Skip to contentSkip to conversation

pinctrl: sunxi: Don't underestimate number of functions

PATCH In review Andre Przywara 3 participants 3 messages Jul 22, 2021 – Aug 10, 2021

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pinctrl: sunxi: Don't underestimate number of functions

When we are building all the various pinctrl structures for the Allwinner pinctrl devices, we do some estimation about the maximum number of distinct function (names) that we will need.

So far we take the number of pins as an upper bound, even though we can actually have up to four special functions per pin. This wasn't a problem until now, since we indeed have typically far more pins than functions, and most pins share common functions.

However the H616 "-r" pin controller has only two pins, but four functions, so we run over the end of the array when we are looking for a matching function name in sunxi_pinctrl_add_function - there is no NULL sentinel left that would terminate the loop:

[    8.200648] Unable to handle kernel paging request at virtual address fffdff7efbefaff5
[    8.209179] Mem abort info:
....
[    8.368456] Call trace:
[    8.370925]  __pi_strcmp+0x90/0xf0
[    8.374559]  sun50i_h616_r_pinctrl_probe+0x1c/0x28
[    8.379557]  platform_probe+0x68/0xd8

Do an actual worst case allocation (4 functions per pin, three common functions and the sentinel) for the initial array allocation. This is now heavily overestimating the number of functions in the common case, but we will reallocate this array later with the actual number of functions, so it's only temporarily.

Fixes 561c1cf17c46 ("pinctrl: sunxi: Add support for the Allwinner H616-R pin controller")
---
 drivers/pinctrl/sunxi/pinctrl-sunxi.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/pinctrl/sunxi/pinctrl-sunxi.c +5 -3
index dc8d39ae045b..9c7679c06dca 100644
@@ -1219,10 +1219,12 @@ static int sunxi_pinctrl_build_state(struct platform_device *pdev)
 	}
 
 	/*
-	 * We suppose that we won't have any more functions than pins,
-	 * we'll reallocate that later anyway
+	 * Find an upper bound for the maximum number of functions: in
+	 * the worst case we have gpio_in, gpio_out, irq and up to four
+	 * special functions per pin, plus one entry for the sentinel.
+	 * We'll reallocate that later anyway.
 	 */
-	pctl->functions = kcalloc(pctl->ngroups,
+	pctl->functions = kcalloc(4 * pctl->ngroups + 4,
 				  sizeof(*pctl->functions),
 				  GFP_KERNEL);
 	if (!pctl->functions)
Signature · 1 line
2.17.6
Acked-by on v1 1/1
#1 lore raw
Quoted 22 lines · Andre Przywara

When we are building all the various pinctrl structures for the Allwinner pinctrl devices, we do some estimation about the maximum number of distinct function (names) that we will need.

So far we take the number of pins as an upper bound, even though we can actually have up to four special functions per pin. This wasn't a problem until now, since we indeed have typically far more pins than functions, and most pins share common functions.

However the H616 "-r" pin controller has only two pins, but four functions, so we run over the end of the array when we are looking for a matching function name in sunxi_pinctrl_add_function - there is no NULL sentinel left that would terminate the loop:

[    8.200648] Unable to handle kernel paging request at virtual address fffdff7efbefaff5
[    8.209179] Mem abort info:
....
[    8.368456] Call trace:
[    8.370925]  __pi_strcmp+0x90/0xf0
[    8.374559]  sun50i_h616_r_pinctrl_probe+0x1c/0x28
[    8.379557]  platform_probe+0x68/0xd8
Quoted 8 lines

Do an actual worst case allocation (4 functions per pin, three common functions and the sentinel) for the initial array allocation. This is now heavily overestimating the number of functions in the common case, but we will reallocate this array later with the actual number of functions, so it's only temporarily.

Fixes 561c1cf17c46 ("pinctrl: sunxi: Add support for the Allwinner H616-R pin controller")

Thanks!
Maxime

#2 lore raw
In reply to Andre Przywara
Quoted 22 lines · Andre Przywara

When we are building all the various pinctrl structures for the Allwinner pinctrl devices, we do some estimation about the maximum number of distinct function (names) that we will need.

So far we take the number of pins as an upper bound, even though we can actually have up to four special functions per pin. This wasn't a problem until now, since we indeed have typically far more pins than functions, and most pins share common functions.

However the H616 "-r" pin controller has only two pins, but four functions, so we run over the end of the array when we are looking for a matching function name in sunxi_pinctrl_add_function - there is no NULL sentinel left that would terminate the loop:

[    8.200648] Unable to handle kernel paging request at virtual address fffdff7efbefaff5
[    8.209179] Mem abort info:
....
[    8.368456] Call trace:
[    8.370925]  __pi_strcmp+0x90/0xf0
[    8.374559]  sun50i_h616_r_pinctrl_probe+0x1c/0x28
[    8.379557]  platform_probe+0x68/0xd8
Quoted 8 lines

Do an actual worst case allocation (4 functions per pin, three common functions and the sentinel) for the initial array allocation. This is now heavily overestimating the number of functions in the common case, but we will reallocate this array later with the actual number of functions, so it's only temporarily.

Fixes 561c1cf17c46 ("pinctrl: sunxi: Add support for the Allwinner H616-R pin controller")

Patch applied for fixes!
Thanks for finding this.

Yours,
Linus Walleij