Message ID | 20171106203520.7880-2-michael.nawrocki@gtri.gatech.edu |
---|---|
State | New |
Headers | show |
Series | Fix eepro100 simple transmission, add i82559 PCI ID | expand |
On 06.11.2017 21:35, Mike Nawrocki wrote: > The simple transmission mode was treating the area immediately after the > transmit command block (TCB) as if it were a transmit buffer descriptor, > when in reality it is simply the packet data. This change simply copies > the data following the TCB into the packet buffer. > > Signed-off-by: Mike Nawrocki <michael.nawrocki@gtri.gatech.edu> > --- > hw/net/eepro100.c | 18 +++--------------- > 1 file changed, 3 insertions(+), 15 deletions(-) > > diff --git a/hw/net/eepro100.c b/hw/net/eepro100.c > index 80b8f47c4b..91dd058010 100644 > --- a/hw/net/eepro100.c > +++ b/hw/net/eepro100.c > @@ -774,23 +774,11 @@ static void tx_command(EEPRO100State *s) > } > assert(tcb_bytes <= sizeof(buf)); > while (size < tcb_bytes) { > - uint32_t tx_buffer_address = ldl_le_pci_dma(&s->dev, tbd_address); > - uint16_t tx_buffer_size = lduw_le_pci_dma(&s->dev, tbd_address + 4); > -#if 0 > - uint16_t tx_buffer_el = lduw_le_pci_dma(&s->dev, tbd_address + 6); > -#endif > - if (tx_buffer_size == 0) { > - /* Prevent an endless loop. */ > - logout("loop in %s:%u\n", __FILE__, __LINE__); > - break; > - } > - tbd_address += 8; > TRACE(RXTX, logout > ("TBD (simplified mode): buffer address 0x%08x, size 0x%04x\n", > - tx_buffer_address, tx_buffer_size)); > - tx_buffer_size = MIN(tx_buffer_size, sizeof(buf) - size); > - pci_dma_read(&s->dev, tx_buffer_address, &buf[size], tx_buffer_size); > - size += tx_buffer_size; > + tbd_address, tcb_bytes)); > + pci_dma_read(&s->dev, tbd_address, &buf[size], tcb_bytes); > + size += tcb_bytes; > } > if (tbd_array == 0xffffffff) { > /* Simplified mode. Was already handled by code above. */ > Hi, I've got no clue why, but apparently this patch introduced a regression in rc1: The tests/pxe-test is now failing on a big endian (s390x) host when testing the i82550 device. It's working fine with commit 8fa5ad6dfb66b6c695aa6a6b54b7031c (and 2.11-rc0), and it's broken starting with commit 1865e288a823c764cd4344d704e33344 ... do you have any clue what might be wrong here now? Thomas
On 16.11.2017 17:59, Thomas Huth wrote: > On 06.11.2017 21:35, Mike Nawrocki wrote: >> The simple transmission mode was treating the area immediately after the >> transmit command block (TCB) as if it were a transmit buffer descriptor, >> when in reality it is simply the packet data. This change simply copies >> the data following the TCB into the packet buffer. >> >> Signed-off-by: Mike Nawrocki <michael.nawrocki@gtri.gatech.edu> >> --- >> hw/net/eepro100.c | 18 +++--------------- >> 1 file changed, 3 insertions(+), 15 deletions(-) >> >> diff --git a/hw/net/eepro100.c b/hw/net/eepro100.c >> index 80b8f47c4b..91dd058010 100644 >> --- a/hw/net/eepro100.c >> +++ b/hw/net/eepro100.c >> @@ -774,23 +774,11 @@ static void tx_command(EEPRO100State *s) >> } >> assert(tcb_bytes <= sizeof(buf)); >> while (size < tcb_bytes) { >> - uint32_t tx_buffer_address = ldl_le_pci_dma(&s->dev, tbd_address); >> - uint16_t tx_buffer_size = lduw_le_pci_dma(&s->dev, tbd_address + 4); >> -#if 0 >> - uint16_t tx_buffer_el = lduw_le_pci_dma(&s->dev, tbd_address + 6); >> -#endif >> - if (tx_buffer_size == 0) { >> - /* Prevent an endless loop. */ >> - logout("loop in %s:%u\n", __FILE__, __LINE__); >> - break; >> - } >> - tbd_address += 8; >> TRACE(RXTX, logout >> ("TBD (simplified mode): buffer address 0x%08x, size 0x%04x\n", >> - tx_buffer_address, tx_buffer_size)); >> - tx_buffer_size = MIN(tx_buffer_size, sizeof(buf) - size); >> - pci_dma_read(&s->dev, tx_buffer_address, &buf[size], tx_buffer_size); >> - size += tx_buffer_size; >> + tbd_address, tcb_bytes)); >> + pci_dma_read(&s->dev, tbd_address, &buf[size], tcb_bytes); >> + size += tcb_bytes; >> } >> if (tbd_array == 0xffffffff) { >> /* Simplified mode. Was already handled by code above. */ >> > > Hi, > > I've got no clue why, but apparently this patch introduced a regression > in rc1: The tests/pxe-test is now failing on a big endian (s390x) host > when testing the i82550 device. > It's working fine with commit 8fa5ad6dfb66b6c695aa6a6b54b7031c (and > 2.11-rc0), and it's broken starting with commit > 1865e288a823c764cd4344d704e33344 ... do you have any clue what might be > wrong here now? This patch seems to fix the issue for me: diff --git a/hw/net/eepro100.c b/hw/net/eepro100.c index a63ed2c..03e00f7 100644 --- a/hw/net/eepro100.c +++ b/hw/net/eepro100.c @@ -756,8 +756,8 @@ static void read_cb(EEPRO100State *s) static void tx_command(EEPRO100State *s) { - uint32_t tbd_array = le32_to_cpu(s->tx.tbd_array_addr); - uint16_t tcb_bytes = (le16_to_cpu(s->tx.tcb_bytes) & 0x3fff); + uint32_t tbd_array = s->tx.tbd_array_addr; + uint16_t tcb_bytes = s->tx.tcb_bytes & 0x3fff; /* Sends larger than MAX_ETH_FRAME_SIZE are allowed, up to 2600 bytes. */ uint8_t buf[2600]; uint16_t size = 0; Does that make sense? I think the fields are already swapped in read_cb(), so no need to swap them in tx_command() again? Thomas
On 11/16/2017 12:43 PM, Thomas Huth wrote: > On 16.11.2017 17:59, Thomas Huth wrote: >> On 06.11.2017 21:35, Mike Nawrocki wrote: >>> The simple transmission mode was treating the area immediately after the >>> transmit command block (TCB) as if it were a transmit buffer descriptor, >>> when in reality it is simply the packet data. This change simply copies >>> the data following the TCB into the packet buffer. >>> >>> Signed-off-by: Mike Nawrocki <michael.nawrocki@gtri.gatech.edu> >>> --- >>> hw/net/eepro100.c | 18 +++--------------- >>> 1 file changed, 3 insertions(+), 15 deletions(-) >>> >>> diff --git a/hw/net/eepro100.c b/hw/net/eepro100.c >>> index 80b8f47c4b..91dd058010 100644 >>> --- a/hw/net/eepro100.c >>> +++ b/hw/net/eepro100.c >>> @@ -774,23 +774,11 @@ static void tx_command(EEPRO100State *s) >>> } >>> assert(tcb_bytes <= sizeof(buf)); >>> while (size < tcb_bytes) { >>> - uint32_t tx_buffer_address = ldl_le_pci_dma(&s->dev, tbd_address); >>> - uint16_t tx_buffer_size = lduw_le_pci_dma(&s->dev, tbd_address + 4); >>> -#if 0 >>> - uint16_t tx_buffer_el = lduw_le_pci_dma(&s->dev, tbd_address + 6); >>> -#endif >>> - if (tx_buffer_size == 0) { >>> - /* Prevent an endless loop. */ >>> - logout("loop in %s:%u\n", __FILE__, __LINE__); >>> - break; >>> - } >>> - tbd_address += 8; >>> TRACE(RXTX, logout >>> ("TBD (simplified mode): buffer address 0x%08x, size 0x%04x\n", >>> - tx_buffer_address, tx_buffer_size)); >>> - tx_buffer_size = MIN(tx_buffer_size, sizeof(buf) - size); >>> - pci_dma_read(&s->dev, tx_buffer_address, &buf[size], tx_buffer_size); >>> - size += tx_buffer_size; >>> + tbd_address, tcb_bytes)); >>> + pci_dma_read(&s->dev, tbd_address, &buf[size], tcb_bytes); >>> + size += tcb_bytes; >>> } >>> if (tbd_array == 0xffffffff) { >>> /* Simplified mode. Was already handled by code above. */ >>> >> >> Hi, >> >> I've got no clue why, but apparently this patch introduced a regression >> in rc1: The tests/pxe-test is now failing on a big endian (s390x) host >> when testing the i82550 device. >> It's working fine with commit 8fa5ad6dfb66b6c695aa6a6b54b7031c (and >> 2.11-rc0), and it's broken starting with commit >> 1865e288a823c764cd4344d704e33344 ... do you have any clue what might be >> wrong here now? > > This patch seems to fix the issue for me: > > diff --git a/hw/net/eepro100.c b/hw/net/eepro100.c > index a63ed2c..03e00f7 100644 > --- a/hw/net/eepro100.c > +++ b/hw/net/eepro100.c > @@ -756,8 +756,8 @@ static void read_cb(EEPRO100State *s) > > static void tx_command(EEPRO100State *s) > { > - uint32_t tbd_array = le32_to_cpu(s->tx.tbd_array_addr); > - uint16_t tcb_bytes = (le16_to_cpu(s->tx.tcb_bytes) & 0x3fff); > + uint32_t tbd_array = s->tx.tbd_array_addr; > + uint16_t tcb_bytes = s->tx.tcb_bytes & 0x3fff; > /* Sends larger than MAX_ETH_FRAME_SIZE are allowed, up to 2600 bytes. */ > uint8_t buf[2600]; > uint16_t size = 0; > > Does that make sense? I think the fields are already swapped in > read_cb(), so no need to swap them in tx_command() again? > > Thomas > That makes sense to me. Not entirely sure why my patch caused the regression, but your patch seems like the correct behavior here. Thanks, Mike
diff --git a/hw/net/eepro100.c b/hw/net/eepro100.c index 80b8f47c4b..91dd058010 100644 --- a/hw/net/eepro100.c +++ b/hw/net/eepro100.c @@ -774,23 +774,11 @@ static void tx_command(EEPRO100State *s) } assert(tcb_bytes <= sizeof(buf)); while (size < tcb_bytes) { - uint32_t tx_buffer_address = ldl_le_pci_dma(&s->dev, tbd_address); - uint16_t tx_buffer_size = lduw_le_pci_dma(&s->dev, tbd_address + 4); -#if 0 - uint16_t tx_buffer_el = lduw_le_pci_dma(&s->dev, tbd_address + 6); -#endif - if (tx_buffer_size == 0) { - /* Prevent an endless loop. */ - logout("loop in %s:%u\n", __FILE__, __LINE__); - break; - } - tbd_address += 8; TRACE(RXTX, logout ("TBD (simplified mode): buffer address 0x%08x, size 0x%04x\n", - tx_buffer_address, tx_buffer_size)); - tx_buffer_size = MIN(tx_buffer_size, sizeof(buf) - size); - pci_dma_read(&s->dev, tx_buffer_address, &buf[size], tx_buffer_size); - size += tx_buffer_size; + tbd_address, tcb_bytes)); + pci_dma_read(&s->dev, tbd_address, &buf[size], tcb_bytes); + size += tcb_bytes; } if (tbd_array == 0xffffffff) { /* Simplified mode. Was already handled by code above. */
The simple transmission mode was treating the area immediately after the transmit command block (TCB) as if it were a transmit buffer descriptor, when in reality it is simply the packet data. This change simply copies the data following the TCB into the packet buffer. Signed-off-by: Mike Nawrocki <michael.nawrocki@gtri.gatech.edu> --- hw/net/eepro100.c | 18 +++--------------- 1 file changed, 3 insertions(+), 15 deletions(-)