| Message ID | 20221013161700.11631-1-fmdefrancesco@gmail.com |
|---|---|
| State | Accepted |
| Headers | show |
| Series | fs/hostfs: Replace kmap() with kmap_local_page() | expand |
On Thu, Oct 13, 2022 at 06:17:00PM +0200, Fabio M. De Francesco wrote: > The use of kmap() is being deprecated in favor of kmap_local_page(). > > There are two main problems with kmap(): (1) It comes with an overhead as > the mapping space is restricted and protected by a global lock for > synchronization and (2) it also requires global TLB invalidation when the > kmap’s pool wraps and it might block when the mapping space is fully > utilized until a slot becomes available. > > With kmap_local_page() the mappings are per thread, CPU local, can take > page faults, and can be called from any context (including interrupts). > It is faster than kmap() in kernels with HIGHMEM enabled. Furthermore, > the tasks can be preempted and, when they are scheduled to run again, the > kernel virtual addresses are restored and still valid. > > Therefore, replace kmap() with kmap_local_page() in hostfs_kern.c, it > being the only file with kmap() call sites currently left in fs/hostfs. > > Cc: "Venkataramanan, Anirudh" <anirudh.venkataramanan@intel.com> > Suggested-by: Ira Weiny <ira.weiny@intel.com> What can we do to move this patch along? I'm not an expert but it looks ok to me. Reviewed-by: Ira Weiny <ira.weiny@intel.com> > Signed-off-by: Fabio M. De Francesco <fmdefrancesco@gmail.com> > --- > > These changes are not tested in a 32 bits VM as I use to do with other more > problematic conversions. Mere code inspection makes me reasonably think > that the rules of local mappings are not violated by this conversion. > > Furthermore, I have no idea how to test this code. If maintainers think > that tests are absolutely necessary, any hints about how to perform them > would be greatly appreciated. > > fs/hostfs/hostfs_kern.c | 15 ++++++++------- > 1 file changed, 8 insertions(+), 7 deletions(-) > > diff --git a/fs/hostfs/hostfs_kern.c b/fs/hostfs/hostfs_kern.c > index 07881b76d42f..0c9f89352601 100644 > --- a/fs/hostfs/hostfs_kern.c > +++ b/fs/hostfs/hostfs_kern.c > @@ -412,7 +412,7 @@ static int hostfs_writepage(struct page *page, struct writeback_control *wbc) > if (page->index >= end_index) > count = inode->i_size & (PAGE_SIZE-1); > > - buffer = kmap(page); > + buffer = kmap_local_page(page); > > err = write_file(HOSTFS_I(inode)->fd, &base, buffer, count); > if (err != count) { > @@ -428,9 +428,9 @@ static int hostfs_writepage(struct page *page, struct writeback_control *wbc) > err = 0; > > out: > - kunmap(page); > - > + kunmap_local(buffer); > unlock_page(page); > + > return err; > } > > @@ -441,7 +441,7 @@ static int hostfs_read_folio(struct file *file, struct folio *folio) > loff_t start = page_offset(page); > int bytes_read, ret = 0; > > - buffer = kmap(page); > + buffer = kmap_local_page(page); > bytes_read = read_file(FILE_HOSTFS_I(file)->fd, &start, buffer, > PAGE_SIZE); > if (bytes_read < 0) { > @@ -458,8 +458,9 @@ static int hostfs_read_folio(struct file *file, struct folio *folio) > > out: > flush_dcache_page(page); > - kunmap(page); > + kunmap_local(buffer); > unlock_page(page); > + > return ret; > } > > @@ -484,9 +485,9 @@ static int hostfs_write_end(struct file *file, struct address_space *mapping, > unsigned from = pos & (PAGE_SIZE - 1); > int err; > > - buffer = kmap(page); > + buffer = kmap_local_page(page); > err = write_file(FILE_HOSTFS_I(file)->fd, &pos, buffer + from, copied); > - kunmap(page); > + kunmap_local(buffer); > > if (!PageUptodate(page) && err == PAGE_SIZE) > SetPageUptodate(page); > -- > 2.37.3 >
diff --git a/fs/hostfs/hostfs_kern.c b/fs/hostfs/hostfs_kern.c index 07881b76d42f..0c9f89352601 100644 --- a/fs/hostfs/hostfs_kern.c +++ b/fs/hostfs/hostfs_kern.c @@ -412,7 +412,7 @@ static int hostfs_writepage(struct page *page, struct writeback_control *wbc) if (page->index >= end_index) count = inode->i_size & (PAGE_SIZE-1); - buffer = kmap(page); + buffer = kmap_local_page(page); err = write_file(HOSTFS_I(inode)->fd, &base, buffer, count); if (err != count) { @@ -428,9 +428,9 @@ static int hostfs_writepage(struct page *page, struct writeback_control *wbc) err = 0; out: - kunmap(page); - + kunmap_local(buffer); unlock_page(page); + return err; } @@ -441,7 +441,7 @@ static int hostfs_read_folio(struct file *file, struct folio *folio) loff_t start = page_offset(page); int bytes_read, ret = 0; - buffer = kmap(page); + buffer = kmap_local_page(page); bytes_read = read_file(FILE_HOSTFS_I(file)->fd, &start, buffer, PAGE_SIZE); if (bytes_read < 0) { @@ -458,8 +458,9 @@ static int hostfs_read_folio(struct file *file, struct folio *folio) out: flush_dcache_page(page); - kunmap(page); + kunmap_local(buffer); unlock_page(page); + return ret; } @@ -484,9 +485,9 @@ static int hostfs_write_end(struct file *file, struct address_space *mapping, unsigned from = pos & (PAGE_SIZE - 1); int err; - buffer = kmap(page); + buffer = kmap_local_page(page); err = write_file(FILE_HOSTFS_I(file)->fd, &pos, buffer + from, copied); - kunmap(page); + kunmap_local(buffer); if (!PageUptodate(page) && err == PAGE_SIZE) SetPageUptodate(page);
The use of kmap() is being deprecated in favor of kmap_local_page(). There are two main problems with kmap(): (1) It comes with an overhead as the mapping space is restricted and protected by a global lock for synchronization and (2) it also requires global TLB invalidation when the kmap’s pool wraps and it might block when the mapping space is fully utilized until a slot becomes available. With kmap_local_page() the mappings are per thread, CPU local, can take page faults, and can be called from any context (including interrupts). It is faster than kmap() in kernels with HIGHMEM enabled. Furthermore, the tasks can be preempted and, when they are scheduled to run again, the kernel virtual addresses are restored and still valid. Therefore, replace kmap() with kmap_local_page() in hostfs_kern.c, it being the only file with kmap() call sites currently left in fs/hostfs. Cc: "Venkataramanan, Anirudh" <anirudh.venkataramanan@intel.com> Suggested-by: Ira Weiny <ira.weiny@intel.com> Signed-off-by: Fabio M. De Francesco <fmdefrancesco@gmail.com> --- These changes are not tested in a 32 bits VM as I use to do with other more problematic conversions. Mere code inspection makes me reasonably think that the rules of local mappings are not violated by this conversion. Furthermore, I have no idea how to test this code. If maintainers think that tests are absolutely necessary, any hints about how to perform them would be greatly appreciated. fs/hostfs/hostfs_kern.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-)