[PATCH v1 1/1] KVM: s390: Fix a deadlock

[PATCH v1 1/1] KVM: s390: Fix a deadlock

[Claudio Imbrenda]

In some scenarios, a deadlock can happen, involving _do_shadow_pte().

Convert all usages of pgste_get_lock() to pgste_get_trylock() in
_do_shadow_pte() and return -EAGAIN. All callers can already deal with
-EAGAIN being returned.

Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Fixes: e38c884df921 ("KVM: s390: Switch to new gmap")
---
 arch/s390/kvm/gaccess.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/arch/s390/kvm/gaccess.c b/arch/s390/kvm/gaccess.c
index 4630b2a067ea..a9da9390867d 100644
--- a/arch/s390/kvm/gaccess.c
+++ b/arch/s390/kvm/gaccess.c
@@ -1434,7 +1434,8 @@ static int _do_shadow_pte(struct gmap *sg, gpa_t raddr, union pte *ptep_h, union
 	if (rc)
 		return rc;
 
-	pgste = pgste_get_lock(ptep_h);
+	if (!pgste_get_trylock(ptep_h, &pgste))
+		return -EAGAIN;
 	newpte = _pte(f->pfn, f->writable, !p, 0);
 	newpte.s.d |= ptep->s.d;
 	newpte.s.sd |= ptep->s.sd;
@@ -1444,7 +1445,8 @@ static int _do_shadow_pte(struct gmap *sg, gpa_t raddr, union pte *ptep_h, union
 	pgste_set_unlock(ptep_h, pgste);
 
 	newpte = _pte(f->pfn, 0, !p, 0);
-	pgste = pgste_get_lock(ptep);
+	if (!pgste_get_trylock(ptep, &pgste))
+		return -EAGAIN;
 	pgste = __dat_ptep_xchg(ptep, pgste, newpte, gpa_to_gfn(raddr), sg->asce, uses_skeys(sg));
 	pgste_set_unlock(ptep, pgste);
 
-- 
2.53.0

Re: [PATCH v1 1/1] KVM: s390: Fix a deadlock

[Christoph Schlameuss]

On Tue Mar 3, 2026 at 6:52 PM CET, Claudio Imbrenda wrote:
> In some scenarios, a deadlock can happen, involving _do_shadow_pte().
>
> Convert all usages of pgste_get_lock() to pgste_get_trylock() in
> _do_shadow_pte() and return -EAGAIN. All callers can already deal with
> -EAGAIN being returned.
>
> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
> Fixes: e38c884df921 ("KVM: s390: Switch to new gmap")
Reviewed-by: Christoph Schlameuss <schlameuss@linux.ibm.com>

Re: [PATCH v1 1/1] KVM: s390: Fix a deadlock

[Janosch Frank]

On 3/3/26 18:52, Claudio Imbrenda wrote:
> In some scenarios, a deadlock can happen, involving _do_shadow_pte().
> 
> Convert all usages of pgste_get_lock() to pgste_get_trylock() in
> _do_shadow_pte() and return -EAGAIN. All callers can already deal with
> -EAGAIN being returned.
> 
> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
> Fixes: e38c884df921 ("KVM: s390: Switch to new gmap")

Reviewed-by: Janosch Frank <frankja@linux.ibm.com>

Re: [PATCH v1 1/1] KVM: s390: Fix a deadlock

[Christian Borntraeger]

Am 03.03.26 um 18:52 schrieb Claudio Imbrenda:
> In some scenarios, a deadlock can happen, involving _do_shadow_pte().
> 
> Convert all usages of pgste_get_lock() to pgste_get_trylock() in
> _do_shadow_pte() and return -EAGAIN. All callers can already deal with
> -EAGAIN being returned.
> 
> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
> Fixes: e38c884df921 ("KVM: s390: Switch to new gmap")

Tested-by: Christian Borntraeger <borntraeger@linux.ibm.com>

I no longer see the rcu stalls.

> ---
>   arch/s390/kvm/gaccess.c | 6 ++++--
>   1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/s390/kvm/gaccess.c b/arch/s390/kvm/gaccess.c
> index 4630b2a067ea..a9da9390867d 100644
> --- a/arch/s390/kvm/gaccess.c
> +++ b/arch/s390/kvm/gaccess.c
> @@ -1434,7 +1434,8 @@ static int _do_shadow_pte(struct gmap *sg, gpa_t raddr, union pte *ptep_h, union
>   	if (rc)
>   		return rc;
>   
> -	pgste = pgste_get_lock(ptep_h);
> +	if (!pgste_get_trylock(ptep_h, &pgste))
> +		return -EAGAIN;
>   	newpte = _pte(f->pfn, f->writable, !p, 0);
>   	newpte.s.d |= ptep->s.d;
>   	newpte.s.sd |= ptep->s.sd;
> @@ -1444,7 +1445,8 @@ static int _do_shadow_pte(struct gmap *sg, gpa_t raddr, union pte *ptep_h, union
>   	pgste_set_unlock(ptep_h, pgste);
>   
>   	newpte = _pte(f->pfn, 0, !p, 0);
> -	pgste = pgste_get_lock(ptep);
> +	if (!pgste_get_trylock(ptep, &pgste))
> +		return -EAGAIN;
>   	pgste = __dat_ptep_xchg(ptep, pgste, newpte, gpa_to_gfn(raddr), sg->asce, uses_skeys(sg));
>   	pgste_set_unlock(ptep, pgste);
>