[Pdbg] [PATCH v2 2/2] adu: Add arugments for block size
Amitay Isaacs
amitay at ozlabs.org
Wed Dec 12 17:30:11 AEDT 2018
On Tue, 2018-12-11 at 16:05 +1100, Alistair Popple wrote:
> Not all memory can be read with the default ADU block size of 8
> bytes. Specifically cache-inhibited access to some MMIO regions such
> as PCIe BAR spaces requires 4 byte accesses to avoid check stopping
> the machine.
>
> This patch adds an argument to the put/getmem commands to allow a
> specific block size to be selected.
>
> Signed-off-by: Alistair Popple <alistair at popple.id.au>
> ---
>
> Changes since v1:
> - Fixed up comments
> - Removed debug code
>
> libpdbg/adu.c | 140 +++++++++++++++++++++++++++++++++++++------
> -----------
> libpdbg/libpdbg.h | 18 ++++---
> libpdbg/target.h | 4 +-
> src/mem.c | 14 +++---
> src/pdbgproxy.c | 4 +-
> src/thread.c | 2 +-
> 6 files changed, 121 insertions(+), 61 deletions(-)
>
> diff --git a/libpdbg/adu.c b/libpdbg/adu.c
> index b67a43e..5396213 100644
> --- a/libpdbg/adu.c
> +++ b/libpdbg/adu.c
> @@ -84,20 +84,48 @@
> #define FBC_ALTD_DATA_DONE PPC_BIT(3)
> #define FBC_ALTD_PBINIT_MISSING PPC_BIT(18)
>
> +/* There are more general implementations of this with a loop and
> more
> + * performant implementations using GCC builtins which aren't
> + * portable. Given we only need a limited domain this is quick, easy
> + * and portable. */
> +uint8_t blog2(uint8_t x)
> +{
> + switch(x) {
> + case 1:
> + return 0;
> + case 2:
> + return 1;
> + case 4:
> + return 2;
> + case 8:
> + return 3;
> + case 16:
> + return 4;
> + case 32:
> + return 5;
> + case 64:
> + return 6;
> + default:
> + assert(0);
> + }
> +}
> +
> int adu_getmem(struct pdbg_target *adu_target, uint64_t start_addr,
> - uint8_t *output, uint64_t size)
> + uint8_t *output, uint64_t size, uint8_t block_size)
> {
> - return __adu_getmem(adu_target, start_addr, output, size,
> false);
> + return __adu_getmem(adu_target, start_addr, output,
> + size, false, block_size);
> }
>
> int adu_getmem_ci(struct pdbg_target *adu_target, uint64_t
> start_addr,
> - uint8_t *output, uint64_t size)
> + uint8_t *output, uint64_t size, uint8_t block_size)
> {
> - return __adu_getmem(adu_target, start_addr, output, size,
> true);
> + return __adu_getmem(adu_target, start_addr, output,
> + size, true, block_size);
> }
>
> int __adu_getmem(struct pdbg_target *adu_target, uint64_t
> start_addr,
> - uint8_t *output, uint64_t size, bool ci)
> + uint8_t *output, uint64_t size, bool ci, uint8_t
> block_size)
> {
> struct adu *adu;
> uint8_t *output0;
> @@ -109,33 +137,34 @@ int __adu_getmem(struct pdbg_target
> *adu_target, uint64_t start_addr,
>
> output0 = output;
>
> - /* Align start address to 8-byte boundary */
> - addr0 = 8 * (start_addr / 8);
> + /* Align start address to block_sized boundary */
> + addr0 = block_size * (start_addr / block_size);
>
> - /* We read data in 8-byte aligned chunks */
> - for (addr = addr0; addr < start_addr + size; addr += 8) {
> + /* We read data in block_sized aligned chunks */
> + for (addr = addr0; addr < start_addr + size; addr +=
> block_size) {
> uint64_t data;
>
> - if (adu->getmem(adu, addr, &data, ci))
> + if (adu->getmem(adu, addr, &data, ci, block_size))
> return -1;
>
> - /* ADU returns data in big-endian form in the register
> */
> + /* ADU returns data in big-endian form in the register.
> */
> data = __builtin_bswap64(data);
> + data >>= (addr & 0x7ull)*8;
The above line will be better if written based block_size (as
discussed).
>
> if (addr < start_addr) {
> size_t offset = start_addr - addr;
> - size_t n = (size <= 8-offset ? size : 8-
> offset);
> + size_t n = (size <= block_size-offset ? size :
> block_size-offset);
>
> memcpy(output, ((uint8_t *) &data) + offset,
> n);
> output += n;
> - } else if (addr + 8 > start_addr + size) {
> + } else if (addr + block_size > start_addr + size) {
> uint64_t offset = start_addr + size - addr;
>
> memcpy(output, &data, offset);
> output += offset;
> } else {
> - memcpy(output, &data, 8);
> - output += 8;
> + memcpy(output, &data, block_size);
> + output += block_size;
> }
>
> pdbg_progress_tick(output - output0, size);
> @@ -147,19 +176,19 @@ int __adu_getmem(struct pdbg_target
> *adu_target, uint64_t start_addr,
> }
>
> int adu_putmem(struct pdbg_target *adu_target, uint64_t start_addr,
> - uint8_t *output, uint64_t size)
> + uint8_t *output, uint64_t size, uint8_t block_size)
> {
> - return __adu_putmem(adu_target, start_addr, output, size,
> false);
> + return __adu_putmem(adu_target, start_addr, output, size,
> false, block_size);
> }
>
> int adu_putmem_ci(struct pdbg_target *adu_target, uint64_t
> start_addr,
> - uint8_t *output, uint64_t size)
> + uint8_t *output, uint64_t size, uint8_t block_size)
> {
> - return __adu_putmem(adu_target, start_addr, output, size,
> true);
> + return __adu_putmem(adu_target, start_addr, output, size, true,
> block_size);
> }
>
> int __adu_putmem(struct pdbg_target *adu_target, uint64_t
> start_addr,
> - uint8_t *input, uint64_t size, bool ci)
> + uint8_t *input, uint64_t size, bool ci, uint8_t
> block_size)
> {
> struct adu *adu;
> int rc = 0, tsize;
> @@ -169,20 +198,25 @@ int __adu_putmem(struct pdbg_target
> *adu_target, uint64_t start_addr,
> adu = target_to_adu(adu_target);
> end_addr = start_addr + size;
> for (addr = start_addr; addr < end_addr; addr += tsize, input
> += tsize) {
> - if ((addr % 8) || (addr + 8 > end_addr)) {
> - /* If the address is not 64-bit aligned we
> - * copy in a byte at a time until it is. */
> + if ((addr % block_size) || (addr + block_size >
> end_addr)) {
> + /* If the address is not aligned to block_size
> + * we copy the data in one byte at a time
> + * until it is aligned. */
> tsize = 1;
>
> - /* Copy the input data in with correct
> alignment */
> + /* Copy the input data in with correct
> + * alignment. Bytes need to aligned to the
> + * correct byte offset in the data register
> + * regardless of address. */
> data = ((uint64_t) *input) << 8*(8 - (addr % 8)
> - 1);
> } else {
> - tsize = 8;
> - memcpy(&data, input, sizeof(data));
> + tsize = block_size;
> + memcpy(&data, input, block_size);
> data = __builtin_bswap64(data);
> + data >>= (addr & 7ull)*8;
Same thing here too.
> }
>
> - adu->putmem(adu, addr, data, tsize, ci);
> + adu->putmem(adu, addr, data, tsize, ci, block_size);
> pdbg_progress_tick(addr - start_addr, size);
> }
>
> @@ -234,7 +268,8 @@ static int adu_reset(struct adu *adu)
> return 0;
> }
>
> -static int p8_adu_getmem(struct adu *adu, uint64_t addr, uint64_t
> *data, int ci)
> +static int p8_adu_getmem(struct adu *adu, uint64_t addr, uint64_t
> *data,
> + int ci, uint8_t block_size)
It might be better to add a new commands instead of modifying the
existing getmem/putmem behaviour.
> {
> uint64_t ctrl_reg, cmd_reg, val;
> int rc = 0;
> @@ -242,12 +277,15 @@ static int p8_adu_getmem(struct adu *adu,
> uint64_t addr, uint64_t *data, int ci)
> CHECK_ERR(adu_lock(adu));
>
> ctrl_reg = P8_TTYPE_TREAD;
> - if (ci)
> + if (ci) {
> /* Do cache inhibited access */
> ctrl_reg = SETFIELD(P8_FBC_ALTD_TTYPE, ctrl_reg,
> P8_TTYPE_CI_PARTIAL_READ);
> - else
> + block_size = (blog2(block_size) + 1) << 1;
> + } else {
> ctrl_reg = SETFIELD(P8_FBC_ALTD_TTYPE, ctrl_reg,
> P8_TTYPE_DMA_PARTIAL_READ);
> - ctrl_reg = SETFIELD(P8_FBC_ALTD_TSIZE, ctrl_reg, 8);
> + block_size = 0;
> + }
> + ctrl_reg = SETFIELD(P8_FBC_ALTD_TSIZE, ctrl_reg, block_size);
>
> CHECK_ERR_GOTO(out, rc = pib_read(&adu->target,
> P8_ALTD_CMD_REG, &cmd_reg));
> cmd_reg |= FBC_ALTD_START_OP;
> @@ -292,19 +330,23 @@ out:
>
> }
>
> -int p8_adu_putmem(struct adu *adu, uint64_t addr, uint64_t data, int
> size, int ci)
> +int p8_adu_putmem(struct adu *adu, uint64_t addr, uint64_t data, int
> size,
> + int ci, uint8_t block_size)
> {
> int rc = 0;
> uint64_t cmd_reg, ctrl_reg, val;
> CHECK_ERR(adu_lock(adu));
>
> ctrl_reg = P8_TTYPE_TWRITE;
> - if (ci)
> + if (ci) {
> /* Do cache inhibited access */
> ctrl_reg = SETFIELD(P8_FBC_ALTD_TTYPE, ctrl_reg,
> P8_TTYPE_CI_PARTIAL_WRITE);
> - else
> + block_size = (blog2(block_size) + 1) << 1;
> + } else {
> ctrl_reg = SETFIELD(P8_FBC_ALTD_TTYPE, ctrl_reg,
> P8_TTYPE_DMA_PARTIAL_WRITE);
> - ctrl_reg = SETFIELD(P8_FBC_ALTD_TSIZE, ctrl_reg, size);
> + block_size <<= 1;
> + }
> + ctrl_reg = SETFIELD(P8_FBC_ALTD_TSIZE, ctrl_reg, block_size);
>
> CHECK_ERR_GOTO(out, rc = pib_read(&adu->target,
> P8_ALTD_CMD_REG, &cmd_reg));
> cmd_reg |= FBC_ALTD_START_OP;
> @@ -349,19 +391,25 @@ out:
> return rc;
> }
>
> -static int p9_adu_getmem(struct adu *adu, uint64_t addr, uint64_t
> *data, int ci)
> +static int p9_adu_getmem(struct adu *adu, uint64_t addr, uint64_t
> *data,
> + int ci, uint8_t block_size)
> {
> uint64_t ctrl_reg, cmd_reg, val;
>
> cmd_reg = P9_TTYPE_TREAD;
> - if (ci)
> + if (ci) {
> /* Do cache inhibited access */
> cmd_reg = SETFIELD(P9_FBC_ALTD_TTYPE, cmd_reg,
> P9_TTYPE_CI_PARTIAL_READ);
> - else
> + block_size = (blog2(block_size) + 1) << 1;
> + } else {
> cmd_reg = SETFIELD(P9_FBC_ALTD_TTYPE, cmd_reg,
> P9_TTYPE_DMA_PARTIAL_READ);
>
> - /* For a read size is apparently always 0 */
> - cmd_reg = SETFIELD(P9_FBC_ALTD_TSIZE, cmd_reg, 0);
> + /* For normal reads the size is ignored as HW always
> + * returns a cache line */
> + block_size = 0;
> + }
> +
> + cmd_reg = SETFIELD(P9_FBC_ALTD_TSIZE, cmd_reg, block_size);
> cmd_reg |= FBC_ALTD_START_OP;
> cmd_reg = SETFIELD(FBC_ALTD_SCOPE, cmd_reg, SCOPE_REMOTE);
> cmd_reg = SETFIELD(FBC_ALTD_DROP_PRIORITY, cmd_reg,
> DROP_PRIORITY_LOW);
> @@ -400,7 +448,8 @@ retry:
> return 0;
> }
>
> -static int p9_adu_putmem(struct adu *adu, uint64_t addr, uint64_t
> data, int size, int ci)
> +static int p9_adu_putmem(struct adu *adu, uint64_t addr, uint64_t
> data, int size,
> + int ci, uint8_t block_size)
> {
> uint64_t ctrl_reg, cmd_reg, val;
>
> @@ -409,12 +458,15 @@ static int p9_adu_putmem(struct adu *adu,
> uint64_t addr, uint64_t data, int size
> size <<= 1;
>
> cmd_reg = P9_TTYPE_TWRITE;
> - if (ci)
> + if (ci) {
> /* Do cache inhibited access */
> cmd_reg = SETFIELD(P9_FBC_ALTD_TTYPE, cmd_reg,
> P9_TTYPE_CI_PARTIAL_WRITE);
> - else
> + block_size = (blog2(block_size) + 1) << 1;
> + } else {
> cmd_reg = SETFIELD(P9_FBC_ALTD_TTYPE, cmd_reg,
> P9_TTYPE_DMA_PARTIAL_WRITE);
> - cmd_reg = SETFIELD(P9_FBC_ALTD_TSIZE, cmd_reg, size);
> + block_size <<= 1;
> + }
> + cmd_reg = SETFIELD(P9_FBC_ALTD_TSIZE, cmd_reg, block_size);
> cmd_reg |= FBC_ALTD_START_OP;
> cmd_reg = SETFIELD(FBC_ALTD_SCOPE, cmd_reg, SCOPE_REMOTE);
> cmd_reg = SETFIELD(FBC_ALTD_DROP_PRIORITY, cmd_reg,
> DROP_PRIORITY_LOW);
> diff --git a/libpdbg/libpdbg.h b/libpdbg/libpdbg.h
> index 301c2c8..9ac8804 100644
> --- a/libpdbg/libpdbg.h
> +++ b/libpdbg/libpdbg.h
> @@ -204,12 +204,18 @@ int htm_status(struct pdbg_target *target);
> int htm_dump(struct pdbg_target *target, char *filename);
> int htm_record(struct pdbg_target *target, char *filename);
>
> -int adu_getmem(struct pdbg_target *target, uint64_t addr, uint8_t
> *ouput, uint64_t size);
> -int adu_putmem(struct pdbg_target *target, uint64_t addr, uint8_t
> *input, uint64_t size);
> -int adu_getmem_ci(struct pdbg_target *target, uint64_t addr, uint8_t
> *ouput, uint64_t size);
> -int adu_putmem_ci(struct pdbg_target *target, uint64_t addr, uint8_t
> *input, uint64_t size);
> -int __adu_getmem(struct pdbg_target *target, uint64_t addr, uint8_t
> *ouput, uint64_t size, bool ci);
> -int __adu_putmem(struct pdbg_target *target, uint64_t addr, uint8_t
> *input, uint64_t size, bool ci);
> +int adu_getmem(struct pdbg_target *target, uint64_t addr,
> + uint8_t *ouput, uint64_t size, uint8_t block_size);
> +int adu_putmem(struct pdbg_target *target, uint64_t addr,
> + uint8_t *input, uint64_t size, uint8_t block_size);
> +int adu_getmem_ci(struct pdbg_target *target, uint64_t addr,
> + uint8_t *ouput, uint64_t size, uint8_t block_size);
> +int adu_putmem_ci(struct pdbg_target *target, uint64_t addr,
> + uint8_t *input, uint64_t size, uint8_t block_size);
> +int __adu_getmem(struct pdbg_target *target, uint64_t addr, uint8_t
> *ouput,
> + uint64_t size, bool ci, uint8_t block_size);
> +int __adu_putmem(struct pdbg_target *target, uint64_t addr, uint8_t
> *input,
> + uint64_t size, bool ci, uint8_t block_size);
>
> int opb_read(struct pdbg_target *target, uint32_t addr, uint32_t
> *data);
> int opb_write(struct pdbg_target *target, uint32_t addr, uint32_t
> data);
> diff --git a/libpdbg/target.h b/libpdbg/target.h
> index 7cc855d..16ae304 100644
> --- a/libpdbg/target.h
> +++ b/libpdbg/target.h
> @@ -107,8 +107,8 @@ struct htm {
>
> struct adu {
> struct pdbg_target target;
> - int (*getmem)(struct adu *, uint64_t, uint64_t *, int);
> - int (*putmem)(struct adu *, uint64_t, uint64_t, int, int);
> + int (*getmem)(struct adu *, uint64_t, uint64_t *, int,
> uint8_t);
> + int (*putmem)(struct adu *, uint64_t, uint64_t, int, int,
> uint8_t);
> };
> #define target_to_adu(x) container_of(x, struct adu, target)
>
> diff --git a/src/mem.c b/src/mem.c
> index ce099c2..b578f95 100644
> --- a/src/mem.c
> +++ b/src/mem.c
> @@ -39,8 +39,10 @@ struct mem_flags {
> };
>
> #define MEM_CI_FLAG ("--ci", ci, parse_flag_noarg, false)
> +#define BLOCK_SIZE (parse_number8_pow2, "8")
>
> -static int getmem(uint64_t addr, uint64_t size, struct mem_flags
> flags)
> +static int getmem(uint64_t addr, uint64_t size,
> + uint8_t block_size, struct mem_flags flags)
> {
> struct pdbg_target *target;
> uint8_t *buf;
> @@ -59,7 +61,7 @@ static int getmem(uint64_t addr, uint64_t size,
> struct mem_flags flags)
>
> pdbg_set_progress_tick(progress_tick);
> progress_init();
> - if (!__adu_getmem(target, addr, buf, size, flags.ci)) {
> + if (!__adu_getmem(target, addr, buf, size, flags.ci,
> block_size)) {
> if (write(STDOUT_FILENO, buf, size) < 0)
> PR_ERROR("Unable to write stdout.\n");
> else
> @@ -74,10 +76,10 @@ static int getmem(uint64_t addr, uint64_t size,
> struct mem_flags flags)
> return rc;
>
> }
> -OPTCMD_DEFINE_CMD_WITH_FLAGS(getmem, getmem, (ADDRESS, DATA),
> +OPTCMD_DEFINE_CMD_WITH_FLAGS(getmem, getmem, (ADDRESS, DATA,
> BLOCK_SIZE),
> mem_flags, (MEM_CI_FLAG));
>
> -static int putmem(uint64_t addr, struct mem_flags flags)
> +static int putmem(uint64_t addr, uint8_t block_size, struct
> mem_flags flags)
> {
> uint8_t *buf;
> int read_size, rc = 0;
> @@ -98,7 +100,7 @@ static int putmem(uint64_t addr, struct mem_flags
> flags)
> if (read_size <= 0)
> break;
>
> - if (__adu_putmem(adu_target, addr, buf, read_size,
> flags.ci)) {
> + if (__adu_putmem(adu_target, addr, buf, read_size,
> flags.ci, block_size)) {
> rc = 0;
> printf("Unable to write memory.\n");
> break;
> @@ -111,5 +113,5 @@ static int putmem(uint64_t addr, struct mem_flags
> flags)
> free(buf);
> return rc;
> }
> -OPTCMD_DEFINE_CMD_WITH_FLAGS(putmem, putmem, (ADDRESS),
> +OPTCMD_DEFINE_CMD_WITH_FLAGS(putmem, putmem, (ADDRESS, BLOCK_SIZE),
> mem_flags, (MEM_CI_FLAG));
> diff --git a/src/pdbgproxy.c b/src/pdbgproxy.c
> index dedea7a..877cd4c 100644
> --- a/src/pdbgproxy.c
> +++ b/src/pdbgproxy.c
> @@ -223,7 +223,7 @@ static void get_mem(uint64_t *stack, void *priv)
>
> linear_map = get_real_addr(addr);
> if (linear_map != -1UL) {
> - if (adu_getmem(adu_target, linear_map, (uint8_t *)
> data, len)) {
> + if (adu_getmem(adu_target, linear_map, (uint8_t *) data, len,
> 8)) {
> PR_ERROR("Unable to read memory\n");
> err = 1;
> }
> @@ -293,7 +293,7 @@ static void put_mem(uint64_t *stack, void *priv)
>
> PR_INFO("put_mem 0x%016" PRIx64 " = 0x%016" PRIx64 "\n", addr,
> stack[2]);
>
> - if (adu_putmem(adu_target, addr, data, len)) {
> + if (adu_putmem(adu_target, addr, data, len, 8)) {
> PR_ERROR("Unable to write memory\n");
> err = 3;
> }
> diff --git a/src/thread.c b/src/thread.c
> index 1fd448d..2879c19 100644
> --- a/src/thread.c
> +++ b/src/thread.c
> @@ -35,7 +35,7 @@ static bool is_real_address(struct thread_regs
> *regs, uint64_t addr)
>
> static int load8(struct pdbg_target *target, uint64_t addr, uint64_t
> *value)
> {
> - if (adu_getmem(target, addr, (uint8_t *)value, 8)) {
> + if (adu_getmem(target, addr, (uint8_t *)value, 8, 8)) {
> pdbg_log(PDBG_ERROR, "Unable to read memory
> address=%016" PRIx64 ".\n", addr);
> return 0;
> }
> --
> 2.11.0
Amitay.
--
It is nice to be important but it is much more important to be NICE.
More information about the Pdbg
mailing list