When Dom0 informs us about MMCFG usability, this may change whether
extended capabilities are available (accessible) for devices. Zap what
might be on record, and re-initialize things.

No synchronization is added for the case where devices may already be in
use. That'll need sorting when (a) DomU support was added and (b) DomU-s
may run already while Dom0 / hwdom still boots (dom0less, Hyperlaunch).

vpci_cleanup_capabilities() also shouldn't have used
pci_find_ext_capability(), as already when the function was introduced
extended config space may not have been (properly) accessible anymore,
no matter whether it was during init. Extended capability cleanup hooks
need to cope with being called when the respective capability doesn't
exist (and hence the corresponding ->init() hook was never called).

Fixes: 70e6dace747e ("vpci: Use cleanup to free capability resource during deassign")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
vpci_reinit_ext_capabilities()'es return value is checked only to log an
error; it doesn't feel quite right to fail the hypercall because of this.
Roger brought up the idea of de-assigning the device in such a case, but
if a driver doesn't use extended capabilities the device would likely
continue to work fine, for Dom0 this probably wouldn't be quite right
anyway, and it's also unclear whether calling deassign_device() could be
done from this context. Something like what pci_check_disable_device()
does may be an option, if we really think we need to "break" the device.

The use of is_hardware_domain() in vpci_cleanup_capabilities() was
uncommented and hence is left so. Shouldn't there be a DomU-related TODO
or FIXME?
---
v5: Don't use pci_find_ext_capability() in vpci_cleanup_capabilities().
    Add assertion in vpci_reinit_ext_capabilities().
v4: Make sure ->cleanup() and ->init() are invoked.
v3: New.

--- a/xen/arch/x86/physdev.c
+++ b/xen/arch/x86/physdev.c
@@ -8,6 +8,8 @@
 #include <xen/guest_access.h>
 #include <xen/iocap.h>
 #include <xen/serial.h>
+#include <xen/vpci.h>
+
 #include <asm/current.h>
 #include <asm/io_apic.h>
 #include <asm/msi.h>
@@ -169,8 +171,17 @@ int cf_check physdev_check_pci_extcfg(st
 
     ASSERT(pdev->seg == info->segment);
     if ( pdev->bus >= info->start_bus && pdev->bus <= info->end_bus )
+    {
+        int rc;
+
         pci_check_extcfg(pdev);
 
+        rc = vpci_reinit_ext_capabilities(pdev);
+        if ( rc )
+            gprintk(XENLOG_ERR, "%pp(%pd): vPCI extcap reinit failed: %d\n",
+                    &pdev->sbdf, pdev->domain, rc);
+    }
+
     return 0;
 }
 #endif /* COMPAT */
--- a/xen/drivers/vpci/cap.c
+++ b/xen/drivers/vpci/cap.c
@@ -285,13 +285,16 @@ static int vpci_init_ext_capability_list
     return 0;
 }
 
-int vpci_init_capabilities(struct pci_dev *pdev)
+int vpci_init_capabilities(struct pci_dev *pdev, bool ext_only)
 {
     int rc;
 
-    rc = vpci_init_capability_list(pdev);
-    if ( rc )
-        return rc;
+    if ( !ext_only )
+    {
+        rc = vpci_init_capability_list(pdev);
+        if ( rc )
+            return rc;
+    }
 
     rc = vpci_init_ext_capability_list(pdev);
     if ( rc )
@@ -305,7 +308,7 @@ int vpci_init_capabilities(struct pci_de
         unsigned int pos = 0;
 
         if ( !is_ext )
-            pos = pci_find_cap_offset(pdev->sbdf, cap);
+            pos = !ext_only ? pci_find_cap_offset(pdev->sbdf, cap) : 0;
         else if ( is_hardware_domain(pdev->domain) )
             pos = pci_find_ext_capability(pdev, cap);
 
@@ -349,22 +352,23 @@ int vpci_init_capabilities(struct pci_de
     return 0;
 }
 
-void vpci_cleanup_capabilities(struct pci_dev *pdev)
+void vpci_cleanup_capabilities(struct pci_dev *pdev, bool ext_only)
 {
     for ( unsigned int i = 0; i < NUM_VPCI_INIT; i++ )
     {
         const vpci_capability_t *capability = &__start_vpci_array[i];
         const unsigned int cap = capability->id;
-        unsigned int pos = 0;
 
         if ( !capability->cleanup )
             continue;
 
-        if ( !capability->is_ext )
-            pos = pci_find_cap_offset(pdev->sbdf, cap);
-        else if ( is_hardware_domain(pdev->domain) )
-            pos = pci_find_ext_capability(pdev, cap);
-        if ( pos )
+        /*
+         * Cannot call pci_find_ext_capability() here, as extended config
+         * space may (no longer) be accessible.
+         */
+        if ( capability->is_ext
+             ? is_hardware_domain(pdev->domain)
+             : !ext_only && pci_find_cap_offset(pdev->sbdf, cap) )
         {
             int rc = capability->cleanup(pdev, false);
 
@@ -376,6 +380,28 @@ void vpci_cleanup_capabilities(struct pc
     }
 }
 
+int vpci_reinit_ext_capabilities(struct pci_dev *pdev)
+{
+    if ( !pdev->vpci )
+        return 0;
+
+    /*
+     * FIXME: DomU support is missing.  For already running domains we may
+     * need to pause them around the entire re-evaluation of extended config
+     * space accessibility.
+     */
+    if ( pdev->domain )
+        ASSERT(pdev->domain == hardware_domain || pdev->domain == dom_io);
+
+    vpci_cleanup_capabilities(pdev, true);
+
+    if ( vpci_remove_registers(pdev->vpci, PCI_CFG_SPACE_SIZE,
+                               PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) )
+        ASSERT_UNREACHABLE();
+
+    return vpci_init_capabilities(pdev, true);
+}
+
 /*
  * Local variables:
  * mode: C
--- a/xen/drivers/vpci/private.h
+++ b/xen/drivers/vpci/private.h
@@ -46,8 +46,8 @@ typedef struct {
 
 int __must_check vpci_init_header(struct pci_dev *pdev);
 
-int vpci_init_capabilities(struct pci_dev *pdev);
-void vpci_cleanup_capabilities(struct pci_dev *pdev);
+int vpci_init_capabilities(struct pci_dev *pdev, bool ext_only);
+void vpci_cleanup_capabilities(struct pci_dev *pdev, bool ext_only);
 
 /* Add/remove a register handler. */
 int __must_check vpci_add_register_mask(struct vpci *vpci,
--- a/xen/drivers/vpci/vpci.c
+++ b/xen/drivers/vpci/vpci.c
@@ -102,7 +102,7 @@ void vpci_deassign_device(struct pci_dev
                     &pdev->domain->vpci_dev_assigned_map);
 #endif
 
-    vpci_cleanup_capabilities(pdev);
+    vpci_cleanup_capabilities(pdev, false);
 
     spin_lock(&pdev->vpci->lock);
     while ( !list_empty(&pdev->vpci->handlers) )
@@ -159,7 +159,7 @@ int vpci_assign_device(struct pci_dev *p
     if ( rc )
         goto out;
 
-    rc = vpci_init_capabilities(pdev);
+    rc = vpci_init_capabilities(pdev, false);
 
  out:
     if ( rc )
--- a/xen/include/xen/vpci.h
+++ b/xen/include/xen/vpci.h
@@ -25,6 +25,8 @@ int __must_check vpci_assign_device(stru
 /* Remove all handlers and free vpci related structures. */
 void vpci_deassign_device(struct pci_dev *pdev);
 
+int vpci_reinit_ext_capabilities(struct pci_dev *pdev);
+
 /* Generic read/write handlers for the PCI config space. */
 uint32_t vpci_read(pci_sbdf_t sbdf, unsigned int reg, unsigned int size);
 void vpci_write(pci_sbdf_t sbdf, unsigned int reg, unsigned int size,
@@ -202,6 +204,11 @@ bool vpci_ecam_read(pci_sbdf_t sbdf, uns
 #else /* !CONFIG_HAS_VPCI */
 struct vpci_vcpu {};
 
+static inline int vpci_reinit_ext_capabilities(struct pci_dev *pdev)
+{
+    return 0;
+}
+
 static inline int vpci_assign_device(struct pci_dev *pdev)
 {
     return 0;

On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
> When Dom0 informs us about MMCFG usability, this may change whether
> extended capabilities are available (accessible) for devices. Zap what
> might be on record, and re-initialize things.
> 
> No synchronization is added for the case where devices may already be in
> use. That'll need sorting when (a) DomU support was added and (b) DomU-s
> may run already while Dom0 / hwdom still boots (dom0less, Hyperlaunch).
> 
> vpci_cleanup_capabilities() also shouldn't have used
> pci_find_ext_capability(), as already when the function was introduced
> extended config space may not have been (properly) accessible anymore,
> no matter whether it was during init. Extended capability cleanup hooks
> need to cope with being called when the respective capability doesn't
> exist (and hence the corresponding ->init() hook was never called).
> 
> Fixes: 70e6dace747e ("vpci: Use cleanup to free capability resource during deassign")
> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> ---
> vpci_reinit_ext_capabilities()'es return value is checked only to log an
> error; it doesn't feel quite right to fail the hypercall because of this.
> Roger brought up the idea of de-assigning the device in such a case, but
> if a driver doesn't use extended capabilities the device would likely
> continue to work fine, for Dom0 this probably wouldn't be quite right
> anyway, and it's also unclear whether calling deassign_device() could be
> done from this context. Something like what pci_check_disable_device()
> does may be an option, if we really think we need to "break" the device.

We may want to add a note there, stating that we have considered all
possible options, and hiding the capability and hoping the owner
domain would continue to work as expected seems the less bad of all of
them?

> The use of is_hardware_domain() in vpci_cleanup_capabilities() was
> uncommented and hence is left so. Shouldn't there be a DomU-related TODO
> or FIXME?

Hm, yes, possibly.  I think limiting extended space availability to
the hardware domain only has been done "just" because we have no
extended capabilities to expose to domUs so far, and I don't think we
even setup the extended capability list in the domU case.

> ---
> v5: Don't use pci_find_ext_capability() in vpci_cleanup_capabilities().
>     Add assertion in vpci_reinit_ext_capabilities().
> v4: Make sure ->cleanup() and ->init() are invoked.
> v3: New.
> 
> --- a/xen/arch/x86/physdev.c
> +++ b/xen/arch/x86/physdev.c
> @@ -8,6 +8,8 @@
>  #include <xen/guest_access.h>
>  #include <xen/iocap.h>
>  #include <xen/serial.h>
> +#include <xen/vpci.h>
> +
>  #include <asm/current.h>
>  #include <asm/io_apic.h>
>  #include <asm/msi.h>
> @@ -169,8 +171,17 @@ int cf_check physdev_check_pci_extcfg(st
>  
>      ASSERT(pdev->seg == info->segment);
>      if ( pdev->bus >= info->start_bus && pdev->bus <= info->end_bus )
> +    {
> +        int rc;
> +
>          pci_check_extcfg(pdev);
>  
> +        rc = vpci_reinit_ext_capabilities(pdev);
> +        if ( rc )
> +            gprintk(XENLOG_ERR, "%pp(%pd): vPCI extcap reinit failed: %d\n",
> +                    &pdev->sbdf, pdev->domain, rc);
> +    }
> +
>      return 0;
>  }
>  #endif /* COMPAT */
> --- a/xen/drivers/vpci/cap.c
> +++ b/xen/drivers/vpci/cap.c
> @@ -285,13 +285,16 @@ static int vpci_init_ext_capability_list
>      return 0;
>  }
>  
> -int vpci_init_capabilities(struct pci_dev *pdev)
> +int vpci_init_capabilities(struct pci_dev *pdev, bool ext_only)
>  {
>      int rc;
>  
> -    rc = vpci_init_capability_list(pdev);
> -    if ( rc )
> -        return rc;
> +    if ( !ext_only )
> +    {
> +        rc = vpci_init_capability_list(pdev);
> +        if ( rc )
> +            return rc;
> +    }
>  
>      rc = vpci_init_ext_capability_list(pdev);
>      if ( rc )
> @@ -305,7 +308,7 @@ int vpci_init_capabilities(struct pci_de
>          unsigned int pos = 0;
>  
>          if ( !is_ext )
> -            pos = pci_find_cap_offset(pdev->sbdf, cap);
> +            pos = !ext_only ? pci_find_cap_offset(pdev->sbdf, cap) : 0;
>          else if ( is_hardware_domain(pdev->domain) )
>              pos = pci_find_ext_capability(pdev, cap);
>  
> @@ -349,22 +352,23 @@ int vpci_init_capabilities(struct pci_de
>      return 0;
>  }
>  
> -void vpci_cleanup_capabilities(struct pci_dev *pdev)
> +void vpci_cleanup_capabilities(struct pci_dev *pdev, bool ext_only)
>  {

You could short-circuit the function here, ie:

if ( ext_only && !is_hardware_domain(pdev->domain) )
    return;

But I'm not sure that would simplify the code of the function much?
Likewise for vpci_init_capabilities().

>      for ( unsigned int i = 0; i < NUM_VPCI_INIT; i++ )
>      {
>          const vpci_capability_t *capability = &__start_vpci_array[i];
>          const unsigned int cap = capability->id;
> -        unsigned int pos = 0;
>  
>          if ( !capability->cleanup )
>              continue;
>  
> -        if ( !capability->is_ext )
> -            pos = pci_find_cap_offset(pdev->sbdf, cap);
> -        else if ( is_hardware_domain(pdev->domain) )
> -            pos = pci_find_ext_capability(pdev, cap);
> -        if ( pos )
> +        /*
> +         * Cannot call pci_find_ext_capability() here, as extended config
> +         * space may (no longer) be accessible.
> +         */
> +        if ( capability->is_ext
> +             ? is_hardware_domain(pdev->domain)
> +             : !ext_only && pci_find_cap_offset(pdev->sbdf, cap) )

Given the changes you have done to the reBAR cleanup function, we
could even call capability->cleanup() on domUs, as the handler has to
deal with uninitialized capabilities either way?

>          {
>              int rc = capability->cleanup(pdev, false);
>  
> @@ -376,6 +380,28 @@ void vpci_cleanup_capabilities(struct pc
>      }
>  }
>  
> +int vpci_reinit_ext_capabilities(struct pci_dev *pdev)
> +{
> +    if ( !pdev->vpci )
> +        return 0;
> +
> +    /*
> +     * FIXME: DomU support is missing.  For already running domains we may
> +     * need to pause them around the entire re-evaluation of extended config
> +     * space accessibility.
> +     */
> +    if ( pdev->domain )
> +        ASSERT(pdev->domain == hardware_domain || pdev->domain == dom_io);

Is this to cope around races?  I don't think it's a valid state to
have pdev->vpci != NULL and pdev->domain == NULL?

Neither you can have pdev->domain == dom_io and pdev->vpci != NULL?

> +
> +    vpci_cleanup_capabilities(pdev, true);
> +
> +    if ( vpci_remove_registers(pdev->vpci, PCI_CFG_SPACE_SIZE,
> +                               PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) )
> +        ASSERT_UNREACHABLE();

Ideally this would better be done the other way around.  We first
remove the handlers, and the cleanup the capabilities.  Just to ensure
no stray handler could end up having cached references to data that's
been freed by vpci_cleanup_capabilities().

And we should take the write_lock(&pdev->domain->pci_lock).

Thanks, Roger.

On 04.03.2026 16:06, Roger Pau Monné wrote:
> On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
>> When Dom0 informs us about MMCFG usability, this may change whether
>> extended capabilities are available (accessible) for devices. Zap what
>> might be on record, and re-initialize things.
>>
>> No synchronization is added for the case where devices may already be in
>> use. That'll need sorting when (a) DomU support was added and (b) DomU-s
>> may run already while Dom0 / hwdom still boots (dom0less, Hyperlaunch).
>>
>> vpci_cleanup_capabilities() also shouldn't have used
>> pci_find_ext_capability(), as already when the function was introduced
>> extended config space may not have been (properly) accessible anymore,
>> no matter whether it was during init. Extended capability cleanup hooks
>> need to cope with being called when the respective capability doesn't
>> exist (and hence the corresponding ->init() hook was never called).
>>
>> Fixes: 70e6dace747e ("vpci: Use cleanup to free capability resource during deassign")
>> Signed-off-by: Jan Beulich <jbeulich@suse.com>
>> ---
>> vpci_reinit_ext_capabilities()'es return value is checked only to log an
>> error; it doesn't feel quite right to fail the hypercall because of this.
>> Roger brought up the idea of de-assigning the device in such a case, but
>> if a driver doesn't use extended capabilities the device would likely
>> continue to work fine, for Dom0 this probably wouldn't be quite right
>> anyway, and it's also unclear whether calling deassign_device() could be
>> done from this context. Something like what pci_check_disable_device()
>> does may be an option, if we really think we need to "break" the device.
> 
> We may want to add a note there, stating that we have considered all
> possible options, and hiding the capability and hoping the owner
> domain would continue to work as expected seems the less bad of all of
> them?

While adding that note it occurred to me that in order to keep the device
as functioning as possible, in the re-init case vpci_init_capabilities()
might better not bail upon encountering a failure, but accumulate the
error while continuing its loop in a best-effort manner. Thoughts? (One
of the two return-s is already guarded by !is_hardware_domain(), so that
could be left alone for the immediate purpose.)

Jan

On Thu, Mar 05, 2026 at 10:00:13AM +0100, Jan Beulich wrote:
> On 04.03.2026 16:06, Roger Pau Monné wrote:
> > On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
> >> When Dom0 informs us about MMCFG usability, this may change whether
> >> extended capabilities are available (accessible) for devices. Zap what
> >> might be on record, and re-initialize things.
> >>
> >> No synchronization is added for the case where devices may already be in
> >> use. That'll need sorting when (a) DomU support was added and (b) DomU-s
> >> may run already while Dom0 / hwdom still boots (dom0less, Hyperlaunch).
> >>
> >> vpci_cleanup_capabilities() also shouldn't have used
> >> pci_find_ext_capability(), as already when the function was introduced
> >> extended config space may not have been (properly) accessible anymore,
> >> no matter whether it was during init. Extended capability cleanup hooks
> >> need to cope with being called when the respective capability doesn't
> >> exist (and hence the corresponding ->init() hook was never called).
> >>
> >> Fixes: 70e6dace747e ("vpci: Use cleanup to free capability resource during deassign")
> >> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> >> ---
> >> vpci_reinit_ext_capabilities()'es return value is checked only to log an
> >> error; it doesn't feel quite right to fail the hypercall because of this.
> >> Roger brought up the idea of de-assigning the device in such a case, but
> >> if a driver doesn't use extended capabilities the device would likely
> >> continue to work fine, for Dom0 this probably wouldn't be quite right
> >> anyway, and it's also unclear whether calling deassign_device() could be
> >> done from this context. Something like what pci_check_disable_device()
> >> does may be an option, if we really think we need to "break" the device.
> > 
> > We may want to add a note there, stating that we have considered all
> > possible options, and hiding the capability and hoping the owner
> > domain would continue to work as expected seems the less bad of all of
> > them?
> 
> While adding that note it occurred to me that in order to keep the device
> as functioning as possible, in the re-init case vpci_init_capabilities()
> might better not bail upon encountering a failure, but accumulate the
> error while continuing its loop in a best-effort manner. Thoughts? (One
> of the two return-s is already guarded by !is_hardware_domain(), so that
> could be left alone for the immediate purpose.)

Right, yes, that would be preferable.  We already print a message for
the failed to init capabilities, so there's no need to print another
one in the caller.

Thanks, Roger.

On 05.03.2026 10:22, Roger Pau Monné wrote:
> On Thu, Mar 05, 2026 at 10:00:13AM +0100, Jan Beulich wrote:
>> On 04.03.2026 16:06, Roger Pau Monné wrote:
>>> On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
>>>> When Dom0 informs us about MMCFG usability, this may change whether
>>>> extended capabilities are available (accessible) for devices. Zap what
>>>> might be on record, and re-initialize things.
>>>>
>>>> No synchronization is added for the case where devices may already be in
>>>> use. That'll need sorting when (a) DomU support was added and (b) DomU-s
>>>> may run already while Dom0 / hwdom still boots (dom0less, Hyperlaunch).
>>>>
>>>> vpci_cleanup_capabilities() also shouldn't have used
>>>> pci_find_ext_capability(), as already when the function was introduced
>>>> extended config space may not have been (properly) accessible anymore,
>>>> no matter whether it was during init. Extended capability cleanup hooks
>>>> need to cope with being called when the respective capability doesn't
>>>> exist (and hence the corresponding ->init() hook was never called).
>>>>
>>>> Fixes: 70e6dace747e ("vpci: Use cleanup to free capability resource during deassign")
>>>> Signed-off-by: Jan Beulich <jbeulich@suse.com>
>>>> ---
>>>> vpci_reinit_ext_capabilities()'es return value is checked only to log an
>>>> error; it doesn't feel quite right to fail the hypercall because of this.
>>>> Roger brought up the idea of de-assigning the device in such a case, but
>>>> if a driver doesn't use extended capabilities the device would likely
>>>> continue to work fine, for Dom0 this probably wouldn't be quite right
>>>> anyway, and it's also unclear whether calling deassign_device() could be
>>>> done from this context. Something like what pci_check_disable_device()
>>>> does may be an option, if we really think we need to "break" the device.
>>>
>>> We may want to add a note there, stating that we have considered all
>>> possible options, and hiding the capability and hoping the owner
>>> domain would continue to work as expected seems the less bad of all of
>>> them?
>>
>> While adding that note it occurred to me that in order to keep the device
>> as functioning as possible, in the re-init case vpci_init_capabilities()
>> might better not bail upon encountering a failure, but accumulate the
>> error while continuing its loop in a best-effort manner. Thoughts? (One
>> of the two return-s is already guarded by !is_hardware_domain(), so that
>> could be left alone for the immediate purpose.)
> 
> Right, yes, that would be preferable.  We already print a message for
> the failed to init capabilities, so there's no need to print another
> one in the caller.

Hmm, that's another aspect I didn't consider. Yes, the log message in the
caller is redundant with the present code structure. If we expect that to
remain like that, I can drop logging anything from
physdev_check_pci_extcfg(). Which then re-raises the question whether
vpci_reinit_ext_capabilities() might better return void. At which point
the comment I put in physdev_check_pci_extcfg() (upon your request) would
want to move there.

But my earlier question went in a different direction, and you didn't
comment on that at all.

Jan

On Thu, Mar 05, 2026 at 10:45:29AM +0100, Jan Beulich wrote:
> On 05.03.2026 10:22, Roger Pau Monné wrote:
> > On Thu, Mar 05, 2026 at 10:00:13AM +0100, Jan Beulich wrote:
> >> On 04.03.2026 16:06, Roger Pau Monné wrote:
> >>> On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
> >>>> When Dom0 informs us about MMCFG usability, this may change whether
> >>>> extended capabilities are available (accessible) for devices. Zap what
> >>>> might be on record, and re-initialize things.
> >>>>
> >>>> No synchronization is added for the case where devices may already be in
> >>>> use. That'll need sorting when (a) DomU support was added and (b) DomU-s
> >>>> may run already while Dom0 / hwdom still boots (dom0less, Hyperlaunch).
> >>>>
> >>>> vpci_cleanup_capabilities() also shouldn't have used
> >>>> pci_find_ext_capability(), as already when the function was introduced
> >>>> extended config space may not have been (properly) accessible anymore,
> >>>> no matter whether it was during init. Extended capability cleanup hooks
> >>>> need to cope with being called when the respective capability doesn't
> >>>> exist (and hence the corresponding ->init() hook was never called).
> >>>>
> >>>> Fixes: 70e6dace747e ("vpci: Use cleanup to free capability resource during deassign")
> >>>> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> >>>> ---
> >>>> vpci_reinit_ext_capabilities()'es return value is checked only to log an
> >>>> error; it doesn't feel quite right to fail the hypercall because of this.
> >>>> Roger brought up the idea of de-assigning the device in such a case, but
> >>>> if a driver doesn't use extended capabilities the device would likely
> >>>> continue to work fine, for Dom0 this probably wouldn't be quite right
> >>>> anyway, and it's also unclear whether calling deassign_device() could be
> >>>> done from this context. Something like what pci_check_disable_device()
> >>>> does may be an option, if we really think we need to "break" the device.
> >>>
> >>> We may want to add a note there, stating that we have considered all
> >>> possible options, and hiding the capability and hoping the owner
> >>> domain would continue to work as expected seems the less bad of all of
> >>> them?
> >>
> >> While adding that note it occurred to me that in order to keep the device
> >> as functioning as possible, in the re-init case vpci_init_capabilities()
> >> might better not bail upon encountering a failure, but accumulate the
> >> error while continuing its loop in a best-effort manner. Thoughts? (One
> >> of the two return-s is already guarded by !is_hardware_domain(), so that
> >> could be left alone for the immediate purpose.)
> > 
> > Right, yes, that would be preferable.  We already print a message for
> > the failed to init capabilities, so there's no need to print another
> > one in the caller.
> 
> Hmm, that's another aspect I didn't consider. Yes, the log message in the
> caller is redundant with the present code structure. If we expect that to
> remain like that, I can drop logging anything from
> physdev_check_pci_extcfg(). Which then re-raises the question whether
> vpci_reinit_ext_capabilities() might better return void. At which point
> the comment I put in physdev_check_pci_extcfg() (upon your request) would
> want to move there.
> 
> But my earlier question went in a different direction, and you didn't
> comment on that at all.

Yes, I think we should accumulate errors.  One device failing doesn't
mean the rest will also fail.  We should continue the loop.

Thanks, Roger.

On 04.03.2026 16:06, Roger Pau Monné wrote:
> On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
>> When Dom0 informs us about MMCFG usability, this may change whether
>> extended capabilities are available (accessible) for devices. Zap what
>> might be on record, and re-initialize things.
>>
>> No synchronization is added for the case where devices may already be in
>> use. That'll need sorting when (a) DomU support was added and (b) DomU-s
>> may run already while Dom0 / hwdom still boots (dom0less, Hyperlaunch).
>>
>> vpci_cleanup_capabilities() also shouldn't have used
>> pci_find_ext_capability(), as already when the function was introduced
>> extended config space may not have been (properly) accessible anymore,
>> no matter whether it was during init. Extended capability cleanup hooks
>> need to cope with being called when the respective capability doesn't
>> exist (and hence the corresponding ->init() hook was never called).
>>
>> Fixes: 70e6dace747e ("vpci: Use cleanup to free capability resource during deassign")
>> Signed-off-by: Jan Beulich <jbeulich@suse.com>
>> ---
>> vpci_reinit_ext_capabilities()'es return value is checked only to log an
>> error; it doesn't feel quite right to fail the hypercall because of this.
>> Roger brought up the idea of de-assigning the device in such a case, but
>> if a driver doesn't use extended capabilities the device would likely
>> continue to work fine, for Dom0 this probably wouldn't be quite right
>> anyway, and it's also unclear whether calling deassign_device() could be
>> done from this context. Something like what pci_check_disable_device()
>> does may be an option, if we really think we need to "break" the device.
> 
> We may want to add a note there, stating that we have considered all
> possible options, and hiding the capability and hoping the owner
> domain would continue to work as expected seems the less bad of all of
> them?

I'll see what I can do.

>> The use of is_hardware_domain() in vpci_cleanup_capabilities() was
>> uncommented and hence is left so. Shouldn't there be a DomU-related TODO
>> or FIXME?
> 
> Hm, yes, possibly.  I think limiting extended space availability to
> the hardware domain only has been done "just" because we have no
> extended capabilities to expose to domUs so far, and I don't think we
> even setup the extended capability list in the domU case.

Considering how many things there are to be done for DomU support, I
think it would help if every place where e.g. is_hardware_domain() is
used only as surrogate would be properly annotated. Or perhaps properly
named predicates could be introduced right away, so one can actually go
hunt for all of them. Then again is_hardware_domain() is also something
one can go hunt for.

>> @@ -349,22 +352,23 @@ int vpci_init_capabilities(struct pci_de
>>      return 0;
>>  }
>>  
>> -void vpci_cleanup_capabilities(struct pci_dev *pdev)
>> +void vpci_cleanup_capabilities(struct pci_dev *pdev, bool ext_only)
>>  {
> 
> You could short-circuit the function here, ie:
> 
> if ( ext_only && !is_hardware_domain(pdev->domain) )
>     return;
> 
> But I'm not sure that would simplify the code of the function much?
> Likewise for vpci_init_capabilities().

Such a short-circuit would need replacing / dropping once DomU support is
added. I was hoping the chosen arrangement would make for a little less
churn at that time. I'll listen to your advice, though, just that the
question gives the impression you're not quite sure either.

>>      for ( unsigned int i = 0; i < NUM_VPCI_INIT; i++ )
>>      {
>>          const vpci_capability_t *capability = &__start_vpci_array[i];
>>          const unsigned int cap = capability->id;
>> -        unsigned int pos = 0;
>>  
>>          if ( !capability->cleanup )
>>              continue;
>>  
>> -        if ( !capability->is_ext )
>> -            pos = pci_find_cap_offset(pdev->sbdf, cap);
>> -        else if ( is_hardware_domain(pdev->domain) )
>> -            pos = pci_find_ext_capability(pdev, cap);
>> -        if ( pos )
>> +        /*
>> +         * Cannot call pci_find_ext_capability() here, as extended config
>> +         * space may (no longer) be accessible.
>> +         */
>> +        if ( capability->is_ext
>> +             ? is_hardware_domain(pdev->domain)
>> +             : !ext_only && pci_find_cap_offset(pdev->sbdf, cap) )
> 
> Given the changes you have done to the reBAR cleanup function, we
> could even call capability->cleanup() on domUs, as the handler has to
> deal with uninitialized capabilities either way?

Hmm, yes, looks like we could.

>> @@ -376,6 +380,28 @@ void vpci_cleanup_capabilities(struct pc
>>      }
>>  }
>>  
>> +int vpci_reinit_ext_capabilities(struct pci_dev *pdev)
>> +{
>> +    if ( !pdev->vpci )
>> +        return 0;
>> +
>> +    /*
>> +     * FIXME: DomU support is missing.  For already running domains we may
>> +     * need to pause them around the entire re-evaluation of extended config
>> +     * space accessibility.
>> +     */
>> +    if ( pdev->domain )
>> +        ASSERT(pdev->domain == hardware_domain || pdev->domain == dom_io);
> 
> Is this to cope around races?  I don't think it's a valid state to
> have pdev->vpci != NULL and pdev->domain == NULL?

Knowing that I had seen pdev-s with NULL domains, I'm perhaps overly cautious.
Yes, ->vpci being non-NULL ought to demand a proper owner.

> Neither you can have pdev->domain == dom_io and pdev->vpci != NULL?

Same here, looks like I went too far.

>> +
>> +    vpci_cleanup_capabilities(pdev, true);
>> +
>> +    if ( vpci_remove_registers(pdev->vpci, PCI_CFG_SPACE_SIZE,
>> +                               PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) )
>> +        ASSERT_UNREACHABLE();
> 
> Ideally this would better be done the other way around.  We first
> remove the handlers, and the cleanup the capabilities.  Just to ensure
> no stray handler could end up having cached references to data that's
> been freed by vpci_cleanup_capabilities().

And maybe not just that: For the hwdom case cleanup_rebar() adds new handlers,
which we'd wrongly purge again right away. (Because we pass "false" for "hide",
this isn't an active issue right now.)

> And we should take the write_lock(&pdev->domain->pci_lock).

Now this is a request that I'm struggling with some. I can see that callers
of vpci_{init,cleanup}_capabilities() assert that the lock is being held, yet
it's not quite clear to me why that's needed. Shouldn't vPCI internals all
synchronize on the vPCI lock of the domain?

Jan

On Wed, Mar 04, 2026 at 04:39:00PM +0100, Jan Beulich wrote:
> On 04.03.2026 16:06, Roger Pau Monné wrote:
> > On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
> >> When Dom0 informs us about MMCFG usability, this may change whether
> >> extended capabilities are available (accessible) for devices. Zap what
> >> might be on record, and re-initialize things.
> >>
> >> No synchronization is added for the case where devices may already be in
> >> use. That'll need sorting when (a) DomU support was added and (b) DomU-s
> >> may run already while Dom0 / hwdom still boots (dom0less, Hyperlaunch).
> >>
> >> vpci_cleanup_capabilities() also shouldn't have used
> >> pci_find_ext_capability(), as already when the function was introduced
> >> extended config space may not have been (properly) accessible anymore,
> >> no matter whether it was during init. Extended capability cleanup hooks
> >> need to cope with being called when the respective capability doesn't
> >> exist (and hence the corresponding ->init() hook was never called).
> >>
> >> Fixes: 70e6dace747e ("vpci: Use cleanup to free capability resource during deassign")
> >> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> >> ---
> >> vpci_reinit_ext_capabilities()'es return value is checked only to log an
> >> error; it doesn't feel quite right to fail the hypercall because of this.
> >> Roger brought up the idea of de-assigning the device in such a case, but
> >> if a driver doesn't use extended capabilities the device would likely
> >> continue to work fine, for Dom0 this probably wouldn't be quite right
> >> anyway, and it's also unclear whether calling deassign_device() could be
> >> done from this context. Something like what pci_check_disable_device()
> >> does may be an option, if we really think we need to "break" the device.
> > 
> > We may want to add a note there, stating that we have considered all
> > possible options, and hiding the capability and hoping the owner
> > domain would continue to work as expected seems the less bad of all of
> > them?
> 
> I'll see what I can do.
> 
> >> The use of is_hardware_domain() in vpci_cleanup_capabilities() was
> >> uncommented and hence is left so. Shouldn't there be a DomU-related TODO
> >> or FIXME?
> > 
> > Hm, yes, possibly.  I think limiting extended space availability to
> > the hardware domain only has been done "just" because we have no
> > extended capabilities to expose to domUs so far, and I don't think we
> > even setup the extended capability list in the domU case.
> 
> Considering how many things there are to be done for DomU support, I
> think it would help if every place where e.g. is_hardware_domain() is
> used only as surrogate would be properly annotated. Or perhaps properly
> named predicates could be introduced right away, so one can actually go
> hunt for all of them. Then again is_hardware_domain() is also something
> one can go hunt for.

I would mind having more concrete predicates, even if right now they
are some kind of dummy checks.  However it might be difficult to
formulate those know without having the full picture of what domU
support requires.

> >> @@ -349,22 +352,23 @@ int vpci_init_capabilities(struct pci_de
> >>      return 0;
> >>  }
> >>  
> >> -void vpci_cleanup_capabilities(struct pci_dev *pdev)
> >> +void vpci_cleanup_capabilities(struct pci_dev *pdev, bool ext_only)
> >>  {
> > 
> > You could short-circuit the function here, ie:
> > 
> > if ( ext_only && !is_hardware_domain(pdev->domain) )
> >     return;
> > 
> > But I'm not sure that would simplify the code of the function much?
> > Likewise for vpci_init_capabilities().
> 
> Such a short-circuit would need replacing / dropping once DomU support is
> added. I was hoping the chosen arrangement would make for a little less
> churn at that time. I'll listen to your advice, though, just that the
> question gives the impression you're not quite sure either.

Yeah, I wasn't fully sure.  IT would be nice if we could add those
short circuits now, and then once domU support is in place we just
remove teh shortcuts and it works for domU also.  But I fear more
changes will be needed anyway, at which point the short-circuit is
not that attractive to use.

> >> +
> >> +    vpci_cleanup_capabilities(pdev, true);
> >> +
> >> +    if ( vpci_remove_registers(pdev->vpci, PCI_CFG_SPACE_SIZE,
> >> +                               PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) )
> >> +        ASSERT_UNREACHABLE();
> > 
> > Ideally this would better be done the other way around.  We first
> > remove the handlers, and the cleanup the capabilities.  Just to ensure
> > no stray handler could end up having cached references to data that's
> > been freed by vpci_cleanup_capabilities().
> 
> And maybe not just that: For the hwdom case cleanup_rebar() adds new handlers,
> which we'd wrongly purge again right away. (Because we pass "false" for "hide",
> this isn't an active issue right now.)
> 
> > And we should take the write_lock(&pdev->domain->pci_lock).
> 
> Now this is a request that I'm struggling with some. I can see that callers
> of vpci_{init,cleanup}_capabilities() assert that the lock is being held, yet
> it's not quite clear to me why that's needed. Shouldn't vPCI internals all
> synchronize on the vPCI lock of the domain?

Right, the callers of the handlers already hold the locks, and the
removal of the handlers should also hold the locks.  The point of
taking the d->pci_lock is to avoid the device from being removed
while there are vPCI accesses against it being done.  The vPCI lock is
fine for vPCI internals, but functions that deal with addition or
removal of devices need the d->pci_lock to avoid races with possibly
freeing pdev->vpci while in use.

I think you are right, and for the usage here (that doesn't add or
remove pdev->vpci itself), the internal vPCI lock should be enough.

Thanks, Roger.

On 04.03.2026 17:53, Roger Pau Monné wrote:
> On Wed, Mar 04, 2026 at 04:39:00PM +0100, Jan Beulich wrote:
>> On 04.03.2026 16:06, Roger Pau Monné wrote:
>>> On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
>>>> @@ -349,22 +352,23 @@ int vpci_init_capabilities(struct pci_de
>>>>      return 0;
>>>>  }
>>>>  
>>>> -void vpci_cleanup_capabilities(struct pci_dev *pdev)
>>>> +void vpci_cleanup_capabilities(struct pci_dev *pdev, bool ext_only)
>>>>  {
>>>
>>> You could short-circuit the function here, ie:
>>>
>>> if ( ext_only && !is_hardware_domain(pdev->domain) )
>>>     return;
>>>
>>> But I'm not sure that would simplify the code of the function much?
>>> Likewise for vpci_init_capabilities().
>>
>> Such a short-circuit would need replacing / dropping once DomU support is
>> added. I was hoping the chosen arrangement would make for a little less
>> churn at that time. I'll listen to your advice, though, just that the
>> question gives the impression you're not quite sure either.
> 
> Yeah, I wasn't fully sure.  IT would be nice if we could add those
> short circuits now, and then once domU support is in place we just
> remove teh shortcuts and it works for domU also.  But I fear more
> changes will be needed anyway, at which point the short-circuit is
> not that attractive to use.

As per your other request (calling ->cleanup() even for DomU-s) the use of
is_hardware_domain() would go away anyway, and the function would be ready
for use for DomU-s as well.

>>>> +
>>>> +    vpci_cleanup_capabilities(pdev, true);
>>>> +
>>>> +    if ( vpci_remove_registers(pdev->vpci, PCI_CFG_SPACE_SIZE,
>>>> +                               PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) )
>>>> +        ASSERT_UNREACHABLE();
>>>
>>> Ideally this would better be done the other way around.  We first
>>> remove the handlers, and the cleanup the capabilities.  Just to ensure
>>> no stray handler could end up having cached references to data that's
>>> been freed by vpci_cleanup_capabilities().
>>
>> And maybe not just that: For the hwdom case cleanup_rebar() adds new handlers,
>> which we'd wrongly purge again right away. (Because we pass "false" for "hide",
>> this isn't an active issue right now.)
>>
>>> And we should take the write_lock(&pdev->domain->pci_lock).
>>
>> Now this is a request that I'm struggling with some. I can see that callers
>> of vpci_{init,cleanup}_capabilities() assert that the lock is being held, yet
>> it's not quite clear to me why that's needed. Shouldn't vPCI internals all
>> synchronize on the vPCI lock of the domain?
> 
> Right, the callers of the handlers already hold the locks, and the
> removal of the handlers should also hold the locks.  The point of
> taking the d->pci_lock is to avoid the device from being removed
> while there are vPCI accesses against it being done.  The vPCI lock is
> fine for vPCI internals, but functions that deal with addition or
> removal of devices need the d->pci_lock to avoid races with possibly
> freeing pdev->vpci while in use.
> 
> I think you are right, and for the usage here (that doesn't add or
> remove pdev->vpci itself), the internal vPCI lock should be enough.

Well, we could take two positions: Either we say that as we're being called
from a context where the PCI device is being operated on anyway, we can
assume it can't go away. Then no further locking would be needed here. Or
we want to explicitly guard against that, in which case (seeing that
nothing is added / removed), d->pci_lock may want read-locking?

Jan

On Thu, Mar 05, 2026 at 09:40:32AM +0100, Jan Beulich wrote:
> On 04.03.2026 17:53, Roger Pau Monné wrote:
> > On Wed, Mar 04, 2026 at 04:39:00PM +0100, Jan Beulich wrote:
> >> On 04.03.2026 16:06, Roger Pau Monné wrote:
> >>> On Wed, Feb 25, 2026 at 12:44:44PM +0100, Jan Beulich wrote:
> >>>> @@ -349,22 +352,23 @@ int vpci_init_capabilities(struct pci_de
> >>>>      return 0;
> >>>>  }
> >>>>  
> >>>> -void vpci_cleanup_capabilities(struct pci_dev *pdev)
> >>>> +void vpci_cleanup_capabilities(struct pci_dev *pdev, bool ext_only)
> >>>>  {
> >>>
> >>> You could short-circuit the function here, ie:
> >>>
> >>> if ( ext_only && !is_hardware_domain(pdev->domain) )
> >>>     return;
> >>>
> >>> But I'm not sure that would simplify the code of the function much?
> >>> Likewise for vpci_init_capabilities().
> >>
> >> Such a short-circuit would need replacing / dropping once DomU support is
> >> added. I was hoping the chosen arrangement would make for a little less
> >> churn at that time. I'll listen to your advice, though, just that the
> >> question gives the impression you're not quite sure either.
> > 
> > Yeah, I wasn't fully sure.  IT would be nice if we could add those
> > short circuits now, and then once domU support is in place we just
> > remove teh shortcuts and it works for domU also.  But I fear more
> > changes will be needed anyway, at which point the short-circuit is
> > not that attractive to use.
> 
> As per your other request (calling ->cleanup() even for DomU-s) the use of
> is_hardware_domain() would go away anyway, and the function would be ready
> for use for DomU-s as well.

OK, so that one is (possibly) sorted then.

> >>>> +
> >>>> +    vpci_cleanup_capabilities(pdev, true);
> >>>> +
> >>>> +    if ( vpci_remove_registers(pdev->vpci, PCI_CFG_SPACE_SIZE,
> >>>> +                               PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) )
> >>>> +        ASSERT_UNREACHABLE();
> >>>
> >>> Ideally this would better be done the other way around.  We first
> >>> remove the handlers, and the cleanup the capabilities.  Just to ensure
> >>> no stray handler could end up having cached references to data that's
> >>> been freed by vpci_cleanup_capabilities().
> >>
> >> And maybe not just that: For the hwdom case cleanup_rebar() adds new handlers,
> >> which we'd wrongly purge again right away. (Because we pass "false" for "hide",
> >> this isn't an active issue right now.)
> >>
> >>> And we should take the write_lock(&pdev->domain->pci_lock).
> >>
> >> Now this is a request that I'm struggling with some. I can see that callers
> >> of vpci_{init,cleanup}_capabilities() assert that the lock is being held, yet
> >> it's not quite clear to me why that's needed. Shouldn't vPCI internals all
> >> synchronize on the vPCI lock of the domain?
> > 
> > Right, the callers of the handlers already hold the locks, and the
> > removal of the handlers should also hold the locks.  The point of
> > taking the d->pci_lock is to avoid the device from being removed
> > while there are vPCI accesses against it being done.  The vPCI lock is
> > fine for vPCI internals, but functions that deal with addition or
> > removal of devices need the d->pci_lock to avoid races with possibly
> > freeing pdev->vpci while in use.
> > 
> > I think you are right, and for the usage here (that doesn't add or
> > remove pdev->vpci itself), the internal vPCI lock should be enough.
> 
> Well, we could take two positions: Either we say that as we're being called
> from a context where the PCI device is being operated on anyway, we can
> assume it can't go away. Then no further locking would be needed here. Or
> we want to explicitly guard against that, in which case (seeing that
> nothing is added / removed), d->pci_lock may want read-locking?

In this context the caller is already holding the pcidevs lock, so
there's no risk of the device being de-assigned.

My reasoning for taking the d->pci_lock in write mode was to prevent
any concurrent access to vPCI while the changes to the emulated config
space is taken place, but you are right that the vPCI lock if used
properly should prevent races.

Thanks, Roger.