[v4] Implement Firmware Assisted Dump for PSeries

[PATCH v4 6/8] hw/ppc: Pass dump-sizes property for fadump in device tree

Posted by Aditya Gupta 10 months, 3 weeks ago

Platform (ie. QEMU) is expected to pass few device tree properties for
details for fadump:

  * "ibm,configure-kernel-dump-sizes": Space required to store dump data
    for firmware provided dump sections (ie. CPU & HPTE regions)
  * "ibm,configure-kernel-dump-version": Versions of fadump supported

Pass the above device tree nodes so that kernel can reserve sufficient
space for preserving the CPU state data

Note: As of this patch, the "kernel-dump" device tree entry is still not
added for the second boot, so after crash, the second kernel will boot
assuming fadump dump is "NOT" active, and try to register for fadump,
but since we already have fadump registered in QEMU internal state, the
register rtas call will fail with: "DUMP ACTIVE"

Signed-off-by: Aditya Gupta <adityag@linux.ibm.com>
---
 hw/ppc/spapr.c | 57 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 57 insertions(+)

diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index a415e51d077a..3cbc6a7409b7 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -900,6 +900,61 @@ static int spapr_dt_rng(void *fdt)
     return ret ? -1 : 0;
 }
 
+static void spapr_dt_rtas_fadump(SpaprMachineState *spapr, void *fdt, int rtas)
+{
+    MachineState *ms = MACHINE(spapr);
+    MachineClass *mc = MACHINE_GET_CLASS(ms);
+
+    uint32_t max_possible_cpus = mc->possible_cpu_arch_ids(ms)->len;
+    uint64_t fadump_cpu_state_size = 0;
+    uint16_t fadump_versions[2] = {
+        FADUMP_VERSION /* min supported version */,
+        FADUMP_VERSION /* max supported version */
+    };
+    uint32_t fadump_rgn_sizes[2][3] = {
+        {
+            cpu_to_be32(FADUMP_CPU_STATE_DATA),
+            0, 0 /* Calculated later */
+        },
+        {
+            cpu_to_be32(FADUMP_HPTE_REGION),
+            0, 0 /* HPTE region not implemented */
+        }
+    };
+
+    /*
+     * CPU State Data contains multiple fields such as header, num_cpus and
+     * register entries
+     *
+     * Calculate the maximum CPU State Data size, according to maximum
+     * possible CPUs the QEMU VM can have
+     *
+     * This calculation must match the 'cpu_state_len' calculation done in
+     * 'populate_cpu_state_data' in spapr_fadump.c
+     */
+    fadump_cpu_state_size += sizeof(struct FadumpRegSaveAreaHeader);
+    fadump_cpu_state_size += 0xc;                      /* padding as in PAPR */
+    fadump_cpu_state_size += sizeof(__be32);           /* num_cpus */
+    fadump_cpu_state_size += max_possible_cpus   *     /* reg entries */
+                             FADUMP_NUM_PER_CPU_REGS *
+                             sizeof(struct FadumpRegEntry);
+
+    /* Set maximum size for CPU state data region */
+    assert(fadump_rgn_sizes[0][0] == cpu_to_be32(FADUMP_CPU_STATE_DATA));
+
+    /* Upper 32 bits of size, usually 0 */
+    fadump_rgn_sizes[0][1] = cpu_to_be32(fadump_cpu_state_size >> 32);
+
+    /* Lower 32 bits of size */
+    fadump_rgn_sizes[0][2] = cpu_to_be32(fadump_cpu_state_size & 0xffffffff);
+
+    /* Add device tree properties required from platform for fadump */
+    _FDT((fdt_setprop(fdt, rtas, "ibm,configure-kernel-dump-version",
+                    fadump_versions, sizeof(fadump_versions))));
+    _FDT((fdt_setprop(fdt, rtas, "ibm,configure-kernel-dump-sizes",
+                    fadump_rgn_sizes, sizeof(fadump_rgn_sizes))));
+}
+
 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
 {
     MachineState *ms = MACHINE(spapr);
@@ -1012,6 +1067,8 @@ static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
                      lrdr_capacity, sizeof(lrdr_capacity)));
 
+    spapr_dt_rtas_fadump(spapr, fdt, rtas);
+
     spapr_dt_rtas_tokens(fdt, rtas);
 }
 
-- 
2.49.0

Re: [PATCH v4 6/8] hw/ppc: Pass dump-sizes property for fadump in device tree

Posted by Sourabh Jain 3 months, 3 weeks ago



On 23/03/25 23:10, Aditya Gupta wrote:
> Platform (ie. QEMU) is expected to pass few device tree properties for
> details for fadump:
>
>    * "ibm,configure-kernel-dump-sizes": Space required to store dump data
>      for firmware provided dump sections (ie. CPU & HPTE regions)
>    * "ibm,configure-kernel-dump-version": Versions of fadump supported
>
> Pass the above device tree nodes so that kernel can reserve sufficient
> space for preserving the CPU state data
>
> Note: As of this patch, the "kernel-dump" device tree entry is still not
> added for the second boot, so after crash, the second kernel will boot
> assuming fadump dump is "NOT" active, and try to register for fadump,
> but since we already have fadump registered in QEMU internal state, the
> register rtas call will fail with: "DUMP ACTIVE"
>
> Signed-off-by: Aditya Gupta <adityag@linux.ibm.com>
> ---
>   hw/ppc/spapr.c | 57 ++++++++++++++++++++++++++++++++++++++++++++++++++
>   1 file changed, 57 insertions(+)
>
> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> index a415e51d077a..3cbc6a7409b7 100644
> --- a/hw/ppc/spapr.c
> +++ b/hw/ppc/spapr.c
> @@ -900,6 +900,61 @@ static int spapr_dt_rng(void *fdt)
>       return ret ? -1 : 0;
>   }
>   
> +static void spapr_dt_rtas_fadump(SpaprMachineState *spapr, void *fdt, int rtas)
> +{
> +    MachineState *ms = MACHINE(spapr);
> +    MachineClass *mc = MACHINE_GET_CLASS(ms);
> +
> +    uint32_t max_possible_cpus = mc->possible_cpu_arch_ids(ms)->len;
> +    uint64_t fadump_cpu_state_size = 0;
> +    uint16_t fadump_versions[2] = {
> +        FADUMP_VERSION /* min supported version */,
> +        FADUMP_VERSION /* max supported version */
> +    };
> +    uint32_t fadump_rgn_sizes[2][3] = {
> +        {
> +            cpu_to_be32(FADUMP_CPU_STATE_DATA),
> +            0, 0 /* Calculated later */
> +        },
> +        {
> +            cpu_to_be32(FADUMP_HPTE_REGION),
> +            0, 0 /* HPTE region not implemented */
> +        }

#nit-pick
Why to advertise if we don't support it? Kernel anyways ignores this for 
now.

> +    };
> +
> +    /*
> +     * CPU State Data contains multiple fields such as header, num_cpus and
> +     * register entries
> +     *
> +     * Calculate the maximum CPU State Data size, according to maximum
> +     * possible CPUs the QEMU VM can have
> +     *
> +     * This calculation must match the 'cpu_state_len' calculation done in
> +     * 'populate_cpu_state_data' in spapr_fadump.c
> +     */
> +    fadump_cpu_state_size += sizeof(struct FadumpRegSaveAreaHeader);
> +    fadump_cpu_state_size += 0xc;                      /* padding as in PAPR */
> +    fadump_cpu_state_size += sizeof(__be32);           /* num_cpus */
> +    fadump_cpu_state_size += max_possible_cpus   *     /* reg entries */
> +                             FADUMP_NUM_PER_CPU_REGS *
> +                             sizeof(struct FadumpRegEntry);
> +
> +    /* Set maximum size for CPU state data region */
> +    assert(fadump_rgn_sizes[0][0] == cpu_to_be32(FADUMP_CPU_STATE_DATA));
> +
> +    /* Upper 32 bits of size, usually 0 */
> +    fadump_rgn_sizes[0][1] = cpu_to_be32(fadump_cpu_state_size >> 32);
> +
> +    /* Lower 32 bits of size */
> +    fadump_rgn_sizes[0][2] = cpu_to_be32(fadump_cpu_state_size & 0xffffffff);
> +
> +    /* Add device tree properties required from platform for fadump */
> +    _FDT((fdt_setprop(fdt, rtas, "ibm,configure-kernel-dump-version",
> +                    fadump_versions, sizeof(fadump_versions))));
> +    _FDT((fdt_setprop(fdt, rtas, "ibm,configure-kernel-dump-sizes",
> +                    fadump_rgn_sizes, sizeof(fadump_rgn_sizes))));
> +}
> +
>   static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
>   {
>       MachineState *ms = MACHINE(spapr);
> @@ -1012,6 +1067,8 @@ static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
>       _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
>                        lrdr_capacity, sizeof(lrdr_capacity)));
>   
> +    spapr_dt_rtas_fadump(spapr, fdt, rtas);
> +
>       spapr_dt_rtas_tokens(fdt, rtas);
>   }
>

Re: [PATCH v4 6/8] hw/ppc: Pass dump-sizes property for fadump in device tree

Posted by Aditya Gupta 3 months, 3 weeks ago

On 25/10/18 04:50PM, Sourabh Jain wrote:
> 
> 
> > <...snip...>
> > +    uint32_t fadump_rgn_sizes[2][3] = {
> > +        {
> > +            cpu_to_be32(FADUMP_CPU_STATE_DATA),
> > +            0, 0 /* Calculated later */
> > +        },
> > +        {
> > +            cpu_to_be32(FADUMP_HPTE_REGION),
> > +            0, 0 /* HPTE region not implemented */
> > +        }
> 
> #nit-pick
> Why to advertise if we don't support it? Kernel anyways ignores this for
> now.

Nice catch.
PAPR doesn't seem to say about HPTE being optional anywhere, nor being
mandatory, so to be on safe side, exported it with 0 size until/if
it's implemented.

PAPR R1–7.3.30–7 says this (trimmed, emphasis mine):

	The platform 'must' present the RTAS property,
	“ibm,configure-kernel-dump-sizes” in the OF device tree, which
	describes space required for 'firmware defined dump sections', where
	the 'firmware defined dump sections' are:
		0x0001 = CPU State Data
		0x0002 = Hardware Page Table for Real Mode Region

Thanks,
- Aditya G

Re: [PATCH v4 6/8] hw/ppc: Pass dump-sizes property for fadump in device tree

Posted by Sourabh Jain 3 months, 3 weeks ago


On 20/10/25 01:00, Aditya Gupta wrote:
> On 25/10/18 04:50PM, Sourabh Jain wrote:
>>
>>> <...snip...>
>>> +    uint32_t fadump_rgn_sizes[2][3] = {
>>> +        {
>>> +            cpu_to_be32(FADUMP_CPU_STATE_DATA),
>>> +            0, 0 /* Calculated later */
>>> +        },
>>> +        {
>>> +            cpu_to_be32(FADUMP_HPTE_REGION),
>>> +            0, 0 /* HPTE region not implemented */
>>> +        }
>> #nit-pick
>> Why to advertise if we don't support it? Kernel anyways ignores this for
>> now.
> Nice catch.
> PAPR doesn't seem to say about HPTE being optional anywhere, nor being
> mandatory, so to be on safe side, exported it with 0 size until/if
> it's implemented.
>
> PAPR R1–7.3.30–7 says this (trimmed, emphasis mine):
>
> 	The platform 'must' present the RTAS property,
> 	“ibm,configure-kernel-dump-sizes” in the OF device tree, which
> 	describes space required for 'firmware defined dump sections', where
> 	the 'firmware defined dump sections' are:
> 		0x0001 = CPU State Data
> 		0x0002 = Hardware Page Table for Real Mode Region
Yeah lets advertise HPTE with zero size.

- Sourabh

[PATCH v4 1/8] hw/ppc: Implement skeleton code for fadump in PSeries
[PATCH v4 2/8] hw/ppc: Implement fadump register command
[PATCH v4 3/8] hw/ppc: Trigger Fadump boot if fadump is registered
[PATCH v4 4/8] hw/ppc: Preserve memory regions registered for fadump
[PATCH v4 5/8] hw/ppc: Implement saving CPU state in Fadump
[PATCH v4 6/8] hw/ppc: Pass dump-sizes property for fadump in device tree
[PATCH v4 7/8] hw/ppc: Enable fadump for PSeries
[PATCH v4 8/8] tests/functional: Add test for fadump in PSeries