Skip to contentSkip to conversation

mmc: sdhci-pci-gli: GL9755: Quirks for Apple ARM platforms

PATCH Superseded Hector Martin 2 participants 7 messages Dec 7, 2021 – Dec 9, 2021 1 message not in the archive

Subsystems (2)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover mmc: sdhci-pci-gli: GL9755: Quirks for Apple ARM platforms

Hi folks,

This short series adds a few quirks needed to make the card readers in Apple M1 Pro/Max MacBook laptops work properly.

The first patch should be straightforward; it just allows configuring the CD/WP polarity based on device tree settings. There is already a standard DT binding for this.

The second patch bugs me. I don't understand why this problem happens on these machines, and not on e.g. x86 laptops (which presumably work with this driver). 8/16-bit MMIO reads work fine on other PCIe devices on these machines, so it is not a generalized problem with the PCIe controller in these SoCs. The problem also happens when running macOS (it also uses 32-bit reads). Ben, is there any chance you might know of some vendor-specific knob somewhere that can fix this issue without requiring the MMIO read workaround? Interestingly, 8/16-bit writes work perfectly fine.

Hector Martin (2):
mmc: sdhci-pci-gli: GL9755: Support for CD/WP inversion on OF
platforms
mmc: sdhci-pci-gli: GL9755: Issue 8/16-bit MMIO reads as 32-bit reads.

 drivers/mmc/host/sdhci-pci-gli.c | 38 ++++++++++++++++++++++++++++++--
 1 file changed, 36 insertions(+), 2 deletions(-)
Signature · 1 line
2.33.0
#1 lore raw

PATCH 1/2 mmc: sdhci-pci-gli: GL9755: Support for CD/WP inversion on OF platforms

This is required on some Apple ARM64 laptops using this controller. As is typical on DT platforms, pull these quirks from the device tree using the standard mmc bindings.

See Documentation/devicetree/bindings/mmc/mmc-controller.yaml

---
 drivers/mmc/host/sdhci-pci-gli.c | 19 +++++++++++++++++--
 1 file changed, 17 insertions(+), 2 deletions(-)
drivers/mmc/host/sdhci-pci-gli.c +17 -2
index 4fd99c1e82ba..ad742743a494 100644
@@ -12,6 +12,7 @@
 #include <linux/pci.h>
 #include <linux/mmc/mmc.h>
 #include <linux/delay.h>
+#include <linux/of.h>
 #include "sdhci.h"
 #include "sdhci-pci.h"
 #include "cqhci.h"
@@ -114,8 +115,10 @@
 #define   GLI_9755_WT_EN_OFF    0x0
 
 #define PCI_GLI_9755_PECONF   0x44
-#define   PCI_GLI_9755_LFCLK    GENMASK(14, 12)
-#define   PCI_GLI_9755_DMACLK   BIT(29)
+#define   PCI_GLI_9755_LFCLK          GENMASK(14, 12)
+#define   PCI_GLI_9755_DMACLK         BIT(29)
+#define   PCI_GLI_9755_INVERT_CD      BIT(30)
+#define   PCI_GLI_9755_INVERT_WP      BIT(31)
 
 #define PCI_GLI_9755_CFG2          0x48
 #define   PCI_GLI_9755_CFG2_L1DLY    GENMASK(28, 24)
@@ -570,6 +573,18 @@ static void gl9755_hw_setting(struct sdhci_pci_slot *slot)
 	gl9755_wt_on(pdev);
 
 	pci_read_config_dword(pdev, PCI_GLI_9755_PECONF, &value);
+#ifdef CONFIG_OF
+	if (pdev->dev.of_node) {
+		/*
+		 * Apple ARM64 platforms using these chips may have
+		 * inverted CD/WP detection.
+		 */
+		if (of_property_read_bool(pdev->dev.of_node, "cd-inverted"))
+			value |= PCI_GLI_9755_INVERT_CD;
+		if (of_property_read_bool(pdev->dev.of_node, "wp-inverted"))
+			value |= PCI_GLI_9755_INVERT_WP;
+	}
+#endif
 	value &= ~PCI_GLI_9755_LFCLK;
 	value &= ~PCI_GLI_9755_DMACLK;
 	pci_write_config_dword(pdev, PCI_GLI_9755_PECONF, value);
Signature · 1 line
2.33.0
#2 lore raw

PATCH 2/2 mmc: sdhci-pci-gli: GL9755: Issue 8/16-bit MMIO reads as 32-bit reads.

In reply to Hector Martin

For some reason, <32-bit reads do not work on Apple ARM64 platforms with these chips (even though they do on other PCIe devices). Issue them as 32-bit reads instead. This is done unconditionally, as it shouldn't hurt even if not necessary.

---
 drivers/mmc/host/sdhci-pci-gli.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
drivers/mmc/host/sdhci-pci-gli.c +19 -0
index ad742743a494..31547fed0952 100644
@@ -906,7 +906,26 @@ static int gli_probe_slot_gl9763e(struct sdhci_pci_slot *slot)
 	return 0;
 }
 
+#define REG_OFFSET_IN_BITS(reg) ((reg) << 3 & 0x18)
+
+static u16 sdhci_gli_readw(struct sdhci_host *host, int reg)
+{
+	u32 val = readl(host->ioaddr + (reg & ~3));
+	u16 word;
+	word = (val >> REG_OFFSET_IN_BITS(reg)) & 0xffff;
+	return word;
+}
+
+static u8 sdhci_gli_readb(struct sdhci_host *host, int reg)
+{
+	u32 val = readl(host->ioaddr + (reg & ~3));
+	u8 byte = (val >> REG_OFFSET_IN_BITS(reg)) & 0xff;
+	return byte;
+}
+
 static const struct sdhci_ops sdhci_gl9755_ops = {
+	.read_w			= sdhci_gli_readw,
+	.read_b			= sdhci_gli_readb,
 	.set_clock		= sdhci_gl9755_set_clock,
 	.enable_dma		= sdhci_pci_enable_dma,
 	.set_bus_width		= sdhci_set_bus_width,
Signature · 1 line
2.33.0
#3 lore raw

Re: [PATCH 2/2] mmc: sdhci-pci-gli: GL9755: Issue 8/16-bit MMIO reads as 32-bit reads.

Hi Hector,

Quoted 16 lines · Hector Martin

For some reason, <32-bit reads do not work on Apple ARM64 platforms with these chips (even though they do on other PCIe devices). Issue them as 32-bit reads instead. This is done unconditionally, as it shouldn't hurt even if not necessary.

---
 drivers/mmc/host/sdhci-pci-gli.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
drivers/mmc/host/sdhci-pci-gli.c +0 -0
index ad742743a494..31547fed0952 100644

drivers/mmc/host/sdhci-pci-gli.c static int gli_probe_slot_gl9763e(struct sdhci_pci_slot *slot) · line 915

Quoted 10 lines
drivers/mmc/host/sdhci-pci-gli.c +7 -0
@@ -906,7 +906,26 @@ static int gli_probe_slot_gl9763e(struct sdhci_pci_slot *slot)
 	return 0;
 }
 
+#define REG_OFFSET_IN_BITS(reg) ((reg) << 3 & 0x18)
+
+static u16 sdhci_gli_readw(struct sdhci_host *host, int reg)
+{
+	u32 val = readl(host->ioaddr + (reg & ~3));
+	u16 word;
+	word = (val >> REG_OFFSET_IN_BITS(reg)) & 0xffff;

checkpatch says "
WARNING: Missing a blank line after declarations
#71: FILE: drivers/mmc/host/sdhci-pci-gli.c:915:
+ u16 word;
+ word = (val >> REG_OFFSET_IN_BITS(reg)) & 0xffff;
"

drivers/mmc/host/sdhci-pci-gli.c static int gli_probe_slot_gl9763e(struct sdhci_pci_slot *slot) · line 928

Quoted 13 lines
drivers/mmc/host/sdhci-pci-gli.c +12 -0
+	return word;
+}
+
+static u8 sdhci_gli_readb(struct sdhci_host *host, int reg)
+{
+	u32 val = readl(host->ioaddr + (reg & ~3));
+	u8 byte = (val >> REG_OFFSET_IN_BITS(reg)) & 0xff;
+	return byte;
+}
+
 static const struct sdhci_ops sdhci_gl9755_ops = {
+	.read_w			= sdhci_gli_readw,
+	.read_b			= sdhci_gli_readb,

I think GL9750 also need this patch.
Can you help to add these two functions to sdhci_gl9750_ops?

Best regards,
Ben

Quoted 5 lines
drivers/mmc/host/sdhci-pci-gli.c +0 -0
 	.set_clock		= sdhci_gl9755_set_clock,
 	.enable_dma		= sdhci_pci_enable_dma,
 	.set_bus_width		= sdhci_set_bus_width,
Signature · 1 line
2.33.0
#4 lore raw
In reply to Hector Martin

Argh, forgot to Cc Ben...

Ben, please let me know what you think about this series. I can resend it CCing you if you want, or you can also find it here:

https://lore.kernel.org/all/[email protected]/

Sorry for missing the Cc...

-Hector

Quoted 27 lines · Hector Martin

Hi folks,

This short series adds a few quirks needed to make the card readers in Apple M1 Pro/Max MacBook laptops work properly.

The first patch should be straightforward; it just allows configuring the CD/WP polarity based on device tree settings. There is already a standard DT binding for this.

The second patch bugs me. I don't understand why this problem happens on these machines, and not on e.g. x86 laptops (which presumably work with this driver). 8/16-bit MMIO reads work fine on other PCIe devices on these machines, so it is not a generalized problem with the PCIe controller in these SoCs. The problem also happens when running macOS (it also uses 32-bit reads). Ben, is there any chance you might know of some vendor-specific knob somewhere that can fix this issue without requiring the MMIO read workaround? Interestingly, 8/16-bit writes work perfectly fine.

Hector Martin (2):
mmc: sdhci-pci-gli: GL9755: Support for CD/WP inversion on OF
platforms
mmc: sdhci-pci-gli: GL9755: Issue 8/16-bit MMIO reads as 32-bit reads.

  drivers/mmc/host/sdhci-pci-gli.c | 38 ++++++++++++++++++++++++++++++--
  1 file changed, 36 insertions(+), 2 deletions(-)
Signature · 2 lines
Hector Martin ([email protected])
Public Key: https://mrcn.st/pub

A message that is not in the archive (sent off-list or not yet received): CACT4zj9Oy_Le8KzPS9WfH+Zx9Re7h0SOwkWFN+G5X1pCF3s3ZA@mail.gmail.com

#6 lore raw

Hi Ben,

Quoted 5 lines · Ben Chuang

Hi Hector,

I got your two patches so you don't need to resend them. About the 8/16-bits MMIO read patch as the second patch, it is necessary. I don't know some vendor-specific knobs, sorry.

Is this only required on these Apple platforms? I'm very curious about what's different about them. I assume these chips don't need the workaround on x86 laptops, right?

Quoted 1 line

ps. Do you make Asahi Linux now?  It's a cool project. :)

Along with a lot of other talented folks, yes :)

-Hector

Signature · 2 lines
Hector Martin ([email protected])
Public Key: https://mrcn.st/pub
#7 lore raw

Hi Hector,

Quoted 6 lines · Hector Martin

Hi Ben,

Quoted 2 lines · Ben Chuang

Hi Hector,

Quoted 7 lines
Quoted 3 lines

I got your two patches so you don't need to resend them. About the 8/16-bits MMIO read patch as the second patch, it is necessary. I don't know some vendor-specific knobs, sorry.

Is this only required on these Apple platforms? I'm very curious about what's different about them. I assume these chips don't need the workaround on x86 laptops, right?

I'm not sure if it only happens on these Apple platforms. Right, the workaround is not needed on x86.

Quoted 10 lines
Quoted 2 lines

ps. Do you make Asahi Linux now? It's a cool project. :)

Along with a lot of other talented folks, yes :)

-Hector

--
Hector Martin ([email protected])
Public Key: https://mrcn.st/pub

Best regards,
Ben