On 15/12/2020 14:06, Dan Carpenter wrote:
Hello John Garry,
+
The patch e15f2fa959f2: "driver core: platform: Add
devm_platform_get_irqs_affinity()" from Dec 2, 2020, leads to the
following static checker warning:
drivers/base/platform.c:370 devm_platform_get_irqs_affinity()
warn: 'nvec' possible negative type promoted to high
drivers/base/platform.c
351 int devm_platform_get_irqs_affinity(struct platform_device *dev,
352 struct irq_affinity *affd,
353 unsigned int minvec,
354 unsigned int maxvec,
355 int **irqs)
356 {
357 struct irq_affinity_devres *ptr;
358 struct irq_affinity_desc *desc;
359 size_t size;
360 int i, ret, nvec;
361
362 if (!affd)
363 return -EPERM;
364
365 if (maxvec < minvec)
366 return -ERANGE;
367
368 nvec = platform_irq_count(dev);
369
370 if (nvec < minvec)
^^^^
If "nvec" is -EPROBE_DEFER then "minvec" is an unsigned int so it gets
type promoted to a very high positive so it's > minvec.
Today this is kinda safe as the only in-tree user calls
platform_get_irq() and tests for -EPROBE_DEFER prior to the
devm_platform_get_irqs_affinity() call.
But, indeed, I should test for -EPROBE_DEFER here. And I can remove that
platform_get_irq() I mentioned, which existed for another reason.
Thanks,
John
371 return -ENOSPC;
372
373 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
374 if (nvec < minvec)
375 return -ENOSPC;
376
377 if (nvec > maxvec)
378 nvec = maxvec;
379
380 size = sizeof(*ptr) + sizeof(unsigned int) * nvec;
regards,
dan carpenter
.
