Skip to contentSkip to conversation

PCI: mediatek-gen3: handle PERST after reset

PATCH In review Frank Wunderlich 4 participants 8 messages Apr 2, 2023 – May 1, 2023

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 PCI: mediatek-gen3: handle PERST after reset

From: Frank Wunderlich <[email protected]>

De-assert PERST in separate step after reset signals to fully comply the PCIe CEM clause 2.2.

This fixes some NVME detection issues on mt7986.

Fixes d3bf75b579b9 ("PCI: mediatek-gen3: Add MediaTek Gen3 driver for MT8192")

---
Patch is taken from user Ruslan aka RRKh61 (permitted me to send it with me as author).

https://forum.banana-pi.org/t/bpi-r3-nvme-connection-issue/14563/17

---
 drivers/pci/controller/pcie-mediatek-gen3.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/pci/controller/pcie-mediatek-gen3.c +7 -1
index b8612ce5f4d0..176b1a04565d 100644
@@ -350,7 +350,13 @@ static int mtk_pcie_startup_port(struct mtk_gen3_pcie *pcie)
 	msleep(100);
 
 	/* De-assert reset signals */
-	val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB | PCIE_PE_RSTB);
+	val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB);
+	writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
+
+	msleep(100);
+
+	/* De-assert PERST# signals */
+	val &= ~(PCIE_PE_RSTB);
 	writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
 
 	/* Check if the link is up or not */
Signature · 1 line
2.34.1
#1 lore raw

Hi

Quoted 27 lines

Gesendet: Sonntag, 02. April 2023 um 15:13 Uhr
Von: "Frank Wunderlich" <[email protected]>
De-assert PERST in separate step after reset signals to fully comply the PCIe CEM clause 2.2.

This fixes some NVME detection issues on mt7986.

Fixes d3bf75b579b9 ("PCI: mediatek-gen3: Add MediaTek Gen3 driver for MT8192")

---
Patch is taken from user Ruslan aka RRKh61 (permitted me to send it with me as author).

https://forum.banana-pi.org/t/bpi-r3-nvme-connection-issue/14563/17

---
 drivers/pci/controller/pcie-mediatek-gen3.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/pci/controller/pcie-mediatek-gen3.c +0 -1
index b8612ce5f4d0..176b1a04565d 100644
@@ -350,7 +350,13 @@ static int mtk_pcie_startup_port(struct mtk_gen3_pcie *pcie)
 	msleep(100);
 
 	/* De-assert reset signals */
-	val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB | PCIE_PE_RSTB);

drivers/pci/controller/pcie-mediatek-gen3.c static int mtk_pcie_startup_port(struct mtk_gen3_pcie *pcie) · line 362

Quoted 10 lines
drivers/pci/controller/pcie-mediatek-gen3.c +7 -0
+	val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB);
+	writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
+
+	msleep(100);
+
+	/* De-assert PERST# signals */
+	val &= ~(PCIE_PE_RSTB);
 	writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
 
 	/* Check if the link is up or not */

Hi

just a friendly reminder....do i miss any recipient?

regards Frank

#2 lore raw

Hi Frank,

Seems this patch has huge impact on boot time, I'm curious about the NVMe detection issue on mt7986, can you share the SSD model that you are using and the bootup logs with that SSD?

Thanks.

Quoted 4 lines · Frank Wunderlich

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 18 lines

Hi

Quoted 16 lines

Gesendet: Sonntag, 02. April 2023 um 15:13 Uhr
Von: "Frank Wunderlich" <[email protected]>
De-assert PERST in separate step after reset signals to fully comply
the PCIe CEM clause 2.2.

This fixes some NVME detection issues on mt7986.

Fixes d3bf75b579b9 ("PCI: mediatek-gen3: Add MediaTek Gen3 driver

for MT8192")

---
Patch is taken from user Ruslan aka RRKh61 (permitted me to send it with me as author).

https://urldefense.com/v3/__https://forum.banana-pi.org/t/bpi-r3-nvme-connection-issue/14563/17__;!!CTRNKA9wMg0ARbw!nCXEM685pkUpoiZYGKptPYccNrWMeN2D3jIO5_irwxZJ7c6ZzEeACIx-V2WeZHAP_0FKlDDIQ0RbDJ892prtoToDv30$

Quoted 33 lines
Quoted 25 lines
---
 drivers/pci/controller/pcie-mediatek-gen3.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/pci/controller/pcie-mediatek-gen3.c +7 -1
index b8612ce5f4d0..176b1a04565d 100644
@@ -350,7 +350,13 @@ static int mtk_pcie_startup_port(struct
 mtk_gen3_pcie *pcie)
      msleep(100);
 
      /* De-assert reset signals */
-     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB |
 PCIE_PE_RSTB);
+     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB);
+     writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
+
+     msleep(100);
+
+     /* De-assert PERST# signals */
+     val &= ~(PCIE_PE_RSTB);
      writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
Quoted 1 line
      /* Check if the link is up or not */

Hi

just a friendly reminder....do i miss any recipient?

regards Frank

#3 lore raw

Am 27. April 2023 10:31:07 MESZ schrieb "Jianjun Wang (王建军)" <[email protected]>:

Quoted 5 lines

Hi Frank,

Seems this patch has huge impact on boot time, I'm curious about the NVMe detection issue on mt7986, can you share the SSD model that you are using and the bootup logs with that SSD?

Which "huge" delay do you get in which setup? It adds a 100ms delay yes,but this seems needed to some devices working.i found several sources talking about the 100ms wake-up time...

I do not have this issue,but some users in bpi-forum reorted it. Pcie-controller on mt7986/bpi-r3 does simply not detect the nvme and returned ETIMEDOUT (110).

# dmesg | grep 'pci'
[ 5.235564] mtk-pcie-gen3 11280000.pcie: host bridge /soc/pcie@11280000 ranges:
[ 5.242938] mtk-pcie-gen3 11280000.pcie: Parsing ranges property...
[ 5.249235] mtk-pcie-gen3 11280000.pcie: MEM 0x0020000000..0x002fffffff -> 0x0020000000 
[ 5.478062] mtk-pcie-gen3 11280000.pcie: PCIe link down, current LTSSM state: detect.active (0x10 00001)
[ 5.487491] mtk-pcie-gen3: probe of 11280000.pcie failed with error -110

One specific hardware is reported as example:

Adata Legend710 512GB x3

Quoted 25 lines

Thanks.

Quoted 4 lines · Frank Wunderlich

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 18 lines

Hi

Quoted 16 lines

Gesendet: Sonntag, 02. April 2023 um 15:13 Uhr
Von: "Frank Wunderlich" <[email protected]>
De-assert PERST in separate step after reset signals to fully comply
the PCIe CEM clause 2.2.

This fixes some NVME detection issues on mt7986.

Fixes d3bf75b579b9 ("PCI: mediatek-gen3: Add MediaTek Gen3 driver

for MT8192")

---
Patch is taken from user Ruslan aka RRKh61 (permitted me to send it with me as author).

Quoted 34 lines

https://urldefense.com/v3/__https://forum.banana-pi.org/t/bpi-r3-nvme-connection-issue/14563/17__;!!CTRNKA9wMg0ARbw!nCXEM685pkUpoiZYGKptPYccNrWMeN2D3jIO5_irwxZJ7c6ZzEeACIx-V2WeZHAP_0FKlDDIQ0RbDJ892prtoToDv30$

Quoted 33 lines
Quoted 25 lines
---
 drivers/pci/controller/pcie-mediatek-gen3.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/pci/controller/pcie-mediatek-gen3.c +7 -1
index b8612ce5f4d0..176b1a04565d 100644
@@ -350,7 +350,13 @@ static int mtk_pcie_startup_port(struct
 mtk_gen3_pcie *pcie)
      msleep(100);
 
      /* De-assert reset signals */
-     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB |
 PCIE_PE_RSTB);
+     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB);
+     writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
+
+     msleep(100);
+
+     /* De-assert PERST# signals */
+     val &= ~(PCIE_PE_RSTB);
      writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
Quoted 1 line
      /* Check if the link is up or not */

Hi

just a friendly reminder....do i miss any recipient?

regards Frank

regards Frank

#4 lore raw

Hi Frank,

Quoted 14 lines · Frank Wunderlich

External email : Please do not click links or open attachments until you have verified the sender or the content.

Am 27. April 2023 10:31:07 MESZ schrieb "Jianjun Wang (王建军)" < [email protected]>:

Quoted 7 lines

Hi Frank,

Seems this patch has huge impact on boot time, I'm curious about the
NVMe detection issue on mt7986, can you share the SSD model that you
are using and the bootup logs with that SSD?

Quoted 3 lines

Which "huge" delay do you get in which setup? It adds a 100ms delay yes,but this seems needed to some devices working.i found several sources talking about the 100ms wake-up time...

Some products are very sensitive to the bootup time, especially the platforms with many PCIe ports, adding this 100ms delay for each port will cause the bootup time be increased by multiple times(depending on the number of PCIe ports it uses), and also the wake-up time, since the mtk_pcie_starup_port() function will be called on resume stage.

Quoted 4 lines

I do not have this issue,but some users in bpi-forum reorted it. Pcie-controller on mt7986/bpi-r3 does simply not detect the nvme and returned ETIMEDOUT (110).

Since we're already comply with the PCIe CEM specification sections 2.2(PERST# signal)[1], and this issue only occurs on a few platforms, I'll inclined to it's might be a signal quality issue. Also I checked the BPI-R3 schematic diagram[2], and noticed that there are no AC Coupling capacitors on the transmitter side, which described in PCIe CEM Spec sections 4.7.1, but this schematic diagram only have part of it, can you help to check the board design or share the full schematic diagram for further analysis?

[1]:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/pcie-mediatek-gen3.c?h=v6.3#n344
[2]:
https://drive.google.com/file/d/1mxKb8CBbnzfNSd_4esmcX_NovxaXjEb8/view

Thanks.

Quoted 12 lines
# dmesg | grep 'pci'
[ 5.235564] mtk-pcie-gen3 11280000.pcie: host bridge 
/soc/pcie@11280000 ranges:
[ 5.242938] mtk-pcie-gen3 11280000.pcie: Parsing ranges property...
[ 5.249235] mtk-pcie-gen3 11280000.pcie: MEM
0x0020000000..0x002fffffff -> 0x0020000000
[ 5.478062] mtk-pcie-gen3 11280000.pcie: PCIe link down, current
LTSSM state: detect.active (0x10 00001)
[ 5.487491] mtk-pcie-gen3: probe of 11280000.pcie failed with error
-110
Quoted 34 lines

One specific hardware is reported as example:

Adata Legend710 512GB x3

Quoted 30 lines

Thanks.

Quoted 25 lines · Frank Wunderlich

External email : Please do not click links or open attachments until
you have verified the sender or the content.

Hi

Quoted 18 lines

Gesendet: Sonntag, 02. April 2023 um 15:13 Uhr
Von: "Frank Wunderlich" <[email protected]>
De-assert PERST in separate step after reset signals to fully comply
the PCIe CEM clause 2.2.

This fixes some NVME detection issues on mt7986.

Fixes d3bf75b579b9 ("PCI: mediatek-gen3: Add MediaTek Gen3

driver
for MT8192")

---
Patch is taken from user Ruslan aka RRKh61 (permitted me to send it
with me as author).

https://urldefense.com/v3/__https://forum.banana-pi.org/t/bpi-r3-nvme-connection-issue/14563/17__;!!CTRNKA9wMg0ARbw!nCXEM685pkUpoiZYGKptPYccNrWMeN2D3jIO5_irwxZJ7c6ZzEeACIx-V2WeZHAP_0FKlDDIQ0RbDJ892prtoToDv30$

Quoted 36 lines
Quoted 25 lines
Quoted 25 lines
Quoted 25 lines
---
 drivers/pci/controller/pcie-mediatek-gen3.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/pci/controller/pcie-mediatek-gen3.c +7 -1
index b8612ce5f4d0..176b1a04565d 100644
@@ -350,7 +350,13 @@ static int mtk_pcie_startup_port(struct
 mtk_gen3_pcie *pcie)
      msleep(100);
 
      /* De-assert reset signals */
-     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB |
 PCIE_PE_RSTB);
+     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB);
+     writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
+
+     msleep(100);
+
+     /* De-assert PERST# signals */
+     val &= ~(PCIE_PE_RSTB);
      writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
Quoted 8 lines
Quoted 8 lines
Quoted 1 line
      /* Check if the link is up or not */

Hi

just a friendly reminder....do i miss any recipient?

regards Frank

regards Frank

#5 lore raw

Hi

Am 29. April 2023 05:15:51 MESZ schrieb "Jianjun Wang (王建军)" <[email protected]>:

Quoted 21 lines

Hi Frank,

Quoted 14 lines · Frank Wunderlich

External email : Please do not click links or open attachments until you have verified the sender or the content.

Am 27. April 2023 10:31:07 MESZ schrieb "Jianjun Wang (王建军)" < [email protected]>:

Quoted 7 lines

Hi Frank,

Seems this patch has huge impact on boot time, I'm curious about the
NVMe detection issue on mt7986, can you share the SSD model that you
are using and the bootup logs with that SSD?

Quoted 3 lines

Which "huge" delay do you get in which setup? It adds a 100ms delay yes,but this seems needed to some devices working.i found several sources talking about the 100ms wake-up time...

Quoted 5 lines

Some products are very sensitive to the bootup time, especially the platforms with many PCIe ports, adding this 100ms delay for each port will cause the bootup time be increased by multiple times(depending on the number of PCIe ports it uses), and also the wake-up time, since the mtk_pcie_starup_port() function will be called on resume stage.

Thanks for taking time for analysing it.

Quoted 5 lines
Quoted 3 lines

I do not have this issue,but some users in bpi-forum reorted it. Pcie-controller on mt7986/bpi-r3 does simply not detect the nvme and returned ETIMEDOUT (110).

Since we're already comply with the PCIe CEM specification sections 2.2(PERST# signal)[1], and this issue only occurs on a few platforms,

I see there is the msleep before the reset,where the patch splits reset and perst (and add the sleep between).

So imho the best way is to move the existing msleep between these "splitted" signals instead of adding a second one. I have to verify it with someone who have the issue currently.

Quoted 6 lines

I'll inclined to it's might be a signal quality issue. Also I checked the BPI-R3 schematic diagram[2], and noticed that there are no AC Coupling capacitors on the transmitter side, which described in PCIe CEM Spec sections 4.7.1, but this schematic diagram only have part of it, can you help to check the board design or share the full schematic diagram for further analysis?

I gave the questions to board vendor in forum.

Quoted 57 lines

[1]:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/pcie-mediatek-gen3.c?h=v6.3#n344
[2]:
https://drive.google.com/file/d/1mxKb8CBbnzfNSd_4esmcX_NovxaXjEb8/view

Thanks.

Quoted 12 lines
# dmesg | grep 'pci'
[ 5.235564] mtk-pcie-gen3 11280000.pcie: host bridge 
/soc/pcie@11280000 ranges:
[ 5.242938] mtk-pcie-gen3 11280000.pcie: Parsing ranges property...
[ 5.249235] mtk-pcie-gen3 11280000.pcie: MEM
0x0020000000..0x002fffffff -> 0x0020000000
[ 5.478062] mtk-pcie-gen3 11280000.pcie: PCIe link down, current
LTSSM state: detect.active (0x10 00001)
[ 5.487491] mtk-pcie-gen3: probe of 11280000.pcie failed with error
-110
Quoted 34 lines

One specific hardware is reported as example:

Adata Legend710 512GB x3

Quoted 30 lines

Thanks.

Quoted 25 lines · Frank Wunderlich

External email : Please do not click links or open attachments until
you have verified the sender or the content.

Hi

Quoted 18 lines

Gesendet: Sonntag, 02. April 2023 um 15:13 Uhr
Von: "Frank Wunderlich" <[email protected]>
De-assert PERST in separate step after reset signals to fully comply
the PCIe CEM clause 2.2.

This fixes some NVME detection issues on mt7986.

Fixes d3bf75b579b9 ("PCI: mediatek-gen3: Add MediaTek Gen3

driver
for MT8192")

---
Patch is taken from user Ruslan aka RRKh61 (permitted me to send it
with me as author).

https://urldefense.com/v3/__https://forum.banana-pi.org/t/bpi-r3-nvme-connection-issue/14563/17__;!!CTRNKA9wMg0ARbw!nCXEM685pkUpoiZYGKptPYccNrWMeN2D3jIO5_irwxZJ7c6ZzEeACIx-V2WeZHAP_0FKlDDIQ0RbDJ892prtoToDv30$

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
---
 drivers/pci/controller/pcie-mediatek-gen3.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

drivers/pci/controller/pcie-mediatek-gen3.c static int mtk_pcie_startup_port(struct · line 351

Quoted 8 lines
drivers/pci/controller/pcie-mediatek-gen3.c +0 -0
index b8612ce5f4d0..176b1a04565d 100644
@@ -350,7 +350,13 @@ static int mtk_pcie_startup_port(struct
 mtk_gen3_pcie *pcie)
      msleep(100);

Maybe the better way is to drop this msleep when adding the new one below the reset asserts?

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
drivers/pci/controller/pcie-mediatek-gen3.c +7 -1
      /* De-assert reset signals */
-     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB |
 PCIE_PE_RSTB);
+     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB);
+     writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
+
+     msleep(100);
+
+     /* De-assert PERST# signals */
+     val &= ~(PCIE_PE_RSTB);
      writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
 
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
      /* Check if the link is up or not */

regards Frank

#6 lore raw

Am 29. April 2023 11:03:07 MESZ schrieb Frank Wunderlich <[email protected]>:

Quoted 35 lines

Hi

Am 29. April 2023 05:15:51 MESZ schrieb "Jianjun Wang (王建军)" <[email protected]>:

Quoted 21 lines

Hi Frank,

Quoted 14 lines · Frank Wunderlich

External email : Please do not click links or open attachments until you have verified the sender or the content.

Am 27. April 2023 10:31:07 MESZ schrieb "Jianjun Wang (王建军)" < [email protected]>:

Quoted 7 lines

Hi Frank,

Seems this patch has huge impact on boot time, I'm curious about the
NVMe detection issue on mt7986, can you share the SSD model that you
are using and the bootup logs with that SSD?

Quoted 3 lines

Which "huge" delay do you get in which setup? It adds a 100ms delay yes,but this seems needed to some devices working.i found several sources talking about the 100ms wake-up time...

Quoted 5 lines

Some products are very sensitive to the bootup time, especially the platforms with many PCIe ports, adding this 100ms delay for each port will cause the bootup time be increased by multiple times(depending on the number of PCIe ports it uses), and also the wake-up time, since the mtk_pcie_starup_port() function will be called on resume stage.

Thanks for taking time for analysing it.

Quoted 3 lines
Quoted 3 lines

I do not have this issue,but some users in bpi-forum reorted it. Pcie-controller on mt7986/bpi-r3 does simply not detect the nvme and returned ETIMEDOUT (110).

Quoted 6 lines
Quoted 2 lines

Since we're already comply with the PCIe CEM specification sections 2.2(PERST# signal)[1], and this issue only occurs on a few platforms,

I see there is the msleep before the reset,where the patch splits reset and perst (and add the sleep between).

So imho the best way is to move the existing msleep between these "splitted" signals instead of adding a second one. I have to verify it with someone who have the issue currently.

Seems this does not make it work...

Here are the bootlogs with different versions of this patch:

https://forum.banana-pi.org/t/bpi-r3-problem-with-pcie/15152/22

Quoted 88 lines
Quoted 6 lines

I'll inclined to it's might be a signal quality issue. Also I checked the BPI-R3 schematic diagram[2], and noticed that there are no AC Coupling capacitors on the transmitter side, which described in PCIe CEM Spec sections 4.7.1, but this schematic diagram only have part of it, can you help to check the board design or share the full schematic diagram for further analysis?

I gave the questions to board vendor in forum.

Quoted 57 lines

[1]:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/pcie-mediatek-gen3.c?h=v6.3#n344
[2]:
https://drive.google.com/file/d/1mxKb8CBbnzfNSd_4esmcX_NovxaXjEb8/view

Thanks.

Quoted 12 lines
# dmesg | grep 'pci'
[ 5.235564] mtk-pcie-gen3 11280000.pcie: host bridge 
/soc/pcie@11280000 ranges:
[ 5.242938] mtk-pcie-gen3 11280000.pcie: Parsing ranges property...
[ 5.249235] mtk-pcie-gen3 11280000.pcie: MEM
0x0020000000..0x002fffffff -> 0x0020000000
[ 5.478062] mtk-pcie-gen3 11280000.pcie: PCIe link down, current
LTSSM state: detect.active (0x10 00001)
[ 5.487491] mtk-pcie-gen3: probe of 11280000.pcie failed with error
-110
Quoted 34 lines

One specific hardware is reported as example:

Adata Legend710 512GB x3

Quoted 30 lines

Thanks.

Quoted 25 lines · Frank Wunderlich

External email : Please do not click links or open attachments until
you have verified the sender or the content.

Hi

Quoted 18 lines

Gesendet: Sonntag, 02. April 2023 um 15:13 Uhr
Von: "Frank Wunderlich" <[email protected]>
De-assert PERST in separate step after reset signals to fully comply
the PCIe CEM clause 2.2.

This fixes some NVME detection issues on mt7986.

Fixes d3bf75b579b9 ("PCI: mediatek-gen3: Add MediaTek Gen3

driver
for MT8192")

---
Patch is taken from user Ruslan aka RRKh61 (permitted me to send it
with me as author).

https://urldefense.com/v3/__https://forum.banana-pi.org/t/bpi-r3-nvme-connection-issue/14563/17__;!!CTRNKA9wMg0ARbw!nCXEM685pkUpoiZYGKptPYccNrWMeN2D3jIO5_irwxZJ7c6ZzEeACIx-V2WeZHAP_0FKlDDIQ0RbDJ892prtoToDv30$

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
---
 drivers/pci/controller/pcie-mediatek-gen3.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

drivers/pci/controller/pcie-mediatek-gen3.c static int mtk_pcie_startup_port(struct · line 351

Quoted 8 lines
drivers/pci/controller/pcie-mediatek-gen3.c +0 -0
index b8612ce5f4d0..176b1a04565d 100644
@@ -350,7 +350,13 @@ static int mtk_pcie_startup_port(struct
 mtk_gen3_pcie *pcie)
      msleep(100);

Maybe the better way is to drop this msleep when adding the new one below the reset asserts?

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
drivers/pci/controller/pcie-mediatek-gen3.c +7 -1
      /* De-assert reset signals */
-     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB |
 PCIE_PE_RSTB);
+     val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB);
+     writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
+
+     msleep(100);
+
+     /* De-assert PERST# signals */
+     val &= ~(PCIE_PE_RSTB);
      writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
 
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
      /* Check if the link is up or not */

regards Frank

#7 lore raw
In reply to Frank Wunderlich
Quoted 4 lines · Frank Wunderlich

From: Frank Wunderlich <[email protected]>

De-assert PERST in separate step after reset signals to fully comply the PCIe CEM clause 2.2.

I guess this refers to PCIe CEM r5.0, sec 2.2.

Quoted 18 lines

This fixes some NVME detection issues on mt7986.

Fixes d3bf75b579b9 ("PCI: mediatek-gen3: Add MediaTek Gen3 driver for MT8192")

---
Patch is taken from user Ruslan aka RRKh61 (permitted me to send it with me as author).

https://forum.banana-pi.org/t/bpi-r3-nvme-connection-issue/14563/17

---
 drivers/pci/controller/pcie-mediatek-gen3.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/pci/controller/pcie-mediatek-gen3.c +0 -0
index b8612ce5f4d0..176b1a04565d 100644

drivers/pci/controller/pcie-mediatek-gen3.c static int mtk_pcie_startup_port(struct mtk_gen3_pcie *pcie) · line 356

Quoted 8 lines
drivers/pci/controller/pcie-mediatek-gen3.c +4 -1
@@ -350,7 +350,13 @@ static int mtk_pcie_startup_port(struct mtk_gen3_pcie *pcie)
 	msleep(100);
 
 	/* De-assert reset signals */
-	val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB | PCIE_PE_RSTB);
+	val &= ~(PCIE_MAC_RSTB | PCIE_PHY_RSTB | PCIE_BRG_RSTB);
+	writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
+
+	msleep(100);

There should be a #define for the 100ms value since it is required by the generic PCIe CEM spec, not by anything specific to mediatek. If one already exists, we should use it. If not, we should add one.

pcie-tegra194.c and pcie-mediatek.c (at least) also have similar delays and should also use the same #define. There are several other drivers that contain "msleep(100)", but I didn't look to see their purpose.

Quoted 9 lines
drivers/pci/controller/pcie-mediatek-gen3.c +2 -0
+	/* De-assert PERST# signals */
+	val &= ~(PCIE_PE_RSTB);
 	writel_relaxed(val, pcie->base + PCIE_RST_CTRL_REG);
 
 	/* Check if the link is up or not */
Signature · 1 line
2.34.1