On 2020-11-25 22:28:58 [+0800], zqiu2000@xxxxxxx wrote:
> From: Zanxiong Qiu <zqiu2000@xxxxxxx>
>
> get_cpu_var() calls preempt_disable(), while on RT kernel,
> pagevec_lru_move_fn() will call spinlock and might schedule
> the context out and hence the schedule bug occurred, issue
> is found on 5.4.70-rt40 and reproducable on 5.4.74-rt41.
>
…
> 2154a0abcc9 ("mm: Revert the DEFINE_PER_CPU_PAGEVEC implementation")
> reverted the lock/unlock_swap_pvec function, however, get_cpu_var()
> was added back in deactivate_page(), actually, get_locked_var()
> shall be used instead to avoid preempt_disable() call for RT.
The commit is
32154a0abcc97 ("mm: Revert the DEFINE_PER_CPU_PAGEVEC implementation")
the reasoning is correct. deactivate_page() was added in v5.4 and I
missed that part when I added back the old patches which did not handle
it :/
Acked-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
> Signed-off-by: Zanxiong Qiu <zqiu2000@xxxxxxx>
> ---
> mm/swap.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/mm/swap.c b/mm/swap.c
> index cdb4f1fa3a48..463cac334fcf 100644
> --- a/mm/swap.c
> +++ b/mm/swap.c
> @@ -666,12 +666,13 @@ void deactivate_file_page(struct page *page)
> void deactivate_page(struct page *page)
> {
> if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
> - struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
> + struct pagevec *pvec = &get_locked_var(swapvec_lock,
> + lru_deactivate_pvecs);
>
> get_page(page);
> if (!pagevec_add(pvec, page) || PageCompound(page))
> pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
> - put_cpu_var(lru_deactivate_pvecs);
> + put_locked_var(swapvec_lock, lru_deactivate_pvecs);
> }
> }
>
Sebastian
