2023-06-13 07:31:27

by Sumitra Sharma

[permalink] [raw]
Subject: [PATCH v2] lib/test_bpf: Call page_address() on page acquired with GFP_KERNEL flag

generate_test_data() acquires a page with alloc_page(GFP_KERNEL). Pages
allocated with GFP_KERNEL cannot come from Highmem. This is why
there is no need to call kmap() on them.

Therefore, use a plain page_address() on that page.

Suggested-by: Fabio M. De Francesco <[email protected]>
Signed-off-by: Sumitra Sharma <[email protected]>
---

Changes in v2:
- Remove the kmap() call and call page_address() instead.
- Change the commit subject and message.

lib/test_bpf.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/lib/test_bpf.c b/lib/test_bpf.c
index ade9ac672adb..70fcd0bcf14b 100644
--- a/lib/test_bpf.c
+++ b/lib/test_bpf.c
@@ -14388,11 +14388,10 @@ static void *generate_test_data(struct bpf_test *test, int sub)
if (!page)
goto err_kfree_skb;

- ptr = kmap(page);
+ ptr = page_address(page);
if (!ptr)
goto err_free_page;
memcpy(ptr, test->frag_data, MAX_DATA);
- kunmap(page);
skb_add_rx_frag(skb, 0, page, 0, MAX_DATA, MAX_DATA);
}

--
2.25.1



2023-06-18 06:32:09

by Fabio M. De Francesco

[permalink] [raw]
Subject: Re: [PATCH v2] lib/test_bpf: Call page_address() on page acquired with GFP_KERNEL flag

On marted? 13 giugno 2023 09:17:56 CEST Sumitra Sharma wrote:
> generate_test_data() acquires a page with alloc_page(GFP_KERNEL). Pages
> allocated with GFP_KERNEL cannot come from Highmem. This is why
> there is no need to call kmap() on them.
>
> Therefore, use a plain page_address() on that page.
>
> Suggested-by: Fabio M. De Francesco <[email protected]>
> Signed-off-by: Sumitra Sharma <[email protected]>
> ---
>
> Changes in v2:
> - Remove the kmap() call and call page_address() instead.

NIT: Give credit to whom asked you for this removal and explain why the
mapping is not required.

> - Change the commit subject and message.
>
> lib/test_bpf.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/lib/test_bpf.c b/lib/test_bpf.c
> index ade9ac672adb..70fcd0bcf14b 100644
> --- a/lib/test_bpf.c
> +++ b/lib/test_bpf.c
> @@ -14388,11 +14388,10 @@ static void *generate_test_data(struct bpf_test
> *test, int sub) if (!page)
> goto err_kfree_skb;
>
> - ptr = kmap(page);
> + ptr = page_address(page);
> if (!ptr)
> goto err_free_page;

What is the reason of this test? Could "ptr" ever be NULL? What is the code
checking just few lines above this latter test?

Please, take a deeper look at this function as a whole.

Fabio

> memcpy(ptr, test->frag_data, MAX_DATA);
> - kunmap(page);
> skb_add_rx_frag(skb, 0, page, 0, MAX_DATA, MAX_DATA);
> }
>
> --
> 2.25.1





2023-06-22 06:15:47

by Sumitra Sharma

[permalink] [raw]
Subject: Re: [PATCH v2] lib/test_bpf: Call page_address() on page acquired with GFP_KERNEL flag

On Sun, Jun 18, 2023 at 07:07:00AM +0200, Fabio M. De Francesco wrote:
> On marted? 13 giugno 2023 09:17:56 CEST Sumitra Sharma wrote:
> > generate_test_data() acquires a page with alloc_page(GFP_KERNEL). Pages
> > allocated with GFP_KERNEL cannot come from Highmem. This is why
> > there is no need to call kmap() on them.
> >
> > Therefore, use a plain page_address() on that page.
> >
> > Suggested-by: Fabio M. De Francesco <[email protected]>
> > Signed-off-by: Sumitra Sharma <[email protected]>
> > ---
> >
> > Changes in v2:
> > - Remove the kmap() call and call page_address() instead.
>
> NIT: Give credit to whom asked you for this removal and explain why the
> mapping is not required.

>
> > - Change the commit subject and message.
> >
> > lib/test_bpf.c | 3 +--
> > 1 file changed, 1 insertion(+), 2 deletions(-)
> >
> > diff --git a/lib/test_bpf.c b/lib/test_bpf.c
> > index ade9ac672adb..70fcd0bcf14b 100644
> > --- a/lib/test_bpf.c
> > +++ b/lib/test_bpf.c
> > @@ -14388,11 +14388,10 @@ static void *generate_test_data(struct bpf_test
> > *test, int sub) if (!page)
> > goto err_kfree_skb;
> >
> > - ptr = kmap(page);
> > + ptr = page_address(page);
> > if (!ptr)
> > goto err_free_page;
>
> What is the reason of this test? Could "ptr" ever be NULL? What is the code
> checking just few lines above this latter test?
>


The code is allocating a page using alloc_page() with the GFP_KERNEL flag to
obtain a kernel page frame. The checks if (!page) and if (!ptr) are verifying
if the page allocation or the mapping operation were successful.

If the pages obtained through page_address() are not from the highmem zone,
the page_address() function will always return a valid kernel virtual address
and will not return NULL. Hence, the check !ptr can be ignored while the !page
must remain.

I will be working on the v2 patch.

I will also add the credits and add new lines to commit message
explaining why the mapping is not required.

Thank you for the help.

Thanks & regards
Sumitra

> Please, take a deeper look at this function as a whole.
>
> Fabio
>
> > memcpy(ptr, test->frag_data, MAX_DATA);
> > - kunmap(page);
> > skb_add_rx_frag(skb, 0, page, 0, MAX_DATA, MAX_DATA);
> > }
> >
> > --
> > 2.25.1
>
>
>
>