On Sat, 2013-07-27 at 22:30 +0100, Ben Hutchings wrote: > On Sat, 2013-07-27 at 20:30 +0100, Luis Henriques wrote: > > Ben Hutchings <bhutchings@xxxxxxxxxxxxxx> writes: > > > > > On Sat, 2013-07-27 at 01:02 +0100, Ben Hutchings wrote: > > >> On Fri, 2013-07-26 at 12:47 -0400, Neil Horman wrote: > > >> > atl1c uses netdev_alloc_skb to refill its rx dma ring, but that call makes no > > >> > guarantees about the suitability of the memory for use in DMA. As a result > > >> > we've gotten reports of atl1c drivers occasionally hanging and needing to be > > >> > reset: > > >> > https://bugzilla.kernel.org/show_bug.cgi?id=54021 > > >> > > > >> > Fix this by modifying the call to use the internal version __netdev_alloc_skb, > > >> > where you can set the gfp_mask explicitly to include GFP_DMA. > > >> > > >> This is a really bad idea. GFP_DMA means allocation from the ISA DMA > > >> region (< 16 MB). pci_map_single() takes care of allocating a bounce > > >> buffer if necessary. > [...] > > Just to add a little bit more context (and hopefully not noise), I > > started seeing this issue on 3.7. Bisection resulted on the following > > first bad commit: > > > > 69b08f6 net: use bigger pages in __netdev_alloc_frag > > > > Reverting this commit (and e5e6730 "skbuff: Move definition of > > NETDEV_FRAG_PAGE_MAX_SIZE") solved the problem. > > > > Note also that I'm seeing this issue on a 32 bits system (64 bits > > isn't supported). This initially made me think the problem could be > > related with this as 69b08f6 log explicitly refers to 32/64 bit > > archs. But I failed to find any obvious issue with the patch. > > Then it seems like this patch works because passing the GFP_DMA flag to > __netdev_alloc_skb() disables the use of __netdev_alloc_frag() and > results in it calling __alloc_skb(). A better workaround would be for > atl1c to call __alloc_skb() directly. > > Perhaps the controller doesn't split RX DMA across PCIe page boundaries > (4K), or some other boundaries at smaller intervals than > NETDEV_FRAG_PAGE_MAX_SIZE. > > But I think that perhaps the use of __netdev_alloc_frag() should be made > opt-in. I doubt this is the only driver whose DMA requirements have > been broken. Since the Linux DMA API lacks any way for devices to > specify boundaries which would then be observed by pci_map_single(), I > don't think this can be considered simply a driver bug. It is a driver bug to assume anything about alloc_skb(), as there is no specification about skb->head being aligned to whatever boundary. The only guarantee is the one provided by kmalloc(), that is 8 bytes. I specifically asked this exact question in https://bugzilla.kernel.org/show_bug.cgi?id=54021#c19 And the Qualcom guys checked and said it was ok. Who should we trust ? I hope you understand kmalloc(~2000 bytes) never made the assumption the area fit a single page. SLOB or even SLUB/SLAB with some debugging features... If a hardware needs frame being in a single 4K page, its driver must do its own allocation, or add appropriate aligning logic. -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html