Re: [PATCH 02/11] PCI: Try to allocate mem64 above 4G at first

[Bjorn Helgaas <bhelgaas@google.com>]

On Wed, May 23, 2012 at 11:40:46AM -0700, Yinghai Lu wrote:
> On Wed, May 23, 2012 at 10:30 AM, Yinghai Lu <yinghai@kernel.org> wrote:
> > On Wed, May 23, 2012 at 8:57 AM, Linus Torvalds
> > <torvalds@linux-foundation.org> wrote:
> >> On Tue, May 22, 2012 at 11:34 PM, Yinghai Lu <yinghai@kernel.org> wrote:
> >>> and will fall back to below 4g if it can not find any above 4g.
> >>
> >> Has this been tested on 32-bit machines without PAE? There might be
> >> things that just happen to work because their allocations were always
> >> done bottom-up.
> >
> > Good point. that problem should be addressed at first before this patch.
> 
> Just checked code for 32bit machines without PAE.
> 
> when X86_PAE is not set, phys_addr_t aka resource_size_t will be 32bit.
> so in drivers/pci/bus.c::pci_bus_alloc_resource_fit()
> will have bottom to 0.
>     resource_size_t bottom = PCIBIOS_MAX_MEM_32 + 1ULL;
> also in arch/x86/kernel/setup.c::setup_arch()
>    iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
> will have iomem_resource.end to 0xffffffff
> 
> when X86_PAE is set, but CPU does not support PAE.
> phys_addr_t aka resource_size_t will be 32bit.

I think you meant phys_addr_t and resource_size_t will be *64* bit
when X86_PAE is set.  Obvious to you, but quite confusing to non-x86
experts like me :)

> so in drivers/pci/bus.c::pci_bus_alloc_resource_fit()
> will have bottom to 4g.
>     resource_size_t bottom = PCIBIOS_MAX_MEM_32 + 1ULL;
> but
> in arch/x86/kernel/setup.c::setup_arch()
>    iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
> will have iomem_resource.end to 0xffffffff, because x86_phys_bits is 32 when PAE
> is not detected in arch/x86/kernel/cpu/common.c::get_cpu_cap.
> that mean first try will fail, so it will go to second try with bottom to 0.
> 
> so both case are safe with this patch.

I don't really like the dependency on PCIBIOS_MAX_MEM_32 + 1ULL
overflowing to zero -- that means the reader has to know what the
value of PCIBIOS_MAX_MEM_32 is, and things would break in non-obvious
ways if we changed it.

What do you think of a patch like the following?  It makes it
explicit that we can only allocate space the CPU can address.

commit feded2ae21d6160292726ccd5128080d42395be4
Author: Bjorn Helgaas <bhelgaas@google.com>
Date:   Thu May 24 22:15:26 2012 -0600

    PCI: try to allocate 64-bit resources above 4GB
    
    If we have a 64-bit resource, try to allocate it above 4GB first.  If that
    fails, either because there's no space or the CPU can't address space above
    4GB (iomem_resource.end is the highest address the CPU supports), we'll
    fall back to allocating space below 4GB.

diff --git a/drivers/pci/bus.c b/drivers/pci/bus.c
index 4ce5ef2..2c56693 100644
--- a/drivers/pci/bus.c
+++ b/drivers/pci/bus.c
@@ -121,14 +121,18 @@ pci_bus_alloc_resource(struct pci_bus *bus, struct resource *res,
 {
 	int i, ret = -ENOMEM;
 	struct resource *r;
-	resource_size_t max = -1;
+	resource_size_t start = 0;
+	resource_size_t end = PCIBIOS_MAX_MEM_32;
 
 	type_mask |= IORESOURCE_IO | IORESOURCE_MEM;
 
-	/* don't allocate too high if the pref mem doesn't support 64bit*/
-	if (!(res->flags & IORESOURCE_MEM_64))
-		max = PCIBIOS_MAX_MEM_32;
+	/* If this is a 64-bit resource, prefer space above 4GB */
+	if (res->flags & IORESOURCE_MEM_64) {
+		start = PCIBIOS_MAX_MEM_32 + 1ULL;
+		end = iomem_resource.end;
+	}
 
+again:
 	pci_bus_for_each_resource(bus, r, i) {
 		if (!r)
 			continue;
@@ -145,12 +149,18 @@ pci_bus_alloc_resource(struct pci_bus *bus, struct resource *res,
 
 		/* Ok, try it out.. */
 		ret = allocate_resource(r, res, size,
-					r->start ? : min,
-					max, align,
+					max(start, r->start ? : min),
+					end, align,
 					alignf, alignf_data);
 		if (ret == 0)
-			break;
+			return 0;
 	}
+
+	if (start != 0) {
+		start = 0;
+		goto again;
+	}
+
 	return ret;
 }