Date: Mon, 27 May 2019 23:21:50 +0300 From: Andriy Gapon <avg@FreeBSD.org> To: "Bjoern A. Zeeb" <bzeeb-lists@lists.zabbadoz.net> Cc: freebsd-arch@FreeBSD.org Subject: Re: proposal: require ivar accessors to succeed Message-ID: <f489acfb-c107-82d1-8d01-19f36a78d1c4@FreeBSD.org> In-Reply-To: <E7700915-34BE-4371-A258-C010638CFA38@lists.zabbadoz.net> References: <2b2ab28f-45c5-1c28-f923-170d95c20c3d@FreeBSD.org> <E7700915-34BE-4371-A258-C010638CFA38@lists.zabbadoz.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On 27/05/2019 21:10, Bjoern A. Zeeb wrote:
> On 27 May 2019, at 5:44, Andriy Gapon wrote:
>
>> __BUS_ACCESSOR() macro is used to define accessors to bus IVAR variables.
>> Unfortunately, accessors defined in such a fashion completely ignore return
>> values of BUS_READ_IVAR() and BUS_WRITE_IVAR() method calls. There is no way to
>> see if a call is successful. Typically, this should not be a problem as a
>> device driver targets a specific bus (sometimes, buses) and it should know what
>> IVARs the bus has. So, the driver should make only those IVAR calls that are
>> supposed to always succeed on the bus.
>> But sometimes things can go wrong as with everything else.
>>
>> So, I am proposing to add some code to __BUS_ACCESSOR to verify the success.
>> For example, we can panic when a call fails. The checks could be under
>> INVARIANTS or under DIAGNOSTICS or under a new kernel option.
>> A less drastic option is to print a warning message on an error.
>>
>> This is mostly intended to help driver writers and maintainers.
>>
>> Opinions, suggestions, etc are welcome.
>
> What about “fixing” the KPI (possibly adding a 2nd one), deprecating the old
> one, and (slowly over time) migrating old stuff over?
I think that the two proposals are not mutually exclusive.
And I think that both make sense.
However, it's hard for me to imagine a desire to replace code like this
devid = pci_get_devid(dev);
with this
err = pci_get2_devid(dev, &devid);
if (err != 0) {
...
}
Especially given that, modulo bugs, dev is going to be a device on the pci bus
and the call is going to succeed.
In other words, in my opinion, the only cases where an accessor is allowed to
fail are:
- a driver somehow attached to a device on an unexpected bus
- uncoordinated changes in between a bus driver and a device driver
So, programming errors.
--
Andriy Gapon
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?f489acfb-c107-82d1-8d01-19f36a78d1c4>