On Tue, May 28, 2013 at 03:58:07PM +0000, Vick, Matthew wrote:

On 5/27/13 2:21 AM, "Richard Cochran" [off-list ref] wrote:

I would prefer it if we did a MAC check before these two TSICR checks,
since we're making some assumptions about the hardware within the
interrupt cases. At the very least, a comment that these are only
applicable to I210/I211 would be nice.

I can respin with a comment that the additional bits are i210 only. I
think this is better than adding more checks into ISR. Since we only
enable these bits for the i210, the checks would be redundant.

quoted

diff --git a/drivers/net/ethernet/intel/igb/igb_ptp.c

b/drivers/net/ethernet/intel/igb/igb_ptp.c
index 5944de0..8cf4b8a 100644

--- a/drivers/net/ethernet/intel/igb/igb_ptp.c
+++ b/drivers/net/ethernet/intel/igb/igb_ptp.c

@@ -23,6 +23,15 @@

#include "igb.h"

+static int igb_input_sdp = 0;
+static int igb_output_sdp = 1;
+module_param(igb_input_sdp, int, 0444);
+module_param(igb_output_sdp, int, 0444);
+MODULE_PARM_DESC(igb_input_sdp,
+		 "The SDP used as an input, to time stamp external events");
+MODULE_PARM_DESC(igb_output_sdp,
+		 "The SDP used to output the programmable periodic signal");
+

Is there any other mechanism we could use to control this? I would imagine
not, but I know module parameters are generally frowned upon.

This the way I handled it for the PHYTER, and I think it is the best
way from our three choices:

1. kconfig option (inflexible)
2. module param
3. ethtool (can o'worms)

quoted

@@ -711,6 +800,68 @@ int igb_ptp_hwtstamp_ioctl(struct net_device *netdev,

		-EFAULT : 0;
}

+static int igb_sdp_init(struct igb_adapter *adapter)
+{
+	struct e1000_hw *hw = &adapter->hw;
+	u32 ctrl, ctrl_ext, tssdp = 0;
+
+	if (igb_input_sdp == igb_output_sdp) {
+		pr_err("SDP %d set as both input and output\n", igb_input_sdp);
+		return -1;

Shouldn't this return -EINVAL?

Maybe we don't need the return value at all. Just the message is
important, I think.
 

quoted

@@ -785,6 +936,10 @@ void igb_ptp_init(struct igb_adapter *adapter)

		struct timespec ts = ktime_to_timespec(ktime_get_real());

		igb_ptp_settime_i210(&adapter->ptp_caps, &ts);
+		igb_sdp_init(adapter);

What if igb_sdp_init fails?

Hmm, maybe we should then disable the extra features... will consider
for V2.

quoted

+		adapter->ptp_caps.n_ext_ts = 1;
+		adapter->ptp_caps.n_per_out = 1;
+		adapter->ptp_caps.pps = 1;

Thanks,
Richard