Re: [PATCH v1 4/4] misc: fastrpc: Add debugfs support for fastrpc

[Greg KH <gregkh@linuxfoundation.org>]

On Mon, Apr 14, 2025 at 12:41:47PM +0530, Deepika Singh wrote:
> 
> 
> On 4/11/2025 1:55 PM, Ekansh Gupta wrote:
> > 
> > 
> > On 12/3/2024 5:27 PM, Dmitry Baryshkov wrote:
> > > On Tue, 3 Dec 2024 at 07:22, Ekansh Gupta <quic_ekangupt@quicinc.com> wrote:
> > > > 
> > > > 
> > > > On 12/2/2024 6:18 PM, Dmitry Baryshkov wrote:
> > > > > On Mon, Dec 02, 2024 at 03:27:43PM +0530, Ekansh Gupta wrote:
> > > > > > On 11/22/2024 12:23 AM, Dmitry Baryshkov wrote:
> > > > > > > On Thu, Nov 21, 2024 at 12:12:17PM +0530, Ekansh Gupta wrote:
> > > > > > > > On 11/18/2024 7:32 PM, Greg KH wrote:
> > > > > > > > > On Mon, Nov 18, 2024 at 02:10:46PM +0530, Ekansh Gupta wrote:
> > > > > > > > > > Add changes to support debugfs. The fastrpc directory will be
> > > > > > > > > > created which will carry debugfs files for all fastrpc processes.
> > > > > > > > > > The information of fastrpc user and channel contexts are getting
> > > > > > > > > > captured as part of this change.
> > > > > > > > > > 
> > > > > > > > > > Signed-off-by: Ekansh Gupta <quic_ekangupt@quicinc.com>
> > > > > > > > > > ---
> > > > > > > > > >   drivers/misc/fastrpc/Makefile        |   3 +-
> > > > > > > > > >   drivers/misc/fastrpc/fastrpc_debug.c | 156 +++++++++++++++++++++++++++
> > > > > > > > > >   drivers/misc/fastrpc/fastrpc_debug.h |  31 ++++++
> > > > > > > > > >   drivers/misc/fastrpc/fastrpc_main.c  |  18 +++-
> > > > > > > > > >   4 files changed, 205 insertions(+), 3 deletions(-)
> > > > > > > > > >   create mode 100644 drivers/misc/fastrpc/fastrpc_debug.c
> > > > > > > > > >   create mode 100644 drivers/misc/fastrpc/fastrpc_debug.h
> > > > > > > > > > 
> > > > > > > > > > diff --git a/drivers/misc/fastrpc/Makefile b/drivers/misc/fastrpc/Makefile
> > > > > > > > > > index 020d30789a80..4ff6b64166ae 100644
> > > > > > > > > > --- a/drivers/misc/fastrpc/Makefile
> > > > > > > > > > +++ b/drivers/misc/fastrpc/Makefile
> > > > > > > > > > @@ -1,3 +1,4 @@
> > > > > > > > > >   # SPDX-License-Identifier: GPL-2.0
> > > > > > > > > >   obj-$(CONFIG_QCOM_FASTRPC)      += fastrpc.o
> > > > > > > > > > -fastrpc-objs    := fastrpc_main.o
> > > > > > > > > > \ No newline at end of file
> > > > > > > > > > +fastrpc-objs    := fastrpc_main.o \
> > > > > > > > > > +                fastrpc_debug.o
> > > > > > > > > Only build this file if debugfs is enabled.
> > > > > > > > > 
> > > > > > > > > And again, "debug.c"?
> > > > > > > > I'll add change to build this only if debugfs is enabled. Going forward I have plans to add
> > > > > > > > few more debug specific changes, maybe then I'll need to change the build rules again.
> > > > > > > > > > diff --git a/drivers/misc/fastrpc/fastrpc_debug.c b/drivers/misc/fastrpc/fastrpc_debug.c
> > > > > > > > > > new file mode 100644
> > > > > > > > > > index 000000000000..cdb4fc6845a8
> > > > > > > > > > --- /dev/null
> > > > > > > > > > +++ b/drivers/misc/fastrpc/fastrpc_debug.c
> > > > > > > > > > @@ -0,0 +1,156 @@
> > > > > > > > > > +// SPDX-License-Identifier: GPL-2.0
> > > > > > > > > > +// Copyright (c) 2024 Qualcomm Innovation Center.
> > > > > > > > > > +
> > > > > > > > > > +#include <linux/debugfs.h>
> > > > > > > > > > +#include <linux/seq_file.h>
> > > > > > > > > > +#include "fastrpc_shared.h"
> > > > > > > > > > +#include "fastrpc_debug.h"
> > > > > > > > > > +
> > > > > > > > > > +#ifdef CONFIG_DEBUG_FS
> > > > > > > > > Please put the #ifdef in the .h file, not in the .c file.
> > > > > > > > Ack
> > > > > > > > > > +void fastrpc_create_user_debugfs(struct fastrpc_user *fl)
> > > > > > > > > > +{
> > > > > > > > > > +        char cur_comm[TASK_COMM_LEN];
> > > > > > > > > > +        int domain_id, size;
> > > > > > > > > > +        char *debugfs_buf;
> > > > > > > > > > +        struct dentry *debugfs_dir = fl->cctx->debugfs_dir;
> > > > > > > > > > +
> > > > > > > > > > +        memcpy(cur_comm, current->comm, TASK_COMM_LEN);
> > > > > > > > > > +        cur_comm[TASK_COMM_LEN-1] = '\0';
> > > > > > > > > > +        if (debugfs_dir != NULL) {
> > > > > > > > > > +                domain_id = fl->cctx->domain_id;
> > > > > > > > > > +                size = snprintf(NULL, 0, "%.10s_%d_%d_%d", cur_comm,
> > > > > > > > > > +                                current->pid, fl->tgid, domain_id) + 1;
> > > > > > > > > > +                debugfs_buf = kzalloc(size, GFP_KERNEL);
> > > > > > > > > > +                if (debugfs_buf == NULL)
> > > > > > > > > > +                        return;
> > > > > > > > > > +                /*
> > > > > > > > > > +                 * Use HLOS process name, HLOS PID, fastrpc user TGID,
> > > > > > > > > > +                 * domain_id in debugfs filename to create unique file name
> > > > > > > > > > +                 */
> > > > > > > > > > +                snprintf(debugfs_buf, size, "%.10s_%d_%d_%d",
> > > > > > > > > > +                        cur_comm, current->pid, fl->tgid, domain_id);
> > > > > > > > > > +                fl->debugfs_file = debugfs_create_file(debugfs_buf, 0644,
> > > > > > > > > > +                                debugfs_dir, fl, &fastrpc_debugfs_fops);
> > > > > > > > > Why are you saving the debugfs file?  What do you need to do with it
> > > > > > > > > that you can't just delete the whole directory, or look up the name
> > > > > > > > > again in the future when removing it?
> > > > > > > > fl structure is specific to a process using fastrpc driver. The reason to save
> > > > > > > > this debugfs file is to delete is when the process releases fastrpc device.
> > > > > > > > If the file is not deleted, it might flood multiple files in debugfs directory.
> > > > > > > > 
> > > > > > > > As part of this change, only the file that is getting created by a process is
> > > > > > > > getting removed when process is releasing device and I don't think we
> > > > > > > > can clean up the whole directory at this point.
> > > > > > > My 2c: it might be better to create a single file that conains
> > > > > > > information for all the processes instead of that. Or use fdinfo data to
> > > > > > > export process / FD information to userspace.
> > > > > > Thanks for your review. The reason of not having single file for all processes is that
> > > > > > I can run 100s of iteration for any process(say calculator) and every time the properties
> > > > > > of the process can differ(like buffer, session etc.). For this reason, I'm creating and
> > > > > > deleting the debugfs files for every process run.
> > > > > > 
> > > > > > Do you see any advantage of using fdinfo over debugfs? I'm not sure if we can add all
> > > > > > the information(like in debugfs) here.
> > > > > Which information is actually useful / interesting for application
> > > > > developers? If not for the fdinfo, I might still vote for a single file
> > > > > rather than a pile of per-process data.
> Let’s say I am trying to do debugfs read when 10+ or more sessions are
> active per channel, then for pushing data of nth process in a single file, I
> would have to wait for n-1 processes, by that time process data might get
> changed. How do you suggest handling this?

I suggest you NEVER use debugfs for anything that you care about this
type of thing for.

debugfs is for debugging.  Don't expect to rely on it for anything
relating to performance, and many/most systems don't even have it
enabled.  It also can NOT be used for anything that actually is a real
functionality of the system, and MUST work properly if it is not enabled
or a failure happens with the creation of a debugfs file.

So why would this even be an issue, as surely you aren't expecting that
debugfs be the main api for your driver, right?  :)

thanks,

greg k-h