From 1a251f52cfdc417c84411a056bc142cbd77baef4 Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Sun, 28 Jul 2024 15:49:18 -0700 Subject: minmax: make generic MIN() and MAX() macros available everywhere This just standardizes the use of MIN() and MAX() macros, with the very traditional semantics. The goal is to use these for C constant expressions and for top-level / static initializers, and so be able to simplify the min()/max() macros. These macro names were used by various kernel code - they are very traditional, after all - and all such users have been fixed up, with a few different approaches: - trivial duplicated macro definitions have been removed Note that 'trivial' here means that it's obviously kernel code that already included all the major kernel headers, and thus gets the new generic MIN/MAX macros automatically. - non-trivial duplicated macro definitions are guarded with #ifndef This is the "yes, they define their own versions, but no, the include situation is not entirely obvious, and maybe they don't get the generic version automatically" case. - strange use case #1 A couple of drivers decided that the way they want to describe their versioning is with #define MAJ 1 #define MIN 2 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) which adds zero value and I just did my Alexander the Great impersonation, and rewrote that pointless Gordian knot as #define DRV_VERSION "1.2" instead. - strange use case #2 A couple of drivers thought that it's a good idea to have a random 'MIN' or 'MAX' define for a value or index into a table, rather than the traditional macro that takes arguments. These values were re-written as C enum's instead. The new function-line macros only expand when followed by an open parenthesis, and thus don't clash with enum use. Happily, there weren't really all that many of these cases, and a lot of users already had the pattern of using '#ifndef' guarding (or in one case just using '#undef MIN') before defining their own private version that does the same thing. I left such cases alone. Cc: David Laight Cc: Lorenzo Stoakes Signed-off-by: Linus Torvalds --- include/linux/minmax.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'include/linux/minmax.h') diff --git a/include/linux/minmax.h b/include/linux/minmax.h index 9c2848abc804..fc384714da45 100644 --- a/include/linux/minmax.h +++ b/include/linux/minmax.h @@ -277,6 +277,8 @@ static inline bool in_range32(u32 val, u32 start, u32 len) * Use these carefully: no type checking, and uses the arguments * multiple times. Use for obvious constants only. */ +#define MIN(a,b) __cmp(min,a,b) +#define MAX(a,b) __cmp(max,a,b) #define MIN_T(type,a,b) __cmp(min,(type)(a),(type)(b)) #define MAX_T(type,a,b) __cmp(max,(type)(a),(type)(b)) -- cgit v1.2.3 From dc1c8034e31b14a2e5e212104ec508aec44ce1b9 Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Sun, 28 Jul 2024 20:24:12 -0700 Subject: minmax: simplify min()/max()/clamp() implementation Now that we no longer have any C constant expression contexts (ie array size declarations or static initializers) that use min() or max(), we can simpify the implementation by not having to worry about the result staying as a C constant expression. So now we can unconditionally just use temporary variables of the right type, and get rid of the excessive expansion that used to come from the use of __builtin_choose_expr(__is_constexpr(...), .. to pick the specialized code for constant expressions. Another expansion simplification is to pass the temporary variables (in addition to the original expression) to our __types_ok() macro. That may superficially look like it complicates the macro, but when we only want the type of the expression, expanding the temporary variable names is much simpler and smaller than expanding the potentially complicated original expression. As a result, on my machine, doing a $ time make drivers/staging/media/atomisp/pci/isp/kernels/ynr/ynr_1.0/ia_css_ynr.host.i goes from real 0m16.621s user 0m15.360s sys 0m1.221s to real 0m2.532s user 0m2.091s sys 0m0.452s because the token expansion goes down dramatically. In particular, the longest line expansion (which was line 71 of that 'ia_css_ynr.host.c' file) shrinks from 23,338kB (yes, 23MB for one single line) to "just" 1,444kB (now "only" 1.4MB). And yes, that line is still the line from hell, because it's doing multiple levels of "min()/max()" expansion thanks to some of them being hidden inside the uDIGIT_FITTING() macro. Lorenzo has a nice cleanup patch that makes that driver use inline functions instead of macros for sDIGIT_FITTING() and uDIGIT_FITTING(), which will fix that line once and for all, but the 16-fold reduction in this case does show why we need to simplify these helpers. Cc: David Laight Cc: Lorenzo Stoakes Signed-off-by: Linus Torvalds --- include/linux/minmax.h | 43 ++++++++++++++++++++----------------------- 1 file changed, 20 insertions(+), 23 deletions(-) (limited to 'include/linux/minmax.h') diff --git a/include/linux/minmax.h b/include/linux/minmax.h index fc384714da45..e3e4353df983 100644 --- a/include/linux/minmax.h +++ b/include/linux/minmax.h @@ -35,10 +35,10 @@ #define __is_noneg_int(x) \ (__builtin_choose_expr(__is_constexpr(x) && __is_signed(x), x, -1) >= 0) -#define __types_ok(x, y) \ - (__is_signed(x) == __is_signed(y) || \ - __is_signed((x) + 0) == __is_signed((y) + 0) || \ - __is_noneg_int(x) || __is_noneg_int(y)) +#define __types_ok(x, y, ux, uy) \ + (__is_signed(ux) == __is_signed(uy) || \ + __is_signed((ux) + 0) == __is_signed((uy) + 0) || \ + __is_noneg_int(x) || __is_noneg_int(y)) #define __cmp_op_min < #define __cmp_op_max > @@ -51,34 +51,31 @@ #define __cmp_once(op, type, x, y) \ __cmp_once_unique(op, type, x, y, __UNIQUE_ID(x_), __UNIQUE_ID(y_)) -#define __careful_cmp_once(op, x, y) ({ \ - static_assert(__types_ok(x, y), \ +#define __careful_cmp_once(op, x, y, ux, uy) ({ \ + __auto_type ux = (x); __auto_type uy = (y); \ + static_assert(__types_ok(x, y, ux, uy), \ #op "(" #x ", " #y ") signedness error, fix types or consider u" #op "() before " #op "_t()"); \ - __cmp_once(op, __auto_type, x, y); }) + __cmp(op, ux, uy); }) -#define __careful_cmp(op, x, y) \ - __builtin_choose_expr(__is_constexpr((x) - (y)), \ - __cmp(op, x, y), __careful_cmp_once(op, x, y)) +#define __careful_cmp(op, x, y) \ + __careful_cmp_once(op, x, y, __UNIQUE_ID(x_), __UNIQUE_ID(y_)) #define __clamp(val, lo, hi) \ ((val) >= (hi) ? (hi) : ((val) <= (lo) ? (lo) : (val))) -#define __clamp_once(val, lo, hi, unique_val, unique_lo, unique_hi) ({ \ - typeof(val) unique_val = (val); \ - typeof(lo) unique_lo = (lo); \ - typeof(hi) unique_hi = (hi); \ +#define __clamp_once(val, lo, hi, uval, ulo, uhi) ({ \ + __auto_type uval = (val); \ + __auto_type ulo = (lo); \ + __auto_type uhi = (hi); \ static_assert(__builtin_choose_expr(__is_constexpr((lo) > (hi)), \ (lo) <= (hi), true), \ "clamp() low limit " #lo " greater than high limit " #hi); \ - static_assert(__types_ok(val, lo), "clamp() 'lo' signedness error"); \ - static_assert(__types_ok(val, hi), "clamp() 'hi' signedness error"); \ - __clamp(unique_val, unique_lo, unique_hi); }) - -#define __careful_clamp(val, lo, hi) ({ \ - __builtin_choose_expr(__is_constexpr((val) - (lo) + (hi)), \ - __clamp(val, lo, hi), \ - __clamp_once(val, lo, hi, __UNIQUE_ID(__val), \ - __UNIQUE_ID(__lo), __UNIQUE_ID(__hi))); }) + static_assert(__types_ok(uval, lo, uval, ulo), "clamp() 'lo' signedness error"); \ + static_assert(__types_ok(uval, hi, uval, uhi), "clamp() 'hi' signedness error"); \ + __clamp(uval, ulo, uhi); }) + +#define __careful_clamp(val, lo, hi) \ + __clamp_once(val, lo, hi, __UNIQUE_ID(v_), __UNIQUE_ID(l_), __UNIQUE_ID(h_)) /** * min - return minimum of two values of the same or compatible types -- cgit v1.2.3 From 22f5468731491e53356ba7c028f0fdea20b18e2c Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Tue, 30 Jul 2024 10:36:47 -0700 Subject: minmax: improve macro expansion and type checking This clarifies the rules for min()/max()/clamp() type checking and makes them a much more efficient macro expansion. In particular, we now look at the type and range of the inputs to see whether they work together, generating a mask of acceptable comparisons, and then just verifying that the inputs have a shared case: - an expression with a signed type can be used for (1) signed comparisons (2) unsigned comparisons if it is statically known to have a non-negative value - an expression with an unsigned type can be used for (3) unsigned comparison (4) signed comparisons if the type is smaller than 'int' and thus the C integer promotion rules will make it signed anyway Here rule (1) and (3) are obvious, and rule (2) is important in order to allow obvious trivial constants to be used together with unsigned values. Rule (4) is not necessarily a good idea, but matches what we used to do, and we have extant cases of this situation in the kernel. Notably with bcachefs having an expression like min(bch2_bucket_sectors_dirty(a), ca->mi.bucket_size) where bch2_bucket_sectors_dirty() returns an 's64', and 'ca->mi.bucket_size' is of type 'u16'. Technically that bcachefs comparison is clearly sensible on a C type level, because the 'u16' will go through the normal C integer promotion, and become 'int', and then we're comparing two signed values and everything looks sane. However, it's not entirely clear that a 'min(s64,u16)' operation makes a lot of conceptual sense, and it's possible that we will remove rule (4). After all, the _reason_ we have these complicated type checks is exactly that the C type promotion rules are not very intuitive. But at least for now the rule is in place for backwards compatibility. Also note that rule (2) existed before, but is hugely relaxed by this commit. It used to be true only for the simplest compile-time non-negative integer constants. The new macro model will allow cases where the compiler can trivially see that an expression is non-negative even if it isn't necessarily a constant. For example, the amdgpu driver does min_t(size_t, sizeof(fru_info->serial), pia[addr] & 0x3F)); because our old 'min()' macro would see that 'pia[addr] & 0x3F' is of type 'int' and clearly not a C constant expression, so doing a 'min()' with a 'size_t' is a signedness violation. Our new 'min()' macro still sees that 'pia[addr] & 0x3F' is of type 'int', but is smart enough to also see that it is clearly non-negative, and thus would allow that case without any complaints. Cc: Arnd Bergmann Cc: David Laight Cc: Lorenzo Stoakes Signed-off-by: Linus Torvalds --- include/linux/minmax.h | 74 ++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 59 insertions(+), 15 deletions(-) (limited to 'include/linux/minmax.h') diff --git a/include/linux/minmax.h b/include/linux/minmax.h index e3e4353df983..41da6f85a407 100644 --- a/include/linux/minmax.h +++ b/include/linux/minmax.h @@ -26,19 +26,63 @@ #define __typecheck(x, y) \ (!!(sizeof((typeof(x) *)1 == (typeof(y) *)1))) -/* is_signed_type() isn't a constexpr for pointer types */ -#define __is_signed(x) \ - __builtin_choose_expr(__is_constexpr(is_signed_type(typeof(x))), \ - is_signed_type(typeof(x)), 0) +/* + * __sign_use for integer expressions: + * bit #0 set if ok for unsigned comparisons + * bit #1 set if ok for signed comparisons + * + * In particular, statically non-negative signed integer + * expressions are ok for both. + * + * NOTE! Unsigned types smaller than 'int' are implicitly + * converted to 'int' in expressions, and are accepted for + * signed conversions for now. This is debatable. + * + * Note that 'x' is the original expression, and 'ux' is + * the unique variable that contains the value. + * + * We use 'ux' for pure type checking, and 'x' for when + * we need to look at the value (but without evaluating + * it for side effects! Careful to only ever evaluate it + * with sizeof() or __builtin_constant_p() etc). + * + * Pointers end up being checked by the normal C type + * rules at the actual comparison, and these expressions + * only need to be careful to not cause warnings for + * pointer use. + */ +#define __signed_type_use(x,ux) (2+__is_nonneg(x,ux)) +#define __unsigned_type_use(x,ux) (1+2*(sizeof(ux)<4)) +#define __sign_use(x,ux) (is_signed_type(typeof(ux))? \ + __signed_type_use(x,ux):__unsigned_type_use(x,ux)) + +/* + * To avoid warnings about casting pointers to integers + * of different sizes, we need that special sign type. + * + * On 64-bit we can just always use 'long', since any + * integer or pointer type can just be cast to that. + * + * This does not work for 128-bit signed integers since + * the cast would truncate them, but we do not use s128 + * types in the kernel (we do use 'u128', but they will + * be handled by the !is_signed_type() case). + * + * NOTE! The cast is there only to avoid any warnings + * from when values that aren't signed integer types. + */ +#ifdef CONFIG_64BIT + #define __signed_type(ux) long +#else + #define __signed_type(ux) typeof(__builtin_choose_expr(sizeof(ux)>4,1LL,1L)) +#endif +#define __is_nonneg(x,ux) statically_true((__signed_type(ux))(x)>=0) -/* True for a non-negative signed int constant */ -#define __is_noneg_int(x) \ - (__builtin_choose_expr(__is_constexpr(x) && __is_signed(x), x, -1) >= 0) +#define __types_ok(x,y,ux,uy) \ + (__sign_use(x,ux) & __sign_use(y,uy)) -#define __types_ok(x, y, ux, uy) \ - (__is_signed(ux) == __is_signed(uy) || \ - __is_signed((ux) + 0) == __is_signed((uy) + 0) || \ - __is_noneg_int(x) || __is_noneg_int(y)) +#define __types_ok3(x,y,z,ux,uy,uz) \ + (__sign_use(x,ux) & __sign_use(y,uy) & __sign_use(z,uz)) #define __cmp_op_min < #define __cmp_op_max > @@ -53,8 +97,8 @@ #define __careful_cmp_once(op, x, y, ux, uy) ({ \ __auto_type ux = (x); __auto_type uy = (y); \ - static_assert(__types_ok(x, y, ux, uy), \ - #op "(" #x ", " #y ") signedness error, fix types or consider u" #op "() before " #op "_t()"); \ + BUILD_BUG_ON_MSG(!__types_ok(x,y,ux,uy), \ + #op"("#x", "#y") signedness error"); \ __cmp(op, ux, uy); }) #define __careful_cmp(op, x, y) \ @@ -70,8 +114,8 @@ static_assert(__builtin_choose_expr(__is_constexpr((lo) > (hi)), \ (lo) <= (hi), true), \ "clamp() low limit " #lo " greater than high limit " #hi); \ - static_assert(__types_ok(uval, lo, uval, ulo), "clamp() 'lo' signedness error"); \ - static_assert(__types_ok(uval, hi, uval, uhi), "clamp() 'hi' signedness error"); \ + BUILD_BUG_ON_MSG(!__types_ok3(val,lo,hi,uval,ulo,uhi), \ + "clamp("#val", "#lo", "#hi") signedness error"); \ __clamp(uval, ulo, uhi); }) #define __careful_clamp(val, lo, hi) \ -- cgit v1.2.3 From 21b136cc63d2a9ddd60d4699552b69c214b32964 Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Tue, 30 Jul 2024 15:44:16 -0700 Subject: minmax: fix up min3() and max3() too David Laight pointed out that we should deal with the min3() and max3() mess too, which still does excessive expansion. And our current macros are actually rather broken. In particular, the macros did this: #define min3(x, y, z) min((typeof(x))min(x, y), z) #define max3(x, y, z) max((typeof(x))max(x, y), z) and that not only is a nested expansion of possibly very complex arguments with all that involves, the typing with that "typeof()" cast is completely wrong. For example, imagine what happens in max3() if 'x' happens to be a 'unsigned char', but 'y' and 'z' are 'unsigned long'. The types are compatible, and there's no warning - but the result is just random garbage. No, I don't think we've ever hit that issue in practice, but since we now have sane infrastructure for doing this right, let's just use it. It fixes any excessive expansion, and also avoids these kinds of broken type issues. Requested-by: David Laight Acked-by: Arnd Bergmann Signed-off-by: Linus Torvalds --- include/linux/minmax.h | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) (limited to 'include/linux/minmax.h') diff --git a/include/linux/minmax.h b/include/linux/minmax.h index 41da6f85a407..98008dd92153 100644 --- a/include/linux/minmax.h +++ b/include/linux/minmax.h @@ -152,13 +152,20 @@ #define umax(x, y) \ __careful_cmp(max, (x) + 0u + 0ul + 0ull, (y) + 0u + 0ul + 0ull) +#define __careful_op3(op, x, y, z, ux, uy, uz) ({ \ + __auto_type ux = (x); __auto_type uy = (y);__auto_type uz = (z);\ + BUILD_BUG_ON_MSG(!__types_ok3(x,y,z,ux,uy,uz), \ + #op"3("#x", "#y", "#z") signedness error"); \ + __cmp(op, ux, __cmp(op, uy, uz)); }) + /** * min3 - return minimum of three values * @x: first value * @y: second value * @z: third value */ -#define min3(x, y, z) min((typeof(x))min(x, y), z) +#define min3(x, y, z) \ + __careful_op3(min, x, y, z, __UNIQUE_ID(x_), __UNIQUE_ID(y_), __UNIQUE_ID(z_)) /** * max3 - return maximum of three values @@ -166,7 +173,8 @@ * @y: second value * @z: third value */ -#define max3(x, y, z) max((typeof(x))max(x, y), z) +#define max3(x, y, z) \ + __careful_op3(max, x, y, z, __UNIQUE_ID(x_), __UNIQUE_ID(y_), __UNIQUE_ID(z_)) /** * min_not_zero - return the minimum that is _not_ zero, unless both are zero -- cgit v1.2.3