On Thu, Aug 15, 2019 at 11:23:16AM -0700, Jonathan Lemon wrote:
On 15 Aug 2019, at 5:13, Ivan Khoronzhuk wrote:
For 64-bit there is no reason to use vmap/vunmap, so use page_address
as it was initially. For 32 bits, in some apps, like in samples
xdpsock_user.c when number of pgs in use is quite big, the kmap
memory can be not enough, despite on this, kmap looks like is
deprecated in such cases as it can block and should be used rather
for dynamic mm.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@xxxxxxxxxx>
---
net/xdp/xdp_umem.c | 36 ++++++++++++++++++++++++++++++------
1 file changed, 30 insertions(+), 6 deletions(-)
diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
index a0607969f8c0..d740c4f8810c 100644
--- a/net/xdp/xdp_umem.c
+++ b/net/xdp/xdp_umem.c
@@ -14,7 +14,7 @@
#include <linux/netdevice.h>
#include <linux/rtnetlink.h>
#include <linux/idr.h>
-#include <linux/highmem.h>
+#include <linux/vmalloc.h>
#include "xdp_umem.h"
#include "xsk_queue.h"
@@ -170,7 +170,30 @@ static void xdp_umem_unmap_pages(struct
xdp_umem *umem)
unsigned int i;
for (i = 0; i < umem->npgs; i++)
- kunmap(umem->pgs[i]);
+ if (PageHighMem(umem->pgs[i]))
+ vunmap(umem->pages[i].addr);
+}
+
+static int xdp_umem_map_pages(struct xdp_umem *umem)
+{
+ unsigned int i;
+ void *addr;
+
+ for (i = 0; i < umem->npgs; i++) {
+ if (PageHighMem(umem->pgs[i]))
+ addr = vmap(&umem->pgs[i], 1, VM_MAP, PAGE_KERNEL);
+ else
+ addr = page_address(umem->pgs[i]);
+
+ if (!addr) {
+ xdp_umem_unmap_pages(umem);
+ return -ENOMEM;
+ }
+
+ umem->pages[i].addr = addr;
+ }
+
+ return 0;
}
You'll want a __xdp_umem_unmap_pages() helper here that takes an
count of the number of pages to unmap, so it can be called from
xdp_umem_unmap_pages() in the normal case, and xdp_umem_map_pages()
in the error case. Otherwise the error case ends up calling
PageHighMem on a null page.
--
Jonathan
Do you mean null address?
If so, then vunmap do nothing if it's null, and addr is null if it's not
assigned... and it's not assigned w/o correct mapping...
If you mean null page, then it is not possible after all they are
pinned above, here: xdp_umem_pin_pages(), thus assigned.
Or I missed smth?
Despite of this, seems like here should be one more patch, adding unpinning page
in error path, but this not related to this change. Will do this in follow up
fix patch, if no objection to my explanation, ofc.
--
Regards,
Ivan Khoronzhuk
