Re: [PATCH V3 6/6] crypto/nx: Add P9 NX support for 842 compression engine

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Sat, Sep 2, 2017 at 4:40 AM, Haren Myneni <haren@xxxxxxxxxxxxxxxxxx> wrote:
> On 08/29/2017 06:58 AM, Dan Streetman wrote:
>> On Sat, Jul 22, 2017 at 1:01 AM, Haren Myneni <haren@xxxxxxxxxxxxxxxxxx> wrote:
>>>
>>> This patch adds P9 NX support for 842 compression engine. Virtual
>>> Accelerator Switchboard (VAS) is used to access 842 engine on P9.
>>>
>>> For each NX engine per chip, setup receive window using
>>> vas_rx_win_open() which configures RxFIFo with FIFO address, lpid,
>>> pid and tid values. This unique (lpid, pid, tid) combination will
>>> be used to identify the target engine.
>>>
>>> For crypto open request, open send window on the NX engine for
>>> the corresponding chip / cpu where the open request is executed.
>>> This send window will be closed upon crypto close request.
>>>
>>> NX provides high and normal priority FIFOs. For compression /
>>> decompression requests, we use only hight priority FIFOs in kernel.
>>>
>>> Each NX request will be communicated to VAS using copy/paste
>>> instructions with vas_copy_crb() / vas_paste_crb() functions.
>>>
>>> Signed-off-by: Haren Myneni <haren@xxxxxxxxxx>
>>> ---
>>>  drivers/crypto/nx/Kconfig          |   1 +
>>>  drivers/crypto/nx/nx-842-powernv.c | 375 ++++++++++++++++++++++++++++++++++++-
>>>  drivers/crypto/nx/nx-842.c         |   2 +-
>>>  3 files changed, 371 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/drivers/crypto/nx/Kconfig b/drivers/crypto/nx/Kconfig
>>> index ad7552a6998c..cd5dda9c48f4 100644
>>> --- a/drivers/crypto/nx/Kconfig
>>> +++ b/drivers/crypto/nx/Kconfig
>>> @@ -38,6 +38,7 @@ config CRYPTO_DEV_NX_COMPRESS_PSERIES
>>>  config CRYPTO_DEV_NX_COMPRESS_POWERNV
>>>         tristate "Compression acceleration support on PowerNV platform"
>>>         depends on PPC_POWERNV
>>> +       depends on PPC_VAS
>>>         default y
>>>         help
>>>           Support for PowerPC Nest (NX) compression acceleration. This
>>> diff --git a/drivers/crypto/nx/nx-842-powernv.c b/drivers/crypto/nx/nx-842-powernv.c
>>> index c0dd4c7e17d3..13089a0b9dfa 100644
>>> --- a/drivers/crypto/nx/nx-842-powernv.c
>>> +++ b/drivers/crypto/nx/nx-842-powernv.c
>>> @@ -23,6 +23,7 @@
>>>  #include <asm/prom.h>
>>>  #include <asm/icswx.h>
>>>  #include <asm/vas.h>
>>> +#include <asm/reg.h>
>>>
>>>  MODULE_LICENSE("GPL");
>>>  MODULE_AUTHOR("Dan Streetman <ddstreet@xxxxxxxx>");
>>> @@ -32,6 +33,9 @@ MODULE_ALIAS_CRYPTO("842-nx");
>>>
>>>  #define WORKMEM_ALIGN  (CRB_ALIGN)
>>>  #define CSB_WAIT_MAX   (5000) /* ms */
>>> +#define VAS_RETRIES    (10)
>>> +/* # of requests allowed per RxFIFO at a time. 0 for unlimited */
>>> +#define MAX_CREDITS_PER_RXFIFO (1024)
>>>
>>>  struct nx842_workmem {
>>>         /* Below fields must be properly aligned */
>>> @@ -42,16 +46,27 @@ struct nx842_workmem {
>>>
>>>         ktime_t start;
>>>
>>> +       struct vas_window *txwin;       /* Used with VAS function */
>>>         char padding[WORKMEM_ALIGN]; /* unused, to allow alignment */
>>>  } __packed __aligned(WORKMEM_ALIGN);
>>>
>>>  struct nx842_coproc {
>>>         unsigned int chip_id;
>>>         unsigned int ct;
>>> -       unsigned int ci;
>>> +       unsigned int ci;        /* Coprocessor instance, used with icswx */
>>> +       struct {
>>> +               struct vas_window *rxwin;
>>> +               int id;
>>> +       } vas;
>>>         struct list_head list;
>>>  };
>>>
>>> +/*
>>> + * Send the request to NX engine on the chip for the corresponding CPU
>>> + * where the process is executing. Use with VAS function.
>>> + */
>>> +static DEFINE_PER_CPU(struct nx842_coproc *, coproc_inst);
>>> +
>>>  /* no cpu hotplug on powernv, so this list never changes after init */
>>>  static LIST_HEAD(nx842_coprocs);
>>>  static unsigned int nx842_ct;  /* used in icswx function */
>>> @@ -513,6 +528,105 @@ static int nx842_exec_icswx(const unsigned char *in, unsigned int inlen,
>>>  }
>>>
>>>  /**
>>> + * nx842_exec_vas - compress/decompress data using the 842 algorithm
>>> + *
>>> + * (De)compression provided by the NX842 coprocessor on IBM PowerNV systems.
>>> + * This compresses or decompresses the provided input buffer into the provided
>>> + * output buffer.
>>> + *
>>> + * Upon return from this function @outlen contains the length of the
>>> + * output data.  If there is an error then @outlen will be 0 and an
>>> + * error will be specified by the return code from this function.
>>> + *
>>> + * The @workmem buffer should only be used by one function call at a time.
>>> + *
>>> + * @in: input buffer pointer
>>> + * @inlen: input buffer size
>>> + * @out: output buffer pointer
>>> + * @outlenp: output buffer size pointer
>>> + * @workmem: working memory buffer pointer, size determined by
>>> + *           nx842_powernv_driver.workmem_size
>>> + * @fc: function code, see CCW Function Codes in nx-842.h
>>> + *
>>> + * Returns:
>>> + *   0         Success, output of length @outlenp stored in the buffer
>>> + *             at @out
>>> + *   -ENODEV   Hardware unavailable
>>> + *   -ENOSPC   Output buffer is to small
>>> + *   -EMSGSIZE Input buffer too large
>>> + *   -EINVAL   buffer constraints do not fix nx842_constraints
>>> + *   -EPROTO   hardware error during operation
>>> + *   -ETIMEDOUT        hardware did not complete operation in reasonable time
>>> + *   -EINTR    operation was aborted
>>> + */
>>> +static int nx842_exec_vas(const unsigned char *in, unsigned int inlen,
>>> +                                 unsigned char *out, unsigned int *outlenp,
>>> +                                 void *workmem, int fc)
>>> +{
>>> +       struct coprocessor_request_block *crb;
>>> +       struct coprocessor_status_block *csb;
>>> +       struct nx842_workmem *wmem;
>>> +       struct vas_window *txwin;
>>> +       int ret, i = 0;
>>> +       u32 ccw;
>>> +       unsigned int outlen = *outlenp;
>>> +
>>> +       wmem = PTR_ALIGN(workmem, WORKMEM_ALIGN);
>>> +
>>> +       *outlenp = 0;
>>> +
>>> +       crb = &wmem->crb;
>>> +       csb = &crb->csb;
>>> +
>>> +       ret = nx842_config_crb(in, inlen, out, outlen, wmem);
>>> +       if (ret)
>>> +               return ret;
>>> +
>>> +       ccw = 0;
>>> +       ccw = SET_FIELD(CCW_FC_842, ccw, fc);
>>> +       crb->ccw = cpu_to_be32(ccw);
>>> +
>>> +       txwin = wmem->txwin;
>>> +       /* shoudn't happen, we don't load without a coproc */
>>> +       if (!txwin) {
>>> +               pr_err_ratelimited("NX-842 coprocessor is not available");
>>> +               return -ENODEV;
>>> +       }
>>> +
>>> +       do {
>>> +               wmem->start = ktime_get();
>>> +               preempt_disable();
>>> +               /*
>>> +                * VAS copy CRB into L2 cache. Refer <asm/vas.h>.
>>> +                * @crb, @offset and @first (must be true)
>>> +                */
>>> +               vas_copy_crb(crb, 0, 1);
>>> +
>>> +               /*
>>> +                * VAS paste previously copied CRB to NX.
>>> +                * @txwin, @offset, @last (must be true) and @re is
>>> +                * expected/assumed to be true for NX windows.
>>> +                */
>>> +               ret = vas_paste_crb(txwin, 0, 1, 1);
>>> +               preempt_enable();
>>> +               /*
>>> +                * Retry copy/paste function for VAS failures.
>>> +                */
>>> +       } while (ret && (i++ < VAS_RETRIES));
>>> +
>>> +       if (ret) {
>>> +               pr_err_ratelimited("VAS copy/paste failed\n");
>>> +               return ret;
>>> +       }
>>> +
>>> +       ret = wait_for_csb(wmem, csb);
>>> +       if (!ret)
>>> +               *outlenp = be32_to_cpu(csb->count);
>>> +
>>> +       return ret;
>>> +}
>>> +
>>> +/**
>>>   * nx842_powernv_compress - Compress data using the 842 algorithm
>>>   *
>>>   * Compression provided by the NX842 coprocessor on IBM PowerNV systems.
>>> @@ -576,6 +690,198 @@ static inline void nx842_add_coprocs_list(struct nx842_coproc *coproc,
>>>         list_add(&coproc->list, &nx842_coprocs);
>>>  }
>>>
>>> +/*
>>> + * Identify chip ID for each CPU and save coprocesor adddress for the
>>> + * corresponding NX engine in percpu coproc_inst.
>>> + * coproc_inst is used in crypto_init to open send window on the NX instance
>>> + * for the corresponding CPU / chip where the open request is executed.
>>> + */
>>> +static void nx842_set_per_cpu_coproc(struct nx842_coproc *coproc)
>>> +{
>>> +       unsigned int i, chip_id;
>>> +
>>> +       for_each_possible_cpu(i) {
>>> +               chip_id = cpu_to_chip_id(i);
>>> +
>>> +               if (coproc->chip_id == chip_id)
>>> +                       per_cpu(coproc_inst, i) = coproc;
>>> +       }
>>> +}
>>> +
>>> +
>>> +static struct vas_window *nx842_alloc_txwin(struct nx842_coproc *coproc)
>>> +{
>>> +       struct vas_window *txwin = NULL;
>>> +       struct vas_tx_win_attr txattr;
>>> +
>>> +       /*
>>> +        * Kernel requests will be high priority. So open send
>>> +        * windows only for high priority RxFIFO entries.
>>> +        */
>>> +       vas_init_tx_win_attr(&txattr, coproc->ct);
>>> +       txattr.lpid = 0;        /* lpid is 0 for kernel requests */
>>> +       txattr.pid = mfspr(SPRN_PID);
>>> +
>>> +       /*
>>> +        * Open a VAS send window which is used to send request to NX.
>>> +        */
>>> +       txwin = vas_tx_win_open(coproc->vas.id, coproc->ct, &txattr);
>>> +       if (IS_ERR(txwin)) {
>>> +               pr_err("ibm,nx-842: Can not open TX window: %ld\n",
>>> +                               PTR_ERR(txwin));
>>> +               return NULL;
>>> +       }
>>> +
>>> +       return txwin;
>>> +}
>>> +
>>> +static int __init vas_cfg_coproc_info(struct device_node *dn, int chip_id,
>>> +                                       int vasid)
>>> +{
>>> +       struct vas_window *rxwin = NULL;
>>> +       struct vas_rx_win_attr rxattr;
>>> +       struct nx842_coproc *coproc;
>>> +       u32 lpid, pid, tid, fifo_size;
>>> +       u64 rx_fifo;
>>> +       const char *priority;
>>> +       int ret;
>>> +
>>> +       ret = of_property_read_u64(dn, "rx-fifo-address", (void *)&rx_fifo);
>>> +       if (ret) {
>>> +               pr_err("Missing rx-fifo-address property\n");
>>> +               return ret;
>>> +       }
>>> +
>>> +       ret = of_property_read_u32(dn, "rx-fifo-size", &fifo_size);
>>> +       if (ret) {
>>> +               pr_err("Missing rx-fifo-size property\n");
>>> +               return ret;
>>> +       }
>>> +
>>> +       ret = of_property_read_u32(dn, "lpid", &lpid);
>>> +       if (ret) {
>>> +               pr_err("Missing lpid property\n");
>>> +               return ret;
>>> +       }
>>> +
>>> +       ret = of_property_read_u32(dn, "pid", &pid);
>>> +       if (ret) {
>>> +               pr_err("Missing pid property\n");
>>> +               return ret;
>>> +       }
>>> +
>>> +       ret = of_property_read_u32(dn, "tid", &tid);
>>> +       if (ret) {
>>> +               pr_err("Missing tid property\n");
>>> +               return ret;
>>> +       }
>>> +
>>> +       ret = of_property_read_string(dn, "priority", &priority);
>>> +       if (ret) {
>>> +               pr_err("Missing priority property\n");
>>> +               return ret;
>>> +       }
>>> +
>>> +       coproc = kzalloc(sizeof(*coproc), GFP_KERNEL);
>>> +       if (!coproc)
>>> +               return -ENOMEM;
>>> +
>>> +       if (!strcmp(priority, "High"))
>>> +               coproc->ct = VAS_COP_TYPE_842_HIPRI;
>>> +       else if (!strcmp(priority, "Normal"))
>>> +               coproc->ct = VAS_COP_TYPE_842;
>>> +       else {
>>> +               pr_err("Invalid RxFIFO priority value\n");
>>> +               ret =  -EINVAL;
>>> +               goto err_out;
>>> +       }
>>> +
>>> +       vas_init_rx_win_attr(&rxattr, coproc->ct);
>>> +       rxattr.rx_fifo = (void *)rx_fifo;
>>> +       rxattr.rx_fifo_size = fifo_size;
>>> +       rxattr.lnotify_lpid = lpid;
>>> +       rxattr.lnotify_pid = pid;
>>> +       rxattr.lnotify_tid = tid;
>>> +       rxattr.wcreds_max = MAX_CREDITS_PER_RXFIFO;
>>> +
>>> +       /*
>>> +        * Open a VAS receice window which is used to configure RxFIFO
>>> +        * for NX.
>>> +        */
>>> +       rxwin = vas_rx_win_open(vasid, coproc->ct, &rxattr);
>>> +       if (IS_ERR(rxwin)) {
>>> +               ret = PTR_ERR(rxwin);
>>> +               pr_err("setting RxFIFO with VAS failed: %d\n",
>>> +                       ret);
>>> +               goto err_out;
>>> +       }
>>> +
>>> +       coproc->vas.rxwin = rxwin;
>>> +       coproc->vas.id = vasid;
>>> +       nx842_add_coprocs_list(coproc, chip_id);
>>> +
>>> +       /*
>>> +        * Kernel requests use only high priority FIFOs. So save coproc
>>> +        * info in percpu coproc_inst which will be used to open send
>>> +        * windows for crypto open requests later.
>>> +        */
>>> +       if (coproc->ct == VAS_COP_TYPE_842_HIPRI)
>>> +               nx842_set_per_cpu_coproc(coproc);
>>> +
>>> +       return 0;
>>> +
>>> +err_out:
>>> +       kfree(coproc);
>>> +       return ret;
>>> +}
>>> +
>>> +
>>> +static int __init nx842_powernv_probe_vas(struct device_node *pn)
>>> +{
>>> +       struct device_node *dn;
>>> +       int chip_id, vasid, ret = 0;
>>> +       int nx_fifo_found = 0;
>>> +
>>> +       chip_id = of_get_ibm_chip_id(pn);
>>> +       if (chip_id < 0) {
>>> +               pr_err("ibm,chip-id missing\n");
>>> +               return -EINVAL;
>>> +       }
>>> +
>>> +       dn = of_find_compatible_node(pn, NULL, "ibm,power9-vas-x");
>>> +
>>> +       if (!dn) {
>>> +               pr_err("Missing VAS device node\n");
>>> +               return -EINVAL;
>>> +       }
>>> +
>>> +       if (of_property_read_u32(dn, "ibm,vas-id", &vasid)) {
>>> +               pr_err("Missing ibm,vas-id device property\n");
>>> +               of_node_put(dn);
>>> +               return -EINVAL;
>>> +       }
>>> +
>>> +       of_node_put(dn);
>>> +
>>> +       for_each_child_of_node(pn, dn) {
>>> +               if (of_device_is_compatible(dn, "ibm,p9-nx-842")) {
>>> +                       ret = vas_cfg_coproc_info(dn, chip_id, vasid);
>>> +                       if (ret) {
>>> +                               of_node_put(dn);
>>> +                               return ret;
>>> +                       }
>>> +                       nx_fifo_found++;
>>> +               }
>>> +       }
>>> +
>>> +       if (!nx_fifo_found) {
>>> +               pr_err("NX842 FIFO nodes are missing\n");
>>> +               ret = -EINVAL;
>>> +       }
>>> +
>>> +       return ret;
>>> +}
>>> +
>>>  static int __init nx842_powernv_probe(struct device_node *dn)
>>>  {
>>>         struct nx842_coproc *coproc;
>>> @@ -622,6 +928,9 @@ static void nx842_delete_coprocs(void)
>>>         struct nx842_coproc *coproc, *n;
>>>
>>>         list_for_each_entry_safe(coproc, n, &nx842_coprocs, list) {
>>> +               if (coproc->vas.rxwin)
>>> +                       vas_win_close(coproc->vas.rxwin);
>>> +
>>>                 list_del(&coproc->list);
>>>                 kfree(coproc);
>>>         }
>>> @@ -643,6 +952,46 @@ static struct nx842_driver nx842_powernv_driver = {
>>>         .decompress =   nx842_powernv_decompress,
>>>  };
>>>
>>> +static int nx842_powernv_crypto_init_vas(struct crypto_tfm *tfm)
>>> +{
>>> +       struct nx842_crypto_ctx *ctx = crypto_tfm_ctx(tfm);
>>> +       struct nx842_workmem *wmem;
>>> +       struct nx842_coproc *coproc;
>>> +       int ret;
>>> +
>>> +       ret = nx842_crypto_init(tfm, &nx842_powernv_driver);
>>> +
>>> +       if (ret)
>>> +               return ret;
>>> +
>>> +       wmem = PTR_ALIGN((struct nx842_workmem *)ctx->wmem, WORKMEM_ALIGN);
>>> +       coproc = per_cpu(coproc_inst, smp_processor_id());
>>
>> this is wrong.  the crypto transform init function is not guaranteed
>> to be called by the same processor that later uses it.  Just because
>> that happens to be how zswap operates doesn't guarantee other crypto
>> users will do the same.
>
> Dan, Sorry missed this comment.
>
> Right, The actual crypto request can be executed on other processor than the CPU when the init is executed. The main goal is open send window on the NX engine which is on the same chip for the corresponding CPU. So we are OK if the request is scheduled on other CPU as long as it belongs to same chip. Otherwise in the worst case we will end up using remote NX.

ok, but there's no guarantee of future crypto calls being on the same
chip either, so i still don't understand why you're doing this.  if
you want each crypto comp/decomp call to be cpu-local or node-local,
then choose which corpco/txwin to use at comp/decomp time, not
transform init time.


>
> Thanks
> Haren
>
>
>
>




[Index of Archives]     [Kernel]     [Gnu Classpath]     [Gnu Crypto]     [DM Crypt]     [Netfilter]     [Bugtraq]

  Powered by Linux