Skip to contentSkip to conversation

hw/sd/sdhci: Do not force sdhci_mmio_*_ops onto all SD controllers

PATCH In review Bernhard Beschow 3 participants 9 messages Jul 9, 2023 – Jul 24, 2023

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hw/sd/sdhci: Do not force sdhci_mmio_*_ops onto all SD controllers

Since commit c0a55a0c9da2 "hw/sd/sdhci: Support big endian SD host controller interfaces" sdhci_common_realize() forces all SD card controllers to use either sdhci_mmio_le_ops or sdhci_mmio_be_ops, depending on the "endianness" property. However, there are device models which use different MMIO ops: TYPE_IMX_USDHC uses usdhc_mmio_ops and TYPE_S3C_SDHCI uses sdhci_s3c_mmio_ops.

Forcing sdhci_mmio_le_ops breaks SD card handling on the "sabrelite" board, for example. Fix this by defaulting the io_ops to little endian and switch to big endian in sdhci_common_realize() only if there is a matchig big endian variant available.

Fixes c0a55a0c9da2 ("hw/sd/sdhci: Support big endian SD host controller

interfaces")

---
 hw/sd/sdhci.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
hw/sd/sdhci.c +7 -1
index 6811f0f1a8..362c2c86aa 100644
@@ -1382,6 +1382,8 @@ void sdhci_initfn(SDHCIState *s)
 
     s->insert_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_raise_insertion_irq, s);
     s->transfer_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_data_transfer, s);
+
+    s->io_ops = &sdhci_mmio_le_ops;
 }
 
 void sdhci_uninitfn(SDHCIState *s)
@@ -1399,9 +1401,13 @@ void sdhci_common_realize(SDHCIState *s, Error **errp)
 
     switch (s->endianness) {
     case DEVICE_LITTLE_ENDIAN:
-        s->io_ops = &sdhci_mmio_le_ops;
+        /* s->io_ops is little endian by default */
         break;
     case DEVICE_BIG_ENDIAN:
+        if (s->io_ops != &sdhci_mmio_le_ops) {
+            error_setg(errp, "SD controller doesn't support big endianness");
+            return;
+        }
         s->io_ops = &sdhci_mmio_be_ops;
         break;
     default:
Signature · 1 line
2.41.0
Tested-by on v1 1/1
#1 lore raw
Quoted 11 lines · Bernhard Beschow

Since commit c0a55a0c9da2 "hw/sd/sdhci: Support big endian SD host controller interfaces" sdhci_common_realize() forces all SD card controllers to use either sdhci_mmio_le_ops or sdhci_mmio_be_ops, depending on the "endianness" property. However, there are device models which use different MMIO ops: TYPE_IMX_USDHC uses usdhc_mmio_ops and TYPE_S3C_SDHCI uses sdhci_s3c_mmio_ops.

Forcing sdhci_mmio_le_ops breaks SD card handling on the "sabrelite" board, for example. Fix this by defaulting the io_ops to little endian and switch to big endian in sdhci_common_realize() only if there is a matchig big endian variant available.

Quoted 4 lines
Fixes c0a55a0c9da2 ("hw/sd/sdhci: Support big endian SD host controller

interfaces")

Quoted 32 lines
---
  hw/sd/sdhci.c | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
hw/sd/sdhci.c +7 -1
index 6811f0f1a8..362c2c86aa 100644
@@ -1382,6 +1382,8 @@ void sdhci_initfn(SDHCIState *s)
 
      s->insert_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_raise_insertion_irq, s);
      s->transfer_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_data_transfer, s);
+
+    s->io_ops = &sdhci_mmio_le_ops;
  }
 
  void sdhci_uninitfn(SDHCIState *s)
@@ -1399,9 +1401,13 @@ void sdhci_common_realize(SDHCIState *s, Error **errp)
 
      switch (s->endianness) {
      case DEVICE_LITTLE_ENDIAN:
-        s->io_ops = &sdhci_mmio_le_ops;
+        /* s->io_ops is little endian by default */
          break;
      case DEVICE_BIG_ENDIAN:
+        if (s->io_ops != &sdhci_mmio_le_ops) {
+            error_setg(errp, "SD controller doesn't support big endianness");
+            return;
+        }
          s->io_ops = &sdhci_mmio_be_ops;
          break;
      default:
#2 lore raw
In reply to Bernhard Beschow
Quoted 33 lines · Bernhard Beschow

Since commit c0a55a0c9da2 "hw/sd/sdhci: Support big endian SD host controller interfaces" sdhci_common_realize() forces all SD card controllers to use either sdhci_mmio_le_ops or sdhci_mmio_be_ops, depending on the "endianness" property. However, there are device models which use different MMIO ops: TYPE_IMX_USDHC uses usdhc_mmio_ops and TYPE_S3C_SDHCI uses sdhci_s3c_mmio_ops.

Forcing sdhci_mmio_le_ops breaks SD card handling on the "sabrelite" board, for example. Fix this by defaulting the io_ops to little endian and switch to big endian in sdhci_common_realize() only if there is a matchig big endian variant available.

Fixes c0a55a0c9da2 ("hw/sd/sdhci: Support big endian SD host controller

interfaces")

---
  hw/sd/sdhci.c | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
hw/sd/sdhci.c +2 -0
index 6811f0f1a8..362c2c86aa 100644
@@ -1382,6 +1382,8 @@ void sdhci_initfn(SDHCIState *s)
 
      s->insert_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_raise_insertion_irq, s);
      s->transfer_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_data_transfer, s);
+
+    s->io_ops = &sdhci_mmio_le_ops;
  }
 
  void sdhci_uninitfn(SDHCIState *s)

hw/sd/sdhci.c void sdhci_common_realize(SDHCIState *s, Error **errp) · line 1401

Quoted 1 line
hw/sd/sdhci.c +0 -0
@@ -1399,9 +1401,13 @@ void sdhci_common_realize(SDHCIState *s, Error **errp)
 

What about simply keeping the same code guarded with 'if (!s->io_ops)'?

Quoted 13 lines
hw/sd/sdhci.c +5 -1
      switch (s->endianness) {
      case DEVICE_LITTLE_ENDIAN:
-        s->io_ops = &sdhci_mmio_le_ops;
+        /* s->io_ops is little endian by default */
          break;
      case DEVICE_BIG_ENDIAN:
+        if (s->io_ops != &sdhci_mmio_le_ops) {
+            error_setg(errp, "SD controller doesn't support big endianness");
+            return;
+        }
          s->io_ops = &sdhci_mmio_be_ops;
          break;
      default:
#3 lore raw
Quoted 20 lines · Philippe Mathieu-Daudé
Quoted 19 lines · Bernhard Beschow

Since commit c0a55a0c9da2 "hw/sd/sdhci: Support big endian SD host controller interfaces" sdhci_common_realize() forces all SD card controllers to use either sdhci_mmio_le_ops or sdhci_mmio_be_ops, depending on the "endianness" property. However, there are device models which use different MMIO ops: TYPE_IMX_USDHC uses usdhc_mmio_ops and TYPE_S3C_SDHCI uses sdhci_s3c_mmio_ops.

Forcing sdhci_mmio_le_ops breaks SD card handling on the "sabrelite" board, for example. Fix this by defaulting the io_ops to little endian and switch to big endian in sdhci_common_realize() only if there is a matchig big endian variant available.

Fixes c0a55a0c9da2 ("hw/sd/sdhci: Support big endian SD host controller

interfaces")

---
  hw/sd/sdhci.c | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
Quoted 15 lines

hw/sd/sdhci.c void sdhci_common_realize(SDHCIState *s, Error **errp)

Quoted 12 lines
hw/sd/sdhci.c +2 -0
index 6811f0f1a8..362c2c86aa 100644
@@ -1382,6 +1382,8 @@ void sdhci_initfn(SDHCIState *s)
      s->insert_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_raise_insertion_irq, s);
      s->transfer_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_data_transfer, s);
+
+    s->io_ops = &sdhci_mmio_le_ops;
  }
  void sdhci_uninitfn(SDHCIState *s)
@@ -1399,9 +1401,13 @@ void sdhci_common_realize(SDHCIState *s, Error **errp)

What about simply keeping the same code guarded with 'if (!s->io_ops)'?

That was my quick fix which I considered a hack, and I didn't submit it because I thought it was a hack ;-).

On the other side, this solution would probably break on big endian systems which have their own io ops, so I am not sure if it is any better.

Guenter

Quoted 14 lines
Quoted 13 lines
hw/sd/sdhci.c +5 -1
      switch (s->endianness) {
      case DEVICE_LITTLE_ENDIAN:
-        s->io_ops = &sdhci_mmio_le_ops;
+        /* s->io_ops is little endian by default */
          break;
      case DEVICE_BIG_ENDIAN:
+        if (s->io_ops != &sdhci_mmio_le_ops) {
+            error_setg(errp, "SD controller doesn't support big endianness");
+            return;
+        }
          s->io_ops = &sdhci_mmio_be_ops;
          break;
      default:
#4 lore raw
In reply to Philippe Mathieu-Daudé

Am 10. Juli 2023 10:16:35 UTC schrieb "Philippe Mathieu-Daudé" <[email protected]>:

Quoted 20 lines · Bernhard Beschow
Quoted 19 lines

Since commit c0a55a0c9da2 "hw/sd/sdhci: Support big endian SD host controller interfaces" sdhci_common_realize() forces all SD card controllers to use either sdhci_mmio_le_ops or sdhci_mmio_be_ops, depending on the "endianness" property. However, there are device models which use different MMIO ops: TYPE_IMX_USDHC uses usdhc_mmio_ops and TYPE_S3C_SDHCI uses sdhci_s3c_mmio_ops.

Forcing sdhci_mmio_le_ops breaks SD card handling on the "sabrelite" board, for example. Fix this by defaulting the io_ops to little endian and switch to big endian in sdhci_common_realize() only if there is a matchig big endian variant available.

Fixes c0a55a0c9da2 ("hw/sd/sdhci: Support big endian SD host controller

interfaces")

---
  hw/sd/sdhci.c | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
Quoted 15 lines

hw/sd/sdhci.c void sdhci_common_realize(SDHCIState *s, Error **errp) · line 1401

Quoted 13 lines
hw/sd/sdhci.c +2 -0
index 6811f0f1a8..362c2c86aa 100644
@@ -1382,6 +1382,8 @@ void sdhci_initfn(SDHCIState *s)
        s->insert_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_raise_insertion_irq, s);
      s->transfer_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_data_transfer, s);
+
+    s->io_ops = &sdhci_mmio_le_ops;
  }
    void sdhci_uninitfn(SDHCIState *s)
@@ -1399,9 +1401,13 @@ void sdhci_common_realize(SDHCIState *s, Error **errp)
 

What about simply keeping the same code guarded with 'if (!s->io_ops)'?

I chose below approach since it provides an error message when one attempts to set one of the other device models to BE rather than just silently ignoring it.

Also, I didn't want to make the assumption that s->io_ops == NULL implied that sdhci_mmio_*_ops is needed. That's similar material the bug fixed is made of, so I wanted to prevent that in the first place by being more explicit.

In combination with the new error message the limitations of the current code become hopefully very apparent now, and at the same time should provide enough hints for adding BE support to the other device models if ever needed.

Best regards,
Bernhard

Quoted 15 lines
Quoted 14 lines
hw/sd/sdhci.c +5 -1
      switch (s->endianness) {
      case DEVICE_LITTLE_ENDIAN:
-        s->io_ops = &sdhci_mmio_le_ops;
+        /* s->io_ops is little endian by default */
          break;
      case DEVICE_BIG_ENDIAN:
+        if (s->io_ops != &sdhci_mmio_le_ops) {
+            error_setg(errp, "SD controller doesn't support big endianness");
+            return;
+        }
          s->io_ops = &sdhci_mmio_be_ops;
          break;
      default:
#5 lore raw

Am 10. Juli 2023 16:01:46 UTC schrieb Bernhard Beschow <[email protected]>:

Quoted 39 lines

Am 10. Juli 2023 10:16:35 UTC schrieb "Philippe Mathieu-Daudé" <[email protected]>:

Quoted 20 lines · Bernhard Beschow
Quoted 19 lines

Since commit c0a55a0c9da2 "hw/sd/sdhci: Support big endian SD host controller interfaces" sdhci_common_realize() forces all SD card controllers to use either sdhci_mmio_le_ops or sdhci_mmio_be_ops, depending on the "endianness" property. However, there are device models which use different MMIO ops: TYPE_IMX_USDHC uses usdhc_mmio_ops and TYPE_S3C_SDHCI uses sdhci_s3c_mmio_ops.

Forcing sdhci_mmio_le_ops breaks SD card handling on the "sabrelite" board, for example. Fix this by defaulting the io_ops to little endian and switch to big endian in sdhci_common_realize() only if there is a matchig big endian variant available.

Fixes c0a55a0c9da2 ("hw/sd/sdhci: Support big endian SD host controller

interfaces")

---
  hw/sd/sdhci.c | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
Quoted 15 lines

hw/sd/sdhci.c void sdhci_common_realize(SDHCIState *s, Error **errp) · line 1401

Quoted 13 lines
hw/sd/sdhci.c +2 -0
index 6811f0f1a8..362c2c86aa 100644
@@ -1382,6 +1382,8 @@ void sdhci_initfn(SDHCIState *s)
        s->insert_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_raise_insertion_irq, s);
      s->transfer_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_data_transfer, s);
+
+    s->io_ops = &sdhci_mmio_le_ops;
  }
    void sdhci_uninitfn(SDHCIState *s)
@@ -1399,9 +1401,13 @@ void sdhci_common_realize(SDHCIState *s, Error **errp)
 

What about simply keeping the same code guarded with 'if (!s->io_ops)'?

Quoted 8 lines

I chose below approach since it provides an error message when one attempts to set one of the other device models to BE rather than just silently ignoring it.

Also, I didn't want to make the assumption that s->io_ops == NULL implied that sdhci_mmio_*_ops is needed. That's similar material the bug fixed is made of, so I wanted to prevent that in the first place by being more explicit.

In combination with the new error message the limitations of the current code become hopefully very apparent now, and at the same time should provide enough hints for adding BE support to the other device models if ever needed.

Best regards,
Bernhard

Ping

+ qemu--stable

Quoted 16 lines
Quoted 16 lines
Quoted 15 lines
hw/sd/sdhci.c +5 -1
      switch (s->endianness) {
      case DEVICE_LITTLE_ENDIAN:
-        s->io_ops = &sdhci_mmio_le_ops;
+        /* s->io_ops is little endian by default */
          break;
      case DEVICE_BIG_ENDIAN:
+        if (s->io_ops != &sdhci_mmio_le_ops) {
+            error_setg(errp, "SD controller doesn't support big endianness");
+            return;
+        }
          s->io_ops = &sdhci_mmio_be_ops;
          break;
      default:
#6 lore raw

Am 16. Juli 2023 19:53:37 UTC schrieb Bernhard Beschow <[email protected]>:

Quoted 44 lines

Am 10. Juli 2023 16:01:46 UTC schrieb Bernhard Beschow <[email protected]>:

Quoted 41 lines

Am 10. Juli 2023 10:16:35 UTC schrieb "Philippe Mathieu-Daudé" <[email protected]>:

Quoted 20 lines · Bernhard Beschow
Quoted 19 lines

Since commit c0a55a0c9da2 "hw/sd/sdhci: Support big endian SD host controller interfaces" sdhci_common_realize() forces all SD card controllers to use either sdhci_mmio_le_ops or sdhci_mmio_be_ops, depending on the "endianness" property. However, there are device models which use different MMIO ops: TYPE_IMX_USDHC uses usdhc_mmio_ops and TYPE_S3C_SDHCI uses sdhci_s3c_mmio_ops.

Forcing sdhci_mmio_le_ops breaks SD card handling on the "sabrelite" board, for example. Fix this by defaulting the io_ops to little endian and switch to big endian in sdhci_common_realize() only if there is a matchig big endian variant available.

Fixes c0a55a0c9da2 ("hw/sd/sdhci: Support big endian SD host controller

interfaces")

---
  hw/sd/sdhci.c | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
Quoted 15 lines

hw/sd/sdhci.c void sdhci_common_realize(SDHCIState *s, Error **errp) · line 1401

Quoted 13 lines
hw/sd/sdhci.c +2 -0
index 6811f0f1a8..362c2c86aa 100644
@@ -1382,6 +1382,8 @@ void sdhci_initfn(SDHCIState *s)
        s->insert_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_raise_insertion_irq, s);
      s->transfer_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_data_transfer, s);
+
+    s->io_ops = &sdhci_mmio_le_ops;
  }
    void sdhci_uninitfn(SDHCIState *s)
@@ -1399,9 +1401,13 @@ void sdhci_common_realize(SDHCIState *s, Error **errp)
 

What about simply keeping the same code guarded with 'if (!s->io_ops)'?

I chose below approach since it provides an error message when one attempts to set one of the other device models to BE rather than just silently ignoring it.

Quoted 8 lines
Quoted 6 lines

Also, I didn't want to make the assumption that s->io_ops == NULL implied that sdhci_mmio_*_ops is needed. That's similar material the bug fixed is made of, so I wanted to prevent that in the first place by being more explicit.

In combination with the new error message the limitations of the current code become hopefully very apparent now, and at the same time should provide enough hints for adding BE support to the other device models if ever needed.

Best regards,
Bernhard

Ping

Ping^2

I would like to have the bug fixed in 8.1.

Best regards,
Bernhard

Quoted 19 lines

+ qemu--stable

Quoted 16 lines
Quoted 16 lines
Quoted 15 lines
hw/sd/sdhci.c +5 -1
      switch (s->endianness) {
      case DEVICE_LITTLE_ENDIAN:
-        s->io_ops = &sdhci_mmio_le_ops;
+        /* s->io_ops is little endian by default */
          break;
      case DEVICE_BIG_ENDIAN:
+        if (s->io_ops != &sdhci_mmio_le_ops) {
+            error_setg(errp, "SD controller doesn't support big endianness");
+            return;
+        }
          s->io_ops = &sdhci_mmio_be_ops;
          break;
      default:
#7 lore raw
Quoted 51 lines · Bernhard Beschow

Am 16. Juli 2023 19:53:37 UTC schrieb Bernhard Beschow <[email protected]>:

Quoted 48 lines

Am 10. Juli 2023 16:01:46 UTC schrieb Bernhard Beschow <[email protected]>:

Quoted 45 lines

Am 10. Juli 2023 10:16:35 UTC schrieb "Philippe Mathieu-Daudé" <[email protected]>:

Quoted 20 lines · Bernhard Beschow
Quoted 19 lines

Since commit c0a55a0c9da2 "hw/sd/sdhci: Support big endian SD host controller interfaces" sdhci_common_realize() forces all SD card controllers to use either sdhci_mmio_le_ops or sdhci_mmio_be_ops, depending on the "endianness" property. However, there are device models which use different MMIO ops: TYPE_IMX_USDHC uses usdhc_mmio_ops and TYPE_S3C_SDHCI uses sdhci_s3c_mmio_ops.

Forcing sdhci_mmio_le_ops breaks SD card handling on the "sabrelite" board, for example. Fix this by defaulting the io_ops to little endian and switch to big endian in sdhci_common_realize() only if there is a matchig big endian variant available.

Fixes c0a55a0c9da2 ("hw/sd/sdhci: Support big endian SD host controller

interfaces")

---
   hw/sd/sdhci.c | 8 +++++++-
   1 file changed, 7 insertions(+), 1 deletion(-)
Quoted 15 lines

hw/sd/sdhci.c void sdhci_common_realize(SDHCIState *s, Error **errp) · line 1401

Quoted 13 lines
hw/sd/sdhci.c +2 -0
index 6811f0f1a8..362c2c86aa 100644
@@ -1382,6 +1382,8 @@ void sdhci_initfn(SDHCIState *s)
         s->insert_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_raise_insertion_irq, s);
       s->transfer_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, sdhci_data_transfer, s);
+
+    s->io_ops = &sdhci_mmio_le_ops;
   }
     void sdhci_uninitfn(SDHCIState *s)
@@ -1399,9 +1401,13 @@ void sdhci_common_realize(SDHCIState *s, Error **errp)
 

What about simply keeping the same code guarded with 'if (!s->io_ops)'?

I chose below approach since it provides an error message when one attempts to set one of the other device models to BE rather than just silently ignoring it.

Also, I didn't want to make the assumption that s->io_ops == NULL implied that sdhci_mmio_*_ops is needed. That's similar material the bug fixed is made of, so I wanted to prevent that in the first place by being more explicit.

In combination with the new error message the limitations of the current code become hopefully very apparent now, and at the same time should provide enough hints for adding BE support to the other device models if ever needed.

Quoted 9 lines
Quoted 4 lines
Quoted 2 lines

Best regards,
Bernhard

Ping

Ping^2

I would like to have the bug fixed in 8.1.

+1

Not that I care too much - I build qemu myself anyway and carry the patch locally - but this really should get fixed.

Guenter

#8 lore raw
Quoted 7 lines · Guenter Roeck
Quoted 4 lines · Bernhard Beschow

Ping^2

I would like to have the bug fixed in 8.1.

Quoted 7 lines

+1

Not that I care too much - I build qemu myself anyway and carry the patch locally - but this really should get fixed.

Guenter

Patch queued, thanks!