Skip to contentSkip to conversation

sunxi: Use clrsetbits_le32 instead of multiple instruction

PATCH Superseded Michael Trimarchi 4 participants 12 messages Feb 14, 2019 – Feb 18, 2019

Subsystems (3)

v2 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/3 sunxi: Use clrsetbits_le32 instead of multiple instruction

---

V1->V2: none

---
 arch/arm/mach-sunxi/dram_sun8i_a33.c | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)
arch/arm/mach-sunxi/dram_sun8i_a33.c +2 -6
index 1da2727f98..83212aaddf 100644
@@ -148,12 +148,8 @@ static void auto_set_timing_para(struct dram_para *para)
 	reg_val = (tcksrx << 24) | (tcksre << 16) | (tckesr << 8) | (tcke << 0);
 	writel(reg_val, &mctl_ctl->dramtmg5);
 	/* Set two rank timing and exit self-refresh timing */
-	reg_val = readl(&mctl_ctl->dramtmg8);
-	reg_val &= ~(0xff << 8);
-	reg_val &= ~(0xff << 0);
-	reg_val |= (0x33 << 8);
-	reg_val |= (0x8 << 0);
-	writel(reg_val, &mctl_ctl->dramtmg8);
+	clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
+			0x33 << 8 | (0x8 << 0));
 	/* Set phy interface time */
 	reg_val = (0x2 << 24) | (t_rdata_en << 16) | (0x1 << 8)
 			| (wr_latency << 0);
Signature · 1 line
2.17.1
#1 lore raw

PATCH v2 2/3 sunxi: Don't change the rank in dram size detection in A33

Change the size create a glitch in the clken signal on second bank. According to the ddr manual the clken need to be sent accros the reset signal coming the cpu. The rank is calculated just before this function is called and the mctl_set_cr should not change this value anymore

---

V1->V2: Adjust commit description

---
 arch/arm/mach-sunxi/dram_sun8i_a33.c | 3 ---
 1 file changed, 3 deletions(-)
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -3
index 83212aaddf..d73a93a132 100644
@@ -45,14 +45,12 @@ static void mctl_set_cr(struct dram_para *para)
 
 static void auto_detect_dram_size(struct dram_para *para)
 {
-	u8 orig_rank = para->rank;
 	int rows, columns;
 
 	/* Row detect */
 	para->page_size = 512;
 	para->seq = 1;
 	para->rows = 16;
-	para->rank = 1;
 	mctl_set_cr(para);
 	for (rows = 11 ; rows < 16 ; rows++) {
 		if (mctl_mem_matches(1 << (rows + 9))) /* row-column */
@@ -69,7 +67,6 @@ static void auto_detect_dram_size(struct dram_para *para)
 	}
 
 	para->seq = 0;
-	para->rank = orig_rank;
 	para->rows = rows;
 	para->page_size = 1 << columns;
 	mctl_set_cr(para);
Signature · 1 line
2.17.1
#2 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

In reply to Michael Trimarchi

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

---

V1->V2: adjust commit message

---
 arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
index d73a93a132..355fe30aba 100644
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
 	writel(reg_val, &mctl_ctl->dramtmg5);
 	/* Set two rank timing and exit self-refresh timing */
 	clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
-			0x33 << 8 | (0x8 << 0));
+			0x33 << 8 | (0x10 << 0));
 	/* Set phy interface time */
 	reg_val = (0x2 << 24) | (t_rdata_en << 16) | (0x1 << 8)
 			| (wr_latency << 0);
Signature · 1 line
2.17.1
#3 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

Quoted 24 lines · Michael Trimarchi

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

---

V1->V2: adjust commit message

---
arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -0
index d73a93a132..355fe30aba 100644

arch/arm/mach-sunxi/dram_sun8i_a33.c static void auto_set_timing_para(struct dram_para *para) · line 149

Quoted 5 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
 	writel(reg_val, &mctl_ctl->dramtmg5);
 	/* Set two rank timing and exit self-refresh timing */
 	clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
-			0x33 << 8 | (0x8 << 0));
+			0x33 << 8 | (0x10 << 0));

How exactly does the change in constants match up with the commit message? What is the field@“<< 0”?

Are you configuring the IOs to 1.5V with this write (as the commit message would suggest)?

Quoted 10 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -0
 	/* Set phy interface time */
 	reg_val = (0x2 << 24) | (t_rdata_en << 16) | (0x1 << 8)
 			| (wr_latency << 0);
Signature · 6 lines
2.17.1

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot
#4 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

Hi Philipp

Quoted 20 lines · Philipp Tomsich
Quoted 20 lines · Michael Trimarchi

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

---

V1->V2: adjust commit message

---
arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 14 lines

arch/arm/mach-sunxi/dram_sun8i_a33.c static void auto_set_timing_para(struct dram_para *para) · line 149

Quoted 10 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
index d73a93a132..355fe30aba 100644
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
      writel(reg_val, &mctl_ctl->dramtmg5);
      /* Set two rank timing and exit self-refresh timing */
      clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
-                     0x33 << 8 | (0x8 << 0));
+                     0x33 << 8 | (0x10 << 0));

How exactly does the change in constants match up with the commit message? What is the field at “<< 0”?

I just reduce the code change. Ok, forget to mention in the commit message that I don't
have documentation of the cpu. So my fix is based on allwinner architecture comparison and
the fact that it fix my issue. I was having a failure rate of 7% before.

Quoted 2 lines

Are you configuring the IOs to 1.5V with this write (as the commit message would suggest)?

No, the model of the memory is LPDDR3 but they can work as DDR3 compatible mode if they are powered at 1.5V. A33 has not support for LPDDR3 so make no sense to try
to let them work as LPDDR3. I will update the commit message

Michael

Quoted 12 lines
Quoted 11 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -1
      /* Set phy interface time */
      reg_val = (0x2 << 24) | (t_rdata_en << 16) | (0x1 << 8)
                      | (wr_latency << 0);
--

2.17.1

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

Signature · 4 lines
| Michael Nazzareno Trimarchi                     Amarula Solutions BV |
| COO  -  Founder                                      Cruquiuskade 47 |
| +31(0)851119172                                 Amsterdam 1018 AM NL |
|                  [`as] http://www.amarulasolutions.com               |
#5 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

In reply to Philipp Tomsich
Quoted 19 lines · Philipp Tomsich
Quoted 19 lines · Michael Trimarchi

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

---

V1->V2: adjust commit message

---
arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 13 lines

arch/arm/mach-sunxi/dram_sun8i_a33.c static void auto_set_timing_para(struct dram_para *para) · line 149

Quoted 10 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
index d73a93a132..355fe30aba 100644
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
 	writel(reg_val, &mctl_ctl->dramtmg5);
 	/* Set two rank timing and exit self-refresh timing */
 	clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
-			0x33 << 8 | (0x8 << 0));
+			0x33 << 8 | (0x10 << 0));

How exactly does the change in constants match up with the commit message? What is the field at “<< 0”?

From looking at the ZynqMP register guide, which is reported to be close to the various Allwinner DRAM controllers, bits [6:0] are tXS: Self refresh exit delay. Increasing that should not hurt, and if I understand it correctly it only affects the time after the self-refresh *exit*, which happens only after (re-)initialisation(?).

But it should be noted that this unconditionally affects all A33 boards.

Quoted 2 lines

Are you configuring the IOs to 1.5V with this write (as the commit message would suggest)?

I think the voltage is totally unrelated here, this is a pure timing register.

Cheers,
Andre.

Quoted 12 lines
Quoted 11 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -0
 	/* Set phy interface time */
 	reg_val = (0x2 << 24) | (t_rdata_en << 16) | (0x1 << 8)
 			| (wr_latency << 0);
Signature · 6 lines
2.17.1

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot
#6 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

Quoted 38 lines · André Przywara
Quoted 19 lines · Philipp Tomsich
Quoted 19 lines · Michael Trimarchi

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

---

V1->V2: adjust commit message

---
arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 13 lines

arch/arm/mach-sunxi/dram_sun8i_a33.c static void auto_set_timing_para(struct dram_para *para) · line 149

Quoted 10 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
index d73a93a132..355fe30aba 100644
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
 	writel(reg_val, &mctl_ctl->dramtmg5);
 	/* Set two rank timing and exit self-refresh timing */
 	clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
-			0x33 << 8 | (0x8 << 0));
+			0x33 << 8 | (0x10 << 0));

How exactly does the change in constants match up with the commit message? What is the field at “<< 0”?

Quoted 5 lines

From looking at the ZynqMP register guide, which is reported to be close to the various Allwinner DRAM controllers, bits [6:0] are tXS: Self refresh exit delay. Increasing that should not hurt, and if I understand it correctly it only affects the time after the self-refresh *exit*, which happens only after (re-)initialisation(?).

The “set two rank timing” comment doesn’t match our expectation that this will be self-refresh timings, as on the ZynqMP or the A80. Self-refresh only happens, if someone puts the DRAM into self-refresh (i.e. “suspend to DRAM”) and then resumes it. I don’t see a reference to the error occurring from this.

That said, if the field indeed was the exit self-refresh timing, this would usually be tXSDLL, encoded as tXSDLL/32. JESD79 specifies tESXR as tXSDLL (which in turn is tDLLK(min)), which is 512 nCK always: 0x10 would be 512 and 0x08 would only be 256… then again, this should only matter when exiting self-refresh.

Note that the ‘auto-enter self-refresh’ functionality is controlled through BIT(3) in PWRCTL on the A33, according to Allwinner’s basic_loader. I don’t know whether that may be turned on by the driver (i.e. didn’t check).

OTOH, when working with an underdocumented DRAM register layout and once one has an educated guess at what register/setting may be affected, a DRAM protocol analyzer can provide conclusive answers by looking at the differences in behaviour with 0x8 and 0x10 in that bitfield…

Just my 2 cents.

Quoted 23 lines

But it should be noted that this unconditionally affects all A33 boards.

Quoted 2 lines

Are you configuring the IOs to 1.5V with this write (as the commit message would suggest)?

I think the voltage is totally unrelated here, this is a pure timing register.

Cheers,
Andre.

Quoted 11 lines
Quoted 11 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -0
 	/* Set phy interface time */
 	reg_val = (0x2 << 24) | (t_rdata_en << 16) | (0x1 << 8)
 			| (wr_latency << 0);
Signature · 6 lines
2.17.1

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de <mailto:[email protected]>
https://lists.denx.de/listinfo/u-boot <https://lists.denx.de/listinfo/u-boot>
#7 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

On Fri, 15 Feb 2019 11:18:38 +0100
Philipp Tomsich <[email protected]> wrote:

Hi Philipp,

Quoted 42 lines
Quoted 36 lines · André Przywara
Quoted 19 lines · Philipp Tomsich
Quoted 19 lines · Michael Trimarchi

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

---

V1->V2: adjust commit message

---
arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 13 lines

arch/arm/mach-sunxi/dram_sun8i_a33.c static void auto_set_timing_para(struct dram_para *para) · line 149

Quoted 10 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
index d73a93a132..355fe30aba 100644
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
 	writel(reg_val, &mctl_ctl->dramtmg5);
 	/* Set two rank timing and exit self-refresh timing */
 	clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
-			0x33 << 8 | (0x8 << 0));
+			0x33 << 8 | (0x10 << 0));  

How exactly does the change in constants match up with the commit message? What is the field at “<< 0”?

Quoted 5 lines

From looking at the ZynqMP register guide, which is reported to be close to the various Allwinner DRAM controllers, bits [6:0] are tXS: Self refresh exit delay. Increasing that should not hurt, and if I understand it correctly it only affects the time after the self-refresh *exit*, which happens only after (re-)initialisation(?).

Quoted 5 lines

The “set two rank timing” comment doesn’t match our expectation that this will be self-refresh timings, as on the ZynqMP or the A80. Self-refresh only happens, if someone puts the DRAM into self-refresh (i.e. “suspend to DRAM”) and then resumes it. I don’t see a reference to the error occurring from this.

So I was wondering about this as well. Indeed we don't seem to *explicitly* enter or exit Self-Refresh anywhere, but maybe this is done implicitly as part of some training phase?
I might be wrong about this, but in some DDR3 documentation I see that changing the DLL state is connected to self refresh, and enabling the DLL is required as part of the initialisation? So is the DRAMC somehow triggering a self refresh exit "magically"? If I understand this correctly, our ranking test does reset the controller?

Quoted 5 lines

That said, if the field indeed was the exit self-refresh timing, this would usually be tXSDLL, encoded as tXSDLL/32. JESD79 specifies tESXR as tXSDLL (which in turn is tDLLK(min)), which is 512 nCK always: 0x10 would be 512 and 0x08 would only be 256… then again, this should only matter when exiting self-refresh.

So I was looking at this:
https://www.xilinx.com/html_docs/registers/ug1087/ddrc___dramtmg8.html
and tXSDLL is bits [14:8], while the non-DLL tXS is bits [6:0].

Quoted 4 lines

Note that the ‘auto-enter self-refresh’ functionality is controlled through BIT(3) in PWRCTL on the A33, according to Allwinner’s basic_loader. I don’t know whether that may be turned on by the driver (i.e. didn’t check).

I don't see us touching pwrctl for the A33 (but for the A23).

Quoted 5 lines

OTOH, when working with an underdocumented DRAM register layout and once one has an educated guess at what register/setting may be affected, a DRAM protocol analyzer can provide conclusive answers by looking at the differences in behaviour with 0x8 and 0x10 in that bitfield…

All I have is a multimeter and no A33 ;-)

Thanks for providing some background!

Cheers,
Andre.

Quoted 25 lines

Just my 2 cents.

Quoted 23 lines

But it should be noted that this unconditionally affects all A33 boards.

Quoted 2 lines

Are you configuring the IOs to 1.5V with this write (as the commit message would suggest)?

I think the voltage is totally unrelated here, this is a pure timing register.

Cheers,
Andre.

Quoted 11 lines
Quoted 11 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -0
 	/* Set phy interface time */
 	reg_val = (0x2 << 24) | (t_rdata_en << 16) | (0x1 << 8)
 			| (wr_latency << 0);
Signature · 6 lines
2.17.1

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de <mailto:[email protected]>
https://lists.denx.de/listinfo/u-boot <https://lists.denx.de/listinfo/u-boot>  
#8 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

Dipl.-Ing. Dr.techn. Philipp Tomsich
Theobroma Systems Design und Consulting GmbH Seestadtstrasse 27 (Aspern IQ), A-1220 Wien, Austria
Phone: +43 1 2369893-401, Fax: +43 1 2369893-9-401
Cell phone: +43 664 8346109
http://www.theobroma-systems.com <http://www.theobroma-systems.com/>

Quoted 49 lines · Andre Przywara

On Fri, 15 Feb 2019 11:18:38 +0100
Philipp Tomsich <[email protected]> wrote:

Hi Philipp,

Quoted 42 lines
Quoted 41 lines · André Przywara
Quoted 19 lines · Philipp Tomsich
Quoted 19 lines · Michael Trimarchi

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

---

V1->V2: adjust commit message

---
arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 13 lines

arch/arm/mach-sunxi/dram_sun8i_a33.c static void auto_set_timing_para(struct dram_para *para) · line 149

Quoted 10 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
index d73a93a132..355fe30aba 100644
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
 	writel(reg_val, &mctl_ctl->dramtmg5);
 	/* Set two rank timing and exit self-refresh timing */
 	clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
-			0x33 << 8 | (0x8 << 0));
+			0x33 << 8 | (0x10 << 0));  

How exactly does the change in constants match up with the commit message? What is the field at “<< 0”?

From looking at the ZynqMP register guide, which is reported to be close to the various Allwinner DRAM controllers, bits [6:0] are tXS: Self refresh exit delay. Increasing that should not hurt, and if I understand it correctly it only affects the time after the self-refresh *exit*, which happens only after (re-)initialisation(?).

Quoted 8 lines
Quoted 5 lines

The “set two rank timing” comment doesn’t match our expectation that this will be self-refresh timings, as on the ZynqMP or the A80. Self-refresh only happens, if someone puts the DRAM into self-refresh (i.e. “suspend to DRAM”) and then resumes it. I don’t see a reference to the error occurring from this.

So I was wondering about this as well. Indeed we don't seem to *explicitly* enter or exit Self-Refresh anywhere, but maybe this is done implicitly as part of some training phase?
I might be wrong about this, but in some DDR3 documentation I see that changing the DLL state is connected to self refresh, and enabling the DLL is required as part of the initialisation? So is the DRAMC somehow triggering a self refresh exit "magically"? If I understand this correctly, our ranking test does reset the controller?

I’d need to reread JESD79 (a much more relaxing read than the daily news), to figure out whether there’s any valid reason to play with self-refresh during the training.

Quoted 17 lines
Quoted 5 lines

That said, if the field indeed was the exit self-refresh timing, this would usually be tXSDLL, encoded as tXSDLL/32. JESD79 specifies tESXR as tXSDLL (which in turn is tDLLK(min)), which is 512 nCK always: 0x10 would be 512 and 0x08 would only be 256… then again, this should only matter when exiting self-refresh.

So I was looking at this:
https://www.xilinx.com/html_docs/registers/ug1087/ddrc___dramtmg8.html
and tXSDLL is bits [14:8], while the non-DLL tXS is bits [6:0].

Quoted 4 lines

Note that the ‘auto-enter self-refresh’ functionality is controlled through BIT(3) in PWRCTL on the A33, according to Allwinner’s basic_loader. I don’t know whether that may be turned on by the driver (i.e. didn’t check).

I don't see us touching pwrctl for the A33 (but for the A23).

Quoted 7 lines
Quoted 5 lines

OTOH, when working with an underdocumented DRAM register layout and once one has an educated guess at what register/setting may be affected, a DRAM protocol analyzer can provide conclusive answers by looking at the differences in behaviour with 0x8 and 0x10 in that bitfield…

All I have is a multimeter and no A33 ;-)

You are aware that was more of a philosophical rambling and not intended as a request to you…
However, I am way more surprised that none of the parties having money riding on their A33 designs has gone through the required effort.

Cheers,
Philipp.

#9 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

Hi all

Quoted 26 lines · Philipp Tomsich

Dipl.-Ing. Dr.techn. Philipp Tomsich Theobroma Systems Design und Consulting GmbH Seestadtstrasse 27 (Aspern IQ), A-1220 Wien, Austria
Phone: +43 1 2369893-401, Fax: +43 1 2369893-9-401 Cell phone: +43 664 8346109
http://www.theobroma-systems.com

On 15.02.2019, at 12:31, Andre Przywara <[email protected]> wrote:

On Fri, 15 Feb 2019 11:18:38 +0100
Philipp Tomsich <[email protected]> wrote:

Hi Philipp,

On 14.02.2019, at 22:24, André Przywara <[email protected]> wrote:

On 14/02/2019 16:36, Philipp Tomsich wrote:

Quoted 8 lines

On 14.02.2019, at 16:58, Michael Trimarchi <[email protected]> wrote:

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

I will fix the commit

Quoted 25 lines

---

V1->V2: adjust commit message

---
arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
index d73a93a132..355fe30aba 100644
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
 writel(reg_val, &mctl_ctl->dramtmg5);
 /* Set two rank timing and exit self-refresh timing */
 clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
- 0x33 << 8 | (0x8 << 0));
+ 0x33 << 8 | (0x10 << 0));
 
 

How exactly does the change in constants match up with the commit message? What is the field at “<< 0”?

Quoted 6 lines

From looking at the ZynqMP register guide, which is reported to be close to the various Allwinner DRAM controllers, bits [6:0] are tXS: Self refresh exit delay. Increasing that should not hurt, and if I understand it correctly it only affects the time after the self-refresh *exit*, which happens only after (re-)initialisation(?).

Seems that affect the initialization sequence.

Quoted 3 lines

The “set two rank timing” comment doesn’t match our expectation that this will be self-refresh timings, as on the ZynqMP or the A80.

The comment is the original one

Quoted 13 lines

Self-refresh only happens, if someone puts the DRAM into self-refresh (i.e. “suspend to DRAM”) and then resumes it. I don’t see a reference to the error occurring from this.

So I was wondering about this as well. Indeed we don't seem to *explicitly* enter or exit Self-Refresh anywhere, but maybe this is done implicitly as part of some training phase?
I might be wrong about this, but in some DDR3 documentation I see that changing the DLL state is connected to self refresh, and enabling the DLL is required as part of the initialisation? So is the DRAMC somehow triggering a self refresh exit "magically"? If I understand this correctly, our ranking test does reset the controller?

I’d need to reread JESD79 (a much more relaxing read than the daily news), to figure out whether there’s any valid reason to play with self-refresh during the training.

Quoted 6 lines

That said, if the field indeed was the exit self-refresh timing, this would usually be tXSDLL, encoded as tXSDLL/32. JESD79 specifies tESXR as tXSDLL (which in turn is tDLLK(min)), which is 512 nCK always: 0x10 would be 512 and 0x08 would only be 256… then again, this should only matter when exiting self-refresh.

What is the best option here? create a better commit or try to go in deep.

Quoted 11 lines

So I was looking at this:
https://www.xilinx.com/html_docs/registers/ug1087/ddrc___dramtmg8.html
and tXSDLL is bits [14:8], while the non-DLL tXS is bits [6:0].

Note that the ‘auto-enter self-refresh’ functionality is controlled through BIT(3) in PWRCTL on the A33, according to Allwinner’s basic_loader. I don’t know whether that may be turned on by the driver (i.e. didn’t check).

Quoted 7 lines

I don't see us touching pwrctl for the A33 (but for the A23).

OTOH, when working with an underdocumented DRAM register layout and once one has an educated guess at what register/setting may be affected, a DRAM protocol analyzer can provide conclusive answers by looking at the differences in behaviour with 0x8 and 0x10 in that bitfield…

Nice to have but It's not my case. The best for me is update the commit message and give more info there. Add the memory components and the layout of my setup, ask to test other boards before get included

Michael

Quoted 13 lines

All I have is a multimeter and no A33 ;-)

You are aware that was more of a philosophical rambling and not intended as a request to you…
However, I am way more surprised that none of the parties having money riding on their A33 designs has gone through the required effort.

Cheers,
Philipp.

Signature · 4 lines
| Michael Nazzareno Trimarchi                     Amarula Solutions BV |
| COO  -  Founder                                      Cruquiuskade 47 |
| +31(0)851119172                                 Amsterdam 1018 AM NL |
|                  [`as] http://www.amarulasolutions.com               |
#10 lore raw

PATCH v2 3/3 sunxi: Fix A33 memory initialization

Quoted 94 lines · On Fri, 15 Feb 2019 17:01:40 +0100 Michael Nazzareno Trimarchi…

Hi all

Quoted 26 lines · Philipp Tomsich

Dipl.-Ing. Dr.techn. Philipp Tomsich Theobroma Systems Design und Consulting GmbH Seestadtstrasse 27 (Aspern IQ), A-1220 Wien, Austria
Phone: +43 1 2369893-401, Fax: +43 1 2369893-9-401 Cell phone: +43 664 8346109
http://www.theobroma-systems.com

On 15.02.2019, at 12:31, Andre Przywara <[email protected]> wrote:

On Fri, 15 Feb 2019 11:18:38 +0100
Philipp Tomsich <[email protected]> wrote:

Hi Philipp,

On 14.02.2019, at 22:24, André Przywara <[email protected]> wrote:

On 14/02/2019 16:36, Philipp Tomsich wrote:

Quoted 8 lines

On 14.02.2019, at 16:58, Michael Trimarchi <[email protected]> wrote:

Set two rank timing and exit self-refresh timing seems not done properly. We know use the same write that we are using on H5 silicon. Tested was done in A33 allwinner cpu, dual rank connection connected with two MT41K512M16HA-125:A memory model. Memory is configure as DDR3 1.5V

I will fix the commit

Quoted 25 lines

---

V1->V2: adjust commit message

---
arch/arm/mach-sunxi/dram_sun8i_a33.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-sunxi/dram_sun8i_a33.c +1 -1
index d73a93a132..355fe30aba 100644
@@ -146,7 +146,7 @@ static void auto_set_timing_para(struct dram_para *para)
 writel(reg_val, &mctl_ctl->dramtmg5);
 /* Set two rank timing and exit self-refresh timing */
 clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
- 0x33 << 8 | (0x8 << 0));
+ 0x33 << 8 | (0x10 << 0));
 
 

How exactly does the change in constants match up with the commit message? What is the field at “<< 0”?

Quoted 6 lines

From looking at the ZynqMP register guide, which is reported to be close to the various Allwinner DRAM controllers, bits [6:0] are tXS: Self refresh exit delay. Increasing that should not hurt, and if I understand it correctly it only affects the time after the self-refresh *exit*, which happens only after (re-)initialisation(?).

Seems that affect the initialization sequence.

Quoted 3 lines

The “set two rank timing” comment doesn’t match our expectation that this will be self-refresh timings, as on the ZynqMP or the A80.

The comment is the original one

Quoted 13 lines

Self-refresh only happens, if someone puts the DRAM into self-refresh (i.e. “suspend to DRAM”) and then resumes it. I don’t see a reference to the error occurring from this.

So I was wondering about this as well. Indeed we don't seem to *explicitly* enter or exit Self-Refresh anywhere, but maybe this is done implicitly as part of some training phase?
I might be wrong about this, but in some DDR3 documentation I see that changing the DLL state is connected to self refresh, and enabling the DLL is required as part of the initialisation? So is the DRAMC somehow triggering a self refresh exit "magically"? If I understand this correctly, our ranking test does reset the controller?

I’d need to reread JESD79 (a much more relaxing read than the daily news), to figure out whether there’s any valid reason to play with self-refresh during the training.

Quoted 8 lines
Quoted 6 lines

That said, if the field indeed was the exit self-refresh timing, this would usually be tXSDLL, encoded as tXSDLL/32. JESD79 specifies tESXR as tXSDLL (which in turn is tDLLK(min)), which is 512 nCK always: 0x10 would be 512 and 0x08 would only be 256… then again, this should only matter when exiting self-refresh.

What is the best option here? create a better commit or try to go in deep.

Well, we don't need to get this out of hand:
1) This change affects all A33 systems, unconditionally. We would need confirmation from owners of those other systems that it still works for them.
2) Yes, the commit message should be changed. Please state the issue you have seen, maybe roughly summarise our guesses here. Then mention that this changes the DRAM init part for all A33 users.

Cheers,
Andre.

Quoted 41 lines
Quoted 12 lines

So I was looking at this:
https://www.xilinx.com/html_docs/registers/ug1087/ddrc___dramtmg8.html
and tXSDLL is bits [14:8], while the non-DLL tXS is bits [6:0].

Note that the ‘auto-enter self-refresh’ functionality is controlled through BIT(3) in PWRCTL on the A33, according to Allwinner’s basic_loader. I don’t know whether that may be turned on by the driver (i.e. didn’t check).

Quoted 7 lines

I don't see us touching pwrctl for the A33 (but for the A23).

OTOH, when working with an underdocumented DRAM register layout and once one has an educated guess at what register/setting may be affected, a DRAM protocol analyzer can provide conclusive answers by looking at the differences in behaviour with 0x8 and 0x10 in that bitfield…

Nice to have but It's not my case. The best for me is update the commit message and give more info there. Add the memory components and the layout of my setup, ask to test other boards before get included

Michael

Quoted 13 lines

All I have is a multimeter and no A33 ;-)

You are aware that was more of a philosophical rambling and not intended as a request to you…
However, I am way more surprised that none of the parties having money riding on their A33 designs has gone through the required effort.

Cheers,
Philipp.

#11 lore raw

PATCH v2 1/3 sunxi: Use clrsetbits_le32 instead of multiple instruction

In reply to Michael Trimarchi
Quoted 15 lines · Michael Trimarchi

---

V1->V2: none

---
 arch/arm/mach-sunxi/dram_sun8i_a33.c | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -0
index 1da2727f98..83212aaddf 100644

arch/arm/mach-sunxi/dram_sun8i_a33.c static void auto_set_timing_para(struct dram_para *para) · line 153

Quoted 6 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -3
@@ -148,12 +148,8 @@ static void auto_set_timing_para(struct dram_para *para)
        reg_val = (tcksrx << 24) | (tcksre << 16) | (tckesr << 8) | (tcke << 0);
        writel(reg_val, &mctl_ctl->dramtmg5);
        /* Set two rank timing and exit self-refresh timing */
-       reg_val = readl(&mctl_ctl->dramtmg8);
-       reg_val &= ~(0xff << 8);
-       reg_val &= ~(0xff << 0);

these are clr

arch/arm/mach-sunxi/dram_sun8i_a33.c static void auto_set_timing_para(struct dram_para *para) · line 155

Quoted 2 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +0 -2
-       reg_val |= (0x33 << 8);
-       reg_val |= (0x8 << 0);

these are set

Quoted 3 lines
arch/arm/mach-sunxi/dram_sun8i_a33.c +2 -1
-       writel(reg_val, &mctl_ctl->dramtmg8);
+       clrsetbits_le32(&mctl_ctl->dramtmg8, (0xff << 8) | (0xff << 0),
+                       0x33 << 8 | (0x8 << 0));