[Qemu-devel] [PATCH] pci: remove explicit check to 64K ioport size

[Qemu-devel] [PATCH] pci: remove explicit check to 64K ioport size

[Hervé Poussineau]

This check is useless, as bigger addresses will be ignored when
added to 'io' MemoryRegion, which has a size of 64K.

However, some architectures don't use the 'io' MemoryRegion, like
the alpha and versatile platforms. They create a PCI I/O region
bigger than 64K, so let them handle PCI I/O BARs in the higher range.

Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
---
 hw/pci/pci.c |    3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/hw/pci/pci.c b/hw/pci/pci.c
index d00682e..a8e2b29 100644
--- a/hw/pci/pci.c
+++ b/hw/pci/pci.c
@@ -1028,8 +1028,7 @@ static pcibus_t pci_bar_address(PCIDevice *d,
         }
         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
         last_addr = new_addr + size - 1;
-        /* NOTE: we have only 64K ioports on PC */
-        if (last_addr <= new_addr || new_addr == 0 || last_addr > UINT16_MAX) {
+        if (last_addr <= new_addr || new_addr == 0) {
             return PCI_BAR_UNMAPPED;
         }
         return new_addr;
-- 
1.7.10.4

Re: [Qemu-devel] [PATCH] pci: remove explicit check to 64K ioport size

[Richard Henderson]

On 09/13/2013 04:58 AM, Hervé Poussineau wrote:
> This check is useless, as bigger addresses will be ignored when
> added to 'io' MemoryRegion, which has a size of 64K.
> 
> However, some architectures don't use the 'io' MemoryRegion, like
> the alpha and versatile platforms. They create a PCI I/O region
> bigger than 64K, so let them handle PCI I/O BARs in the higher range.
> 
> Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
> ---
>  hw/pci/pci.c |    3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)

Reviewed-by: Richard Henderson <rth@twiddle.net>


r~

Re: [Qemu-devel] [PATCH] pci: remove explicit check to 64K ioport size

[Michael S. Tsirkin]

On Fri, Sep 13, 2013 at 01:58:44PM +0200, Hervé Poussineau wrote:
> This check is useless, as bigger addresses will be ignored when
> added to 'io' MemoryRegion, which has a size of 64K.
> 
> However, some architectures don't use the 'io' MemoryRegion, like
> the alpha and versatile platforms. They create a PCI I/O region
> bigger than 64K, so let them handle PCI I/O BARs in the higher range.
> 
> Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
> ---
>  hw/pci/pci.c |    3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/hw/pci/pci.c b/hw/pci/pci.c
> index d00682e..a8e2b29 100644
> --- a/hw/pci/pci.c
> +++ b/hw/pci/pci.c
> @@ -1028,8 +1028,7 @@ static pcibus_t pci_bar_address(PCIDevice *d,
>          }
>          new_addr = pci_get_long(d->config + bar) & ~(size - 1);
>          last_addr = new_addr + size - 1;
> -        /* NOTE: we have only 64K ioports on PC */
> -        if (last_addr <= new_addr || new_addr == 0 || last_addr > UINT16_MAX) {
> +        if (last_addr <= new_addr || new_addr == 0) {
>              return PCI_BAR_UNMAPPED;
>          }
>          return new_addr;
> -- 
> 1.7.10.4

We still didn't fix the overlap for PCI BARs properly
so it looks like the following is needed on top.
Could you please check whether this works for you on versatile/alpha?

diff --git a/hw/pci/pci.c b/hw/pci/pci.c
index d452135..3118854 100644
--- a/hw/pci/pci.c
+++ b/hw/pci/pci.c
@@ -1028,7 +1028,10 @@ static pcibus_t pci_bar_address(PCIDevice *d,
         }
         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
         last_addr = new_addr + size - 1;
-        if (last_addr <= new_addr || new_addr == 0) {
+        /* Check if BAR is being sized explicitly.
+         * TODO: make priorities correct and remove this work around.
+         */
+        if (last_addr <= new_addr || new_addr == 0 || last_addr >= UINT32_MAX) {
             return PCI_BAR_UNMAPPED;
         }
         return new_addr;

Re: [Qemu-devel] [PATCH] pci: remove explicit check to 64K ioport size

[Hervé Poussineau]

Michael S. Tsirkin a écrit :
> On Fri, Sep 13, 2013 at 01:58:44PM +0200, Hervé Poussineau wrote:
>> This check is useless, as bigger addresses will be ignored when
>> added to 'io' MemoryRegion, which has a size of 64K.
>>
>> However, some architectures don't use the 'io' MemoryRegion, like
>> the alpha and versatile platforms. They create a PCI I/O region
>> bigger than 64K, so let them handle PCI I/O BARs in the higher range.
>>
>> Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
>> ---
>>  hw/pci/pci.c |    3 +--
>>  1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/hw/pci/pci.c b/hw/pci/pci.c
>> index d00682e..a8e2b29 100644
>> --- a/hw/pci/pci.c
>> +++ b/hw/pci/pci.c
>> @@ -1028,8 +1028,7 @@ static pcibus_t pci_bar_address(PCIDevice *d,
>>          }
>>          new_addr = pci_get_long(d->config + bar) & ~(size - 1);
>>          last_addr = new_addr + size - 1;
>> -        /* NOTE: we have only 64K ioports on PC */
>> -        if (last_addr <= new_addr || new_addr == 0 || last_addr > UINT16_MAX) {
>> +        if (last_addr <= new_addr || new_addr == 0) {
>>              return PCI_BAR_UNMAPPED;
>>          }
>>          return new_addr;
>> -- 
>> 1.7.10.4
> 
> We still didn't fix the overlap for PCI BARs properly
> so it looks like the following is needed on top.
> Could you please check whether this works for you on versatile/alpha?

Yes, alpha and versatile still seem to work.

Hervé

> 
> diff --git a/hw/pci/pci.c b/hw/pci/pci.c
> index d452135..3118854 100644
> --- a/hw/pci/pci.c
> +++ b/hw/pci/pci.c
> @@ -1028,7 +1028,10 @@ static pcibus_t pci_bar_address(PCIDevice *d,
>          }
>          new_addr = pci_get_long(d->config + bar) & ~(size - 1);
>          last_addr = new_addr + size - 1;
> -        if (last_addr <= new_addr || new_addr == 0) {
> +        /* Check if BAR is being sized explicitly.
> +         * TODO: make priorities correct and remove this work around.
> +         */
> +        if (last_addr <= new_addr || new_addr == 0 || last_addr >= UINT32_MAX) {
>              return PCI_BAR_UNMAPPED;
>          }
>          return new_addr;
>