RE: [RFC PATCH V3 net-next 1/5] ethtool: Allow network drivers to dump arbitrary EEPROM data

["Don Bollinger" <don@thebollingers.org>]

On 3/16/2021 11:24 AM, Moshe Shemesh wrote:
> On 3/16/2021 12:31 AM, Don Bollinger wrote:
> > On Mon, 15 Mar 2021 10:12:39 +0700 Moshe Shemesh wrote:
> >> From: Vladyslav Tarasiuk <vladyslavt@nvidia.com>
> >>
> >> Define get_module_eeprom_data_by_page() ethtool callback and
> >> implement netlink infrastructure.
> >>

<snip>

> >> +     request->offset =
> >> nla_get_u32(tb[ETHTOOL_A_EEPROM_DATA_OFFSET]);
> >> +     request->length =
> >> nla_get_u32(tb[ETHTOOL_A_EEPROM_DATA_LENGTH]);
> >> +     if (tb[ETHTOOL_A_EEPROM_DATA_PAGE] &&
> >> +         dev->ethtool_ops->get_module_eeprom_data_by_page &&
> >> +         request->offset + request->length >
> >> ETH_MODULE_EEPROM_PAGE_LEN)
> >> +             return -EINVAL;
> > If a request exceeds the length of EEPROM (in legacy), we truncate it
> > to EEPROM length.  I suggest if a request exceeds the length of the
> > page (in the new KAPI), then we truncate at the end of the page.  Thus
> > rather than 'return -EINVAL;' I suggest
> >
> >            request->length = ETH_MODULE_EEPROM_PAGE_LEN -
> > request->offset;
> 
> I was not sure about that, isn't it better that once user specified page he has
> to be in the page boundary and let him know the command failed ?

On further review, I agree with your original code.  Reading across a page boundary will require explicit tracking of offset and page to read from.  If you get a short read, you *might* have to update the page on the next read.  If you have to keep track of where the page boundary is and adjust the offset and page together, then you should be explicitly managing not to try to read across a page boundary.  Put another way, if you try to read across a page boundary you probably have a bug in your code and will appreciate an immediate error.

I withdraw my previous suggestion, don't change it.

Don