[PATCH] x86/hyperv: Consider NUMA affinity when allocating memory for per-CPU vmsa

[PATCH] x86/hyperv: Consider NUMA affinity when allocating memory for per-CPU vmsa

[Li RongQing]

per-CPU vmsa are dominantly accessed from their own local CPUs,
so allocate them node-local to improve performance.

And reorganized variables to be reverse christmas tree order

Signed-off-by: Li RongQing <lirongqing@baidu.com>
---
 arch/x86/hyperv/ivm.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/arch/x86/hyperv/ivm.c b/arch/x86/hyperv/ivm.c
index 768d73d..5bc2430 100644
--- a/arch/x86/hyperv/ivm.c
+++ b/arch/x86/hyperv/ivm.c
@@ -291,16 +291,18 @@ static void snp_cleanup_vmsa(struct sev_es_save_area *vmsa)
 
 int hv_snp_boot_ap(u32 cpu, unsigned long start_ip)
 {
-	struct sev_es_save_area *vmsa = (struct sev_es_save_area *)
-		__get_free_page(GFP_KERNEL | __GFP_ZERO);
+	struct hv_enable_vp_vtl *start_vp_input;
 	struct sev_es_save_area *cur_vmsa;
+	struct sev_es_save_area *vmsa;
 	struct desc_ptr gdtr;
-	u64 ret, retry = 5;
-	struct hv_enable_vp_vtl *start_vp_input;
 	unsigned long flags;
+	u64 ret, retry = 5;
+	struct page *p;
 
-	if (!vmsa)
+	p = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL | __GFP_ZERO, 0);
+	if (!p)
 		return -ENOMEM;
+	vmsa = (struct sev_es_save_area *)page_address(p);
 
 	native_store_gdt(&gdtr);
 
-- 
2.9.4

Re: [PATCH] x86/hyperv: Consider NUMA affinity when allocating memory for per-CPU vmsa

[Wei Liu]

On Mon, Apr 15, 2024 at 04:57:12PM +0800, Li RongQing wrote:
> per-CPU vmsa are dominantly accessed from their own local CPUs,
> so allocate them node-local to improve performance.
> 
> And reorganized variables to be reverse christmas tree order
> 
> Signed-off-by: Li RongQing <lirongqing@baidu.com>

Applied. Thanks.