| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <fdc778be-907a-49bd-bf10-086f45716181@huawei.com>
Date: Mon, 22 Jul 2024 20:55:27 +0800
From: Yunsheng Lin <linyunsheng@...wei.com>
To: Alexander H Duyck <alexander.duyck@...il.com>, <davem@...emloft.net>,
<kuba@...nel.org>, <pabeni@...hat.com>
CC: <netdev@...r.kernel.org>, <linux-kernel@...r.kernel.org>, Andrew Morton
<akpm@...ux-foundation.org>, <linux-mm@...ck.org>
Subject: Re: [RFC v11 08/14] mm: page_frag: some minor refactoring before
adding new API
On 2024/7/22 7:40, Alexander H Duyck wrote:
> On Fri, 2024-07-19 at 17:33 +0800, Yunsheng Lin wrote:
>> Refactor common codes from __page_frag_alloc_va_align()
>> to __page_frag_cache_refill(), so that the new API can
>> make use of them.
>>
>> CC: Alexander Duyck <alexander.duyck@...il.com>
>> Signed-off-by: Yunsheng Lin <linyunsheng@...wei.com>
>> ---
>> include/linux/page_frag_cache.h | 2 +-
>> mm/page_frag_cache.c | 93 +++++++++++++++++----------------
>> 2 files changed, 49 insertions(+), 46 deletions(-)
>>
>> diff --git a/include/linux/page_frag_cache.h b/include/linux/page_frag_cache.h
>> index 12a16f8e8ad0..5aa45de7a9a5 100644
>> --- a/include/linux/page_frag_cache.h
>> +++ b/include/linux/page_frag_cache.h
>> @@ -50,7 +50,7 @@ static inline void *encoded_page_address(unsigned long encoded_va)
>>
>> static inline void page_frag_cache_init(struct page_frag_cache *nc)
>> {
>> - nc->encoded_va = 0;
>> + memset(nc, 0, sizeof(*nc));
>> }
>>
>
> I do not like requiring the entire structure to be reset as a part of
> init. If encoded_va is 0 then we have reset the page and the flags.
> There shouldn't be anything else we need to reset as remaining and bias
> will be reset when we reallocate.
The argument is about aoviding one checking for fast path by doing the
memset in the slow path, which you might already know accroding to your
comment in previous version.
It is just sometimes hard to understand your preference for maintainability
over performance here as sometimes your comment seems to perfer performance
over maintainability, like the LEA trick you mentioned and offset count-down
before this patchset. It would be good to be more consistent about this,
otherwise it is sometimes confusing when doing the refactoring.
>
>> static inline bool page_frag_cache_is_pfmemalloc(struct page_frag_cache *nc)
>> diff --git a/mm/page_frag_cache.c b/mm/page_frag_cache.c
>> index 7928e5d50711..d9c9cad17af7 100644
>> --- a/mm/page_frag_cache.c
>> +++ b/mm/page_frag_cache.c
>> @@ -19,6 +19,28 @@
>> #include <linux/page_frag_cache.h>
>> #include "internal.h"
>>
>> +static struct page *__page_frag_cache_recharge(struct page_frag_cache *nc)
>> +{
>> + unsigned long encoded_va = nc->encoded_va;
>> + struct page *page;
>> +
>> + page = virt_to_page((void *)encoded_va);
>> + if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
>> + return NULL;
>> +
>> + if (unlikely(encoded_page_pfmemalloc(encoded_va))) {
>> + VM_BUG_ON(compound_order(page) !=
>> + encoded_page_order(encoded_va));
>> + free_unref_page(page, encoded_page_order(encoded_va));
>> + return NULL;
>> + }
>> +
>> + /* OK, page count is 0, we can safely set it */
>> + set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
>> +
>> + return page;
>> +}
>> +
>> static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
>> gfp_t gfp_mask)
>> {
>> @@ -26,6 +48,14 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
>> struct page *page = NULL;
>> gfp_t gfp = gfp_mask;
>>
>> + if (likely(nc->encoded_va)) {
>> + page = __page_frag_cache_recharge(nc);
>> + if (page) {
>> + order = encoded_page_order(nc->encoded_va);
>> + goto out;
>> + }
>> + }
>> +
>
> This code has no business here. This is refill, you just dropped
> recharge in here which will make a complete mess of the ordering and be
> confusing to say the least.
>
> The expectation was that if we are calling this function it is going to
> overwrite the virtual address to NULL on failure so we discard the old
> page if there is one present. This changes that behaviour. What you
> effectively did is made __page_frag_cache_refill into the recharge
> function.
The idea is to reuse the below for both __page_frag_cache_refill() and
__page_frag_cache_recharge(), which seems to be about maintainability
to not having duplicated code. If there is a better idea to avoid that
duplicated code while keeping the old behaviour, I am happy to change
it.
/* reset page count bias and remaining to start of new frag */
nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
nc->remaining = PAGE_SIZE << order;
>
>> #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
>> gfp_mask = (gfp_mask & ~__GFP_DIRECT_RECLAIM) | __GFP_COMP |
>> __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC;
>> @@ -35,7 +65,7 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
>> if (unlikely(!page)) {
>> page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
>> if (unlikely(!page)) {
>> - nc->encoded_va = 0;
>> + memset(nc, 0, sizeof(*nc));
>> return NULL;
>> }
>>
>
> The memset will take a few more instructions than the existing code
> did. I would prefer to keep this as is if at all possible.
It will not take more instructions for arm64 as it has 'stp' instruction for
__HAVE_ARCH_MEMSET is set.
There is something similar for x64?
>
>> @@ -45,6 +75,16 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
>> nc->encoded_va = encode_aligned_va(page_address(page), order,
>> page_is_pfmemalloc(page));
>>
>> + /* Even if we own the page, we do not use atomic_set().
>> + * This would break get_page_unless_zero() users.
>> + */
>> + page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
>> +
>> +out:
>> + /* reset page count bias and remaining to start of new frag */
>> + nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
>> + nc->remaining = PAGE_SIZE << order;
>> +
>> return page;
>> }
>>
>
> Why bother returning a page at all? It doesn't seem like you don't use
> it anymore. It looks like the use cases you have for it in patch 11/12
> all appear to be broken from what I can tell as you are adding page as
> a variable when we don't need to be passing internal details to the
> callers of the function when just a simple error return code would do.
It would be good to be more specific about the 'broken' part here.
Powered by blists - more mailing lists