[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <2a3a8e6a-e636-268b-c2af-0404bc3cd388@huawei.com>
Date: Tue, 9 Oct 2018 22:04:58 +0800
From: YueHaibing <yuehaibing@...wei.com>
To: Michael Ellerman <mpe@...erman.id.au>, <benh@...nel.crashing.org>,
<paulus@...ba.org>, <nfont@...ux.vnet.ibm.com>
CC: <linux-kernel@...r.kernel.org>, <linuxppc-dev@...ts.ozlabs.org>
Subject: Re: [PATCH v2 -next] powerpc/pseries/memory-hotplug: Fix return value
type of find_aa_index
On 2018/10/9 15:00, Michael Ellerman wrote:
> YueHaibing <yuehaibing@...wei.com> writes:
>> 'aa_index' is defined as an unsigned value, but find_aa_index
>> may return -1 when dlpar_clone_property fails. So we use an rc
>> value to track the validation of finding the aa_index instead
>> of the 'aa_index' value itself
>>
>> Fixes: c05a5a40969e ("powerpc/pseries: Dynamic add entires to associativity lookup array")
>> Signed-off-by: YueHaibing <yuehaibing@...wei.com>
>> ---
>> v2: use 'rc' track the validation of aa_index
>
> Thanks for sending a v2, some more comments ...
>
>> diff --git a/arch/powerpc/platforms/pseries/hotplug-memory.c b/arch/powerpc/platforms/pseries/hotplug-memory.c
>> index 9a15d39..796e68b 100644
>> --- a/arch/powerpc/platforms/pseries/hotplug-memory.c
>> +++ b/arch/powerpc/platforms/pseries/hotplug-memory.c
>> @@ -101,13 +101,12 @@ static struct property *dlpar_clone_property(struct property *prop,
>> return new_prop;
>> }
>>
>> -static u32 find_aa_index(struct device_node *dr_node,
>> - struct property *ala_prop, const u32 *lmb_assoc)
>> +static int find_aa_index(struct device_node *dr_node, struct property *ala_prop,
>> + const u32 *lmb_assoc, u32 *aa_index)
>> {
>> u32 *assoc_arrays;
>> - u32 aa_index;
>> int aa_arrays, aa_array_entries, aa_array_sz;
>> - int i, index;
>> + int i, index, rc = -1;
>
> It's preferable to leave rc uninitialised until we actually need to
> initialise it, that gives the compiler the chance to warn us if we use
> it inadvertently before that.
>
>>
>> /*
>> * The ibm,associativity-lookup-arrays property is defined to be
>> @@ -121,18 +120,18 @@ static u32 find_aa_index(struct device_node *dr_node,
>> aa_array_entries = be32_to_cpu(assoc_arrays[1]);
>> aa_array_sz = aa_array_entries * sizeof(u32);
>>
>> - aa_index = -1;
>
> So that would be here:
> rc = -1;
>
> But ..
>
>> for (i = 0; i < aa_arrays; i++) {
>> index = (i * aa_array_entries) + 2;
>>
>> if (memcmp(&assoc_arrays[index], &lmb_assoc[1], aa_array_sz))
>> continue;
>>
>> - aa_index = i;
>> + *aa_index = i;
>> + rc = 0;
>> break;
>> }
>
> The 'rc' variable is basically a boolean now, it means "we found something".
>
> And all we do with it in the found case (rc = 0) is test it below and return.
>
> So can't we just return directly in the for loop above, rather than breaking?
>
> In which case we don't need the rc variable at all.
>
> And the whole function may as well return bool, rather than int.
>
> Does that make sense?
Yes, will do that in v3.
>
> cheers
>
>> - if (aa_index == -1) {
>> + if (rc == -1) {
>> struct property *new_prop;
>> u32 new_prop_size;
>>
>> @@ -157,10 +156,11 @@ static u32 find_aa_index(struct device_node *dr_node,
>> * number of entries - 1 since we added its associativity
>> * to the end of the lookup array.
>> */
>> - aa_index = be32_to_cpu(assoc_arrays[0]) - 1;
>> + *aa_index = be32_to_cpu(assoc_arrays[0]) - 1;
>> + rc = 0;
>> }
>>
>> - return aa_index;
>> + return rc;
>> }
>>
>> static int update_lmb_associativity_index(struct drmem_lmb *lmb)
>
> .
>
Powered by blists - more mailing lists