Message ID | fb8d4f70908311348p4b63220bqb661705b8c3d3f3b@mail.gmail.com |
---|---|
State | Superseded |
Headers | show |
On Mon, Aug 31, 2009 at 11:48 PM, Artyom Tarasenko<atar4qemu@googlemail.com> wrote: > Trying to find out what could be the mysterious scsi commands 60, 80, > c0 and e0, I found out that OBP during dma exchange has problems with > endianness. > OBP sends commands like this: > 12 20 00 00 05 00 or this 12 80 00 00 05 00, but esp.c expects > that the second byte is the command and the first one is the target. > > The following patch solves the problem, but breaks OpenBIOS. The > question is where do we have wrong endianness? It's not necessarily > esp, it may be something on the way like dma or mmu. Any ideas? In addition to OpenBIOS, I'd also think Linux, NetBSD and OpenBSD should break. Lance needs similar operation, but there the bus is little endian and one bit configures byte swapping for some of data. Usually the first byte is bus id. What if you changed (as a hack) here: DPRINTF("do_cmd: busid 0x%x\n", buf[0]); lun = buf[0] & 7; datalen = s->current_dev->send_command(s->current_dev, 0, &buf[1], lun); &buf[1] to &buf[0]?
2009/9/1 Blue Swirl <blauwirbel@gmail.com>: > Usually the first byte is bus id. What if you changed (as a hack) here: > DPRINTF("do_cmd: busid 0x%x\n", buf[0]); > lun = buf[0] & 7; > datalen = s->current_dev->send_command(s->current_dev, 0, &buf[1], lun); > &buf[1] to &buf[0]? Yep, figured that out too, but did a little more complex hack: lun = buf[1] >>5; datalen = s->current_dev->send_command(s->current_dev, 0, &buf[0], lun); with this hack OBP checks proper LUNs. So, the problem is not endianness, but the missing bus id. But how the bus id should get into the dma buffer? It should be generated by esp, right?
On Tue, Sep 1, 2009 at 8:08 PM, Artyom Tarasenko<atar4qemu@googlemail.com> wrote: > 2009/9/1 Blue Swirl <blauwirbel@gmail.com>: >> Usually the first byte is bus id. What if you changed (as a hack) here: >> DPRINTF("do_cmd: busid 0x%x\n", buf[0]); >> lun = buf[0] & 7; >> datalen = s->current_dev->send_command(s->current_dev, 0, &buf[1], lun); >> &buf[1] to &buf[0]? > > Yep, figured that out too, but did a little more complex hack: > > lun = buf[1] >>5; > datalen = s->current_dev->send_command(s->current_dev, 0, &buf[0], lun); > > with this hack OBP checks proper LUNs. > > So, the problem is not endianness, but the missing bus id. But how the > bus id should get into the dma buffer? > It should be generated by esp, right? > It could be pushed into FIFO with PIO.
diff --git a/hw/esp.c b/hw/esp.c index aad547e..aceee48 100644 --- a/hw/esp.c +++ b/hw/esp.c @@ -169,11 +169,17 @@ static uint32_t get_cmd(ESPState *s, uint8_t *buf) { uint32_t dmalen; int target; - - target = s->wregs[ESP_WBUSID] & BUSID_DID; - if (s->dma) { - dmalen = s->rregs[ESP_TCLO] | (s->rregs[ESP_TCMID] << 8); - s->dma_memory_read(s->dma_opaque, buf, dmalen); + uint8_t tmp,i; + + target = s->wregs[ESP_WBUSID] & BUSID_DID; + if (s->dma) { + dmalen = s->rregs[ESP_TCLO] | (s->rregs[ESP_TCMID] << 8); + s->dma_memory_read(s->dma_opaque, buf, dmalen); + for(i=0;i<dmalen;i+=2) { + tmp=buf[i]; + buf[i]=buf[i+1]; + buf[i+1]=tmp; + } } else { dmalen = s->ti_size; memcpy(buf, s->ti_buf, dmalen);