Re: [PATCH v2] cpufreq: powernv: fix stack bloat and NR_CPUS limitation
John Hubbard writes: > On 10/30/19 7:39 PM, Michael Ellerman wrote: >> Hi John, >> >> Sorry I didn't reply to this sooner, too many patches :/ >> >> John Hubbard writes: >>> The following build warning occurred on powerpc 64-bit builds: >>> >>> drivers/cpufreq/powernv-cpufreq.c: In function 'init_chip_info': >>> drivers/cpufreq/powernv-cpufreq.c:1070:1: warning: the frame size of 1040 >>> bytes is larger than 1024 bytes [-Wframe-larger-than=] >> >> Oddly I don't see that warning in my builds, eg with GCC9: >> >>https://travis-ci.org/linuxppc/linux/jobs/604870722 > > This is with a cross-compiler based on gcc 8.1.0, which I got from: >https://mirrors.edge.kernel.org/pub/tools/crosstool/files/bin/x86_64/8.1.0/ > > I'll put that in the v3 commit description. > >> >>> This is due to putting 1024 bytes on the stack: >>> >>> unsigned int chip[256]; >>> >>> ...and while looking at this, it also has a bug: it fails with a stack >>> overrun, if CONFIG_NR_CPUS > 256. >> >> It _probably_ doesn't, because it only increments the index when the >> chip_id of the CPU changes, ie. it doesn't create a chip for every CPU. >> But I agree it's flaky the way it's written. > > I'll soften up the wording accordingly. > >> >>> Fix both problems by dynamically allocating based on CONFIG_NR_CPUS. >> >> Shouldn't it use num_possible_cpus() ? >> >> Given the for loop is over possible CPUs that seems like the upper >> bound. In practice it should be lower because some CPUs will share a >> chip. >> > > OK, I see, that's more consistent with the code, I'll change to that. Thanks. cheers
Re: [PATCH v2] cpufreq: powernv: fix stack bloat and NR_CPUS limitation
On 10/30/19 7:39 PM, Michael Ellerman wrote:
Hi John,
Sorry I didn't reply to this sooner, too many patches :/
John Hubbard writes:
The following build warning occurred on powerpc 64-bit builds:
drivers/cpufreq/powernv-cpufreq.c: In function 'init_chip_info':
drivers/cpufreq/powernv-cpufreq.c:1070:1: warning: the frame size of 1040 bytes
is larger than 1024 bytes [-Wframe-larger-than=]
Oddly I don't see that warning in my builds, eg with GCC9:
https://travis-ci.org/linuxppc/linux/jobs/604870722
This is with a cross-compiler based on gcc 8.1.0, which I got from:
https://mirrors.edge.kernel.org/pub/tools/crosstool/files/bin/x86_64/8.1.0/
I'll put that in the v3 commit description.
This is due to putting 1024 bytes on the stack:
unsigned int chip[256];
...and while looking at this, it also has a bug: it fails with a stack
overrun, if CONFIG_NR_CPUS > 256.
It _probably_ doesn't, because it only increments the index when the
chip_id of the CPU changes, ie. it doesn't create a chip for every CPU.
But I agree it's flaky the way it's written.
I'll soften up the wording accordingly.
Fix both problems by dynamically allocating based on CONFIG_NR_CPUS.
Shouldn't it use num_possible_cpus() ?
Given the for loop is over possible CPUs that seems like the upper
bound. In practice it should be lower because some CPUs will share a
chip.
OK, I see, that's more consistent with the code, I'll change to that.
thanks,
--
John Hubbard
NVIDIA
Fixes: 053819e0bf840 ("cpufreq: powernv: Handle throttling due to Pmax capping at
chip level")
Cc: Shilpasri G Bhat
Cc: Preeti U Murthy
Cc: Viresh Kumar
Cc: Rafael J. Wysocki
Cc: linux...@vger.kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: John Hubbard
---
Changes since v1: includes Viresh's review commit fixes.
drivers/cpufreq/powernv-cpufreq.c | 17 +
1 file changed, 13 insertions(+), 4 deletions(-)
diff --git a/drivers/cpufreq/powernv-cpufreq.c
b/drivers/cpufreq/powernv-cpufreq.c
index 6061850e59c9..5b2e968cb5ea 100644
--- a/drivers/cpufreq/powernv-cpufreq.c
+++ b/drivers/cpufreq/powernv-cpufreq.c
@@ -1041,9 +1041,14 @@ static struct cpufreq_driver powernv_cpufreq_driver = {
static int init_chip_info(void)
{
- unsigned int chip[256];
+ unsigned int *chip;
unsigned int cpu, i;
unsigned int prev_chip_id = UINT_MAX;
+ int ret = 0;
+
+ chip = kcalloc(CONFIG_NR_CPUS, sizeof(*chip), GFP_KERNEL);
+ if (!chip)
+ return -ENOMEM;
for_each_possible_cpu(cpu) {
unsigned int id = cpu_to_chip_id(cpu);
@@ -1055,8 +1060,10 @@ static int init_chip_info(void)
}
chips = kcalloc(nr_chips, sizeof(struct chip), GFP_KERNEL);
- if (!chips)
- return -ENOMEM;
+ if (!chips) {
+ ret = -ENOMEM;
+ goto free_and_return;
+ }
for (i = 0; i < nr_chips; i++) {
chips[i].id = chip[i];
@@ -1066,7 +1073,9 @@ static int init_chip_info(void)
per_cpu(chip_info, cpu) = [i];
}
- return 0;
+free_and_return:
+ kfree(chip);
+ return ret;
}
static inline void clean_chip_info(void)
--
2.23.0
Re: [PATCH v2] cpufreq: powernv: fix stack bloat and NR_CPUS limitation
Hi John,
Sorry I didn't reply to this sooner, too many patches :/
John Hubbard writes:
> The following build warning occurred on powerpc 64-bit builds:
>
> drivers/cpufreq/powernv-cpufreq.c: In function 'init_chip_info':
> drivers/cpufreq/powernv-cpufreq.c:1070:1: warning: the frame size of 1040
> bytes is larger than 1024 bytes [-Wframe-larger-than=]
Oddly I don't see that warning in my builds, eg with GCC9:
https://travis-ci.org/linuxppc/linux/jobs/604870722
> This is due to putting 1024 bytes on the stack:
>
> unsigned int chip[256];
>
> ...and while looking at this, it also has a bug: it fails with a stack
> overrun, if CONFIG_NR_CPUS > 256.
It _probably_ doesn't, because it only increments the index when the
chip_id of the CPU changes, ie. it doesn't create a chip for every CPU.
But I agree it's flaky the way it's written.
> Fix both problems by dynamically allocating based on CONFIG_NR_CPUS.
Shouldn't it use num_possible_cpus() ?
Given the for loop is over possible CPUs that seems like the upper
bound. In practice it should be lower because some CPUs will share a
chip.
cheers
> Fixes: 053819e0bf840 ("cpufreq: powernv: Handle throttling due to Pmax
> capping at chip level")
> Cc: Shilpasri G Bhat
> Cc: Preeti U Murthy
> Cc: Viresh Kumar
> Cc: Rafael J. Wysocki
> Cc: linux...@vger.kernel.org
> Cc: linuxppc-dev@lists.ozlabs.org
> Signed-off-by: John Hubbard
> ---
>
> Changes since v1: includes Viresh's review commit fixes.
>
> drivers/cpufreq/powernv-cpufreq.c | 17 +
> 1 file changed, 13 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/cpufreq/powernv-cpufreq.c
> b/drivers/cpufreq/powernv-cpufreq.c
> index 6061850e59c9..5b2e968cb5ea 100644
> --- a/drivers/cpufreq/powernv-cpufreq.c
> +++ b/drivers/cpufreq/powernv-cpufreq.c
> @@ -1041,9 +1041,14 @@ static struct cpufreq_driver powernv_cpufreq_driver = {
>
> static int init_chip_info(void)
> {
> - unsigned int chip[256];
> + unsigned int *chip;
> unsigned int cpu, i;
> unsigned int prev_chip_id = UINT_MAX;
> + int ret = 0;
> +
> + chip = kcalloc(CONFIG_NR_CPUS, sizeof(*chip), GFP_KERNEL);
> + if (!chip)
> + return -ENOMEM;
>
> for_each_possible_cpu(cpu) {
> unsigned int id = cpu_to_chip_id(cpu);
> @@ -1055,8 +1060,10 @@ static int init_chip_info(void)
> }
>
> chips = kcalloc(nr_chips, sizeof(struct chip), GFP_KERNEL);
> - if (!chips)
> - return -ENOMEM;
> + if (!chips) {
> + ret = -ENOMEM;
> + goto free_and_return;
> + }
>
> for (i = 0; i < nr_chips; i++) {
> chips[i].id = chip[i];
> @@ -1066,7 +1073,9 @@ static int init_chip_info(void)
> per_cpu(chip_info, cpu) = [i];
> }
>
> - return 0;
> +free_and_return:
> + kfree(chip);
> + return ret;
> }
>
> static inline void clean_chip_info(void)
> --
> 2.23.0
Re: [PATCH v2] cpufreq: powernv: fix stack bloat and NR_CPUS limitation
On Friday, October 18, 2019 7:07:12 AM CET Viresh Kumar wrote:
> On 17-10-19, 21:55, John Hubbard wrote:
> > The following build warning occurred on powerpc 64-bit builds:
> >
> > drivers/cpufreq/powernv-cpufreq.c: In function 'init_chip_info':
> > drivers/cpufreq/powernv-cpufreq.c:1070:1: warning: the frame size of 1040
> > bytes is larger than 1024 bytes [-Wframe-larger-than=]
> >
> > This is due to putting 1024 bytes on the stack:
> >
> > unsigned int chip[256];
> >
> > ...and while looking at this, it also has a bug: it fails with a stack
> > overrun, if CONFIG_NR_CPUS > 256.
> >
> > Fix both problems by dynamically allocating based on CONFIG_NR_CPUS.
> >
> > Fixes: 053819e0bf840 ("cpufreq: powernv: Handle throttling due to Pmax
> > capping at chip level")
> > Cc: Shilpasri G Bhat
> > Cc: Preeti U Murthy
> > Cc: Viresh Kumar
> > Cc: Rafael J. Wysocki
> > Cc: linux...@vger.kernel.org
> > Cc: linuxppc-dev@lists.ozlabs.org
> > Signed-off-by: John Hubbard
> > ---
> >
> > Changes since v1: includes Viresh's review commit fixes.
> >
> > drivers/cpufreq/powernv-cpufreq.c | 17 +
> > 1 file changed, 13 insertions(+), 4 deletions(-)
>
> Acked-by: Viresh Kumar
>
>
Applying as 5.5 material, thanks!
Re: [PATCH v2] cpufreq: powernv: fix stack bloat and NR_CPUS limitation
On 17-10-19, 21:55, John Hubbard wrote:
> The following build warning occurred on powerpc 64-bit builds:
>
> drivers/cpufreq/powernv-cpufreq.c: In function 'init_chip_info':
> drivers/cpufreq/powernv-cpufreq.c:1070:1: warning: the frame size of 1040
> bytes is larger than 1024 bytes [-Wframe-larger-than=]
>
> This is due to putting 1024 bytes on the stack:
>
> unsigned int chip[256];
>
> ...and while looking at this, it also has a bug: it fails with a stack
> overrun, if CONFIG_NR_CPUS > 256.
>
> Fix both problems by dynamically allocating based on CONFIG_NR_CPUS.
>
> Fixes: 053819e0bf840 ("cpufreq: powernv: Handle throttling due to Pmax
> capping at chip level")
> Cc: Shilpasri G Bhat
> Cc: Preeti U Murthy
> Cc: Viresh Kumar
> Cc: Rafael J. Wysocki
> Cc: linux...@vger.kernel.org
> Cc: linuxppc-dev@lists.ozlabs.org
> Signed-off-by: John Hubbard
> ---
>
> Changes since v1: includes Viresh's review commit fixes.
>
> drivers/cpufreq/powernv-cpufreq.c | 17 +
> 1 file changed, 13 insertions(+), 4 deletions(-)
Acked-by: Viresh Kumar
--
viresh
