Re: [Qemu-ppc] [PATCH v3 2/2] target-ppc: Implement rtas_get_sysparm(PROCESSOR_MODULE_INFO)

[David Gibson]

On Fri, Nov 13, 2015 at 06:13:08PM -0800, Sukadev Bhattiprolu wrote:
> Implement RTAS_SYSPARM_PROCESSOR_MODULE_INFO parameter to rtas_get_sysparm()
> call in qemu. This call returns the processor module (socket), chip and core
> information as specified in section 7.3.16.17 of LoPAPR v2.7.
> 
> We walk the /proc/device-tree to determine the number of modules(aka sockets),
> chips and cores in the _host_ system and return this info to the guest
> application that makes the rtas_get_sysparm() call.
> 
> We currently hard code the number of module_types to 1 (we currently don't
> have systems with more than one module type) but we should determine that
> dynamically somehow later.
> 
> Thanks to input from Nishanth Aravamudan, Alexey Kardashevskiy, David Gibson,
> Thomas Huth.
> 
> Signed-off-by: Sukadev Bhattiprolu <address@hidden>
> ---
> Changelog[v3]:
>         [David Gibson] Use glob() to simplify pattern matching path names.
>             Move new code into target-ppc/{kvm.c,kvm_ppc.h} to not impact
>             other targets and avoid creating new files.
>         [Alexey Kardashevskiy] Fix indentation, error messages; fold new
>             code into existing files and avoid creating new files for this
>             parameter; since the keys are integer, use g_direct_hash() and
>             GINT_TO_POINTER() and avoid allocating/freeing memory.
> 
> Changelog[v2]:
>         [Alexey Kardashevskiy] Use existing interfaces like ldl_be_p(),
>             stw_be_phys(), g_hash_table_new_full(), error_report() rather
>             than re-inventing; fix indentation, function prottypes etc;
>             Drop the fts() interface (which doesn't seem to be available
>             on mingw32/mingw64) and use opendir() to walk specific
>             directories in the directory tree.
> ---
>  hw/ppc/spapr_rtas.c    |  31 +++++++++++
>  include/hw/ppc/spapr.h |  13 +++++
>  target-ppc/kvm.c       | 141 
> +++++++++++++++++++++++++++++++++++++++++++++++++
>  target-ppc/kvm_ppc.h   |   8 +++
>  4 files changed, 193 insertions(+)
> 
> diff --git a/hw/ppc/spapr_rtas.c b/hw/ppc/spapr_rtas.c
> index 34b12a3..44b2537 100644
> --- a/hw/ppc/spapr_rtas.c
> +++ b/hw/ppc/spapr_rtas.c
> @@ -38,6 +38,7 @@
>  
>  #include <libfdt.h>
>  #include "hw/ppc/spapr_drc.h"
> +#include "kvm_ppc.h"
>  
>  /* #define DEBUG_SPAPR */
>  
> @@ -240,6 +241,36 @@ static void rtas_ibm_get_system_parameter(PowerPCCPU 
> *cpu,
>      target_ulong ret = RTAS_OUT_SUCCESS;
>  
>      switch (parameter) {
> +    case RTAS_SYSPARM_PROCESSOR_MODULE_INFO: {
> +        struct sPAPRRTASModuleInfo modinfo;
> +        int i, size = sizeof(modinfo), offset = 0;
> +
> +        memset(&modinfo, 0, size);
> +        if (kvmppc_rtas_get_module_info(&modinfo)) {
> +            ret = RTAS_OUT_HW_ERROR;
> +            break;
> +        }
> +
> +        stw_be_phys(&address_space_memory, buffer+offset, size);

You're still advertising the full structure size to the guest, even
though it may be only partially populated.

That will probably work in practice, but I think we should be
PAPRishly correct and only output the size that we actually use here.

> +        offset += 2;
> +
> +        stw_be_phys(&address_space_memory, buffer+offset, 
> modinfo.module_types);
> +        offset += 2;
> +
> +        for (i = 0; i < modinfo.module_types; i++) {
> +            stw_be_phys(&address_space_memory, buffer+offset,
> +                        modinfo.si[i].sockets);
> +            offset += 2;
> +            stw_be_phys(&address_space_memory, buffer+offset,
> +                        modinfo.si[i].chips);
> +            offset += 2;
> +            stw_be_phys(&address_space_memory, buffer+offset,
> +                        modinfo.si[i].cores_per_chip);
> +            offset += 2;
> +        }
> +        break;
> +    }
> +
>      case RTAS_SYSPARM_SPLPAR_CHARACTERISTICS: {
>          char *param_val = g_strdup_printf("MaxEntCap=%d,"
>                                            "DesMem=%llu,"
> diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
> index 5baa906..f793465 100644
> --- a/include/hw/ppc/spapr.h
> +++ b/include/hw/ppc/spapr.h
> @@ -463,6 +463,7 @@ int spapr_allocate_irq_block(int num, bool lsi, bool msi);
>  /* RTAS ibm,get-system-parameter token values */
>  #define RTAS_SYSPARM_SPLPAR_CHARACTERISTICS      20
>  #define RTAS_SYSPARM_DIAGNOSTICS_RUN_MODE        42
> +#define RTAS_SYSPARM_PROCESSOR_MODULE_INFO       43
>  #define RTAS_SYSPARM_UUID                        48
>  
>  /* RTAS indicator/sensor types
> @@ -646,4 +647,16 @@ int spapr_rng_populate_dt(void *fdt);
>   */
>  #define SPAPR_LMB_FLAGS_ASSIGNED 0x00000008
>  
> +#define SPAPR_MAX_MODULE_TYPES  1
> +
> +struct sPAPRRTASSocketInfo {
> +    unsigned short sockets;
> +    unsigned short chips;
> +    unsigned short cores_per_chip;
> +};
> +struct sPAPRRTASModuleInfo {
> +    unsigned short module_types;
> +    struct sPAPRRTASSocketInfo si[SPAPR_MAX_MODULE_TYPES];
> +};
> +
>  #endif /* !defined (__HW_SPAPR_H__) */
> diff --git a/target-ppc/kvm.c b/target-ppc/kvm.c
> index 9940a90..bbef78c 100644 7
> --- a/target-ppc/kvm.c
> +++ b/target-ppc/kvm.c
> @@ -15,6 +15,7 @@
>   */
>  
>  #include <dirent.h>
> +#include <glob.h>
>  #include <sys/types.h>
>  #include <sys/ioctl.h>
>  #include <sys/mman.h>
> @@ -2518,3 +2519,143 @@ int kvmppc_enable_hwrng(void)
>  
>      return kvmppc_enable_hcall(kvm_state, H_RANDOM);
>  }
> +
> +/* Read an identifier from the file @path and add the identifier
> + * to the hash table @gt unless its already in the table.
> + */
> +static int kvmppc_hash_file_contents(GHashTable *gt, char *path)
> +{
> +    uint32_t idx;
> +
> +    idx = kvmppc_read_int_dt(path);
> +    if (idx == -1) {
> +        return -1;
> +    }
> +
> +    if (g_hash_table_contains(gt, GINT_TO_POINTER(idx))) {
> +        return 0;
> +    }
> +
> +    if (!g_hash_table_insert(gt, GINT_TO_POINTER(idx), NULL)) {
> +        fprintf(stderr, "%s() Unable to add key %d\n", __func__, idx);
> +        return -1;
> +    }
> +
> +    return 0;
> +}
> +
> +static int kvmppc_glob_count_ids_dt(const char *pattern, int *count)
> +{
> +    int i, rc;
> +    glob_t dtglob;
> +    GHashTable *htbl;
> +
> +    rc = glob(pattern, GLOB_NOSORT, NULL, &dtglob);
> +    if (rc) {
> +        fprintf(stderr, "%s() glob(%s) returns %d, errno %d\n", __func__,
> +                pattern, rc, errno);
> +        return -1;
> +    }
> +
> +    rc = -1;
> +    htbl = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL, NULL);
> +
> +    for (i = 0; i < dtglob.gl_pathc; i++) {
> +        if (kvmppc_hash_file_contents(htbl, dtglob.gl_pathv[i])) {
> +            goto cleanup;
> +        }
> +    }
> +
> +    *count = g_hash_table_size(htbl);
> +    rc = 0;
> +
> +cleanup:
> +    globfree(&dtglob);
> +    g_hash_table_remove_all(htbl);
> +    g_hash_table_destroy(htbl);
> +
> +    return rc;
> +}
> +
> +/* Each socket's (aka module's) id is contained in the 'ibm,hw-module-id'
> + * file in an "xscom" directory (/proc/device-tree/xscom*). Similarly each
> + * chip's id is contained in the 'ibm,chip-id' file in an xscom directory.
> + *
> + * Since a module can have more than one chip and a chip can have more than
> + * one core, there are likely to be duplicates in the module and chip ids
> + * i.e more than one xscom directory can contain the same module/chip id.
> + *
> + * Search the xscom directories and count the number of _UNIQUE_ modules
> + * and chips in the system.
> + *
> + * Return 0 if one or more modules and chips each are found. Return -1
> + * otherwise.
> + */
> +static int kvmppc_count_sockets_chips_dt(int *num_sockets, int *num_chips)
> +{
> +    const char *chip_pattern = 
> "/proc/device-tree/address@hidden/ibm,chip-id";
> +    const char *module_pattern = 
> "/proc/device-tree/address@hidden/ibm,hw-module-id";
> +
> +    if (kvmppc_glob_count_ids_dt(module_pattern, num_sockets)
> +        || kvmppc_glob_count_ids_dt(chip_pattern, num_chips)) {
> +        return -1;
> +    }
> +
> +    if (*num_sockets == 0 || *num_chips == 0) {
> +        return -1;
> +    }
> +
> +    return 0;
> +}
> +
> +/* Each core in the system is represented by a directory with the prefix
> + * 'PowerPC,POWER' in directory /proc/device-tree/cpus/.  Process that
> + * directory and count the number of cores in the system.
> + *
> + * Return 0 if one or more cores are found. Return -1 otherwise.
> + */
> +static int kvmppc_count_cores_dt(int *num_cores)
> +{
> +    int rc;
> +    glob_t dtglob;
> +    const char *cpus_pattern = "/proc/device-tree/cpus/PowerPC,POWER*";

Under KVM PR, this could still be too specific to IBM machines.  I
think it's probably safer to just use /proc/device-tree/cpus/*, I
don't *think* we get anything under /cpus that isn't a cpu node.

In a number of ways I'd actually prefer to move to /cpus/address@hidden in
general, since that follows the OF generic names recommendation we
follow for most other nodes.

> +    rc = glob(cpus_pattern, GLOB_NOSORT, NULL, &dtglob);
> +    if (rc) {
> +        fprintf(stderr, "%s() glob(%s) returns %d, errno %d\n", __func__,
> +                cpus_pattern, rc, errno);
> +        return -1;
> +    }
> +
> +    *num_cores = dtglob.gl_pathc;
> +    globfree(&dtglob);
> +
> +    if (*num_cores == 0) {
> +        return -1;
> +    }
> +
> +    return 0;
> +}
> +
> +int kvmppc_rtas_get_module_info(struct sPAPRRTASModuleInfo *modinfo)
> +{
> +    int cores, chips, sockets;
> +
> +    if (kvmppc_count_sockets_chips_dt(&sockets, &chips)
> +        || kvmppc_count_cores_dt(&cores)) {
> +        return -1;
> +    }
> +
> +    /* TODO: Although the interface allows multiple module types, Power
> +     *       systems currently support only one module type per system -
> +     *       either Single Chip Module(SCM) or Dual Chip Module(DCM),
> +     *       but not both types in the same system. Hard code module_types
> +     *       for now till we can determine it dynamically.
> +     */
> +    modinfo->module_types = 1;
> +    modinfo->si[0].sockets = sockets;
> +    modinfo->si[0].chips = chips;
> +    modinfo->si[0].cores_per_chip = cores / chips;
> +
> +    return 0;
> +}
> diff --git a/target-ppc/kvm_ppc.h b/target-ppc/kvm_ppc.h
> index 309cbe0..46a2768 100644
> --- a/target-ppc/kvm_ppc.h
> +++ b/target-ppc/kvm_ppc.h
> @@ -10,6 +10,7 @@
>  #define __KVM_PPC_H__
>  
>  #define TYPE_HOST_POWERPC_CPU "host-" TYPE_POWERPC_CPU
> +#include "hw/ppc/spapr.h"
>  
>  #ifdef CONFIG_KVM
>  
> @@ -55,6 +56,7 @@ void kvmppc_hash64_write_pte(CPUPPCState *env, target_ulong 
> pte_index,
>                               target_ulong pte0, target_ulong pte1);
>  bool kvmppc_has_cap_fixup_hcalls(void);
>  int kvmppc_enable_hwrng(void);
> +int kvmppc_rtas_get_module_info(struct sPAPRRTASModuleInfo *modinfo);
>  
>  #else
>  
> @@ -256,6 +258,12 @@ static inline int kvmppc_enable_hwrng(void)
>  {
>      return -1;
>  }
> +
> +static inline int kvmppc_rtas_get_module_info(struct sPAPRRTASModuleInfo *mi)
> +{
> +    return -1;
> +}
> +
>  #endif
>  
>  #ifndef CONFIG_KVM

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature