Message-ID: <20150905005011.GO15099@codeaurora.org>
Date:	Fri, 4 Sep 2015 17:50:11 -0700
From:	Stephen Boyd <sboyd@...eaurora.org>
To:	Gilad Avidov <gavidov@...eaurora.org>
Cc:	sdharia@...eaurora.org, mlocke@...eaurora.org,
	linux-arm-msm@...r.kernel.org, gregkh@...uxfoundation.org,
	svarbanov@...sol.com, wsa@...-dreams.de,
	devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
	iivanov@...sol.com, agross@...eaurora.org
Subject: Re: [PATCH] spmi-pmic-arb: support configurable number of peripherals

On 09/04, Gilad Avidov wrote:
> On Thu, 3 Sep 2015 17:16:30 -0700
> Stephen Boyd <sboyd@...eaurora.org> wrote:
> 
> > On 09/03, Gilad Avidov wrote:
> > > +			 supported by HW. Default (minimum
> > > supported) is 128. +
> > > +Example V1 PMIC-Arbiter:
> > >  
> > >  	spmi {
> > >  		compatible = "qcom,spmi-pmic-arb";
> > > @@ -62,4 +66,32 @@ Example:
> > >  
> > >  		interrupt-controller;
> > >  		#interrupt-cells = <4>;
> > > +
> > > +		qcom,max-peripherals = <256>;
> > 
> > If it's v1 isn't it always 128? So having 256 here is just
> > confusing.
> 
> Hi Stephen,
> 
> Actually some v1 chipsets, such as Dragonboard APQ8074, support 256
> peripherals.

Then the commit text should be updated. It only mentions that v2
HW has sub-versions which support more than 128.

> 
> > 
> > > +	};
> > > +
> > > @@ -129,14 +131,15 @@ struct spmi_pmic_arb_dev {
> > >  	u8			channel;
> > >  	int			irq;
> > >  	u8			ee;
> > > -	u8			min_apid;
> > > -	u8			max_apid;
> > > -	u32
> > > mapping_table[SPMI_MAPPING_TABLE_LEN];
> > > +	u16			min_irq_apid;
> > > +	u16			max_irq_apid;
> > > +	u16			max_apid;
> > > +	u32			*mapping_table;
> > >  	struct irq_domain	*domain;
> > >  	struct spmi_controller	*spmic;
> > > -	u16			apid_to_ppid[256];
> > > +	u16			*irq_apid_to_ppid;
> > 
> > Please drop all this renaming noise, or at the least, put it in a
> > different patch. More than half the patch is just changing the
> > names of these variables for what seems like no reason.
> > 
> 
> Regarding the change of max/min_apid to max/min_irq_apid:
> max_apid was already used but the name does not make good sense. Since
> really it is not the max_apid supported. Instead it is the largest apid
> which interrupt is currently requested for. But now we need a value that
> is actually the maximum supported apid. This is why repurposed max_apid
> and corrected the previous naming.

max_apid in this patch is only used in probe, so there's no need
for that variable to be a member of the structure. Meaning we can
leave max_apid alone and call this new variable max_periphs or
something like that and leave it local to the probe function.

> 
> > >  	const struct pmic_arb_ver_ops *ver_ops;
> > > -	u8			*ppid_to_chan;
> > > +	u16			*ppid_to_chan;
> > >  };
> > >  
> > >  	struct spmi_pmic_arb_dev *pa = irq_get_handler_data(irq);
> > >  	struct irq_chip *chip = irq_get_chip(irq);
> > >  	void __iomem *intr = pa->intr;
> > > -	int first = pa->min_apid >> 5;
> > > -	int last = pa->max_apid >> 5;
> > > +	int first = pa->min_irq_apid >> 5;
> > > +	int last = pa->max_irq_apid >> 5;
> > >  	u32 status;
> > >  	int i, id;
> > >  
> > > @@ -903,14 +915,30 @@ static int spmi_pmic_arb_probe(struct
> > > platform_device *pdev) 
> > >  	pa->ee = ee;
> > >  
> > > -	for (i = 0; i < ARRAY_SIZE(pa->mapping_table); ++i)
> > > -		pa->mapping_table[i] = readl_relaxed(
> > > -				pa->cnfg +
> > > SPMI_MAPPING_TABLE_REG(i));
> > > +	pa->irq_apid_to_ppid = devm_kzalloc(&ctrl->dev,
> > > pa->max_apid *
> > > +
> > > sizeof(*pa->irq_apid_to_ppid),
> > > +					    GFP_KERNEL);
> > > +	if (!pa->irq_apid_to_ppid) {
> > > +		err = -ENOMEM;
> > > +		goto err_put_ctrl;
> > > +	}
> > > +
> > > +	pa->mapping_table = devm_kzalloc(&ctrl->dev,
> > > +					(pa->max_apid - 1) *
> > > sizeof(u32),
> > > +					GFP_KERNEL);
> > > +	if (!pa->mapping_table) {
> > > +		err = -ENOMEM;
> > > +		goto err_put_ctrl;
> > > +	}
> > > +
> > > +	for (i = 0; i < (pa->max_apid - 1); ++i)
> > > +		pa->mapping_table[i] = readl_relaxed(pa->cnfg +
> > > +
> > > SPMI_MAPPING_TABLE_REG(i));
> 
> > we're searching through the mapping table we can cache the value
> > from the register if the entry isn't 0. This delays the
> 
> Greedy approach (reading upto 512 registers from hw) have very small
> upfront penalty since sequential reads take place very fast
> and HW acceleration (prefetching) will improve it. A lazy approach will
> required much more complex algorithm (for example any value
> including zero is valid. So we will need to augment a marker for
> unassigned entries) and prefetching will not help.

That's solvable with a bitmap of 512 entries, not too much of a
complicated algorithm.

We're sort of screwed now that we have to find the channel for a
ppid by reading X number of PMIC_ARB_REG_CHNL registers. We could
lazily do that too though, where we fill in the array linearly
with what we read until we find the ppid we're looking for. We
would need to do a bitmap trick again, but since we know ppid is
at most going to use 12 bits, we can set the highest bit to
indicate the ppid_to_chan entry is valid.

> 
> > processing to when we're mapping irqs, hopefully speeding up
> > probe for the case where you have a handful of irqs to map.
> 
> Please note that we traverse this mapping table every time a PMIC
> driver registers for an interrupt. This means that even if we speed up
> the probe of pmic-arb driver we will not speed the system boot since
> the probing of the PMIC drivers will register for interrupts.

Do we need to read all 512 entries for a typical system? If we do
then I agree reading it all upfront is probably better, but if we
only read a sparse amount of entries then we can defer it to when
it's going to be used. On upstream 8974 dts files with all the
devices we have in DT right now, I see 75 unique register reads
during boot on apq8074 dragonboard.

> 
> > 
> > The DT property wouldn't be necessary then. Arguably it's being
> > added there to optimize the size of the mapping table and isn't
> > really necessary otherwise.
> > 
> 
> We need to know this value for both memory allocation
> and for stopping before reading over the bounds of the mapping_table in
> HW.
> 

Presumably the hardware is correct and isn't telling us to read
entries in the mapping table that are beyond the bounds, so
stepping over the bounds of the mapping table cache isn't
possible if we always make it the size of the biggest table.

Here's the untested patch. I hope we don't ever have more than
512 periphs.

BTW, we shouldn't make qcom,channel required on v2 hardware. That
doesn't make any sense so it should be changed to required only
on v1 devices and we should remove it from the 8916 dtsi file.

----8<-----
Signed-off-by: Stephen Boyd <sboyd@...eaurora.org>

 drivers/spmi/spmi-pmic-arb.c | 98 +++++++++++++++++++++++++++++---------------
 1 file changed, 66 insertions(+), 32 deletions(-)

diff --git a/drivers/spmi/spmi-pmic-arb.c b/drivers/spmi/spmi-pmic-arb.c
index c7aa6f1a898e..cbe963f829d7 100644
--- a/drivers/spmi/spmi-pmic-arb.c
+++ b/drivers/spmi/spmi-pmic-arb.c
@@ -10,6 +10,7 @@
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  * GNU General Public License for more details.
  */
+#include <linux/bitmap.h>
 #include <linux/delay.h>
 #include <linux/err.h>
 #include <linux/interrupt.h>
@@ -47,9 +48,9 @@
 #define SPMI_MAPPING_BIT_IS_1_FLAG(X)	(((X) >> 8) & 0x1)
 #define SPMI_MAPPING_BIT_IS_1_RESULT(X)	(((X) >> 0) & 0xFF)
 
-#define SPMI_MAPPING_TABLE_LEN		255
 #define SPMI_MAPPING_TABLE_TREE_DEPTH	16	/* Maximum of 16-bits */
-#define PPID_TO_CHAN_TABLE_SZ		BIT(12)	/* PPID is 12bit chan is 1byte*/
+#define PMIC_ARB_MAX_PPID		BIT(12) /* PPID is 12bit */
+#define PMIC_ARB_PPID_VALID		BIT(15)
 
 /* Ownership Table */
 #define SPMI_OWNERSHIP_TABLE_REG(N)	(0x0700 + (4 * (N)))
@@ -85,9 +86,7 @@ enum pmic_arb_cmd_op_code {
 };
 
 /* Maximum number of support PMIC peripherals */
-#define PMIC_ARB_MAX_PERIPHS		256
-#define PMIC_ARB_MAX_CHNL		128
-#define PMIC_ARB_PERIPH_ID_VALID	(1 << 15)
+#define PMIC_ARB_MAX_PERIPHS		512
 #define PMIC_ARB_TIMEOUT_US		100
 #define PMIC_ARB_MAX_TRANS_BYTES	(8)
 
@@ -125,18 +124,21 @@ struct spmi_pmic_arb_dev {
 	void __iomem		*wr_base;
 	void __iomem		*intr;
 	void __iomem		*cnfg;
+	void __iomem		*core;
 	raw_spinlock_t		lock;
 	u8			channel;
 	int			irq;
 	u8			ee;
-	u8			min_apid;
-	u8			max_apid;
-	u32			mapping_table[SPMI_MAPPING_TABLE_LEN];
+	u16			min_apid;
+	u16			max_apid;
+	u32			*mapping_table;
+	DECLARE_BITMAP(mapping_table_valid, PMIC_ARB_MAX_PERIPHS);
 	struct irq_domain	*domain;
 	struct spmi_controller	*spmic;
-	u16			apid_to_ppid[256];
+	u16			*apid_to_ppid;
 	const struct pmic_arb_ver_ops *ver_ops;
-	u8			*ppid_to_chan;
+	u16			*ppid_to_chan;
+	u16			last_channel;
 };
 
 /**
@@ -614,6 +616,10 @@ static int search_mapping_table(struct spmi_pmic_arb_dev *pa,
 	u32 data;
 
 	for (i = 0; i < SPMI_MAPPING_TABLE_TREE_DEPTH; ++i) {
+		if (!test_and_set_bit(index, pa->mapping_table_valid))
+			mapping_table[index] = readl_relaxed(pa->cnfg +
+						SPMI_MAPPING_TABLE_REG(index));
+
 		data = mapping_table[index];
 
 		if (ppid & (1 << SPMI_MAPPING_BIT_INDEX(data))) {
@@ -706,11 +712,37 @@ static u32 pmic_arb_offset_v1(struct spmi_pmic_arb_dev *pa, u8 sid, u16 addr)
 	return 0x800 + 0x80 * pa->channel;
 }
 
+static u16 find_chan(struct spmi_pmic_arb_dev *pa, u16 ppid)
+{
+	u32 regval;
+	u16 chan;
+	u16 _ppid;
+
+	for (chan = pa->last_channel; ; chan++, pa->last_channel++) {
+		regval = readl_relaxed(pa->core + PMIC_ARB_REG_CHNL(chan));
+		if (!regval)
+			continue;
+
+		_ppid = (regval >> 8) & PMIC_ARB_PPID_MASK;
+		pa->ppid_to_chan[_ppid] = chan | PMIC_ARB_PPID_VALID;
+		if (_ppid == ppid)
+			break;
+	}
+
+	return chan;
+}
+
+
 /* v2 offset per ppid (chan) and per ee */
 static u32 pmic_arb_offset_v2(struct spmi_pmic_arb_dev *pa, u8 sid, u16 addr)
 {
 	u16 ppid = (sid << 8) | (addr >> 8);
-	u8  chan = pa->ppid_to_chan[ppid];
+	u16 chan;
+
+	chan = pa->ppid_to_chan[ppid];
+	if (!(chan & PMIC_ARB_PPID_VALID))
+		chan = find_chan(pa, ppid);
+	chan &= ~PMIC_ARB_PPID_MASK;
 
 	return 0x1000 * pa->ee + 0x8000 * chan;
 }
@@ -797,7 +829,7 @@ static int spmi_pmic_arb_probe(struct platform_device *pdev)
 	struct resource *res;
 	void __iomem *core;
 	u32 channel, ee, hw_ver;
-	int err, i;
+	int err;
 	bool is_v1;
 
 	ctrl = spmi_controller_alloc(&pdev->dev, sizeof(*pa));
@@ -825,10 +857,7 @@ static int spmi_pmic_arb_probe(struct platform_device *pdev)
 		pa->wr_base = core;
 		pa->rd_base = core;
 	} else {
-		u8  chan;
-		u16 ppid;
-		u32 regval;
-
+		pa->core = core;
 		pa->ver_ops = &pmic_arb_v2;
 
 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
@@ -847,23 +876,17 @@ static int spmi_pmic_arb_probe(struct platform_device *pdev)
 			goto err_put_ctrl;
 		}
 
-		pa->ppid_to_chan = devm_kzalloc(&ctrl->dev,
-					PPID_TO_CHAN_TABLE_SZ, GFP_KERNEL);
-		if (!pa->ppid_to_chan) {
-			err = -ENOMEM;
-			goto err_put_ctrl;
-		}
 		/*
 		 * PMIC_ARB_REG_CHNL is a table in HW mapping channel to ppid.
 		 * ppid_to_chan is an in-memory invert of that table.
 		 */
-		for (chan = 0; chan < PMIC_ARB_MAX_CHNL; ++chan) {
-			regval = readl_relaxed(core + PMIC_ARB_REG_CHNL(chan));
-			if (!regval)
-				continue;
-
-			ppid = (regval >> 8) & 0xFFF;
-			pa->ppid_to_chan[ppid] = chan;
+		pa->ppid_to_chan = devm_kcalloc(&ctrl->dev,
+						PMIC_ARB_MAX_PPID,
+						sizeof(*pa->ppid_to_chan),
+						GFP_KERNEL);
+		if (!pa->ppid_to_chan) {
+			err = -ENOMEM;
+			goto err_put_ctrl;
 		}
 	}
 
@@ -915,9 +938,20 @@ static int spmi_pmic_arb_probe(struct platform_device *pdev)
 
 	pa->ee = ee;
 
-	for (i = 0; i < ARRAY_SIZE(pa->mapping_table); ++i)
-		pa->mapping_table[i] = readl_relaxed(
-				pa->cnfg + SPMI_MAPPING_TABLE_REG(i));
+	pa->apid_to_ppid = devm_kcalloc(&ctrl->dev, PMIC_ARB_MAX_PERIPHS,
+					    sizeof(*pa->apid_to_ppid),
+					    GFP_KERNEL);
+	if (!pa->apid_to_ppid) {
+		err = -ENOMEM;
+		goto err_put_ctrl;
+	}
+
+	pa->mapping_table = devm_kcalloc(&ctrl->dev, PMIC_ARB_MAX_PERIPHS - 1,
+					sizeof(*pa->mapping_table), GFP_KERNEL);
+	if (!pa->mapping_table) {
+		err = -ENOMEM;
+		goto err_put_ctrl;
+	}
 
 	/* Initialize max_apid/min_apid to the opposite bounds, during
 	 * the irq domain translation, we are sure to update these */
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists