[6.6][Bugfix] mm: readahead: fix potential max_order overflow in select_new_order · Pull Request !7628 · anolis/cloud-kernel

bugzilla: https://bugzilla.openanolis.cn/show_bug.cgi?id=36706

In page_cache_ra_order(), ilog2(ra->size) is passed as the max_order to select_new_order(). When ra->size is 0, ilog2() returns -1, which causes `orders &= BIT(max_order + 1) - 1` to evaluate BIT(0) - 1 = 0, zeroing orders and triggering the VM_WARN_ON(!orders) immediately below.

Clamp max_order to [0, MAX_PAGECACHE_ORDER] before use, falling back to min(old_order, MAX_PAGECACHE_ORDER) when out of range. This is the 6.6 equivalent of commit 93f8daa4cae2 ("anolis: mm: readahead: fix potential max_order overflow") already on devel-7.0.

## Test

Static verification: the condition `max_order < 0 || max_order > MAX_PAGECACHE_ORDER` triggers exactly when ilog2(ra->size) returns -1 (ra->size==0) or wraps to a large value, and the clamp restores a safe order. No new codepath is introduced; existing readahead behaviour is unchanged for valid max_order values.

Fixes: d5282eb85aa0 ("mm: mTHP user controls to configure pagecache large folio sizes")

类型	指派人员	状态
审查	zhongjiang vxuyu baolin.wang	已完成 (0/0人)
测试	小龙	进行中 (0/1人)

GVPanolis/cloud-kernel

搜索帮助

GVP anolis/cloud-kernel