Skip to contentSkip to conversation

ARM: EDMA: Fix clearing of unused list for DT DMA resources

PATCH In review Joel Fernandes 3 participants 7 messages Sep 26, 2013 – Sep 27, 2013

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

PATCH v3 1/1 ARM: EDMA: Fix clearing of unused list for DT DMA resources

HWMOD removal for MMC is breaking edma_start as the events are being manually triggered due to unused channel list not being clear.

The above issue is fixed by reading the "dmas" property from the DT node if it exists and clearing the bits in the unused channel list if the dma controller used by any device is EDMA. For this purpose we use the of_* helpers to parse the arguments in the dmas phandle list.

Also introduced is a minor clean up of a checkpatch error in old code.

Cc Sekhar Nori <[email protected]>
Cc Tony Lindgren <[email protected]>
Cc Olof Johansson <[email protected]>
Cc Nishanth Menon <[email protected]>
Cc Pantel Antoniou <[email protected]>
Cc Jason Kridner <[email protected]>
Cc Koen Kooi <[email protected]>

---
Just resending this patch after discussing with Sekhar and Olof.

AM335x is being booted by many users such as the beaglebone community. DT is the only boot method available for all these users. EDMA is required for the operation for many common peripherals in AM335x SoC such as McASP, MMC and Crypto.

Although EDMA DT nodes are going in only for 3.13, in reality the kernel has been used for more than a year with EDMA code and out of tree EDMA DTS patches. Hence though the DT nodes are still not in mainline, this patch can be still be considered a critical fix as such and it would be great if it could be included in 3.12-rc release. Thanks.

More details about why this broke an existing feature folks were using:
Previously the DMA resources for platform devices were being populated through HWMOD, however with the recent clean ups with HWMOD, this data has been moved to Device tree. The EDMA code though is not aware of this so it fails to fetch the DMA resources correctly which it needs to prepare the unused channel list (basically doesn't properly clear the channels that are in use, in the unused list).

 arch/arm/common/edma.c | 38 +++++++++++++++++++++++++++++++-------
 1 file changed, 31 insertions(+), 7 deletions(-)
arch/arm/common/edma.c +31 -7
index 117f955..8e1a024 100644
@@ -269,6 +269,11 @@ static const struct edmacc_param dummy_paramset = {
 	.ccnt = 1,
 };
 
+static const struct of_device_id edma_of_ids[] = {
+	{ .compatible = "ti,edma3", },
+	{}
+};
+
 /*****************************************************************************/
 
 static void map_dmach_queue(unsigned ctlr, unsigned ch_no,
@@ -560,14 +565,38 @@ static int reserve_contiguous_slots(int ctlr, unsigned int id,
 static int prepare_unused_channel_list(struct device *dev, void *data)
 {
 	struct platform_device *pdev = to_platform_device(dev);
-	int i, ctlr;
+	int i, count, ctlr;
+	struct of_phandle_args  dma_spec;
 
+	if (dev->of_node) {
+		count = of_property_count_strings(dev->of_node, "dma-names");
+		if (count < 0)
+			return 0;
+		for (i = 0; i < count; i++) {
+			if (of_parse_phandle_with_args(dev->of_node, "dmas",
+						       "#dma-cells", i,
+						       &dma_spec))
+				continue;
+
+			if (!of_match_node(edma_of_ids, dma_spec.np)) {
+				of_node_put(dma_spec.np);
+				continue;
+			}
+
+			clear_bit(EDMA_CHAN_SLOT(dma_spec.args[0]),
+				  edma_cc[0]->edma_unused);
+			of_node_put(dma_spec.np);
+		}
+		return 0;
+	}
+
+	/* For non-OF case */
 	for (i = 0; i < pdev->num_resources; i++) {
 		if ((pdev->resource[i].flags & IORESOURCE_DMA) &&
 				(int)pdev->resource[i].start >= 0) {
 			ctlr = EDMA_CTLR(pdev->resource[i].start);
 			clear_bit(EDMA_CHAN_SLOT(pdev->resource[i].start),
-					edma_cc[ctlr]->edma_unused);
+				  edma_cc[ctlr]->edma_unused);
 		}
 	}
 
@@ -1762,11 +1791,6 @@ static int edma_probe(struct platform_device *pdev)
 	return 0;
 }
 
-static const struct of_device_id edma_of_ids[] = {
-	{ .compatible = "ti,edma3", },
-	{}
-};
-
 static struct platform_driver edma_driver = {
 	.driver = {
 		.name	= "edma",
Signature · 1 line
1.8.1.2
#1 lore raw
Quoted 20 lines · Joel Fernandes

HWMOD removal for MMC is breaking edma_start as the events are being manually triggered due to unused channel list not being clear.

The above issue is fixed by reading the "dmas" property from the DT node if it exists and clearing the bits in the unused channel list if the dma controller used by any device is EDMA. For this purpose we use the of_* helpers to parse the arguments in the dmas phandle list.

Also introduced is a minor clean up of a checkpatch error in old code.

Cc Sekhar Nori <[email protected]>
Cc Tony Lindgren <[email protected]>
Cc Olof Johansson <[email protected]>
Cc Nishanth Menon <[email protected]>
Cc Pantel Antoniou <[email protected]>
Cc Jason Kridner <[email protected]>
Cc Koen Kooi <[email protected]>
Quoted 2 lines

---
Just resending this patch after discussing with Sekhar and Olof.

Actually, the patch you talked to me about was v3 of this. It seems that you have reposted v6 but labelled it v3. This is very confusing.

Quoted 10 lines

AM335x is being booted by many users such as the beaglebone community. DT is the only boot method available for all these users. EDMA is required for the operation for many common peripherals in AM335x SoC such as McASP, MMC and Crypto.

Although EDMA DT nodes are going in only for 3.13, in reality the kernel has been used for more than a year with EDMA code and out of tree EDMA DTS patches. Hence though the DT nodes are still not in mainline, this patch can be still be considered a critical fix as such and it would be great if it could be included in 3.12-rc release. Thanks.

This is really the root of this problem. If you sit on code out of tree for a year, and something breaks that we couldn't even have detected since we didn't have the out-of-tree pieces. We'll help you the first few times (such as now) but we will eventually stop caring.

If I was in a worse mood, then I'd just say that since your users already has to have out-of-tree code to even use this functionality, they could just add this fix on top of that stack of patches. But I'm in a slightly better mood than that and I'll pick it up this time. :)

Quoted 7 lines

More details about why this broke an existing feature folks were using:
Previously the DMA resources for platform devices were being populated through HWMOD, however with the recent clean ups with HWMOD, this data has been moved to Device tree. The EDMA code though is not aware of this so it fails to fetch the DMA resources correctly which it needs to prepare the unused channel list (basically doesn't properly clear the channels that are in use, in the unused list).

So that we can learn for next time: What should we (as in us maintainers and you TI) have done differently to avoid this?

-Olof

#2 lore raw
Quoted 24 lines · Olof Johansson
Quoted 20 lines · Joel Fernandes

HWMOD removal for MMC is breaking edma_start as the events are being manually triggered due to unused channel list not being clear.

The above issue is fixed by reading the "dmas" property from the DT node if it exists and clearing the bits in the unused channel list if the dma controller used by any device is EDMA. For this purpose we use the of_* helpers to parse the arguments in the dmas phandle list.

Also introduced is a minor clean up of a checkpatch error in old code.

Cc Sekhar Nori <[email protected]>
Cc Tony Lindgren <[email protected]>
Cc Olof Johansson <[email protected]>
Cc Nishanth Menon <[email protected]>
Cc Pantel Antoniou <[email protected]>
Cc Jason Kridner <[email protected]>
Cc Koen Kooi <[email protected]>
Quoted 2 lines

---
Just resending this patch after discussing with Sekhar and Olof.

Quoted 2 lines

Actually, the patch you talked to me about was v3 of this. It seems that you have reposted v6 but labelled it v3. This is very confusing.

Sorry about this. :-( This is indeed v6.

Quoted 11 lines
Quoted 10 lines

AM335x is being booted by many users such as the beaglebone community. DT is the only boot method available for all these users. EDMA is required for the operation for many common peripherals in AM335x SoC such as McASP, MMC and Crypto.

Although EDMA DT nodes are going in only for 3.13, in reality the kernel has been used for more than a year with EDMA code and out of tree EDMA DTS patches. Hence though the DT nodes are still not in mainline, this patch can be still be considered a critical fix as such and it would be great if it could be included in 3.12-rc release. Thanks.

Quoted 4 lines

This is really the root of this problem. If you sit on code out of tree for a year, and something breaks that we couldn't even have detected since we didn't have the out-of-tree pieces. We'll help you the first few times (such as now) but we will eventually stop caring.

When I started looking at EDMA in June, I noticed that a lot had already been merged. EDMA DMA Engine driver itself was merged last year, no worries there. but the long pending list of fixes to be made to the driver had to written and rewritten multiple times which took a long time.

Due to this, the EDMA device tree entries could not be merged in fear that doing so would cause problems such as MMC/SD corruption etc.

Quoted 4 lines

If I was in a worse mood, then I'd just say that since your users already has to have out-of-tree code to even use this functionality, they could just add this fix on top of that stack of patches. But I'm in a slightly better mood than that and I'll pick it up this time. :)

Thank you! :)

Quoted 10 lines
Quoted 7 lines

More details about why this broke an existing feature folks were using:
Previously the DMA resources for platform devices were being populated through HWMOD, however with the recent clean ups with HWMOD, this data has been moved to Device tree. The EDMA code though is not aware of this so it fails to fetch the DMA resources correctly which it needs to prepare the unused channel list (basically doesn't properly clear the channels that are in use, in the unused list).

So that we can learn for next time: What should we (as in us maintainers and you TI) have done differently to avoid this?

I think a little on both sides can be improved.

For TI, a bit more work toward explaining patches better in change logs so that maintainers understand and are willing to help to get the code upstream would help. A lot of improvement to internal policies have been made for developing on upstream, and that's certainly a good thing but there's still more room for improvement.

For maintainers, EDMA code would have been kept breathing all these months (atleast 8 months) if those temporary fixes mentioned above would have been merged into the kernel instead of not applied. With those fixes in place, dts could have been enabled and EDMA would be fully working all these months. This would have certainly made things a lot easier. Rewriting stuff the right way is OK but if it is a _lot_ of effort, then to keep things alive, merging stuff "for now" specially if they are one-liners could be made acceptable.

EDMA fixes have now been written the "right way", so those temporary fixes are no longer needed :) but it certainly took a long time to do it the right way during which things were dead. Only thing pending for working EDMA now is the dts patches which are already in Benoit's for-3.13 tree and this patch that you're picking up.

Thanks Olof for your help! :)

regards,

-Joel

#3 lore raw
Quoted 80 lines · Joel Fernandes
Quoted 24 lines · Olof Johansson
Quoted 20 lines · Joel Fernandes

HWMOD removal for MMC is breaking edma_start as the events are being manually triggered due to unused channel list not being clear.

The above issue is fixed by reading the "dmas" property from the DT node if it exists and clearing the bits in the unused channel list if the dma controller used by any device is EDMA. For this purpose we use the of_* helpers to parse the arguments in the dmas phandle list.

Also introduced is a minor clean up of a checkpatch error in old code.

Cc Sekhar Nori <[email protected]>
Cc Tony Lindgren <[email protected]>
Cc Olof Johansson <[email protected]>
Cc Nishanth Menon <[email protected]>
Cc Pantel Antoniou <[email protected]>
Cc Jason Kridner <[email protected]>
Cc Koen Kooi <[email protected]>
Quoted 2 lines

---
Just resending this patch after discussing with Sekhar and Olof.

Quoted 2 lines

Actually, the patch you talked to me about was v3 of this. It seems that you have reposted v6 but labelled it v3. This is very confusing.

Sorry about this. :-( This is indeed v6.

Quoted 11 lines
Quoted 10 lines

AM335x is being booted by many users such as the beaglebone community. DT is the only boot method available for all these users. EDMA is required for the operation for many common peripherals in AM335x SoC such as McASP, MMC and Crypto.

Although EDMA DT nodes are going in only for 3.13, in reality the kernel has been used for more than a year with EDMA code and out of tree EDMA DTS patches. Hence though the DT nodes are still not in mainline, this patch can be still be considered a critical fix as such and it would be great if it could be included in 3.12-rc release. Thanks.

Quoted 4 lines

This is really the root of this problem. If you sit on code out of tree for a year, and something breaks that we couldn't even have detected since we didn't have the out-of-tree pieces. We'll help you the first few times (such as now) but we will eventually stop caring.

When I started looking at EDMA in June, I noticed that a lot had already been merged. EDMA DMA Engine driver itself was merged last year, no worries there. but the long pending list of fixes to be made to the driver had to written and rewritten multiple times which took a long time.

Due to this, the EDMA device tree entries could not be merged in fear that doing so would cause problems such as MMC/SD corruption etc.

Quoted 4 lines

If I was in a worse mood, then I'd just say that since your users already has to have out-of-tree code to even use this functionality, they could just add this fix on top of that stack of patches. But I'm in a slightly better mood than that and I'll pick it up this time. :)

Thank you! :)

Quoted 10 lines
Quoted 7 lines

More details about why this broke an existing feature folks were using:
Previously the DMA resources for platform devices were being populated through HWMOD, however with the recent clean ups with HWMOD, this data has been moved to Device tree. The EDMA code though is not aware of this so it fails to fetch the DMA resources correctly which it needs to prepare the unused channel list (basically doesn't properly clear the channels that are in use, in the unused list).

So that we can learn for next time: What should we (as in us maintainers and you TI) have done differently to avoid this?

I think a little on both sides can be improved.

For TI, a bit more work toward explaining patches better in change logs so that maintainers understand and are willing to help to get the code upstream would help. A lot of improvement to internal policies have been made for developing on upstream, and that's certainly a good thing but there's still more room for improvement.

Quoted 7 lines

For maintainers, EDMA code would have been kept breathing all these months (atleast 8 months) if those temporary fixes mentioned above would have been merged into the kernel instead of not applied. With those fixes in place, dts could have been enabled and EDMA would be fully working all these months. This would have certainly made things a lot easier. Rewriting stuff the right way is OK but if it is a _lot_ of effort, then to keep things alive, merging stuff "for now" specially if they are one-liners could be made acceptable.

Joel, can you give a link to the "one-liner" "temporary fix" that was was not merged? I am unable to put it in context.

Thanks,
Sekhar

#4 lore raw
Quoted 89 lines · Sekhar Nori
Quoted 80 lines · Joel Fernandes
Quoted 24 lines · Olof Johansson
Quoted 20 lines · Joel Fernandes

HWMOD removal for MMC is breaking edma_start as the events are being manually triggered due to unused channel list not being clear.

The above issue is fixed by reading the "dmas" property from the DT node if it exists and clearing the bits in the unused channel list if the dma controller used by any device is EDMA. For this purpose we use the of_* helpers to parse the arguments in the dmas phandle list.

Also introduced is a minor clean up of a checkpatch error in old code.

Cc Sekhar Nori <[email protected]>
Cc Tony Lindgren <[email protected]>
Cc Olof Johansson <[email protected]>
Cc Nishanth Menon <[email protected]>
Cc Pantel Antoniou <[email protected]>
Cc Jason Kridner <[email protected]>
Cc Koen Kooi <[email protected]>
Quoted 2 lines

---
Just resending this patch after discussing with Sekhar and Olof.

Quoted 2 lines

Actually, the patch you talked to me about was v3 of this. It seems that you have reposted v6 but labelled it v3. This is very confusing.

Sorry about this. :-( This is indeed v6.

Quoted 11 lines
Quoted 10 lines

AM335x is being booted by many users such as the beaglebone community. DT is the only boot method available for all these users. EDMA is required for the operation for many common peripherals in AM335x SoC such as McASP, MMC and Crypto.

Although EDMA DT nodes are going in only for 3.13, in reality the kernel has been used for more than a year with EDMA code and out of tree EDMA DTS patches. Hence though the DT nodes are still not in mainline, this patch can be still be considered a critical fix as such and it would be great if it could be included in 3.12-rc release. Thanks.

Quoted 4 lines

This is really the root of this problem. If you sit on code out of tree for a year, and something breaks that we couldn't even have detected since we didn't have the out-of-tree pieces. We'll help you the first few times (such as now) but we will eventually stop caring.

When I started looking at EDMA in June, I noticed that a lot had already been merged. EDMA DMA Engine driver itself was merged last year, no worries there. but the long pending list of fixes to be made to the driver had to written and rewritten multiple times which took a long time.

Due to this, the EDMA device tree entries could not be merged in fear that doing so would cause problems such as MMC/SD corruption etc.

Quoted 4 lines

If I was in a worse mood, then I'd just say that since your users already has to have out-of-tree code to even use this functionality, they could just add this fix on top of that stack of patches. But I'm in a slightly better mood than that and I'll pick it up this time. :)

Thank you! :)

Quoted 10 lines
Quoted 7 lines

More details about why this broke an existing feature folks were using:
Previously the DMA resources for platform devices were being populated through HWMOD, however with the recent clean ups with HWMOD, this data has been moved to Device tree. The EDMA code though is not aware of this so it fails to fetch the DMA resources correctly which it needs to prepare the unused channel list (basically doesn't properly clear the channels that are in use, in the unused list).

So that we can learn for next time: What should we (as in us maintainers and you TI) have done differently to avoid this?

I think a little on both sides can be improved.

For TI, a bit more work toward explaining patches better in change logs so that maintainers understand and are willing to help to get the code upstream would help. A lot of improvement to internal policies have been made for developing on upstream, and that's certainly a good thing but there's still more room forard improvement.

Quoted 7 lines

For maintainers, EDMA code would have been kept breathing all these months (atleast 8 months) if those temporary fixes mentioned above would have been merged into the kernel instead of not applied. With those fixes in place, dts could have been enabled and EDMA would be fully working all these months. This would have certainly made things a lot easier. Rewriting stuff the right way is OK but if it is a _lot_ of effort, then to keep things alive, merging stuff "for now" specially if they are one-liners could be made acceptable.

Quoted 2 lines

Joel, can you give a link to the "one-liner" "temporary fix" that was was not merged? I am unable to put it in context.

Sure, not exactly a one-line but maybe couple of lines (see below). Also wasn't referring to anything you're maintaining. I was just continuing the original discussion of where we can improve as a community.

My point was very trivial changes that keep things working such as this one should be merged in time to keep things working:
http://lkml.indiana.edu/hypermail/linux/kernel/1210.2/01476.html

That one patch was reworked several times, and then was decided to do the right way and finally merged with the limit-less series. All I am saying is all this while the above patch(es) could be merged to keep things alive, and that way we could avoid a lot of the pain and frustration that folks felt.

Its just my opinion though, I don't feel strongly about it but I think this may be a helpful approach that's all.

regards,

-Joel

#5 lore raw
In reply to Joel Fernandes
Quoted 69 lines · Joel Fernandes
Quoted 24 lines · Olof Johansson
Quoted 20 lines · Joel Fernandes

HWMOD removal for MMC is breaking edma_start as the events are being manually triggered due to unused channel list not being clear.

The above issue is fixed by reading the "dmas" property from the DT node if it exists and clearing the bits in the unused channel list if the dma controller used by any device is EDMA. For this purpose we use the of_* helpers to parse the arguments in the dmas phandle list.

Also introduced is a minor clean up of a checkpatch error in old code.

Cc Sekhar Nori <[email protected]>
Cc Tony Lindgren <[email protected]>
Cc Olof Johansson <[email protected]>
Cc Nishanth Menon <[email protected]>
Cc Pantel Antoniou <[email protected]>
Cc Jason Kridner <[email protected]>
Cc Koen Kooi <[email protected]>
Quoted 2 lines

---
Just resending this patch after discussing with Sekhar and Olof.

Quoted 2 lines

Actually, the patch you talked to me about was v3 of this. It seems that you have reposted v6 but labelled it v3. This is very confusing.

Sorry about this. :-( This is indeed v6.

Quoted 11 lines
Quoted 10 lines

AM335x is being booted by many users such as the beaglebone community. DT is the only boot method available for all these users. EDMA is required for the operation for many common peripherals in AM335x SoC such as McASP, MMC and Crypto.

Although EDMA DT nodes are going in only for 3.13, in reality the kernel has been used for more than a year with EDMA code and out of tree EDMA DTS patches. Hence though the DT nodes are still not in mainline, this patch can be still be considered a critical fix as such and it would be great if it could be included in 3.12-rc release. Thanks.

Quoted 4 lines

This is really the root of this problem. If you sit on code out of tree for a year, and something breaks that we couldn't even have detected since we didn't have the out-of-tree pieces. We'll help you the first few times (such as now) but we will eventually stop caring.

When I started looking at EDMA in June, I noticed that a lot had already been merged. EDMA DMA Engine driver itself was merged last year, no worries there. but the long pending list of fixes to be made to the driver had to written and rewritten multiple times which took a long time.

Due to this, the EDMA device tree entries could not be merged in fear that doing so would cause problems such as MMC/SD corruption etc.

Quoted 4 lines

If I was in a worse mood, then I'd just say that since your users already has to have out-of-tree code to even use this functionality, they could just add this fix on top of that stack of patches. But I'm in a slightly better mood than that and I'll pick it up this time. :)

Thank you! :)

Quoted 8 lines
Quoted 7 lines

More details about why this broke an existing feature folks were using:
Previously the DMA resources for platform devices were being populated through HWMOD, however with the recent clean ups with HWMOD, this data has been moved to Device tree. The EDMA code though is not aware of this so it fails to fetch the DMA resources correctly which it needs to prepare the unused channel list (basically doesn't properly clear the channels that are in use, in the unused list).

Quoted 4 lines
Quoted 2 lines

So that we can learn for next time: What should we (as in us maintainers and you TI) have done differently to avoid this?

I think a little on both sides can be improved.

Since we are in lessons learnt mode, I think as developers we need to learn to prioritize fixes over other features we are working on. I went back to the chronology of this patch series.

22nd July 2013: v2 posted
29th July 2013: Discussion on whether the patch can wait till *v3.12*
merge window.
29th July 2013: comments given on v2
22nd Aug 2013: Pull request's sent by Sekhar for v3.12
24th Aug 2013: another v2 posted (all comments given earlier not
addressed, received some comments on build warnings)
27th Aug 2013: another v3 posted (all comments given on 29th July not
addressed)
10th Sept 2013: another v3 posted (all comments given on 29th July not
addressed)
[some discussion on comments and why this cannot wait until v3.13]
17th Sept 2013: Final version ready for merge posted.
26th Sept 2013: Another v3 posted, this time for Olof to send into

		v3.12-rc

See, early on, the patch was actually in consideration for v3.12 merge. The barrier of entry into -rc cycle is pretty high. So if you have an opportunity to hit a merge window, utilize that by prioritizing this work over anything else you may be doing.

I know you got busy with adding support for SG lists and all, but clearly this patch is critical in your mind. Plus the comments were not tough to fix. There is a need to keep looking at what provides the best return on time invested.

Thanks,
Sekhar

#6 lore raw
Quoted 70 lines · Sekhar Nori
Quoted 69 lines · Joel Fernandes
Quoted 24 lines · Olof Johansson
Quoted 20 lines · Joel Fernandes

HWMOD removal for MMC is breaking edma_start as the events are being manually triggered due to unused channel list not being clear.

The above issue is fixed by reading the "dmas" property from the DT node if it exists and clearing the bits in the unused channel list if the dma controller used by any device is EDMA. For this purpose we use the of_* helpers to parse the arguments in the dmas phandle list.

Also introduced is a minor clean up of a checkpatch error in old code.

Cc Sekhar Nori <[email protected]>
Cc Tony Lindgren <[email protected]>
Cc Olof Johansson <[email protected]>
Cc Nishanth Menon <[email protected]>
Cc Pantel Antoniou <[email protected]>
Cc Jason Kridner <[email protected]>
Cc Koen Kooi <[email protected]>
Quoted 2 lines

---
Just resending this patch after discussing with Sekhar and Olof.

Quoted 2 lines

Actually, the patch you talked to me about was v3 of this. It seems that you have reposted v6 but labelled it v3. This is very confusing.

Sorry about this. :-( This is indeed v6.

Quoted 11 lines
Quoted 10 lines

AM335x is being booted by many users such as the beaglebone community. DT is the only boot method available for all these users. EDMA is required for the operation for many common peripherals in AM335x SoC such as McASP, MMC and Crypto.

Although EDMA DT nodes are going in only for 3.13, in reality the kernel has been used for more than a year with EDMA code and out of tree EDMA DTS patches. Hence though the DT nodes are still not in mainline, this patch can be still be considered a critical fix as such and it would be great if it could be included in 3.12-rc release. Thanks.

Quoted 4 lines

This is really the root of this problem. If you sit on code out of tree for a year, and something breaks that we couldn't even have detected since we didn't have the out-of-tree pieces. We'll help you the first few times (such as now) but we will eventually stop caring.

When I started looking at EDMA in June, I noticed that a lot had already been merged. EDMA DMA Engine driver itself was merged last year, no worries there. but the long pending list of fixes to be made to the driver had to written and rewritten multiple times which took a long time.

Due to this, the EDMA device tree entries could not be merged in fear that doing so would cause problems such as MMC/SD corruption etc.

Quoted 4 lines

If I was in a worse mood, then I'd just say that since your users already has to have out-of-tree code to even use this functionality, they could just add this fix on top of that stack of patches. But I'm in a slightly better mood than that and I'll pick it up this time. :)

Thank you! :)

Quoted 8 lines
Quoted 7 lines

More details about why this broke an existing feature folks were using:
Previously the DMA resources for platform devices were being populated through HWMOD, however with the recent clean ups with HWMOD, this data has been moved to Device tree. The EDMA code though is not aware of this so it fails to fetch the DMA resources correctly which it needs to prepare the unused channel list (basically doesn't properly clear the channels that are in use, in the unused list).

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

So that we can learn for next time: What should we (as in us maintainers and you TI) have done differently to avoid this?

I think a little on both sides can be improved.

Since we are in lessons learnt mode, I think as developers we need to learn to prioritize fixes over other features we are working on. I went back to the chronology of this patch series.

Sure, I agree with this. Will definitely work on it.

Quoted 22 lines

22nd July 2013: v2 posted
29th July 2013: Discussion on whether the patch can wait till *v3.12*
merge window.
29th July 2013: comments given on v2
22nd Aug 2013: Pull request's sent by Sekhar for v3.12
24th Aug 2013: another v2 posted (all comments given earlier not
addressed, received some comments on build warnings)
27th Aug 2013: another v3 posted (all comments given on 29th July not
addressed)
10th Sept 2013: another v3 posted (all comments given on 29th July not
addressed)
[some discussion on comments and why this cannot wait until v3.13]
17th Sept 2013: Final version ready for merge posted.
26th Sept 2013: Another v3 posted, this time for Olof to send into

		v3.12-rc

See, early on, the patch was actually in consideration for v3.12 merge. The barrier of entry into -rc cycle is pretty high. So if you have an opportunity to hit a merge window, utilize that by prioritizing this work over anything else you may be doing.

Quoted 4 lines

I know you got busy with adding support for SG lists and all, but clearly this patch is critical in your mind. Plus the comments were not tough to fix. There is a need to keep looking at what provides the best return on time invested.

Yes, but we are not referring to this particular patch at all. Olof was asking about the code that was not been merged and/or the reasons things are not working. I was just responding to that. But thanks for the advice, will def keep it in mind.

Regards,

-Joel