| Message ID | 20240406022737.3898763-1-tangnianyao@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [RESPIN] irqchip/gic-v4.1:Fix VSYNC referencing an unmapped VPE | expand |
On Sat, 06 Apr 2024 03:27:37 +0100, Nianyao Tang <tangnianyao@huawei.com> wrote: > > As per the GICv4.1 spec (Arm IHI 0069H, 5.3.19): > > "A VMAPP with {V, Alloc}=={0, x} is self-synchronizing, This means the > ITS command queue does not show the command as consumed until all of > its effects are completed." > > Furthermore, VSYNC is allowed to deliver an SError when referencing a > non existent VPE. > > By these definitions, a VMAPP followed by a VSYNC is a bug, as the > later references a VPE that has been unmapped by the former. > > Fix it by eliding the VSYNC in this scenario. > > Fixes: 64edfaa9a234 ("irqchip/gic-v4.1: Implement the v4.1 flavour of VMAPP") > > Signed-off-by: Nianyao Tang <tangnianyao@huawei.com> > Reviewed-by: Marc Zyngier <maz@kernel.org> Thanks for having taken my remarks into account, it looks good now. Thomas, can you please take this as a fix for 6.9? Thanks, M.
On 2024/4/6 10:27, Nianyao Tang wrote: > As per the GICv4.1 spec (Arm IHI 0069H, 5.3.19): > > "A VMAPP with {V, Alloc}=={0, x} is self-synchronizing, This means the > ITS command queue does not show the command as consumed until all of > its effects are completed." > > Furthermore, VSYNC is allowed to deliver an SError when referencing a > non existent VPE. > > By these definitions, a VMAPP followed by a VSYNC is a bug, as the > later references a VPE that has been unmapped by the former. > > Fix it by eliding the VSYNC in this scenario. > > Fixes: 64edfaa9a234 ("irqchip/gic-v4.1: Implement the v4.1 flavour of VMAPP") > > Signed-off-by: Nianyao Tang <tangnianyao@huawei.com> > Reviewed-by: Marc Zyngier <maz@kernel.org> Reviewed-by: Zenghui Yu <yuzenghui@huawei.com> Thanks.
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c index fca888b36680..2a537cbfcb07 100644 --- a/drivers/irqchip/irq-gic-v3-its.c +++ b/drivers/irqchip/irq-gic-v3-its.c @@ -786,6 +786,7 @@ static struct its_vpe *its_build_vmapp_cmd(struct its_node *its, struct its_cmd_block *cmd, struct its_cmd_desc *desc) { + struct its_vpe *vpe = valid_vpe(its, desc->its_vmapp_cmd.vpe); unsigned long vpt_addr, vconf_addr; u64 target; bool alloc; @@ -798,6 +799,11 @@ static struct its_vpe *its_build_vmapp_cmd(struct its_node *its, if (is_v4_1(its)) { alloc = !atomic_dec_return(&desc->its_vmapp_cmd.vpe->vmapp_count); its_encode_alloc(cmd, alloc); + /* + * Unmapping a VPE is self-synchronizing on GICv4.1, + * no need to issue a VSYNC. + */ + vpe = NULL; } goto out; @@ -832,7 +838,7 @@ static struct its_vpe *its_build_vmapp_cmd(struct its_node *its, out: its_fixup_cmd(cmd); - return valid_vpe(its, desc->its_vmapp_cmd.vpe); + return vpe; } static struct its_vpe *its_build_vmapti_cmd(struct its_node *its,
As per the GICv4.1 spec (Arm IHI 0069H, 5.3.19): "A VMAPP with {V, Alloc}=={0, x} is self-synchronizing, This means the ITS command queue does not show the command as consumed until all of its effects are completed." Furthermore, VSYNC is allowed to deliver an SError when referencing a non existent VPE. By these definitions, a VMAPP followed by a VSYNC is a bug, as the later references a VPE that has been unmapped by the former. Fix it by eliding the VSYNC in this scenario. Fixes: 64edfaa9a234 ("irqchip/gic-v4.1: Implement the v4.1 flavour of VMAPP") Signed-off-by: Nianyao Tang <tangnianyao@huawei.com> Reviewed-by: Marc Zyngier <maz@kernel.org> --- drivers/irqchip/irq-gic-v3-its.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)