[Skiboot] [RFC PATCH v3 4/6] xscoms: read/write xscoms using ucall

[Oliver O'Halloran]

On Wed, 2020-01-22 at 10:13 -0500, Ryan Grimm wrote:
> From: Madhavan Srinivasan <maddy at linux.vnet.ibm.com>
> 
> xscom registers are in the secure memory area when secure mode is
> enabled. These registers cannot be accessed directly and need to use
> ultravisor services using ultracall.
> 
> Signed-off-by: Madhavan Srinivasan <maddy at linux.vnet.ibm.com>
> Signed-off-by: Santosh Sivaraj <santosh at fossix.org>
> [ linuxram: Set uv_present just after starting UV ]
> Signed-off-by: Ram Pai <linuxram at us.ibm.com>
> [ grimm: Don't check MSR in xscom read/write ]
> Signed-off-by: Ryan Grimm <grimm at linux.ibm.com>
> ---
>  hw/ultravisor.c      |  4 ++++
>  include/ultravisor.h | 22 ++++++++++++++++++++++
>  include/xscom.h      | 11 +++++++++--
>  3 files changed, 35 insertions(+), 2 deletions(-)
> 
> diff --git a/hw/ultravisor.c b/hw/ultravisor.c
> index d55d752b..a5553757 100644
> --- a/hw/ultravisor.c
> +++ b/hw/ultravisor.c
> @@ -17,6 +17,7 @@
>  #include <libfdt/libfdt.h>
>  #include <libstb/container.h>
>  
> +bool uv_present = false;
>  static char *uv_image = NULL;
>  static size_t uv_image_size;
>  struct xz_decompress *uv_xz = NULL;
> @@ -251,6 +252,9 @@ int start_ultravisor(void)
>  
>  	cpu_start_ultravisor((void *)uv_opal);
>  
> +	/* From now on XSCOM must go through ucall */
> +	uv_present = true;
> +
>  	while (i > 0)
>  		cpu_wait_job(jobs[--i], true);
>  
> diff --git a/include/ultravisor.h b/include/ultravisor.h
> index daeb99b6..d212f463 100644
> --- a/include/ultravisor.h
> +++ b/include/ultravisor.h
> @@ -12,10 +12,15 @@
>   * for the secure virtual machines */
>  #define UV_SECURE_MEM_BIT              (PPC_BIT(15))
>  #define MAX_COMPRESSED_UV_IMAGE_SIZE 0x40000 /* 256 Kilobytes */
> +#define UV_READ_SCOM  0xF114
> +#define UV_WRITE_SCOM 0xF118
> +#define UCALL_BUFSIZE 4

>From 1/6:

> Return value of all ultracalls is in register R3. Other output values
> from the ultracall, if any, are returned in registers R4 through R12.

Which is up to 9 returned values. So where did 4 come from? This is an
accident waiting to happen.

>  #define UV_ACCESS_BIT		0x1ULL << 48
>  #define UV_LOAD_MAX_SIZE	0x200000
>  #define UV_FDT_MAX_SIZE		0x100000
>  
> +extern bool uv_present;
> +
>  extern int start_uv(uint64_t entry, struct uv_opal *uv_opal);
>  extern bool uv_add_mem_range(__be64 start, __be64 end);
>  extern void uv_preload_image(void);
> @@ -24,4 +29,21 @@ extern void init_uv(void);
>  extern int start_ultravisor(void);
>  extern long ucall(unsigned long opcode, unsigned long *retbuf, ...);
>  
> +static inline int uv_xscom_read(u64 partid, u64 pcb_addr, u64 *val)
> +{
> +	long rc;
> +	unsigned long retbuf[UCALL_BUFSIZE];
> +
> +	rc = ucall(UV_READ_SCOM, retbuf, partid, pcb_addr);
> +	*val = retbuf[0];
> +	return rc;
> +}
> +
> +static inline int uv_xscom_write(u64 partid, u64 pcb_addr, u64 val)
> +{
> +	unsigned long retbuf[UCALL_BUFSIZE];
> +
> +	return ucall(UV_WRITE_SCOM, retbuf, partid, pcb_addr, val);
> +}

Do we need to be logging xscom errors in skiboot?

> +
>  #endif /* __ULTRAVISOR_H */
> diff --git a/include/xscom.h b/include/xscom.h
> index 8a466d56..2f4c819b 100644
> --- a/include/xscom.h
> +++ b/include/xscom.h
> @@ -7,6 +7,7 @@
>  #include <stdint.h>
>  #include <processor.h>
>  #include <cpu.h>
> +#include <ultravisor.h>
>  
>  /*
>   * SCOM "partID" definitions:

> @@ -174,10 +175,16 @@ extern void _xscom_unlock(void);
>  /* Targeted SCOM access */
>  static inline int xscom_read(uint32_t partid, uint64_t pcb_addr, uint64_t *val)
>  {
> -	return _xscom_read(partid, pcb_addr, val, true);
> +	if (!uv_present)
> +		return _xscom_read(partid, pcb_addr, val, true);
> +
> +	return uv_xscom_read(partid, pcb_addr, val);
>  }
>  static inline int xscom_write(uint32_t partid, uint64_t pcb_addr, uint64_t val) {
> -	return _xscom_write(partid, pcb_addr, val, true);
> +	if (!uv_present)
> +		return _xscom_write(partid, pcb_addr, val, true);
> +
> +	return uv_xscom_write(partid, pcb_addr, val);

Seems kind of backwards since having a UV is the exceptional case. IMO
do:

	if (uv_present)
		return do_uv_scom()
	return _scom_write();

Less noise in the diff too.

>  }
>  extern int xscom_write_mask(uint32_t partid, uint64_t pcb_addr, uint64_t val, uint64_t mask);
>