This patch revert commit c83900393aa1 ("tee: Remove vmalloc page support")

The firmware framework uses vmalloc page to store an image of a firmware,
got from the file system.
To be able to give this firmware to OP-TEE without an extra copy,
the vmalloc page support needs to be reintroduce.

Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
---
 drivers/tee/tee_shm.c | 24 ++++++++++++++++++------
 1 file changed, 18 insertions(+), 6 deletions(-)

diff --git a/drivers/tee/tee_shm.c b/drivers/tee/tee_shm.c
index 673cf0359494..b2d349ac17b4 100644
--- a/drivers/tee/tee_shm.c
+++ b/drivers/tee/tee_shm.c
@@ -28,14 +28,26 @@ static int shm_get_kernel_pages(unsigned long start, size_t page_count,
 	struct page *page;
 	size_t n;
 
-	if (WARN_ON_ONCE(is_vmalloc_addr((void *)start) ||
-			 is_kmap_addr((void *)start)))
+	if (WARN_ON_ONCE(is_kmap_addr((void *)start)))
 		return -EINVAL;
 
-	page = virt_to_page((void *)start);
-	for (n = 0; n < page_count; n++) {
-		pages[n] = page + n;
-		get_page(pages[n]);
+	if (is_vmalloc_addr((void *)start)) {
+		struct page *page;
+
+		for (n = 0; n < page_count; n++) {
+			page = vmalloc_to_page((void *)(start + PAGE_SIZE * n));
+			if (!page)
+				return -ENOMEM;
+
+			get_page(page);
+			pages[n] = page;
+		}
+	}  else {
+		page = virt_to_page((void *)start);
+		for (n = 0; n < page_count; n++) {
+			pages[n] = page + n;
+			get_page(pages[n]);
+		}
 	}
 
 	return page_count;
-- 
2.25.1

On Tue, May 23, 2023 at 11:13:47AM +0200, Arnaud Pouliquen wrote:
> This patch revert commit c83900393aa1 ("tee: Remove vmalloc page support")

As per the discussion back then: don't just blindly do the same dumb
thing again and fix the interfae to actually pass in a page array,
or iov_iter or an actually useful container that fits.

Hello Christoph,

On 5/24/23 08:46, Christoph Hellwig wrote:
> On Tue, May 23, 2023 at 11:13:47AM +0200, Arnaud Pouliquen wrote:
>> This patch revert commit c83900393aa1 ("tee: Remove vmalloc page support")
> 
> As per the discussion back then: don't just blindly do the same dumb
> thing again and fix the interfae to actually pass in a page array,
> or iov_iter or an actually useful container that fits.
> 

I suppose your are speaking about this discussion:
https://lore.kernel.org/all/20221002002326.946620-3-ira.weiny@intel.com/

If I'm not mistaken, I should modify at tee_shm_register_kernel_buf API and
register_shm_helper inernal function, right?

Seems that Jens has also pointed out the free part...

What about having equivalent of shm_get_kernel_pages in an external helper (to
defined where to put it), could it be an alternative of the upadate of the
tee_shm API?

Thanks,
Arnaud

On Wed, May 24, 2023 at 04:01:14PM +0200, Arnaud POULIQUEN wrote:
> > As per the discussion back then: don't just blindly do the same dumb
> > thing again and fix the interfae to actually pass in a page array,
> > or iov_iter or an actually useful container that fits.
> > 
> 
> I suppose your are speaking about this discussion:
> https://lore.kernel.org/all/20221002002326.946620-3-ira.weiny@intel.com/

Yes.

> 
> If I'm not mistaken, I should modify at tee_shm_register_kernel_buf API and
> register_shm_helper inernal function, right?
> 

> What about having equivalent of shm_get_kernel_pages in an external helper (to
> defined where to put it), could it be an alternative of the upadate of the
> tee_shm API?

I think the fundamentally right thing is to pass an iov_iter to
register_shm_helper, and then use the new as of 6.3
iov_iter_extract_pages helper to extract the pages from that.  For
the kernel users you can then simply pass down an ITER_BVEC iter
that you can fill with vmalloc pages if you want.

On 5/26/23 14:37, Christoph Hellwig wrote:
> On Wed, May 24, 2023 at 04:01:14PM +0200, Arnaud POULIQUEN wrote:
>>> As per the discussion back then: don't just blindly do the same dumb
>>> thing again and fix the interfae to actually pass in a page array,
>>> or iov_iter or an actually useful container that fits.
>>>
>>
>> I suppose your are speaking about this discussion:
>> https://lore.kernel.org/all/20221002002326.946620-3-ira.weiny@intel.com/
> 
> Yes.
> 
>>
>> If I'm not mistaken, I should modify at tee_shm_register_kernel_buf API and
>> register_shm_helper inernal function, right?
>>
> 
>> What about having equivalent of shm_get_kernel_pages in an external helper (to
>> defined where to put it), could it be an alternative of the upadate of the
>> tee_shm API?
> 
> I think the fundamentally right thing is to pass an iov_iter to
> register_shm_helper, and then use the new as of 6.3
> iov_iter_extract_pages helper to extract the pages from that.  For
> the kernel users you can then simply pass down an ITER_BVEC iter
> that you can fill with vmalloc pages if you want.
> 

Thanks for the advice!

Regards,
Arnaud