Merge tag 'dma-mapping-6.11-2024-08-10' of git://git.infradead.org/users/hch/dma-mapping

0409cc53 · Linus Torvalds <[email protected]> · 2024-08-10 10:19:05 -07:00

Pull dma-mapping fix from Christoph Hellwig:

 - avoid a deadlock with dma-debug and netconsole (Rik van Riel)

* tag 'dma-mapping-6.11-2024-08-10' of git://git.infradead.org/users/hch/dma-mapping:
  dma-debug: avoid deadlock between dma debug vs printk and netconsole

Commit: 0409cc53c42129cfeeaea610d0ebb91934491dca
Parents: 31b24446 · bd44ca3d

Merge commit. Changes below are compared with the first parent: 31b24446.

Changes

1 files changed · +4 −1

File	Status	Add	Del	Content
kernel/dma/debug.c	modified	+4	−1	View file

kernel/dma/debug.c

View file

diff --git a/kernel/dma/debug.c b/kernel/dma/debug.c
index a6e3792..d570535 100644
--- a/kernel/dma/debug.c
+++ b/kernel/dma/debug.c
@@ -416,8 +416,11 @@ static unsigned long long phys_addr(struct dma_debug_entry *entry)
  * dma_active_cacheline entry to track per event.  dma_map_sg(), on the
  * other hand, consumes a single dma_debug_entry, but inserts 'nents'
  * entries into the tree.
+ *
+ * Use __GFP_NOWARN because the printk from an OOM, to netconsole, could end
+ * up right back in the DMA debugging code, leading to a deadlock.
  */
-static RADIX_TREE(dma_active_cacheline, GFP_ATOMIC);
+static RADIX_TREE(dma_active_cacheline, GFP_ATOMIC | __GFP_NOWARN);
 static DEFINE_SPINLOCK(radix_lock);
 #define ACTIVE_CACHELINE_MAX_OVERLAP ((1 << RADIX_TREE_MAX_TAGS) - 1)
 #define CACHELINE_PER_PAGE_SHIFT (PAGE_SHIFT - L1_CACHE_SHIFT)