diff mbox series

[v6,6/7] xen/arm: process pending vPCI map/unmap operations

Message ID 20211105063326.939843-7-andr2000@gmail.com (mailing list archive)
State Superseded
Headers show
Series PCI devices passthrough on Arm, part 2 | expand

Commit Message

Oleksandr Andrushchenko Nov. 5, 2021, 6:33 a.m. UTC 
From: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>

vPCI may map and unmap PCI device memory (BARs) being passed through which
may take a lot of time. For this those operations may be deferred to be
performed later, so that they can be safely preempted.

Currently this deferred processing is happening in common IOREQ code
which doesn't seem to be the right place for x86 and is even more
doubtful because IOREQ may not be enabled for Arm at all.
So, for Arm the pending vPCI work may have no chance to be executed
if the processing is left as is in the common IOREQ code only.
For that reason make vPCI processing happen in arch specific code.

Please be aware that there are a few outstanding TODOs affecting this
code path, see xen/drivers/vpci/header.c:map_range and
xen/drivers/vpci/header.c:vpci_process_pending.

Signed-off-by: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>
---
Cc: Andrew Cooper <andrew.cooper3@citrix.com>
Cc: Paul Durrant <paul@xen.org>

Since v5:
 - check_for_vcpu_work: vPCI addition is moved before the
   vcpu_ioreq__handle_completion(v). This is to avoid differences
   with the x86 version. (Julien)
Since v2:
 - update commit message with more insight on x86, IOREQ and Arm
 - restored order of invocation for IOREQ and vPCI processing (Jan)
Since v1:
 - Moved the check for pending vpci work from the common IOREQ code
   to hvm_do_resume on x86
 - Re-worked the code for Arm to ensure we don't miss pending vPCI work
---
 xen/arch/arm/traps.c   | 13 +++++++++++++
 xen/arch/x86/hvm/hvm.c |  6 ++++++
 xen/common/ioreq.c     |  9 ---------
 3 files changed, 19 insertions(+), 9 deletions(-)

Comments

Jan Beulich Nov. 5, 2021, 7:40 a.m. UTC | #1 
On 05.11.2021 07:33, Oleksandr Andrushchenko wrote:
> --- a/xen/arch/x86/hvm/hvm.c
> +++ b/xen/arch/x86/hvm/hvm.c
> @@ -546,6 +546,12 @@ void hvm_do_resume(struct vcpu *v)
>  
>      pt_restore_timer(v);
>  
> +    if ( has_vpci(v->domain) && vpci_process_pending(v) )
> +    {
> +        raise_softirq(SCHEDULE_SOFTIRQ);
> +        return;
> +    }
> +
>      if ( !vcpu_ioreq_handle_completion(v) )
>          return;
>  

For this part of the change:
Acked-by: Jan Beulich <jbeulich@suse.com>

Jan
Julien Grall Nov. 17, 2021, 9:26 p.m. UTC | #2 
Hi Oleksandr,

On 05/11/2021 06:33, Oleksandr Andrushchenko wrote:
> From: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>
> 
> vPCI may map and unmap PCI device memory (BARs) being passed through which
> may take a lot of time. For this those operations may be deferred to be
> performed later, so that they can be safely preempted.
> 
> Currently this deferred processing is happening in common IOREQ code
> which doesn't seem to be the right place for x86 and is even more
> doubtful because IOREQ may not be enabled for Arm at all.
> So, for Arm the pending vPCI work may have no chance to be executed
> if the processing is left as is in the common IOREQ code only.
> For that reason make vPCI processing happen in arch specific code.
> 
> Please be aware that there are a few outstanding TODOs affecting this
> code path, see xen/drivers/vpci/header.c:map_range and
> xen/drivers/vpci/header.c:vpci_process_pending.
> 
> Signed-off-by: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>

Reviewed-by: Julien Grall <jgrall@amazon.com>

Cheers,
diff mbox series

Patch

diff --git a/xen/arch/arm/traps.c b/xen/arch/arm/traps.c
index 219ab3c3fbde..8757210a798b 100644
--- a/xen/arch/arm/traps.c
+++ b/xen/arch/arm/traps.c
@@ -34,6 +34,7 @@ 
 #include <xen/symbols.h>
 #include <xen/version.h>
 #include <xen/virtual_region.h>
+#include <xen/vpci.h>
 
 #include <public/sched.h>
 #include <public/xen.h>
@@ -2290,6 +2291,18 @@  static bool check_for_vcpu_work(void)
 {
     struct vcpu *v = current;
 
+    if ( has_vpci(v->domain) )
+    {
+        bool pending;
+
+        local_irq_enable();
+        pending = vpci_process_pending(v);
+        local_irq_disable();
+
+        if ( pending )
+            return true;
+    }
+
 #ifdef CONFIG_IOREQ_SERVER
     if ( domain_has_ioreq_server(v->domain) )
     {
diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index eee365711d63..096a61b7ea02 100644
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -546,6 +546,12 @@  void hvm_do_resume(struct vcpu *v)
 
     pt_restore_timer(v);
 
+    if ( has_vpci(v->domain) && vpci_process_pending(v) )
+    {
+        raise_softirq(SCHEDULE_SOFTIRQ);
+        return;
+    }
+
     if ( !vcpu_ioreq_handle_completion(v) )
         return;
 
diff --git a/xen/common/ioreq.c b/xen/common/ioreq.c
index d732dc045df9..689d256544c8 100644
--- a/xen/common/ioreq.c
+++ b/xen/common/ioreq.c
@@ -25,9 +25,7 @@ 
 #include <xen/lib.h>
 #include <xen/paging.h>
 #include <xen/sched.h>
-#include <xen/softirq.h>
 #include <xen/trace.h>
-#include <xen/vpci.h>
 
 #include <asm/guest_atomics.h>
 #include <asm/ioreq.h>
@@ -212,19 +210,12 @@  static bool wait_for_io(struct ioreq_vcpu *sv, ioreq_t *p)
 
 bool vcpu_ioreq_handle_completion(struct vcpu *v)
 {
-    struct domain *d = v->domain;
     struct vcpu_io *vio = &v->io;
     struct ioreq_server *s;
     struct ioreq_vcpu *sv;
     enum vio_completion completion;
     bool res = true;
 
-    if ( has_vpci(d) && vpci_process_pending(v) )
-    {
-        raise_softirq(SCHEDULE_SOFTIRQ);
-        return false;
-    }
-
     while ( (sv = get_pending_vcpu(v, &s)) != NULL )
         if ( !wait_for_io(sv, get_ioreq(s, v)) )
             return false;