SQL> select sql_text,executions,to_char(buffer_gets) from v$sql where buffer_gets <0;

SQL_TEXT
------------------------------------------------------------------------------------------------------------------------
EXECUTIONS TO_CHAR(BUFFER_GETS)
---------- ----------------------------------------
select count(*) as col_0_0_ from HS_forummessage forummessa0_ where (forummessa0_.forumID=:1 )
205681 -1321805203

select count(info0_.NUMINFOGUID) as col_0_0_ from HS_INFO info0_, HS_C3_LEVEL level1_ where info0_.INTLEVEL=level1_.NUML
EVELID and info0_.INTENABLEDFLAG=1 and info0_.INTPUBLISHSTATE=1 and info0_.DATPUBLISHDATE<=:1 and info0_.NUMCATALOGGUID=
:2 and (level1_.INTLEVEL in (1 , 2 , 3 , 4 , 5)) order by info0_.DATPUBLISHDATE DESC, info0_.NUMORDER DESC
14718967 -1261787636

select count(info0_.NUMINFOGUID) as col_0_0_ from HS_INFO info0_ where info0_.INTENABLEDFLAG=1 and info0_.INTPUBLISHSTAT
E=1 and info0_.DATPUBLISHDATE<=:1 and (info0_.NUMCATALOGGUID in (select catalog1_.NUMCATALOGGUID from HS_CATALOG catalog
1_ where catalog1_.INTCATALOGTYPE=1 and catalog1_.NUMAPPGUID=:2)) and (info0_.VC2INDEXWORDS like :3)
359138 -1950604219

也就是说v$sql视图中的buffer_gets最多能表示:

SQL> select power(2,32)/2 from dual;

POWER(2,32)/2
-------------
2147483648

基本上超过了2 billion，就要转为负数了。

那么这是不是一个Bug呢？
搜索Metalink可以发现很多相关的Bug报告，例如bug no:5689640

但是最终Oracle不认为这是一个Bug，因为即使底层表x$kglob的相关值定义从KQFCINT() 变更位 KQFCUIN()。也只不过将这个限制扩大了一倍而已，这个限制早晚还会被达到。
所以，这个修正请求被Oracle定义位增强而不是Bug。据说在Oracle11g中这个增强被加入。

http://www.eygle.com/archives/2007/08/vsql_buffer_gets_negative.html