[U-Boot] [EXT] Re: [PATCH 7/7] scsi: dts: a3700: add scsi node

[Stefan Roese]

Hi Simon, Hi Ken,

On 01.04.2017 06:22, Simon Glass wrote:
> On 27 March 2017 at 02:28, Ken Ma <make at marvell.com> wrote:
>> Hi Stefan
>>
>>
>>
>> Thanks a lot for your kind reply.
>>
>>
>>
>> But I still do not think it's very good to change sata's uclass id from
>> "UCLASS_AHCI" to "UCLASS_SCSI".
>>
>> If we do such change, UCLASS_AHCI is lost since from the sata.c codes, it
>> does the AHCI initialization work but not SCSI initialization work.
>>
>> If u-boot supports ISIS scanner which supports SCSI, I think its uclass id
>> should be like UCLASS_ISIS but not UCLASS_SCSI.
>>
>> And if we set sata's uclass id as "UCLASS_SCSI", it should provide basic
>> SCSI function, then why can’t we connect a parallel SCSI device like SCSI
>> scanner or cd-rom to the SATA interface?
>>
>> From https://en.wikipedia.org/wiki/Serial_ATA#SATA_and_SCSI
>>
>> In general, SATA devices link compatibly to SAS enclosures and adapters,
>> whereas SCSI devices cannot be directly connected to a SATA bus
>>
>>
>>
>>
>>
>> Actually Marvell’s sata controller is SAS(Serial Attached SCSI system), it
>> integrates SCSI and SATA(AHCI); SAS provides a SCSI bus which works only in
>> SAS range(for example, 2 sata ports in SAS), so actually the SCSI bus
>> controller is not "virtual" controllers but has the same device base
>> register as SATA.
>>
>>
>>
>> From https://en.wikipedia.org/wiki/Serial_Attached_SCSI
>>
>> A typical Serial Attached SCSI system consists of the following basic
>> components:
>>
>> 1.    An initiator: a device that originates device-service and
>> task-management requests for processing by a target device and receives
>> responses for the same requests from other target devices. Initiators may be
>> provided as an on-board component on the motherboard (as is the case with
>> many server-oriented motherboards) or as an add-on host bus adapter.
>>
>> 2.    A target: …
>>
>> So in my opinion, there are two ways to implement SAS as below
>>
>> A.  If our codes provide SAS controller as an on-board component – then a
>> uclass id of UCLASS_SAS should be defined and then in scsi_scan() of
>> scsi_scan.c, both devices of UCLASS_SCSI and UCLASS_SAS should be scanned.
>> In such implementation, UCLASS_SCSI is for parallel SCSI while UCLASS_SAS is
>> for serial attached SCSI;
>>
>> B.  SAS works as an add-on host bus adapter as above said, SAS’s SCSI
>> controller and AHCI controller are both itself as below - SCSI controller is
>> not a virtual device, it exists and only works in SAS internal range(since
>> there is no UCLASS_SAS, I take this way);
>>
>> Although the SAS’s SCSI controller does not need to any special hardware
>> configuration; but actually I think there is something to do, we should bind
>> special scsi_exec() to SCSI devices in SCSI driver or SAS driver (For
>> different SCSI controls, SAS must have different implementation of
>> scsi_exec() comparing to SCSI scanner, or other SCSI devices)
>>
>> By the way, I think we should move the work of creating block devices to
>> scsi-uclass.c
>>
>> scsi: scsi at e0000 {
>>
>>                         compatible = "marvell,mvebu-scsi";
>>
>> reg = <0xe0000 0x2000>;
>>
>>                         sata: sata at e0000 {
>>
>>                               compatible = "marvell,armada-3700-ahci";
>>
>>                               reg = <0xe0000 0x2000>;
>>
>>                               interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
>>
>>                         };
>
> Does this match the Linux DT?

No, and this is my main concern. This patch series introduces a new
"scsi" DT node and moves the controller(s) one level up into this
newly created DT node (Ken please correct me if I'm wrong here).
We simply can't make such changes here in U-Boot without this
being first discussed and decided on the Linux DT mailing list
with the DT maintainers.

Ken, how is this problem solved / handled in Linux without this DT
changes up until now?

Thanks,
Stefan