Missing Diag/Debug Registers in lsi53c895a emulation

On 12.01.2015 10:28, Peter Lieven wrote:
> Hi,
>
> while trying to run an ASA1000V appliance I found that registers 0x09 and 0x3c-3f are read during boot up. Which
> currently results in an exit(1). The appliance uses an quite ancient 2.6.29.6 Linux kernel.
> These registers are mainly for diag and debug purposes so I came up with the following patch. The system works
> with the following patch, but I wonder if this is a correct workaround:

Maybe this is better:

On 12/01/2015 10:39, Peter Lieven wrote:
> On 12.01.2015 10:28, Peter Lieven wrote:
>> Hi,
>>
>> while trying to run an ASA1000V appliance I found that registers 0x09
>> and 0x3c-3f are read during boot up. Which
>> currently results in an exit(1). The appliance uses an quite ancient
>> 2.6.29.6 Linux kernel.
>> These registers are mainly for diag and debug purposes so I came up
>> with the following patch. The system works
>> with the following patch, but I wonder if this is a correct workaround:
> 
> Maybe this is better:
> 
> diff --git a/hw/scsi/lsi53c895a.c b/hw/scsi/lsi53c895a.c
> index ec92048..db7d4b8 100644
> --- a/hw/scsi/lsi53c895a.c
> +++ b/hw/scsi/lsi53c895a.c
> @@ -277,6 +277,7 @@ typedef struct {
>      uint32_t csbc;
>      uint32_t scratch[18]; /* SCRATCHA-SCRATCHR */
>      uint8_t sbr;
> +    uint32_t adder;
> 
>      /* Script ram is stored as 32-bit words in host byteorder.  */
>      uint32_t script_ram[2048];
> @@ -1389,6 +1390,7 @@ again:
>                  switch ((insn >> 27) & 7) {
>                  case 0: /* Jump */
>                      DPRINTF("Jump to 0x%08x\n", addr);
> +                    s->adder = addr;
>                      s->dsp = addr;
>                      break;
>                  case 1: /* Call */
> @@ -1513,6 +1515,8 @@ static uint8_t lsi_reg_readb(LSIState *s, int offset)
>          return 0x7f;
>      case 0x08: /* Revision ID */
>          return 0x00;
> +    case 0x09: /* SOCL */
> +        return s->socl;
>      case 0xa: /* SSID */
>          return s->ssid;
>      case 0xb: /* SBCL */
> @@ -1577,6 +1581,8 @@ static uint8_t lsi_reg_readb(LSIState *s, int offset)
>          return s->sbr;
>      case 0x3b: /* DCNTL */
>          return s->dcntl;
> +    /* ADDER Output (Debug of relative jump address) */
> +    CASE_GET_REG32(adder, 0x3c)
>      case 0x40: /* SIEN0 */
>          return s->sien0;
>      case 0x41: /* SIEN1 */
> 

This looks good.  Can you send it with SoB and git-format-patch?

Paolo

On 12.01.2015 10:42, Paolo Bonzini wrote:
>
> On 12/01/2015 10:39, Peter Lieven wrote:
>> On 12.01.2015 10:28, Peter Lieven wrote:
>>> Hi,
>>>
>>> while trying to run an ASA1000V appliance I found that registers 0x09
>>> and 0x3c-3f are read during boot up. Which
>>> currently results in an exit(1). The appliance uses an quite ancient
>>> 2.6.29.6 Linux kernel.
>>> These registers are mainly for diag and debug purposes so I came up
>>> with the following patch. The system works
>>> with the following patch, but I wonder if this is a correct workaround:
>> Maybe this is better:
>>
>> diff --git a/hw/scsi/lsi53c895a.c b/hw/scsi/lsi53c895a.c
>> index ec92048..db7d4b8 100644
>> --- a/hw/scsi/lsi53c895a.c
>> +++ b/hw/scsi/lsi53c895a.c
>> @@ -277,6 +277,7 @@ typedef struct {
>>       uint32_t csbc;
>>       uint32_t scratch[18]; /* SCRATCHA-SCRATCHR */
>>       uint8_t sbr;
>> +    uint32_t adder;
>>
>>       /* Script ram is stored as 32-bit words in host byteorder.  */
>>       uint32_t script_ram[2048];
>> @@ -1389,6 +1390,7 @@ again:
>>                   switch ((insn >> 27) & 7) {
>>                   case 0: /* Jump */
>>                       DPRINTF("Jump to 0x%08x\n", addr);
>> +                    s->adder = addr;
>>                       s->dsp = addr;
>>                       break;
>>                   case 1: /* Call */
>> @@ -1513,6 +1515,8 @@ static uint8_t lsi_reg_readb(LSIState *s, int offset)
>>           return 0x7f;
>>       case 0x08: /* Revision ID */
>>           return 0x00;
>> +    case 0x09: /* SOCL */
>> +        return s->socl;
>>       case 0xa: /* SSID */
>>           return s->ssid;
>>       case 0xb: /* SBCL */
>> @@ -1577,6 +1581,8 @@ static uint8_t lsi_reg_readb(LSIState *s, int offset)
>>           return s->sbr;
>>       case 0x3b: /* DCNTL */
>>           return s->dcntl;
>> +    /* ADDER Output (Debug of relative jump address) */
>> +    CASE_GET_REG32(adder, 0x3c)
>>       case 0x40: /* SIEN0 */
>>           return s->sien0;
>>       case 0x41: /* SIEN1 */
>>
> This looks good.  Can you send it with SoB and git-format-patch?

Sure. I think full support of the adder debug register would be more complicated,
but as the spec says its main purpose is debugging of relative jump address calculation
it should be fine. In fact if anything had read from this register before qemu would
have aborted ;-)

Peter

>
> Paolo