Re: [PATCH hotfix 6.11] minmax: reduce egregious min/max macro expansion

On Wed, Sep 11, 2024 at 06:24:54PM GMT, Hans de Goede wrote:

Hi Lorenzo,

On 9/11/24 5:34 PM, Lorenzo Stoakes wrote:

quoted

Avoid nested min()/max() which results in egregious macro expansion.

This issue was introduced by commit 867046cc7027 ("minmax: relax check to
allow comparison between unsigned arguments and signed constants") [2].

Work has been done to address the issue of egregious min()/max() macro
expansion in commit 22f546873149 ("minmax: improve macro expansion and type
checking") and related, however it appears that some issues remain on more
tightly constrained systems.

Adjust a few known-bad cases of deeply nested macros to avoid doing so to
mitigate this. Porting the patch first proposed in [1] to Linus's tree.

Running an allmodconfig build using the methodology described in [2] we
observe a 35 MiB reduction in generated code.

The difference is much more significant prior to recent minmax fixes which
were not backported. As per [1] prior these the reduction is more like 200
MiB.

This resolves an issue with slackware 15.0 32-bit compilation as reported
by Richard Narron.

Presumably the min/max fixups would be difficult to backport, this patch
should be easier and fix's Richard's problem in 5.15.

[0]:https://lore.kernel.org/all/b97faef60ad24922b530241c5d7c933c@AcuMS.aculab.com/ (local)
[1]:https://lore.kernel.org/lkml/5882b96e-1287-4390-8174-3316d39038ef@lucifer.local/ (local)
[2]:https://lore.kernel.org/linux-mm/36aa2cad-1db1-4abf-8dd2-fb20484aabc3@lucifer.local/ (local)

Reported-by: Richard Narron <redacted>
Closes: https://lore.kernel.org/all/4a5321bd-b1f-1832-f0c-cea8694dc5aa@aaazen.com/ (local)
Fixes: 867046cc7027 ("minmax: relax check to allow comparison between unsigned arguments and signed constants")
Cc: stable@vger.kernel.org
Signed-off-by: Lorenzo Stoakes <redacted>

Thank you for your patch.

I must say that I'm not a fan of that this is patching 3 totally
unrelated files here in a single patch.

This is e.g. going to be a problem if we need to revert one of
the changes because of regressions...

So I would prefer this to be split into 3 patches.

Well, I was doing this as a favour to Richard between other work so put
this together quickly, but you're right this is going to be a pain to
backport/revert if issues so absolutely - will do.

Since this is a hotfix I'm going to risk annoying people and shoot out
a v2 on same day as v1. Sorry in advance.

One review comment for the atomisp bits inline / below.

quoted

---
 drivers/net/ethernet/marvell/mvpp2/mvpp2.h    |  2 +-
 .../staging/media/atomisp/pci/sh_css_frac.h   | 26 ++++++++++++++-----
 include/linux/skbuff.h                        |  6 ++++-
 3 files changed, 25 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
index e809f91c08fb..8b431f90efc3 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h

@@ -23,7 +23,7 @@
 /* The PacketOffset field is measured in units of 32 bytes and is 3 bits wide,
  * so the maximum offset is 7 * 32 = 224
  */
-#define MVPP2_SKB_HEADROOM	min(max(XDP_PACKET_HEADROOM, NET_SKB_PAD), 224)
+#define MVPP2_SKB_HEADROOM	clamp_t(int, XDP_PACKET_HEADROOM, NET_SKB_PAD, 224)

 #define MVPP2_XDP_PASS		0
 #define MVPP2_XDP_DROPPED	BIT(0)

diff --git a/drivers/staging/media/atomisp/pci/sh_css_frac.h b/drivers/staging/media/atomisp/pci/sh_css_frac.h
index b90b5b330dfa..a973394c5bc0 100644
--- a/drivers/staging/media/atomisp/pci/sh_css_frac.h
+++ b/drivers/staging/media/atomisp/pci/sh_css_frac.h

@@ -32,12 +32,24 @@
 #define uISP_VAL_MAX		      ((unsigned int)((1 << uISP_REG_BIT) - 1))

 /* a:fraction bits for 16bit precision, b:fraction bits for ISP precision */
-#define sDIGIT_FITTING(v, a, b) \
-	min_t(int, max_t(int, (((v) >> sSHIFT) >> max(sFRACTION_BITS_FITTING(a) - (b), 0)), \
-	  sISP_VAL_MIN), sISP_VAL_MAX)
-#define uDIGIT_FITTING(v, a, b) \
-	min((unsigned int)max((unsigned)(((v) >> uSHIFT) \
-	>> max((int)(uFRACTION_BITS_FITTING(a) - (b)), 0)), \
-	  uISP_VAL_MIN), uISP_VAL_MAX)
+static inline int sDIGIT_FITTING(short v, int a, int b)
+{

drivers/staging/media/atomisp/pci/isp/kernels/s3a/s3a_1.0/ia_css_s3a.host.c

calls this with ia_css_3a_config.af_fir1_coef / .af_fir2_coef
as first argument those are of the ia_css_s0_15 type which is:

/* Signed fixed point value, 0 integer bits, 15 fractional bits */
typedef s32 ia_css_s0_15;

please replace the "short v" with "int v"

Yeah I think you're right, it's odd, because it seems that the shift value
and the comments implies that this is a short, but perhaps it's more so
that values are shifted as to obtain 16 bits of precision.

I think that you can then also replace clamp_t() with clamp()

The use of clamp_t() is to avoid egregious macro expansion in
clamp(). After the series improving min/max the clamp() is probably
equivalent. But in 5.15 it will likely not be. So this is, in line with the
purpose of this change, I believe necesasry.

quoted

+	int fit_shift = sFRACTION_BITS_FITTING(a) - b;
+
+	v >>= sSHIFT;
+	v >>= fit_shift > 0 ? fit_shift : 0;
+
+	return clamp_t(int, v, sISP_VAL_MIN, sISP_VAL_MAX);
+}
+
+static inline unsigned int uDIGIT_FITTING(unsigned int v, int a, int b)
+{
+	int fit_shift = uFRACTION_BITS_FITTING(a) - b;
+
+	v >>= uSHIFT;
+	v >>= fit_shift > 0 ? fit_shift : 0;
+
+	return clamp_t(unsigned int, v, uISP_VAL_MIN, uISP_VAL_MAX);
+}

Regular clamp() should work here ? all parameters are already
unsigned ints.

See above.

Regards,

Hans

quoted

 #endif /* __SH_CSS_FRAC_H */

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 29c3ea5b6e93..d53b296df504 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h

@@ -3164,7 +3164,11 @@ static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
  * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
  */
 #ifndef NET_SKB_PAD
-#define NET_SKB_PAD	max(32, L1_CACHE_BYTES)
+#if L1_CACHE_BYTES < 32
+#define NET_SKB_PAD	32
+#else
+#define NET_SKB_PAD	L1_CACHE_BYTES
+#endif
 #endif

 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
--

2.46.0