To: =?UTF-8?Q?Roger_Pau_Monn=C3=A9?= <roger.pau@citrix.com> Cc: =?UTF-8?Q?Roger_Pau_Monn=C3=A9?= <royger@freebsd.org>, "src-committers@freebsd.org" <src-committers@freebsd.org>, "svn-src-all@freebsd.org" <svn-src-all@freebsd.org>, "svn-src-head@freebsd.org" <svn-src-head@freebsd.org> Subject: Re: svn commit: r289834 - head/sys/x86/x86 Message-ID: <CAJ-Vmok2WnSe2DA6jf6eLsL=OGsUD2_1LgLFRJi3LAhYk5aMwA@mail.gmail.com> In-Reply-To: <562A564D.9080703@citrix.com> References: <201510231539.t9NFdxPe018130@repo.freebsd.org> <562A564D.9080703@citrix.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi, I've started seeing panics on -head with ATA code doing a dmamap load -> panic. I'll test by reverting this patch and see what happens, but when it /does/ happen I can't get a crashdump, so debugging will be less easy. Has anyone else seen this? -a On 23 October 2015 at 08:46, Roger Pau Monn=C3=A9 <roger.pau@citrix.com> wr= ote: > El 23/10/15 a les 17.39, Roger Pau Monn=C3=A9 ha escrit: >> Author: royger >> Date: Fri Oct 23 15:39:59 2015 >> New Revision: 289834 >> URL: https://svnweb.freebsd.org/changeset/base/289834 >> >> Log: >> x86/dma_bounce: rework _bus_dmamap_load_ma implementation >> >> The implementation of bus_dmamap_load_ma_triv currently calls >> _bus_dmamap_load_phys on each page that is part of the passed in buffe= r. >> Since each page is treated as an individual buffer, the resulting beha= viour >> is different from the behaviour of _bus_dmamap_load_buffer. This break= s >> certain drivers, like Xen blkfront. >> >> If an unmapped buffer of size 4096 that starts at offset 13 into the f= irst >> page is passed to the current _bus_dmamap_load_ma implementation (so t= he ma >> array contains two pages), the result is that two segments are created= , one >> with a size of 4083 and the other with size 13 (because two independan= t >> calls to _bus_dmamap_load_phys are performed, one for each physical pa= ge). >> If the same is done with a mapped buffer and calling _bus_dmamap_load_= buffer >> the result is that only one segment is created, with a size of 4096. >> >> This patch relegates the usage of bus_dmamap_load_ma_triv in x86 bounc= e >> buffer code to drivers requesting BUS_DMA_KEEP_PG_OFFSET and implement= s >> _bus_dmamap_load_ma so that it's behaviour is the same as the mapped v= ersion >> (_bus_dmamap_load_buffer). This patch only modifies the x86 bounce buf= fer >> code, other arches are left untouched. >> >> Reviewed by: kib, jah >> Differential Revision: https://reviews.freebsd.org/D888 >> Sponsored by: Citrix Systems R&D > > Forgot to add: > > MFC after: 2 weeks > > Roger. >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAJ-Vmok2WnSe2DA6jf6eLsL=OGsUD2_1LgLFRJi3LAhYk5aMwA>