Thread (12 messages) flat view 12 messages, 5 authors, 2008-03-25

Re: [PATCH 1/2 v2] Driver for Freescale 8610 and 5121 DIU

From: Peter Zijlstra <hidden>
Date: 2008-03-20 23:02:51
Also in: linux-fbdev, lkml
Subsystem: locking primitives, the rest · Maintainers: Peter Zijlstra, Ingo Molnar, Will Deacon, Boqun Feng, Linus Torvalds

On Thu, 2008-03-20 at 15:27 -0700, Andrew Morton wrote:
quoted hunk ↗ jump to hunk
quoted
+static struct diu_hw dr = {
+	.mode = MFB_MODE1,
+	.reg_lock = __SPIN_LOCK_UNLOCKED(old_style_spin_init),
+};
I'm not clear on what's supposed to happen with __SPIN_LOCK_UNLOCKED().  I
do know that its documentation is crap.

I guess you should have used SPIN_LOCK_UNLOCKED there rather than
open-coding its equivalent.  And SPIN_LOCK_UNLOCKED _is_ documented.  It
says "don't use this".

Now, s/documentation/guesswork-and-grep/ says that you're supposed to pick
a tree-wide-unique string here as your lockdep key.

So I did this:
--- a/drivers/video/fsl-diu-fb.c~fbdev-driver-for-freescale-8610-and-5121-diu-fix
+++ a/drivers/video/fsl-diu-fb.c
@@ -274,7 +274,7 @@ static struct mfb_info mfb_template[] = 
 
 static struct diu_hw dr = {
 	.mode = MFB_MODE1,
-	.reg_lock = __SPIN_LOCK_UNLOCKED(old_style_spin_init),
+	.reg_lock = __SPIN_LOCK_UNLOCKED(diu_hw.reg_lock),
 };
#define __SPIN_LOCK_UNLOCKED(lockname)

seems pretty suggestive to me, also DEFINE_SPINLOCK(x) shows the proper
usage; the right thing would have been:

static struct diu_hw dr = {
 	.mode = MFB_MODE1,
-	.reg_lock = __SPIN_LOCK_UNLOCKED(old_style_spin_init),
+	.reg_lock = __SPIN_LOCK_UNLOCKED(dr.reg_lock),
};

Where 'dr.reg_lock' is the expression to locate the actual object.

Does something like this make sense to you:

---
Subject: lockdep: document __SPIN_LOCK_UNLOCKED() usage

Small comment clarifying the intended usage of __SPIN_LOCK_UNLOCKED()

[ for the observant readers; yes this prescribes a usage that is more
  than strictly needed, it does however enable interesting future uses ]

Signed-off-by: Peter Zijlstra <redacted>
---
diff --git a/include/linux/spinlock_types.h b/include/linux/spinlock_types.h
index 68d88f7..4278558 100644
--- a/include/linux/spinlock_types.h
+++ b/include/linux/spinlock_types.h
@@ -63,6 +63,10 @@ typedef struct {
 # define RW_DEP_MAP_INIT(lockname)
 #endif
 
+/*
+ * __*_LOCK_UNLOCKED(lockname), where 'lockname' is a valid C expression
+ * that evaluates to the actual object being initialized.
+ */
 #ifdef CONFIG_DEBUG_SPINLOCK
 # define __SPIN_LOCK_UNLOCKED(lockname)					\
 	(spinlock_t)	{	.raw_lock = __RAW_SPIN_LOCK_UNLOCKED,	\
---
Perhaps I should make the macros do something like:

  typecheck(spinlock_t, lockname)

And sweep the tree to make it compile again.
 static struct diu_pool pool;
quoted
+static struct diu_pool pool;
+
+/*	To allocate memory for framebuffer. First try __get_free_pages(). If it
+ *	fails, try rh_alloc. The reason is __get_free_pages() cannot allocate
+ *	very large memory (more than 4MB). We don't want to allocate all memory
+ *	in rheap since small memory allocation/deallocation will fragment the
+ *	rheap and make the furture large allocation fail.
+ */
+
+void *fsl_diu_alloc(unsigned long size, phys_addr_t *phys)
+{
+	void *virt;
+
+	pr_debug("size=%lu\n", size);
+
+	virt = (void *)__get_free_pages(GFP_DMA | GFP_ATOMIC, get_order(size));
GFP_DMA implies GFP_ATOMIC, but it's appropriate for documentation purposes.
FWIW, I prefer the form: GFP_type | __GFP_modifiers

For instance: 'GFP_KERNEL | __GFP_DMA | __GFP_ZERO'
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help