Michael Ellerman <m...@ellerman.id.au> wrote on 11/06/2017 11:45:34 AM:
> From: Michael Ellerman <m...@ellerman.id.au> > To: Alexey Kardashevskiy <a...@ozlabs.ru>, David Gibson > <da...@gibson.dropbear.id.au> > Cc: linuxppc-dev@lists.ozlabs.org, Jonas Pfefferle1 > <j...@zurich.ibm.com>, Nicholas Piggin <npig...@gmail.com> > Date: 11/06/2017 11:45 AM > Subject: Re: [PATCH kernel] powerpc/powernv/ioda: Relax max DMA > window size check > > Alexey Kardashevskiy <a...@ozlabs.ru> writes: > > > On 31/10/17 15:04, Alexey Kardashevskiy wrote: > >> DMA windows can only have a size of power of two on IODA2 hardware and > >> using memory_hotplug_max() to determine the upper limit won't work > >> correcly if it returns not power of two value. > >> > >> This relaxes the check by rounding up the value returned by > >> memory_hotplug_max(). > >> > >> It is expected to impact DPDK on machines with non-power-of-two RAM size, > >> mostly. KVM guests are less likely to be affected as usually guests get > >> less than half of hosts RAM. > > > > > > It was pointed out that this check is quite useless anyway as the vm_locked > > memory limit should hit first, and if that is not set or the user got the > > root privilege level, then there are easier ways to crash the host so I am > > thinking of: > > > > > > diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c > > b/arch/powerpc/platforms/powernv/pci-ioda.c > > index 269f119e4b3c..a47e4cf343b2 100644 > > --- a/arch/powerpc/platforms/powernv/pci-ioda.c > > +++ b/arch/powerpc/platforms/powernv/pci-ioda.c > > @@ -2769,7 +2769,7 @@ static long pnv_pci_ioda2_table_alloc_pages(int nid, > > __u64 bus_offset, > > if (!levels || (levels > POWERNV_IOMMU_MAX_LEVELS)) > > return -EINVAL; > > > > - if ((window_size > memory_hotplug_max()) || > > !is_power_of_2(window_size)) > > + if (!is_power_of_2(window_size)) > > return -EINVAL; > > > > > > > > Makes sense? > > Sounds reasonable. > > Execpt where is the vm_locked check? I think it's in the VFIO driver? If > so I guess the only concern is that this code might be called via some > other path that doesn't do that check. > > cheers > The vm_locked is incremented here: http://elixir.free-electrons.com/linux/v4.13.11/source/drivers/vfio/vfio_iommu_spapr_tce.c#L176 resp. http://elixir.free-electrons.com/linux/v4.13.11/source/arch/powerpc/mm/mmu_context_iommu.c#L124 on VFIO_IOMMU_SPAPR_REGISTER_MEMORY. From my understanding only pages that have been registered through here can be mapped with MAP_DMA. Cheers, Jonas