2021-10-06 12:52:34

by Vitaly Kuznetsov

Subject: [PATCH] x86/hyperv: Avoid erroneously sending IPI to 'self'

__send_ipi_mask_ex() uses an optimization: when the target CPU mask is
equal to 'cpu_present_mask' it uses 'HV_GENERIC_SET_ALL' format to avoid
converting the specified cpumask to VP_SET. This case was overlooked when
'exclude_self' parameter was added. As the result, a spurious IPI to
'self' can be send.

Reported-by: Thomas Gleixner <[email protected]>
Fixes: dfb5c1e12c28 ("x86/hyperv: remove on-stack cpumask from hv_send_ipi_mask_allbutself")
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
arch/x86/hyperv/hv_apic.c | 20 +++++++++++++++-----
1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/arch/x86/hyperv/hv_apic.c b/arch/x86/hyperv/hv_apic.c
index 32a1ad356c18..db2d92fb44da 100644
--- a/arch/x86/hyperv/hv_apic.c
+++ b/arch/x86/hyperv/hv_apic.c
@@ -122,17 +122,27 @@ static bool __send_ipi_mask_ex(const struct cpumask *mask, int vector,
ipi_arg->reserved = 0;
ipi_arg->vp_set.valid_bank_mask = 0;

- if (!cpumask_equal(mask, cpu_present_mask)) {
+ /*
+ * Use HV_GENERIC_SET_ALL and avoid converting cpumask to VP_SET
+ * when the IPI is sent to all currently present CPUs.
+ */
+ if (!cpumask_equal(mask, cpu_present_mask) || exclude_self) {
ipi_arg->vp_set.format = HV_GENERIC_SET_SPARSE_4K;
if (exclude_self)
nr_bank = cpumask_to_vpset_noself(&(ipi_arg->vp_set), mask);
else
nr_bank = cpumask_to_vpset(&(ipi_arg->vp_set), mask);
- }
- if (nr_bank < 0)
- goto ipi_mask_ex_done;
- if (!nr_bank)
+
+ /*
+ * 'nr_bank <= 0' means some CPUs in cpumask can't be
+ * represented in VP_SET. Return an error and fall back to
+ * native (architectural) method of sending IPIs.
+ */
+ if (nr_bank <= 0)
+ goto ipi_mask_ex_done;
+ } else {
ipi_arg->vp_set.format = HV_GENERIC_SET_ALL;
+ }

status = hv_do_rep_hypercall(HVCALL_SEND_IPI_EX, 0, nr_bank,
ipi_arg, NULL);
--
2.31.1

2021-10-06 15:17:23

by Michael Kelley (LINUX)

Subject: RE: [PATCH] x86/hyperv: Avoid erroneously sending IPI to 'self'

From: Vitaly Kuznetsov <[email protected]> Sent: Wednesday, October 6, 2021 5:50 AM
>
> __send_ipi_mask_ex() uses an optimization: when the target CPU mask is
> equal to 'cpu_present_mask' it uses 'HV_GENERIC_SET_ALL' format to avoid
> converting the specified cpumask to VP_SET. This case was overlooked when
> 'exclude_self' parameter was added. As the result, a spurious IPI to
> 'self' can be send.
>
> Reported-by: Thomas Gleixner <[email protected]>
> Fixes: dfb5c1e12c28 ("x86/hyperv: remove on-stack cpumask from hv_send_ipi_mask_allbutself")
> Signed-off-by: Vitaly Kuznetsov <[email protected]>
> ---
> arch/x86/hyperv/hv_apic.c | 20 +++++++++++++++-----
> 1 file changed, 15 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/hyperv/hv_apic.c b/arch/x86/hyperv/hv_apic.c
> index 32a1ad356c18..db2d92fb44da 100644
> --- a/arch/x86/hyperv/hv_apic.c
> +++ b/arch/x86/hyperv/hv_apic.c
> @@ -122,17 +122,27 @@ static bool __send_ipi_mask_ex(const struct cpumask *mask, int vector,
> ipi_arg->reserved = 0;
> ipi_arg->vp_set.valid_bank_mask = 0;
>
> - if (!cpumask_equal(mask, cpu_present_mask)) {
> + /*
> + * Use HV_GENERIC_SET_ALL and avoid converting cpumask to VP_SET
> + * when the IPI is sent to all currently present CPUs.
> + */
> + if (!cpumask_equal(mask, cpu_present_mask) || exclude_self) {
> ipi_arg->vp_set.format = HV_GENERIC_SET_SPARSE_4K;
> if (exclude_self)
> nr_bank = cpumask_to_vpset_noself(&(ipi_arg->vp_set), mask);
> else
> nr_bank = cpumask_to_vpset(&(ipi_arg->vp_set), mask);
> - }
> - if (nr_bank < 0)
> - goto ipi_mask_ex_done;
> - if (!nr_bank)
> +
> + /*
> + * 'nr_bank <= 0' means some CPUs in cpumask can't be
> + * represented in VP_SET. Return an error and fall back to
> + * native (architectural) method of sending IPIs.
> + */
> + if (nr_bank <= 0)
> + goto ipi_mask_ex_done;
> + } else {
> ipi_arg->vp_set.format = HV_GENERIC_SET_ALL;
> + }
>
> status = hv_do_rep_hypercall(HVCALL_SEND_IPI_EX, 0, nr_bank,
> ipi_arg, NULL);
> --
> 2.31.1

Reviewed-by: Michael Kelley <[email protected]>

2021-10-06 15:59:55

by Wei Liu

Subject: Re: [PATCH] x86/hyperv: Avoid erroneously sending IPI to 'self'

On Wed, Oct 06, 2021 at 03:15:22PM +0000, Michael Kelley wrote:
> From: Vitaly Kuznetsov <[email protected]> Sent: Wednesday, October 6, 2021 5:50 AM
[...]
> > --
> > 2.31.1
>
> Reviewed-by: Michael Kelley <[email protected]>

LGTM.

Applied to hyperv-fixes. Thanks.

Wei.

>