| Message ID | 1508123061-6600-9-git-send-email-yi.y.sun@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
>>> On 16.10.17 at 05:04, <yi.y.sun@linux.intel.com> wrote: > This patch implements set value flow for MBA including its callback > function and domctl interface. > > Signed-off-by: Yi Sun <yi.y.sun@linux.intel.com> Reviewed-by: Jan Beulich <jbeulich@suse.com> > v8: > - restore some unnecessary changes in 'cat_check_cbm'. > (suggested by Jan Beulich) This reads as if you did exactly the opposite thing of what you actually did. > +static bool mba_sanitize_thrtl(const struct feat_node *feat, uint32_t *thrtl) > +{ > + if ( *thrtl > feat->mba.thrtl_max ) > + return false; Wouldn't it be better to do this check after ... > + /* > + * Per SDM (chapter "Memory Bandwidth Allocation Configuration"): > + * 1. Linear mode: In the linear mode the input precision is defined > + * as 100-(MBA_MAX). For instance, if the MBA_MAX value is 90, the > + * input precision is 10%. Values not an even multiple of the > + * precision (e.g., 12%) will be rounded down (e.g., to 10% delay > + * applied). > + * 2. Non-linear mode: Input delay values are powers-of-two from zero > + * to the MBA_MAX value from CPUID. In this case any values not a > + * power of two will be rounded down the next nearest power of two. > + */ > + if ( feat->mba.linear ) > + *thrtl -= *thrtl % (100 - feat->mba.thrtl_max); > + else > + { > + /* Not power of 2. */ > + if ( *thrtl & (*thrtl - 1) ) > + *thrtl = 1 << (fls(*thrtl) - 1); > + } ... these adjustments? That way someone specifying e.g. a linear value of 95% would get 90% just like for 85% (s)he would get 80%. > + return true; If so, the return statement could simply become return *thrtl <= feat->mba.thrtl_max; My R-b also applies if you decide to make this change. Jan
On 17-10-16 06:49:49, Jan Beulich wrote: > >>> On 16.10.17 at 05:04, <yi.y.sun@linux.intel.com> wrote: > > This patch implements set value flow for MBA including its callback > > function and domctl interface. > > > > Signed-off-by: Yi Sun <yi.y.sun@linux.intel.com> > > Reviewed-by: Jan Beulich <jbeulich@suse.com> > > > v8: > > - restore some unnecessary changes in 'cat_check_cbm'. > > (suggested by Jan Beulich) > > This reads as if you did exactly the opposite thing of what you > actually did. > > > +static bool mba_sanitize_thrtl(const struct feat_node *feat, uint32_t *thrtl) > > +{ > > + if ( *thrtl > feat->mba.thrtl_max ) > > + return false; > > Wouldn't it be better to do this check after ... > > > + /* > > + * Per SDM (chapter "Memory Bandwidth Allocation Configuration"): > > + * 1. Linear mode: In the linear mode the input precision is defined > > + * as 100-(MBA_MAX). For instance, if the MBA_MAX value is 90, the > > + * input precision is 10%. Values not an even multiple of the > > + * precision (e.g., 12%) will be rounded down (e.g., to 10% delay > > + * applied). > > + * 2. Non-linear mode: Input delay values are powers-of-two from zero > > + * to the MBA_MAX value from CPUID. In this case any values not a > > + * power of two will be rounded down the next nearest power of two. > > + */ > > + if ( feat->mba.linear ) > > + *thrtl -= *thrtl % (100 - feat->mba.thrtl_max); > > + else > > + { > > + /* Not power of 2. */ > > + if ( *thrtl & (*thrtl - 1) ) > > + *thrtl = 1 << (fls(*thrtl) - 1); > > + } > > ... these adjustments? That way someone specifying e.g. a linear > value of 95% would get 90% just like for 85% (s)he would get > 80%. > > > + return true; > > If so, the return statement could simply become > > return *thrtl <= feat->mba.thrtl_max; > > My R-b also applies if you decide to make this change. > Thank you for the suggestion! It is not worthy to send whole patch set out. I will just update this patch. > Jan
diff --git a/xen/arch/x86/domctl.c b/xen/arch/x86/domctl.c index b726eae..4bca15d 100644 --- a/xen/arch/x86/domctl.c +++ b/xen/arch/x86/domctl.c @@ -1465,6 +1465,12 @@ long arch_do_domctl( PSR_TYPE_L2_CBM); break; + case XEN_DOMCTL_PSR_SET_MBA_THRTL: + ret = psr_set_val(d, domctl->u.psr_alloc.target, + domctl->u.psr_alloc.data, + PSR_TYPE_MBA_THRTL); + break; + #define domctl_psr_get_val(d, domctl, type, copyback) ({ \ uint32_t v_; \ int r_ = psr_get_val((d), (domctl)->u.psr_alloc.target, \ diff --git a/xen/arch/x86/psr.c b/xen/arch/x86/psr.c index 549f21b..04dd4a1 100644 --- a/xen/arch/x86/psr.c +++ b/xen/arch/x86/psr.c @@ -138,6 +138,12 @@ static const struct feat_props { /* write_msr is used to write out feature MSR register. */ void (*write_msr)(unsigned int cos, uint32_t val, enum psr_type type); + + /* + * sanitize is used to check if input val fulfills SDM requirement. + * And change it to valid value if SDM allows. + */ + bool (*sanitize)(const struct feat_node *feat, uint32_t *val); } *feat_props[FEAT_TYPE_NUM]; /* @@ -274,11 +280,14 @@ static enum psr_feat_type psr_type_to_feat_type(enum psr_type type) return feat_type; } -static bool psr_check_cbm(unsigned int cbm_len, unsigned long cbm) +/* Implementation of allocation features' functions. */ +static bool cat_check_cbm(const struct feat_node *feat, uint32_t *val) { unsigned int first_bit, zero_bit; + unsigned int cbm_len = feat->cat.cbm_len; + unsigned long cbm = *val; - /* Set bits should only in the range of [0, cbm_len]. */ + /* Set bits should only in the range of [0, cbm_len). */ if ( cbm & (~0ul << cbm_len) ) return false; @@ -297,7 +306,6 @@ static bool psr_check_cbm(unsigned int cbm_len, unsigned long cbm) return true; } -/* Implementation of allocation features' functions. */ static bool cat_init_feature(const struct cpuid_leaf *regs, struct feat_node *feat, struct psr_socket_info *info, @@ -434,6 +442,7 @@ static const struct feat_props l3_cat_props = { .alt_type = PSR_TYPE_UNKNOWN, .get_feat_info = cat_get_feat_info, .write_msr = l3_cat_write_msr, + .sanitize = cat_check_cbm, }; /* L3 CDP props */ @@ -464,6 +473,7 @@ static const struct feat_props l3_cdp_props = { .alt_type = PSR_TYPE_L3_CBM, .get_feat_info = l3_cdp_get_feat_info, .write_msr = l3_cdp_write_msr, + .sanitize = cat_check_cbm, }; /* L2 CAT props */ @@ -479,6 +489,7 @@ static const struct feat_props l2_cat_props = { .alt_type = PSR_TYPE_UNKNOWN, .get_feat_info = cat_get_feat_info, .write_msr = l2_cat_write_msr, + .sanitize = cat_check_cbm, }; /* MBA props */ @@ -499,6 +510,35 @@ static bool mba_get_feat_info(const struct feat_node *feat, static void mba_write_msr(unsigned int cos, uint32_t val, enum psr_type type) { + wrmsrl(MSR_IA32_PSR_MBA_MASK(cos), val); +} + +static bool mba_sanitize_thrtl(const struct feat_node *feat, uint32_t *thrtl) +{ + if ( *thrtl > feat->mba.thrtl_max ) + return false; + + /* + * Per SDM (chapter "Memory Bandwidth Allocation Configuration"): + * 1. Linear mode: In the linear mode the input precision is defined + * as 100-(MBA_MAX). For instance, if the MBA_MAX value is 90, the + * input precision is 10%. Values not an even multiple of the + * precision (e.g., 12%) will be rounded down (e.g., to 10% delay + * applied). + * 2. Non-linear mode: Input delay values are powers-of-two from zero + * to the MBA_MAX value from CPUID. In this case any values not a + * power of two will be rounded down the next nearest power of two. + */ + if ( feat->mba.linear ) + *thrtl -= *thrtl % (100 - feat->mba.thrtl_max); + else + { + /* Not power of 2. */ + if ( *thrtl & (*thrtl - 1) ) + *thrtl = 1 << (fls(*thrtl) - 1); + } + + return true; } static const struct feat_props mba_props = { @@ -507,6 +547,7 @@ static const struct feat_props mba_props = { .alt_type = PSR_TYPE_UNKNOWN, .get_feat_info = mba_get_feat_info, .write_msr = mba_write_msr, + .sanitize = mba_sanitize_thrtl, }; static bool __init parse_psr_bool(const char *s, const char *delim, @@ -972,7 +1013,7 @@ static int insert_val_into_array(uint32_t val[], if ( array_len < props->cos_num ) return -ENOSPC; - if ( !psr_check_cbm(feat->cat.cbm_len, new_val) ) + if ( !props->sanitize(feat, &new_val) ) return -EINVAL; /* diff --git a/xen/include/public/domctl.h b/xen/include/public/domctl.h index 002a26f..b312c69 100644 --- a/xen/include/public/domctl.h +++ b/xen/include/public/domctl.h @@ -1070,6 +1070,7 @@ struct xen_domctl_psr_alloc { #define XEN_DOMCTL_PSR_GET_L3_DATA 5 #define XEN_DOMCTL_PSR_SET_L2_CBM 6 #define XEN_DOMCTL_PSR_GET_L2_CBM 7 +#define XEN_DOMCTL_PSR_SET_MBA_THRTL 8 #define XEN_DOMCTL_PSR_GET_MBA_THRTL 9 uint32_t cmd; /* IN: XEN_DOMCTL_PSR_* */ uint32_t target; /* IN */
This patch implements set value flow for MBA including its callback function and domctl interface. Signed-off-by: Yi Sun <yi.y.sun@linux.intel.com> --- CC: Jan Beulich <jbeulich@suse.com> CC: Andrew Cooper <andrew.cooper3@citrix.com> CC: Wei Liu <wei.liu2@citrix.com> CC: Roger Pau Monné <roger.pau@citrix.com> CC: Chao Peng <chao.p.peng@linux.intel.com> v8: - restore some unnecessary changes in 'cat_check_cbm'. (suggested by Jan Beulich) - use 'fls()' but not 'flsl()'. (suggested by Jan Beulich) - use plain '=' to assign value for thrtl in 'mba_sanitize_thrtl'. (suggested by Jan Beulich) v7: - change name of 'check_val' to 'sanitize'. (suggested by Jan Beulich) - fix comments. (suggested by Jan Beulich) - add parentheses and change '== 0' to '!'. (suggested by Jan Beulich) - remove unnecessary check of 'mba.thrtl_max'. (suggested by Jan Beulich) - remove unnecessary intermediate variable 'mod'. (suggested by Jan Beulich) - refine an assignement sentence to use '&='. (suggested by Jan Beulich) - change type of last parameter of 'sanitize' to 'uint32_t' and apply same change to 'cat_check_cbm'. (suggested by Jan Beulich) v6: - split co-exist features' values setting flow to a new patch. (suggested by Jan Beulich) - restore codes related to 'mba_check_thrtl' and 'check_value'. (suggested by Jan Beulich) v5: - adjust position of 'cat_check_cbm' to not to make changes so big. (suggested by Roger Pau Monné) - remove 'props' from 'struct cos_write_info'. (suggested by Roger Pau Monné) - make a single return statement in 'mba_check_thrtl'. (suggested by Jan Beulich) v4: - remove 'ALLOC_' from macro names. (suggested by Roger Pau Monné) - join two checks into a single if. (suggested by Roger Pau Monné) - remove redundant local variable 'array_len'. (suggested by Roger Pau Monné) v3: - modify commit message to make it clear. (suggested by Roger Pau Monné) - modify functionality of 'check_val' to make it simple to only check value. Change the last parameter type from 'unsigned long *' to 'unsigned long'. (suggested by Roger Pau Monné) - call rdmsrl to get value just written into MSR for MBA. Because HW can automatically change input value to what it wants. (suggested by Roger Pau Monné) - change type of 'write_msr' to 'uint32_t' to return the value actually written into MSR. Then, change 'do_write_psr_msrs' to set the returned value into 'cos_reg_val[]' - move the declaration of 'j' into loop in 'do_write_psr_msrs'. (suggested by Roger Pau Monné) - change 'mba_info' to 'mba'. (suggested by Roger Pau Monné) - change 'cat_info' to 'cat'. (suggested by Roger Pau Monné) - rename 'psr_cat/PSR_CAT' to 'psr_alloc/PSR_ALLOC' and remove 'op/OP' from name. (suggested by Roger Pau Monné) - change 'PSR_VAL_TYPE_MBA' to 'PSR_TYPE_MBA_THRTL'. (suggested by Roger Pau Monné) v2: - remove linear mode 'thrtl_max' check in 'mba_check_thrtl' because it has been checked in 'mba_init_feature'. (suggested by Chao Peng) - for non-linear mode, check if '*thrtl' is not 0 in 'mba_check_thrtl'. If it is 0, we do not need to change it. (suggested by Chao Peng) - move comments to explain changes of 'cos_write_info' from psr.c to commit message. (suggested by Chao Peng) --- xen/arch/x86/domctl.c | 6 ++++++ xen/arch/x86/psr.c | 49 +++++++++++++++++++++++++++++++++++++++++---- xen/include/public/domctl.h | 1 + 3 files changed, 52 insertions(+), 4 deletions(-)