On Mon Sep 14, 2026 at 7:22 PM JST, Gary Guo wrote: > On Mon Sep 14, 2026 at 10:58 AM BST, Danilo Krummrich wrote: >> On Mon Sep 14, 2026 at 2:45 AM CEST, Alexandre Courbot wrote: >>> On Mon Sep 14, 2026 at 6:08 AM JST, Matteo Kloiber wrote: >>>> On Mon Sep 7, 2026 at 11:43 AM JST, Alexandre Courbot wrote: >>>>> It also means that without patch 1, nova-core would split the firmware >>>>> into hundreds of 64KB SG entries, which is not breaking but still >>>>> something we want to avoid. The correct fix is to make sure that >>>>> `dma_set_max_seg_size` is called by the driver, and while we are at it >>>>> we also want every driver to call `dma_set_mask_and_coherent`. Ideally >>>>> we would use the type system to make sure that both functions are called >>>>> before any DMA operation can take place (using a safe interface), but >>>>> I'm not quite sure yet how we can do this. >>>> >>>> This sounds sensible indeed. Should I open a thread regarding that on >>>> Zulip? >>> >>> Probably not necessary, the mailing-list has a larger audience and is >>> the right place for this. I expect people will jump in here with their >>> thoughts. >> >> The problem with those is not that they must strictly be called before >> allocating DMA memory, but they must not be called concurrently with other >> DMA >> operations, such as allocating DMA memory, as it would technically be a data >> race. > > Do they really have to be called *before* allocating DMA memory, not do they > just need not be called *concurrent* to DMA memory allocation? > > If it's the former, we can require these to require mutable reference instead, > so the probe takes `Pin<&'bound mut Device<Core<'_>>>` which still derefs to > `&'bound Device<Bound>`, but Rust will require the shared reference to not > co-exist with the mutable reference.
As Danilo pointed out [1] from a strict safety perspective they must not be called concurrently to DMA allocations. So there is not a hard need to call them before. But my point is different: until patch 1 of this series, nova-core didn't do anything particularly wrong but was operating with an inaccurate DMA segment size. Which is currently inconsequential as long as you don't set `CONFIG_DMA_API_DEBUG`, but will become sub-optimal if patch 2 gets merged (which it should be, otherwise more DMA-limited devices could start misbehaving). The thing is that every driver *really* should set their DMA mask and max segment size ASAP, but at the moment this is done through two unsafe functions that are very easy to overlook. And practically speaking, is there a reason *not* to do that setup before allocating objects? So I am thinking that maybe we could have a kind of `DeviceDma` type which construction involves setting the core DMA parameters (or acknowledging the defaults), and to which all allocations would be tied. IOW, Danilo's DMA allocations rework [2] would tie to that `DeviceDma` object (which itself would be tied to the bound device) instead of the device itself. [1] https://lore.kernel.org/[email protected] [2] https://lore.kernel.org/[email protected]
