qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Qemu-devel] Re: [PATCH] pci: fix pci_find_bus().


From: Michael S. Tsirkin
Subject: [Qemu-devel] Re: [PATCH] pci: fix pci_find_bus().
Date: Sun, 11 Apr 2010 13:46:16 +0300
User-agent: Mutt/1.5.19 (2009-01-05)

On Fri, Apr 09, 2010 at 07:13:24PM +0900, Isaku Yamahata wrote:
> When looking down child bus, it looked parent bridge's
> bus number.
> It should look child bridge's.
> 
> Cc: Blue Swirl <address@hidden>
> Cc: "Michael S. Tsirkin" <address@hidden>
> Signed-off-by: Isaku Yamahata <address@hidden>
> ---
>  hw/pci.c |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/hw/pci.c b/hw/pci.c
> index 0dbca17..2f6907b 100644
> --- a/hw/pci.c
> +++ b/hw/pci.c
> @@ -1557,9 +1557,9 @@ PCIBus *pci_find_bus(PCIBus *bus, int bus_num)
>  
>      /* try child bus */
>      QLIST_FOREACH(sec, &bus->child, sibling) {
> -        if (!bus->parent_dev /* pci host bridge */
> +        if (!sec->parent_dev /* pci host bridge */

I don't understand this first test. As far as I can tell
secondary bus must always have a device, as set by
pci_register_secondary_bus.  Should this
rather be assert(sec->parent_dev)?

>              || (pci_bus_num(sec) <= bus_num &&


And so the above should just use
sec->parent_dev->config[PCI_SECONDARY_BUS]
instead of a wrapper that tests sec->parent_dev.

> -                bus_num <= bus->parent_dev->config[PCI_SUBORDINATE_BUS]) ) {
> +                bus_num <= sec->parent_dev->config[PCI_SUBORDINATE_BUS]) ) {


>              ret = pci_find_bus(sec, bus_num);
>              if (ret) {

I think that in the above, we can just as well do
return pci_find_bus() - if multiple children claim
the same bus range, on real hardware only one
of them will claim the transaction.

>                  return ret;

I find the use of recursion here confusing.
Since as pointed out above this can be a tail recursion,
it can easily be converted to a loop.
How does the below look, for example?
Note: completely untested, likely broken:

diff --git a/hw/pci.c b/hw/pci.c
index b6abd67..e9d6def 100644
--- a/hw/pci.c
+++ b/hw/pci.c
@@ -1546,26 +1546,33 @@ static void pci_bridge_write_config(PCIDevice *d,
 
 PCIBus *pci_find_bus(PCIBus *bus, int bus_num)
 {
-    PCIBus *sec, *ret;
+    PCIBus *sec;
+    bool found;
 
-    if (!bus)
+    if (!bus) {
         return NULL;
+    }
 
     if (pci_bus_num(bus) == bus_num) {
         return bus;
     }
 
     /* try child bus */
-    QLIST_FOREACH(sec, &bus->child, sibling) {
-        if (!bus->parent_dev /* pci host bridge */
-            || (pci_bus_num(sec) <= bus_num &&
-                bus_num <= bus->parent_dev->config[PCI_SUBORDINATE_BUS]) ) {
-            ret = pci_find_bus(sec, bus_num);
-            if (ret) {
-                return ret;
+    do {
+        found = false;
+        QLIST_FOREACH(sec, &bus->child, sibling) {
+            assert(sec->parent_dev);
+            if (sec->parent_dev->config[PCI_SECONDARY_BUS] == bus_num) {
+                return sec;
+            }
+            if (sec->parent_dev->config[PCI_SECONDARY_BUS] < bus_num &&
+                bus_num <= sec->parent_dev->config[PCI_SUBORDINATE_BUS]) {
+                bus = sec;
+                found = true;
+                break;
             }
         }
-    }
+    } while (found);
 
     return NULL;
 }

> -- 
> 1.6.6.1
> 




reply via email to

[Prev in Thread] Current Thread [Next in Thread]