Skip to contentSkip to conversation

atl1e:fix 2.6.31-git4 -- ATL1E 0000:03:00.0: DMA-API: device driver frees DMA

PATCH In review [email protected] 4 participants 7 messages Sep 16, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 atl1e:fix 2.6.31-git4 -- ATL1E 0000:03:00.0: DMA-API: device driver frees DMA

[   25.059969] WARNING: at lib/dma-debug.c:816 check_unmap+0x383/0x55c()
[   25.059976] Hardware name: 1000HE
[   25.059984] ATL1E 0000:03:00.0: DMA-API: device driver frees DMA
    memory with wrong function [device address=0x0000000036b92802] 
    [size=90 bytes] [mapped as single] [unmapped as page]

use the wrong API when free dma. So when map dma use a flag to demostrate whether it is 'pci_map_single' or 'pci_map_page'. When free the dma, check the flags to select the right APIs('pci_unmap_single' or 'pci_unmap_page').

---

drivers/net/atl1e/atl1e.h +9 -0
index ba48220..d63be5c 100644
@@ -377,10 +377,19 @@ struct atl1e_hw {
  */
 struct atl1e_tx_buffer {
    struct sk_buff *skb;
+   unsigned long flags;
+#define ATL1E_TX_PCIMAP_SINGLE     0x0001
+#define ATL1E_TX_PCIMAP_PAGE       0x0002
+#define ATL1E_TX_PCIMAP_TYPE_MASK  0x3
    u16 length;
    dma_addr_t dma;
 };
 
+#define ATL1E_SET_PCIMAP_TYPE(tx_buff, type) do {      \
+   ((tx_buff)->flags) &= ~ATL1E_TX_PCIMAP_TYPE_MASK;   \
+   ((tx_buff)->flags) |= (type);               \
+   } while (0)
+
 struct atl1e_rx_page {
    dma_addr_t  dma;    /* receive rage DMA address */
    u8      *addr;   /* receive rage virtual address */
drivers/net/atl1e/atl1e_main.c +13 -2
index 69b830f..955da73 100644
@@ -635,7 +635,11 @@ static void atl1e_clean_tx_ring(struct atl1e_adapter *adapter)
    for (index = 0; index < ring_count; index++) {
        tx_buffer = &tx_ring->tx_buffer[index];
        if (tx_buffer->dma) {
-           pci_unmap_page(pdev, tx_buffer->dma,
+           if (tx_buffer->flags & ATL1E_TX_PCIMAP_SINGLE)
+               pci_unmap_single(pdev, tx_buffer->dma,
+                   tx_buffer->length, PCI_DMA_TODEVICE);
+           else if (tx_buffer->flags & ATL1E_TX_PCIMAP_PAGE)
+               pci_unmap_page(pdev, tx_buffer->dma,
                    tx_buffer->length, PCI_DMA_TODEVICE);
            tx_buffer->dma = 0;
        }
@@ -1220,7 +1224,11 @@ static bool atl1e_clean_tx_irq(struct atl1e_adapter *adapter)
    while (next_to_clean != hw_next_to_clean) {
        tx_buffer = &tx_ring->tx_buffer[next_to_clean];
        if (tx_buffer->dma) {
-           pci_unmap_page(adapter->pdev, tx_buffer->dma,
+           if (tx_buffer->flags & ATL1E_TX_PCIMAP_SINGLE)
+               pci_unmap_single(adapter->pdev, tx_buffer->dma,
+                   tx_buffer->length, PCI_DMA_TODEVICE);
+           else if (tx_buffer->flags & ATL1E_TX_PCIMAP_PAGE)
+               pci_unmap_page(adapter->pdev, tx_buffer->dma,
                    tx_buffer->length, PCI_DMA_TODEVICE);
            tx_buffer->dma = 0;
        }
@@ -1741,6 +1749,7 @@ static void atl1e_tx_map(struct atl1e_adapter *adapter,
        tx_buffer->length = map_len;
        tx_buffer->dma = pci_map_single(adapter->pdev,
                    skb->data, hdr_len, PCI_DMA_TODEVICE);
+       ATL1E_SET_PCIMAP_TYPE(tx_buffer, ATL1E_TX_PCIMAP_SINGLE);
        mapped_len += map_len;
        use_tpd->buffer_addr = cpu_to_le64(tx_buffer->dma);
        use_tpd->word2 = (use_tpd->word2 & (~TPD_BUFLEN_MASK)) |
@@ -1766,6 +1775,7 @@ static void atl1e_tx_map(struct atl1e_adapter *adapter,
        tx_buffer->dma =
            pci_map_single(adapter->pdev, skb->data + mapped_len,
                    map_len, PCI_DMA_TODEVICE);
+       ATL1E_SET_PCIMAP_TYPE(tx_buffer, ATL1E_TX_PCIMAP_SINGLE);
        mapped_len  += map_len;
        use_tpd->buffer_addr = cpu_to_le64(tx_buffer->dma);
        use_tpd->word2 = (use_tpd->word2 & (~TPD_BUFLEN_MASK)) |
@@ -1801,6 +1811,7 @@ static void atl1e_tx_map(struct atl1e_adapter *adapter,
                        (i * MAX_TX_BUF_LEN),
                        tx_buffer->length,
                        PCI_DMA_TODEVICE);
+           ATL1E_SET_PCIMAP_TYPE(tx_buffer, ATL1E_TX_PCIMAP_PAGE);
            use_tpd->buffer_addr = cpu_to_le64(tx_buffer->dma);
            use_tpd->word2 = (use_tpd->word2 & (~TPD_BUFLEN_MASK)) |
                    ((cpu_to_le32(tx_buffer->length) &
maintainer
#1 lore raw

From: <[email protected]>
Date: Wed, 16 Sep 2009 14:21:53 +0800

Quoted 7 lines
[   25.059969] WARNING: at lib/dma-debug.c:816 check_unmap+0x383/0x55c()
[   25.059976] Hardware name: 1000HE
[   25.059984] ATL1E 0000:03:00.0: DMA-API: device driver frees DMA
    memory with wrong function [device address=0x0000000036b92802] 
    [size=90 bytes] [mapped as single] [unmapped as page]
Quoted 5 lines

use the wrong API when free dma. So when map dma use a flag to demostrate whether it is 'pci_map_single' or 'pci_map_page'. When free the dma, check the flags to select the right APIs('pci_unmap_single' or 'pci_unmap_page').

An 'unsigned long' is an enormous type to use just to store 3 bits of information. Use a u16 or similar to compact the size of struct atl1e_tx_buffer

Also, it looks terribly ugle to define the flag macros, some with three leading zeros in the hexadecimal values and one without.

Please make them consistent, thanks.

Generally, I see that this change was put together very hastily and without very much care.

#2 lore raw
In reply to [email protected]

drivers/net/atl1e/atl1e_main.c static void atl1e_clean_tx_ring(struct atl1e_adapter *adapter) · line 642

Quoted 6 lines · On Wed, 2009-09-16 at 14:21 +0800, [email protected] wrote:
drivers/net/atl1e/atl1e_main.c +5 -1
-           pci_unmap_page(pdev, tx_buffer->dma,
+           if (tx_buffer->flags & ATL1E_TX_PCIMAP_SINGLE)
+               pci_unmap_single(pdev, tx_buffer->dma,
+                   tx_buffer->length, PCI_DMA_TODEVICE);
+           else if (tx_buffer->flags & ATL1E_TX_PCIMAP_PAGE)
+               pci_unmap_page(pdev, tx_buffer->dma,

Could you run this through checkpatch, and fix any errors if find? It looks like you uses spaces for tabs in the code above, maybe in the other block too..

Daniel

author
#3 lore raw
Quoted 8 lines · Daniel Walker

drivers/net/atl1e/atl1e_main.c static void atl1e_clean_tx_ring(struct atl1e_adapter *adapter) · line 642

Quoted 6 lines · On Wed, 2009-09-16 at 14:21 +0800, [email protected] wrote:
drivers/net/atl1e/atl1e_main.c +5 -1
-           pci_unmap_page(pdev, tx_buffer->dma,
+           if (tx_buffer->flags & ATL1E_TX_PCIMAP_SINGLE)
+               pci_unmap_single(pdev, tx_buffer->dma,
+                   tx_buffer->length, PCI_DMA_TODEVICE);
+           else if (tx_buffer->flags & ATL1E_TX_PCIMAP_PAGE)
+               pci_unmap_page(pdev, tx_buffer->dma,
Quoted 7 lines

Could you run this through checkpatch, and fix any errors if find? It looks like you uses spaces for tabs in the code above, maybe in the other block too..

Daniel

ou, my mistake, just resend.

Best wishes
jie

maintainer
#4 lore raw
In reply to Daniel Walker

From: Daniel Walker <[email protected]>
Date: Wed, 16 Sep 2009 02:10:51 -0700

Quoted 8 lines · On Wed, 2009-09-16 at 14:21 +0800, [email protected] wrote:

drivers/net/atl1e/atl1e_main.c static void atl1e_clean_tx_ring(struct atl1e_adapter *adapter) · line 642

Quoted 6 lines
drivers/net/atl1e/atl1e_main.c +5 -1
-           pci_unmap_page(pdev, tx_buffer->dma,
+           if (tx_buffer->flags & ATL1E_TX_PCIMAP_SINGLE)
+               pci_unmap_single(pdev, tx_buffer->dma,
+                   tx_buffer->length, PCI_DMA_TODEVICE);
+           else if (tx_buffer->flags & ATL1E_TX_PCIMAP_PAGE)
+               pci_unmap_page(pdev, tx_buffer->dma,
Quoted 4 lines

Could you run this through checkpatch, and fix any errors if find? It looks like you uses spaces for tabs in the code above, maybe in the other block too..

It's because of his email client, it corrupted the whole patch like that.

#5 lore raw

[David Miller - Wed, Sep 16, 2009 at 02:47:59AM -0700]

| From: Daniel Walker <[email protected]>
| Date: Wed, 16 Sep 2009 02:10:51 -0700
| 
| > On Wed, 2009-09-16 at 14:21 +0800, [email protected] wrote:
| >> -           pci_unmap_page(pdev, tx_buffer->dma,
| >> +           if (tx_buffer->flags & ATL1E_TX_PCIMAP_SINGLE)
| >> +               pci_unmap_single(pdev, tx_buffer->dma,
| >> +                   tx_buffer->length, PCI_DMA_TODEVICE);
| >> +           else if (tx_buffer->flags & ATL1E_TX_PCIMAP_PAGE)
| >> +               pci_unmap_page(pdev, tx_buffer->dma,
| > 
| > Could you run this through checkpatch, and fix any errors if find? It
| > looks like you uses spaces for tabs in the code above, maybe in the
| > other block too..
| > 
| 
| It's because of his email client, it corrupted the whole patch
| like that.

well, client seems to be

| X-Mailer: git-send-email 1.5.2.2

more probably the text editor is the reason.

-- Cyrill

author
#6 lore raw
Quoted 22 lines · Cyrill Gorcunov

[David Miller - Wed, Sep 16, 2009 at 02:47:59AM -0700]

| From: Daniel Walker <[email protected]>
| Date: Wed, 16 Sep 2009 02:10:51 -0700
|
| > On Wed, 2009-09-16 at 14:21 +0800, [email protected] wrote:
| >> -           pci_unmap_page(pdev, tx_buffer->dma,
| >> +           if (tx_buffer->flags & ATL1E_TX_PCIMAP_SINGLE)
| >> +               pci_unmap_single(pdev, tx_buffer->dma,
| >> +                   tx_buffer->length, PCI_DMA_TODEVICE);
| >> +           else if (tx_buffer->flags & ATL1E_TX_PCIMAP_PAGE)
| >> +               pci_unmap_page(pdev, tx_buffer->dma,
| >
| > Could you run this through checkpatch, and fix any errors
if find?
| > It looks like you uses spaces for tabs in the code above,
maybe in
| > the other block too..
| >
|
| It's because of his email client, it corrupted the whole patch like
| that.
Quoted 8 lines

well, client seems to be

| X-Mailer: git-send-email 1.5.2.2

more probably the text editor is the reason.

-- Cyrill

yes, the editor 'vi' have configed "set expandtab".