Re: [PATCH] erofs: fix compacted_{4b_initial, 2b} when compacted_4b_initial > totalidx

[Yue Hu]

On Mon, 13 Sep 2021 19:19:16 +0800
Gao Xiang <hsiangkao@xxxxxxxxxxxxxxxxx> wrote:

> Hi Yue,
> 
> On Mon, Sep 13, 2021 at 06:58:36PM +0800, Yue Hu wrote:
> > Hi Xiang,
> > 
> > On Mon, 13 Sep 2021 17:11:24 +0800
> > Gao Xiang <hsiangkao@xxxxxxxxxxxxxxxxx> wrote:
> >   
> > > On Mon, Sep 13, 2021 at 05:00:16PM +0800, Yue Hu wrote:  
> > > > On Mon, 13 Sep 2021 16:48:45 +0800
> > > > Gao Xiang <hsiangkao@xxxxxxxxxxxxxxxxx> wrote:
> > > >     
> > > > > Hi Yue,
> > > > > 
> > > > > On Mon, Sep 13, 2021 at 03:24:05PM +0800, Yue Hu wrote:    
> > > > > > From: Yue Hu <huyue2@xxxxxxxxxx>
> > > > > > 
> > > > > > mkfs.erofs will treat compacted_4b_initial & compacted_2b as 0 if
> > > > > > compacted_4b_initial > totalidx, kernel should be aligned with it
> > > > > > accordingly.      
> > > > > 
> > > > > There is no difference between compacted_4b_initial or compacted_4b_end
> > > > > for compacted 4B. Since in this way totalidx for compact 2B won't larger
> > > > > than 16 (number of lclusters in a compacted 2B pack.)    
> > > > 
> > > > However, we can see compacted_2b is a big number for this case. It should
> > > > be pointless.    
> > > 
> > > Does it has some real impact?  
> > 
> > No real impact to correct result.
> >   
> > > 
> > > compacted_4b_initial is only used for the alignment use for the
> > > first compacted_2b so that each compacted_2b pack won't cross
> > > the block (page) boundary. And compacted_4b_end is for the last
> > > lclusters aren't fitted in any compacted_2b pack.
> > > 
> > > If compacted_4b_initial > totalidx, I think the whole indexes
> > > would be compacted 4B and handled in
> > > 
> > > 	if (lcn < compacted_4b_initial) {
> > > 		amortizedshift = 2;
> > > 		goto out;
> > > 	}  
> > 
> > Yes, it is. 
> > 
> > My point is why we need compacted_2b here for this case. If it's
> > not helpful/used for next code logic, we should remove/avoid it.
> > I think that may cause some misunderstanding and consume unneeded
> > CPU resources.  
> 
> Okay, make sense. If the number of compacted_2b misleads, how about
> just
> 
> 	if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
> 	    compacted_4b_initial <= totalidx) {
> 		compacted_2b = ...;
> 	} else {
> 		compacted_2b = 0;
> 	}

Looks good. So, the patch title may change a little.

> 
> , and refine the commit message to point out the following facts for
> other folks:
> 
> - compacted_4b_initial is used contain the very first lclusters in order
>   to fulfill the alignment of the first compacted_2b pack;
> 
> - compacted_4b_end is used for the last lclusters which aren't fitted in
>   the previous compacted_2b packs;
> 
> - if compacted_4b_initial > totalidx, the whole indexes will be compacted
>   4B and handled with compacted_4b_initial.

Ok, will add these facts above in v2.

Thanks.

> 
> Thanks,
> Gao Xiang
> 
> > 
> > Thanks.