Sun4m : SCSI ESP controller

On 16/02/14 23:13, Olivier Danet wrote:

> Two small fixes for the ESP (AM53C94) SCSI controller
>
> * Signal the end of the DMA transfer after a SCSI command.
>
> * The status register (RSTAT) is cleared after reading the interrupt
> status register (RINTR), except for the TC bit (=Count To Zero) and the
> scsi phase bits, which mirror SCSI signals levels.
>
> Fixes the bug "esp0: !TC on DATA XFER" with NetBSD
> https://bugs.launchpad.net/qemu/+bug/1055090
>
> Signed-off-by: Olivier Danet <odanet@caramail.com>
> ---
> hw/scsi/esp.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
> index 2d150bf..5e91077 100644
> --- a/hw/scsi/esp.c
> +++ b/hw/scsi/esp.c
> @@ -246,6 +246,7 @@ static void esp_do_dma(ESPState *s)
> s->cmdlen = 0;
> s->do_cmd = 0;
> do_cmd(s, s->cmdbuf);
> + esp_dma_done(s);
> return;
> }
> if (s->async_len == 0) {
> @@ -417,10 +418,9 @@ uint64_t esp_reg_read(ESPState *s, uint32_t saddr)
> except TC */
> old_val = s->rregs[ESP_RINTR];
> s->rregs[ESP_RINTR] = 0;
> - s->rregs[ESP_RSTAT] &= ~STAT_TC;
> s->rregs[ESP_RSEQ] = SEQ_CD;
> esp_lower_irq(s);
> -
> + s->rregs[ESP_RSTAT] &= STAT_TC | STAT_MI;
> return old_val;
> default:
> break;

Hi Olivier,

I've applied the non-whitespace damaged patch (attached) but 
unfortunately I still see the "!TC on DATA XFER" bug with my NetBSD 5 
ISO under qemu-system-sparc :/  Note that the second part of the patch 
showed some fuzz so I'm wondering if this patch is complete or whether 
part of it is still missing?

Also ESP patches should have an esp or scsi/esp prefix in the subject 
line and should CC the current SCSI maintainer (see MAINTAINERS file) 
which is currently Paolo.


HTH,

Mark.

Il 20/02/2014 09:58, Mark Cave-Ayland ha scritto:
>
> Hi Olivier,
>
> I've applied the non-whitespace damaged patch (attached) but
> unfortunately I still see the "!TC on DATA XFER" bug with my NetBSD 5
> ISO under qemu-system-sparc :/  Note that the second part of the patch
> showed some fuzz so I'm wondering if this patch is complete or whether
> part of it is still missing?
>
> Also ESP patches should have an esp or scsi/esp prefix in the subject
> line and should CC the current SCSI maintainer (see MAINTAINERS file)
> which is currently Paolo.
>

BTW, Mark, should you add yourself to MAINTAINERS for sparc?

Paolo

On 20/02/2014 09:58, Mark Cave-Ayland wrote:
> On 16/02/14 23:13, Olivier Danet wrote:
>
>> Two small fixes for the ESP (AM53C94) SCSI controller
>>
>> * Signal the end of the DMA transfer after a SCSI command.
>>
>> * The status register (RSTAT) is cleared after reading the interrupt
>> status register (RINTR), except for the TC bit (=Count To Zero) and the
>> scsi phase bits, which mirror SCSI signals levels.
>>
>> Fixes the bug "esp0: !TC on DATA XFER" with NetBSD
>> https://bugs.launchpad.net/qemu/+bug/1055090
>>
>> Signed-off-by: Olivier Danet <odanet@caramail.com>
>> ---
>> hw/scsi/esp.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
>> index 2d150bf..5e91077 100644
>> --- a/hw/scsi/esp.c
>> +++ b/hw/scsi/esp.c
>> @@ -246,6 +246,7 @@ static void esp_do_dma(ESPState *s)
>> s->cmdlen = 0;
>> s->do_cmd = 0;
>> do_cmd(s, s->cmdbuf);
>> + esp_dma_done(s);
>> return;
>> }
>> if (s->async_len == 0) {
>> @@ -417,10 +418,9 @@ uint64_t esp_reg_read(ESPState *s, uint32_t saddr)
>> except TC */
>> old_val = s->rregs[ESP_RINTR];
>> s->rregs[ESP_RINTR] = 0;
>> - s->rregs[ESP_RSTAT] &= ~STAT_TC;
>> s->rregs[ESP_RSEQ] = SEQ_CD;
>> esp_lower_irq(s);
>> -
>> + s->rregs[ESP_RSTAT] &= STAT_TC | STAT_MI;
>> return old_val;
>> default:
>> break;
>
> Hi Olivier,
>
> I've applied the non-whitespace damaged patch (attached) but 
> unfortunately I still see the "!TC on DATA XFER" bug with my NetBSD 5 
> ISO under qemu-system-sparc :/  Note that the second part of the patch 
> showed some fuzz so I'm wondering if this patch is complete or whether 
> part of it is still missing?
>
> Also ESP patches should have an esp or scsi/esp prefix in the subject 
> line and should CC the current SCSI maintainer (see MAINTAINERS file) 
> which is currently Paolo.
>
>
> HTH,
>
> Mark.

I don't get these errors installing NetBSD 5 (tried 5.02 & 5.1).
With NetBSD 6, the patch makes a lot of difference :

http://temlib.org/pub/qemu_tc_patch.png
http://temlib.org/pub/qemu_tc_nopatch.png

(the default terminal during install is terrible)

* The dma() part is a coherence bug in the ESP driver which does not
update the DMA registers after a command transfer.

* The RINTR part makes the driver match more closely the datasheet:
http://sensi.org/download/doc/am53c94.pdf
Page 19/63

These modifications are not particularly tailored for NetBSD nor SPARCs.

Olivier

(Sorry about the whitespaces, I will try to find out the cause !)

On 20/02/14 08:58, Mark Cave-Ayland wrote:

> On 16/02/14 23:13, Olivier Danet wrote:
>
>> Two small fixes for the ESP (AM53C94) SCSI controller
>>
>> * Signal the end of the DMA transfer after a SCSI command.
>>
>> * The status register (RSTAT) is cleared after reading the interrupt
>> status register (RINTR), except for the TC bit (=Count To Zero) and the
>> scsi phase bits, which mirror SCSI signals levels.
>>
>> Fixes the bug "esp0: !TC on DATA XFER" with NetBSD
>> https://bugs.launchpad.net/qemu/+bug/1055090
>>
>> Signed-off-by: Olivier Danet <odanet@caramail.com>
>> ---
>> hw/scsi/esp.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
>> index 2d150bf..5e91077 100644
>> --- a/hw/scsi/esp.c
>> +++ b/hw/scsi/esp.c
>> @@ -246,6 +246,7 @@ static void esp_do_dma(ESPState *s)
>> s->cmdlen = 0;
>> s->do_cmd = 0;
>> do_cmd(s, s->cmdbuf);
>> + esp_dma_done(s);
>> return;
>> }
>> if (s->async_len == 0) {
>> @@ -417,10 +418,9 @@ uint64_t esp_reg_read(ESPState *s, uint32_t saddr)
>> except TC */
>> old_val = s->rregs[ESP_RINTR];
>> s->rregs[ESP_RINTR] = 0;
>> - s->rregs[ESP_RSTAT] &= ~STAT_TC;
>> s->rregs[ESP_RSEQ] = SEQ_CD;
>> esp_lower_irq(s);
>> -
>> + s->rregs[ESP_RSTAT] &= STAT_TC | STAT_MI;
>> return old_val;
>> default:
>> break;
>
> I've applied the non-whitespace damaged patch (attached) but
> unfortunately I still see the "!TC on DATA XFER" bug with my NetBSD 5
> ISO under qemu-system-sparc :/ Note that the second part of the patch
> showed some fuzz so I'm wondering if this patch is complete or whether
> part of it is still missing?
>
> Also ESP patches should have an esp or scsi/esp prefix in the subject
> line and should CC the current SCSI maintainer (see MAINTAINERS file)
> which is currently Paolo.

Hi Olivier,

I've done some more testing on this patch today, and what I've found is 
that there are 2 types of error emitted from the NetBSD ESP driver: the 
majority of the "!TC on DATA XFER" messages go away under NetBSD 6 with 
the patch, but I still get the occasional one on the console, 
particularly with NetBSD 5. There are still messages about DMA 
interrupts appearing as can be seen at 
http://www.ilande.co.uk/tmp/netbsd5-esp.png.

Both Linux and NetBSD seem to install fine with the patch applied, so no 
regressions there - I haven't had a chance to test Solaris yet. If 
Artyom could give this a quick test on his collection of Solaris images 
then I'd say the patch is okay to apply.

More interesting is that with or without the patch, if you boot a 
standard Debian etch .qcow2 HD image and then shut it down then you get 
the following kernel panic: http://www.ilande.co.uk/tmp/debian-esp.png.

So my current conclusion is that this patch doesn't appear to cause any 
noticeable regressions outside of Solaris, however there is a still a 
problem somewhere with the ESP DMA interrupts on SPARC32.


ATB,

Mark.

On 20/02/14 11:13, Paolo Bonzini wrote:

> Il 20/02/2014 09:58, Mark Cave-Ayland ha scritto:
>>
>> Hi Olivier,
>>
>> I've applied the non-whitespace damaged patch (attached) but
>> unfortunately I still see the "!TC on DATA XFER" bug with my NetBSD 5
>> ISO under qemu-system-sparc :/ Note that the second part of the patch
>> showed some fuzz so I'm wondering if this patch is complete or whether
>> part of it is still missing?
>>
>> Also ESP patches should have an esp or scsi/esp prefix in the subject
>> line and should CC the current SCSI maintainer (see MAINTAINERS file)
>> which is currently Paolo.
>>
>
> BTW, Mark, should you add yourself to MAINTAINERS for sparc?

If people are okay to apply my patches then I don't mind :) I'll send 
this request through separately.


ATB,

Mark.