Julien,
On 20/12/2018 15:21, Julien Grall wrote:
From: Suzuki K Poulose <suzuki.poulose@xxxxxxx>
With 16K pagesize, the hugepage size is 32M. Align the guest
memory to the hugepagesize for 16K.
Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
Cc: Andre Przywara <andre.przywara@xxxxxxx>
Cc: Will Deacon <will.deacon@xxxxxxx>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@xxxxxxx>
Signed-off-by: Julien Grall <julien.grall@xxxxxxx>
---
arm/kvm.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)
diff --git a/arm/kvm.c b/arm/kvm.c
index b824f63..5b1f73c 100644
--- a/arm/kvm.c
+++ b/arm/kvm.c
@@ -59,14 +59,22 @@ void kvm__arch_set_cmdline(char *cmdline, bool video)
void kvm__arch_init(struct kvm *kvm, const char *hugetlbfs_path, u64 ram_size)
{
+ unsigned long alignment;
+
/*
* Allocate guest memory. We must align our buffer to 64K to
* correlate with the maximum guest page size for virtio-mmio.
* If using THP, then our minimal alignment becomes 2M.
* 2M trumps 64K, so let's go with that.
+ * If we are running with 16K page size, align the memory to
+ * 32M, so that we can make use of the THP.
*/
One issue that I realized with the above adjustment is when we specify a
hugetlbfs path, e.g, 1GB hugepages with 4K (with recently added PUD huge
page support at Stage2), we fail to allocate memory.
I think the whole alignment game should be performed only for the anon_mmap
case. If the hugetlbfs path is specified, we are guaranteed to get aligned
memory banks. Do you think you could fix it with this series (of course as
a separate patch), as it seems more relevant ? You could may be do :
if (!hugetlbfs_path) {
+ if (sysconf(_SC_PAGESIZE) == SZ_16K)
+ alignment = SZ_32M;
+ else
+ alignment = SZ_2M;
} else {
alignment = 0;
}
kvm->ram_size = min(ram_size, (u64)ARM_MAX_MEMORY(kvm));
- kvm->arch.ram_alloc_size = kvm->ram_size + SZ_2M;
+ kvm->arch.ram_alloc_size = kvm->ram_size + alignment;
kvm->arch.ram_alloc_start = mmap_anon_or_hugetlbfs(kvm, hugetlbfs_path,
kvm->arch.ram_alloc_size);
@@ -75,7 +83,7 @@ void kvm__arch_init(struct kvm *kvm, const char *hugetlbfs_path, u64 ram_size)
kvm->arch.ram_alloc_size, errno);
kvm->ram_start = (void *)ALIGN((unsigned long)kvm->arch.ram_alloc_start,
- SZ_2M);
+ alignment);
Cheers
Suzuki
