svn commit: r304142 - head/usr.sbin/bsdinstall/partedit

[Nathan Whitehorn]

On 08/18/16 22:33, Warner Losh wrote:
>> On Aug 18, 2016, at 11:21 PM, Nathan Whitehorn <nwhitehorn at freebsd.org> wrote:
>>
>>
>>
>> On 08/18/16 21:15, Warner Losh wrote:
>>> On Thu, Aug 18, 2016 at 6:56 AM, Allan Jude <allanjude at freebsd.org> wrote:
>>>> On 08/18/16 05:50 AM, Dag-Erling Smørgrav wrote:
>>>>> Nathan Whitehorn <nwhitehorn at freebsd.org> writes:
>>>>>> OK. In which configurations? My Dell servers, for instance, don't do
>>>>>> this. How are they set up? What drivers are being used? Is this
>>>>>> something that affects passthrough disks, RAIDs, disk images?
>>>>> Most LSI MegaRAID controllers don't have real passthrough, only JBOD.
>>>>> You can query the drive with "camcontrol identify passX", but the
>>>>> controller does not report a stripe size for the volume (mfidY).
>>>>>
>>>>>> The point is that *if the reported stripe size is wrong*, more things
>>>>>> than partition alignment in the installer will suffer for it.
>>>>> It's not wrong, it's non-existent, and I'm getting really tired of
>>>>> repeating myself.
>>>>>
>>>>>> Fixing the installer with a bandaid in the run-up to a release is
>>>>>> fine, but *we need to fix the underlying problem*.
>>>>> We can't, because hardware sucks, and I'm getting really tired of
>>>>> repeating myself.
>>>>>
>>>>> DES
>>>>>
>>>> Which makes more sense:
>>>>
>>>> A) If stripesize == 0, use some sane value like 4096
>>> I don't like this.
>>>
>>>> B) Some other combination that uses the reported stripe size, unless it
>>>> is 0, in which case it uses 4096 (or some other value controlled by a
>>>> different new sysctl)
>>> Don't like this so much.
>>>
>>>> C) create kern.geom.min_stripe_size with a default of 512, but users can
>>>> set 4096 if they use only 4k devices. (doesn't really solve the problem
>>>> for the installer)
>>> Default it to 4k, and allow users to set it to 512. If the drive
>>> reports < this value
>>> report this value instead. You'll need to make this a tunable. Then the upper
>>> layers wouldn't care. There's a small chance that some SD cards might be
>>> reporting values that are too large. But I think it is confined to SD cards and
>>> if I see too many more I'll do something specific in the SD driver.
>>>
>>> Warner
>>>
>>>
>> That sounds good to me and I think can clean up a lot of code and potential foot-shooting. Who is planning to make the patch? I'm happy to do anything that would be helpful.
> The patch is super-easy, but I need to get the concept validated and make sure that it does not have unintended side effects.
>
> Warner
>
Sounds great. There is no urgency here -- we have a good solution for 
11.0 already -- so taking time to do it right sounds good. I believe 
stripesize is only consumed by disk formatting tools, so unintended side 
effects at least should be minimal.
-Nathan