1. Patchew
  2. linux
  3. View series

[PATCH] iommu/amd: Add NUMA node affinity for IOMMU log buffers

lirongqing posted 1 patch 1 month, 2 weeks ago 
drivers/iommu/amd/init.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
[PATCH] iommu/amd: Add NUMA node affinity for IOMMU log buffers
Posted by lirongqing 1 month, 2 weeks ago 
From: Li RongQing <lirongqing@baidu.com>

Currently, PPR Log and GA logs for AMD IOMMU are allocated using
iommu_alloc_pages_sz(), which does not account for NUMA affinity. This can
lead to remote memory access latencies if the memory is allocated on a
different node than the IOMMU hardware.

Switch to iommu_alloc_pages_node_sz() to ensure that these data structures
are allocated on the same NUMA node as the IOMMU device. If the node
information is unavailable, it defaults to NUMA_NO_NODE.

Signed-off-by: Li RongQing <lirongqing@baidu.com>
---
 drivers/iommu/amd/init.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/iommu/amd/init.c b/drivers/iommu/amd/init.c
index 384c90b..721d9d0 100644
--- a/drivers/iommu/amd/init.c
+++ b/drivers/iommu/amd/init.c
@@ -851,10 +851,11 @@ static void __init free_command_buffer(struct amd_iommu *iommu)
 void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, gfp_t gfp,
 				  size_t size)
 {
+	int nid = iommu->dev ? dev_to_node(&iommu->dev->dev) : NUMA_NO_NODE;
 	void *buf;
 
 	size = PAGE_ALIGN(size);
-	buf = iommu_alloc_pages_sz(gfp, size);
+	buf = iommu_alloc_pages_node_sz(nid, gfp, size);
 	if (!buf)
 		return NULL;
 	if (check_feature(FEATURE_SNP) &&
@@ -957,14 +958,16 @@ static int iommu_ga_log_enable(struct amd_iommu *iommu)
 
 static int iommu_init_ga_log(struct amd_iommu *iommu)
 {
+	int nid = iommu->dev ? dev_to_node(&iommu->dev->dev) : NUMA_NO_NODE;
+
 	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
 		return 0;
 
-	iommu->ga_log = iommu_alloc_pages_sz(GFP_KERNEL, GA_LOG_SIZE);
+	iommu->ga_log = iommu_alloc_pages_node_sz(nid, GFP_KERNEL, GA_LOG_SIZE);
 	if (!iommu->ga_log)
 		goto err_out;
 
-	iommu->ga_log_tail = iommu_alloc_pages_sz(GFP_KERNEL, 8);
+	iommu->ga_log_tail = iommu_alloc_pages_node_sz(nid, GFP_KERNEL, 8);
 	if (!iommu->ga_log_tail)
 		goto err_out;
 
-- 
2.9.4
Re: [PATCH] iommu/amd: Add NUMA node affinity for IOMMU log buffers
Posted by Joerg Roedel 2 weeks, 2 days ago 
On Tue, Feb 10, 2026 at 10:58:39PM -0500, lirongqing wrote:
> From: Li RongQing <lirongqing@baidu.com>
> 
> Currently, PPR Log and GA logs for AMD IOMMU are allocated using
> iommu_alloc_pages_sz(), which does not account for NUMA affinity. This can
> lead to remote memory access latencies if the memory is allocated on a
> different node than the IOMMU hardware.
> 
> Switch to iommu_alloc_pages_node_sz() to ensure that these data structures
> are allocated on the same NUMA node as the IOMMU device. If the node
> information is unavailable, it defaults to NUMA_NO_NODE.
> 
> Signed-off-by: Li RongQing <lirongqing@baidu.com>
> ---
>  drivers/iommu/amd/init.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)

Applied, thanks.
Re: [PATCH] iommu/amd: Add NUMA node affinity for IOMMU log buffers
Posted by Vasant Hegde 1 month, 1 week ago 
On 2/11/2026 9:28 AM, lirongqing wrote:
> From: Li RongQing <lirongqing@baidu.com>
> 
> Currently, PPR Log and GA logs for AMD IOMMU are allocated using
> iommu_alloc_pages_sz(), which does not account for NUMA affinity. This can
> lead to remote memory access latencies if the memory is allocated on a
> different node than the IOMMU hardware.
> 
> Switch to iommu_alloc_pages_node_sz() to ensure that these data structures
> are allocated on the same NUMA node as the IOMMU device. If the node
> information is unavailable, it defaults to NUMA_NO_NODE.
> 
> Signed-off-by: Li RongQing <lirongqing@baidu.com>

Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>

-Vasant
Re: [PATCH] iommu/amd: Add NUMA node affinity for IOMMU log buffers
Posted by Ankit Soni 1 month, 2 weeks ago 
On Tue, Feb 10, 2026 at 10:58:39PM -0500, lirongqing wrote:
> From: Li RongQing <lirongqing@baidu.com>
> 
> Currently, PPR Log and GA logs for AMD IOMMU are allocated using
> iommu_alloc_pages_sz(), which does not account for NUMA affinity. This can
> lead to remote memory access latencies if the memory is allocated on a
> different node than the IOMMU hardware.
> 
> Switch to iommu_alloc_pages_node_sz() to ensure that these data structures
> are allocated on the same NUMA node as the IOMMU device. If the node
> information is unavailable, it defaults to NUMA_NO_NODE.
> 

 Reviewed-by: Ankit Soni <Ankit.Soni@amd.com>

> Signed-off-by: Li RongQing <lirongqing@baidu.com>
> ---
>  drivers/iommu/amd/init.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/iommu/amd/init.c b/drivers/iommu/amd/init.c
> index 384c90b..721d9d0 100644
> --- a/drivers/iommu/amd/init.c
> +++ b/drivers/iommu/amd/init.c
> @@ -851,10 +851,11 @@ static void __init free_command_buffer(struct amd_iommu *iommu)
>  void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, gfp_t gfp,
>  				  size_t size)
>  {
> +	int nid = iommu->dev ? dev_to_node(&iommu->dev->dev) : NUMA_NO_NODE;
>  	void *buf;
>  
>  	size = PAGE_ALIGN(size);
> -	buf = iommu_alloc_pages_sz(gfp, size);
> +	buf = iommu_alloc_pages_node_sz(nid, gfp, size);
>  	if (!buf)
>  		return NULL;
>  	if (check_feature(FEATURE_SNP) &&
> @@ -957,14 +958,16 @@ static int iommu_ga_log_enable(struct amd_iommu *iommu)
>  
>  static int iommu_init_ga_log(struct amd_iommu *iommu)
>  {
> +	int nid = iommu->dev ? dev_to_node(&iommu->dev->dev) : NUMA_NO_NODE;
> +
>  	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
>  		return 0;
>  
> -	iommu->ga_log = iommu_alloc_pages_sz(GFP_KERNEL, GA_LOG_SIZE);
> +	iommu->ga_log = iommu_alloc_pages_node_sz(nid, GFP_KERNEL, GA_LOG_SIZE);
>  	if (!iommu->ga_log)
>  		goto err_out;
>  
> -	iommu->ga_log_tail = iommu_alloc_pages_sz(GFP_KERNEL, 8);
> +	iommu->ga_log_tail = iommu_alloc_pages_node_sz(nid, GFP_KERNEL, 8);
>  	if (!iommu->ga_log_tail)
>  		goto err_out;
>  
> -- 
> 2.9.4
>

Patchew 2.1-devel-ea86937

© 2016 - 2026 Red Hat, Inc.