[PATCH] makedumpfile: Fix a bug found during regression testing for v1.5.6

[Petr Tesarik]

On Wed, 16 Apr 2014 08:11:27 +0000
Atsushi Kumagai <kumagai-atsushi at mxc.nes.nec.co.jp> wrote:

> Hello,
> 
> I have to do regression testing again for this issue and the OOM
> issue reported by Vivek, so the release of v1.5.6 will be put off
> to next week.
> 
> Thanks for your patience.
> 
> Atsushi Kumagai
> 
> --
> From: Atsushi Kumagai <kumagai-atsushi at mxc.nes.nec.co.jp>
> Date: Wed, 16 Apr 2014 14:56:01 +0900
> Subject: [PATCH] Exclude small cycle regions wrongly.
> 
> In bitmap initialization, if a cycle region is smaller than 8 pages
> and its start_pfn is aligned to multiple of 8 pages, it will be
> excluded wrongly.
> 
>   Example case:
>     pfn_start=0x103ff80, pfn_start_roundup=0x103ff80
>     pfn_end=0x103ff84, pfn_end_round=0x103ff80
> 
> This case doesn't meet any conditions to set bits in
> create_1st_bitmap_cyclic() and initialize_2nd_bitmap_cyclic().
> We should admit the case where pfn_start is equal to pfn_end_round
> like above.

Hello Kumagai-san,

FWIW I cleaned up these bit operations in makedumpfile, completely
removing this code and instead adding functions for efficient setting
and clearing of a range of bits in a bitmap:

set_bits(char *buf, size_t start, size_t end);
clear_bits(char *buf, size_t start, size_t end);

The whole branch is still sitting on my disk, because I based it on
my other patch series (generic handling of multi-page exclusions), and
there were quite some conflicts when I tried to rebase it on devel.

But if you include it in v1.5.6, I will do the rebase nevertheless,
so it doesn't have to wait for your decision about the other series.

Can we do it that way?

Petr T