| Message ID | 5719ECCB02000078000E486F@prv-mh.provo.novell.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 22/04/16 08:20, Jan Beulich wrote: > When a guest unmasks MSI-X interrupts before enabling MSI-X on the > device, so far nothing updates the {host,guest}_masked internal state; > this to date only gets done when MSI-X is already enabled. This is why > half way recent Linux works (as it enables MSI-X first), while Windows > doesn't (as it enables MSI-X only after having set up und unmasked all > vectors). Since with a successful write to the vector control field > everything is ready internally, we should also update internal tracking > state there, regardless of the device's MSI-X enabled state. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
>>> On 22.04.16 at 18:35, <andrew.cooper3@citrix.com> wrote: > On 22/04/16 08:20, Jan Beulich wrote: >> When a guest unmasks MSI-X interrupts before enabling MSI-X on the >> device, so far nothing updates the {host,guest}_masked internal state; >> this to date only gets done when MSI-X is already enabled. This is why >> half way recent Linux works (as it enables MSI-X first), while Windows >> doesn't (as it enables MSI-X only after having set up und unmasked all >> vectors). Since with a successful write to the vector control field >> everything is ready internally, we should also update internal tracking >> state there, regardless of the device's MSI-X enabled state. >> >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > > Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Wei, just noticed I forgot to Cc you on this one, which definitely is intended for 4.7 (along with the other two MSI related ones). Jan
On Mon, Apr 25, 2016 at 05:24:43AM -0600, Jan Beulich wrote: > >>> On 22.04.16 at 18:35, <andrew.cooper3@citrix.com> wrote: > > On 22/04/16 08:20, Jan Beulich wrote: > >> When a guest unmasks MSI-X interrupts before enabling MSI-X on the > >> device, so far nothing updates the {host,guest}_masked internal state; > >> this to date only gets done when MSI-X is already enabled. This is why > >> half way recent Linux works (as it enables MSI-X first), while Windows > >> doesn't (as it enables MSI-X only after having set up und unmasked all > >> vectors). Since with a successful write to the vector control field > >> everything is ready internally, we should also update internal tracking > >> state there, regardless of the device's MSI-X enabled state. > >> > >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > > > > Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> > Release-acked-by: Wei Liu <wei.liu2@citrix.com>
--- a/xen/arch/x86/msi.c +++ b/xen/arch/x86/msi.c @@ -434,8 +434,13 @@ static bool_t msi_set_mask_bit(struct ir { writel(flag, entry->mask_base + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET); readl(entry->mask_base + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET); + if ( likely(control & PCI_MSIX_FLAGS_ENABLE) ) break; + + entry->msi_attrib.host_masked = host; + entry->msi_attrib.guest_masked = guest; + flag = 1; } else if ( flag && !(control & PCI_MSIX_FLAGS_MASKALL) )
When a guest unmasks MSI-X interrupts before enabling MSI-X on the device, so far nothing updates the {host,guest}_masked internal state; this to date only gets done when MSI-X is already enabled. This is why half way recent Linux works (as it enables MSI-X first), while Windows doesn't (as it enables MSI-X only after having set up und unmasked all vectors). Since with a successful write to the vector control field everything is ready internally, we should also update internal tracking state there, regardless of the device's MSI-X enabled state. Signed-off-by: Jan Beulich <jbeulich@suse.com> x86/MSI-X: correctly track interrupt masking state When a guest unmasks MSI-X interrupts before enabling MSI-X on the device, so far nothing updates the {host,guest}_masked internal state; this to date only gets done when MSI-X is already enabled. This is why half way recent Linux works (as it enables MSI-X first), while Windows doesn't (as it enables MSI-X only after having set up und unmasked all vectors). Since with a successful write to the vector control field everything is ready internally, we should also update internal tracking state there, regardless of the device's MSI-X enabled state. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- a/xen/arch/x86/msi.c +++ b/xen/arch/x86/msi.c @@ -434,8 +434,13 @@ static bool_t msi_set_mask_bit(struct ir { writel(flag, entry->mask_base + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET); readl(entry->mask_base + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET); + if ( likely(control & PCI_MSIX_FLAGS_ENABLE) ) break; + + entry->msi_attrib.host_masked = host; + entry->msi_attrib.guest_masked = guest; + flag = 1; } else if ( flag && !(control & PCI_MSIX_FLAGS_MASKALL) )