Re: [virtio-comment] [RFC PATCH] virtio-rpmb: fix spec land mine in max_[wr|rd]_cnt

[Harald Mommer <harald.mommer@opensynergy.com>]

Hello,

this damned "unlimited" makes me crazy. Nothing in a computer is unlimited.

How easy could life have been if the virtio spec had used 16 bit values 
for max_wr_cnt and max_rd_cnt in the config space but it is as it is and 
this cannot be changed any more.

On 19.06.23 14:01, Alex BennÃe wrote:
> Even if 0 meant no limit we are still limited by the field size of the
> request. That said for a maximum sized partition (* 80 128 1024) you
> could only actually request 40960 blocks before running out of device.
> Perhaps it would be better to mark 0 as invalid?

Where comes this 80 from? I saw a 0x80 in the virtio specification, 
which would be 128 and not 80 decimal.

> Cc: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
> Cc: Harald Mommer <hmo@opensynergy.com>
> Cc: Will Deacon <will@kernel.org>
> Signed-off-by: Alex BennÃe <alex.bennee@linaro.org>
> ---
>   device-types/rpmb/description.tex | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/device-types/rpmb/description.tex b/device-types/rpmb/description.tex
> index 1dae3fd..2ce8a5b 100644
> --- a/device-types/rpmb/description.tex
> +++ b/device-types/rpmb/description.tex
> @@ -37,7 +37,7 @@ \subsection{Device configuration layout}\label{sec:Device Types / RPMB Device /
>      The values MUST range between 0x00 and 0x80 inclusive.
>   \item[\field{max_wr_cnt and max_rd_cnt}] are the maximum numbers of RPMB
>      block count (256B) that can be performed to device in one request. 0 implies
> -   no limitation.
> +   no limitation other than the maximum value you can store in \field{block_count} (65535).
>   \end{description}

Yes, 65535 is what fits in the 2 byte block count of various underlying 
storage technology specs. So 65535 is an upper limit. Smallest upper 
limit which makes sense? I guess no as there is underlying storage 
technology.

JESD85-B51 for eMMC:

Looking at 6.6.22.4.3 Authenticated Data Write I see that at least for 
write 2 or 3 sizes may be supported:

1 block, 2 blocks and optionally 32 blocks but nothing in between of 2 
and 32.

This is strange. If I understand this correctly (Maybe but I'm not sure) 
the limits of the underlying storage technology cannot be always 
represented cleanly by the virtio config space definitions due to the 
missing 3..31 value range.

JESD220C-2.2 for UFS is more friendly:

14.1.4.4 Geometry descriptor

bRPMB_ReadWriteSize is an 8 bit value, 0 seems not to be foreseen. 255 
is the theoretical maximum and we don't have to cope with 0 means 
"unlimited".

And then I found something in an NVMe spec (NVM Express Base 
Specification, revision 2.0b)

Figure 275 says that in bytes 315:312 some bits 31:24 (8 bits)

"Access Size: If the Number of RPMB Units field is non-zero, then this 
field indicates the maximum number of 512B units of data that may be 
read or written per RPMB access by Security Send or Security Receive 
commands for the controller. This is a 0âs based value. A value of 0h 
indicates support for one unit of 512B of data."

So if I understand this correctly NVMe has the biggest limit with 255 + 1.

Considering current storage technology it could be that in practice for 
now the best implementation choice when interpreting "unlimited" may be 
to interpret this as "256".

Replacing the word "unlimited" by "65535" may not be what should be done 
here.

>   
>   \devicenormative{\subsection}{Device Initialization}{Device Types / RPMB Device / Device Initialization}