2008-10-24 18:48:58

by Carl Love

[permalink] [raw]
Subject: [Cbe-oss-dev] [PATCH] Cell OProfile: Incorrect local array size in activate spu profiling function


The size of the pm_signal_local array should be equal to the
number of SPUs being configured in the call. Currently, the
array is of size 4 (NR_PHYS_CTRS) but being indexed by a for
loop from 0 to 7 (NUM_SPUS_PER_NODE).

Signed-off-by: Carl Love <[email protected]>


Index: Cell_kernel_10_24_2008/arch/powerpc/oprofile/op_model_cell.c
===================================================================
--- Cell_kernel_10_24_2008.orig/arch/powerpc/oprofile/op_model_cell.c
+++ Cell_kernel_10_24_2008/arch/powerpc/oprofile/op_model_cell.c
@@ -830,7 +830,7 @@ static int calculate_lfsr(int n)
static int pm_rtas_activate_spu_profiling(u32 node)
{
int ret, i;
- struct pm_signal pm_signal_local[NR_PHYS_CTRS];
+ struct pm_signal pm_signal_local[NUM_SPUS_PER_NODE];

/*
* Set up the rtas call to configure the debug bus to


2008-10-25 04:40:12

by Michael Ellerman

[permalink] [raw]
Subject: Re: [Cbe-oss-dev] [PATCH] Cell OProfile: Incorrect local array size in activate spu profiling function

On Fri, 2008-10-24 at 11:47 -0700, Carl Love wrote:
> The size of the pm_signal_local array should be equal to the
> number of SPUs being configured in the call. Currently, the
> array is of size 4 (NR_PHYS_CTRS) but being indexed by a for
> loop from 0 to 7 (NUM_SPUS_PER_NODE).

While you're at it you should change to using ARRAY_SIZE() in the for
loop, and use sizeof in the rtas call - that way you'll only have the
size in one place.

I notice pm_rtas_activate_signals() also allocates a pm_signal_local on
the stack and iterates over it. I take it something guarantees that
count in that routine will not exceed NR_PHYS_CTRS ?

cheers

--
Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person

Attachments:
signature.asc (189.00 B)
This is a digitally signed message part
Subject: Re: [Cbe-oss-dev] [PATCH] Cell OProfile: Incorrect local array size in activate spu profiling function

On 24.10.08 11:47:29, Carl Love wrote:
>
> The size of the pm_signal_local array should be equal to the
> number of SPUs being configured in the call. Currently, the
> array is of size 4 (NR_PHYS_CTRS) but being indexed by a for
> loop from 0 to 7 (NUM_SPUS_PER_NODE).
>
> Signed-off-by: Carl Love <[email protected]>

I applied your patch to oprofile/oprofile-for-tip. Thanks Carl.

-Robert

--
Advanced Micro Devices, Inc.
Operating System Research Center
email: [email protected]

2008-10-28 15:19:00

by Carl Love

[permalink] [raw]
Subject: [UDATED PATCH] Cell OProfile: Incorrect local array size in activate spu profiling function

Updated the patch to address comments by Michael Ellerman.
Specifically, changed upper limit in for loop to
ARRAY_SIZE() macro and added a check to make sure the
number of events specified by the user, which is used as
the max for indexing various arrays, is no bigger then the
declared size of the arrays.

The size of the pm_signal_local array should be equal to the
number of SPUs being configured in the array. Currently, the
array is of size 4 (NR_PHYS_CTRS) but being indexed by a for
loop from 0 to 7 (NUM_SPUS_PER_NODE).

Signed-off-by: Carl Love <[email protected]>


Index: Cell_kernel_10_24_2008/arch/powerpc/oprofile/op_model_cell.c
===================================================================
--- Cell_kernel_10_24_2008.orig/arch/powerpc/oprofile/op_model_cell.c
+++ Cell_kernel_10_24_2008/arch/powerpc/oprofile/op_model_cell.c
@@ -582,6 +582,13 @@ static int cell_reg_setup(struct op_coun

num_counters = num_ctrs;

+ if (unlikely(num_ctrs > NR_PHYS_CTRS)) {
+ printk(KERN_ERR
+ "%s: Oprofile, number of specified events " \
+ "exceeds number of physical counters\n",
+ __func__);
+ return -EIO;
+ }
pm_regs.group_control = 0;
pm_regs.debug_bus_control = 0;

@@ -830,13 +837,13 @@ static int calculate_lfsr(int n)
static int pm_rtas_activate_spu_profiling(u32 node)
{
int ret, i;
- struct pm_signal pm_signal_local[NR_PHYS_CTRS];
+ struct pm_signal pm_signal_local[NUM_SPUS_PER_NODE];

/*
* Set up the rtas call to configure the debug bus to
* route the SPU PCs. Setup the pm_signal for each SPU
*/
- for (i = 0; i < NUM_SPUS_PER_NODE; i++) {
+ for (i = 0; i < ARRAY_SIZE(pm_signal_local); i++) {
pm_signal_local[i].cpu = node;
pm_signal_local[i].signal_group = 41;
/* spu i on word (i/2) */

2008-10-29 00:52:42

by Michael Ellerman

[permalink] [raw]
Subject: Re: [UDATED PATCH] Cell OProfile: Incorrect local array size in activate spu profiling function

On Tue, 2008-10-28 at 08:17 -0700, Carl Love wrote:
> Updated the patch to address comments by Michael Ellerman.
> Specifically, changed upper limit in for loop to
> ARRAY_SIZE() macro and added a check to make sure the
> number of events specified by the user, which is used as
> the max for indexing various arrays, is no bigger then the
> declared size of the arrays.
>
> The size of the pm_signal_local array should be equal to the
> number of SPUs being configured in the array. Currently, the
> array is of size 4 (NR_PHYS_CTRS) but being indexed by a for
> loop from 0 to 7 (NUM_SPUS_PER_NODE).
>
> Signed-off-by: Carl Love <[email protected]>
>
>
> Index: Cell_kernel_10_24_2008/arch/powerpc/oprofile/op_model_cell.c
> ===================================================================
> --- Cell_kernel_10_24_2008.orig/arch/powerpc/oprofile/op_model_cell.c
> +++ Cell_kernel_10_24_2008/arch/powerpc/oprofile/op_model_cell.c
> @@ -582,6 +582,13 @@ static int cell_reg_setup(struct op_coun
>
> num_counters = num_ctrs;
>
> + if (unlikely(num_ctrs > NR_PHYS_CTRS)) {
> + printk(KERN_ERR
> + "%s: Oprofile, number of specified events " \
> + "exceeds number of physical counters\n",
> + __func__);
> + return -EIO;
> + }

I wouldn't bother with the unlikely, but whatever, looks good.

> pm_regs.group_control = 0;
> pm_regs.debug_bus_control = 0;
>
> @@ -830,13 +837,13 @@ static int calculate_lfsr(int n)
> static int pm_rtas_activate_spu_profiling(u32 node)
> {
> int ret, i;
> - struct pm_signal pm_signal_local[NR_PHYS_CTRS];
> + struct pm_signal pm_signal_local[NUM_SPUS_PER_NODE];
>
> /*
> * Set up the rtas call to configure the debug bus to
> * route the SPU PCs. Setup the pm_signal for each SPU
> */
> - for (i = 0; i < NUM_SPUS_PER_NODE; i++) {
> + for (i = 0; i < ARRAY_SIZE(pm_signal_local); i++) {
> pm_signal_local[i].cpu = node;
> pm_signal_local[i].signal_group = 41;
> /* spu i on word (i/2) */

Looks good, but ..

You should patch the call to rtas_ibm_cbe_perftools() as well, like:

ret = rtas_ibm_cbe_perftools(SUBFUNC_ACTIVATE,
PASSTHRU_ENABLE, pm_signal_local,
sizeof(pm_signal_local));


cheers

--
Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person

Attachments:
signature.asc (189.00 B)
This is a digitally signed message part

2008-10-29 15:07:43

by Carl Love

[permalink] [raw]
Subject: Re: [UPDATED PATCH VER2] Cell OProfile: Incorrect local array size in activate spu profiling function

Updated the patch to address comments by Michael Ellerman.
Specifically, changed upper limit in for loop to
ARRAY_SIZE() macro and added a check to make sure the
number of events specified by the user, which is used as
the max for indexing various arrays, is no bigger then the
declared size of the arrays.

The size of the pm_signal_local array should be equal to the
number of SPUs being configured in the array. Currently, the
array is of size 4 (NR_PHYS_CTRS) but being indexed by a for
loop from 0 to 7 (NUM_SPUS_PER_NODE).

Signed-off-by: Carl Love <[email protected]>


Index: Cell_kernel_10_24_2008/arch/powerpc/oprofile/op_model_cell.c
===================================================================
--- Cell_kernel_10_24_2008.orig/arch/powerpc/oprofile/op_model_cell.c
+++ Cell_kernel_10_24_2008/arch/powerpc/oprofile/op_model_cell.c
@@ -582,6 +582,13 @@ static int cell_reg_setup(struct op_coun

num_counters = num_ctrs;

+ if (unlikely(num_ctrs > NR_PHYS_CTRS)) {
+ printk(KERN_ERR
+ "%s: Oprofile, number of specified events " \
+ "exceeds number of physical counters\n",
+ __func__);
+ return -EIO;
+ }
pm_regs.group_control = 0;
pm_regs.debug_bus_control = 0;

@@ -830,13 +837,13 @@ static int calculate_lfsr(int n)
static int pm_rtas_activate_spu_profiling(u32 node)
{
int ret, i;
- struct pm_signal pm_signal_local[NR_PHYS_CTRS];
+ struct pm_signal pm_signal_local[NUM_SPUS_PER_NODE];

/*
* Set up the rtas call to configure the debug bus to
* route the SPU PCs. Setup the pm_signal for each SPU
*/
- for (i = 0; i < NUM_SPUS_PER_NODE; i++) {
+ for (i = 0; i < ARRAY_SIZE(pm_signal_local); i++) {
pm_signal_local[i].cpu = node;
pm_signal_local[i].signal_group = 41;
/* spu i on word (i/2) */
@@ -848,7 +855,7 @@ static int pm_rtas_activate_spu_profilin

ret = rtas_ibm_cbe_perftools(SUBFUNC_ACTIVATE,
PASSTHRU_ENABLE, pm_signal_local,
- (NUM_SPUS_PER_NODE
+ (ARRAY_SIZE(pm_signal_local)
* sizeof(struct pm_signal)));

if (unlikely(ret)) {

Subject: Re: [UPDATED PATCH VER2] Cell OProfile: Incorrect local array size in activate spu profiling function

On 29.10.08 08:06:45, Carl Love wrote:
> Updated the patch to address comments by Michael Ellerman.
> Specifically, changed upper limit in for loop to
> ARRAY_SIZE() macro and added a check to make sure the
> number of events specified by the user, which is used as
> the max for indexing various arrays, is no bigger then the
> declared size of the arrays.
>
> The size of the pm_signal_local array should be equal to the
> number of SPUs being configured in the array. Currently, the
> array is of size 4 (NR_PHYS_CTRS) but being indexed by a for
> loop from 0 to 7 (NUM_SPUS_PER_NODE).
>
> Signed-off-by: Carl Love <[email protected]>

I applied your patch to oprofile/oprofile-for-tip. Thanks Carl.

-Robert

--
Advanced Micro Devices, Inc.
Operating System Research Center
email: [email protected]