Re: [PATCH v2 2/2] x86_64: enable SWIOTLB if system has SRAT memory regions above MAX_DMA32_PFN

[Igor Mammedov <imammedo@redhat.com>]

On Fri, 4 Dec 2015 12:49:49 +0100
Ingo Molnar <mingo@kernel.org> wrote:

> 
> * Igor Mammedov <imammedo@redhat.com> wrote:
> 
> > when memory hotplug enabled system is booted with less
> > than 4GB of RAM and then later more RAM is hotplugged
> > 32-bit devices stop functioning with following error:
> > 
> >  nommu_map_single: overflow 327b4f8c0+1522 of device mask ffffffff
> > 
> > the reason for this is that if x86_64 system were booted
> > with RAM less than 4GB, it doesn't enable SWIOTLB and
> > when memory is hotplugged beyond MAX_DMA32_PFN, devices
> > that expect 32-bit addresses can't handle 64-bit addresses.
> > 
> > Fix it by tracking max possible PFN when parsing
> > memory affinity structures from SRAT ACPI table and
> > enable SWIOTLB if there is hotpluggable memory
> > regions beyond MAX_DMA32_PFN.
> > 
> > It fixes KVM guests when they use emulated devices
> > (reproduces with ata_piix, e1000 and usb devices,
> >  RHBZ: 1275941, 1275977, 1271527)
> > It also fixes the HyperV, VMWare with emulated devices
> > which are affected by this issue as well.
> > 
> > Signed-off-by: Igor Mammedov <imammedo@redhat.com>
> > ---
> >  arch/x86/kernel/pci-swiotlb.c | 2 +-
> >  arch/x86/mm/srat.c            | 3 +++
> >  2 files changed, 4 insertions(+), 1 deletion(-)
> > 
> > diff --git a/arch/x86/kernel/pci-swiotlb.c b/arch/x86/kernel/pci-swiotlb.c
> > index adf0392..7c577a1 100644
> > --- a/arch/x86/kernel/pci-swiotlb.c
> > +++ b/arch/x86/kernel/pci-swiotlb.c
> > @@ -88,7 +88,7 @@ int __init pci_swiotlb_detect_4gb(void)
> >  {
> >  	/* don't initialize swiotlb if iommu=off (no_iommu=1) */
> >  #ifdef CONFIG_X86_64
> > -	if (!no_iommu && max_pfn > MAX_DMA32_PFN)
> > +	if (!no_iommu && max_possible_pfn > MAX_DMA32_PFN)
> >  		swiotlb = 1;
> 
> Ok, this series looks a lot better!
> 
> >  #endif
> >  	return swiotlb;
> > diff --git a/arch/x86/mm/srat.c b/arch/x86/mm/srat.c
> > index c2aea63..a26bdbe 100644
> > --- a/arch/x86/mm/srat.c
> > +++ b/arch/x86/mm/srat.c
> > @@ -203,6 +203,9 @@ acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
> >  		pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n",
> >  			(unsigned long long)start, (unsigned long long)end - 1);
> >  
> > +	if (max_possible_pfn < PFN_UP(end - 1))
> > +		max_possible_pfn = PFN_UP(end - 1);
> 
> Small nit, why not write this as something like:
> 
> 	max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));
That doesn't pass strict type check:

include/linux/kernel.h:730:17: warning: comparison of distinct pointer types lacks a cast [enabled by default]
  (void) (&_max1 == &_max2);  \
                 ^
arch/x86/mm/srat.c:206:21: note: in expansion of macro ‘max’
  max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));

I can change max_possible_pfn to u64 to match PFN_UP(end - 1) type.

> 
> ?
> 
> I'd also move this second hunk to the first patch, because logically it belongs 
> there (or into a third patch).
> 
> Thanks,
> 
> 	Ingo