pwm: Fix division by 0 issues
Subsystems (3)
cover pwm: Fix division by 0 issues
Hello,
triggered by Josh Poimboeuf's patch to fix a warning by objtool due to a division by zero in the pwm-mediatek driver, I reviewed all pwm drivers for similar issues.
Josh's patch is resent here, the two additional patches are new. objtool didn't warn about these even though these are quite similar divisions by zero. I didn't debug why these were not catched by clang/objtool.
I haven't made up my mind yet if I'll send them for inclusion in 6.15 given that they are new and only build-tested so far. Also the commits that introduced these problems are already old (v4.12-rc1, v4.4-rc1 and v5.3-rc1 respectively) and there is no indication this problem triggers on real hardware.
I'll put them in my for-next branch anyhow for wider exposure. I happily accept test reports to (hopefully) increase my confidence in these patches.
Best regards
Uwe
Josh Poimboeuf (1):
pwm: mediatek: Prevent divide-by-zero in pwm_mediatek_config()
Uwe Kleine-König (2):
pwm: rcar: Improve register calculation
pwm: fsl-ftm: Handle clk_get_rate() returning 0
drivers/pwm/pwm-fsl-ftm.c | 6 ++++++ drivers/pwm/pwm-mediatek.c | 8 ++++++-- drivers/pwm/pwm-rcar.c | 23 ++++++++++++----------- 3 files changed, 24 insertions(+), 13 deletions(-)
base-commit: 6df320abbb40654085d7258de33d78481e93ac8d
Signature · 1 line
2.47.2
PATCH 1/3 pwm: mediatek: Prevent divide-by-zero in pwm_mediatek_config()
From: Josh Poimboeuf <[email protected]>
With CONFIG_COMPILE_TEST && !CONFIG_HAVE_CLK, pwm_mediatek_config() has a divide-by-zero in the following line:
do_div(resolution, clk_get_rate(pc->clk_pwms[pwm->hwpwm]));
due to the fact that the !CONFIG_HAVE_CLK version of clk_get_rate() returns zero.
This is presumably just a theoretical problem: COMPILE_TEST overrides the dependency on RALINK which would select COMMON_CLK. Regardless it's a good idea to check for the error explicitly to avoid divide-by-zero.
Fixes the following warning:
drivers/pwm/pwm-mediatek.o: warning: objtool: .text: unexpected end of section
--- drivers/pwm/pwm-mediatek.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/pwm/pwm-mediatek.c +6 -2
@@ -121,21 +121,25 @@ static int pwm_mediatek_config(struct pwm_chip *chip, struct pwm_device *pwm,struct pwm_mediatek_chip *pc =to_pwm_mediatek_chip (chip);u32 clkdiv =0 , cnt_period, cnt_duty, reg_width =PWMDWIDTH , reg_thres =PWMTHRES ; +unsigned long clk_rate;u64 resolution;int ret; ret =pwm_mediatek_clk_enable (chip, pwm); -if (ret <0 )return ret; + clk_rate =clk_get_rate (pc->clk_pwms[pwm->hwpwm]); +if (!clk_rate) +return -EINVAL ; +/* Make sure we use the bus clock and not the 26MHz clock */ if (pc->soc->has_ck_26m_sel)writel (0 , pc->regs +PWM_CK_26M_SEL );/* Using resolution in picosecond gets accuracy higher */ resolution = (u64 )NSEC_PER_SEC *1000 ; -do_div (resolution,clk_get_rate (pc->clk_pwms[pwm->hwpwm])); +do_div (resolution, clk_rate); cnt_period =DIV_ROUND_CLOSEST_ULL ((u64 )period_ns *1000 , resolution);while (cnt_period >8191 ) {
Signature · 1 line
2.47.2
PATCH 2/3 pwm: rcar: Improve register calculation
In reply to Uwe Kleine-KönigThere were several issues in the function rcar_pwm_set_counter():
- The u64 values period_ns and duty_ns were cast to int on function call which might loose bits on 32 bit architectures.
Fix: Make parameters to rcar_pwm_set_counter() u64
- The algorithm divided by the result of a division which looses precision.
Fix: Make use of mul_u64_u64_div_u64()
- The calculated values were just masked to fit the respective register fields which again might loose bits.
Fix: Explicitly check for overlow
Implement the respective fixes.
--- drivers/pwm/pwm-rcar.c | 23 ++++++++++++----------- 1 file changed, 12 insertions(+), 11 deletions(-)
drivers/pwm/pwm-rcar.c +12 -11
@@ -102,23 +102,24 @@ static void rcar_pwm_set_clock_control(struct rcar_pwm_chip *rp,rcar_pwm_write (rp, value,RCAR_PWMCR ); } -static int rcar_pwm_set_counter (struct rcar_pwm_chip *rp,int div,int duty_ns, -int period_ns) +static int rcar_pwm_set_counter (struct rcar_pwm_chip *rp,int div,u64 duty_ns, +u64 period_ns) { -unsigned long long one_cycle, tmp;/* 0.01 nanoseconds */ +unsigned long long tmp;unsigned long clk_rate =clk_get_rate (rp->clk);u32 cyc, ph; - one_cycle =NSEC_PER_SEC *100ULL << div; -do_div (one_cycle, clk_rate); +/* div <= 24 == RCAR_PWM_MAX_DIVISION, so the shift doesn't overflow. */ + tmp =mul_u64_u64_div_u64 (period_ns, clk_rate, (u64 )NSEC_PER_SEC << div); +if (tmp >FIELD_MAX (RCAR_PWMCNT_CYC0_MASK )) + tmp =FIELD_MAX (RCAR_PWMCNT_CYC0_MASK ); - tmp = period_ns *100ULL ; -do_div (tmp, one_cycle); - cyc = (tmp <<RCAR_PWMCNT_CYC0_SHIFT ) &RCAR_PWMCNT_CYC0_MASK ; + cyc =FIELD_PREP (RCAR_PWMCNT_CYC0_MASK , tmp); - tmp = duty_ns *100ULL ; -do_div (tmp, one_cycle); - ph = tmp &RCAR_PWMCNT_PH0_MASK ; + tmp =mul_u64_u64_div_u64 (duty_ns, clk_rate, (u64 )NSEC_PER_SEC << div); +if (tmp >FIELD_MAX (RCAR_PWMCNT_PH0_MASK )) + tmp =FIELD_MAX (RCAR_PWMCNT_PH0_MASK ); + ph =FIELD_PREP (RCAR_PWMCNT_PH0_MASK , tmp);/* Avoid prohibited setting */ if (cyc ==0 || ph ==0 )
Signature · 1 line
2.47.2
Re: [PATCH 2/3] pwm: rcar: Improve register calculation
Quoted 12 lines · Uwe Kleine-König
There were several issues in the function
rcar_pwm_set_counter():- The u64 values
period_nsandduty_nswere cast to int on function call which might loose bits on 32 bit architectures.
Fix: Make parameters torcar_pwm_set_counter()u64
- The algorithm divided by the result of a division which looses precision.
Fix: Make use ofmul_u64_u64_div_u64()
- The calculated values were just masked to fit the respective register fields which again might loose bits.
Fix: Explicitly check for overlow
Quoted 4 lines
Implement the respective fixes.
Fixes ed6c1476bf7f ("pwm: Add support for R-Car PWM Timer")Signed-off-by Uwe Kleine-König <[email protected]>
This is now commit b41a4921d409c909 ("pwm: rcar: Improve register calculation"), and has the missing include.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Re: [PATCH 2/3] pwm: rcar: Improve register calculation
Hello Geert,
Quoted 15 lines · Geert Uytterhoeven
Quoted 14 lines · Uwe Kleine-König
There were several issues in the function
rcar_pwm_set_counter():- The u64 values
period_nsandduty_nswere cast to int on function call which might loose bits on 32 bit architectures.
Fix: Make parameters torcar_pwm_set_counter()u64
- The algorithm divided by the result of a division which looses precision.
Fix: Make use ofmul_u64_u64_div_u64()
- The calculated values were just masked to fit the respective register fields which again might loose bits.
Fix: Explicitly check for overlowImplement the respective fixes.
Quoted 7 lines
Quoted 2 lines
Fixes ed6c1476bf7f ("pwm: Add support for R-Car PWM Timer")Signed-off-by Uwe Kleine-König <[email protected]>This is now commit
b41a4921d409c909("pwm: rcar: Improve register calculation"), and has the missing include.Reviewed-by Geert Uytterhoeven <[email protected]>
I added your tag, the commit is now e7327c193014.
Thanks!
Uwe
PATCH 3/3 pwm: fsl-ftm: Handle clk_get_rate() returning 0
In reply to Uwe Kleine-KönigConsidering that the driver doesn't enable the used clocks (and also that clk_get_rate() returns 0 if CONFIG_HAVE_CLK is unset) better check the return value of clk_get_rate() for being non-zero before dividing by it.
--- drivers/pwm/pwm-fsl-ftm.c | 6 ++++++ 1 file changed, 6 insertions(+)
drivers/pwm/pwm-fsl-ftm.c +6 -0
@@ -118,6 +118,9 @@ static unsigned int fsl_pwm_ticks_to_ns(struct fsl_pwm_chip *fpc,unsigned long long exval; rate =clk_get_rate (fpc->clk[fpc->period.clk_select]); +if (rate >> fpc->period.clk_ps ==0 ) +return 0 ; + exval = ticks; exval *=1000000000UL ;do_div (exval, rate >> fpc->period.clk_ps);
@@ -190,6 +193,9 @@ static unsigned int fsl_pwm_calculate_duty(struct fsl_pwm_chip *fpc,unsigned int period = fpc->period.mod_period +1 ;unsigned int period_ns =fsl_pwm_ticks_to_ns (fpc, period); +if (!period_ns) +return 0 ; + duty = (unsigned long long )duty_ns * period;do_div (duty, period_ns);
Signature · 1 line
2.47.2
Hello,
Quoted 3 lines · Uwe Kleine-König
I'll put them in my for-next branch anyhow for wider exposure. I happily accept test reports to (hopefully) increase my confidence in these patches.
FTR: I did that. The 0day build bot already found a problem with it in the rcar patch that is already fixed.
Best regards
Uwe