diff mbox

[3/4] brcmfmac: rework headroom check in .start_xmit()

Message ID 1498125664-25980-4-git-send-email-arend.vanspriel@broadcom.com (mailing list archive)
State Accepted
Commit 270a6c1f65fe68a28a5d39cd405592c550b496c7
Delegated to: Kalle Valo
Headers show

Commit Message

Arend van Spriel June 22, 2017, 10:01 a.m. UTC
Since commit 9cc4b7cb86cb ("brcmfmac: Make skb header writable
before use") the headroom usage has been fixed. However, the
driver was keeping statistics that got lost. So reworking the
code so we get those driver statistics back for debugging.

Cc: James Hughes <james.hughes@raspberrypi.org>
Reviewed-by: Hante Meuleman <hante.meuleman@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieter-paul.giesberts@broadcom.com>
Reviewed-by: Franky Lin <franky.lin@broadcom.com>
Signed-off-by: Arend van Spriel <arend.vanspriel@broadcom.com>
---
 .../net/wireless/broadcom/brcm80211/brcmfmac/bus.h | 15 ++++++++++++--
 .../wireless/broadcom/brcm80211/brcmfmac/core.c    | 23 +++++++++++++++-------
 .../wireless/broadcom/brcm80211/brcmfmac/sdio.c    | 13 +++++++-----
 3 files changed, 37 insertions(+), 14 deletions(-)

Comments

Kalle Valo June 27, 2017, 2:09 p.m. UTC | #1
Arend van Spriel <arend.vanspriel@broadcom.com> writes:

> Since commit 9cc4b7cb86cb ("brcmfmac: Make skb header writable
> before use") the headroom usage has been fixed. However, the
> driver was keeping statistics that got lost. So reworking the
> code so we get those driver statistics back for debugging.
>
> Cc: James Hughes <james.hughes@raspberrypi.org>
> Reviewed-by: Hante Meuleman <hante.meuleman@broadcom.com>
> Reviewed-by: Pieter-Paul Giesberts <pieter-paul.giesberts@broadcom.com>
> Reviewed-by: Franky Lin <franky.lin@broadcom.com>
> Signed-off-by: Arend van Spriel <arend.vanspriel@broadcom.com>
> ---
>  .../net/wireless/broadcom/brcm80211/brcmfmac/bus.h | 15 ++++++++++++--
>  .../wireless/broadcom/brcm80211/brcmfmac/core.c    | 23 +++++++++++++++-------
>  .../wireless/broadcom/brcm80211/brcmfmac/sdio.c    | 13 +++++++-----
>  3 files changed, 37 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
> index e1a4d9e..163ddc4 100644
> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
> @@ -113,6 +113,17 @@ struct brcmf_bus_msgbuf {
>  
>  
>  /**
> + * struct brcmf_bus_stats - bus statistic counters.
> + *
> + * @pktcowed: packets cowed for extra headroom/unorphan.
> + * @pktcow_failed: packets dropped due to failed cow-ing.
> + */
> +struct brcmf_bus_stats {
> +	atomic_t pktcowed;
> +	atomic_t pktcow_failed;
> +};

Same question as in the previous patch. I only see updates for these
variables, but nobody reading them?
Arend van Spriel June 27, 2017, 10:07 p.m. UTC | #2
On 27-06-17 16:09, Kalle Valo wrote:
> Arend van Spriel <arend.vanspriel@broadcom.com> writes:
> 
>> Since commit 9cc4b7cb86cb ("brcmfmac: Make skb header writable
>> before use") the headroom usage has been fixed. However, the
>> driver was keeping statistics that got lost. So reworking the
>> code so we get those driver statistics back for debugging.
>>
>> Cc: James Hughes <james.hughes@raspberrypi.org>
>> Reviewed-by: Hante Meuleman <hante.meuleman@broadcom.com>
>> Reviewed-by: Pieter-Paul Giesberts <pieter-paul.giesberts@broadcom.com>
>> Reviewed-by: Franky Lin <franky.lin@broadcom.com>
>> Signed-off-by: Arend van Spriel <arend.vanspriel@broadcom.com>
>> ---
>>  .../net/wireless/broadcom/brcm80211/brcmfmac/bus.h | 15 ++++++++++++--
>>  .../wireless/broadcom/brcm80211/brcmfmac/core.c    | 23 +++++++++++++++-------
>>  .../wireless/broadcom/brcm80211/brcmfmac/sdio.c    | 13 +++++++-----
>>  3 files changed, 37 insertions(+), 14 deletions(-)
>>
>> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
>> index e1a4d9e..163ddc4 100644
>> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
>> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
>> @@ -113,6 +113,17 @@ struct brcmf_bus_msgbuf {
>>  
>>  
>>  /**
>> + * struct brcmf_bus_stats - bus statistic counters.
>> + *
>> + * @pktcowed: packets cowed for extra headroom/unorphan.
>> + * @pktcow_failed: packets dropped due to failed cow-ing.
>> + */
>> +struct brcmf_bus_stats {
>> +	atomic_t pktcowed;
>> +	atomic_t pktcow_failed;
>> +};
> 
> Same question as in the previous patch. I only see updates for these
> variables, but nobody reading them?

Hi Kalle,

You are right. My intention was to expose these to debugfs, but clearly
did not include it in this patch series. So how do you want to handle
this. Feel free to drop patch 2 and 3. I can resubmit them with
subsequent patches dealing with exposing it to debugfs if that makes
more sense.

Regards,
Arend
Kalle Valo June 28, 2017, 4:04 p.m. UTC | #3
Arend van Spriel <arend.vanspriel@broadcom.com> writes:

> On 27-06-17 16:09, Kalle Valo wrote:
>> Arend van Spriel <arend.vanspriel@broadcom.com> writes:
>> 
>>> Since commit 9cc4b7cb86cb ("brcmfmac: Make skb header writable
>>> before use") the headroom usage has been fixed. However, the
>>> driver was keeping statistics that got lost. So reworking the
>>> code so we get those driver statistics back for debugging.
>>>
>>> Cc: James Hughes <james.hughes@raspberrypi.org>
>>> Reviewed-by: Hante Meuleman <hante.meuleman@broadcom.com>
>>> Reviewed-by: Pieter-Paul Giesberts <pieter-paul.giesberts@broadcom.com>
>>> Reviewed-by: Franky Lin <franky.lin@broadcom.com>
>>> Signed-off-by: Arend van Spriel <arend.vanspriel@broadcom.com>
>>> ---
>>>  .../net/wireless/broadcom/brcm80211/brcmfmac/bus.h | 15 ++++++++++++--
>>>  .../wireless/broadcom/brcm80211/brcmfmac/core.c    | 23 +++++++++++++++-------
>>>  .../wireless/broadcom/brcm80211/brcmfmac/sdio.c    | 13 +++++++-----
>>>  3 files changed, 37 insertions(+), 14 deletions(-)
>>>
>>> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
>>> b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
>>> index e1a4d9e..163ddc4 100644
>>> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
>>> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
>>> @@ -113,6 +113,17 @@ struct brcmf_bus_msgbuf {
>>>  
>>>  
>>>  /**
>>> + * struct brcmf_bus_stats - bus statistic counters.
>>> + *
>>> + * @pktcowed: packets cowed for extra headroom/unorphan.
>>> + * @pktcow_failed: packets dropped due to failed cow-ing.
>>> + */
>>> +struct brcmf_bus_stats {
>>> +	atomic_t pktcowed;
>>> +	atomic_t pktcow_failed;
>>> +};
>> 
>> Same question as in the previous patch. I only see updates for these
>> variables, but nobody reading them?
>
> Hi Kalle,
>
> You are right. My intention was to expose these to debugfs, but clearly
> did not include it in this patch series. So how do you want to handle
> this. Feel free to drop patch 2 and 3. I can resubmit them with
> subsequent patches dealing with exposing it to debugfs if that makes
> more sense.

No need to resend if you are going to add the debugfs interface soon.
But the general rule is that we don't add any dead code to kernel and
there was no explanation in the commit log, that's why I asked.
diff mbox

Patch

diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
index e1a4d9e..163ddc4 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h
@@ -113,6 +113,17 @@  struct brcmf_bus_msgbuf {
 
 
 /**
+ * struct brcmf_bus_stats - bus statistic counters.
+ *
+ * @pktcowed: packets cowed for extra headroom/unorphan.
+ * @pktcow_failed: packets dropped due to failed cow-ing.
+ */
+struct brcmf_bus_stats {
+	atomic_t pktcowed;
+	atomic_t pktcow_failed;
+};
+
+/**
  * struct brcmf_bus - interface structure between common and bus layer
  *
  * @bus_priv: pointer to private bus device.
@@ -120,8 +131,8 @@  struct brcmf_bus_msgbuf {
  * @dev: device pointer of bus device.
  * @drvr: public driver information.
  * @state: operational state of the bus interface.
+ * @stats: statistics shared between common and bus layer.
  * @maxctl: maximum size for rxctl request message.
- * @tx_realloc: number of tx packets realloced for headroom.
  * @chip: device identifier of the dongle chip.
  * @always_use_fws_queue: bus wants use queue also when fwsignal is inactive.
  * @wowl_supported: is wowl supported by bus driver.
@@ -137,8 +148,8 @@  struct brcmf_bus {
 	struct device *dev;
 	struct brcmf_pub *drvr;
 	enum brcmf_bus_state state;
+	struct brcmf_bus_stats stats;
 	uint maxctl;
-	atomic_t tx_realloc;
 	u32 chip;
 	u32 chiprev;
 	bool always_use_fws_queue;
diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c
index a88da5a..790151b 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c
@@ -199,6 +199,7 @@  static netdev_tx_t brcmf_netdev_start_xmit(struct sk_buff *skb,
 	struct brcmf_if *ifp = netdev_priv(ndev);
 	struct brcmf_pub *drvr = ifp->drvr;
 	struct ethhdr *eh;
+	int head_delta;
 
 	brcmf_dbg(DATA, "Enter, bsscfgidx=%d\n", ifp->bsscfgidx);
 
@@ -211,13 +212,21 @@  static netdev_tx_t brcmf_netdev_start_xmit(struct sk_buff *skb,
 		goto done;
 	}
 
-	/* Make sure there's enough writable headroom*/
-	ret = skb_cow_head(skb, drvr->hdrlen);
-	if (ret < 0) {
-		brcmf_err("%s: skb_cow_head failed\n",
-			  brcmf_ifname(ifp));
-		dev_kfree_skb(skb);
-		goto done;
+	/* Make sure there's enough writeable headroom */
+	if (skb_headroom(skb) < drvr->hdrlen || skb_header_cloned(skb)) {
+		head_delta = drvr->hdrlen - skb_headroom(skb);
+
+		brcmf_dbg(INFO, "%s: insufficient headroom (%d)\n",
+			  brcmf_ifname(ifp), head_delta);
+		atomic_inc(&drvr->bus_if->stats.pktcowed);
+		ret = pskb_expand_head(skb, ALIGN(head_delta, NET_SKB_PAD), 0,
+				       GFP_ATOMIC);
+		if (ret < 0) {
+			brcmf_err("%s: failed to expand headroom\n",
+				  brcmf_ifname(ifp));
+			atomic_inc(&drvr->bus_if->stats.pktcow_failed);
+			goto done;
+		}
 	}
 
 	/* validate length for ether packet */
diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c
index a280536..e745bc1 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c
@@ -2037,6 +2037,7 @@  static uint brcmf_sdio_readframes(struct brcmf_sdio *bus, uint maxframes)
 
 static int brcmf_sdio_txpkt_hdalign(struct brcmf_sdio *bus, struct sk_buff *pkt)
 {
+	struct brcmf_bus_stats *stats;
 	u16 head_pad;
 	u8 *dat_buf;
 
@@ -2046,16 +2047,18 @@  static int brcmf_sdio_txpkt_hdalign(struct brcmf_sdio *bus, struct sk_buff *pkt)
 	head_pad = ((unsigned long)dat_buf % bus->head_align);
 	if (head_pad) {
 		if (skb_headroom(pkt) < head_pad) {
-			atomic_inc(&bus->sdiodev->bus_if->tx_realloc);
-			head_pad = 0;
-			if (skb_cow(pkt, head_pad))
+			stats = &bus->sdiodev->bus_if->stats;
+			atomic_inc(&stats->pktcowed);
+			if (skb_cow_head(pkt, head_pad)) {
+				atomic_inc(&stats->pktcow_failed);
 				return -ENOMEM;
+			}
 		}
 		skb_push(pkt, head_pad);
 		dat_buf = (u8 *)(pkt->data);
-		memset(dat_buf, 0, head_pad + bus->tx_hdrlen);
 	}
-	return head_pad;
+	memset(dat_buf, 0, head_pad + bus->tx_hdrlen);
+	return 0;
 }
 
 /**