Date: Tue, 22 Dec 2015 11:37:23 +0200 From: Max Gurtovoy <maxg@mellanox.com> To: Sagi Grimberg <sagig@dev.mellanox.co.il>, Konstantin Belousov <kostikbel@gmail.com> Cc: Sagi Grimberg <sagig@mellanox.com>, <freebsd-scsi@freebsd.org>, "Hans Petter Selasky" <hanss@mellanox.com>, Oren Duer <oren@mellanox.com> Subject: Re: splitting iovecs to bios Message-ID: <567919D3.6050004@mellanox.com> In-Reply-To: <566D2C91.9050900@dev.mellanox.co.il> References: <56696E03.8050202@mellanox.com> <20151210150210.GY82577@kib.kiev.ua> <566D2C91.9050900@dev.mellanox.co.il>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi Konstantin, > >> There might be indeed a reason, it could be that some drivers expect >> blocking to be done by the userspace. The drivers could have some >> restrictions on transfer sizes and atomicity of transfer, which would >> be broken by the unconditional merge. I cannot give you an example >> of such driver, known block-aware drivers like sa(4) only require the >> bio size to be multiple of the basic block size. > > I'm surprised to learn that the generic access layer splits IO requests > just because some block drivers cannot handle it. I'd expect that this > sort of limitation would be communicated by the drivers in the form of > device flag SI_NOMERGE. > >> OTOH, I see no issue with adding a SI_PHYSIOMERGE flag and doing the >> merges for the driver in physio(), when unmapped request has consequtive >> iov elements ending and starting at the page boundary. > > I'd say it should be the other way around, physio would always strive > to append/merge iov elements but wouldn't in case the device does not > support it. Moreover, some modern devices does not even require the page > boundary alignment you mentioned. These devices can execute IO to/from > any arbitrary scatter list of buffers. Do you know if this issue is on someone's plate ? If it doesn't, maybe we can try to advance it and start implementing some solutions. As I said earlier and as Sagi mentioned, this feature can improve the performance of modern devices. Thanks, Max.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?567919D3.6050004>