Re: [PATCH v2] staging: mt29f_spinand: fix memory leak while programming pages

From: Jheng-Jhong Wu
Date: Thu Aug 02 2018 - 12:02:48 EST

Dear greg k-h,

Before device is removed and freed memory automatically, programming
pages may run many many times.
Assume we erase and rewrite a large part of the flash, then
spinand_program_page() might exhaust memory if memory is not large
We may not remove and re-add the device between each programming page, right?
In fact, OOM indeed occurred when I tested programming multi-pages by
mtd_debug tool.
Erased first, then programmed pages.

Best Regards,
Jheng-Jhong Wu (Victor Wu)
E-mail: goodwater.wu@xxxxxxxxx
Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> æ 2018å8æ2æ éå äå4:03åéï
> On Thu, Aug 02, 2018 at 03:51:30PM +0800, Jheng-Jhong Wu wrote:
> > In spinand_program_page(), it uses devm_kzalloc() to allocate memory to
> > wbuf dynamically if internal ECC is on, but it doesn't free memory
> > allocated to wbuf at the end of this function. Before the spinand device
> > is removed and frees memory automatically, programming pages may run many
> > times. This leads to a memory leak issue when internal ECC is on.
> How is this a memory leak? The memory will be freed when the struct
> device is removed from the system. How did you test that there was a
> leak?
> > Changelog:
> >
> > v2:
> > - use kzalloc()/kfree() to replace devm_kzalloc()/devm_kfree()
> > - add some descriptions to commit message
> this changelog goes below the --- line.
> thanks,
> greg k-h