Skip to contentSkip to conversation

net: mvneta_bm: fix gen_pool_free address for BPPI

PATCH In review Chenguang Zhao 2 participants 4 messages Aug 5 – Aug 6

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net: mvneta_bm: fix gen_pool_free address for BPPI

From: Chenguang Zhao <[email protected]>

gen_pool_free() expects the virtual address returned by gen_pool_dma_alloc(), not the physical address. Passing phys can miss the chunk and trigger BUG() on remove or probe rollback.

Fixes dc35a10f68d3 ("net: mvneta: bm: add support for hardware buffer management")
---
 drivers/net/ethernet/marvell/mvneta_bm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/marvell/mvneta_bm.c +1 -1
index 6bb380494919..2a114a69a0ab 100644
@@ -389,7 +389,7 @@ static int mvneta_bm_get_sram(struct device_node *dn,
 
 static void mvneta_bm_put_sram(struct mvneta_bm *priv)
 {
-	gen_pool_free(priv->bppi_pool, priv->bppi_phys_addr,
+	gen_pool_free(priv->bppi_pool, (unsigned long)priv->bppi_virt_addr,
 		      MVNETA_BM_BPPI_SIZE);
 }
 
Signature · 1 line
2.25.1
#1 lore raw
Quoted 17 lines · Chenguang Zhao

From: Chenguang Zhao <[email protected]>

gen_pool_free() expects the virtual address returned by gen_pool_dma_alloc(), not the physical address. Passing phys can miss the chunk and trigger BUG() on remove or probe rollback.

Fixes dc35a10f68d3 ("net: mvneta: bm: add support for hardware buffer management")
---
 drivers/net/ethernet/marvell/mvneta_bm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/marvell/mvneta_bm.c +0 -0
index 6bb380494919..2a114a69a0ab 100644

drivers/net/ethernet/marvell/mvneta_bm.c static int mvneta_bm_get_sram(struct device_node *dn, · line 392

Quoted 5 lines
drivers/net/ethernet/marvell/mvneta_bm.c +1 -1
@@ -389,7 +389,7 @@ static int mvneta_bm_get_sram(struct device_node *dn,
 
 static void mvneta_bm_put_sram(struct mvneta_bm *priv)
 {
-	gen_pool_free(priv->bppi_pool, priv->bppi_phys_addr,
+	gen_pool_free(priv->bppi_pool, (unsigned long)priv->bppi_virt_addr,

The cast is ugly, but it also seems correct. However, can the API be improved. If the intention is

gen_pool_free(priv->bppi_pool,

              gen_pool_dma_alloc(priv->bppi_pool,
                                 MVNETA_BM_BPPI_SIZE,
                                 &priv->bppi_phys_addr),
              MVNETA_BM_BPPI_SIZE);

maybe gen_pool_free() should be changed to take a void *?

Andrew

#2 lore raw
Quoted 27 lines · Andrew Lunn
Quoted 17 lines · Chenguang Zhao

From: Chenguang Zhao <[email protected]>

gen_pool_free() expects the virtual address returned by gen_pool_dma_alloc(), not the physical address. Passing phys can miss the chunk and trigger BUG() on remove or probe rollback.

Fixes dc35a10f68d3 ("net: mvneta: bm: add support for hardware buffer management")
---
 drivers/net/ethernet/marvell/mvneta_bm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/marvell/mvneta_bm.c +0 -0
index 6bb380494919..2a114a69a0ab 100644

drivers/net/ethernet/marvell/mvneta_bm.c static int mvneta_bm_get_sram(struct device_node *dn, · line 392

Quoted 5 lines
drivers/net/ethernet/marvell/mvneta_bm.c +1 -1
@@ -389,7 +389,7 @@ static int mvneta_bm_get_sram(struct device_node *dn,
 
 static void mvneta_bm_put_sram(struct mvneta_bm *priv)
 {
-	gen_pool_free(priv->bppi_pool, priv->bppi_phys_addr,
+	gen_pool_free(priv->bppi_pool, (unsigned long)priv->bppi_virt_addr,

The cast is ugly, but it also seems correct. However, can the API be improved. If the intention is

gen_pool_free(priv->bppi_pool,

Quoted 8 lines
              gen_pool_dma_alloc(priv->bppi_pool,
                                 MVNETA_BM_BPPI_SIZE,
                                 &priv->bppi_phys_addr),
              MVNETA_BM_BPPI_SIZE);

maybe gen_pool_free() should be changed to take a void *?

Andrew

Thanks for the suggestion.

The cast is indeed a bit ugly, but several other drivers free memory from gen_pool_dma_alloc() in the same way today. Changing gen_pool_free() itself to take a void * would touch quite a few call sites, including ones that use genalloc for non-pointer cookies, so that may be a larger change than we want for this fix.

If we do want to improve the API, perhaps a small helper would be enough, for example:

static inline void gen_pool_dma_free(struct gen_pool *pool, void *vaddr, size_t size)
{
    gen_pool_free(pool, (unsigned long)vaddr, size);
}

Does that sound reasonable to you? I'm happy to keep this bugfix as-is

for now, or follow up with such a helper if preferred.

Chenguang

#3 lore raw
Quoted 9 lines
Quoted 9 lines

The cast is ugly, but it also seems correct. However, can the API be improved. If the intention is

gen_pool_free(priv->bppi_pool,

              gen_pool_dma_alloc(priv->bppi_pool,
                                 MVNETA_BM_BPPI_SIZE,
                                 &priv->bppi_phys_addr),
              MVNETA_BM_BPPI_SIZE);
Quoted 7 lines
Quoted 3 lines

maybe gen_pool_free() should be changed to take a void *?

Andrew

Thanks for the suggestion.

The cast is indeed a bit ugly, but several other drivers free memory from gen_pool_dma_alloc() in the same way today.

Flip it around. Are there many calls which don't have the cast?

Quoted 4 lines

Changing gen_pool_free()
itself to take a void * would touch quite a few call sites, including ones that use genalloc for non-pointer cookies, so that may be a larger change than we want for this fix.

I agree it is more than just the fix here. But this fix is adding yet another instance where maybe bad design is forcing a cast.

It should also be quite a mechanical change. Can Coccinelle do the change needed?

Andrew