Re: [PATCH v1 5/6] riscv/virt: Add the PFlash CFI01 device

[Philippe Mathieu-Daudé]

On 9/25/19 2:55 AM, Alistair Francis wrote:
> On Tue, Sep 24, 2019 at 2:32 AM Philippe Mathieu-Daudé
> <address@hidden> wrote:
>>
>> On 9/23/19 11:46 PM, Peter Maydell wrote:
>>> On Fri, 20 Sep 2019 at 23:23, Alistair Francis <address@hidden> wrote:
>>>> On Thu, Sep 19, 2019 at 10:15 PM Bin Meng <address@hidden> wrote:
>>>>> I don't think we should mirror what is used on ARM virt board to
>>>>> create 2 flash for sifive_u. For ARM virt, there are 2 flashes because
>>>>> they need distinguish secure and non-secure. For sifive_u, only one is
>>>>> enough.
>>>>
>>>> I went back and forward about 1 or 2. Two seems more usable as maybe
>>>> someone wants to include two pflash files? The Xilinx machine also has
>>>> two so I'm kind of used to 2, but I'm not really fussed.
>>
>> The Xilinx machine has 2 because it matches the hardware.
>>
>>> One of the reasons for having 2 on the Arm board (we do this
>>> even if we're not supporting secure vs non-secure) is that
>>> then you can use one for a fixed read-only BIOS image (backed
>>> by a file on the host filesystem shared between all VMs), and
>>> one backed by a read-write per-VM file providing permanent
>>> storage for BIOS environment variables. Notably UEFI likes to
>>> work this way, but the idea applies in theory to other
>>> boot loader or BIOSes I guess.
>>
>> IIRC Laszlo's explanation, the only reason it is that way is because the
>> pflash_cfi01 model still doesn't implement sector locking. At the OVMF
>> emerged from EDK2, to have a safe ROM vs DATA storage it was decided to
>> use 2 different flashes.
>> When I understood when this config layout started, I suggested Laszlo to
>> use a real ROM to store the OVMF CODE since it is pointless to do
>> firmware upgrade in virtualized environment, but he said it was too late
>> to change the design.
>>
>> If you don't plan to run UEFI "Capsule Update" on your Virt board, I
>> suggest using memory_region_init_rom() with your firmware CODE, and a
>> parallel/SPI flash for the data VARStore.
> 
> We might run that one day, who knows :)

You certainly want to run EDK2, I'm following RISCV progress on the list.

What doesn't make sense on a virtualized platform is the "Capsule
Update" feature IMO. Where it makes sense is on the SiFive E/U boards
models.

>>> I would suggest also checking with Markus that your code
>>> for instantiating the flash devices follows the current
>>> recommendations so the backing storage can be configured
>>> via -blockdev. (This is a fairly recent change from June or
>>> so; current-in-master virt and sbsa boards provide an example
>>> of doing the right thing, I think.)
>>>
>>> thanks
>>> -- PMM
>>>