EDAC/mc: Use kzalloc_flex()
PATCH v3 1/1 EDAC/mc: Use kzalloc_flex()
Convert struct mem_ctl_info to use flex array and use the new flex array helpers to simplify initialization.
Add __counted_by for extra runtime analysis when requested.
Move counting assignment immediately after allocation as required by __counted_by.
Move memcpy to make it clear that this should have been kmemdup_array.
---
v3: move memcpy after n_layers assignment
v2: reword message slightly.
drivers/edac/edac_mc.c | 10 +++------- include/linux/edac.h | 23 ++++++++++++----------- 2 files changed, 15 insertions(+), 18 deletions(-)
drivers/edac/edac_mc.c +3 -7
@@ -203,7 +203,6 @@ static void mci_release(struct device *dev)kfree (mci->csrows); }kfree (mci->pvt_info); -kfree (mci->layers);kfree (mci); }
@@ -361,13 +360,12 @@ struct mem_ctl_info *edac_mc_alloc(unsigned int mc_num, per_rank = true; } - mci = kzalloc_obj(struct mem_ctl_info); + mci =kzalloc_flex (*mci, layers, n_layers);if (!mci)return NULL ; - mci->layers = kzalloc_objs(struct edac_mc_layer, n_layers); -if (!mci->layers) - goto error; + mci->n_layers = n_layers; +memcpy (mci->layers, layers,sizeof (*layer) * n_layers); mci->pvt_info =kzalloc (sz_pvt,GFP_KERNEL );if (!mci->pvt_info)
@@ -379,8 +377,6 @@ struct mem_ctl_info *edac_mc_alloc(unsigned int mc_num,/* setup index and various internal pointers */ mci->mc_idx = mc_num; mci->tot_dimms = tot_dimms; - mci->n_layers = n_layers; -memcpy (mci->layers, layers,sizeof (*layer) * n_layers); mci->nr_csrows = tot_csrows; mci->num_cschannel = tot_channels; mci->csbased = per_rank;
include/linux/edac.h +12 -11
@@ -541,17 +541,6 @@ struct mem_ctl_info {struct csrow_info **csrows;unsigned int nr_csrows, num_cschannel; -/* -* Memory Controller hierarchy -* -* There are basically two types of memory controller: the ones that -* sees memory sticks ("dimms"), and the ones that sees memory ranks. -* All old memory controllers enumerate memories per rank, but most -* of the recent drivers enumerate memories per DIMM, instead. -* When the memory controller is per rank, csbased is true. -*/ -unsigned int n_layers; -struct edac_mc_layer *layers;bool csbased;/*
@@ -609,6 +598,18 @@ struct mem_ctl_info {u8 fake_inject_layer[EDAC_MAX_LAYERS ];bool fake_inject_ue;u16 fake_inject_count; + +/* +* Memory Controller hierarchy +* +* There are basically two types of memory controller: the ones that +* sees memory sticks ("dimms"), and the ones that sees memory ranks. +* All old memory controllers enumerate memories per rank, but most +* of the recent drivers enumerate memories per DIMM, instead. +* When the memory controller is per rank, csbased is true. +*/ +unsigned int n_layers; + struct edac_mc_layer layers[] __counted_by(n_layers); };#define mci_for_each_dimm (mci, dimm) \
--
2.53.0
Borislav Petkov
maintainerQuoted 2 lines · Rosen Penev
Convert struct
mem_ctl_infoto use flex array and use the new flex array helpers to simplify initialization.
Meh, simplify... rather to teach an old dog like me new tricks... :)
Quoted 4 lines
Add
__counted_byfor extra runtime analysis when requested.Move counting assignment immediately after allocation as required by
__counted_by.
So far so good.
Quoted 1 line
Move memcpy to make it clear that this should have been
kmemdup_array.
But you lost me here. What do you mean with the "should have been kmemdup_array"?
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 12 lines · Borislav Petkov
Quoted 2 lines · Rosen Penev
Convert struct
mem_ctl_infoto use flex array and use the new flex array helpers to simplify initialization.Meh, simplify... rather to teach an old dog like me new tricks... :)
Quoted 4 lines
Add
__counted_byfor extra runtime analysis when requested.Move counting assignment immediately after allocation as required by
__counted_by.
Quoted 6 lines
So far so good.
Quoted 1 line
Move memcpy to make it clear that this should have been
kmemdup_array.But you lost me here. What do you mean with the "should have been
kmemdup_array"?
kzalloc_flex == kzalloc + kcalloc.
kmemdup == kzalloc + memcpy.kmemdup_array == kcalloc + memcpy.
So all together,
kzalloc + kcalloc + memcpy.
Or
kzalloc_flex + memcpy.
Placing the memcpy right after counting variable assignment enables coccinelle to transform this whole thing into a single
kmemdup_flex
call if such a thing were to eventually be introduced.
Quoted 8 lines
Thx.
--
Regards/Gruss,
Boris.
Borislav Petkov
maintainerQuoted 1 line · Rosen Penev
call if such a thing were to eventually be introduced.
Oh, hypothetical what-if. Yeah, pls drop all that gunk from commit messages pls and simply concentrate on why the patch exists.
But you don't have to for this one - I'll massage it.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 6 lines · Borislav Petkov
Quoted 1 line · Rosen Penev
call if such a thing were to eventually be introduced.
Oh, hypothetical what-if. Yeah, pls drop all that gunk from commit messages pls and simply concentrate on why the patch exists.
Not sure what you mean.
Quoted 2 lines
But you don't have to for this one - I'll massage it.
Feel free.
Quoted 8 lines
Thx.
--
Regards/Gruss,
Boris.
Borislav Petkov
maintainerQuoted 3 lines · Rosen Penev
Quoted 2 lines
Oh, hypothetical what-if. Yeah, pls drop all that gunk from commit messages pls and simply concentrate on why the patch exists.
Not sure what you mean.
I mean this: a commit message should simply state why a patch exists. Here's what I did with yours:
Convert struct mem_ctl_info to use flex array and use the new flex array helpers to enable runtime bounds checking, including annotating the array length member with __counted_by() for extra runtime analysis when requested.
Move counter assignment immediately after allocation as required by __counted_by().
Move memcpy() after the counter assignment so that it is initialized before the first reference to the flex array, as the new attribute requires.
The idea is that when one reads the commit message months, or even years from now - something we all have to do on a daily basis - it should have all the necessary information why the change was done.
And nothing else. The emphasis being on the latter part. In this case, what this should have been and what some tools can do when lines are magically ordered doesn't really matter. The change must be worth to exist for itself. In this case, runtime bounds checking, which is something we all want.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 13 lines · Borislav Petkov
Quoted 3 lines · Rosen Penev
Quoted 2 lines
Oh, hypothetical what-if. Yeah, pls drop all that gunk from commit messages pls and simply concentrate on why the patch exists.
Not sure what you mean.
I mean this: a commit message should simply state why a patch exists. Here's what I did with yours:
Convert struct
mem_ctl_infoto use flex array and use the new flex array helpers to enable runtime bounds checking, including annotating the array length member with__counted_by()for extra runtime analysis when requested.
Quoted 5 lines
Move counter assignment immediately after allocation as required by
__counted_by().Move
memcpy()after the counter assignment so that it is initialized before the first reference to the flex array, as the new attribute requires.
Looks great.
Quoted 17 lines
The idea is that when one reads the commit message months, or even years from now - something we all have to do on a daily basis - it should have all the necessary information why the change was done.
And nothing else. The emphasis being on the latter part. In this case, what this should have been and what some tools can do when lines are magically ordered doesn't really matter. The change must be worth to exist for itself. In this case, runtime bounds checking, which is something we all want.
Thx.
--
Regards/Gruss,
Boris.
Quoted 14 lines · Rosen Penev
Quoted 13 lines · Borislav Petkov
Quoted 3 lines · Rosen Penev
Quoted 2 lines
Oh, hypothetical what-if. Yeah, pls drop all that gunk from commit messages pls and simply concentrate on why the patch exists.
Not sure what you mean.
I mean this: a commit message should simply state why a patch exists. Here's what I did with yours:
Convert struct
mem_ctl_infoto use flex array and use the new flex array helpers to enable runtime bounds checking, including annotating the array length member with__counted_by()for extra runtime analysis when requested.
Quoted 15 lines
Quoted 5 lines
Move counter assignment immediately after allocation as required by
__counted_by().Move
memcpy()after the counter assignment so that it is initialized before the first reference to the flex array, as the new attribute requires.Looks great.
Quoted 9 lines
The idea is that when one reads the commit message months, or even years from now - something we all have to do on a daily basis - it should have all the necessary information why the change was done.
And nothing else. The emphasis being on the latter part. In this case, what this should have been and what some tools can do when lines are magically ordered doesn't really matter. The change must be worth to exist for itself. In this case, runtime bounds checking, which is something we all want.
Applied, thanks.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 11 lines · Borislav Petkov
Quoted 10 lines · Rosen Penev
Quoted 9 lines · Borislav Petkov
Quoted 3 lines · Rosen Penev
Quoted 2 lines
Oh, hypothetical what-if. Yeah, pls drop all that gunk from commit messages pls and simply concentrate on why the patch exists.
Not sure what you mean.
I mean this: a commit message should simply state why a patch exists. Here's what I did with yours:
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Convert struct
mem_ctl_infoto use flex array and use the new flex array helpers to enable runtime bounds checking, including annotating the array length member with__counted_by()for extra runtime analysis when requested.Move counter assignment immediately after allocation as required by
__counted_by().
This is misinformation and should be phrased differently[1]
-Gustavo
[1] https://lore.kernel.org/linux-hardening/[email protected]/
Quoted 16 lines
Quoted 13 lines
Quoted 3 lines
Move
memcpy()after the counter assignment so that it is initialized before the first reference to the flex array, as the new attribute requires.Looks great.
Quoted 9 lines
The idea is that when one reads the commit message months, or even years from now - something we all have to do on a daily basis - it should have all the necessary information why the change was done.
And nothing else. The emphasis being on the latter part. In this case, what this should have been and what some tools can do when lines are magically ordered doesn't really matter. The change must be worth to exist for itself. In this case, runtime bounds checking, which is something we all want.
Applied, thanks.
Borislav Petkov
maintainerQuoted 5 lines · Gustavo A. R. Silva
This is misinformation and should be phrased differently[1]
-Gustavo
[1] https://lore.kernel.org/linux-hardening/[email protected]/
Yah, I was wondering about that but thought there are some other requirements I don't know about.
Btw, can you do me a favor pls and summarize your very good blog article:
https://people.kernel.org/gustavoars/how-to-use-the-new-counted_by-attribute-in-c-and-linux
into a text file in Documentation/ so that we have the valid requirements for using this compiler glue properly, in the kernel?
Thx.
Now lemme zap that line.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette