Project

General

Profile

Actions
Copy link

Bug #1839

open

Ldv-manager can't find bug in driver

Added by Ilja Zakharov about 13 years ago. Updated about 13 years ago.

Status:
Open
Priority:
Normal
Assignee:
Ilja Zakharov
Category:
Environment Generation
Start date:
09/28/2011
Due date:
% Done:

0%

Estimated time:
Detected in build:
svn
Platform:
Linux x86
Published in build:

Description

In driver gpio/pca953x.c was added artificial bug on 32 rule (lock-unlock), but result of cheking was safe.

Checked various changes, one of them:

static void pca953x_irq_bus_sync_unlock(struct irq_data *d) {
struct pca953x_chip chip = irq_data_get_irq_chip_data(d);
uint16_t new_irqs;
uint16_t level;
mutex_lock(&chip->irq_lock);
/ Look for any newly setup interrupt */
new_irqs = chip->irq_trig_fall | chip->irq_trig_raise;
new_irqs &= ~chip->reg_direction;

while (new_irqs) {
level = __ffs(new_irqs);
pca953x_gpio_direction_input(&chip->gpio_chip, level);
new_irqs &= ~(1 << level);
}

// mutex_unlock(&chip->irq_lock);
}

Files

pca953x.tar.bz2 (3.83 KB) pca953x.tar.bz2 Ilja Zakharov, 10/04/2011 02:05 PM
Actions
Copy link
 #1

Updated by Pavel Shved about 13 years ago

  • Status changed from New to Feedback

What function did you make the change to?

Will it compile with the default main() generator?

What happens if you add the bug in the first function your generated main invokes?

Actions
Copy link
 #2

Updated by Ilja Zakharov about 13 years ago

At the moment I wrote patterns for paired functions in all two structures in driver, it seems that problem of incorrect determination of artificial bugs is in wrong modeling of drivers with two or more structures. This assumption is confirmed by addition tests and false-positives in "clean" driver without any changes that appeared while i tried to run driver separate from kernel.

Actions
Copy link
 #3

Updated by Pavel Shved about 13 years ago

If you have a separate driver, please, attach it, write the command line how you invoke the tools, the result you've got, and the result you expected, so that others could reproduce it and fix it if applicable.

(This applies not to this bug only).

Actions
Copy link
 #4

Updated by Ilja Zakharov about 13 years ago

i run as usual:

ldv-manager "LDV_DEBUG=100" "envs=linux-2.6.39.tar.bz2" "drivers=pca953x.tar.bz2" "name=x1" "rule_models=32_7 39_7 68_1 08_1"

I got false-positive result "unsafe" on 32_7 rule, this false-positive connected with incorrect order of using functions from structures irq_chip and i2c_driver. This bug can't be fixed by configuration only as i understand, because of absence of opportunity in configuration file to control order of functions from different structures together.

Actions
Copy link
 #5

Updated by Ilja Zakharov about 13 years ago

Simple scheme of using strucures and their functions in this driver:

init ---> struct 12c_driver:
--> calling probe ---> struct irq_chip
---> various functions from struct irq_chip
--> calling remove.

According this scheme may be appropriate to use functions from different structures using chain-order of declaration structures:
firstly functions from structure that is used in init()
Then according mentions of second structure in functions of previous.

Actions
Copy link
 #6

Updated by Pavel Shved about 13 years ago

I can't reproduce this bug with the driver attached. It reports "safe" on all four rules. My toolset version is 32b02f5f. Do you use a non-standard config or something?

Actions
Copy link
 #7

Updated by Pavel Shved about 13 years ago

  • Status changed from Feedback to Open
  • Platform set to Linux x86

The bug is only reproducible on 32-bit architecture, as it seemingly turns on the option CONFIG_GPIO_PCA953X_IRQ by default.

Actions
Copy link

Also available in: Atom PDF