| Message ID | 1601219284-13275-1-git-send-email-jrdr.linux@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | media: atomisp: Fixed error handling path | expand |
On Sun, Sep 27, 2020 at 08:38:04PM +0530, Souptick Joarder wrote: > Inside alloc_user_pages() based on flag value either pin_user_pages() > or get_user_pages_fast() will be called. However, these API might fail. > > But free_user_pages() called in error handling path doesn't bother > about return value and will try to unpin bo->pgnr pages, which is > incorrect. > > Fix this by passing the page_nr to free_user_pages(). If page_nr > 0 > pages will be unpinned based on bo->mem_type. This will also take care > of non error handling path. > > Fixes: 14a638ab96c5 ("media: atomisp: use pin_user_pages() for memory > allocation") > Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com> > Cc: John Hubbard <jhubbard@nvidia.com> > Cc: Ira Weiny <ira.weiny@intel.com> > Cc: Dan Carpenter <dan.carpenter@oracle.com> > --- > drivers/staging/media/atomisp/pci/hmm/hmm_bo.c | 13 ++++++++----- > 1 file changed, 8 insertions(+), 5 deletions(-) > > diff --git a/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c b/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > index f13af23..0168f98 100644 > --- a/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > +++ b/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > @@ -857,16 +857,17 @@ static void free_private_pages(struct hmm_buffer_object *bo, > kfree(bo->page_obj); > } > > -static void free_user_pages(struct hmm_buffer_object *bo) > +static void free_user_pages(struct hmm_buffer_object *bo, > + unsigned int page_nr) > { > int i; > > hmm_mem_stat.usr_size -= bo->pgnr; ^^^^^^^^^^^ This is still a problem. It needs to be hmm_mem_stat.usr_size -= page_nr. regards, dan carpenter > > if (bo->mem_type == HMM_BO_MEM_TYPE_PFN) { > - unpin_user_pages(bo->pages, bo->pgnr); > + unpin_user_pages(bo->pages, page_nr); > } else { > - for (i = 0; i < bo->pgnr; i++) > + for (i = 0; i < page_nr; i++) > put_page(bo->pages[i]); > } > kfree(bo->pages);
Hi Dan, On Mon, Sep 28, 2020 at 2:08 PM Dan Carpenter <dan.carpenter@oracle.com> wrote: > > On Sun, Sep 27, 2020 at 08:38:04PM +0530, Souptick Joarder wrote: > > Inside alloc_user_pages() based on flag value either pin_user_pages() > > or get_user_pages_fast() will be called. However, these API might fail. > > > > But free_user_pages() called in error handling path doesn't bother > > about return value and will try to unpin bo->pgnr pages, which is > > incorrect. > > > > Fix this by passing the page_nr to free_user_pages(). If page_nr > 0 > > pages will be unpinned based on bo->mem_type. This will also take care > > of non error handling path. > > > > Fixes: 14a638ab96c5 ("media: atomisp: use pin_user_pages() for memory > > allocation") > > Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com> > > Cc: John Hubbard <jhubbard@nvidia.com> > > Cc: Ira Weiny <ira.weiny@intel.com> > > Cc: Dan Carpenter <dan.carpenter@oracle.com> > > --- > > drivers/staging/media/atomisp/pci/hmm/hmm_bo.c | 13 ++++++++----- > > 1 file changed, 8 insertions(+), 5 deletions(-) > > > > diff --git a/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c b/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > > index f13af23..0168f98 100644 > > --- a/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > > +++ b/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > > @@ -857,16 +857,17 @@ static void free_private_pages(struct hmm_buffer_object *bo, > > kfree(bo->page_obj); > > } > > > > -static void free_user_pages(struct hmm_buffer_object *bo) > > +static void free_user_pages(struct hmm_buffer_object *bo, > > + unsigned int page_nr) > > { > > int i; > > > > hmm_mem_stat.usr_size -= bo->pgnr; > ^^^^^^^^^^^ > This is still a problem. It needs to be hmm_mem_stat.usr_size -= page_nr. In failure path, it is hmm_mem_stat.usr_size += bo->pgnr. I think, hmm_mem_stat.usr_size -= bo->pgnr is correct as per existing code. Do you think that needs to be changed ? > > regards, > dan carpenter > > > > > if (bo->mem_type == HMM_BO_MEM_TYPE_PFN) { > > - unpin_user_pages(bo->pages, bo->pgnr); > > + unpin_user_pages(bo->pages, page_nr); > > } else { > > - for (i = 0; i < bo->pgnr; i++) > > + for (i = 0; i < page_nr; i++) > > put_page(bo->pages[i]); > > } > > kfree(bo->pages); >
On Tue, Sep 29, 2020 at 07:34:39AM +0530, Souptick Joarder wrote: > Hi Dan, > > > On Mon, Sep 28, 2020 at 2:08 PM Dan Carpenter <dan.carpenter@oracle.com> wrote: > > > > On Sun, Sep 27, 2020 at 08:38:04PM +0530, Souptick Joarder wrote: > > > Inside alloc_user_pages() based on flag value either pin_user_pages() > > > or get_user_pages_fast() will be called. However, these API might fail. > > > > > > But free_user_pages() called in error handling path doesn't bother > > > about return value and will try to unpin bo->pgnr pages, which is > > > incorrect. > > > > > > Fix this by passing the page_nr to free_user_pages(). If page_nr > 0 > > > pages will be unpinned based on bo->mem_type. This will also take care > > > of non error handling path. > > > > > > Fixes: 14a638ab96c5 ("media: atomisp: use pin_user_pages() for memory > > > allocation") > > > Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com> > > > Cc: John Hubbard <jhubbard@nvidia.com> > > > Cc: Ira Weiny <ira.weiny@intel.com> > > > Cc: Dan Carpenter <dan.carpenter@oracle.com> > > > --- > > > drivers/staging/media/atomisp/pci/hmm/hmm_bo.c | 13 ++++++++----- > > > 1 file changed, 8 insertions(+), 5 deletions(-) > > > > > > diff --git a/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c b/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > > > index f13af23..0168f98 100644 > > > --- a/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > > > +++ b/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c > > > @@ -857,16 +857,17 @@ static void free_private_pages(struct hmm_buffer_object *bo, > > > kfree(bo->page_obj); > > > } > > > > > > -static void free_user_pages(struct hmm_buffer_object *bo) > > > +static void free_user_pages(struct hmm_buffer_object *bo, > > > + unsigned int page_nr) > > > { > > > int i; > > > > > > hmm_mem_stat.usr_size -= bo->pgnr; > > ^^^^^^^^^^^ > > This is still a problem. It needs to be hmm_mem_stat.usr_size -= page_nr. > > In failure path, it is hmm_mem_stat.usr_size += bo->pgnr. > I think, hmm_mem_stat.usr_size -= bo->pgnr is correct as per existing code. > Do you think that needs to be changed ? > Yeah. Crud. I'm sorry. You had it right. Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com> regards, dan carpenter
diff --git a/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c b/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c index f13af23..0168f98 100644 --- a/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c +++ b/drivers/staging/media/atomisp/pci/hmm/hmm_bo.c @@ -857,16 +857,17 @@ static void free_private_pages(struct hmm_buffer_object *bo, kfree(bo->page_obj); } -static void free_user_pages(struct hmm_buffer_object *bo) +static void free_user_pages(struct hmm_buffer_object *bo, + unsigned int page_nr) { int i; hmm_mem_stat.usr_size -= bo->pgnr; if (bo->mem_type == HMM_BO_MEM_TYPE_PFN) { - unpin_user_pages(bo->pages, bo->pgnr); + unpin_user_pages(bo->pages, page_nr); } else { - for (i = 0; i < bo->pgnr; i++) + for (i = 0; i < page_nr; i++) put_page(bo->pages[i]); } kfree(bo->pages); @@ -942,6 +943,8 @@ static int alloc_user_pages(struct hmm_buffer_object *bo, dev_err(atomisp_dev, "get_user_pages err: bo->pgnr = %d, pgnr actually pinned = %d.\n", bo->pgnr, page_nr); + if (page_nr < 0) + page_nr = 0; goto out_of_mem; } @@ -954,7 +957,7 @@ static int alloc_user_pages(struct hmm_buffer_object *bo, out_of_mem: - free_user_pages(bo); + free_user_pages(bo, page_nr); return -ENOMEM; } @@ -1037,7 +1040,7 @@ void hmm_bo_free_pages(struct hmm_buffer_object *bo) if (bo->type == HMM_BO_PRIVATE) free_private_pages(bo, &dynamic_pool, &reserved_pool); else if (bo->type == HMM_BO_USER) - free_user_pages(bo); + free_user_pages(bo, bo->pgnr); else dev_err(atomisp_dev, "invalid buffer type.\n"); mutex_unlock(&bo->mutex);
Inside alloc_user_pages() based on flag value either pin_user_pages() or get_user_pages_fast() will be called. However, these API might fail. But free_user_pages() called in error handling path doesn't bother about return value and will try to unpin bo->pgnr pages, which is incorrect. Fix this by passing the page_nr to free_user_pages(). If page_nr > 0 pages will be unpinned based on bo->mem_type. This will also take care of non error handling path. Fixes: 14a638ab96c5 ("media: atomisp: use pin_user_pages() for memory allocation") Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com> Cc: John Hubbard <jhubbard@nvidia.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Dan Carpenter <dan.carpenter@oracle.com> --- drivers/staging/media/atomisp/pci/hmm/hmm_bo.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)