Skip to contentSkip to conversation

dw_dmac: return proper residue value

PATCH Superseded Andy Shevchenko 2 participants 11 messages Jan 23, 2013 – Jan 24, 2013 dmaengine

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover dw_dmac: return proper residue value

The patch series is targeted for getting proper residue value.

Since v1:
- everything is rewritten to address Viresh's and Vinod's comments.

Andy Shevchenko (4):
dw_dmac: remove unnecessary tx_list field in dw_dma_chan
dw_dmac: introduce total_len field in struct dw_desc
dw_dmac: fill individual length of descriptor
dw_dmac: return proper residue value

 drivers/dma/dw_dmac.c      |   86 ++++++++++++++++++++++++++++++++++++--------
 drivers/dma/dw_dmac_regs.h |    3 +-
 2 files changed, 74 insertions(+), 15 deletions(-)
Signature · 1 line
1.7.10.4
#1 lore raw

PATCH v2 1/4 dw_dmac: remove unnecessary tx_list field in dw_dma_chan

The soft LLP mode is working for active descriptor only. So, we do not need to have a copy of its pointer.

---
 drivers/dma/dw_dmac.c      |   17 ++++++++++++-----
 drivers/dma/dw_dmac_regs.h |    1 -
 2 files changed, 12 insertions(+), 6 deletions(-)
drivers/dma/dw_dmac.c +12 -5
index 72e6316..82e7c58 100644
@@ -284,9 +284,9 @@ static void dwc_dostart(struct dw_dma_chan *dwc, struct dw_desc *first)
 
 		dwc_initialize(dwc);
 
-		dwc->tx_list = &first->tx_list;
 		dwc->tx_node_active = &first->tx_list;
 
+		/* Submit first block */
 		dwc_do_single_block(dwc, first);
 
 		return;
@@ -404,15 +404,22 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 		dma_writel(dw, CLEAR.XFER, dwc->mask);
 
 		if (test_bit(DW_DMA_IS_SOFT_LLP, &dwc->flags)) {
-			if (dwc->tx_node_active != dwc->tx_list) {
-				desc = to_dw_desc(dwc->tx_node_active);
+			/*
+			 * We are inside first active descriptor.
+			 * Otherwise something is really wrong.
+			 */
+			desc = dwc_first_active(dwc);
+
+			if (dwc->tx_node_active != &desc->tx_list) {
+				child = to_dw_desc(dwc->tx_node_active);
 
 				/* Submit next block */
-				dwc_do_single_block(dwc, desc);
-				spin_unlock_irqrestore(&dwc->lock, flags);
+				dwc_do_single_block(dwc, child);
 
+				spin_unlock_irqrestore(&dwc->lock, flags);
 				return;
 			}
+
 			/* We are done here */
 			clear_bit(DW_DMA_IS_SOFT_LLP, &dwc->flags);
 		}
drivers/dma/dw_dmac_regs.h +0 -1
index fef296d..13000d2 100644
@@ -194,7 +194,6 @@ struct dw_dma_chan {
 	bool				initialized;
 
 	/* software emulation of the LLP transfers */
-	struct list_head	*tx_list;
 	struct list_head	*tx_node_active;
 
 	spinlock_t		lock;
Signature · 1 line
1.7.10.4
#3 lore raw

PATCH v2 2/4 dw_dmac: introduce total_len field in struct dw_desc

In reply to Andy Shevchenko

By this new field we distinguish a total length of the chain and the individual length of each descriptor in the chain.

---
 drivers/dma/dw_dmac.c      |   12 ++++++------
 drivers/dma/dw_dmac_regs.h |    1 +
 2 files changed, 7 insertions(+), 6 deletions(-)
drivers/dma/dw_dmac.c +6 -6
index 82e7c58..c3062a5 100644
@@ -335,18 +335,18 @@ dwc_descriptor_complete(struct dw_dma_chan *dwc, struct dw_desc *desc,
 		if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
 			if (txd->flags & DMA_COMPL_DEST_UNMAP_SINGLE)
 				dma_unmap_single(parent, desc->lli.dar,
-						desc->len, DMA_FROM_DEVICE);
+					desc->total_len, DMA_FROM_DEVICE);
 			else
 				dma_unmap_page(parent, desc->lli.dar,
-						desc->len, DMA_FROM_DEVICE);
+					desc->total_len, DMA_FROM_DEVICE);
 		}
 		if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
 			if (txd->flags & DMA_COMPL_SRC_UNMAP_SINGLE)
 				dma_unmap_single(parent, desc->lli.sar,
-						desc->len, DMA_TO_DEVICE);
+					desc->total_len, DMA_TO_DEVICE);
 			else
 				dma_unmap_page(parent, desc->lli.sar,
-						desc->len, DMA_TO_DEVICE);
+					desc->total_len, DMA_TO_DEVICE);
 		}
 	}
 
@@ -773,7 +773,7 @@ dwc_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
 
 	prev->lli.llp = 0;
 	first->txd.flags = flags;
-	first->len = len;
+	first->total_len = len;
 
 	return &first->txd;
 
@@ -936,7 +936,7 @@ slave_sg_fromdev_fill_desc:
 		prev->lli.ctllo |= DWC_CTLL_INT_EN;
 
 	prev->lli.llp = 0;
-	first->len = total_len;
+	first->total_len = total_len;
 
 	return &first->txd;
 
drivers/dma/dw_dmac_regs.h +1 -0
index 13000d2..833b4cf 100644
@@ -296,6 +296,7 @@ struct dw_desc {
 	struct list_head		tx_list;
 	struct dma_async_tx_descriptor	txd;
 	size_t				len;
+	size_t				total_len;
 };
 
 #define to_dw_desc(h)	list_entry(h, struct dw_desc, desc_node)
Signature · 1 line
1.7.10.4
#5 lore raw

PATCH v2 3/4 dw_dmac: fill individual length of descriptor

In reply to Andy Shevchenko

It will be useful to have the length of the transfer in the descriptor. The cyclic transfer functions remained untouched.

---
 drivers/dma/dw_dmac.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/dma/dw_dmac.c +3 -1
index c3062a5..026a3c7 100644
@@ -756,6 +756,7 @@ dwc_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
 		desc->lli.dar = dest + offset;
 		desc->lli.ctllo = ctllo;
 		desc->lli.ctlhi = xfer_count;
+		desc->len = xfer_count << src_width;
 
 		if (!first) {
 			first = desc;
@@ -852,8 +853,8 @@ slave_sg_todev_fill_desc:
 				dlen = len;
 				len = 0;
 			}
-
 			desc->lli.ctlhi = dlen >> mem_width;
+			desc->len = dlen;
 
 			if (!first) {
 				first = desc;
@@ -912,6 +913,7 @@ slave_sg_fromdev_fill_desc:
 				len = 0;
 			}
 			desc->lli.ctlhi = dlen >> reg_width;
+			desc->len = dlen;
 
 			if (!first) {
 				first = desc;
Signature · 1 line
1.7.10.4
Acked-by on v2 3/4
#6 lore raw

Re: [PATCH v2 3/4] dw_dmac: fill individual length of descriptor

Quoted 21 lines · Andy Shevchenko

It will be useful to have the length of the transfer in the descriptor. The cyclic transfer functions remained untouched.

---
 drivers/dma/dw_dmac.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/dma/dw_dmac.c +1 -0
index c3062a5..026a3c7 100644
@@ -756,6 +756,7 @@ dwc_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
                desc->lli.dar = dest + offset;
                desc->lli.ctllo = ctllo;
                desc->lli.ctlhi = xfer_count;
+               desc->len = xfer_count << src_width;
 
                if (!first) {
                        first = desc;

drivers/dma/dw_dmac.c slave_sg_todev_fill_desc: · line 855

Quoted 4 lines
drivers/dma/dw_dmac.c +0 -1
@@ -852,8 +853,8 @@ slave_sg_todev_fill_desc:
                                dlen = len;
                                len = 0;
                        }
-

:(

drivers/dma/dw_dmac.c slave_sg_fromdev_fill_desc: · line 919

Quoted 13 lines
drivers/dma/dw_dmac.c +2 -0
                        desc->lli.ctlhi = dlen >> mem_width;
+                       desc->len = dlen;
 
                        if (!first) {
                                first = desc;
@@ -912,6 +913,7 @@ slave_sg_fromdev_fill_desc:
                                len = 0;
                        }
                        desc->lli.ctlhi = dlen >> reg_width;
+                       desc->len = dlen;
 
                        if (!first) {
                                first = desc;

After fixing that.

#7 lore raw

PATCH v2 4/4 dw_dmac: return proper residue value

In reply to Andy Shevchenko

Currently the driver returns full length of the active descriptor which is wrong. We have to go throught the active descriptor and substract the length of each sent children in the chain from the total length along with the actual data in the DMA channel registers.

The cyclic case is not handled by this patch due to len field in the descriptor structure is left untouched by the original code.

---
 drivers/dma/dw_dmac.c      |   53 ++++++++++++++++++++++++++++++++++++++++++--
 drivers/dma/dw_dmac_regs.h |    1 +
 2 files changed, 52 insertions(+), 2 deletions(-)
drivers/dma/dw_dmac.c +51 -2
index 026a3c7..639d4cc 100644
@@ -387,6 +387,35 @@ static void dwc_complete_all(struct dw_dma *dw, struct dw_dma_chan *dwc)
 		dwc_descriptor_complete(dwc, desc, true);
 }
 
+/* Returns how many bytes were already received from source */
+static inline u32 dwc_get_sent(struct dw_dma_chan *dwc)
+{
+	u32 ctlhi = channel_readl(dwc, CTL_HI);
+	u32 ctllo = channel_readl(dwc, CTL_LO);
+
+	return (ctlhi & DWC_CTLH_BLOCK_TS_MASK) * (1 << (ctllo >> 4 & 7));
+}
+
+static void dwc_update_residue(struct dw_dma_chan *dwc, struct dw_desc *desc)
+{
+	struct list_head *head = &desc->tx_list, *active;
+
+	/* First node is active */
+	if (dwc->tx_node_active == head->next) {
+		dwc->residue = desc->total_len - dwc_get_sent(dwc);
+		return;
+	}
+
+	/* Somewhere in the middle */
+	dwc->residue = desc->total_len - desc->len;
+	list_for_each(active, head) {
+		if (active == dwc->tx_node_active->prev)
+			break;
+		dwc->residue -= to_dw_desc(active)->len;
+	}
+	dwc->residue -= dwc_get_sent(dwc);
+}
+
 static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 {
 	dma_addr_t llp;
@@ -399,6 +428,8 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 	llp = channel_readl(dwc, LLP);
 	status_xfer = dma_readl(dw, RAW.XFER);
 
+	dwc->residue = 0;
+
 	if (status_xfer & dwc->mask) {
 		/* Everything we've submitted is done */
 		dma_writel(dw, CLEAR.XFER, dwc->mask);
@@ -410,6 +441,8 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 			 */
 			desc = dwc_first_active(dwc);
 
+			dwc_update_residue(dwc, desc);
+
 			if (dwc->tx_node_active != &desc->tx_list) {
 				child = to_dw_desc(dwc->tx_node_active);
 
@@ -436,6 +469,9 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 
 	if (test_bit(DW_DMA_IS_SOFT_LLP, &dwc->flags)) {
 		dev_vdbg(chan2dev(&dwc->chan), "%s: soft LLP mode\n", __func__);
+
+		dwc_update_residue(dwc, dwc_first_active(dwc));
+
 		spin_unlock_irqrestore(&dwc->lock, flags);
 		return;
 	}
@@ -444,6 +480,9 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 			(unsigned long long)llp);
 
 	list_for_each_entry_safe(desc, _desc, &dwc->active_list, desc_node) {
+		/* initial residue value */
+		dwc->residue = desc->total_len;
+
 		/* check first descriptors addr */
 		if (desc->txd.phys == llp) {
 			spin_unlock_irqrestore(&dwc->lock, flags);
@@ -453,16 +492,21 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 		/* check first descriptors llp */
 		if (desc->lli.llp == llp) {
 			/* This one is currently in progress */
+			dwc->residue -= dwc_get_sent(dwc);
 			spin_unlock_irqrestore(&dwc->lock, flags);
 			return;
 		}
 
-		list_for_each_entry(child, &desc->tx_list, desc_node)
+		dwc->residue -= desc->len;
+		list_for_each_entry(child, &desc->tx_list, desc_node) {
 			if (child->lli.llp == llp) {
 				/* Currently in progress */
+				dwc->residue -= dwc_get_sent(dwc);
 				spin_unlock_irqrestore(&dwc->lock, flags);
 				return;
 			}
+			dwc->residue -= child->len;
+		}
 
 		/*
 		 * No descriptors so far seem to be in progress, i.e.
@@ -1058,6 +1102,7 @@ dwc_tx_status(struct dma_chan *chan,
 	      struct dma_tx_state *txstate)
 {
 	struct dw_dma_chan	*dwc = to_dw_dma_chan(chan);
+	unsigned long		flags;
 	enum dma_status		ret;
 
 	ret = dma_cookie_status(chan, cookie, txstate);
@@ -1067,8 +1112,12 @@ dwc_tx_status(struct dma_chan *chan,
 		ret = dma_cookie_status(chan, cookie, txstate);
 	}
 
+	spin_lock_irqsave(&dwc->lock, flags);
+
 	if (ret != DMA_SUCCESS)
-		dma_set_residue(txstate, dwc_first_active(dwc)->len);
+		dma_set_residue(txstate, dwc->residue);
+
+	spin_unlock_irqrestore(&dwc->lock, flags);
 
 	if (dwc->paused)
 		return DMA_PAUSED;
drivers/dma/dw_dmac_regs.h +1 -0
index 833b4cf..88dd8eb 100644
@@ -203,6 +203,7 @@ struct dw_dma_chan {
 	struct list_head	active_list;
 	struct list_head	queue;
 	struct list_head	free_list;
+	u32			residue;
 	struct dw_cyclic_desc	*cdesc;
 
 	unsigned int		descs_allocated;
Signature · 1 line
1.7.10.4
#8 lore raw

drivers/dma/dw_dmac.c static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc) · line 447

Quoted 13 lines · Andy Shevchenko
drivers/dma/dw_dmac.c +2 -0
 
 static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 {
        dma_addr_t llp;
@@ -410,6 +441,8 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
                         */
                        desc = dwc_first_active(dwc);
 
+                       dwc_update_residue(dwc, desc);
+
                        if (dwc->tx_node_active != &desc->tx_list) {
                                child = to_dw_desc(dwc->tx_node_active);

Is there a point updating residue here? I don't have a very good knowledge of nollp transfers but this is what i know...

The above "if" will pass if we are still doing transfers and fail if all transfers are done.
After the end of each LLI we receive an interrupt, where we queue next LLI. Better
would be to initialize dwc->residue at dwc_dostart() with total length, start decrementing
it with desc->len for every lli interrupt we get and if call for getting residue comes in
middle of transfer, simple return residue - dwc_get_sent(desc) without updating residue
field...

drivers/dma/dw_dmac.c static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc) · line 474

Quoted 7 lines
drivers/dma/dw_dmac.c +3 -0
@@ -436,6 +469,9 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 
        if (test_bit(DW_DMA_IS_SOFT_LLP, &dwc->flags)) {
                dev_vdbg(chan2dev(&dwc->chan), "%s: soft LLP mode\n", __func__);
+
+               dwc_update_residue(dwc, dwc_first_active(dwc));
+

same is applicable here too and so you can get rid of dwc_update_residue() routine.

Quoted 45 lines
drivers/dma/dw_dmac.c +10 -1
                spin_unlock_irqrestore(&dwc->lock, flags);
                return;
        }
@@ -444,6 +480,9 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
                        (unsigned long long)llp);
 
        list_for_each_entry_safe(desc, _desc, &dwc->active_list, desc_node) {
+               /* initial residue value */
+               dwc->residue = desc->total_len;
+
                /* check first descriptors addr */
                if (desc->txd.phys == llp) {
                        spin_unlock_irqrestore(&dwc->lock, flags);
@@ -453,16 +492,21 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
                /* check first descriptors llp */
                if (desc->lli.llp == llp) {
                        /* This one is currently in progress */
+                       dwc->residue -= dwc_get_sent(dwc);
                        spin_unlock_irqrestore(&dwc->lock, flags);
                        return;
                }
 
-               list_for_each_entry(child, &desc->tx_list, desc_node)
+               dwc->residue -= desc->len;
+               list_for_each_entry(child, &desc->tx_list, desc_node) {
                        if (child->lli.llp == llp) {
                                /* Currently in progress */
+                               dwc->residue -= dwc_get_sent(dwc);
                                spin_unlock_irqrestore(&dwc->lock, flags);
                                return;
                        }
+                       dwc->residue -= child->len;
+               }
 
                /*
                 * No descriptors so far seem to be in progress, i.e.
@@ -1058,6 +1102,7 @@ dwc_tx_status(struct dma_chan *chan,
              struct dma_tx_state *txstate)
 {
        struct dw_dma_chan      *dwc = to_dw_dma_chan(chan);
+       unsigned long           flags;
        enum dma_status         ret;
 
        ret = dma_cookie_status(chan, cookie, txstate);

drivers/dma/dw_dmac.c dwc_tx_status(struct dma_chan *chan, · line 1116

Quoted 5 lines
drivers/dma/dw_dmac.c +2 -0
@@ -1067,8 +1112,12 @@ dwc_tx_status(struct dma_chan *chan,
                ret = dma_cookie_status(chan, cookie, txstate);
        }
 
+       spin_lock_irqsave(&dwc->lock, flags);
+

why do you need locking here?

Quoted 20 lines
drivers/dma/dw_dmac.c +3 -1
        if (ret != DMA_SUCCESS)
-               dma_set_residue(txstate, dwc_first_active(dwc)->len);
+               dma_set_residue(txstate, dwc->residue);
+
+       spin_unlock_irqrestore(&dwc->lock, flags);
 
        if (dwc->paused)
                return DMA_PAUSED;
drivers/dma/dw_dmac_regs.h +1 -0
index 833b4cf..88dd8eb 100644
@@ -203,6 +203,7 @@ struct dw_dma_chan {
        struct list_head        active_list;
        struct list_head        queue;
        struct list_head        free_list;
+       u32                     residue;
        struct dw_cyclic_desc   *cdesc;
 
        unsigned int            descs_allocated;
#9 lore raw
Quoted 19 lines · Viresh Kumar

drivers/dma/dw_dmac.c static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc) · line 447

Quoted 13 lines · Andy Shevchenko
drivers/dma/dw_dmac.c +2 -0
 
 static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 {
        dma_addr_t llp;
@@ -410,6 +441,8 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
                         */
                        desc = dwc_first_active(dwc);
 
+                       dwc_update_residue(dwc, desc);
+
                        if (dwc->tx_node_active != &desc->tx_list) {
                                child = to_dw_desc(dwc->tx_node_active);

Is there a point updating residue here? I don't have a very good knowledge of nollp transfers but this is what i know...

Quoted 7 lines

The above "if" will pass if we are still doing transfers and fail if all transfers are done.
After the end of each LLI we receive an interrupt, where we queue next LLI. Better
would be to initialize dwc->residue at dwc_dostart() with total length, start decrementing
it with desc->len for every lli interrupt we get

It's mostly okay, but we have to handle few cases:
- we have only first (master) descriptor
- we have a chain of the descriptors: master + children
- we have finished last transfer

Quoted 1 line

From my point of view we can't fully get rid of dwc_update_residue(),

but modify it a bit (drop away for loop).

Quoted 5 lines

and if call for
getting residue comes in
middle of transfer, simple return residue - dwc_get_sent(desc) without updating residue
field...

Where? In the tx_status to call something which returns dwc->residue - dwc_get_sent() ?

Quoted 11 lines

drivers/dma/dw_dmac.c static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc) · line 474

Quoted 8 lines
drivers/dma/dw_dmac.c +3 -0
@@ -436,6 +469,9 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
 
        if (test_bit(DW_DMA_IS_SOFT_LLP, &dwc->flags)) {
                dev_vdbg(chan2dev(&dwc->chan), "%s: soft LLP mode\n", __func__);
+
+               dwc_update_residue(dwc, dwc_first_active(dwc));
+

same is applicable here too and so you can get rid of dwc_update_residue() routine.

Same thoughts as above are applicable to this.

Quoted 48 lines
Quoted 48 lines
drivers/dma/dw_dmac.c +10 -1
                spin_unlock_irqrestore(&dwc->lock, flags);
                return;
        }
@@ -444,6 +480,9 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
                        (unsigned long long)llp);
 
        list_for_each_entry_safe(desc, _desc, &dwc->active_list, desc_node) {
+               /* initial residue value */
+               dwc->residue = desc->total_len;
+
                /* check first descriptors addr */
                if (desc->txd.phys == llp) {
                        spin_unlock_irqrestore(&dwc->lock, flags);
@@ -453,16 +492,21 @@ static void dwc_scan_descriptors(struct dw_dma *dw, struct dw_dma_chan *dwc)
                /* check first descriptors llp */
                if (desc->lli.llp == llp) {
                        /* This one is currently in progress */
+                       dwc->residue -= dwc_get_sent(dwc);
                        spin_unlock_irqrestore(&dwc->lock, flags);
                        return;
                }
 
-               list_for_each_entry(child, &desc->tx_list, desc_node)
+               dwc->residue -= desc->len;
+               list_for_each_entry(child, &desc->tx_list, desc_node) {
                        if (child->lli.llp == llp) {
                                /* Currently in progress */
+                               dwc->residue -= dwc_get_sent(dwc);
                                spin_unlock_irqrestore(&dwc->lock, flags);
                                return;
                        }
+                       dwc->residue -= child->len;
+               }
 
                /*
                 * No descriptors so far seem to be in progress, i.e.
@@ -1058,6 +1102,7 @@ dwc_tx_status(struct dma_chan *chan,
              struct dma_tx_state *txstate)
 {
        struct dw_dma_chan      *dwc = to_dw_dma_chan(chan);
+       unsigned long           flags;
        enum dma_status         ret;
 
        ret = dma_cookie_status(chan, cookie, txstate);
@@ -1067,8 +1112,12 @@ dwc_tx_status(struct dma_chan *chan,
                ret = dma_cookie_status(chan, cookie, txstate);
        }
 
Quoted 4 lines

drivers/dma/dw_dmac.c dwc_tx_status(struct dma_chan *chan, · line 1116

Quoted 2 lines
drivers/dma/dw_dmac.c +2 -0
+       spin_lock_irqsave(&dwc->lock, flags);
+

why do you need locking here?

What about the case when one CPU is getting an interrupt and runs scan_descriptors when the other, for example, in the middle of tx_status? So, I'm afraid the dma_set_residue(txstate, dwc->residue) is not atomic and we might end up with random numbers here.

Quoted 20 lines
Quoted 20 lines
drivers/dma/dw_dmac.c +3 -1
        if (ret != DMA_SUCCESS)
-               dma_set_residue(txstate, dwc_first_active(dwc)->len);
+               dma_set_residue(txstate, dwc->residue);
+
+       spin_unlock_irqrestore(&dwc->lock, flags);
 
        if (dwc->paused)
                return DMA_PAUSED;
drivers/dma/dw_dmac_regs.h +1 -0
index 833b4cf..88dd8eb 100644
@@ -203,6 +203,7 @@ struct dw_dma_chan {
        struct list_head        active_list;
        struct list_head        queue;
        struct list_head        free_list;
+       u32                     residue;
        struct dw_cyclic_desc   *cdesc;
 
        unsigned int            descs_allocated;
Signature · 2 lines
Andy Shevchenko <[email protected]>
Intel Finland Oy
#10 lore raw
Quoted 13 lines · Andy Shevchenko
Quoted 10 lines · Viresh Kumar

Is there a point updating residue here? I don't have a very good knowledge of nollp transfers but this is what i know...

The above "if" will pass if we are still doing transfers and fail if all transfers are done.
After the end of each LLI we receive an interrupt, where we queue next LLI. Better
would be to initialize dwc->residue at dwc_dostart() with total length, start decrementing
it with desc->len for every lli interrupt we get

Quoted 7 lines

It's mostly okay, but we have to handle few cases:
- we have only first (master) descriptor
- we have a chain of the descriptors: master + children
- we have finished last transfer

From my point of view we can't fully get rid of dwc_update_residue(), but modify it a bit (drop away for loop).

I feel the residue mechanism for all three cases can be handled by the code i asked you to try. Lets see how it goes, after we see first level of draft from you.

Quoted 8 lines
Quoted 5 lines

and if call for
getting residue comes in
middle of transfer, simple return residue - dwc_get_sent(desc) without updating residue
field...

Where? In the tx_status to call something which returns dwc->residue - dwc_get_sent() ?

Yes. My point was not to update dwc->residue with whatever data we get out of dwc_get_sent().

Quoted 9 lines
Quoted 4 lines

Review

Quoted 2 lines
(quoted diff) +2 -0
+       spin_lock_irqsave(&dwc->lock, flags);
+

why do you need locking here?

What about the case when one CPU is getting an interrupt and runs scan_descriptors when the other, for example, in the middle of tx_status? So, I'm afraid the dma_set_residue(txstate, dwc->residue) is not atomic and we might end up with random numbers here.

I am not against the lock, but want both of us to know why it is there :) So, i don't think that we need to protect dma_set_residue() at all. That's not our job.

What we need to make sure is value read from dwc->residue is consistent. And so, i believe we need a lock to protect read and write to dwc->residue.