[Qemu-devel] [PATCH] e1000: pre-initialize RAH/RAL registers

[Qemu-devel] [PATCH] e1000: pre-initialize RAH/RAL registers

[Gabriel L. Somlo]

Some guest operating systems' drivers (Mac OS X in particular) fail to
properly initialize the Receive Address registers (probably expecting
them to be pre-initialized by an earlier component, such as a proprietary
BIOS). This patch pre-initializes the RA registers, allowing OS X
networking to function properly. Other guest operating systems are not
affected, and free to (re)initialize these registers during boot.

Signed-off-by: Gabriel Somlo <somlo@cmu.edu>
---
 hw/e1000.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/hw/e1000.c b/hw/e1000.c
index e4f1ffe..6478ff3 100644
--- a/hw/e1000.c
+++ b/hw/e1000.c
@@ -278,6 +278,10 @@ static void e1000_reset(void *opaque)
     if (d->nic->nc.link_down) {
         e1000_link_down(d);
     }
+
+  /* Some guests expect pre-initialized RAH/RAL (AddrValid flag + MACaddr) */
+  d->mac_reg[RA+1] = E1000_RAH_AV;
+  memmove(&d->mac_reg[RA], &d->conf.macaddr, sizeof(struct MACAddr));
 }
 
 static void
-- 
1.7.7.6

Re: [Qemu-devel] [PATCH] e1000: pre-initialize RAH/RAL registers

[Stefan Hajnoczi]

On Tue, Oct 30, 2012 at 01:20:40PM -0400, Gabriel L. Somlo wrote:
> Some guest operating systems' drivers (Mac OS X in particular) fail to
> properly initialize the Receive Address registers (probably expecting
> them to be pre-initialized by an earlier component, such as a proprietary
> BIOS). This patch pre-initializes the RA registers, allowing OS X
> networking to function properly. Other guest operating systems are not
> affected, and free to (re)initialize these registers during boot.
> 
> Signed-off-by: Gabriel Somlo <somlo@cmu.edu>
> ---
>  hw/e1000.c |    4 ++++
>  1 files changed, 4 insertions(+), 0 deletions(-)
> 
> diff --git a/hw/e1000.c b/hw/e1000.c
> index e4f1ffe..6478ff3 100644
> --- a/hw/e1000.c
> +++ b/hw/e1000.c
> @@ -278,6 +278,10 @@ static void e1000_reset(void *opaque)
>      if (d->nic->nc.link_down) {
>          e1000_link_down(d);
>      }
> +
> +  /* Some guests expect pre-initialized RAH/RAL (AddrValid flag + MACaddr) */
> +  d->mac_reg[RA+1] = E1000_RAH_AV;

Please use 4 space indentation (QEMU coding style).

> +  memmove(&d->mac_reg[RA], &d->conf.macaddr, sizeof(struct MACAddr));

When the host is big-endian the filter code will byteswap and the MAC address
will not match:

  for (rp = s->mac_reg + RA; rp < s->mac_reg + RA + 32; rp += 2) {
      if (!(rp[1] & E1000_RAH_AV))
          continue;
      ra[0] = cpu_to_le32(rp[0]);
      ra[1] = cpu_to_le32(rp[1]);
      if (!memcmp(buf, (uint8_t *)ra, 6)) {

Stefan

[Qemu-devel] [PATCH v2] e1000: pre-initialize RAH/RAL registers

[Gabriel L. Somlo]

Some guest operating systems' drivers (Mac OS X in particular) fail to
properly initialize the Receive Address registers (probably expecting
them to be pre-initialized by an earlier component, such as a specific
proprietary BIOS). This patch pre-initializes the RA registers, allowing
OS X networking to function properly. Other guest operating systems are
not affected, and free to (re)initialize these registers during boot.

Signed-off-by: Gabriel Somlo <somlo@cmu.edu>
---

On Wed, Oct 31, 2012 at 09:03:51AM +0100, Stefan Hajnoczi wrote:
> Please use 4 space indentation (QEMU coding style).

My bad, fixed in this version.

> > +  memmove(&d->mac_reg[RA], &d->conf.macaddr, sizeof(struct MACAddr));
>
> When the host is big-endian the filter code will byteswap and the MAC address
> will not match:
> 
>   for (rp = s->mac_reg + RA; rp < s->mac_reg + RA + 32; rp += 2) {
>       if (!(rp[1] & E1000_RAH_AV))
>           continue;
>       ra[0] = cpu_to_le32(rp[0]);
>       ra[1] = cpu_to_le32(rp[1]);
>       if (!memcmp(buf, (uint8_t *)ra, 6)) {

OK, I guess I could just memmove from d->eeprom_data instead of
d->conf.macaddr.a, on the assumption that pci_e1000_init() already
did the endian-proofing for me. But probably the safest/most paranoid
method is best, see below...

Thanks,
  Gabriel

 hw/e1000.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)

diff --git a/hw/e1000.c b/hw/e1000.c
index e4f1ffe..a29b844 100644
--- a/hw/e1000.c
+++ b/hw/e1000.c
@@ -266,6 +266,8 @@ rxbufsize(uint32_t v)
 static void e1000_reset(void *opaque)
 {
     E1000State *d = opaque;
+    uint8_t *macaddr = d->conf.macaddr.a;
+    int i;
 
     qemu_del_timer(d->autoneg_timer);
     memset(d->phy_reg, 0, sizeof d->phy_reg);
@@ -278,6 +280,14 @@ static void e1000_reset(void *opaque)
     if (d->nic->nc.link_down) {
         e1000_link_down(d);
     }
+
+    /* Some guests expect pre-initialized RAH/RAL (AddrValid flag + MACaddr) */
+    d->mac_reg[RA] = 0;
+    d->mac_reg[RA+1] = E1000_RAH_AV;
+    for (i = 0; i < 4; i++) {
+        d->mac_reg[RA] |= macaddr[i]<<(8*i);
+        d->mac_reg[RA+1] |= (i < 2) ? macaddr[i+4]<<(8*i) : 0;
+    }
 }
 
 static void
-- 
1.7.7.6

Re: [Qemu-devel] [PATCH v2] e1000: pre-initialize RAH/RAL registers

[Stefan Hajnoczi]

On Wed, Oct 31, 2012 at 02:15:39PM -0400, Gabriel L. Somlo wrote:
> Some guest operating systems' drivers (Mac OS X in particular) fail to
> properly initialize the Receive Address registers (probably expecting
> them to be pre-initialized by an earlier component, such as a specific
> proprietary BIOS). This patch pre-initializes the RA registers, allowing
> OS X networking to function properly. Other guest operating systems are
> not affected, and free to (re)initialize these registers during boot.
> 
> Signed-off-by: Gabriel Somlo <somlo@cmu.edu>
> ---
> 
> On Wed, Oct 31, 2012 at 09:03:51AM +0100, Stefan Hajnoczi wrote:
> > Please use 4 space indentation (QEMU coding style).
> 
> My bad, fixed in this version.
> 
> > > +  memmove(&d->mac_reg[RA], &d->conf.macaddr, sizeof(struct MACAddr));
> >
> > When the host is big-endian the filter code will byteswap and the MAC address
> > will not match:
> > 
> >   for (rp = s->mac_reg + RA; rp < s->mac_reg + RA + 32; rp += 2) {
> >       if (!(rp[1] & E1000_RAH_AV))
> >           continue;
> >       ra[0] = cpu_to_le32(rp[0]);
> >       ra[1] = cpu_to_le32(rp[1]);
> >       if (!memcmp(buf, (uint8_t *)ra, 6)) {
> 
> OK, I guess I could just memmove from d->eeprom_data instead of
> d->conf.macaddr.a, on the assumption that pci_e1000_init() already
> did the endian-proofing for me. But probably the safest/most paranoid
> method is best, see below...
> 
> Thanks,
>   Gabriel
> 
>  hw/e1000.c |   10 ++++++++++
>  1 files changed, 10 insertions(+), 0 deletions(-)

Although real hardware doesn't behave like this we can probably get away
with it.  I will test Linux and Windows guests with e1000 for the QEMU
1.3 release - if we run into problems this patch may be reverted.

Thanks, applied to the net tree:
https://github.com/stefanha/qemu/commits/net

Stefan