jeremytrimble
Status?
At risk of abusing the issues system, I'm curious to know what the current status is of this driver.
I too would very much like to have such a device driver for pushing data from user-space over DMA. Have you made any more attempts to get it added to the mainline kernel tree?
Cheers, Henry
Henry,
I started a preliminary discussion on the dmaengine kernel mailing list [1], but the dmaengine subsystem maintainer seemed unreceptive to the concept.
In other discussions, it was suggested that perhaps I could reimagine ezdma as an addition to the "uio" (userspace I/O) interface in the kernel, though I haven't had time to really pursue this.
That said, a few other folks have contacted me about using ezdma in their projects and may be doing so as we speak. I'd be interested in hearing how you might want to use it, and/or glad to take pull requests or answer questions if I'm able.
[1] http://www.spinics.net/lists/dmaengine/msg04599.html
Hi Jeremy,
Yeah, it was that thread that prompted the question. Was the uio stuff discussed offline, or do you have a reference for that? I too am thinking along exactly these lines - how best to access DMAd hardware from userspace. Clearly with something like a Zynq there is any number of weird and wonderful hardware widgets that can be created for which no subsystem yet exists. I don't know what the correct method for handling this is, but uio seems to be a plausible route.
Henry,
Sorry for the delay -- this fell off my radar for a bit.
Pasting from my IRC logs in #kernelnewbies on OFTC:
2015-06-18 14:05:54 jtrimble imMute: you might want to check out: https://github.com/jeremytrimble/ezdma
2015-06-18 14:06:07 jtrimble (disclaimer, I wrote it) :-)
2015-06-18 14:07:06 jtrimble the way I handle that is calling dmaengine_terminate_all() in my release()
2015-06-18 14:13:27 derRichard jtrimble: why didn't you add dma support do uio?
2015-06-18 14:14:16 imMute ah, I should also note that I'm not using any of the built-in DMA channel stuff (existing code, didn't know about them, $more_excuses)
2015-06-18 14:18:47 jtrimble derRichard: i guess i didn't know i could
2015-06-18 14:19:46 jtrimble i've started a conversation about adding ezdma to the mainline, but the dmaengine maintainer's taken some issue with it -- do you think that the functionality implemented in ezdma maybe belongs in uio instead?
2015-06-18 14:20:41 derRichard yes. your driver is an ad hoc uio.
2015-06-18 14:21:13 derRichard we have uio for exactly that reason. to stop these (often horrible) ad hoc character drivers which implement mmap()
2015-06-18 14:22:34 derRichard jtrimble: also see vfio
2015-06-18 14:24:41 jtrimble derRichard: ah ok. ezdma doesn't implement mmap() though, it uses a blocking read()/write() (and dma_map()s the userspace pages directly during those calls)
2015-06-18 14:25:18 derRichard jtrimble: yeah, i see. but it's just a minor implementation detail
2015-06-18 14:27:17 jtrimble derRichard: are you saying i basically re-implemented something that could have been done with uio, or do you think that it's something I should try to add to uio?
2015-06-18 14:29:30 derRichard the latter
2015-06-18 14:29:37 derRichard currently uio has no dma support
2015-06-18 14:29:47 derRichard vfio would work, but it depends on iommu
2015-06-18 14:31:52 jtrimble derRichard: ok, cool. i'll look into it
2015-06-18 14:41:10 derRichard jtrimble: adding dma to uio isn't that hard
2015-06-18 14:41:32 derRichard but there are many dma "modes". device->device, device->mem, mem->mem, etc...
2015-06-18 14:41:45 derRichard you'd have to find a way which models all
2015-06-18 14:44:56 jtrimble derRichard: hmm yes. ezdma currently supports device->mem and mem->device. i'll think about how to incorporate the others
2015-06-18 14:45:36 jtrimble i've started a conversation on the dmaengine mailing list about adding ezdma as-is there, but the maintainer seems less than thrilled (though I haven't submitted the patch yet as I want to re-test)
2015-06-18 14:45:48 jtrimble but i have gotten word from a few folks that they're using it in their projects
2015-06-18 14:45:58 jtrimble maybe uio is where this functionality ultimately belongs
2015-06-18 14:46:04 derRichard jtrimble: TBH, i'm also not thrilled. it is an ad hoc solution and nothing generic
2015-06-18 14:48:39 jtrimble derRichard: that's fair. i've been writing (out-of-tree, probably disgusting) drivers for a while but still have a lot to learn, especially about the mainline kernel design philosophies
2015-06-18 14:51:09 jtrimble i'd like to find a way to contribute this to the mainline though, as this is the kind of thing that comes up quite often in my field and it would be great to "give back" to the open-source world that's given me so much
2015-06-18 14:51:36 jtrimble so i'll definitely start thinking about how it could fit into uio
... so much optimism. And then my life took over.