Message ID | 20120510083717.45992.7495.stgit@amd-6168-8-1.englab.nay.redhat.com |
---|---|
State | New |
Headers | show |
On Thu, May 10, 2012 at 04:37:22PM +0800, Jason Wang wrote: > This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because > the linux 8139cp driver would leave the card in "Config Register Write Enable" > mode after the eeprom were read or write ( which is unexpected in the spec > ). Could you show where this happens please? > Also a physical 8139 card can still DMA into host memory in modes other than > Normal mode, so we need revert this commit to align with the behavior of > physical card. > > The issue of 8139cp driver should be fixed in linux seperately. > > Signed-off-by: Jason Wang <jasowang@redhat.com> It's admittedly a bit of a hack but I thought the point is to work with unmodified drivers? What do windows drivers do? Can you check pls? > --- > hw/rtl8139.c | 9 --------- > 1 files changed, 0 insertions(+), 9 deletions(-) > > diff --git a/hw/rtl8139.c b/hw/rtl8139.c > index eb22d04..2413bc3 100644 > --- a/hw/rtl8139.c > +++ b/hw/rtl8139.c > @@ -791,9 +791,6 @@ static int rtl8139_can_receive(VLANClientState *nc) > return 1; > if (!rtl8139_receiver_enabled(s)) > return 1; > - /* network/host communication happens only in normal mode */ > - if ((s->Cfg9346 & Chip9346_op_mask) != Cfg9346_Normal) > - return 0; > > if (rtl8139_cp_receiver_enabled(s)) { > /* ??? Flow control not implemented in c+ mode. > @@ -836,12 +833,6 @@ static ssize_t rtl8139_do_receive(VLANClientState *nc, const uint8_t *buf, size_ > return -1; > } > > - /* check whether we are in normal mode */ > - if ((s->Cfg9346 & Chip9346_op_mask) != Cfg9346_Normal) { > - DPRINTF("not in normal op mode\n"); > - return -1; > - } > - > /* XXX: check this */ > if (s->RxConfig & AcceptAllPhys) { > /* promiscuous: receive all */
On 05/10/2012 04:54 PM, Michael S. Tsirkin wrote: > On Thu, May 10, 2012 at 04:37:22PM +0800, Jason Wang wrote: >> This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because >> the linux 8139cp driver would leave the card in "Config Register Write Enable" >> mode after the eeprom were read or write ( which is unexpected in the spec >> ). > Could you show where this happens please? Hi Michael: According to the spec: """ Normal(0x00): RTL8139C(L)+ network/host communication mode. ... Config. Register Write Enable(0x11): Before writing to the CONFIG0, 1, 3, 4 registers, and bits 13, 12, 8 of BMCR (offset 62h-63h), the RTL8139C(L)+ must be placed in this mode. This will protect the RTL8139C(L)+ configuration from accidental change. """ So If I am reading it correctly, guest should place the card in "Normal mode" during transmission and reception. But linux driver would reset the mode to 11 after each read or write to the eeprom, see eeprom_cmd_end() in 8139cp.c. >> Also a physical 8139 card can still DMA into host memory in modes other than >> Normal mode, so we need revert this commit to align with the behavior of >> physical card. >> >> The issue of 8139cp driver should be fixed in linux seperately. >> >> Signed-off-by: Jason Wang<jasowang@redhat.com> > It's admittedly a bit of a hack but I thought the point is > to work with unmodified drivers? Yes, and as the physical 8139 card would still doing transmission and reception when it's not in normal mode, so we need revert this patch to let unmodified driver work. > What do windows drivers do? Can you check pls? Windows guest would let the card in normal mode after eeprom access. > >> --- >> hw/rtl8139.c | 9 --------- >> 1 files changed, 0 insertions(+), 9 deletions(-) >> >> diff --git a/hw/rtl8139.c b/hw/rtl8139.c >> index eb22d04..2413bc3 100644 >> --- a/hw/rtl8139.c >> +++ b/hw/rtl8139.c >> @@ -791,9 +791,6 @@ static int rtl8139_can_receive(VLANClientState *nc) >> return 1; >> if (!rtl8139_receiver_enabled(s)) >> return 1; >> - /* network/host communication happens only in normal mode */ >> - if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) >> - return 0; >> >> if (rtl8139_cp_receiver_enabled(s)) { >> /* ??? Flow control not implemented in c+ mode. >> @@ -836,12 +833,6 @@ static ssize_t rtl8139_do_receive(VLANClientState *nc, const uint8_t *buf, size_ >> return -1; >> } >> >> - /* check whether we are in normal mode */ >> - if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) { >> - DPRINTF("not in normal op mode\n"); >> - return -1; >> - } >> - >> /* XXX: check this */ >> if (s->RxConfig& AcceptAllPhys) { >> /* promiscuous: receive all */
On Thu, May 10, 2012 at 05:31:56PM +0800, Jason Wang wrote: > On 05/10/2012 04:54 PM, Michael S. Tsirkin wrote: > >On Thu, May 10, 2012 at 04:37:22PM +0800, Jason Wang wrote: > >>This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because > >>the linux 8139cp driver would leave the card in "Config Register Write Enable" > >>mode after the eeprom were read or write ( which is unexpected in the spec > >>). > >Could you show where this happens please? > > Hi Michael: > > According to the spec: > """ > Normal(0x00): RTL8139C(L)+ network/host communication mode. > ... > Config. Register Write Enable(0x11): Before writing to the CONFIG0, > 1, 3, 4 registers, and bits 13, 12, 8 of BMCR (offset 62h-63h), the > RTL8139C(L)+ must be placed in this mode. This will protect the > RTL8139C(L)+ configuration from accidental change. > """ > So If I am reading it correctly, guest should place the card in > "Normal mode" during transmission and reception. But linux driver > would reset the mode to 11 after each read or write to the eeprom, > see eeprom_cmd_end() in 8139cp.c. Which version? The one I see upstream just clears chip select ... > >>Also a physical 8139 card can still DMA into host memory in modes other than > >>Normal mode, so we need revert this commit to align with the behavior of > >>physical card. > >> > >>The issue of 8139cp driver should be fixed in linux seperately. > >> > >>Signed-off-by: Jason Wang<jasowang@redhat.com> > >It's admittedly a bit of a hack but I thought the point is > >to work with unmodified drivers? > > Yes, and as the physical 8139 card would still doing transmission > and reception when it's not in normal mode, so we need revert this > patch to let unmodified driver work. But it *won't* work - the reason we applied this hack is because guest memory got corrupted because of the guest bug. eeprom writes are rare enough so this seems like the lesser evil. > >What do windows drivers do? Can you check pls? > > Windows guest would let the card in normal mode after eeprom access. Does it have the rx ring programming bug too? > > > >>--- > >> hw/rtl8139.c | 9 --------- > >> 1 files changed, 0 insertions(+), 9 deletions(-) > >> > >>diff --git a/hw/rtl8139.c b/hw/rtl8139.c > >>index eb22d04..2413bc3 100644 > >>--- a/hw/rtl8139.c > >>+++ b/hw/rtl8139.c > >>@@ -791,9 +791,6 @@ static int rtl8139_can_receive(VLANClientState *nc) > >> return 1; > >> if (!rtl8139_receiver_enabled(s)) > >> return 1; > >>- /* network/host communication happens only in normal mode */ > >>- if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) > >>- return 0; > >> > >> if (rtl8139_cp_receiver_enabled(s)) { > >> /* ??? Flow control not implemented in c+ mode. > >>@@ -836,12 +833,6 @@ static ssize_t rtl8139_do_receive(VLANClientState *nc, const uint8_t *buf, size_ > >> return -1; > >> } > >> > >>- /* check whether we are in normal mode */ > >>- if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) { > >>- DPRINTF("not in normal op mode\n"); > >>- return -1; > >>- } > >>- > >> /* XXX: check this */ > >> if (s->RxConfig& AcceptAllPhys) { > >> /* promiscuous: receive all */
On 05/10/2012 07:05 PM, Michael S. Tsirkin wrote: > On Thu, May 10, 2012 at 05:31:56PM +0800, Jason Wang wrote: >> On 05/10/2012 04:54 PM, Michael S. Tsirkin wrote: >>> On Thu, May 10, 2012 at 04:37:22PM +0800, Jason Wang wrote: >>>> This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because >>>> the linux 8139cp driver would leave the card in "Config Register Write Enable" >>>> mode after the eeprom were read or write ( which is unexpected in the spec >>>> ). >>> Could you show where this happens please? >> Hi Michael: >> >> According to the spec: >> """ >> Normal(0x00): RTL8139C(L)+ network/host communication mode. >> ... >> Config. Register Write Enable(0x11): Before writing to the CONFIG0, >> 1, 3, 4 registers, and bits 13, 12, 8 of BMCR (offset 62h-63h), the >> RTL8139C(L)+ must be placed in this mode. This will protect the >> RTL8139C(L)+ configuration from accidental change. >> """ >> So If I am reading it correctly, guest should place the card in >> "Normal mode" during transmission and reception. But linux driver >> would reset the mode to 11 after each read or write to the eeprom, >> see eeprom_cmd_end() in 8139cp.c. > Which version? The one I see upstream just clears chip select ... It use writeb(), so in the meantime it also set the high two bits of 9346. >>>> Also a physical 8139 card can still DMA into host memory in modes other than >>>> Normal mode, so we need revert this commit to align with the behavior of >>>> physical card. >>>> >>>> The issue of 8139cp driver should be fixed in linux seperately. >>>> >>>> Signed-off-by: Jason Wang<jasowang@redhat.com> >>> It's admittedly a bit of a hack but I thought the point is >>> to work with unmodified drivers? >> Yes, and as the physical 8139 card would still doing transmission >> and reception when it's not in normal mode, so we need revert this >> patch to let unmodified driver work. > But it *won't* work - the reason we applied this hack is > because guest memory got corrupted because of > the guest bug. eeprom writes are rare enough so this > seems like the lesser evil. > >>> What do windows drivers do? Can you check pls? >> Windows guest would let the card in normal mode after eeprom access. > Does it have the rx ring programming bug too? It does not have such a bug according to debug log. >>>> --- >>>> hw/rtl8139.c | 9 --------- >>>> 1 files changed, 0 insertions(+), 9 deletions(-) >>>> >>>> diff --git a/hw/rtl8139.c b/hw/rtl8139.c >>>> index eb22d04..2413bc3 100644 >>>> --- a/hw/rtl8139.c >>>> +++ b/hw/rtl8139.c >>>> @@ -791,9 +791,6 @@ static int rtl8139_can_receive(VLANClientState *nc) >>>> return 1; >>>> if (!rtl8139_receiver_enabled(s)) >>>> return 1; >>>> - /* network/host communication happens only in normal mode */ >>>> - if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) >>>> - return 0; >>>> >>>> if (rtl8139_cp_receiver_enabled(s)) { >>>> /* ??? Flow control not implemented in c+ mode. >>>> @@ -836,12 +833,6 @@ static ssize_t rtl8139_do_receive(VLANClientState *nc, const uint8_t *buf, size_ >>>> return -1; >>>> } >>>> >>>> - /* check whether we are in normal mode */ >>>> - if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) { >>>> - DPRINTF("not in normal op mode\n"); >>>> - return -1; >>>> - } >>>> - >>>> /* XXX: check this */ >>>> if (s->RxConfig& AcceptAllPhys) { >>>> /* promiscuous: receive all */
On Mon, May 14, 2012 at 01:31:22PM +0800, Jason Wang wrote: > On 05/10/2012 07:05 PM, Michael S. Tsirkin wrote: > >On Thu, May 10, 2012 at 05:31:56PM +0800, Jason Wang wrote: > >>On 05/10/2012 04:54 PM, Michael S. Tsirkin wrote: > >>>On Thu, May 10, 2012 at 04:37:22PM +0800, Jason Wang wrote: > >>>>This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because > >>>>the linux 8139cp driver would leave the card in "Config Register Write Enable" > >>>>mode after the eeprom were read or write ( which is unexpected in the spec > >>>>). > >>>Could you show where this happens please? > >>Hi Michael: > >> > >>According to the spec: > >>""" > >>Normal(0x00): RTL8139C(L)+ network/host communication mode. > >>... > >>Config. Register Write Enable(0x11): Before writing to the CONFIG0, > >>1, 3, 4 registers, and bits 13, 12, 8 of BMCR (offset 62h-63h), the > >>RTL8139C(L)+ must be placed in this mode. This will protect the > >>RTL8139C(L)+ configuration from accidental change. > >>""" > >>So If I am reading it correctly, guest should place the card in > >>"Normal mode" during transmission and reception. But linux driver > >>would reset the mode to 11 after each read or write to the eeprom, > >>see eeprom_cmd_end() in 8139cp.c. > >Which version? The one I see upstream just clears chip select ... > > It use writeb(), so in the meantime it also set the high two bits of 9346. OK, so that's the bug, right? And it's later cleared during normal init, but reading eeprom will set it? > >>>>Also a physical 8139 card can still DMA into host memory in modes other than > >>>>Normal mode, so we need revert this commit to align with the behavior of > >>>>physical card. > >>>> > >>>>The issue of 8139cp driver should be fixed in linux seperately. > >>>> > >>>>Signed-off-by: Jason Wang<jasowang@redhat.com> > >>>It's admittedly a bit of a hack but I thought the point is > >>>to work with unmodified drivers? > >>Yes, and as the physical 8139 card would still doing transmission > >>and reception when it's not in normal mode, so we need revert this > >>patch to let unmodified driver work. > >But it *won't* work - the reason we applied this hack is > >because guest memory got corrupted because of > >the guest bug. eeprom writes are rare enough so this > >seems like the lesser evil. > > > >>>What do windows drivers do? Can you check pls? > >>Windows guest would let the card in normal mode after eeprom access. > >Does it have the rx ring programming bug too? > > It does not have such a bug according to debug log. > >>>>--- > >>>> hw/rtl8139.c | 9 --------- > >>>> 1 files changed, 0 insertions(+), 9 deletions(-) > >>>> > >>>>diff --git a/hw/rtl8139.c b/hw/rtl8139.c > >>>>index eb22d04..2413bc3 100644 > >>>>--- a/hw/rtl8139.c > >>>>+++ b/hw/rtl8139.c > >>>>@@ -791,9 +791,6 @@ static int rtl8139_can_receive(VLANClientState *nc) > >>>> return 1; > >>>> if (!rtl8139_receiver_enabled(s)) > >>>> return 1; > >>>>- /* network/host communication happens only in normal mode */ > >>>>- if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) > >>>>- return 0; > >>>> > >>>> if (rtl8139_cp_receiver_enabled(s)) { > >>>> /* ??? Flow control not implemented in c+ mode. > >>>>@@ -836,12 +833,6 @@ static ssize_t rtl8139_do_receive(VLANClientState *nc, const uint8_t *buf, size_ > >>>> return -1; > >>>> } > >>>> > >>>>- /* check whether we are in normal mode */ > >>>>- if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) { > >>>>- DPRINTF("not in normal op mode\n"); > >>>>- return -1; > >>>>- } > >>>>- > >>>> /* XXX: check this */ > >>>> if (s->RxConfig& AcceptAllPhys) { > >>>> /* promiscuous: receive all */
On 05/14/2012 02:20 PM, Michael S. Tsirkin wrote: > On Mon, May 14, 2012 at 01:31:22PM +0800, Jason Wang wrote: >> > On 05/10/2012 07:05 PM, Michael S. Tsirkin wrote: >>> > >On Thu, May 10, 2012 at 05:31:56PM +0800, Jason Wang wrote: >>>> > >>On 05/10/2012 04:54 PM, Michael S. Tsirkin wrote: >>>>> > >>>On Thu, May 10, 2012 at 04:37:22PM +0800, Jason Wang wrote: >>>>>> > >>>>This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because >>>>>> > >>>>the linux 8139cp driver would leave the card in "Config Register Write Enable" >>>>>> > >>>>mode after the eeprom were read or write ( which is unexpected in the spec >>>>>> > >>>>). >>>>> > >>>Could you show where this happens please? >>>> > >>Hi Michael: >>>> > >> >>>> > >>According to the spec: >>>> > >>""" >>>> > >>Normal(0x00): RTL8139C(L)+ network/host communication mode. >>>> > >>... >>>> > >>Config. Register Write Enable(0x11): Before writing to the CONFIG0, >>>> > >>1, 3, 4 registers, and bits 13, 12, 8 of BMCR (offset 62h-63h), the >>>> > >>RTL8139C(L)+ must be placed in this mode. This will protect the >>>> > >>RTL8139C(L)+ configuration from accidental change. >>>> > >>""" >>>> > >>So If I am reading it correctly, guest should place the card in >>>> > >>"Normal mode" during transmission and reception. But linux driver >>>> > >>would reset the mode to 11 after each read or write to the eeprom, >>>> > >>see eeprom_cmd_end() in 8139cp.c. >>> > >Which version? The one I see upstream just clears chip select ... >> > >> > It use writeb(), so in the meantime it also set the high two bits of 9346. > OK, so that's the bug, right? > And it's later cleared during normal init, but reading > eeprom will set it? > Right, that's exactly what I mean.
On 05/10/2012 04:37 PM, Jason Wang wrote: > This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because > the linux 8139cp driver would leave the card in "Config Register Write Enable" > mode after the eeprom were read or write ( which is unexpected in the spec > ). Also a physical 8139 card can still DMA into host memory in modes other than > Normal mode, so we need revert this commit to align with the behavior of > physical card. > > The issue of 8139cp driver should be fixed in linux seperately. > > Signed-off-by: Jason Wang<jasowang@redhat.com> Applied. Thanks. Regards, Anthony Liguori > --- > hw/rtl8139.c | 9 --------- > 1 files changed, 0 insertions(+), 9 deletions(-) > > diff --git a/hw/rtl8139.c b/hw/rtl8139.c > index eb22d04..2413bc3 100644 > --- a/hw/rtl8139.c > +++ b/hw/rtl8139.c > @@ -791,9 +791,6 @@ static int rtl8139_can_receive(VLANClientState *nc) > return 1; > if (!rtl8139_receiver_enabled(s)) > return 1; > - /* network/host communication happens only in normal mode */ > - if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) > - return 0; > > if (rtl8139_cp_receiver_enabled(s)) { > /* ??? Flow control not implemented in c+ mode. > @@ -836,12 +833,6 @@ static ssize_t rtl8139_do_receive(VLANClientState *nc, const uint8_t *buf, size_ > return -1; > } > > - /* check whether we are in normal mode */ > - if ((s->Cfg9346& Chip9346_op_mask) != Cfg9346_Normal) { > - DPRINTF("not in normal op mode\n"); > - return -1; > - } > - > /* XXX: check this */ > if (s->RxConfig& AcceptAllPhys) { > /* promiscuous: receive all */ > > >
On 06/04/2012 01:48 PM, Anthony Liguori wrote: > On 05/10/2012 04:37 PM, Jason Wang wrote: >> This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is >> because >> the linux 8139cp driver would leave the card in "Config Register >> Write Enable" >> mode after the eeprom were read or write ( which is unexpected in the >> spec >> ). Also a physical 8139 card can still DMA into host memory in modes >> other than >> Normal mode, so we need revert this commit to align with the behavior of >> physical card. >> >> The issue of 8139cp driver should be fixed in linux seperately. >> >> Signed-off-by: Jason Wang<jasowang@redhat.com> > > Applied. Thanks. > > Regards, > > Anthony Liguori Hi Anthony: Michael has some concern on this patch, so I post another patch of validating the rx buf instead of checking the opmode, please see http://lists.gnu.org/archive/html/qemu-devel/2012-05/msg02385.html. So maybe we can apply that or I need to send another patch on top of this reverting? Thanks
On 06/04/2012 01:58 PM, Jason Wang wrote: > On 06/04/2012 01:48 PM, Anthony Liguori wrote: >> On 05/10/2012 04:37 PM, Jason Wang wrote: >>> This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because >>> the linux 8139cp driver would leave the card in "Config Register Write Enable" >>> mode after the eeprom were read or write ( which is unexpected in the spec >>> ). Also a physical 8139 card can still DMA into host memory in modes other than >>> Normal mode, so we need revert this commit to align with the behavior of >>> physical card. >>> >>> The issue of 8139cp driver should be fixed in linux seperately. >>> >>> Signed-off-by: Jason Wang<jasowang@redhat.com> >> >> Applied. Thanks. >> >> Regards, >> >> Anthony Liguori > > Hi Anthony: > > Michael has some concern on this patch, so I post another patch of validating > the rx buf instead of checking the opmode, please see > http://lists.gnu.org/archive/html/qemu-devel/2012-05/msg02385.html. So maybe we > can apply that or I need to send another patch on top of this reverting? I'll work it all out. Regards, Anthony Liguori > > Thanks >
On 06/04/2012 03:28 PM, Anthony Liguori wrote: >> Michael has some concern on this patch, so I post another patch of >> validating >> the rx buf instead of checking the opmode, please see >> http://lists.gnu.org/archive/html/qemu-devel/2012-05/msg02385.html. So >> maybe we >> can apply that or I need to send another patch on top of this reverting? > > I'll work it all out. I see the revert reverted, but I don't see the followup patch applied? A laboriously bisected regression here suggests this is the problem.
diff --git a/hw/rtl8139.c b/hw/rtl8139.c index eb22d04..2413bc3 100644 --- a/hw/rtl8139.c +++ b/hw/rtl8139.c @@ -791,9 +791,6 @@ static int rtl8139_can_receive(VLANClientState *nc) return 1; if (!rtl8139_receiver_enabled(s)) return 1; - /* network/host communication happens only in normal mode */ - if ((s->Cfg9346 & Chip9346_op_mask) != Cfg9346_Normal) - return 0; if (rtl8139_cp_receiver_enabled(s)) { /* ??? Flow control not implemented in c+ mode. @@ -836,12 +833,6 @@ static ssize_t rtl8139_do_receive(VLANClientState *nc, const uint8_t *buf, size_ return -1; } - /* check whether we are in normal mode */ - if ((s->Cfg9346 & Chip9346_op_mask) != Cfg9346_Normal) { - DPRINTF("not in normal op mode\n"); - return -1; - } - /* XXX: check this */ if (s->RxConfig & AcceptAllPhys) { /* promiscuous: receive all */
This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because the linux 8139cp driver would leave the card in "Config Register Write Enable" mode after the eeprom were read or write ( which is unexpected in the spec ). Also a physical 8139 card can still DMA into host memory in modes other than Normal mode, so we need revert this commit to align with the behavior of physical card. The issue of 8139cp driver should be fixed in linux seperately. Signed-off-by: Jason Wang <jasowang@redhat.com> --- hw/rtl8139.c | 9 --------- 1 files changed, 0 insertions(+), 9 deletions(-)