Re: [PATCH V3] KVM: SEV: fix wrong pinning of pages

[Tom Lendacky]

On 1/11/25 00:15, yangge1116@xxxxxxx wrote:
> From: yangge <yangge1116@xxxxxxx>
> 
> In the sev_mem_enc_register_region() function, we need to call
> sev_pin_memory() to pin memory for the long term. However, when
> calling sev_pin_memory(), the FOLL_LONGTERM flag is not passed, causing
> the allocated pages not to be migrated out of MIGRATE_CMA/ZONE_MOVABLE,
> violating these mechanisms to avoid fragmentation with unmovable pages,
> for example making CMA allocations fail.
> 
> To address the aforementioned problem, we should add the FOLL_LONGTERM
> flag when calling sev_pin_memory() within the sev_mem_enc_register_region()
> function.

Seems reasonable, some minor comments below, otherwise:

Reviewed-by: Tom Lendacky <thomas.lendacky@xxxxxxx>

> 
> Signed-off-by: yangge <yangge1116@xxxxxxx>
> ---
> 
> V3:
> - the fix only needed for sev_mem_enc_register_region()
> 
> V2:
> - update code and commit message suggested by David
> 
>  arch/x86/kvm/svm/sev.c | 16 +++++++++-------
>  1 file changed, 9 insertions(+), 7 deletions(-)
> 
> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> index 943bd07..04a125c 100644
> --- a/arch/x86/kvm/svm/sev.c
> +++ b/arch/x86/kvm/svm/sev.c
> @@ -622,7 +622,7 @@ static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
>  
>  static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
>  				    unsigned long ulen, unsigned long *n,
> -				    int write)
> +				    int flags)
>  {
>  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
>  	unsigned long npages, size;
> @@ -663,7 +663,7 @@ static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
>  		return ERR_PTR(-ENOMEM);
>  
>  	/* Pin the user virtual address. */
> -	npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
> +	npinned = pin_user_pages_fast(uaddr, npages, flags, pages);
>  	if (npinned != npages) {
>  		pr_err("SEV: Failure locking %lu pages.\n", npages);
>  		ret = -ENOMEM;
> @@ -751,7 +751,7 @@ static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
>  	vaddr_end = vaddr + size;
>  
>  	/* Lock the user memory. */
> -	inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
> +	inpages = sev_pin_memory(kvm, vaddr, size, &npages, FOLL_WRITE);
>  	if (IS_ERR(inpages))
>  		return PTR_ERR(inpages);
>  
> @@ -1250,7 +1250,8 @@ static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
>  		if (IS_ERR(src_p))
>  			return PTR_ERR(src_p);
>  
> -		dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
> +		dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n,
> +					FOLL_WRITE);

You can keep this as one line.

>  		if (IS_ERR(dst_p)) {
>  			sev_unpin_memory(kvm, src_p, n);
>  			return PTR_ERR(dst_p);
> @@ -1316,7 +1317,7 @@ static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
>  	if (copy_from_user(&params, u64_to_user_ptr(argp->data), sizeof(params)))
>  		return -EFAULT;
>  
> -	pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
> +	pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, FOLL_WRITE);
>  	if (IS_ERR(pages))
>  		return PTR_ERR(pages);
>  
> @@ -1798,7 +1799,7 @@ static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
>  
>  	/* Pin guest memory */
>  	guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
> -				    PAGE_SIZE, &n, 1);
> +				    PAGE_SIZE, &n, FOLL_WRITE);
>  	if (IS_ERR(guest_page)) {
>  		ret = PTR_ERR(guest_page);
>  		goto e_free_trans;
> @@ -2696,7 +2697,8 @@ int sev_mem_enc_register_region(struct kvm *kvm,
>  		return -ENOMEM;
>  
>  	mutex_lock(&kvm->lock);
> -	region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
> +	region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages,
> +				FOLL_WRITE | FOLL_LONGTERM);

Need proper alignment of these parameters.

Thanks,
Tom

>  	if (IS_ERR(region->pages)) {
>  		ret = PTR_ERR(region->pages);
>  		mutex_unlock(&kvm->lock);