kvm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Thomas Huth <thuth@redhat.com>
To: David Hildenbrand <david@redhat.com>,
	Claudio Imbrenda <imbrenda@linux.ibm.com>,
	kvm@vger.kernel.org
Cc: linux-s390@vger.kernel.org, borntraeger@de.ibm.com,
	frankja@linux.ibm.com
Subject: Re: [kvm-unit-tests PATCH v8 6/6] s390x: SCLP unit test
Date: Wed, 22 Jan 2020 13:13:08 +0100	[thread overview]
Message-ID: <f2e11864-d3d2-9ebb-64a7-b673bb0e8765@redhat.com> (raw)
In-Reply-To: <88ac3980-9ae5-24df-ca01-89a8b32b7b22@redhat.com>

On 22/01/2020 12.20, David Hildenbrand wrote:
> On 22.01.20 11:40, David Hildenbrand wrote:
>> On 22.01.20 11:39, Thomas Huth wrote:
>>> On 22/01/2020 11.32, David Hildenbrand wrote:
>>>> On 22.01.20 11:31, Thomas Huth wrote:
>>>>> On 22/01/2020 11.22, David Hildenbrand wrote:
>>>>>> On 22.01.20 11:10, David Hildenbrand wrote:
>>>>>>> On 20.01.20 19:42, Claudio Imbrenda wrote:
>>>>>>>> SCLP unit test. Testing the following:
>>>>>>>>
>>>>>>>> * Correctly ignoring instruction bits that should be ignored
>>>>>>>> * Privileged instruction check
>>>>>>>> * Check for addressing exceptions
>>>>>>>> * Specification exceptions:
>>>>>>>>   - SCCB size less than 8
>>>>>>>>   - SCCB unaligned
>>>>>>>>   - SCCB overlaps prefix or lowcore
>>>>>>>>   - SCCB address higher than 2GB
>>>>>>>> * Return codes for
>>>>>>>>   - Invalid command
>>>>>>>>   - SCCB too short (but at least 8)
>>>>>>>>   - SCCB page boundary violation
>>>>>>>>
>>>>>>>> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
>>>>>>>> Reviewed-by: Thomas Huth <thuth@redhat.com>
>>>>>>>> Acked-by: Janosch Frank <frankja@linux.ibm.com>
>>>>>>>> ---
>>>>>>>>  s390x/Makefile      |   1 +
>>>>>>>>  s390x/sclp.c        | 474 ++++++++++++++++++++++++++++++++++++++++++++
>>>>>>>>  s390x/unittests.cfg |   8 +
>>>>>>>>  3 files changed, 483 insertions(+)
>>>>>>>>  create mode 100644 s390x/sclp.c
>>>>>>>>
>>>>>>>> diff --git a/s390x/Makefile b/s390x/Makefile
>>>>>>>> index 3744372..ddb4b48 100644
>>>>>>>> --- a/s390x/Makefile
>>>>>>>> +++ b/s390x/Makefile
>>>>>>>> @@ -16,6 +16,7 @@ tests += $(TEST_DIR)/diag288.elf
>>>>>>>>  tests += $(TEST_DIR)/stsi.elf
>>>>>>>>  tests += $(TEST_DIR)/skrf.elf
>>>>>>>>  tests += $(TEST_DIR)/smp.elf
>>>>>>>> +tests += $(TEST_DIR)/sclp.elf
>>>>>>>>  tests_binary = $(patsubst %.elf,%.bin,$(tests))
>>>>>>>>  
>>>>>>>>  all: directories test_cases test_cases_binary
>>>>>>>> diff --git a/s390x/sclp.c b/s390x/sclp.c
>>>>>>>> new file mode 100644
>>>>>>>> index 0000000..215347e
>>>>>>>> --- /dev/null
>>>>>>>> +++ b/s390x/sclp.c
>>>>>>>> @@ -0,0 +1,474 @@
>>>>>>>> +/*
>>>>>>>> + * Service Call tests
>>>>>>>> + *
>>>>>>>> + * Copyright (c) 2019 IBM Corp
>>>>>>>> + *
>>>>>>>> + * Authors:
>>>>>>>> + *  Claudio Imbrenda <imbrenda@linux.ibm.com>
>>>>>>>> + *
>>>>>>>> + * This code is free software; you can redistribute it and/or modify it
>>>>>>>> + * under the terms of the GNU General Public License version 2.
>>>>>>>> + */
>>>>>>>> +
>>>>>>>> +#include <libcflat.h>
>>>>>>>> +#include <asm/page.h>
>>>>>>>> +#include <asm/asm-offsets.h>
>>>>>>>> +#include <asm/interrupt.h>
>>>>>>>> +#include <sclp.h>
>>>>>>>> +
>>>>>>>> +#define PGM_NONE	1
>>>>>>>> +#define PGM_BIT_SPEC	(1ULL << PGM_INT_CODE_SPECIFICATION)
>>>>>>>> +#define PGM_BIT_ADDR	(1ULL << PGM_INT_CODE_ADDRESSING)
>>>>>>>> +#define PGM_BIT_PRIV	(1ULL << PGM_INT_CODE_PRIVILEGED_OPERATION)
>>>>>>>> +#define MKPTR(x) ((void *)(uint64_t)(x))
>>>>>>>> +
>>>>>>>> +#define LC_SIZE (2 * PAGE_SIZE)
>>>>>>>> +
>>>>>>>> +static uint8_t pagebuf[LC_SIZE] __attribute__((aligned(LC_SIZE)));	/* scratch pages used for some tests */
>>>>>>>> +static uint8_t prefix_buf[LC_SIZE] __attribute__((aligned(LC_SIZE)));	/* temporary lowcore for test_sccb_prefix */
>>>>>>>> +static uint8_t sccb_template[PAGE_SIZE];				/* SCCB template to be used */
>>>>>>>> +static uint32_t valid_code;						/* valid command code for READ SCP INFO */
>>>>>>>> +static struct lowcore *lc;
>>>>>>>> +
>>>>>>>> +/**
>>>>>>>> + * Perform one service call, handling exceptions and interrupts.
>>>>>>>> + */
>>>>>>>> +static int sclp_service_call_test(unsigned int command, void *sccb)
>>>>>>>> +{
>>>>>>>> +	int cc;
>>>>>>>> +
>>>>>>>> +	sclp_mark_busy();
>>>>>>>> +	sclp_setup_int();
>>>>>>>> +	cc = servc(command, __pa(sccb));
>>>>>>>> +	if (lc->pgm_int_code) {
>>>>>>>> +		sclp_handle_ext();
>>>>>>>> +		return 0;
>>>>>>>> +	}
>>>>>>>> +	if (!cc)
>>>>>>>> +		sclp_wait_busy();
>>>>>>>> +	return cc;
>>>>>>>> +}
>>>>>>>> +
>>>>>>>> +/**
>>>>>>>> + * Perform one test at the given address, optionally using the SCCB template,
>>>>>>>> + * checking for the expected program interrupts and return codes.
>>>>>>>> + *
>>>>>>>> + * The parameter buf_len indicates the number of bytes of the template that
>>>>>>>> + * should be copied to the test address, and should be 0 when the test
>>>>>>>> + * address is invalid, in which case nothing is copied.
>>>>>>>> + *
>>>>>>>> + * The template is used to simplify tests where the same buffer content is
>>>>>>>> + * used many times in a row, at different addresses.
>>>>>>>> + *
>>>>>>>> + * Returns true in case of success or false in case of failure
>>>>>>>> + */
>>>>>>>> +static bool test_one_sccb(uint32_t cmd, uint8_t *addr, uint16_t buf_len, uint64_t exp_pgm, uint16_t exp_rc)
>>>>>>>> +{
>>>>>>>> +	SCCBHeader *h = (SCCBHeader *)addr;
>>>>>>>> +	int res, pgm;
>>>>>>>> +
>>>>>>>> +	/* Copy the template to the test address if needed */
>>>>>>>> +	if (buf_len)
>>>>>>>> +		memcpy(addr, sccb_template, buf_len);
>>>>>>>> +	if (exp_pgm != PGM_NONE)
>>>>>>>> +		expect_pgm_int();
>>>>>>>> +	/* perform the actual call */
>>>>>>>> +	res = sclp_service_call_test(cmd, h);
>>>>>>>> +	if (res) {
>>>>>>>> +		report_info("SCLP not ready (command %#x, address %p, cc %d)", cmd, addr, res);
>>>>>>>> +		return false;
>>>>>>>> +	}
>>>>>>>> +	pgm = clear_pgm_int();
>>>>>>>> +	/* Check if the program exception was one of the expected ones */
>>>>>>>> +	if (!((1ULL << pgm) & exp_pgm)) {
>>>>>>>> +		report_info("First failure at addr %p, buf_len %d, cmd %#x, pgm code %d",
>>>>>>>> +				addr, buf_len, cmd, pgm);
>>>>>>>> +		return false;
>>>>>>>> +	}
>>>>>>>> +	/* Check if the response code is the one expected */
>>>>>>>> +	if (exp_rc && exp_rc != h->response_code) {
>>>>>>>> +		report_info("First failure at addr %p, buf_len %d, cmd %#x, resp code %#x",
>>>>>>>> +				addr, buf_len, cmd, h->response_code);
>>>>>>>> +		return false;
>>>>>>>> +	}
>>>>>>>> +	return true;
>>>>>>>> +}
>>>>>>>> +
>>>>>>>> +/**
>>>>>>>> + * Wrapper for test_one_sccb to be used when the template should not be
>>>>>>>> + * copied and the memory address should not be touched.
>>>>>>>> + */
>>>>>>>> +static bool test_one_ro(uint32_t cmd, uint8_t *addr, uint64_t exp_pgm, uint16_t exp_rc)
>>>>>>>> +{
>>>>>>>> +	return test_one_sccb(cmd, addr, 0, exp_pgm, exp_rc);
>>>>>>>> +}
>>>>>>>> +
>>>>>>>> +/**
>>>>>>>> + * Wrapper for test_one_sccb to set up a simple SCCB template.
>>>>>>>> + *
>>>>>>>> + * The parameter sccb_len indicates the value that will be saved in the SCCB
>>>>>>>> + * length field of the SCCB, buf_len indicates the number of bytes of
>>>>>>>> + * template that need to be copied to the actual test address. In many cases
>>>>>>>> + * it's enough to clear/copy the first 8 bytes of the buffer, while the SCCB
>>>>>>>> + * itself can be larger.
>>>>>>>> + *
>>>>>>>> + * Returns true in case of success or false in case of failure
>>>>>>>> + */
>>>>>>>> +static bool test_one_simple(uint32_t cmd, uint8_t *addr, uint16_t sccb_len,
>>>>>>>> +			uint16_t buf_len, uint64_t exp_pgm, uint16_t exp_rc)
>>>>>>>> +{
>>>>>>>> +	memset(sccb_template, 0, sizeof(sccb_template));
>>>>>>>> +	((SCCBHeader *)sccb_template)->length = sccb_len;
>>>>>>>> +	return test_one_sccb(cmd, addr, buf_len, exp_pgm, exp_rc);
>>>>>>>
>>>>>>> Doing a fresh ./configure + make on RHEL7 gives me
>>>>>>>
>>>>>>> [linux1@rhkvm01 kvm-unit-tests]$ make
>>>>>>> gcc  -std=gnu99 -ffreestanding -I /home/linux1/git/kvm-unit-tests/lib -I /home/linux1/git/kvm-unit-tests/lib/s390x -I lib -O2 -march=zEC12 -fno-delete-null-pointer-checks -g -MMD -MF s390x/.sclp.d -Wall -Wwrite-strings -Wempty-body -Wuninitialized -Wignored-qualifiers -Werror  -fomit-frame-pointer    -Wno-frame-address   -fno-pic    -Wclobbered  -Wunused-but-set-parameter  -Wmissing-parameter-type  -Wold-style-declaration -Woverride-init -Wmissing-prototypes -Wstrict-prototypes   -c -o s390x/sclp.o s390x/sclp.c
>>>>>>> s390x/sclp.c: In function 'test_one_simple':
>>>>>>> s390x/sclp.c:121:2: error: dereferencing type-punned pointer will break strict-aliasing rules [-Werror=strict-aliasing]
>>>>>>>   ((SCCBHeader *)sccb_template)->length = sccb_len;
>>>>>>>   ^
>>>>>>> s390x/sclp.c: At top level:
>>>>>>> cc1: error: unrecognized command line option "-Wno-frame-address" [-Werror]
>>>>>>> cc1: all warnings being treated as errors
>>>>>>> make: *** [s390x/sclp.o] Error 1
>>>>>>
>>>>>> The following makes it work:
>>>>>>
>>>>>>
>>>>>> diff --git a/s390x/sclp.c b/s390x/sclp.c
>>>>>> index c13fa60..0b8117a 100644
>>>>>> --- a/s390x/sclp.c
>>>>>> +++ b/s390x/sclp.c
>>>>>> @@ -117,8 +117,10 @@ static bool test_one_ro(uint32_t cmd, uint8_t *addr, uint64_t exp_pgm, uint16_t
>>>>>>  static bool test_one_simple(uint32_t cmd, uint8_t *addr, uint16_t sccb_len,
>>>>>>                         uint16_t buf_len, uint64_t exp_pgm, uint16_t exp_rc)
>>>>>>  {
>>>>>> +       SCCBHeader *header = (void *)sccb_template;
>>>>>> +
>>>>>>         memset(sccb_template, 0, sizeof(sccb_template));
>>>>>> -       ((SCCBHeader *)sccb_template)->length = sccb_len;
>>>>>> +       header->length = sccb_len;
>>>>>
>>>>> While that might silence the compiler warning, we still might get
>>>>> aliasing problems here, I think.
>>>>> The right way to solve this problem is to turn sccb_template into a
>>>>> union of the various structs/arrays that you want to use and then access
>>>>> the fields through the union instead ("type-punning through union").
>>>>
>>>> We do have the exact same thing in lib/s390x/sclp.c already, no?
>>>
>>> Maybe we should carefully check that code, too...
>>>
>>>> Especially, new compilers don't seem to care?
>>>
>>> I've seen horrible bugs due to these aliasing problems in the past -
>>> without compiler warnings showing up! Certain versions of GCC assume
>>> that they can re-order code with pointers that point to types of
>>> different sizes, i.e. in the above example, I think they could assume
>>> that they could re-order the memset() and the header->length = ... line.
>>> I'd feel better if we play safe and use a union here.
>>
>> Should we simply allow type-punning?
> 
> For now I squashed
> 
> diff --git a/s390x/sclp.c b/s390x/sclp.c
> index c13fa60..7d92bf3 100644
> --- a/s390x/sclp.c
> +++ b/s390x/sclp.c
> @@ -26,7 +26,12 @@
> 
>  static uint8_t pagebuf[LC_SIZE] __attribute__((aligned(LC_SIZE)));	/*
> scratch pages used for some tests */
>  static uint8_t prefix_buf[LC_SIZE] __attribute__((aligned(LC_SIZE)));
> /* temporary lowcore for test_sccb_prefix */
> -static uint8_t sccb_template[PAGE_SIZE];				/* SCCB template to be used */
> +/* SCCB template to be used */
> +static union {
> +	uint8_t raw[PAGE_SIZE];
> +	SCCBHeader header;
> +	WriteEventData data;
> +} sccb_template;
>  static uint32_t valid_code;						/* valid command code for READ SCP INFO */
>  static struct lowcore *lc;
> 
> @@ -69,7 +74,7 @@ static bool test_one_sccb(uint32_t cmd, uint8_t *addr,
> uint16_t buf_len, uint64_
> 
>  	/* Copy the template to the test address if needed */
>  	if (buf_len)
> -		memcpy(addr, sccb_template, buf_len);
> +		memcpy(addr, sccb_template.raw, buf_len);
>  	if (exp_pgm != PGM_NONE)
>  		expect_pgm_int();
>  	/* perform the actual call */
> @@ -117,8 +122,8 @@ static bool test_one_ro(uint32_t cmd, uint8_t *addr,
> uint64_t exp_pgm, uint16_t
>  static bool test_one_simple(uint32_t cmd, uint8_t *addr, uint16_t sccb_len,
>  			uint16_t buf_len, uint64_t exp_pgm, uint16_t exp_rc)
>  {
> -	memset(sccb_template, 0, sizeof(sccb_template));
> -	((SCCBHeader *)sccb_template)->length = sccb_len;
> +	memset(sccb_template.raw, 0, sizeof(sccb_template.raw));
> +	sccb_template.header.length = sccb_len;
>  	return test_one_sccb(cmd, addr, buf_len, exp_pgm, exp_rc);
>  }
> 
> @@ -299,10 +304,10 @@ static void test_boundary(void)
>  {
>  	const uint32_t cmd = SCLP_CMD_WRITE_EVENT_DATA;
>  	const uint16_t res = SCLP_RC_SCCB_BOUNDARY_VIOLATION;
> -	WriteEventData *sccb = (WriteEventData *)sccb_template;
> +	WriteEventData *sccb = &sccb_template.data;
>  	int len, offset;
> 
> -	memset(sccb_template, 0, sizeof(sccb_template));
> +	memset(sccb_template.raw, 0, sizeof(sccb_template.raw));
>  	sccb->h.function_code = SCLP_FC_NORMAL_WRITE;
>  	for (len = 32; len <= 4096; len++) {
>  		offset = len & 7 ? len & ~7 : len - 8;
> @@ -323,10 +328,10 @@ static void test_toolong(void)
>  {
>  	const uint32_t cmd = SCLP_CMD_WRITE_EVENT_DATA;
>  	const uint16_t res = SCLP_RC_SCCB_BOUNDARY_VIOLATION;
> -	WriteEventData *sccb = (WriteEventData *)sccb_template;
> +	WriteEventData *sccb = &sccb_template.data;
>  	int len;
> 
> -	memset(sccb_template, 0, sizeof(sccb_template));
> +	memset(sccb_template.raw, 0, sizeof(sccb_template.raw));
>  	sccb->h.function_code = SCLP_FC_NORMAL_WRITE;
>  	for (len = 4097; len < 8192; len++) {
>  		sccb->h.length = len;

Thanks, that looks better now!

 Thomas


  reply	other threads:[~2020-01-22 12:13 UTC|newest]

Thread overview: 27+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-01-20 18:42 [kvm-unit-tests PATCH v8 0/6] s390x: SCLP Unit test Claudio Imbrenda
2020-01-20 18:42 ` [kvm-unit-tests PATCH v8 1/6] s390x: export sclp_setup_int Claudio Imbrenda
2020-01-20 18:42 ` [kvm-unit-tests PATCH v8 2/6] s390x: sclp: add service call instruction wrapper Claudio Imbrenda
2020-01-20 18:42 ` [kvm-unit-tests PATCH v8 3/6] s390x: lib: fix stfl wrapper asm Claudio Imbrenda
2020-01-21  6:15   ` Thomas Huth
2020-01-21  9:17   ` David Hildenbrand
2020-01-21 12:42   ` Janosch Frank
2020-01-20 18:42 ` [kvm-unit-tests PATCH v8 4/6] s390x: lib: add SPX and STPX instruction wrapper Claudio Imbrenda
2020-01-20 18:42 ` [kvm-unit-tests PATCH v8 5/6] s390x: lib: fix program interrupt handler if sclp_busy was set Claudio Imbrenda
2020-01-21  6:19   ` Thomas Huth
2020-01-21  9:18   ` David Hildenbrand
2020-01-20 18:42 ` [kvm-unit-tests PATCH v8 6/6] s390x: SCLP unit test Claudio Imbrenda
2020-01-22  9:55   ` David Hildenbrand
2020-01-22 10:10   ` David Hildenbrand
2020-01-22 10:22     ` David Hildenbrand
2020-01-22 10:31       ` Thomas Huth
2020-01-22 10:32         ` David Hildenbrand
2020-01-22 10:39           ` Thomas Huth
2020-01-22 10:40             ` David Hildenbrand
2020-01-22 11:20               ` David Hildenbrand
2020-01-22 12:13                 ` Thomas Huth [this message]
2020-01-22 12:16               ` Thomas Huth
2020-01-22 14:15                 ` strict aliasing in kvm-unit-tests (was: Re: [kvm-unit-tests PATCH v8 6/6] s390x: SCLP unit test) Thomas Huth
2020-01-22 14:42                   ` Paolo Bonzini
2020-01-22 15:01                     ` Laurent Vivier
2020-01-22  9:55 ` [kvm-unit-tests PATCH v8 0/6] s390x: SCLP Unit test David Hildenbrand
2020-01-22 10:02   ` David Hildenbrand

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=f2e11864-d3d2-9ebb-64a7-b673bb0e8765@redhat.com \
    --to=thuth@redhat.com \
    --cc=borntraeger@de.ibm.com \
    --cc=david@redhat.com \
    --cc=frankja@linux.ibm.com \
    --cc=imbrenda@linux.ibm.com \
    --cc=kvm@vger.kernel.org \
    --cc=linux-s390@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).