Skip to contentSkip to conversation

riscv: Generate devicetree only after machine initialization is complete

PATCH Superseded Guenter Roeck 3 participants 7 messages Jul 3, 2023 RISC-V

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 riscv: Generate devicetree only after machine initialization is complete

If the devicetree is created before machine initialization is complete, it misses dynamic devices. Specifically, the tpm device is not added to the devicetree file and is therefore not instantiated in Linux. Create devicetree in virt_machine_done() to solve the problem.

Cc Alistair Francis <[email protected]>
Fixes 325b7c4e75 hw/riscv: Enable TPM backends
---
 hw/riscv/virt.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
hw/riscv/virt.c +6 -3
index ed4c27487e..08876284f5 100644
@@ -1248,6 +1248,11 @@ static void virt_machine_done(Notifier *notifier, void *data)
     uint64_t kernel_entry = 0;
     BlockBackend *pflash_blk0;
 
+    /* create devicetree if not provided */
+    if (!machine->dtb) {
+        create_fdt(s, memmap);
+    }
+
     /*
      * Only direct boot kernel is currently supported for KVM VM,
      * so the "-bios" parameter is not supported when KVM is enabled.
@@ -1508,15 +1513,13 @@ static void virt_machine_init(MachineState *machine)
     }
     virt_flash_map(s, system_memory);
 
-    /* load/create device tree */
+    /* load device tree */
     if (machine->dtb) {
         machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
         if (!machine->fdt) {
             error_report("load_device_tree() failed");
             exit(1);
         }
-    } else {
-        create_fdt(s, memmap);
     }
 
     s->machine_done.notify = virt_machine_done;
Signature · 1 line
2.39.2
#1 lore raw
Quoted 4 lines · Guenter Roeck

If the devicetree is created before machine initialization is complete, it misses dynamic devices. Specifically, the tpm device is not added to the devicetree file and is therefore not instantiated in Linux. Create devicetree in virt_machine_done() to solve the problem.

This makes sense, but what about the other archs/machines? Shouldn't we fix this generically?

Quoted 40 lines
Cc Alistair Francis <[email protected]>
Fixes 325b7c4e75 hw/riscv: Enable TPM backends
---
  hw/riscv/virt.c | 9 ++++++---
  1 file changed, 6 insertions(+), 3 deletions(-)
hw/riscv/virt.c +6 -3
index ed4c27487e..08876284f5 100644
@@ -1248,6 +1248,11 @@ static void virt_machine_done(Notifier *notifier, void *data)
      uint64_t kernel_entry = 0;
      BlockBackend *pflash_blk0;
 
+    /* create devicetree if not provided */
+    if (!machine->dtb) {
+        create_fdt(s, memmap);
+    }
+
      /*
       * Only direct boot kernel is currently supported for KVM VM,
       * so the "-bios" parameter is not supported when KVM is enabled.
@@ -1508,15 +1513,13 @@ static void virt_machine_init(MachineState *machine)
      }
      virt_flash_map(s, system_memory);
 
-    /* load/create device tree */
+    /* load device tree */
      if (machine->dtb) {
          machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
          if (!machine->fdt) {
              error_report("load_device_tree() failed");
              exit(1);
          }
-    } else {
-        create_fdt(s, memmap);
      }
 
      s->machine_done.notify = virt_machine_done;
#2 lore raw
Quoted 9 lines · Philippe Mathieu-Daudé
Quoted 4 lines · Guenter Roeck

If the devicetree is created before machine initialization is complete, it misses dynamic devices. Specifically, the tpm device is not added to the devicetree file and is therefore not instantiated in Linux. Create devicetree in virt_machine_done() to solve the problem.

This makes sense, but what about the other archs/machines? Shouldn't we fix this generically?

I had a brief look into various implementations. To me the code looks all very machine specific. I don't think it can be solved generically (if other machines are even affected), but then I am not a qemu expert and may be wrong.

Guenter

Quoted 38 lines
Quoted 37 lines
Cc Alistair Francis <[email protected]>
Fixes 325b7c4e75 hw/riscv: Enable TPM backends

---
  hw/riscv/virt.c | 9 ++++++---

  1 file changed, 6 insertions(+), 3 deletions(-)
hw/riscv/virt.c +6 -3
index ed4c27487e..08876284f5 100644
@@ -1248,6 +1248,11 @@ static void virt_machine_done(Notifier *notifier, void *data)
       uint64_t kernel_entry = 0;
       BlockBackend *pflash_blk0;
+    /* create devicetree if not provided */
+    if (!machine->dtb) {
+        create_fdt(s, memmap);
+    }
+
       /*
        * Only direct boot kernel is currently supported for KVM VM,
        * so the "-bios" parameter is not supported when KVM is enabled.
@@ -1508,15 +1513,13 @@ static void virt_machine_init(MachineState *machine)
       }
       virt_flash_map(s, system_memory);
-    /* load/create device tree */
+    /* load device tree */
       if (machine->dtb) {
           machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
           if (!machine->fdt) {
               error_report("load_device_tree() failed");
               exit(1);
           }
-    } else {
-        create_fdt(s, memmap);
       }
       s->machine_done.notify = virt_machine_done;
#3 lore raw
In reply to Philippe Mathieu-Daudé
Quoted 8 lines · Philippe Mathieu-Daudé
Quoted 4 lines · Guenter Roeck

If the devicetree is created before machine initialization is complete, it misses dynamic devices. Specifically, the tpm device is not added to the devicetree file and is therefore not instantiated in Linux. Create devicetree in virt_machine_done() to solve the problem.

This makes sense, but what about the other archs/machines? Shouldn't we fix this generically?

As far as other archs goes I can say that ARM isn't affected by it because the fdt creation is done by arm_load_dtb() during virt_machine_done time. I'm not aware of how x86 handles TPM. And pseries/ppc64 does a completely different thing (per usual).

Inside hw/riscv the only TPM capable board is 'virt'. So I think this patch has an adequate scope.

Thanks,

Daniel

Quoted 39 lines
Quoted 38 lines
Cc Alistair Francis <[email protected]>
Fixes 325b7c4e75 hw/riscv: Enable TPM backends

---
  hw/riscv/virt.c | 9 ++++++---

  1 file changed, 6 insertions(+), 3 deletions(-)
hw/riscv/virt.c +6 -3
index ed4c27487e..08876284f5 100644
@@ -1248,6 +1248,11 @@ static void virt_machine_done(Notifier *notifier, void *data)
       uint64_t kernel_entry = 0;
       BlockBackend *pflash_blk0;
+    /* create devicetree if not provided */
+    if (!machine->dtb) {
+        create_fdt(s, memmap);
+    }
+
       /*
        * Only direct boot kernel is currently supported for KVM VM,
        * so the "-bios" parameter is not supported when KVM is enabled.
@@ -1508,15 +1513,13 @@ static void virt_machine_init(MachineState *machine)
       }
       virt_flash_map(s, system_memory);
-    /* load/create device tree */
+    /* load device tree */
       if (machine->dtb) {
           machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
           if (!machine->fdt) {
               error_report("load_device_tree() failed");
               exit(1);
           }
-    } else {
-        create_fdt(s, memmap);
       }
       s->machine_done.notify = virt_machine_done;
#4 lore raw
In reply to Guenter Roeck
Quoted 17 lines · Guenter Roeck

If the devicetree is created before machine initialization is complete, it misses dynamic devices. Specifically, the tpm device is not added to the devicetree file and is therefore not instantiated in Linux. Create devicetree in virt_machine_done() to solve the problem.

Cc Alistair Francis <[email protected]>
Fixes 325b7c4e75 hw/riscv: Enable TPM backends
---
  hw/riscv/virt.c | 9 ++++++---
  1 file changed, 6 insertions(+), 3 deletions(-)
hw/riscv/virt.c +0 -0
index ed4c27487e..08876284f5 100644

hw/riscv/virt.c static void virt_machine_done(Notifier *notifier, void *data) · line 1255

Quoted 8 lines
hw/riscv/virt.c +5 -0
@@ -1248,6 +1248,11 @@ static void virt_machine_done(Notifier *notifier, void *data)
      uint64_t kernel_entry = 0;
      BlockBackend *pflash_blk0;
 
+    /* create devicetree if not provided */
+    if (!machine->dtb) {
+        create_fdt(s, memmap);
+    }
+

I suggest moving the entire load/create DT code from virt_machine_init() to the start of virt_machine_done():

    /* load/create device tree */
    if (machine->dtb) {
        machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
        if (!machine->fdt) {
            error_report("load_device_tree() failed");
            exit(1);
        }
    } else {
        create_fdt(s, memmap);
    }

This way we don't have to look in to 2 different functions to wonder what happens in case machine->dtb is NULL.

Thanks,

Daniel

Quoted 20 lines
hw/riscv/virt.c +1 -3
      /*
       * Only direct boot kernel is currently supported for KVM VM,
       * so the "-bios" parameter is not supported when KVM is enabled.
@@ -1508,15 +1513,13 @@ static void virt_machine_init(MachineState *machine)
      }
      virt_flash_map(s, system_memory);
 
-    /* load/create device tree */
+    /* load device tree */
      if (machine->dtb) {
          machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
          if (!machine->fdt) {
              error_report("load_device_tree() failed");
              exit(1);
          }
-    } else {
-        create_fdt(s, memmap);
      }
 
      s->machine_done.notify = virt_machine_done;
#5 lore raw
Quoted 40 lines · Daniel Henrique Barboza
Quoted 17 lines · Guenter Roeck

If the devicetree is created before machine initialization is complete, it misses dynamic devices. Specifically, the tpm device is not added to the devicetree file and is therefore not instantiated in Linux. Create devicetree in virt_machine_done() to solve the problem.

Cc Alistair Francis <[email protected]>
Fixes 325b7c4e75 hw/riscv: Enable TPM backends

---
  hw/riscv/virt.c | 9 ++++++---

  1 file changed, 6 insertions(+), 3 deletions(-)
hw/riscv/virt.c +0 -0
index ed4c27487e..08876284f5 100644

hw/riscv/virt.c static void virt_machine_done(Notifier *notifier, void *data) · line 1254

Quoted 7 lines
hw/riscv/virt.c +5 -0
@@ -1248,6 +1248,11 @@ static void virt_machine_done(Notifier *notifier, void *data)
       uint64_t kernel_entry = 0;
       BlockBackend *pflash_blk0;
+    /* create devicetree if not provided */
+    if (!machine->dtb) {
+        create_fdt(s, memmap);
+    }
+

I suggest moving the entire load/create DT code from virt_machine_init() to the start of virt_machine_done():

    /* load/create device tree */
    if (machine->dtb) {
        machine->fdt = load_device_tree(machine->dtb, &s->fdt_size); if (!machine->fdt) { error_report("load_device_tree() failed"); exit(1);
        }
    } else {
        create_fdt(s, memmap);
    }

Quoted 3 lines

This way we don't have to look in to 2 different functions to wonder what happens in case machine->dtb is NULL.

I can do that, but I don't know how to test it. Is there a working dtb/machine combination for riscv which would let me test loading a devicetree file ?

Guenter

#6 lore raw
Quoted 41 lines · Guenter Roeck
Quoted 40 lines · Daniel Henrique Barboza
Quoted 17 lines · Guenter Roeck

If the devicetree is created before machine initialization is complete, it misses dynamic devices. Specifically, the tpm device is not added to the devicetree file and is therefore not instantiated in Linux. Create devicetree in virt_machine_done() to solve the problem.

Cc Alistair Francis <[email protected]>
Fixes 325b7c4e75 hw/riscv: Enable TPM backends

---
  hw/riscv/virt.c | 9 ++++++---

  1 file changed, 6 insertions(+), 3 deletions(-)
hw/riscv/virt.c +0 -0
index ed4c27487e..08876284f5 100644

hw/riscv/virt.c static void virt_machine_done(Notifier *notifier, void *data) · line 1254

Quoted 7 lines
hw/riscv/virt.c +5 -0
@@ -1248,6 +1248,11 @@ static void virt_machine_done(Notifier *notifier, void *data)
       uint64_t kernel_entry = 0;
       BlockBackend *pflash_blk0;
+    /* create devicetree if not provided */
+    if (!machine->dtb) {
+        create_fdt(s, memmap);
+    }
+

I suggest moving the entire load/create DT code from virt_machine_init() to the start of virt_machine_done():

     /* load/create device tree */
     if (machine->dtb) {
         machine->fdt = load_device_tree(machine->dtb, &s->fdt_size); if (!machine->fdt) { error_report("load_device_tree() failed"); exit(1);
         }
     } else {
         create_fdt(s, memmap);
     }

Quoted 6 lines
Quoted 3 lines

This way we don't have to look in to 2 different functions to wonder what happens in case machine->dtb is NULL.

I can do that, but I don't know how to test it. Is there a working dtb/machine combination for riscv which would let me test loading a devicetree file ?

I recommend using your own setup with TPM (I'm assuming you're using a TPM setup), generate a .dtb from it, and then launch it using '-dtb'.

First you need the patch applied (otherwise there won't be a TPM in the FDT). After that, relaunch the same machine again but appending in the end of the command line:

-machine dumpdtb=file.dtb

This will create a 'file.dtb' file in the working dir and exit. After that re-launch the machine again but now append:

-dtb file.dtb

And you should be able to boot a 'virt' machine with TPM support.

Thanks,

Daniel

Quoted 3 lines

Guenter