Subject: [PATCH] xen-pciback: Allow setting PCI_MSIX_FLAGS_MASKALL too

When Xen domain configures MSI-X, the usual approach is to enable MSI-X
together with masking all of them via the config space, then fill the
table and only then clear PCI_MSIX_FLAGS_MASKALL. Allow doing this via
QEMU running in a stub domain.

Previously, when changing PCI_MSIX_FLAGS_MASKALL was not allowed, the
whole write was aborted, preventing change to the PCI_MSIX_FLAGS_ENABLE
bit too.

Note the Xen hypervisor intercepts this write anyway, and may keep the
PCI_MSIX_FLAGS_MASKALL bit set if it wishes to. It will store the
guest-requested state and will apply it eventually.

Signed-off-by: Marek Marczykowski-Górecki <[email protected]>
---
drivers/xen/xen-pciback/conf_space_capability.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/xen/xen-pciback/conf_space_capability.c b/drivers/xen/xen-pciback/conf_space_capability.c
index 5e53b4817f16..097316a74126 100644
--- a/drivers/xen/xen-pciback/conf_space_capability.c
+++ b/drivers/xen/xen-pciback/conf_space_capability.c
@@ -190,13 +190,16 @@ static const struct config_field caplist_pm[] = {
};

static struct msi_msix_field_config {
- u16 enable_bit; /* bit for enabling MSI/MSI-X */
- unsigned int int_type; /* interrupt type for exclusiveness check */
+ u16 enable_bit; /* bit for enabling MSI/MSI-X */
+ u16 allowed_bits; /* bits allowed to be changed */
+ unsigned int int_type; /* interrupt type for exclusiveness check */
} msi_field_config = {
.enable_bit = PCI_MSI_FLAGS_ENABLE,
+ .allowed_bits = PCI_MSI_FLAGS_ENABLE,
.int_type = INTERRUPT_TYPE_MSI,
}, msix_field_config = {
.enable_bit = PCI_MSIX_FLAGS_ENABLE,
+ .allowed_bits = PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL,
.int_type = INTERRUPT_TYPE_MSIX,
};

@@ -229,7 +232,7 @@ static int msi_msix_flags_write(struct pci_dev *dev, int offset, u16 new_value,
return 0;

if (!dev_data->allow_interrupt_control ||
- (new_value ^ old_value) & ~field_config->enable_bit)
+ (new_value ^ old_value) & ~field_config->allowed_bits)
return PCIBIOS_SET_FAILED;

if (new_value & field_config->enable_bit) {
--
2.37.3



2022-11-14 11:09:33

by Jan Beulich

[permalink] [raw]
Subject: Re: [PATCH] xen-pciback: Allow setting PCI_MSIX_FLAGS_MASKALL too

On 14.11.2022 11:31, Marek Marczykowski-Górecki wrote:
> When Xen domain configures MSI-X, the usual approach is to enable MSI-X
> together with masking all of them via the config space, then fill the
> table and only then clear PCI_MSIX_FLAGS_MASKALL. Allow doing this via
> QEMU running in a stub domain.
>
> Previously, when changing PCI_MSIX_FLAGS_MASKALL was not allowed, the
> whole write was aborted, preventing change to the PCI_MSIX_FLAGS_ENABLE
> bit too.
>
> Note the Xen hypervisor intercepts this write anyway, and may keep the
> PCI_MSIX_FLAGS_MASKALL bit set if it wishes to. It will store the
> guest-requested state and will apply it eventually.
>
> Signed-off-by: Marek Marczykowski-Górecki <[email protected]>

Reviewed-by: Jan Beulich <[email protected]>