On Tue, Apr 11, 2017 at 8:22 AM, Deyoung Hong (dhong) <dhong@xxxxxxxxxx> wrote: > I would think both ->setup and ->get_file_size should be called before get_file_size is not being called if the engine provides setup() method. If it doesn't, fio calls get_file_size too early, to my recollection even before the engine has been properly initialized, and that's why setup() workaround exists. > ->iomem_alloc, but I have an fio job file (case problem reported by the > user) where ->iomem_alloc is called before ->setup. if you see this happening (although, from my year-old recollection, that's not the case with setup()), just submit a bug report showing your job file and call sequence. Regards, Andrey > My main question is > that an fio bug (minor)? If someone would like to fix it, I can send a test > job file. > > > > Thanks, > > Deyoung > > > > > > From: Andrey Kuzmin [mailto:andrey.v.kuzmin@xxxxxxxxx] > Sent: Monday, April 10, 2017 2:40 PM > To: Deyoung Hong (dhong) <dhong@xxxxxxxxxx> > Cc: fio@xxxxxxxxxxxxxxx > Subject: RE: Adding IO engine > > > > > > > > On Apr 11, 2017 00:21, "Deyoung Hong (dhong)" <dhong@xxxxxxxxxx> wrote: > > The concept of "file" can represent a special device and driver unknown to > the system, and get_file_size() and/or setup() are usually the first APIs > that fio should call where the device will be initialized and set its max > size so that fio can know the device "file" capacity. Such device has to be > initialized prior to IO (DMA) buffer allocation/free that is specific only > to that device. There isn't any other fio plugin APIs that precede > iomem_alloc, and I think iomem_alloc/free were added specifically to support > that special device. > > > > > > No, the API was added to support I/O engines with special requirements re > I/O buffer memory, with no relation to special files. If you have specific > file setup requirements, your engine should provide setup method. As far as > I recall, it occurs before iomem initialization. > > > > Regards, > > Andrey > > > How is that a misconception? > > > > -----Original Message----- > From: fio-owner@xxxxxxxxxxxxxxx [mailto:fio-owner@xxxxxxxxxxxxxxx] On Behalf > Of Andrey Kuzmin > Sent: Monday, April 10, 2017 12:44 PM > To: Deyoung Hong (dhong) <dhong@xxxxxxxxxx> > Cc: fio@xxxxxxxxxxxxxxx > Subject: Re: Adding IO engine > > On Mon, Apr 10, 2017 at 8:45 PM, Deyoung Hong (dhong) <dhong@xxxxxxxxxx> > wrote: >> Hi Jens, >> >> I have a couple basic questions: >> >> 1) If I want to submit a new IO engine with fio and it has a dependency >> on an external library, how does that work when fio is compiled? Please >> also point me to info how to submit a new IO engine to be packaged with fio? >> >> 2) I've seen cases where the ->iomem_alloc() function (i.e. for a custom >> IO engine) is invoked first by fio before ->get_file_size() or ->setup(). >> This causes an issue since IO memory allocation can have dependency on >> certain initialization (for the custom IO engine) that is done in >> ->get_file_size(). I just wonder if that's an fio bug or how it should be? > > It sounds more like a misconception re iomem_{alloc,free} API. It is > about buffers I/O goes to/from, not files. As an example (and the root > cause behind the API proposal), I wanted to be able to allocate > buffers in DMAable memory as my engine required that to interact with > hardware directly. > > Being unrelated to files by design, buffer initialization occurs way > before file setup as it did in fio before iomem_* API was introduced. > > Regards, > Andrey > > > >> My current work around is to check in ->iomem_alloc() each time and do >> initializion first if has not. >> >> Thanks, >> Deyoung >> >> -- >> To unsubscribe from this list: send the line "unsubscribe fio" in >> the body of a message to majordomo@xxxxxxxxxxxxxxx >> More majordomo info at http://vger.kernel.org/majordomo-info.html > -- > To unsubscribe from this list: send the line "unsubscribe fio" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- To unsubscribe from this list: send the line "unsubscribe fio" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
