lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200506084558.tcayd2fuzoe6rsfm@rric.localdomain>
Date:   Wed, 6 May 2020 10:45:59 +0200
From:   Robert Richter <rrichter@...vell.com>
To:     Borislav Petkov <bp@...en8.de>
CC:     Mauro Carvalho Chehab <mchehab@...nel.org>,
        Tony Luck <tony.luck@...el.com>,
        James Morse <james.morse@....com>,
        Aristeu Rozanski <aris@...hat.com>,
        Matthias Brugger <mbrugger@...e.com>,
        <linux-edac@...r.kernel.org>, <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2 08/10] EDAC/ghes: Carve out MC device handling into
 separate functions

On 27.04.20 18:38:56, Borislav Petkov wrote:
> On Wed, Apr 22, 2020 at 01:58:12PM +0200, Robert Richter wrote:

> > +static int ghes_mc_add_or_free(struct mem_ctl_info *mci,
> > +			struct list_head *dimm_list)
> 
> No, I think we talked about this already. This function should be
> called:
> 
> 	ghes_mc_add()
> 
> and should do one thing and one thing only in good old unix tradition:
> add the MC.
> 
> > +{
> >  	unsigned long flags;
> > -	int idx = -1;
> > +	int rc;
> > +
> > +	rc = edac_mc_add_mc(mci);
> > +	if (rc < 0) {
> 
> > +		ghes_dimm_release(dimm_list);
> > +		edac_mc_free(mci);
> > +		return rc;
> 
> Those last three lines should be called by the *caller* of
> ghes_mc_add(), when latter returns an error value.

These direct operations are nothing a caller should deal with.

The caller does now:

	mci = ghes_mc_create(...);
	... /* prepare dimms */
	return ghes_mc_add_or_free(...);

To shut it down we just use:

	ghes_mc_free();

Pretty simple.

Now, lets look at your suggestion to put it out of the function. A
caller always needs to free the mci and dimms, so we will get:

	int rc;
	mci = ghes_mc_create(...);
	... /* prepare dimms */
	rc = ghes_mc_add(...);
	if (rc < 0) {
		/* free mci */
		/* free dimms */
		...
	}
	return rc;

We loose the tail call and simplicity here. Note this duplicates code
as there are 2 users of ghes_mc_add().

Now, the caller does not know the implementation details, so we need
to provide another release function (let's call it *_release() here):

	mci = ghes_mc_create(...);
	... /* prepare dimms */
	rc = ghes_mc_add(...);
	if (rc < 0) {
		ghes_mc_release(mci);
		ghes_dimm_release(dimm_list);
	}
	return rc;

Ok, now there is another function needed to release everything.

This design also impacts ghes_mc_free(). So the shutdown
implementation would turn to:

	struct mem_ctl_info *mci;
	...
	mci = ghes_mc_del();
	ghes_mc_release(mci);
	...

I don't see any benefit. See also below for the delta of an
implementation of the suggested changes.

> 
> > +	}
> > +
> > +	spin_lock_irqsave(&ghes_lock, flags);
> > +	ghes_pvt = mci->pvt_info;
> > +	list_splice_tail(dimm_list, &ghes_dimm_list);
> > +	spin_unlock_irqrestore(&ghes_lock, flags);
> > +
> > +	return 0;
> > +}
> > +
> > +static void ghes_mc_free(void)
> > +{
> > +	struct mem_ctl_info *mci;
> > +	unsigned long flags;
> > +	LIST_HEAD(dimm_list);
> > +
> > +	/*
> > +	 * Wait for the irq handler being finished.
> > +	 */
> > +	spin_lock_irqsave(&ghes_lock, flags);
> > +	mci = ghes_pvt ? ghes_pvt->mci : NULL;
> > +	ghes_pvt = NULL;
> > +	list_splice_init(&ghes_dimm_list, &dimm_list);
> > +	spin_unlock_irqrestore(&ghes_lock, flags);
> > +
> > +	ghes_dimm_release(&dimm_list);
> > +
> > +	if (!mci)
> > +		return;
> > +
> > +	mci = edac_mc_del_mc(mci->pdev);
> > +	if (mci)
> > +		edac_mc_free(mci);
> > +}
> 
> This function needs to do only freeing of the mc. The list splicing and
> dimm releasing needs to be done by its caller, before calling it.

ghes_mc_free() is the counterpart to ghes_mc_add() and thus needs to
also handle the dimm_list here. This cannot be left to the caller.

Considering all the above, I don't see how your suggestions to the
interface could improve the code. Hmm...

Below an implementation that illustrates the changes.

Thanks,

-Robert

---
 drivers/edac/ghes_edac.c | 41 +++++++++++++++++++++++++++-------------
 1 file changed, 28 insertions(+), 13 deletions(-)

diff --git a/drivers/edac/ghes_edac.c b/drivers/edac/ghes_edac.c
index 7f39346d895b..896d7b488fc2 100644
--- a/drivers/edac/ghes_edac.c
+++ b/drivers/edac/ghes_edac.c
@@ -576,18 +576,14 @@ static struct mem_ctl_info *ghes_mc_create(struct device *dev, int mc_idx,
 	return mci;
 }
 
-static int ghes_mc_add_or_free(struct mem_ctl_info *mci,
-			       struct list_head *dimms)
+static int ghes_mc_add(struct mem_ctl_info *mci, struct list_head *dimms)
 {
 	unsigned long flags;
 	int rc;
 
 	rc = edac_mc_add_mc(mci);
-	if (rc < 0) {
-		dimm_release(dimms);
-		edac_mc_free(mci);
+	if (rc < 0)
 		return rc;
-	}
 
 	spin_lock_irqsave(&ghes_lock, flags);
 	ghes_pvt = mci->pvt_info;
@@ -597,7 +593,7 @@ static int ghes_mc_add_or_free(struct mem_ctl_info *mci,
 	return 0;
 }
 
-static void ghes_mc_free(void)
+static struct mem_ctl_info *ghes_mc_del(void)
 {
 	struct mem_ctl_info *mci;
 	unsigned long flags;
@@ -614,10 +610,14 @@ static void ghes_mc_free(void)
 
 	dimm_release(&dimms);
 
-	if (!mci)
-		return;
+	if (mci)
+		mci = edac_mc_del_mc(mci->pdev);
 
-	mci = edac_mc_del_mc(mci->pdev);
+	return mci;
+}
+
+static void ghes_mc_release(struct mem_ctl_info *mci)
+{
 	if (mci)
 		edac_mc_free(mci);
 }
@@ -627,6 +627,7 @@ static int ghes_edac_register_fake(struct device *dev)
 	struct mem_ctl_info *mci;
 	struct dimm_info *dimm;
 	LIST_HEAD(empty);
+	int rc;
 
 	mci = ghes_mc_create(dev, 0, 1);
 	if (!mci)
@@ -642,13 +643,18 @@ static int ghes_edac_register_fake(struct device *dev)
 
 	snprintf(dimm->label, sizeof(dimm->label), "unknown memory");
 
-	return ghes_mc_add_or_free(mci, &empty);
+	rc = ghes_mc_add(mci, &empty);
+	if (rc < 0)
+		ghes_mc_free(mci);
+
+	return rc;
 }
 
 static int ghes_edac_register_one(struct device *dev, int mc_idx, int num_dimm)
 {
 	struct dimm_fill dimm_fill;
 	struct mem_ctl_info *mci;
+	int rc;
 
 	mci = ghes_mc_create(dev, mc_idx, num_dimm);
 	if (!mci)
@@ -660,7 +666,13 @@ static int ghes_edac_register_one(struct device *dev, int mc_idx, int num_dimm)
 
 	dmi_walk(ghes_edac_dmidecode, &dimm_fill);
 
-	return ghes_mc_add_or_free(mci, &dimm_fill.dimms);
+	rc = ghes_mc_add(mci, &dimm_fill.dimms);
+	if (rc < 0) {
+		dimm_release(&dimm_fill.dimms);
+		ghes_mc_release(mci);
+	}
+
+	return rc;
 }
 
 int ghes_edac_register(struct ghes *ghes, struct device *dev)
@@ -740,10 +752,13 @@ int ghes_edac_register(struct ghes *ghes, struct device *dev)
 
 void ghes_edac_unregister(struct ghes *ghes)
 {
+	struct mem_ctl_info *mci;
+
 	mutex_lock(&ghes_reg_mutex);
 
 	if (refcount_dec_and_test(&ghes_refcount)) {
-		ghes_mc_free();
+		mci = ghes_mc_del();
+		ghes_mc_release(mci);
 		dimm_pool_destroy();
 	}
 
-- 
2.20.1

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ