[Qemu-devel] TBL register permissions for PPC

[Dima Ilyevsky]

Hello All,

I have a question about read permissions of TBL SPR for all ppc processors:

I have discovered that my application, compiled by WindRiver diab compiler and running in vxworks OS on ppc405 architecture bumps into exception generated when trying to read TBL or TBU registers:

program

Exception current instruction address: 0x0003eb28

Machine Status Register: 0x00088200

Condition Register: 0x40000440

<<<<<Registers>>>>>

.....

<<<<<Disassembly>>>>>

 0x3eb08  4e800020    blr

 0x3eb0c  3860ffff    li          r3,0xffff # -1

 0x3eb10  4e800020    blr

 0x3eb14  38a00000    li          r5,0x0 # 0

 0x3eb18  7cbc43a6    mtspr       TBL,r5

 0x3eb1c  7c7d43a6    mtspr       TBU,r3

 0x3eb20  7c9c43a6    mtspr       TBL,r4

 0x3eb24  4e800020    blr

*0x3eb28  7cbd42a6    mfspr       r5,TBU

 0x3eb2c  7cdc42a6    mfspr       r6,TBL

 0x3eb30  7cfd42a6     mfspr       r7,TBU

 0x3eb34  7c053800    cmp         crf0,0,r5,r7

 0x3eb38  4082fff0       bc          0x4,2, 0x3eb28 # vxTimeBaseGet

 0x3eb3c  90a30000    stw         r5,0(r3)

 0x3eb40  90c40000    stw         r6,0(r4)

 0x3eb44  4e800020    blr

<<<<<Traceback>>>>>

......

I looked into qemu mfspr instruction implementation and discovered that for some reason qemu had set NOACCESS permissions for reading this register upon CPU initialization:

/* Generic PowerPC time base */

static void gen_tbl (CPUPPCState *env)

{

    spr_register(env, SPR_VTBL,  "TBL",

                 &spr_read_tbl, SPR_NOACCESS,

                 &spr_read_tbl, SPR_NOACCESS,

                 0x00000000);

    spr_register(env, SPR_TBL,   "TBL",

                 SPR_NOACCESS, SPR_NOACCESS,

                 SPR_NOACCESS, &spr_write_tbl,

                 0x00000000);

    spr_register(env, SPR_VTBU,  "TBU",

                 &spr_read_tbu, SPR_NOACCESS,

                 &spr_read_tbu, SPR_NOACCESS,

                 0x00000000);

    spr_register(env, SPR_TBU,   "TBU",

                 SPR_NOACCESS, SPR_NOACCESS,

                 SPR_NOACCESS, &spr_write_tbu,

                 0x00000000);

}

Is this a right thing to do? it's certainly either qemu bug or diab bug (which i don't think is likely, cause this code has been running on the real hw without any issues)

BR,

Dmitry I.
--


Jonathan Swift  - "May you live every day of your life."